Contents - ä¸å½æ£®æçç©å¤æ ·æ§çæµç½ç»

庆研究员、曹敏研究员、米湘成副研究员、李先琨研究员、金光泽教授、王希华教授、丁炳杨教授、杜晓军博士等对论文搜集工作的帮助。感谢全体 CForBio相关研究人员和研究生们的艰苦努力和卓有成效的贡献 ; 感谢中山大学何芳良教授、台湾东华大学孙义方教授、美国 Smithsonian 热带研究所热带森林科学研究中心 (CTFS/SIGEO) 和 Earthwatch Institute 的大力支持 ; 感谢科技部、基金委、中国科学院和汇丰银行的项目资助。马克平二零一二年十月二十日于香山

PrefaceOrganized by Biodiversity Committee, Chinese Academy of Sciences, the ChineseForest Biodiversity Monitoring Network (CForBio) began to establish in 2004. It is aresearch base for monitoring changes in biodiversity of forest ecosystems in China,and also an important part of global forest biodiversity monitoring network. It coversmajor forest vegetation types at different latitudes, including coniferous andbroadleaved mixed forest, evergreen broad-leaved forest and tropical rainforest.Twelve permanent plots with area more than 9 ha have been set up by CForBio till2012, including eight mainly by institutes of CAS, the other fourby East ChinaNormal University and Northeast Forestry University. The plots are as belows: 30-hamixed broadleaved-Korean pine forest plot at Fenglin in Xiaoxing’an Mountains inHeilongjiang; 9-ha mixed broadleaved-Korean pine forest plot and 9.12-ha Spruce-firvalley forest plot at Liangshui in Xiaoxing’an Mountains in Heilongjiang; 25-hadeciduous broadleaved Korean pine forest plot at Changbai Mountain in Jilin; 20-hadeciduous broad-leaved forest plot at Dongling Mountain in Beijing; 25-ha deciduousbroad-leaved forest plot at Baotianman in Henan; 25-ha mid-subtropical mountainevergreen and deciduous broadleaved mixed forest plot at Badagong Mountain inHunan; 20- ha subtropical evergreen broad-leaved forest plot at Tiantong Mountain inZhejiang; 24-ha subtropical evergreen broadleaved forest plot at Gutian Mountain inZhejiang; 20-ha lower subtropical evergreen broadleaved forest plot at DinghuMountain in Guangdong; 15-ha karst seasonal rain forest plot at Nonggang inGuangxi; 20-ha tropical rain forest plot at Xishuangbanna in YunnanUp to now, CForBio has made significant progress. Research discoveries have beenpublished in high profile ecological journals, such as Ecology Letters, Ecology,Journal of Ecology, Oikos. From 2011 to 2012, 71 research papers have been

published based on CForBio platform, among which about 44 were published in SCIjournals and 27 CSCD journals.This collection is a second edition of proceedings of Chinese Forest BiodiversityMonitoring Network, the content of the collection compiled the research paperspublished in 2011- 2012, and had a supplement containing the papers before 2011.Compared with the first edition, the papers of Liangshui plot, Tiantong plot andBaishanzu plot were also compiled. The collection of SCI journal papers withappendix of CSCD-paper list (2011- 2012), SCI journal paper list (2007-2010) andCSCD-paper list (2007-2010) were prepared. Related research progress and publishedpapers will be posted to the website http://www.cfbiodiv.org/ in succession.Xuehong Xu, Yinan Liu andProf. Keping Ma, etc. compiled the proceedings. Wewould like to thank Profs. Wanhui Ye, Zhanqing Hao, Min Cao, Xiangcheng Mi,Xiankun Li, Guangze Jin, Xihua Wang, Bingyang Ding and Dr. Xiaojun Du for theirhelp in checking publication list, the whole team of CForBio for their efforts andcontributions to promoting research. We would also like to thank Prof. Fangliang Hefrom Sun Yat-sen University, Prof. I-Fang Sun from Donghua University, support ofCTFS and the Earthwatch Institute. Finally, we are grateful to the Ministry of Scienceand Technology, National Natural Science Foundation of China, Chinese Academy ofSciences, and HSBC for their financial support.Keping MaFragrant Hill October 20th, 2012

论文目录ContentsSCI 刊物上发表的文章目录Research papers published in SCI journalsDensity dependence across multiple life stages in a temperate old-growth forest of NortheastChina. (Oecologia)Tiefeng Piao, Liza S. Comita, Guangze Jin* and Ji Hong Kim ................................................1Estimate of leaf area index in an old-growth mixed broadleaved-Korean pine forest innortheastern China. ( PLoS ONE)Zhili Liu, Guangze Jin* and Yujiao Qi...................................................................................12Optical and litter collection methods for measuring leaf area index in an old-growthtemperate forest in northeastern China. ( Journal of Forest Research)Yujiao Qi, Guangze Jin* and Zhili Liu...................................................................................23Effects of local biotic neighbors and habitat heterogeneity on tree and shrub seedlingsurvival in an old-growth temperate forest. (Oecologia)Xuejiao Bai, Simon A. Queenborough, Xugao Wang, Jian Zhang, Buhang Li, Zuoqiang Yuan,Dingliang Xing, Fei Lin, Ji Ye and Zhanqing Hao .................................................................33Testing the independent species’ arrangement assertion made by theories of stochasticgeometry of biodiversity. (Proceedings of the Royal Society, B: Biological Sciences)Thorsten Wiegand, Andreas Huth, Stephan Getzin, Xugao Wang, Zhanqing Hao, C. V.Savitri, Gunatilleke and I. A. U. Nimal Gunatilleke ...............................................................44Local-scale drivers of tree survival in a temperate forest. (PLoS ONE)Xugao Wang, Liza S. Comita, Zhanqing Hao*, Stuart J. Davies, Ji Ye, Fei Lin and ZuoqiangYuan........................................................................................................................................53Effects of soil water and nitrogen on growth and photosynthetic responses of manchurianash (Fraxinus mandshurica) seedlings in northeastern China. (PLoS ONE)Miao Wang , Shuai Shi, Fei Lin, Zhanqing Hao, Ping Jiang and Guanhua Dai .....................63What happens below the canopy? Direct and indirect influences of the dominant species onforest vertical layers. (Oikos)Zuoqiang Yuan, Antonio Gazol, Xugao Wang, Dingliang Xing, Fei Lin, Xuejiao Bai,Yuqiang Zhao, Buhang Li and Zhanqing Hao ........................................................................75Seed rain dynamics reveals strong dispersal limitation, different reproductive strategies andresponse to climate in a temperate forest in Northeast China. (Journal of Vegetation Science)Buhang Li, Zhanqing Hao*, Yue Bin, Jian Zhang and Miao Wang .......................................84Spatial patterns of tree species richness in two temperate forests. (Journal of Ecology)Xugao Wang, Thorsten Wiegand, Amy Wolf, Robert Howe, Stuart J. Davies and ZhanqingHao..........................................................................................................................................93Scale specific determinants of tree diversity in an old growth temperate forest in China.(Basic and Applied Ecology)Zuoqiang Yuan, Xugao Wang,Antonio Gazol,Buhang Li,Fei Lin,Ji Ye,Xuejiao Bai andZhanqing Hao........................................................................................................................105I

Spatial distributions of species in an old-growth temperate forest, northeastern China.(Canadian Journal of Forest Research)Xugao Wang, Ji Ye, Buhang Li, Jian Zhang, Fei Lin and Zhanqing Hao.............................113Species associations in an old-growth temperate forest in north-eastern China. (Journal ofEcology)Xugao Wang, Thorsten Wiegand, Zhanqing Hao, Buhang Li, Ji Ye and Fei Lin.................122Spatial patterns and associations of six congeneric species in an old-growth temperate forest.(Acta Oecologica)Jian Zhang, Bo Song, Buhang Li, Ji Ye, Xugao Wang and Zhanqing Hao ..........................135Fine-scale species co-occurrence patterns in an old-growth temperate forest. (Forest Ecologyand Management)Jian Zhang, Zhanqing Hao, Bo Song, Buhang Li, Xugao Wang and Ji Ye ..........................145The contribution of rare species to community phylogenetic diversity across a globalnetwork of forest plots. (The American Naturalist)Xiangcheng Mi, Nathan G. Swenson, Renato Valencia, John Kress, David L. Erickson,Álvaro Pérez-Castañeda, Haibao Ren, Sheng-Hsin Su, Nimal Gunatilleke, Savi Gunatilleke,Zhanqing Hao, Wanhui Ye, Min Cao, H S Suresh, H S Dattaraja, S Sukumar and Keping Ma...............................................................................................................................................151Covariation in plant functional traits and soil fertility within two species-rich forests. (PLoSONE)Xiaojuan Liu, Nathan G. Swenson, S. Joseph Wright, Liwen Zhang, Kai Song, Yanjun Du,Jinlong Zhang, Xiangcheng Mi, Haibao Ren and Keping Ma ..............................................165Age and radial growth pattern of four tree species in a subtropical forest of China. (Trees)Pei Xing, Qibin Zhang and Patrick J. Baker .........................................................................174Comparison of phylobetadiversity indices based on community data from Gutianshan forestplot. (Chinese Science Bulletin)Gang Feng, Jinlong Zhang, Nancai, Mide Rao, Xiangcheng Mi,Haibao Ren and Keping Ma...............................................................................................................................................182The effects of ice storm on seed rain and seed limitation in an evergreen broad-leaved forestin east China. (Acta Oecologica)Yanjun Du, Xiangcheng Mi, Xiaojuan Liu and Keping Ma .................................................190Comparison of seed rain and seed limitation between community understory and gaps in asubtropical evergreen forest. (Acta Oecologica)Yanjun Du, Xiangcheng Mi and Keping Ma ........................................................................197A genome evolution-based framework for measures of originality for clades. (Journal ofTheoretical Biology)Jianxiong Huang, Xiangcheng Mi and Keping Ma...............................................................206Strong plant-soil associations in a heterogeneous subtropical broad-leaved forest. (Plant andSoil)Liwen Zhang, Xiangcheng Mi, Hongbo Shao and Keping Ma.............................................213Separating the effect of mechanisms shaping species-abundance distributions at multiplescales in a subtropical forest. (Oikos)Jiajia Cheng, Xiangcheng Mi, Jintun Zhang and Keping Ma ...............................................223Density dependence is prevalent in a heterogeneous subtropical forest. (Oikos)II

Yan Zhu, Xiangcheng Mi, Haibao Ren and Keping Ma .......................................................232Clonal integration increases performance of ramets of the fern Diplopterygium glaucum inan evergreen forest in southeastern China. (Flora)Juan Du, Ning Wang, Peter Alpert, MingjianYu, Feihai Yu and Ming Dong ......................243Factors affecting detection probability in plant distribution studies. (Journal of Ecology)Guoke Chen, Marc Kéry, Jinlong Zhang and Keping Ma.....................................................248Species-area relationships explained by the joint effects of dispersal limitation and habitatheterogeneity. (Ecology)Guochun Shen, Mingjian Yu, Xinsheng Hu, Xiangcheng Mi, Haibao Ren, I-fang Sun andKeping Ma.............................................................................................................................255Seed dispersal phenology and dispersal syndromes in a subtropical broad-leaved forest ofChina. (Forest Ecology and Management)Yanjun Du, Xiangcheng Mi, Haibao Ren, Xiaojuan Liu, Lei Chen, Teng Fang, Yan Zhu andKeping Ma.............................................................................................................................268Spatial associations of tree species in a subtropical evergreen broad-leaved forest. (Journalof Plant Ecology)Zhengrong Luo, Mingjian Yu, DeL Chen, Yougui Wu and Bingyang Ding........................274Density dependence is not very prevalent in a heterogeneous subtropical forest. (Oikos)Zhengrong Luo, Xiangcheng Mi, Xiaorong Chen, Zhenlin Ye and Bingyang Ding ............284Genetic groups in the common plant species Castanopsis chinensis and their associationswith topographic habitats. (Oikos)Zhengfeng Wang, Juyu Lian, Guomin Huang, Wanhui Ye, Honglin Cao and Zhangmin Wang...............................................................................................................................................296Isolation and characterization of 36 polymorphic microsatellite makers in Schima superba(Theaceae). (American Journal of Botany)Hongyu Niu, Xiaoyi Li, Wanhui Ye, Zhengfeng Wang, Honglin Cao and Zhangmin Wang...............................................................................................................................................304Topographic effects on fine-scale spatial genetic structure in Castanopsis chinensis Hance(Fagaceae). (Plant Species Biology)Jian He, Xiaoyi Li, Dandan Gao, Peng Zhu, Zhengfeng Wang*, Zhangming Wang, WanhuiYe and Honglin Cao..............................................................................................................308Multifractal analysis of diversity scaling laws in a subtropical forest. (Ecological Complexity)Shiguang Wei, Lin Li, Zhongliang Huang, Wanhui Ye, Guiquan gong, Xiaoyong Zhou andJuyu Lian...............................................................................................................................315Seedling recruitment patterns in a 20 ha subtropical forest plot: hints for niche-basedprocesses and negative density dependence. (European Journal of Forest Research)Bin Yue, Guojun Lin, Buhang Li, Linfang Wu, Yong Shen and Wanhui Ye.......................322Tree mortality and recruitment in a subtropical broadleaved monsoon forest in South China.(Journal of Tropical Forest Science)Bin Yue, Lian Juyu, Wang Zhangming, Ye Wanhui *, Cao Honglin...................................331Exploring tree-habitat associations in a Chinese subtropical forest plot using a molecularphylogeny generated from DNA barcode Loci. (PLoS ONE)Nancai Pei, Juyu Lian, David L. Erickson, Nathan G. Swenson, W. John Kress, Wanhui Yeand Xuejun Ge ......................................................................................................................341III

Isolation and characterization of microsatellite loci in Castanopsis fissa in lower subtropicalChina. (Silvae Genetica)Lei Dong, Zhengfeng Wang, Peng Zhu and Wanhui Ye ......................................................350Spatial distribution and interspecific associations of tree species in a tropical seasonal rainforest of China. (PLoS ONE)Guoyu Lan, Stephan Getzin, Thorsten Wiegand, Yuehua Hu, Guishui Xie, Hua Zhu and MinCao ........................................................................................................................................352Buttress trees in a 20-hectare tropical dipterocarp rainforest in Xishuangbanna, SW China.(Journal of Plant Ecology)Zhiyuan He, Yong Tang, Xiaobao Deng and Min Cao.........................................................361Seasonal differentiation in density-dependent seedling survival in a tropical rain forest.(Journal of Ecology)Luxiang Lin, Lisa S. Comita, Zheng Zheng and Min Cao....................................................367Dominant species and dispersal limitation regulate tree species distributions in a 20-ha plotin Xishuangbanna, Southwest China. (Oikos)Yuehua Hu, Liqing Sha, F. Guillaume Blanchet, Jiaolin Zhang, Yong Tang, Guoyu Lan andMin Cao ................................................................................................................................377Strong neutral spatial effects shape tree species distributions across life stages at multiplescales. (PLoS ONE)Yuehua Hu,, Guoyu Lan, Liqing Sha, Min Cao, Yong Tang, Yide Li and Daping Xu ........386Tree species diversity of a 20-ha plot in a tropical seasonal rainforest in Xishuangbanna,southwest China. (Journal of Forest Research)Guoyu Lan, Hua Zhu and Min Cao.......................................................................................395Lianas as structural parasites: a re-evaluation.(Chinese Science Bulletin)Yong Tang, Roger L. Kitching and Min Cao........................................................................403Topography related spatial distribution of dominant tree species in a tropical seasonal rainforest in China. (Forest Ecology and Management)Guoyu Lan, Yuehua Hua, Min Cao and Hua Zhu.................................................................409附录 Ⅰ: 中文及其它期刊发表的文章目录 ...............................................................................416AppendixⅠ: Research papers published in non-SCI journals......................................................419附录 Ⅱ:2006-2010 发表论文目录 ............................................................................................423AppendixⅡ: Research papers in 2006-2010IV

OecologiaDOI 10.1007/s00442-012-2481-yCOMMUNITY ECOLOGY - ORIGINAL RESEARCHDensity dependence across multiple life stages in a temperateold-growth forest of northeast ChinaTiefeng Piao • Liza S. Comita • Guangze Jin •Ji Hong KimReceived: 12 November 2011 / Accepted: 10 September 2012Ó The Author(s) 2012. This article is published with open access at Springerlink.comAbstract Recent studies on species coexistence suggestthat density dependence is an important mechanism regulatingplant populations. However, there have been few studiesof density dependence conducted for more than one life-historystage or that control for habitat heterogeneity, which mayinfluence spatial patterns of survival and mask densitydependence. We explored the prevalence of density dependenceacross multiple life stages, and the effects of controllingfor habitat heterogeneity, in a temperate forest in northeastChina. We used generalized linear mixed-effects models totest for density-dependent mortality of seedlings and spatialpoint pattern analysis to detect density dependence forCommunicated by Miguel Franco.Electronic supplementary material The online version of thisarticle (doi:10.1007/s00442-012-2481-y) contains supplementarymaterial, which is available to authorized users.T. Piao G. Jin (&)Center for Ecological Research, Northeast Forestry University,Harbin 150040, Chinae-mail: taxus@126.comPresent Address:T. PiaoCollege of Forest and Environmental Sciences,Kangwon National University, Chuncheon 200-701, KoreaL. S. ComitaDepartment of Evolution, Ecology and Organismal Biology,The Ohio State University, Columbus, OH 43210, USAL. S. ComitaSmithsonian Tropical Research Institute, Box 0843-03092,Balboa, Ancón, Republic of PanamáJ. H. KimCollege of Forest and Environmental Sciences, KangwonNational University, Chuncheon 200-701, Koreasapling-to-juvenile transitions. Conspecific neighbors had anegative effect on survival of plants in both life stages. At theseedling stage, we found a negative effect of conspecificseedling neighbors on survival when analyzing all speciescombined. However, in species-level analyses, only 2 of 11focal species were negatively impacted by conspecificneighbors, indicating wide variation among species in thestrength of density dependence. Controlling for habitat heterogeneitydid not alter our findings of density dependence atthe seedling stage. For the sapling-to-juvenile transition stage,11 of 15 focal species showed patterns of local scale (\10 m)conspecific thinning, consistent with negative densitydependence. The results varied depending on whether wecontrolled for habitat heterogeneity, indicating that a failure toaccount for habitat heterogeneity can obscure patterns ofdensity dependence. We conclude that density dependencemay promote tree species coexistence by acting across multiplelife-history stages in this temperate forest.Keywords Species coexistence Janzen–Connellhypothesis Liangshui FDP Pinus koraiensis Habitat heterogeneityIntroductionUnderstanding the mechanisms of population size regulationis of vital importance in the study of species coexistenceand biodiversity maintenance. Recent studies haveprovided strong evidence that density-dependent processesplay a role in shaping plant communities (Wills and Condit1999; Harms et al. 2000; Hille Ris Lambers et al. 2002;Comita et al. 2010). Density-dependent mortality andgrowth can be generated by intraspecific competition forresources (Wright 2002). In addition, since Janzen (1970)1231

Oecologiaand Connell (1971) reported that host-specific naturalenemies reduce survival when a species occurs at highlocal densities, specialized herbivore and pathogen-inducednegative density dependence has also been considered apotentially important mechanism regulating populationdynamics and facilitating species coexistence in diversetree communities (Wright 2002).Numerous studies have examined the importance ofdensity dependence in forests. For example, Harms et al.(2000) found widespread negative density dependence overthe seed-to-seedling transition for 53 species on BarroColorado Island (BCI), Panama. Metz et al. (2010) found astrong negative impact of conspecific seedling densities andadult abundance on first-year seedling survival in a study of163 species in a lowland Amazonian rain forest. Yamazakiet al. (2009) in their study of eight tree species of a temperateforest found that six of them showed distance- and/ordensity-dependent seedling mortality caused by diseasesand rodents. Comita and Hubbell (2009) tracked establishedseedlings of 235 species in the BCI 50-ha forest dynamicsplot (FDP) over 3 years and also found negative effects ofconspecific neighbors on survival. Together, these andadditional studies (e.g. Webb and Peart 1999; Hille RisLambers et al. 2002; Queenborough et al. 2007; Pigot andLeather 2008) provide evidence for an important role ofnegative density dependence at early life stages.Negative density dependence has also been detected atlater life stages in tree communities. For example, Stoll andNewbery (2005) studied the growth of medium-sized (10 to\100 cm diameter at breast height, dbh) trees of ten abundantoverstory dipterocarp species and found strong negativeeffects of neighbors on their growth in a lowland forest inBorneo. Similarly, Zhang et al. (2009) found tree survivalwas negatively correlated with conspecific basal area for 8 of13 focal species with dbh C 1 cm in the temperate forest ofChangbaishan Mountain, China. Thus, previous studiesindicate that negative density dependence exists at both earlyand later life-history stages of trees. Therefore, to test theprevalence of density dependence in a community, we mustconsider multiple size classes. Mortality patterns in seedlingscan usually be analyzed by direct observation, due to theirhigh mortality rates caused by susceptibility to natural enemiesand environmental stressors. For larger trees that havelower mortality rates and can have a lifespan of severalhundreds of years, however, several years of observation islikely too short to detect effects of density dependence(Ratikainen et al. 2008). However, if we can assume thatpopulations of larger trees are in an equilibrium stage, spatialpoint pattern analysis can be an effective approach for thedetection of lagged effects of density dependence, by lookingat changes in aggregation of each species from early to laterlife-history stages. This is possible because pollen and seeddispersal limitation, which are quite common in plantcommunities (e.g., Hubbell et al. 1999, Cázares-Martínezet al. 2010), may cause spatial aggregation in recruitment(Wright 2002). If there is strong negative density dependence,the degree of conspecific aggregation will declinewith increasing size class, due to lower survival of individualsgrowing in high density patches of conspecifics (Sterneret al. 1986; Barot et al. 1999; Condit et al. 2000).One confounding factor in the analysis of densitydependence is habitat heterogeneity, since a species willtend to perform better when growing in its preferred habitat(Getzin et al. 2008; Murrell 2009). If a species has highlocal density in its preferred habitat and low local densityin marginal habitats, and host-specific natural enemies orintraspecific competition do not offset the habitat advantages,a positive relationship between conspecific densityand performance would be found, despite underlying negativedensity-dependent effects. Therefore, tests for densitydependence must account for habitat heterogeneity. However,this is made difficult by the fact that numerousenvironmental covariates are difficult to quantify (He andDuncan 2000; Wiegand et al. 2007). Getzin et al. (2008)developed a simple method to solve this problem. By usingadult trees as ‘‘controls’’, they factored out habitat heterogeneityand were able to detect conspecific densitydependentthinning in western hemlock populations.In this study, we explore the prevalence of densitydependence across multiple life stages in a temperate forestof northeast China. We use data on 5,762 seedlings of 34woody species to examine the effects of conspecific densityon the survival of established seedlings (early life-historystage). In addition, we use data on 15 abundant woodyspecies with dbh C 1 cm (later life-history stages) toexamine conspecific density-dependent thinning from saplingto juvenile stages using spatial point pattern analysis.For all life-history stages studied, we accounted for habitatheterogeneity that may mask density-dependent effects.We test the hypothesis that density dependence is prevalentboth in early and later life-history stages of trees. Specifically,for seedlings, we test the hypothesis that survival in theseedling bank declines with increasing local conspecificneighbor density and the effect of conspecific neighbors differsfrom that of heterospecifics. For larger trees, we test thehypothesis that the extent of aggregation declines from saplingto juvenile stages, indicating the existence of negativedensity dependence at later life-history stages.Materials and methodsStudy site and data collectionOur study site, termed Liangshui forest dynamics plot (FDP)is located in Liangshui national reserve (47°10 0 50 00 N,1232

Oecologia128°53 0 20 00 E) of northeastern China (Online Resource 1).The reserve is characterized by rolling mountainous terrainwith elevation ranging from 280 to 707 m. a.s.l. Meanannual temperature is -0.3 °C with mean daily maximumtemperature of 7.5 °C and minimum temperature of-6.6 °C. Mean annual surface soil temperature is 1.2 °Cwith 100–120 frost-free days. Mean annual precipitation is676 mm with 78 % relative humidity and an evaporationrate of 805 mm (Jin et al. 2006).The 9-ha (300 9 300 m) FDP was established in 2005in a typical mixed broadleaved-Korean pine (Pinus koraiensis)forest. All woody stems C2 cm dbh in the plot weremapped, measured, identified to species, and tagged in2005. In 2010, we recensused the plot and additionallyincluded woody stems 1–2 cm dbh. In the 2010 census, wedocumented 21,775 free-standing live individuals C1 cmdbh belonging to 18 families, 32 genera and 46 species(species identifications based on Chou et al. 1985). In thispaper, we used data on live trees C1 cm dbh from the 2010census. These trees were grouped by maximum attainableheight into five growth forms: shrubs (S; B5 m), smallunderstory tree species (US; 5 to B10 m), large understorytree species (UL; 10 to B20 m), small canopy tree species(CS; 20 to B30 m) and large canopy tree species(CL; [30 m). In turn, each growth form was divided intothree dbh size classes to define life-history stages: sapling,juvenile, and adult stages (Table 1). For analyses with treesC1 cm dbh, 15 species that had C40 individuals at each ofthese life stages were selected as focal species (OnlineResource 2). These species comprised 89 % of total stemsC1 cm dbh in the plot.In 2005, we established a permanently marked 4-m 2seedling plot in the northwest corner of each 10 9 10 msubplot of the 9-ha FDP. All free-standing, woody seedlingsand small saplings C30 cm tall and \1 cm dbh(hereafter referred to as seedlings) were tagged, measured,and identified to species within each seedling plot. Werecensused the 900 seedling plots in 2007, 2008, and 2010.To estimate light conditions above the seedling plots,hemispherical photographs were taken using a fisheye lens(Nikon FC-E8) mounted on a Nikon camera (Coolpix4500) at a height of 1.3 m over the center of each seedlingplot during August 2005. We analyzed the hemisphericalphotos using Hemiview canopy analysis software v.2.1(Delta-T De-vices, UK, 1999) to calculate the percentcanopy openness above each seedling plot. To investigateeffects of topography, the entire study plot was first dividedinto 3,600 contiguous 5 9 5 m quadrats, and then thetopographic position (ridge, upper slope, lower slope, andvalley) for each quadrat containing a seedling plot wasassessed visually by comparing the topography to surroundingquadrats.Data analysisWe tested for conspecific negative density dependence attwo life stages: the established seedling stage (early-stage)and the sapling-to-juvenile transition (later-stage). However,because species habitat preferences may obscureunderlying density-dependent processes, we first examinedevidence for effects of habitat heterogeneity.Test of habitat heterogeneitySpatial patterns of mature trees can be used as an indicatorof strong environmental habitat preferences, under theassumption that mature trees have undergone thinning overtime due to environmental filtering (Getzin et al. 2008).Although dispersal limitation can also leave a signature onthe spatial pattern of trees, mature individuals likely representthose who lived in sites most favorable for thespecies (Condit et al. 2000) and thus their spatial patternscan be used to detect underlying habitat heterogeneity. Totest for habitat preferences at our study site, we analyzedthe spatial pattern of adult trees for the 15 focal species,using the cumulative L-function (the transformed K-func-qffiffiffiffiffiffiffition, LðrÞ ¼KðrÞpr, where r is the variable radiussampled around each tree of each focal species; furthermethodological details are explained in ‘‘Later-stage conspecificdensity dependence’’, below) (Ripley 1976; Stoyanand Stoyan 1994; Illian et al. 2008) with the homogeneousPoisson process as a null model. Stoyan and Penttinen(2000) suggested that, in mature boreal forests, tree–treeinteractions are independent at scales [10 m, and, beyondthis scale, spatial patterns of trees are influenced byTable 1 Life stage classifications based on dbh (cm) for trees of different growth forms: shrubs (S), small understory tree species (US), largeunderstory tree species (UL), small canopy tree species (CS), and large canopy tree species (CL), respectivelyLife stage S US UL CS CLSapling 1.0-1.5 1.0-2.0 1.0-2.5 1.0-5.0 1.0-8.0Juvenile 1.6-2.0 2.1-4.0 2.6-6.0 5.1-10.0 8.1-15.0Adult [2.0 [4.0 [6.0 [10.0 [15.0Dbh cut-offs were selected to ensure adequate sample sizes for the spatial point pattern analysis1233

Oecologiaenvironmental factors. In this study, we also assumed thattree–tree interactions can be neglected beyond the scale of10 m, and consider aggregated patterns of adult trees atscales [10 m as a sign of habitat heterogeneity.Early-stage density dependenceTo test for density dependence at the seedling stage, weexamined the effect of neighbor density on seedling survivalusing generalized linear mixed-effects models(GLMMs) with binomial errors. We modeled the probabilityof an individual seedling surviving across the2005–2010 census intervals as a function of the density andidentity of seedling and tree (C1 cm dbh) neighbors. Localseedling densities were obtained by counting the number ofconspecific (S CON ) and heterospecific (S HET ) seedlings inthe same 4-m 2 quadrat as the focal seedling in 2005, andlocal densities of trees were calculated by summing up thebasal area (B) of all conspecific (B CON ) and heterospecific(B HET ) trees C1 cm dbh in the 2010 census within a radiusof 10 m. A radius of 10 m was selected since it yielded thelowest AIC value compared with 5-, 15-, and 20-m radii inpreliminary analyses using the full model (model 9 inTable 2). In addition to conspecific and heterospecificneighbor densities, initial seedling height (H) was includedas a fixed effect, since seedling size is usually significantlyand positively correlated with survival.We included seedling plot as a random effect in themodel to account for spatial autocorrelation in survival.Including a plot term should be sufficient, since all seedlingplots are spaced 10 m apart and spatial autocorrelation ofmodel residuals was found to be negligible beyond 10 m(Online Resource 3). We also included species as a randomeffect in the community-level models, in order to allow fordifferences among species in their baseline survival rates(i.e. the model intercept term). We excluded Sorbariasorbifolia and Spiraea salicifolia from all analyses, sincethey did not have stems of dbh C 1 cm in the study plot.To assess the role of conspecific and heterospecificneighbor densities on seedling survival, nine models wereconstructed according to Comita and Hubbell (2009)(Table 2). These nine models fall into three classes: (1) adensity-independent model, (2) models in which there is aneffect of overall seedling or tree neighbor densities, with nodifferentiation between conspecifics and heterospecifics,and (3) models in which the effect of conspecifics differsfrom heterospecifics for seedling or tree neighbors. Modelswere compared using Akaike’s information criterion (AIC;Burnham and Anderson 2003).We examined the effect of neighbors on seedling survivalat three levels. First, we examined seedling survivalon a species-by-species basis for 11 abundant species(n [ 99 seedlings). Second, we examined seedling survivalfor all species combined in the whole dataset. Third, weexcluded those species that showed density-dependentmortality in the first level analysis and examined patternsof seedling survival for the remaining species combined, inorder to determine whether community-wide results werebeing driven by a few species.To test whether species’ habitat preferences affected ourability to detect density dependence, we repeated the aboveanalyses with the abiotic variables of canopy openness andtopography position as covariates to control for habitatheterogeneity in each of the nine models. In the specieslevelanalyses, we included canopy openness and topographyposition as fixed effects. In the community-wideanalyses, because we expected species to vary in theirresponses to canopy openness and topography position, weincluded the two variables as random effects that variedamong species. We also ran the models using altitude (as acontinuous variable) instead of the topographic categories,but the results were qualitatively similar, so we onlyTable 2 Nine models compared to determine effects of conspecific and heterospecific neighbor densities on established seedling survival in theLiangshui FDPModel class Model Model structureDensity independent 1 a ? b 9 HEffect of conspecific density = effect of heterospecific density 2 a ? b 9 H ? c 9 S TOTAL3 a ? b 9 H ? c 9 S TOTAL ? d 9 B TOTAL4 a ? b 9 H ? d 9 B TOTALEffect of conspecific density = effect of heterospecific density 5 a ? b 9 H ? c 1 9 S CON ? c 2 9 S HET6 a ? b 9 H ? c 1 9 S CON ? c 2 9 S HET ? d 9 B TOTAL7 a ? b 9 H ? d 1 9 B CON ? d 2 9 B HET8 a ? b 9 H ? c 9 S TOTAL ? d 1 9 B CON ? d 2 9 B HET9 a ? b 9 H ? c 1 9 S CON ? c 2 9 S HET ? d 1 9 B CON ? d 2 9 B HETH seedling height; S CON , S HET and S TOTAL number of conspecific, heterospecific and overall seedling neighbors of the focal seedling in the 4-m 2seedling plot; B CON , B HET and B TOTAL basal area of conspecific, heterospecific and overall tree neighbors (C 1 cm dbh) within a radius of 10 m;a, b, c, c 1 , c 2 , d, d 1 and d 2 model coefficients1234

OecologiaFig. 1 Examples of conspecific density-dependent analysis usingPinus koraiensis a saplings, b juveniles as cases and c decline ofadditional aggregation from the sapling to juvenile stage. Themaximum strength of conspecific thinning (d max ) took place at thescale of 0 m. Results for the analysis of all 15 focal species:d saplings, e juveniles as cases and f number of focal species showingdensity dependence at each scale. In (d) and (e), solid circlespresent the results using topographic categories, since theybetter capture the microhabitat variation in the plot.GLMMs were fitted by the lmer() function of the ‘lme4’package in R 2.13.0 (R Development Core Team 2011)with the recommended Laplace method (Bates et al. 2008;Bolker et al. 2009).Later-stage conspecific density dependenceFor each of the 15 focal species selected for later-stagedensity dependence analysis, we applied the method ofrandom-labeling null model within a case–control design(Getzin et al. 2008) to estimate conspecific density-dependentthinning from the sapling to juvenile stage, utilizing thebivariate pair correlation g-function (Stoyan and Stoyan1994; Illian et al. 2008). We used saplings and juveniles ascases (pattern i) and adults as controls (pattern j) to accountfor habitat heterogeneity. The g-functions are invariantunder random thinning of trees, hence we would expectg ij ðrÞ ¼g ji ðrÞ ¼g ii ðrÞ ¼g jj ðrÞ, where r denotes distancescale. We used a i ðrÞ ¼g ij ðrÞ g ii ðrÞ as a test statistic to testwhether cases i show an additional pattern that is independentfrom the controls j. If a i (r) \ 0, cases can be said toexhibit additional aggregated patterns relative to adults,irrespective of whether habitat heterogeneity is present ornot (Getzin et al. 2006; Watson et al. 2007). The change inadditional aggregation from sapling to juvenile stages can beexpressed by the formula: dðrÞ ¼a juveniles ðrÞ a saplings ðrÞ(Zhu et al. 2010), where a juveniles (r) is the additional aggregationof juveniles relative to adults over the scales r, anda saplings (r) is the additional aggregation of saplings. For aparticular species, if a saplings (r) \ 0 and d(r) [ 0, we wouldrepresent the number of species with the test statistic g ij (r) -g ii (r) \ 0 (i.e. cases show additional aggregation relative to adults),open circles represent the number of species with g ij (r) - g ii (r) [ 0(i.e. cases are less aggregated than adults), and open squares representthe number of species with g ij (r) - g ii (r) = 0 (i.e. patterns for casesand adults are created by the same stochastic process)infer that conspecific density-dependent thinning takes placefrom cohorts of saplings to cohorts of juveniles for thatspecies. We focused on the scale of 0–10 m for the analysisabove, because we assume the effects from tree–tree interactionscould be efficiently indicated by this scale, andchanges in spatial pattern beyond a scale of 10 m could becaused by other environmental factors (i.e. large-scale habitatheterogeneity; Stoyan and Penttinen 2000). d max was themaximum strength of conspecific thinning, when d(r) takesthe maximal value at the scale of 0–30 m. We provide anexample to illustrate this part of the analysis using Pinuskoraiensis (Fig. 1a–c).To assess the importance of controlling for habitatpreference in the analysis of density dependence, we alsoadditionally randomized the locations of adult trees foreach species and used them as pattern j in place of theactual adult pattern which controls habitat preference andrepeated the analysis described above.All spatial point pattern analysis was done in the gridbasedsoftware Programita (Wiegand and Moloney 2004),using resolutions of a grid size of 1 m 2 and a ring width of3 m for analysis of tree–tree interactions and habitat heterogeneityat scales of 0–30 m. The resolutions wereselected based on the size of our 300 9 300 m plot and themeasurement uncertainty of point coordinates, and theyshould be sufficient to capture detailed variation in the paircorrelationfunction over the range of scales where weexpected significant effects (effects from tree–tree interactionsand habitat heterogeneity) up to 30 m (Wiegandand Moloney 2004; Zhu et al. 2010).For all spatial point pattern analysis, we performed 999Monte Carlo simulations of the null model and used the1235

Oecologiafifth-lowest and fifth-highest values (i.e., extreme 0.5 %simulated cases at either end) as simulation envelopes.However, because the simulation tests are performed atdifferent scales concurrently, this simulation inferenceyields an underestimated Type I error rate (Loosmore andFord 2006). We therefore combined this simulation envelopemethod with a goodness-of-fit test (GOF) (Diggle2003). Further analysis were only performed for those datasets where the observed GOF’s P \ 0.005 (Loosmore andFord 2006; Wiegand et al. 2007).ResultsTest of habitat heterogeneityExcept for Ulmus laciniata, all focal species showed significantaggregation up to 30 m (i.e. P \ 0.005 for GOFtests). Adults of 12 of the 15 focal species showed increasingaggregation at scales r [ 10 m (Online Resource 4). Thisindicated that most of the focal species exhibited habitatpreference caused by large-scale habitat heterogeneity,suggesting that we should account for habitat heterogeneityin our analysis of density dependence.Early-stage density dependenceFor a total of 5,762 seedlings, mortality was 29.8 % from2005–2010, thus averaging *6 % per year. Among the 11abundant focal species, percent seedling mortality between2005 and 2010 ranged from 11.0 to 55.4 % (mean =38.8 %).Before controlling for habitat preference, for 5 of the 11focal species, the best-fit model was the density-independentmodel, indicating that neither seedling nor tree neighborsinfluenced seedling survival. For 4 species, the best-fit modelincluded overall seedling density or basal area of trees, withno difference between the effects of conspecifics and heterospecifics.For the remaining 2 species, the best-fit modelincluded separate terms for conspecific and heterospecificneighbors. For Philadelphus schrenkii, the best-fit model(model 5) included conspecific and heterospecific seedlingdensities, but not tree basal area, and the effect of conspecificseedling density was significantly negative. For Deutziaglabrata, the best-fit model was the full model (model 9),which included separate terms for conspecific and heterospecificseedling and tree neighbors. The effect of conspecificneighbors was significantly negative for both seedling densityand basal area of trees C1 cm dbh. In contrast, the effectof heterospecific basal area of trees C1 cmdbhwassignificantlypositive (Online Resource 5). Thus, of the 11 focalspecies, only 2 showed patterns of seedling survival consistentwith conspecific negative density dependence.Nonetheless, in the community-wide analysis, wedetected significant conspecific negative density dependence.With all species combined, the probability ofseedling survival was best described by model 6, whichincluded separate conspecific and heterospecific seedlingterms and overall basal area of trees. The effect of conspecificseedling density was significantly negative. Theeffect of overall basal area of trees C1 cm dbh was significantlypositive (Table 3).However, after removing the two species that showednegative density dependence in the species-level analysis(Deutzia glabrata and Philadelphus schrenkii), community-wideseedling survival was best fitted by model 4,indicating that there was an effect of overall basal area oftrees C1 cm dbh, but no effect of seedling neighbors. Theeffect of overall basal area of trees C1 cm dbh remainedsignificantly positive (Table 3).Controlling for habitat heterogeneity did not qualitativelyalter the observed patterns of conspecific negativedensity dependence at the seedling stage. Including canopyopenness and topographic position as covariates in themodels did change the best-fit models for 6 of the 11 focalspecies (Online Resource 6 vs. 5). However, for all 6 ofthose species, the best-fit models did not include separateterms for conspecific and heterospecific neighbors,regardless of whether the model controlled for habitatheterogeneity. For the community-level analyses, the bestfitmodels did not change when including canopy opennessand topographic position as covariates, and the coefficientvalues for neighbor effects were similar compared withmodels that did not include these habitat variables(Table 3).Later-stage conspecific density dependenceWe calculated the number of species showing aggregated,random and regular patterns at the sapling and juvenilestage at each (1 m) scale up to 10 m. For saplings, 11 of 15species exhibited additional aggregation relative to adults(i.e. saplings were more clustered than adults), 7 speciesshowed random patterns (i.e. not significantly differentfrom the adults), and 1 species showed more regular patterns(i.e. saplings were less aggregated than the adults) atscales up to 10 m (Fig. 1d). For juveniles, 8 of 15 speciesexhibited additional aggregation relative to adults, 13species showed random patterns, and no species showedmore regular patterns up to 10 m (Fig. 1e).The 11 species (73 %) that exhibited additional aggregationrelative to adults in the sapling stage were all foundto have a decline in the strength of additional clusteringfrom the sapling to juvenile stage at the scale of 0–10 m,indicating that the majority of abundant species showedconspecific density dependence across the study area1236

OecologiaTable 3 Effects of local-scale seedling and adult neighbor densities on survival of established seedlings with and without controlling for habitatpreference in the 9-ha plot (see ‘‘Materials and methods’’)Dataset Whole dataset (BF = 6, n = 5,762, SP = 34) Subset (BF = 4, n = 3,987, SP = 32)Control for habitat preference No Yes No YesParameter values (standard error)H 0.283 (0.033) 0.282 (0.033) 0.255 (0.041) 0.257 (0.041)S CON 20.085 (0.038) 20.098 (0.038) – –S HET -0.014 (0.045) -0.012 (0.045) – –S TOTAL – – – –B CON – – – –B HET – – – –B TOTAL 0.106 (0.041) 0.098 (0.042) 0.090 (0.046) 0.084 (0.046)AICModel 1 6,779.2 6,790.7 4,718.7 4,732.4Model 2 6,780.9 6,792.3 4,720.6 4,734.4Model 3 6,775.9 6,787.5 4,718.5 4,732.9Model 4 6,775.1 6,787.5 4,716.9 4,731.3Model 5 6,778.7 6,789.3 4,722.4 4,736.0Model 6 6,774.2 6,785.3 4,720.3 4,734.6Model 7 6,777 6,789.4 4,718.9 4,733.3Model 8 6,777.8 6,790.1 4,720.5 4,734.9Model 9 6,776.2 6,787.7 4,722.3 4,736.6Subset is the dataset excluding the two species showing negative density dependence in the species-level analysis. Bold values denote significanteffects (P \ 0.05). AIC values are presented for each of the nine models compared (see Table 2)BF best-fit model, n number of seedlings used in the analysis, SP number of seedling species used in the analysisTable 4 Values of conspecific thinning, d(r), for the 11 species that exhibited thinning effects from the sapling to juvenile stage at the scale of0–10 m when habitat heterogeneity was factored outScale r (m) 0 1 2 3 4 5 6 7 8 9 10Abies nephrolepis 29.2 16.5 12.7 6.5 3.9 2.3 1.7 0.7 0.4 – –Acer mono 0.9 0.2 0.2 0.1 0.2 0.2 0.1 0.0 0.1 – 0.0Acer tegmentosum – – – 0.1 0.6 0.7 0.8 0.6 0.2 0.1 –Acer ukurunduense – – – – 0.2 0.3 – – – – –Betula costata 17.8 17.2 12.9 9.5 6.6 4.9 4.2 4.2 3.3 3.2 2.2Euonymus pauciflorus – 1.1 1.2 – – – – – – – –Fraxinus mandshurica 1.2 – 0.2 3.7 4.6 4.6 3.8 3.0 1.6 1.2 0.6Pinus koraiensis 58.1 36.4 27.6 17.5 11.6 10.5 10.6 10.2 8.0 6.9 5.3Tilia amurensis 1.3 – – – – – – – – – –Tilia mandshurica – – – – 8.4 11.9 11.4 6.9 – – –Ulmus laciniata 3.7 3.1 1.7 0.3 – – 0.1 – 0.0 0.2 0.4Bold values denote the maximum strength of conspecific thinning (d max ) at the scale of 0–30 mduring the sapling to juvenile transition (Table 4). For the 4species that did not exhibit additional aggregation relativeto adults in the sapling stage at the scale of 0–10 m(Acanthopanax senticosus, Corylus mandshurica, Philadelphusschrenkii and Syringa reticulata var. mandshurica),we cannot test whether they experienced densitydependence from the sapling to juvenile stage with ourmethods. However, since saplings were not more aggregatedthan adult trees for these species, it is unlikely thatconspecific density-dependent thinning occurred beyondthe sapling stage in these species.The number of species showing conspecific densitydependentthinning decreased with increasing spatial scale(Fig. 1f). Six of the 11 species that exhibited conspecific1237

Oecologiadensity-dependent thinning at the scale of 0–10 m reacheda maximum strength of thinning at the scale of 0 m (in a1 9 1 m grid cell) (Table 4). Furthermore, the largestradius at which maximum thinning occurred was only 6 m(for Acer tegmentosum; Table 4). The thinning intensityalso had a trend of decreasing with increasing scales formost species (Table 4). Together, these results imply thatconspecific density-dependent thinning occurred predominantlyat close distances among neighbors.The above analyses accounted for habitat heterogeneityby using mature tree distributions as controls. When habitatheterogeneity was not controlled for, we found an increase inthe number of species showing conspecific thinning at largerscales ([20 m) (Online Resource 7c). As a result, we did notsee a decrease in the number of species showing conspecificdensity-dependent thinning with increasing spatial scale, aswas found when controlling for habitat heterogeneity.Nonetheless, there was no difference in the overall numberof species (11 of 15 focal species) found to have a decline inthe strength of additional aggregation relative to the randomizedadult locations from sapling to juvenile stage at thescale of 0–10 m (Online Resource 8). In other words, thesame number of species exhibited conspecific densitydependentthinning at local (\10 m) scales regardless ofwhether habitat heterogeneity was controlled for; however,there were differences in the individual species exhibitingconspecific thinning. Tilia amurensis, which exhibitedconspecific density-dependent thinning from the sapling tojuvenile stage when habitat heterogeneity was factored out,did not show that trend when habitat heterogeneity wasunaccounted for. Philadelphus schrenkii, which did notshow conspecific density-dependent thinning when habitatheterogeneity was factored out, was found to exhibit conspecificdensity-dependent thinning when habitat heterogeneitywas not accounted for (Online Resource 8; Table 4). Inaddition, for species that showed significant conspecificthinning regardless of whether habitat heterogeneity wasaccounted for, there were often differences in the scale ofmaximum conspecific thinning. For example, Acer ukurunduenseand Tilia mandshurica reached their maximumstrength of conspecific thinning, d max , at scales of r [ 15 mwhen habitat heterogeneity was not accounted for, but bothreached d max at the scale of 5 m when habitat heterogeneitywas accounted for (Online Resource 8; Table 4).DiscussionDensity dependence has been hypothesized to be one of themost prominent mechanisms contributing to the maintenanceof diversity (Janzen 1970; Connell 1971; Hooper1998; Chesson 2000; Volkov et al. 2005). Though manystudies have examined how this mechanism operates, mosthave focused on a single life-history stage (e.g., Bell et al.2006; Diez 2007; Queenborough et al. 2009) and fewstudies have factored out the potentially confoundinginfluence of habitat heterogeneity (but see He and Duncan2000; Zhu et al. 2010; Chen et al. 2010). As far as weknow, this is the first study of density dependence in treesto include more than one life-history stage while controllingfor habitat heterogeneity. Our study of the impact ofthe biotic neighborhood on the established seedling andsapling to juvenile transition stages showed that conspecificneighbors tend to have a negative impact on survival.In addition, our analyses demonstrate the influence ofhabitat heterogeneity on analyses of density dependence.Habitat heterogeneityThe increase in establishment driven by favorable habitatmay offset the thinning of conspecific trees due to negativedensity dependence (Wright 2002). Our results demonstratethe importance of controlling for habitat heterogeneitywhen testing for density dependence in plantcommunities. For both early and later life stages, wecompared the results with and without controlling forhabitat heterogeneity. For the seedling stage, we found thatcontrolling for habitat heterogeneity did not qualitativelyalter our results of negative density dependence at thecommunity level or the proportion of focal species exhibitingnegative density-dependent seedling survival. At theseedling stage, survival may be more strongly affected byconspecific density than abiotic conditions. However, ourmodels only included canopy openness and topographicposition to control for possible habitat preferences.Therefore, we cannot rule out the possibility that unmeasuredhabitat variables, such as soil nutrients and temperature,may be masking patterns of density dependence. Forthe sapling to juvenile stage, the same number of speciesshowed density dependence when controlling and notcontrolling for habitat heterogeneity. However, severalspecies showed differing patterns in terms of the significanceor scale of density dependence when factoring outhabitat heterogeneity. When habitat heterogeneity was notaccounted for, conspecific thinning tended to be detected atlarger spatial scales, likely driven by unfavorable habitatand not tree–tree interactions. Previous studies have alsopointed out the confounding effects of habitat heterogeneityon density dependence analysis. For example, He andDuncan (2000) tested intra- and interspecific densitydependenteffects on survival of three species in an oldgrowthDouglas fir (Pseudotsuga menziesii) forest, whereelevation gradients control local variation in site conditions.They found that, after controlling for elevation, theprobability of western hemlock (Tsuga heterophylla) survivalwas no longer significantly higher in less dense1238

Oecologiapatches of Douglas fir. Zhu et al. (2010), in a study of 47abundant species in a subtropical forest, also used therandom labeling method and found that the number ofspecies showing density dependence was different whenfactoring out habitat heterogeneity. In a study of densitydependentseedling survival in a subtropical forest, Chenet al. (2010) found that habitat heterogeneity explainedmore variation among species in seedling survival thanspecies abundance, and concluded that tests for community-levelconsequences of density dependence mustaccount for habitat heterogeneity. Those findings, togetherwith our study, indicate that failing to consider habitatheterogeneity could lead to incorrect inferences aboutdensity-dependent effects, and therefore controlling forhabitat heterogeneity is necessary in studies of densitydependence.Density dependence across multiple life stagesIn this study, we analyzed density dependence for bothearlier (the established seedling stage) and later life-historystages (the sapling to juvenile transition) of tree species inthe Lianghsui FDP. For seedlings, we found a significantnegative effect of local conspecific seedling density onsurvival when analyzing all species in the communitytogether, consistent with predictions of the Janzen–Connellhypothesis. However, in separate species-level analyses,we detected negative effects of conspecific seedlingneighbors on focal seedling survival for only two species,Deutzia glabrata (which accounted for 22.1 % of totalseedlings) and Philadelphus schrenkii (which accountedfor 8.7 % of total seedlings). This suggests that our communitylevel finding of density dependence was likelydriven by these two abundant species. Indeed, when weremoved these species from the dataset, we did not detectsignificant conspecific density dependence. Thus, ourresults emphasize how community-level analyses canconceal variation among species in density dependence.This is consistent with recent studies that have found widevariation among species in the strength of density dependence(Comita et al. 2010; Kobe and Vriesendorp 2011).Relatively few studies of density dependence have beenconducted for saplings and larger trees at the communitylevel (Carson et al. 2008). In this study, we examineddensity dependence across the sapling-to-juvenile transition.At this later stage, we found that 11 of 15 focalspecies showed conspecific density-dependent thinning.For those 11 species, we found that the thinning occurredpredominantly at very small scales: the maximum strengthof thinning occurred only up to 6 m, and 6 species reacheda maximum strength of thinning at the scale of\1 m.Thismay due to the decline in aggregation of saplings withincreasing distance from parent trees, likely caused bydispersal limitation (Hubbell and Foster 1983; He et al.1997). The observed local-scale conspecific thinning couldhave resulted from strong intraspecific competition forresources or host-specific natural enemy attack leading todensity-dependent mortality. At the smallest spatial scales,we also cannot rule out the possibility that thinning wascaused by physical space constraints.Studies of density dependence at later life stages in otherforests have similarly found that the majority of speciestested exhibit significant conspecific density dependence(e.g., Wills et al. 1997; Peters 2003; Stoll and Newbery2005; Zhang et al. 2009). For example, Wills et al. (1997)found that for 67 of 84 focal species in Panama, recruitmentof C1 cm dbh saplings was negatively correlated withconspecific basal area in at least one quadrat size, andintraspecific effects were stronger than interspecific effects.Similarly, in a study of neighbor effects on saplings andtrees (dbh C 1 cm), Peters (2003) found evidence of density-dependentmortality for saplings and trees of[75 % ofthe species tested at sites in Pasoh, Malaysia, and BCI,Panama. These studies, together with the results presentedhere, show that density-dependent effects can be prevalentat later life stages and should not be ignored.Nonetheless, density dependence is often found to bemore prevalent at earlier compared to later life stages intree communities (e.g., Hille Ris Lambers et al. 2002;Comita and Hubbell 2009). This pattern may occur ifdensity dependence during earlier stages is sufficientlystrong to thin out conspecifics to levels below which negativeeffects of density are not detectable at later stages.However, we found significant negative conspecific densityeffects for only 2 of the 11 abundant species in our analysisof density-dependent survival at the established seedlingstage, but significant conspecific density-dependent thinningfrom the sapling to juvenile stage for 11 of the 15focal species, including 4 of the species that did not showdensity dependence effects at the seedling stage. Thus, ourresults do not support the idea that density dependence isstronger at earlier stages. However, the higher incidence ofdensity dependence at later stages in our study may reflectthe different methods used to analyze density dependenceat earlier and later life stages. We examined densitydependence at later life-history stages using spatial patternanalysis, which takes advantage of the accumulated effectsof density-dependent mortality over time as individuals ofa species move from one size class to another. In contrast,for seedlings, we used direct observations of seedlingmortality over 5 years, which may not have been sufficientto detect density dependence for some species.The prevalence of density-dependent mortality found inour study has important implications beyond our understandingof the forces that structure this particular temperateforest. Janzen–Connell effects were assumed to be1239

Oecologiastronger in tropical than temperate forests because ofhigher numbers of specialist predators and pathogens in theformer (Janzen 1970; Connell 1971; Coley and Barone1996; Givnish 1999; Harms et al. 2000; Dyer et al. 2007),and the latitudinal gradient in tree diversity has beenhypothesized to be caused by decreasing Janzen–Connelleffects with increasing latitude. However, studies havefound that density-dependent mortality is as common intemperate forests as in tropical forests. For example, HilleRis Lambers et al. (2002) found the proportion of speciesaffected by negative density dependence at the seed andseedling stages in a North American temperate forest wasequivalent to that reported for tropical forests, though theyacknowledged that the magnitude of density-dependenteffects may be higher in tropical forests. Our study ofdensity dependence in Liangshui FDP also supports theidea that density dependence may be as important intemperate forests as in tropical forests, and densitydependence alone is unlikely to explain latitudinal treediversity differences in the world’s forests (but see Johnsonet al. 2012).ConclusionIn combination with previous studies from tropical andsubtropical forests, our results from an old-growth temperateforest indicate that significant negative effects oflocal conspecific density occur at multiple life stages intree communities. This suggests that density dependenceplays an important role in enhancing community diversityof forests in different latitudes. Although we found negativeconspecific effects at multiple stages, species exhibitingnegative density dependence at one life stage did notnecessarily exhibit it at other stages. Thus, analyses thatfocus on a single life stage may underestimate the prevalenceand importance of density dependence in tree communities.Similarly, our results suggest that studies that failto take into account confounding factors, such as habitatheterogeneity and species-level variation, may also mischaracterizethe role of density dependence in shapingplant communities. Therefore, we recommend that futurestudies take habitat heterogeneity and other potentiallyconfounding factors into account and also test for effects ofconspecific neighbors across all life stages.Acknowledgments This study was financially supported by theNational Natural Science Foundation of China (No. 30770350,31270473), and the Natural Science Foundation of HeilongjiangProvince, China (No. ZJN0706). Many critical comments on themanuscript were received in the 2011 Smithsonian CTFS/SIGEO andCForBio Workshop in Changbaishan (US NSF grant DEB-1046113).We are grateful to Keping Ma and Zhanqing Hao for their efforts inorganizing that workshop, and the technological support fromSmithsonian Tropical Research Institute. We also thank ThorstenWiegand, Stephan Getzin and Yan Zhu for providing useful suggestions.The conduction of this study complied with the current laws ofChina.Open Access This article is distributed under the terms of theCreative Commons Attribution License which permits any use, distribution,and reproduction in any medium, provided the originalauthor(s) and the source are credited.ReferencesBarot S, Gignoux J, Menaut JC (1999) Demography of a savannapalm tree: predictions from comprehensive spatial patternsanalysis. Ecology 80:1987–2000Bates D, Maechler M, Dai B (2008) lme4: linear mixed-effectsmodels using S4 classes. R package version 0.999375-28.Available at: http://lme4.r-forge.r-project.org/Bell T, Freckleton RP, Lewis OT (2006) Plant pathogens drivedensity-dependent seedling mortality in a tropical tree. Ecol Lett9:569–574Bolker BM, Brooks ME, Clark CJ, Geange SW, Poulsen JR, StevensMHH, White JSS (2009) Generalized linear mixed models: apractical guide for ecology and evolution. Trends Ecol Evol24:127–135Burnham KP, Anderson DR (2003) Model selection and multimodelinference: a practical information-theoretic approach, 2nd edn.Springer, New YorkCarson WP, Anderson JT, Egbert G, Leigh J, Schnitzer SA (2008)Challenges associated with testing and falsifying the Janzen-Connell hypothesis: a review and critique. In: Carson WP,Schnitzer SA, STR I (eds) Tropical forest community ecology.Blackwell, Oxford, pp 210–241Cázares-Martínez J, Montaña C, Franco M (2010) The role of pollenlimitation on the coexistence of two dioecious, wind-pollinated,closely related shrubs in a fluctuating environment. Oecologia164:679–687Chen L, Mi XC, Comita LS, Zhang LW, Ren HB, Ma KP (2010)Community-level consequences of density dependence andhabitat heterogeneity in a subtropical broad-leaved forest. EcolLett 13:695–704Chesson P (2000) Mechanisms of maintenance of species diversity.Annu Rev Ecol Evol Syst 31:343–366Chou YL, Tung SL, Nie SQ (1985) Ligneous Flora of Heilongjiang(in Chinese). Heilongjiang Science and Technology Press,HarbinColey PD, Barone JA (1996) Herbivory and plant defenses in tropicalforests. Annu Rev Ecol Syst 27:305–335Comita LS, Hubbell SP (2009) Local neighborhood and species’shade tolerance influence survival in a diverse seedling bank.Ecology 90:328–334Comita LS, Muller-Landau HC, Aguilar S, Hubbell SP (2010)Asymmetric density dependence shapes species abundances in atropical tree community. Science 329:330–332Condit R, Ashton PS, Baker P, Bunyavejchewin S, Gunatilleke S,Gunatilleke N, Hubbell SP, Foster RB, Itoh A, LaFrankie JV,Lee HS, Losos E, Manokaran N, Sukumar R, Yamakura T(2000) Spatial patterns in the distribution of tropical tree species.Science 288:1414–1418Connell JH (1971) On the role of natural enemies in preventingcompetitive exclusion in some marine animals and in rain foresttrees. In: Den Boer PJ, Gradwell GR (eds) Dynamics ofpopulations. PUDOC, Wageningen, pp 298–31212310

OecologiaDiez JM (2007) Hierarchical patterns of symbiotic orchid germinationlinked to adult proximity and environmental gradients. J Ecol95:159–170Diggle P (2003) Statistical analysis of spatial point patterns. Arnold,LondonDyer LA, Singer MS, Lill JT, Stireman JO, Gentry GL, Marquis RJ,Ricklefs RE, Greeney HF, Wagner DL, Morais HC, Diniz IR,Kursar TA, Coley PD (2007) Host specificity of Lepidoptera intropical and temperate forests. Nature 448:696–699Getzin S, Dean C, He F, Trofymow JA, Wiegand K, Wiegand T(2006) Spatial patterns and competition of tree species in aDouglas-fir chronosequence on Vancouver Island. Ecography29:671–682Getzin S, Wiegand T, Wiegand K, He FL (2008) Heterogeneityinfluences spatial patterns and demographics in forest stands.J Ecol 96:807–820Givnish TJ (1999) On the causes of gradients in tropical tree diversity.J Ecol 87:193–210Harms KE, Wright SJ, Calderon O, Hernandez A, Herre EA (2000)Pervasive density-dependent recruitment enhances seedlingdiversity in a tropical forest. Nature 404:493–495He FL, Duncan RP (2000) Density-dependent effects on tree survivalin an old-growth Douglas-fir forest. J Ecol 88:676–688He FL, Legendre P, LaFrankie JV (1997) Distribution patterns of treespecies in a malaysian tropical rain forest. J Veg Sci 8:105–114Hille Ris Lambers J, Clark JS, Beckage B (2002) Density-dependentmortality and the latitudinal gradient in species diversity. Nature417:732–735Hooper DU (1998) The role of complementarity and competition inecosystem responses to variation in plant diversity. Ecology79:704–719Hubbell SP, Foster RB (1983) Diversity of canopy trees in aneotropical forest and implications for the conservation oftropical trees. In: Sutton SJ, Whitmore TC, Chadwick AC (eds)Tropical rainforest ecology and management. Blackwell,Oxford, pp. 314–329Hubbell SP, Foster RB, O’Brien ST, Harms KE, Condit R, WeschlerB, Wright SJ, Loo de Lao S (1999) Light gap disturbances,recruitment limitation, and tree diversity in a neotropical forest.Science 283:554–557Illian J, Penttinen A, Stoyan H, Stoyan D (2008) Statistical analysisand modelling of spatial point patterns. Wiley, LondonJanzen DH (1970) Herbivores and number of tree species in tropicalforests. Am Nat 104:501–528Jin G, Xie X, Tian Y, Kim JH (2006) The pattern of seed rain in thebroadleaved-Korean pine mixed forest of Xiaoxing’an Mountains,China. J Korean For Soc 95:621–627Johnson DJ, Beaulieu WT, Bever JD, Clay K (2012) Conspecificnegative density dependence and forest diversity. Science336:904–907Kobe RK, Vriesendorp CF (2011) Conspecific density dependence inseedlings varies with species shade tolerance in a wet tropicalforest. Ecol Lett 14:503–510Loosmore NB, Ford ED (2006) Statistical inference using the G or Kpoint pattern spatial statistics. Ecology 87:1925–1931Metz MR, Sousa WP, Valencia R (2010) Widespread densitydependentseedling mortality promotes species coexistence in ahighly diverse Amazonian rain forest. Ecology 91:3675–3685Murrell DJ (2009) On the emergent spatial structure of size-structuredpopulations: when does self-thinning lead to a reduction inclustering? J Ecol 97:256–266Peters HA (2003) Neighbour-regulated mortality: the influence ofpositive and negative density dependence on tree populations inspecies-rich tropical forests. Ecol Lett 6:757–765Pigot AL, Leather SR (2008) Invertebrate predators drive distancedependentpatterns of seedling mortality in a temperate tree Acerpseudoplatanus. Oikos 117:521–530Queenborough SA, Burslem DFRP, Garwood NC, Valencia R (2007)Neighborhood and community interactions determine the spatialpattern of tropical tree seedling survival. Ecology 88:2248–2258Queenborough SA, Burslem DFRP, Garwood NC, Valencia R (2009)Taxonomic scale-dependence of habitat niche partitioning andbiotic neighbourhood on survival of tropical tree seedlings. ProcR Soc Lond B 276:4197–4205R Development Core Team (2011). R: a language and environmentfor statistical computing. R foundation for statistical computing,Vienna. ISBN 3-900051-07-0. Available at http://www.R-project.org. Accessed on 13 Apr 2011Ratikainen I, Gill JA, Gunnarsson TG, Sutherland WJ, Kokko H(2008) When density dependence is not instantaneous: theoreticaldevelopments and management implications. Ecol Lett11:184–198Ripley BD (1976) The second-order analysis of stationary pointprocesses. J Appl Prob 13:255–266Sterner RW, Ribic CA, Schatz GE (1986) Testing for life historicalchanges in spatial patterns of four tropical tree species. J Ecol74:621–633Stoll P, Newbery DM (2005) Evidence of species-specific neighborhoodeffects in the Dipterocarpaceae of a Bornean rain forest.Ecology 86:3048–3062Stoyan D, Penttinen A (2000) Recent applications of point processmethods in forestry statistics. Stat Sci 15:61–78Stoyan D, Stoyan H (1994) Fractals, random shapes, and point fields:methods of geometrical statistics. Wiley, ChichesterVolkov I, Banavar JR, He FL, Hubbell SP, Maritan A (2005) Densitydependence explains tree species abundance and diversity intropical forests. Nature 438:658–661Watson DM, Roshier DA, Wiegand T (2007) Spatial ecology of aparasitic shrub: patterns and predictions. Austral Ecol 32:359–369Webb CO, Peart DR (1999) Seedling density dependence promotescoexistence of Bornean rain forest. Ecology 80:2006–2017Wiegand T, Moloney KA (2004) Rings, circles, and null-models forpoint pattern analysis in ecology. Oikos 104:209–229Wiegand T, Gunatilleke S, Gunatilleke N (2007) Species associationsin a heterogeneous Srilankan dipterocarp forest. Am Nat 170:E77–E95Wills C, Condit R (1999) Similar non-random processes maintaindiversity in two tropical forests. Proc Biol Sci 266:1445–1552Wills C, Condit R, Foster RB, Hubbell SP (1997) Strong density- anddiversity-related effects help to maintain tree species diversity ina neotropical forest. Proc Natl Acad Sci USA 94:1252–1257Wright SJ (2002) Plant diversity in tropical forests: a review ofmechanisms of species coexistence. Oecologia 130:1–14Yamazaki M, Iwamoto S, Seiwa K (2009) Distance- and densitydependentseedling morality caused by several diseases in eighttree species co-occurring in a temperate forest. Plant Ecol201:181–196Zhang J, Hao ZQ, Sun IF, Song B, Ye J, Li BH, Wang XG (2009)Density dependence on tree survival in an old-growth temperateforest in northeastern China. Ann For Sci 66:1–9Zhu Y, Mi XC, Ren HB, Ma KP (2010) Density dependence isprevalent in a heterogeneous subtropical forest. Oikos 119:109–11912311

Estimate of Leaf Area Index in an Old-Growth MixedBroadleaved-Korean Pine Forest in Northeastern ChinaZhili Liu, Guangze Jin*, Yujiao QiCenter for Ecological Research, Northeast Forestry University, Harbin, ChinaAbstractLeaf area index (LAI) is an important variable in the study of forest ecosystem processes, but very few studies are designedto monitor LAI and the seasonal variability in a mixed forest using non-destructive sampling. In this study, first, true LAI fromMay 1 st and November 15 th was estimated by making several calibrations to LAI as measured from the WinSCANOPY 2006Plant Canopy Analyzer. These calibrations include a foliage element (shoot, that is considered to be a collection of needles)clumping index measured directly from the optical instrument, TRAC (Tracing Radiation and Architecture of Canopies); aneedle-to-shoot area ratio obtained from shoot samples; and a woody-to-total area ratio. Second, by periodically combiningtrue LAI (May 1 st ) with the seasonality of LAI for deciduous and coniferous species throughout the leaf-expansion season(from May to August), we estimated LAI of each investigation period in the leaf-expansion season. Third, by combining trueLAI (November 15 th ) with litter trap data (both deciduous and coniferous species), we estimated LAI of each investigationperiod during the leaf-fall season (from September to mid-November). Finally, LAI for the entire canopy then was derivedfrom the initial leaf expansion to the leaf fall. The results showed that LAI reached its peak with a value of 6.53 m 2 m 22 (acorresponding value of 3.83 m 2 m 22 from optical instrument) in early August, and the mean LAI was 4.97 m 2 m 22 fromMay to November using the proposed method. The optical instrument method underestimated LAI by an average of41.64% (SD = 6.54) throughout the whole study period compared to that estimated by the proposed method. The result ofthe present work implied that our method would be suitable for measuring LAI, for detecting the seasonality of LAI in amixed forest, and for measuring LAI seasonality for each species.Citation: Liu Z, Jin G, Qi Y (2012) Estimate of Leaf Area Index in an Old-Growth Mixed Broadleaved-Korean Pine Forest in Northeastern China. PLoS ONE 7(3):e32155. doi:10.1371/journal.pone.0032155Editor: Ben Bond-Lamberty, DOE Pacific Northwest National Laboratory, United States of AmericaReceived October 12, 2011; Accepted January 19, 2012; Published March 9, 2012Copyright: ß 2012 Liu et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricteduse, distribution, and reproduction in any medium, provided the original author and source are credited.Funding: This work was financially supported by the Forestry Science and Technology Supporting Program of China (No. 2011BAD37B01)(http://www.most.gov.cn/), the National Natural Science Foundation of China (No. 30770350)(http://www.nsfc.gov.cn), and by the Northeast Forestry University Graduate Thesis FundedProjects (No. GRAM09)(http://gra.nefu.edu.cn/xwgg.asp?page = 1). The funders had no role in study design, data collection and analysis, decision to publish, orpreparation of the manuscript.Competing Interests: The authors have declared that no competing interests exist.* E-mail: taxus@126.comIntroductionLeaf area index (LAI) is one of the most importantcharacteristics of plant canopy structure and has attracted manyscholars’ attention [1]. LAI is defined as half the total green leafarea per unit ground surface area (m 2 m 22 ) [2], and it directlyinfluences both the amount of solar radiation that can beintercepted and the plant-atmosphere exchange of CO 2 , O 2 ,water and energy [3–5]. LAI is required as an input variable inmost ecosystem models simulating carbon and water cycles [6],and it often serves as a convenient surrogate measure of grossprimary productivity (GPP) [7]. In addition, an accuratemeasurement of LAI is essential for converting leaf-level processesto the canopy level [8].Traditional measurements of LAI are generally divided intodirect and indirect methods [9–12]. Direct methods includedestructive sampling, allometry, and litter traps [13]. Destructivesampling method is the best method to obtain LAI, but it is notsuitable for measuring LAI in a large area and for dynamicmonitoring because it is destructive and time-consuming, and wecannot repeatedly destroy sample forest stands [14]. An allometricapproach can replace destructive sampling, but it remains difficultto monitor seasonal changes [15,16]. Litter traps method isnon-destructive, and collecting leaf litter to determine LAI is veryaccurate [17,18]. However, litter traps method is more successfulin deciduous forests that have a single leaf-fall season than inevergreen or mixed forests that have more continuous leaf loss andreplacement, and it provides little information about LAI duringthe leaf-expansion season. As a result, litter traps method shouldnot be used to monitor LAI seasonality in a mixed forest stand[15,19].The indirect ground method (optical instruments) infers LAI bymeasuring radiation transmission through the canopy [20]. Themain instruments include Tracing Radiation and Architecture ofCanopies (TRAC), LAI-2000 Plant Canopy Analyzer, Sunfleckceptometer, DEMON and Hemispherical photography [10,21].For a mixed forest, it is possible to overcome some problems withdirect methods by using this method. However, LAI is calculatedby most optical instruments with the assumption that leaves have arandom spatial distribution, and it is difficult to distinguish foliagefrom woody tissue. Thus, we use the term ‘‘effective LAI (L e )’’ todescribe LAI derived optically [22]. To find true LAI, L e fromoptical instruments must be calibrated properly (including woodytissue and the clumping effect).In this study, we propose a practical field measurement methodfor LAI in the canopy of a mixed forest using non-destructivePLoS ONE | www.plosone.org 1 March 2012 | Volume 7 | Issue 3 | e3215512

Estimate of Leaf Area Indexsampling. We implemented this method to obtain LAI for theentire canopy from the initial leaf expansion to the leaf fall in anold-growth mixed broadleaved-Korean pine forest in northeastChina. Based on the results from the proposed method, theaccuracy of the conventional indirect optical method (Hemisphericalphotography) for measuring LAI was investigated.Materials and MethodsEthics StatementThe field studies was conducted at the Liangshui NationalNature Reserve (47u109500N, 128u539200E), which is located onthe south side of the Xiaoxing’an Mountains, in northeasternChina. It is a practice base for the researchers (including studentsand teachers) of the Northeast Forestry University. Thus we couldconduct experiments there without specific permits. The experimentsconducted in this study do not involve or impactendangered or protected species.2.1 Study siteThe study site is an old-growth, mixed broadleaved-Koreanpine forest in the Liangshui National Nature Reserve, which wasestablished in 1980, and joined China’s Man and the BiosphereReserve Network in September 1997. It was promoted to anational nature reserve with the approval of the Chinese StateCouncil in December 1997 to protect the old growth mixedbroadleaved-Korean pine forest ecosystem. The topography iscomplex, and the highest mountain elevation is 707.3 m. Theannual mean air temperature and the mean annual rainfall are20.3uC and 676 mm, respectively. The site is covered by snow for130–150 days, and the frost-free period is 100–120 days. Thezonal vegetation is mixed broadleaved-Korean pine forest. Thespecies composition of the tree canopy at the study site are asfollows: the needle tree group (evergreen species) consists of Pinuskoraiensis, Abies nephrolepis, and Picea spp., with a mean diameter atbreast height (DBH) of 27.58 cm and a density of 313 trees ha 21 ;the broad-leaved species (deciduous species) mainly include Acermono, Betula costata, Fraxinus mandshurica, Tilia amurensis, Ulmuslaciniata, Acer tegmentosum, Acer ukurunduense, Ulmus japonica, Tiliamandshurica, Corylus mandshurica, and Prunus padus, with a meanDBH of 8.36 cm and a density of 1391 trees ha 21 .The study was conducted in a permanent sampling plot of anarea of 9 ha (300 m6300 m), divided into 900 sub-plots,10 m610 m each. We measured DBH and tree height andrecorded the coordinates of all plants with DBH$2 cm in eachsub-plot. Aluminum tree brands with tree numbers were nailed1.4 m above the root; however, brands were fixed by copper wirefor plants with DBH,8 cm to reduce the influence on plantgrowth [23]. At the center (160 m6160 m) of the permanentsampling plot, litter traps and hemispherical photography wereperformed at the same points, on a 868 grid (64 total samplepoints) with each point separated by 20 m.LAI observations were carried out from early May to November15, 2007. For details, the observations for collecting litter andtaking hemispherical photographs were all made on May 1 st , May14 th , May 22 nd , May 30 th , June 5 th , June14 th , July 1 st , July 15 th ,August 1 st , September 1 st , September 15 th , October 1 st , October15 th , November 1 st and November 15 th . The leaf seasonalityobservations were made from May to July (contemporaneous withcollecting the litter or taking hemispherical photographs).2.2 Optical leaf area indexIn our study, hemispherical photographs of the sample pointswere taken 1.3 m above ground using a WinSCANOPY 2006Plant Canopy Analyzer (contains a digital camera (Coolpix 4500,Nikon, Tokyo, Japan), a 180u fisheye lens (Nikon FC-E8), selflevelinggimbals and a tripod). The photographs were taken withautomatic exposure under uniform sky conditions, such as shortlybefore sunset or sunrise or when it was evenly overcast. Weestimated L e using hemispherical photography from May 1 st toNovember 15 th , which has been widely utilized to measure canopystructures [24–26], using the software DHP (Digital HemisphericalPhotography) [27]. To avoid missing small gaps in DHP at largezenith angles, L e was calculated using a zenith angle range of 0–60u, although the hemispherical photographs cover the range of 0–90u.2.3 The proposed methodOur method combines three components. First, L e of May 1 stand November 15 th are calibrated using the three components (thewoody-to-total area ratio, a; the element clumping index, V E ; theneedle-to-shoot area ratio, c E ), and then they represent true LAI ofMay 1 st and November 15 th , respectively. Second, by periodicallycombining true LAI (May 1 st ) with the seasonality of LAI fordeciduous and coniferous species throughout the leaf-expansionseason (from May to August), we can estimate LAI of eachinvestigation period in the leaf-expansion season. Third, bycombining true LAI (November 15 th ) with litter trap data (bothdeciduous and coniferous species), we can estimate LAI of eachinvestigation period during the leaf-fall season (from September tomid-November). LAI for the entire canopy then can be derivedaccordingly from the initial leaf expansion to the leaf fall.2.3.1 From L e to true LAI for May 1 st and November15 th . Most optical instruments measuring LAI assume aspatially-random distribution of foliage elements [22]. In thiscase, L e can be calculated from the gap fraction by adoptingMiller’s theorem [28], summarized in the following equation:L e ~2ð p=201ln½ Š cosh sin hdhPðhÞ ð1Þwhere P(h) is the gap fraction at the zenith view angle. However,most leaves in plant canopies are not randomly distributed inspace. Their distribution is in close relation to the distribution oftree crowns and branches in forests [29], especially for conifers, theclumping effect beyond and within the shoots and woody tissuemust be taken into account. Based on the development andvalidation of this theory [22,29], true LAI (L) is calculated usingthe following equation:cL~ ð1{aÞL EeV Ewhere a is the woody-to-total area ratio, c E is the needle-to-shootarea ratio quantifying the effect of foliage clumping within theshoots, V E is the element clumping index quantifying the effect offoliage clumping at scales larger than the shoots, and L e is effectiveleaf area index that directly obtained from optical instruments.Thus we could obtain true LAI of May 1 st and November 15 thusing equation (2), with the effects of the broadleaves ignored(because there are almost no leaves for deciduous species at thosetimes).The woody tissue was differentiated from the greenery usingPhotoshop software. The clumping effect included clumpingbeyond and within the shoots, and the former was quantified bya clumping index directly obtained from DHP and TRAC,whereas the latter was quantified by a needle-to-shoot area ratioð2ÞPLoS ONE | www.plosone.org 2 March 2012 | Volume 7 | Issue 3 | e3215513

Estimate of Leaf Area Indexobtained through laboratory analysis of shoot samples followingthe method described in Chen [22].Measurement of woody-to-total area ratio. The woody-tototalarea ratio (a) equals the woody area index (WAI) divided bythe plant area index (PAI). Traditional measurements of the ratioare generally divided into direct and indirect methods [9,15,30].The direct destructive sampling method usually involvesmeasuring the woody area of representative trees within foreststands, similar to the procedure used for direct measurements ofLAI. However, the direct method was impossible because it wastime-consuming and labor-intensive, and particularly, logging isprohibited in national nature reserves. Indirect methods (e.g., LAI-2000 or Hemispherical photography) are convenient and efficient,taking measurements during the leafless periods, but there are noleafless periods in a mixed forest stand, thus green and non-greenmaterials cannot be separated optically. In this study, almost nobroad leaves existed on May 1 st and November 15 th , so theinfluence on the deciduous species could be ignored. We removedthe errors from woody tissue with Photoshop 7.0 software. Whenprocessing the hemispherical photographs from May 1 st andNovember 15 th , first, we obtained the total LAI (L total ) ofphotograph using DHP software; second, a generic badge toolwas used to replace the stems and branches (both deciduous andevergreen) with nearby background parts, then we obtained LAI ofgreen materials (L green ) of photograph using DHP software withthe same threshold as above; third, the woody-to-total area ratiowas then derived accordingly (a =(L total 2L green )/L total ). Then, themean values of May 1 st and November 15 th were 0.096 and 0.092,respectively.Observation of clumping index. The effect of foliageclumping beyond the shoots is considered using a clumpingindex (V E ) because most of the branches and leaves of plantcanopies are not randomly distributed. With the development oftechnology, the method developed by Chen and Cihlar [31](briefly CC method) has been widely used to calibrate theclumping effect beyond the shoots. The final equation based onthe gap size distribution theory for calculating V E is:V E (h)~ In½Fm(0,h)Š ½1{Fmr(0,h)ŠIn½Fmr(0,h)Š ½1{Fm(0,h)Šwhere F m (0,h) is the total canopy gap fraction at zenith angleè, i.e.the accumulated gap fraction from the largest to smallest gaps; andF mr (0,è) is the total canopy gap fraction after removing large gapsresulting from the non-random foliage element distribution due tocanopy structures such as tree crowns and branches.The clumping index (V E ) for calibrating the clumping effectbeyond the shoots can be estimated using DHP and TRACWin3.9.1 software, and this method has been validated [15]. Then, thehemispherical photographic images from May 1 st and November 15 thwere calibrated using a clumping index. The clumping index from theDHP was computed within the zenith angle range of 40–45u.Measurement of needle-to-shoot area ratio. The needleto-shootarea ratio (c E ) was used to quantify the clumping effectwithin the shoots [29]. The c E was measured according to Chen’smethod [22]. The clumping effect within the shoots in conifers wasmostly determined by the tree’s growth condition, so we tried tosample within different conditions. To obtain an average c E valuefor the stand, P. koraiensis at the study site were first grouped intothree categories by their DBH: dominant (D), co-dominant (M)and suppressed (S), and three trees were selected from each class.From each tree, shoot samples (a shoot was the sampling unit)were taken at three heights: top (T), middle (M) and low (L), andnine shoots were sampled from each class, thus creating nine shootð3Þclasses: DT, DM, DL, MT, MM, ML, ST, SM and SL, for thestand. Therefore, we obtained a total of 243 shoot samples, whichwere taken back to the laboratory for further analysis. Based onthe theoretical development by Chen [22], the needle-to-shootarea ratio (c E ) is calculated as follows:c E ~A n =A swhere A n is half the total needle area (including all sides) in a shootand A s is half the shoot area. A n was obtained by the volumedisplacement method described by Chen [22].Chen’s approach [22] was used to measure the projected shootarea (A p ) at just three camera incidence angles: 0u, 45u and 90u.The following equation is used to calculate half of the total shootarea (A s ):A s ~2 cos(150 )A p (0 0 ,0 0 )zcos(45 0 )A p (45 0 ,0 0 )zcos(75 0 )A p (90 0 ,0 0 )ð5Þcos(15 0 )zcos(45 0 )zcos(75 0 )We obtained c E by combining equation (4) with (5). For Piceaspp. and A. nephrolepis and deciduous forests, individual leaves wereconsidered to be foliage elements, so c E =1.2.3.2. Seasonality of LAI in the leaf-expansion season.Leaf (needle) seasonality observations. We carried out leaf(needle) seasonality observations with periodic in situ observationsof sample foliage from 14 species in leaf-expansion season. Underthe influence on light, water conditions and nutrients, the growthrate of single foliage may be influenced by its position within thecanopy and the stand. Three trees were sampled from each specieswithin different conditions, and five leaves for broad-leaved speciesand fifteen needles for coniferous species were chosen from eachdirection (east, south, west and north), with different heights, forone tree. On May 1 st , May 14 th , May 22 nd , May 30 th , June 5 th ,June 14 th , July 1 st and July 15 th , we obtained the size (length andwidth or thickness) of each sample leaf (needle), and the elementsof the sample needles were measured in several places.For broad-leaved species, the area of a single leaf is not easilymultiplied length by width because of the irregular shape of theleaves. Thus we used an adjustment coefficient to adjust the leafarea based on the length and width of single leaf. To obtain theadjustment coefficient (m), the following equation is used:m~ SLDwhere S is the area of a single leaf, L is the length of the leaf, D isthe width of the leaf. To obtain the values, 20 leaves were collectedfrom each species, and we were able to calculate the half leaf areaby scanning. We obtained m value by combining the half leaf areawith the length and width from observations. The mean leaf areaof each species was obtained periodically by combining the lengthand the width from observations with m value that was assumed tobe unchanged in the leaf-expansion season.For coniferous species, the width and thickness of needles, whichwere determined through periodic monitoring, did not changedramatically. Therefore, we can assume that the width andthickness of needles kept the same value during the leaf-expansionseason, and the area of single needle can be calculated as follows:ð4Þð6ÞPinus koraiensis S~3al ð7ÞPicea spp: S~4al ð8ÞPLoS ONE | www.plosone.org 3 March 2012 | Volume 7 | Issue 3 | e3215514

Estimate of Leaf Area Indexwhere S is the area of single needle, a is the side of the crosssection,the values are 1.00 mm and 0.98 mm, respectively, and lis the length of the needle. The value of l is obtained from periodicobservations.Abies nephrolepis S~2a(bzc)lwhere S is the area of a single needle, b is the width of a needle, thevalue is 1.33 mm, c is the thickness of the needle, the value is0.44 mm, and l is the length of needle. The value of l was alsodetermined from periodic observations. By periodically obtainingin suit observations of sample needles, we could estimate the meanarea of a single needle of each coniferous species using equations(7)–(9).Seasonality of LAI for deciduous species. For broadleavedspecies, if the mean leaf area and the number of leaf foreach species during the leaf-expansion season are obtained, we canestimate LAI of each species for one sample point. Assuming thatthe total leaf number remains the same throughout the leafexpansionseason, thus, the total leaf number can be obtained bycombining little-trap data (the total leaf area of litter of the wholestudy period) with the mean leaf area of mature leaves (that fromlate leaf-expansion season) for each species that is calculated byequation (6). The total leaf area for each species could be estimatedusing the following two equations:S Total{i ~D i S iS Total{i ~N i S Mean{ið9Þð10Þð11Þwhere S Total-i is the total leaf area of species i in the late leafexpansionseason (the maximum LAI period), D i is the total massof species i throughout the whole study period, S i is the SLA ofspecies i, N i is the total leaf number of species i in the leafexpansionseason, and S Mean-i is the mean leaf area for matureleaves of species i. With equations (6), (10) and (11) and the littertrap area, we can then calculate LAI seasonality of all broadleavedspecies on each sample point during the leaf-expansionseason.Seasonality of LAI for coniferous species. For coniferousspecies, we can similarly assume that the total needle numberremains the same in the leaf-expansion season. Based on the meanleaf area of mature needle of each species for one sample pointthat can be estimated using equations (7)–(9), by combining theincreased total LAI for each species throughout the investigationperiod, we can calculate the total needle number. This increasedtotal LAI can be estimated from the needle growth and fall duringinvestigation period for coniferous species using the followingequation:DLAI~L a zL b {L cð12Þwhere L a is true LAI of November 15 th , L b is the total LAI from theneedle litter (from May 1 st to November 15 th ), and L c is true LAI ofMay 1 st .The increased total LAI of each coniferous species is obtainedby combining equation (12) with the ratio of LAI from the litter ofeach to all species throughout the investigation period. Then, wecan obtain the total leaf area of each coniferous species on eachsample point using the following equation:S Total{i ~L i Að13Þwhere S Total-i is the total leaf area of coniferous species i, L i is theincreased total LAI of coniferous species i, and A is the area of thelitter trap. By adding equations from (7)–(13), we can estimate LAIseasonality of all coniferous species on each sample point duringthe leaf-expansion season.Then, combining LAI of broad-leaved species with conifers, wecan estimate the increased LAI of the whole canopy from the startof the investigation period through using the following equation:XLi ~ N i|S Mean{iAð14Þwhere L i is the increased LAI of species i, N i is the total leaf (needle)number of species i, S Mean-i is the mean leaf (needle) area of speciesi, and A is the total area of all sampling areas (litter traps).For example, when true LAI of May 1 st as a benchmark isadded to the increased LAI of May 14 th , minus LAI from theneedle litter during the period, we obtain true LAI of May 14 th .Byanalogy, true LAI of the study stand can be obtained for othersampling times during the leaf-expansion season.2.3.3 Seasonality of LAI in the leaf-fall season.Observation of specific leaf area. Specific leaf area (SLA)(the leaf area per unit of dry mass, cm 2 g 21 ) is determined inrelation to the species and living conditions of sample trees, and tothe positions of sample leaves within the crown [32–34]. However,SLA (single species) has been shown to change only slightly duringthe leaf-expansion season [35]. To accurately obtain SLA of eachspecies, the above factors were considered when sampling themature leaves (needles) of the main species.Non-flat leaves were not collected when sampling broad leaves.The flat one-sided areas of leaves were measured by scanning. Thearea was calculated by counting the number of leafy pixels andmultiplying the number by the pixel size. To reduce error, SLA ofuncertain broad-leaved fragments from the litter was obtainedfrom the mean SLA of other broad-leaved species.Needle age (current year versus one-year-old needles) wasconsidered when sampling needles [36,37]. The areas of needleswere measured by the volume displacement method [22]. First,the whole shoot with the stem was immersed in water in acontainer that was resting on a sensitive balance and that was largeenough to prevent the shoot from touching the side or the bottomof the container (moderate detergent was put into the water toreduce the water’s surface tension). The displaced water volume ismeasured as the increase in weight if the shoot is not touching theside or the bottom of the container because the displaced waterexerts forces equally in all directions including the bottom of thecontainer. Then, we could obtain the entire shoot volume (V 1 )through increasing the weight. Second, the stem volume (V 2 ) wasmeasured in the same way, with the needles removed, andsimultaneously, the number of needles and the average length ofthe needles were measured. To obtain the volume of the needles(V), the total volume was reduced by the stem volume, brieflyV~V 1 {V 2 .The top of the needles was found to be acuminate, and we couldignore the areas at the top because they were negligibly small. Theshape of P. koraiensis approximates a triangular prism, and Piceaspp. and A. nephrolepis are cuboid. The equations of half the totalneedle area are easily obtained for P. koraiensis and Picea spp.because their cross-sections are an equilateral triangle and square,respectively. For A. nephrolepis, the ratio width and thickness of eachneedle were measured by Vernier calipers. To reduce the error,the width and thickness were taken as the average of 1/4, 1/2 and3/4 of each needle. The equations of half the total needle area (A)are calculated as follows:PLoS ONE | www.plosone.org 4 March 2012 | Volume 7 | Issue 3 | e3215515

Estimate of Leaf Area IndexpPinus koreaiensis : A~2:28 ffiffiffiffiffiffinvlpPicea spp: : A~2:00 ffiffiffiffiffiffinvlpAbies nephrolepis : A~2:31 ffiffiffiffiffiffinvlð15Þð16Þð17ÞTable 1. Element clumping index, quantifying the effect offoliage clumping at scales larger than the shoot, as a functionof the solar zenith angle for the study stand.Date Mean ± SD Maximum Minimum SampleMay 1 st 0.9160.04 1.00 0.80 64November 15 th 0.9160.04 0.97 0.81 64where v is the displaced volume (cm 3 or g) of the needles in a shoot,n is the total number of needles submerged, and l is the averagelength (cm).The mass of all the sampling leaves (needles) was measured afterdrying (65uC for 48 h). Then, we can calculate SLA using thefollowing equation:S i ~PAiPWið18Þwhere S i is the specific leaf area of species i, A i is the area of speciesi, and W i is the dry mass of species i.Litter trap observations. Each litter trap had a squareaperture of 0.5 m 2 , and its base was approximately 0.5 m abovethe ground. From May to September, litter was recovered at thesame time that the leaf expansion was surveyed, and fromSeptember to mid-November, litter was recovered bimonthly.During each litter collection, we sorted the litter in each trap intothe leaves of each species in time to avoid affecting the results dueto the decomposition of the leaves. After the litter was weighed, thesampled leaves were dried at 65uC for 48 h, and the total dry massof all the leaves was obtained. LAI was then calculated throughSLA. The leaves of the deciduous species from May to Augustcould be discarded because the number of leaves collected wassmall. In fact, LAI from deciduous species during that periodoccupied only 0.04% of the leaves collected during the whole studyperiod.Therefore, when true LAI of November 15 th is used as abenchmark and is added to LAI of November 15 th from the litter(both deciduous and coniferous species), LAI of November 1 st isobtained. By analogy, true LAI of the study stand can be obtainedfor other sampling times during the leaf-fall season.Finally, by adding the component LAI of all species, we canthen obtain true LAI for the entire canopy from the initial leafexpansion to the leaf fall.Results3.1 Clumping indexThe clumping index of each sample point on May 1 st andNovember 15 th was directly obtained from DHP and TRACWin3.9.1 software, and the difference in the results was not significant(Table 1).3.2 Needle-to-shoot area ratio (c E )The values of c E for the P. koraiensis ranged from 1.48 to 2.68,and the standard deviation of each c E was less than 0.52 (Table 2).The average c E increased with the increasing height level, but thevariations found within the top were dominant, co-dominant andsuppressed, and the values were 2.37 (SD = 0.48), 1.92 (SD = 0.40)and 1.81 (SD = 0.37), respectively. However, the variations withinthe middle and low portion of each canopy class were small, andthe mean value was 1.64 (SD = 0.04). Generally, dominant treeshad the largest values, followed by co-dominant and suppressedNote: All the values of clumping index were unitless.doi:10.1371/journal.pone.0032155.t001trees. These large differences were mostly determined by thegrowth conditions, such as light and water [32], suggesting that theseparation of the canopy classes was necessary for the samplingstrategy. In this study, the mean value of 1.77 was used as theneedle-to-shoot area ratio of the P. koraiensis.3.3 Specific leaf area (SLA)SLA largely varied with tree species (Table 3). SLA of F.mandshurica was largest with a value of 385.96 cm 2 g 21 , but SLA ofPicea spp. was only 49.84 cm 2 g 21 . Generally, SLA of coniferousspecies was smaller with a mean value of 66.60 cm 2 g 21(SD = 15.06), which was a quarter of the broad-leaved species.3.4 Adjustment coefficientSlight differences were found within species because of thedifferent shapes (Table 4). The adjustment coefficient values forthe study species ranged from 0.48 to 0.74, and an average valuewas 0.65 (SD = 0.08). We found that the shape of the leavesdetermined the value of adjustment coefficient, the product oflength and width of heart-shaped (or elliptical) leaves was moreclose to true leaf area (namely the adjustment coefficient value wasbigger) than leaves of other shapes, such as in T. amurensis and T.mandshurica. However, the palm-shaped A. mono had an adjustmentcoefficient value of only 0.48.Table 2. Mean needle-to-shoot area ratio and standarddeviation of P. koraiensis for nine trees and height classes innorth-eastern China.Canopy Sample Top (T) Middle (M) Low (L)Dominant (D) a 2.6860.33 1.7860.15 1.5660.15b 2.3860.35 1.6360.22 1.5860.14c 2.0360.52 1.6560.20 1.7860.20Co-dominant (M) d 1.5960.15 1.5760.24 1.5460.21e 1.9560.35 1.6360.21 1.6060.23f 2.2360.37 1.8560.22 1.6460.25Suppressed (S) g 1.7760.25 1.6860.25 1.7060.10h 2.1560.35 1.6060.14 1.6960.16i 1.4960.14 1.4960.15 1.4860.11Mean 1.7760.37Note: In the stand, 243 shoot samples were taken from nine trees: threedominant (D), three co-dominant (M), and three suppressed (S), at threeheights: top (T), middle (M), and bottom (L), forming nine classes with 27 shootsamples each: DT, DM, DL, MT, MM, ML, ST, SM, and SL; all these values wereunitless.doi:10.1371/journal.pone.0032155.t002PLoS ONE | www.plosone.org 5 March 2012 | Volume 7 | Issue 3 | e3215516

Estimate of Leaf Area IndexTable 3. Specific leaf area of main tree species, obtained fromsample foliage in the study stand.Species SLA (cm 2 g 21 )Pinus koraiensis 79.00Picea spp. 49.84Abies nephrolepis 70.96Acer mono 315.16Fraxinus mandshurica 385.96Tilia amurensis 163.3Betula costata 197.44Acer tegmentosum 241.28Corylus mandshurica 382.94Acer ukurunduense 378.99Ulmus laciniata 300.16Tilia mandshurica 354.49Ulmus japonica 212.42Prunus padus 123.95Populus ussuriensis 125.78Quercus mongolica 280.05Note: Specific leaf area (SLA) of uncertain broad-leaved fragments from thelitter was obtained from the average SLA of other broad-leaved species.doi:10.1371/journal.pone.0032155.t0033.5 Leaf seasonality observationsAll the species showed clear seasonality of the single leaf area(Figure 1). As seen in the changing leaf area of the species, mostspecies except F. mandshurica and A. mono had a single leaf flush.The first species that reached leaf flush were U. japonica and P.padus, lasting from early May to approximately May 22 nd . T.mandshurica and U. laciniata reached the flush in mid-May, and itlasted approximately half a month. As of May 22 nd , more specieswere reaching flush, such as C. mandshurica, B. costata, A.Table 4. Adjustment coefficient, adjusting the leaf area basedon the length and width of single leaf, obtained from 20sample leaves for each broad-leaved species in the studystand.Species Adjustment SDcoefficientPrunus padus 0.72 0.05Betula costata 0.62 0.01Acer ukurunduense 0.54 0.09Tilia mandshurica 0.73 0.02Ulmus laciniata 0.64 0.04Corylus mandshurica 0.72 0.03Acer tegmentosum 0.67 0.04Fraxinus mandshurica 0.63 0.02Acer mono 0.48 0.03Ulmus japonica 0.62 0.04Tilia amurensis 0.74 0.06Note: All the values of adjustment coefficient were unitless.doi:10.1371/journal.pone.0032155.t004ukurunduense, A. tegmentosum, and T. amurensis. In contrast, F.mandshurica and A. mono showed two leaf flushes: the first onearound May 22 nd and the second in early June, each lasting forabout ten days. Generally, coniferous species had a lateremergence of leaves, such as P. koraiensis and Picea spp., and wereall after mid-May. Only the growth of A. mono was irregular (theleaf area decreased in late May and recovered in mid-June),probably because of herbivory by insects (mainly aphids), andwhen the leaves in the second flush grew larger than those eaten byinsects, the single leaf area recovered.3.6 Litter trap observationsThe total area from litter of all species collected throughout theinvestigation period was 1.65 m 2 , and the area from the mainspecies showed clear differences (Table 5). Generally, the areafrom the broad-leaved species was about three times as large asthat of the coniferous species. However, P. koraiensis accounted for21.35% (maximum ratio) of the total area. The standard error ofthe total area per species were all less than 0.04, suggesting that theexperimental strategy (the number of traps (64) and measurementof SLA) was reliable for the study site.3.7 LAI of main species in all seasonsBy combining the three components: true LAI of May 1 st andNovember 15 th , litter trap data and the seasonality of the leaf area,we were able to estimate LAI of the major broad-leaved speciesduring all seasons (Figure 2). Every broad-leaved species reachedits peak in July, and A. mono had the largest peak LAI with a valueof 0.43 (SE = 0.05) m 2 m 22 , followed by F. mandshurica with a valueof 0.37 (SE = 0.10) m 2 m 22 . P. padu was the smallest, only as largeas approximately one percent of the largest. All broad-leavedspecies, except for U. laciniata, began to fall in early September,and U. laciniata fell in late August. All species had a rapid falling ofleaves in late September, and the leaves of P. padu and A.tegmentosum fell in early October, but the others fell in mid-October.LAI of the three coniferous species were always greater than 0because they were evergreen. By combining leaf seasonalityobservations and litter trap data, we were able to estimate thedynamic variations of LAI in increased and decreased (DLAI)during all seasons (Figure 3). For Picea spp. and A. nephrolepis, thechanges in DLAI had a single flush in mid-July. However, for P.koraiensis, the DLAI decreased in mid-June and recovered in earlyJuly showed that the range of LAI increased from leaf seasonalityobservations was less than the range of LAI decreased from littertrap data during this period. In general, for the three coniferousspecies, the increased LAI from leaf seasonality observations werenearly the same as the decreased LAI from the litter trap data.The total LAI of all broad-leaved species showed clear seasonalchanges with a maximum of 2.17 m 2 m 22 on July 15 th (Figure 4).Although the total LAI of all coniferous species had the largestpeak with the value of 3.89 m 2 m 22 on July 15 th , it only increased23% more than the minimum.3.8 LAI estimation using the indirect optical methodLAI estimated from the optical method for the studied speciesranged from 1.79 to 3.83 m 2 m 22 , obviously lower (the meanunderestimated 41.64%) than those provided by our methodthroughout the study period (Figure 5). LAI from both methodspeaked in early August. Moreover, the pattern of seasonal changewas different. In late May, the optical method showed a lowerincreasing speed than our method. From September 15 th toOctober 15 th , the optical method showed a lower decreasing speedthan our method, whereas the seasonality of LAI from the opticalmethod showed little variation over the entire study period.PLoS ONE | www.plosone.org 6 March 2012 | Volume 7 | Issue 3 | e3215517

Estimate of Leaf Area IndexFigure 1. Seasonality of mean leaf area for the tree species, obtained from 60 sample leaves (needle samplings were moderatelyincreased) of 3 individuals for each species. Each time series for the data was normalized using the annual maximum value set to 1.0.doi:10.1371/journal.pone.0032155.g001PLoS ONE | www.plosone.org 7 March 2012 | Volume 7 | Issue 3 | e3215518

Estimate of Leaf Area IndexTable 5.Total area from litter accounted for by the majorspecies at the study site in the whole investigation period.Species Area (m 2 ) ± SE Fraction (%)Prunus padus 0.0060.00 0.11Ulmus japonica 0.0260.01 1.49Abies nephrolepis 0.0360.00 1.70Acer tegmentosum 0.0360.01 1.83Picea spp. 0.0460.00 2.26Tilia mandshurica 0.0460.01 2.50Corylus mandshurica 0.0860.02 4.76Acer ukurunduense 0.0860.02 4.93Ulmus laciniata 0.1360.02 7.97Betula costata 0.1360.02 8.17Tilia amurensis 0.1660.03 9.84Fraxinus mandshurica 0.1760.04 10.35Acer mono 0.2160.02 13.00Pinus koraiensis 0.3060.03 21.35other 0.16 9.70total 1.65 100.00Note: Number of litter traps per species n = 64.doi:10.1371/journal.pone.0032155.t005Discussion4.1 Reliability of the proposed methodIn our study, 11 broad-leaved species that we selected for leafseasonality observations accounted for 87.0% of LAI (August 1 st )estimated from the litter trap data. This percentage suggests thatthe seasonality of about 13% of the total LAI was uncertain,probably a result of other broad-leaved species without leafexpansionobservations (the basal area of other broad-leavedspecies accounts for 12.8% of the total broad-leaved species). Asimilar result was shown by Nasahara et al. [38], in which speciesselected for the leaf seasonality observations accounted for 84% ofLAI estimated from the litter trap data. Our result did not containconiferous species because those leaves were falling during theentire study period. Therefore, we could probably measure moreaccurate assessments of the seasonality of LAI by obtaining leafexpansionobservations for more species in future studies.In present study, we could not calculate the increased LAI foreach coniferous species if we only used litter traps during the leaffallseason because needles fell during the entire study period. Wesolved this problem successfully. First, we estimated the total LAIof all conifers from the litter collected throughout the study period,added true LAI of November 15 th , and subtracted true LAI ofMay 1 st before obtaining the increased LAI of all coniferousspecies, simplified as equation (12). By combining the ratio of litterfrom each coniferous species of all species in the entire studyperiod, we obtained the increased LAI of each coniferous species.Figure 2. LAI of total broad-leaved species in the study site estimated by three components: true LAI of May 1 st and November 15 th ,litter trap data and leaf seasonality observations. Error bars represent the standard error.doi:10.1371/journal.pone.0032155.g002PLoS ONE | www.plosone.org 8 March 2012 | Volume 7 | Issue 3 | e3215519

Estimate of Leaf Area IndexFigure 3. LAI of total coniferous species in the study site estimated by the increased LAI from leaf seasonality observations minusthe decreased LAI from litter trap data. Error bars represent the standard error.doi:10.1371/journal.pone.0032155.g003Figure 4. The total LAI of broad-leaved and coniferous speciesestimated throughout the study period. For broad-leaved species,LAI was estimated using the leaf seasonality observations and litter trapdata during the study period; and for coniferous species, in addition tothese data, LAI was also estimated based on true LAI on May 1 st andNovember 15 th because there are no leafless periods in a mixed forest.Error bars represent the standard error.doi:10.1371/journal.pone.0032155.g004There are some weaknesses in the present study. First, we calibratedthe Picea spp. and A. nephrolepis by needle-to-shoot area ratio by mistakewhile obtaining true LAI of May 1 st and November 15 th because thetwo species did not exhibit the clumping effect within the shoots.However, we determined the error, with a value of 7%, according to thelitter proportion of the two species (LAI from the litter of the two specieswas 15.6% of the total coniferous species), and obviously, we coulddiscard the error relative to the 41.64% underestimated by the opticalmethod. Second, although we considered factors that could affect leafmeasurements (e.g., shape, size, and growth conditions) of the sampleleaves during our measurements to adjust the coefficients of all species,the sample leaves were not collected at the same time as the leafexpansionobservation; thus, measuring whether or not the adjustmentcoefficients were influenced by the leaf collection method was notvalidated in this study. Finally, we ignored the influence on perennialneedles when estimating LAI. If we could eliminate these parts of theprocess, our method would become more accurately.4.2 Measurement of main parametersLAI obtained during the leafless period by the opticalmethods was assumed to represent the woody area index,which was used in previous studies. However, that is not suitableFigure 5. LAI of the canopy estimated by two methods: The method proposed in the present study (which was estimated by threecomponents: LAI of May 1 st and November 15 th , litter trap data and leaf seasonality observations) and the hemisphericalphotography. Error bars represent the standard error.doi:10.1371/journal.pone.0032155.g005PLoS ONE | www.plosone.org 9 March 2012 | Volume 7 | Issue 3 | e3215520

Estimate of Leaf Area Indexfor a coniferous forest because there are no leafless periods in anevergreen forest stand [27,30,39]. We can measure the woodyto-totalarea ratio using Photoshop software and the opticalmethod, and to overcome the huge investment in labor andtime, this method is more reasonable because it is forbidden todestroy the national nature reserve. The mean values of May 1 stand November 15 th were 0.096 and 0.092, respectively. Thosevalues conform to the published values of a range from 0.03 to0.40 [30,40].The clumping effect exists not only beyond the shoots but alsowithin the shoots (especially in conifer forest). The measurement ofthe clumping index (beyond the shoots) is an exciting topic, andthe methods mainly include CI LX (clumping index fromlogarithmic gap averaging); CI W (clumping index from modifiedlogarithmic gap averaging); CI CC (clumping index from gap sizedistribution); CI CLX (clumping index from combination of gap sizeand logarithmic averaging); and CI PCS (clumping index fromPielou’s coefficient of spatial segregation) [41]. We obtained theclumping index directly from the DHP-TRAC software that waswidely used by other studies [39,40]. By making comparativestudies of these methods in future studies, we will determine whichone is more suitable for this research site. The clumping effectwithin the shoots exists in conifer species and varies with the livingconditions of trees [39]. To reduce the error, we took the averageof a large quantity of shoot sample within different conditions. Theneedle-to-shoot area ratio was 1.77. Similar results have beenpublished. For instance, Chen et al. [29] realized a needle-to-shootarea ratio that ranged from 1.4 to 1.8 for a boreal conifer forest,and Bréda et al. [15] estimated c E values of 1.2–2.0 for a stand ofconiferous forest.4.3 Comparison with optical methodAt present, optical methods are widely used to estimate LAI andits dynamic changes because they are easier and quicker to carry out.However, the hemispherical photography technique tends toReferences1. Muraoka H, Saigusa N, Nasahara KN, Noda H, Yoshino J, et al. (2010) Effectsof seasonal and interannual variations in leaf photosynthesis and canopy leafarea index on gross primary production of a cool-temperate deciduous broadleafforest in Takayama Japan. J Plant Res 123: 563–576.2. Chen JM, Black TA (1992) Defining leaf area index for non-flat leaves. PlantCell Environ 15: 421–429.3. Behera SK, Srivastava P, Pathre UV, Tuli R (2010) An indirect method ofestimating leaf area index in Jatropha curcas L using LAI-2000 Plant CanopyAnalyzer. Agric For Meteorol 150: 307–311.4. Sonnentag O, Chen JM, Roberts DA, Talbot J, Halligan KQ, et al. (2007)Mapping tree and shrub leaf area indices in an ombrotrophic peatland throughmultiple endmember spectral unmixing. Remote Sens Environ 109: 342–360.5. Sprintsin M, Karnieli A, Berliner P, Rotenberg E, Yakir D, et al. (2007) Theeffect of spatial resolution on the accuracy of leaf area index estimation for aforest planted in the desert transition zone. Remote Sens Environ 109: 416–428.6. Asner GP, Scurlock JMO, Hicke JA (2003) Global synthesis of leaf area indexobservations: implications for ecological and remote sensing studies. Global EcolBiogeogr 12: 191–205.7. Barr AG, Black TA, Hogg EH, Kljun N, Morgenstern K, et al. (2004) Interannualvariability in the leaf area index of a boreal aspen-hazelnut forest inrelation to net ecosystem production. Agric For Meteorol 126: 237–255.8. Gower ST, Norman JM (1991) Rapid estimation of leaf area index in coniferand broad-leaf plantations. Ecology 72: 1896–1900.9. Deblonde G, Penner M, Royer A (1994) Measuring leaf area index with the LI-COR LAI-2000 in pine stands. Ecology 75: 1507–1511.10. Jonckheere I, Fleck S, Nackaerts K, Muys B, Coppin P, et al. (2004) Review ofmethods for in situ leaf area index determination: Part I Theories, sensors andhemispherical photography. Agric For Meteorol 121: 19–35.11. Pinto-Júnior OB, Sanches L, Lobo FdA, Brandão AA, Nogueira JdS (2010) Leafarea index of a tropical semi-deciduous forest of the southern. Amazon BasinInt J Biometeorol;doi 101007/s00484-010-0317-1.12. Sonnentag O, Talbot J, Chen JM, Roulet NT (2007) Using direct and indirectmeasurements of leaf area index to characterize the shrub canopy in anombrotrophic peatland. Agric For Meteorol 144: 200–212.underestimate LAI [25,42,43]. In comparison with our method,the hemispherical photography method provided lower LAI values(a range from 27.12% to 51.07%) and a smaller seasonal variationamplitude. Underestimates of hemispherical photography have beenreported in other studies. For instance, Zhang et al. [43]demonstrated that digital hemispherical photographs taken withautomatic exposure are not reliable, causing L e underestimations by16–71%, and Van Gardingen et al. [44] found that thehemispherical photography method resulted in an underestimateof 50% compared to a destructive harvest. So, it is necessary tovalidate and improve indirect optical methods. To learn the reasonsfor the discrepancy between the proposed method and opticalmethod, we will need to validate each step in the derivation of LAI ineach method and present that in a further study on this stand.ConclusionsThis proposed method can provide not only the total LAI butalso LAI for each species and its seasonal changes. By contrast, theoptical method average underestimated LAI by 41.64%(SD = 6.54). Based on reasonably calibrating LAI from the opticalmethod, by combining leaf-expansion observations with litter trapsto estimate LAI and its seasonal changes in a mixed broadleaved-Korean pine forest, this method will become an effective methodin the future.AcknowledgmentsWe thank Dr. J. M. Chen for providing us with the technology to collectand analyze the data. We greatly appreciate input of editor and twoanonymous reviewers for improving this manuscript.Author ContributionsConceived and designed the experiments: GJ. Performed the experiments:ZL YQ. Analyzed the data: ZL YQ. Wrote the paper: ZL GJ.13. Ryu Y, Sonnentag O, Nilson T, Vargas R, Kobayashi H, et al. (2010) How toquantify tree leaf area index in an open savanna ecosystem: A multi-instrumentand multi-model approach. Agric For Meteorol 150: 63–76.14. Macfarlane C, Grigg A, Evangelista C (2007) Estimating forest leaf area usingcover and fullframe fisheye photography: thinking inside the circle. Agric ForMeteorol 146: 1–12.15. Bréda NJJ (2003) Ground-based measurements of leaf area index: a review ofmethods, instruments and current controversies. J Exp Bot 54: 2403–2417.16. Marshall JD, Waring RH (1986) Comparison of methods of estimating leaf-areaindex in old-growth Douglas-fir. Ecology 67: 975–979.17. Neumann HH, Den Hartog G, Shaw RH (1989) Leaf area measurements basedon hemispheric photographs and leaf-litter collection in a deciduous forestduring autumn leaf-fall. Agric For Meteorol 45: 325–345.18. Mussche S, Samson R, Nachtergale L, De Schrijver A, Lemeur R, et al. (2001) Acomparison of optical and direct methods for monitoring the seasonal dynamicsof leaf area index in deciduous forests. Silva Fennica 35: 373–384.19. Cutini A, Matteucci G, Mugnozza GS (1998) Estimation of leaf area indexwith the Li-Cor LAI 2000 in deciduous forests. Forest Ecol Manag 105: 55–65.20. Wasseige C, Bastin D, Defourny P (2003) Seasonal variation of tropical forestLAI based on field measurements in Central African Republic. Agric ForMeteorol 119: 181–194.21. Pierce LL, Running SW (1988) Rapid estimation of coniferous forest leaf areaindex using a portable integrating radiometer. Ecology 69: 1762–1767.22. Chen JM (1996) Optically-based methods for measuring seasonal variationof leaf area index in boreal conifer stands. Agric For Meteorol 80: 135–163.23. Jin GZ, Xie XC, Tian YY, Kim JH (2006) The pattern of seed rain in thebroadleaved-Korean pine mixed forest of Xiao xing an Mountains, China. J KorFor Soc 95: 621–627.24. Garrigues S, Shabanov NV, Swanson K, Morisette JT, Baret F, et al. (2008)Intercomparison and sensitivity analysis of Leaf Area Index retrievals from LAI-2000, AccuPAR, and digital hemispherical photography over croplands. AgricFor Meteorol 148: 1193–1209.PLoS ONE | www.plosone.org 10 March 2012 | Volume 7 | Issue 3 | e3215521

Estimate of Leaf Area Index25. Montes F, Pita P, Rubio A, Cañellas I (2007) Leaf area index estimation inmountain even-aged Pinus silvestris L stands from hemispherical photographs.Agric For Meteorol 145: 215–228.26. Schleppi P, Conedera M, Sedivy I, Thimonier A (2007) Correcting non-linearityand slope effects in the estimation of the leaf area index of forests fromhemispherical photographs. Agric For Meteorol 144: 236–242.27. Leblanc SG, Chen JM, Fernandes R, Deering DW, Conley A (2005) Methodologycomparison for canopy structure parameters extraction from digital hemisphericalphotography in boreal forests. Agric For Meteorol 129: 187–207.28. Miller JB (1967) A formula for average foliage density. Aust J Bot 15: 141–144.29. Chen JM, Rich PM, Gower ST, Norman JM, Plummer S (1997) Leaf area indexof boreal forests: theory, techniques, and measurements. J Geophys Res 102:29429–29443.30. Zou J, Yan GJ, Zhu L, Zhang WM (2009) Woody-to-total area ratiodetermination with a multispectral canopy imager. Tree Physiol 29: 1069–1080.31. Chen JM, Cihlar J (1996) Retrieving leaf area index of boreal conifer forestsusing Landsat TM images. Remote Sens Environ 55: 153–162.32. Bouriaud O, Soudani K, Bréda N (2003) Leaf area index from litter collection:impact of specific leaf area variability within a beech stand. Can J For Res 29:371–380.33. Gower ST, Kucharik CJ, Norman JM (1999) Direct and indirect estimation ofleaf area index, fAPAR, and net primary production of terrestrial ecosystems.Remote Sens Environ 70: 29–51.34. Nouvellon Y, Laclau JP, Epron D, Kinana A, Mabiala A, et al. (2010) Withinstandand seasonal variations of specific leaf area in a clonal Eucalyptusplantation in the Republic of Congo. Forest Ecol Manag 259: 1796–1807.35. Eriksson H, Eklundh L, Hall K, Lindroth A (2005) Estimating LAI in deciduousforest stands. Agric For Meteorol 129: 27–37.36. Ishii H, Ford ED, Boscolo ME, Manriquez AC, Wilson ME, et al. (2002)Variation in specific needle area of old-growth Douglas-fir in relation to needleage, within-crown position and epicormic shoot production. Tree Physiol 22:31–40.37. Wang XC, Janssens IA, Curiel Yuste J, Ceulemans R (2006) Variation of specificleaf area and upscaling to leaf area index in mature Scots pine. Trees-structFunct 20: 304–310.38. Nasahara KN, Muraoka H, Nagai S, Mikami H (2008) Vertical integration ofleaf area index in a Japanese deciduous broad-leaved forest. Agric For Meteorol148: 1136–1146.39. Chen JM, Govind A, Sonnentag O, Zhang YQ, Barr A, et al. (2006) Leaf areaindex measurements at Fluxnet-Canada forest sites. Agric For Meteorol 140:257–268.40. Macfarlane C, Hoffman M, Eamus D, Kerp N, Higginson S, et al. (2007)Estimation of leaf area index in eucalypt forest using digital photography. AgricFor Meteorol 143: 176–188.41. Gonsamo A, Pellikka P (2009) The computation of foliage clumping index usinghemispherical photography. Agric For Meteorol 149: 1781–1787.42. Thimonier A, Sedivy I, Schleppi P (2010) Estimating leaf area index in differenttypes of mature forest stands in Switzerland: a comparison of methods.Eur J Forest Res 129: 543–562.43. Zhang YQ, Chen JM, Miller JR (2005) Determining digital hemisphericalphotograph exposure for leaf area index estimation. Agric For Meteorol 133:166–181.44. Van Gardingen PR, Jackson GE, Hernandez-Daumas S, Russell G, Sharp L(1999) Leaf area index estimates obtained for clumped canopies usinghemispherical photography. Agric For Meteorol 94: 243–257.PLoS ONE | www.plosone.org 11 March 2012 | Volume 7 | Issue 3 | e3215522

J For ResDOI 10.1007/s10310-012-0370-1ORIGINAL ARTICLEOptical and litter collection methods for measuring leaf area indexin an old-growth temperate forest in northeastern ChinaYujiao Qi • Guangze Jin • Zhili LiuReceived: 23 November 2011 / Accepted: 19 July 2012Ó The Japanese Forest Society and Springer 2012Abstract Hemispherical photographs combined withlitter collection were applied to determine seasonaldynamics of leaf area index (LAI) between the period ofmaximum leaf area and the leafless period from an oldgrowthtemperate forest in the Xiaoxing’an Mountains,northeastern China. Our objective is to explore the changein the relationship between ‘‘true’’ LAI and effective LAI(calculated only from hemispherical photography) and tofind the best LAI estimation models. Effective LAI inNovember is corrected for contribution of woody materialand clumping at shoot and beyond shoot levels, to giveminimum ‘‘true’’ LAI. The ‘‘true’’ LAI in each period isestimated as a sum of the minimum ‘‘true’’ LAI and littercollection LAI in each period. Power function regressioncalibration models were then carried out between ‘‘true’’LAI and effective LAI in each period and the entire litterfallperiod. Then, significance tests were applied to detectthe differences among different models. The resultsshowed that the average ‘‘true’’ LAI ranged from2.74 ± 0.54 on November 1 to 6.64 ± 1.34 on July 1. Forthe entire season, average effective LAI was 53.16 %lower than the average ‘‘true’’ LAI. After significance tests,calibration models were classified into two types: (1)maximum LAI period and the period of maximum leaf fall;(2) the period during which leaves began falling and alldeciduous leaves had fallen. Based on our experience, webelieve that the classified models can produce reliable andaccurate LA1 values for the needle and broad-leaved mixedforest stands under the non-destructive condition.Keywords An old-growth temperate forest Correctionfor LAI Hemispherical photography LAI Litter fallcollectionList of symbolsCC Clumping index calculated by Chen and Cihlar(1995)LX Clumping index calculated by Lang and Xiang(1986)CLX Clumping index that combines CC and CLXmethod suggested by Leblanc et al. (2005)L t Total plant area index that includes woodycomponentsL Leaf area indexa Woody-to-total area ratioX E Element clumping indexc E Needle-to-shoot area ratioD Dominant layerC Co-dominant layerS Suppressed layerT Top height of a treeM Middle height of a treeL Low height of a treeY. QiSchool of Forestry, Northeast Forestry University,Harbin 150040, ChinaG. Jin (&) Z. LiuCenter for Ecological Research, Northeast ForestryUniversity, Harbin 150040, Chinae-mail: taxus@126.comIntroductionLeaf area index (LAI), defined as half the total leaf area perunit ground surface area (Chen and Black 1992), is relatedto many biophysical and physiological processes, includingphotosynthesis, respiration, transpiration, carbon cycling,12323

J For Resnet primary productivity, and energy exchange (Bonan1993), and is an important factor in ecosystem models(Running and Gower 1991; Weiss et al. 2004).LAI can be measured directly or indirectly (Gower et al.1999; Levy and Jarvis 1999; Mussche et al. 2001; Jonckheereet al. 2004; Sonnentag et al. 2007; Behera et al.2010; Ryu et al. 2010; Sprintsin et al. 2011). Direct measurementsinclude the harvest method (Brenner et al. 1995;Whitford et al. 1995), and the litter collection method(Neumann et al. 1989; Cutini et al. 1998; Ishihara andHiura 2011), among others. The harvest method is generallysimple and reliable, but the procedure is destructive,laborious, and time consuming. The litter collectionmethod is useful for deciduous forests with adequate spatialand temporal sampling schemes (Neumann et al. 1989).Compared with direct methods, indirect methods arequicker and more efficient, and they are usually based onoptical methods that measure the canopy gap fraction as afunction of the zenith angle (Wang and Miller 1987;Gazarini et al. 1990; Chen 1996; Walter and Himmler1996). Optical measurements are made mainly by the followinginstruments: LAI-2000, Sunfleck Ceptometer,Demon, TRAC, and Hemispherical Photography, as well asothers. Many optical instruments (such as WinSCANO-PY2006, LAI-2000) measure LAI by inversion of the gapfraction, always ignoring the clumping effect, and oftenunderestimate LAI, especially in coniferous forests (Chen1996; Chen et al. 1997). Thus, methods that directlymeasure LAI are used to calibrate indirect optical methods(Fukushima et al. 1998; Eriksson et al. 2005; Kalácskaet al. 2005).At present, there are two major sources of errors associatedwith optical LAI measurements: the clumping effectand non-photosynthetic components. First, the clumpingeffect arises from both beyond-shoot clumping and withinshootclumping. Different techniques have been developedto quantify beyond-shoot clumping (element clumpingindex, X E ), including CC, which is based on the gap sizeand fraction analysis (Chen and Cihlar 1995), LX, whichuses a logarithmic gap averaging technique (Lang andXiang 1986), CLX, which is based on a combination ofconcepts used in CC and LX (Leblanc et al. 2005), andPCS-Pielou’s coefficient of spatial segregation (Pielou1962), among others. Unlike other methods, the CC correctionmethod has the advantages of both gap fraction andgap size information, which can be applied to all types ofplant canopies without need for spatial pattern assumptionsabout canopy elements. Within-shoot clumping (needle-toshootarea ratio, c E ) can be measured using the geometricmeasurement method (Johnson 1984; Oker-Blom andSmolander 1988) or the volume displacement method(Chen 1996; Chen et al. 1997), both of which requirefield sampling. Second, in forests, non-photosyntheticcomponents mainly refer to the stems and branches. Atpresent, methods for measuring the woody-to-total arearatio (a = woody area index/plant area index) can also beclassified into three types, namely direct methods, visualestimation methods, and indirect methods. Direct methodsrely on destructive sampling and usually involve measuringthe woody area of representative trees within forest stands.The direct method has rarely been used to estimate abecause of the huge investment in labor and time. Visualestimation does not require tree cutting, but estimate biasmay yield less reliable values. Indirect methods rely onradiation sensors and photos (Kucharik et al. 1998; Zouet al. 2009), the principle is based on differences in spectralcharacteristics of various tree parts, such as leaves, stems,and branches, and thus these methods still depend on fieldmeasurements.The goal of this study is to obtain a more accurate LAIunder the non-destructive condition in mixed broadleaved–pine forest and analyze its relationship with effective LAI.To achieve this goal, we (1) measured a, X E , c E , effectiveLAI, and ‘‘true’’ LAI, (2) constructed an empirical modelbased on the correlation between ‘‘true’’ LAI and effectiveLAI in each period, and (3) explored how different LAIchanged and what were the laws in the relationship ofdifferent models. And our final objective is to develop aless labor-intensive but accurate prediction method todetermine LAI using the litter trap and optical methods inan old-growth temperate forest in the Xiaoxing’an Mountainsin northeastern China.Materials and methodsSite descriptionThe study area is located in the Liangshui National Reserveof the Xiaoxing’an Mountains, northeastern China(47°10 0 50 00 N, 128°53 0 20 00 E). The Reserve is characterizedby rolling mountainous terrain, ranging from 300 to707.4 m asl (highest peak) with a typical slope of 10°–15°.It covers 12,133 ha with about 1.7 million m 3 of growingstock and 98 % canopy coverage. The reserve has one ofthe most concentrated and well-conserved mixed broadleaved–Koreanpine forest in China. The forest vegetationis primarily composed of Pinus koraiensis, Picea koraiensis,Abies nephrolepis, Tilia amurensis, T. mandshurica,Acer mono, Fraxinus mandshurica, Ulmus laciniata, Betulacostata, B. platyphylla, Quercus mongolica, Larixgmelini, Juglans mandshurica, A. ukurunduense, and A.tegmentosum. The mean annual temperature is -0.3 °Cwith a highest mean temperature of 7.5 °C and lowestmean temperature of -6.6 °C. Mean annual precipitation is676 mm (Jin et al. 2006).12324

J For ResResearch methodsPlot settingThe study was carried out in a large permanent plot (9 ha,300 m 9 300 m) within the Reserve. Every tree of [2 cmdiameter at breast height (DBH) was marked with a numberedtag (for DBH [8 cm, the tree was nailed with analuminum tag or wired with a copper tag), and its DBH,height, and coordinates were measured (Jin et al. 2006).A core area of 160 m 9 160 m was established at thecenter of the permanent plot. In the core area, and each20 m 9 20 m quadrat was treated as a site. Litter traps andhemispherical photography were performed at the samepoints (in the center of these sites), on an 8 9 8 grid (64total sample points).Litter collection and specific leaf areaIn 2009, leaf litter was collected once a month from earlyAugust through early September, and collected once every2 weeks from mid-September through early November.Litter traps were surrounded with wire of 8 mm and nylonmesh (pore size 1 mm, depth 0.5–0.6 m). The traps wereround, with area 0.5 m 2 , located 1 m above ground. Uponlitter collection, the dead leaves/needles were bagged,tagged, and taken back to the laboratory for species identification.Leaves/needles from each entire bag wereweighed after drying at 65 °C for 48 h.Specific leaf area (SLA) is defined as leaf area per unit dryweight. We adopted the definition of LAI used by Chen andBlack (1992) and measured unilateral area and half surfacearea for broadleaves and needles, respectively. In September2009, 30–50 leaves as a sample were collected, one samplewas taken to measure SLA by scanning (leaf area was calculatedby multiplying the number of leaf pixels by the pixelsize) for each deciduous species; and 200–300 perennialmature needles as a sample were measured, three sampleswere taken to measure SLA by the volume displacementmethod for each conifer species. To reduce error, we used themean SLA of other broad-leaved species to calculate SLA ofuncertain broad-leaved fragments from the litter.The shapes of needles of Pinus koraiensis, Picea spp., andAbies nephrolepis are a prism, diamond column, and rectangular,respectively; the needles are tapered, so theexceedingly small surface area at both ends was ignored.Needles and attached branches were immersed in a watercontainer, which was large enough to avoid needles touchingthe wall and bottom. The additional volume of water in thecontainer was the volume of needles and branches (V 1 ). Theneedles were then removed from branches, and the volumeof branches (V 2 ) was measured by the same method; theneedle volume was thus: V = V 1 - V 2 . Because each prismor diamond column is equilateral, the semi-surface areaformulas for Pinus koraiensis and Picea spp. were easilyderived. For Abies nephrolepis, the needles are rectangular,and thus we took 30 samples of Abies nephrolepis needlesand used a caliper to measure each needle’s length andwidth, and the average ratio of width and thickness weretaken. To reduce the error, the width and thickness weretaken as averages of 1/4, 1/2, and 3/4 of each needle. Finally,we determined the semi-surface area equation for Abiesnephrolepis. We used the following formulas to calculate thesemi-surface area (A) of the needles.pA ¼ 2:28ffiffiffiffiffiffinvl for Pinus koraiensisð1ÞpA ¼ 2:00ffiffiffiffiffiffinvl for Picea spp:ð2ÞpA ¼ 2:31ffiffiffiffiffiffinvl for Abies nephrolepisð3Þwhere v is the displaced volume (cm 3 or g) of needles in ashoot, n is the total number of needles submerged, and l isthe average length (in cm). Needles (without branches)were dried and weighed.SLA was calculated as follows:S i ¼PAiPWið4Þwhere S i is the SLA of species i, A i is the semi-surface areaof leaf i, and W i is the dry weight of leaf i.Litter collection LAI during a given period was calculatedusingXLi ¼ D i S i =Að5Þwhere L i is the reduced LAI of tree species i, D i is the dryweight of the litter of tree species i, S i is the SLA of treespecies i, and A is the area of the litter trap.Image acquisitionHemispherical photographs were acquired with a WinS-CANOPY2006 Canopy Analyzer (Nikon Coolpix P1 digitalcamera with a 180° fish-eye lens; Regent Instruments, Canada).The camera and lens were mounted on a sturdy, leveledtripod, and the fish-eye was at the same height of the littertraps. The photographs were taken with automatic exposureunder conditions of diffuse skylight, normally soon aftersunrise or immediately before sunset. Taking hemisphericalphotographs and collecting litter were done simultaneously.TheoryTheory of deriving LAIThe angular distribution of canopy gap fraction, G(h),where h is the zenith angle, is generally described as follows(Nilson 1971):12325

J For ResGðhÞ ¼exp½ KðhÞX E L t = cos hŠ ð6Þwhere K(h) is the projection coefficient characterizing thefoliage angle distribution, L t is the plant area indexincluding leaf and woody areas, and X E is the total foliageclumping index. When the foliage spatial distributionis random, X E is 1.0. If foliage is uniformly distributed,X E is [ 1.0. When foliage is clumped, X E is \1.0.Because many optical instruments measure G(h), fromwhich only the product of X E and L t is obtained, X E L t istherefore called the effective LAI denoted as L e (Chenet al. 1991).To obtain the true LAI, three corrections must be madeto L e (Chen 1996).L ¼ð1 aÞL e c E =X E ð7Þwhere L is true LAI, L e is effective leaf area index thatdirectly obtained from optical instruments, a is thewoody-to-total area ratio, c E is the needle-to-shoot arearatio quantifying the effect of foliage clumping within theshoots, X E is clumping index. For deciduous forests,individual leaves are considered as the foliage elements,and c E = 1. For needle-leaved forests, it is larger than1.0. X E includes the effect of foliage clumping at scaleslarger than the shoot (it decreases with increasedclumping).Woody-to-total area ratio (a)The woody-to-total area ratio (a) equals the woody areaindex (WAI) divided by the plant area index (PAI). InNovember at our study site, all deciduous leaves hadalready fallen, and the rest were needles, stems, andbranches. Thus, the LAI that was measured by hemisphericalphotography was actually the area index of needles,stems, and branches. Because it is difficult todistinguish woody cover from greenery when an opticalinstrument is used to measure LAI, we use Photoshopto remove the woody cover. That was before analyzinghemispherical photography with WinSCANOPY2006, the Clone Stamp Tool in Photoshop was used toreplace the stems and branches with their backgrounds(sky and leaves), which avoided removing the leavesbehind stems when just only using sky as the backgroundsand reduced the error. Each photo process took about2–3 min.Element clumping index (X E )Element clumping index (X E ) was computed based on thegap size and fraction analysis (Chen and Cihlar 1995, 2002;Leblanc 2002; Leblanc et al. 2005):X E ðhÞ ¼ In ½ F mð0; hÞŠ½1 F mr ð0; hÞŠð8ÞIn½F mr ð0; hÞŠ½1 F m ð0; hÞŠwhere F m (0,h) is the measured accumulated gap fractionlarger than zero, i.e., the canopy gap fraction, and F mr (0, h)is the gap fraction for the canopy when large gaps that arenot theoretically possible in a random canopy have beenremoved for a given LAI and foliage element width. Anglesthat range close to the zenith result in short segments, whichmay produce erroneous estimates of the gap fraction andsize distributions. On the other hand, those angles near tothe horizon yield a high proportion of mixed pixels owing tolight scattering and coarse image resolution. We usedangles ranging from 0° to 60° to measure LAI (Leblanc andChen 2001; Gonsamo and Pellikka 2009).Needle-to-shoot area ratio (c E )The needle-to-shoot area ratio is used to quantify foliageclumping within shoots. We adopted the volume replacementmethod proposed by Chen (1996) to measure c E .Toobtain an average value for a stand, trees were first groupedinto three categories by their height and size: dominant (D),co-dominant (C), and suppressed (S), and three trees wererandomly selected from each category for shoot samples.From each tree, shoot samples were taken at three heights:top (T), middle (M), and low (L), and nine shoots werecollected at each height of a tree, thus creating nine shootclasses: DT, DM, DL, CT, CM, CL, ST, SM, and SL; thisapproach yielded a total of 243 shoots. Samples were takenback to the laboratory for analysis. The clumping withinshoots was quantified using the needle-to-shoot area ratio(c E ) as follows:c E ¼ A n =A sð9Þwhere A n is half the total needle area (including all sides) ina shoot, and A s is half the shoot area.To calculate the half-surface area of Pinus koraiensis(A n ), formula (1) was used; A s was scanned from threeangles, 0°, 45°, and 90°, and formula (8) was used. A p (0°,0°), A p (45°, 0°), and A p (90°, 0°) were 0°, 45°, and 90° ofthe projected area, respectively.A s ¼ 2 cosð15 ÞA p ð0 ; 0 Þþcosð45 ÞA p ð45 ; 0 Þþcosð75 ÞA p ð90 ; 0 Þcosð15 Þþcosð45 Þþcosð75 Þð10Þ12326

J For ResData processing and analysisThis study area is a mixed broadleaf–conifer forest, wedetermine LAI using litter collection and corrected hemisphericalphotography before and after all deciduous leaveshad fallen (the intersect was about in early November),respectively, so we regard LAI, sum of LAI in November 1(corrected by a, X E , and c E ) and LAI from leaf litter collectionin each period (litter traps emptied seven times duringthe whole period from early November to early August) asthe ‘‘true’’ LAI for each period. Regression analysis was alsocarried out between ‘‘true’’ LAI and effective LAI (calculatedonly from hemispherical photography) in each period.WinSCANOPY2006 was used to derive clumping indexbetween shoots and effective LAI based on the Beer-Lambert-Campbell ellipsoid distribution algorithm(Campbell 1986). We used Duncan’s multiple comparisontest to compare needle-to-shoot area ratio in differentlayers and different LAIs for each period. A paired-samplest test was used to compare the ‘‘true’’ LAI and correctedLAI from the classified model for each period. All statisticalanalyses used were part of SPSS 15.0.ResultsMain parametersDifferent species had different SLA (Table 1). The SLA ofleaves was generally large, and of the needles was relativelysmall.Table 1 Specific leaf area (SLA) for each tree speciesTree speciesSLA (cm 2 /g)Pinus koraiensis 79.00 ± 2.84Abies nephrolepis 70.96 ± 2.42Picea spp. (Picea koraiensis and P. jezoensis) 49.84 ± 4.21Ulmus japonica 212.42Betula costata 197.44Fraxinus mandshurica 385.96Acer tegmentosum 241.28Quercus mongolica 280.05Populus ussuriensis 125.78Acer ukurunduense 378.99Ulmus laciniata 300.16Acer mono 315.16Tilia amurensis 163.30Tilia mandshurica 354.49Prunus padus 123.95Corylus mandshurica 382.94Others 326.77Figure 1 shows two images, one is the original and theother is the photo after masking of stems and branches. Ascan be seen by comparing the photographs, the stems andbranches were essentially completely removed. LAI wasestimated from hemispherical photography in November(before and after masking of stems and branches). Thiscomputer-adjusted LAI was smaller than the effective LAI,as expected (Fig. 2). Among all the sites, the maximumdifference between the computer-adjusted LAI and theeffective LAI was 0.39, the minimum difference was 0.03,and the average difference was 0.11 ± 0.060. And theaverage woody-to-total area ratio (a) was 0.096, the maximumwas 0.378, the minimum was 0.024.X E was calculated by the CC method. The maximumvalue was 0.89, the minimum 0.46, and the average0.69 ± 0.11. After the correction both by X E and a, themaximum LAI value was 2.25, the minimum 0.99, and theaverage 1.55 ± 0.30.Table 2 presents values of c E for nine cases (see‘‘Materials and Methods’’). DT had the maximum value of2.37, ML and SM both had a minimum value of 1.59, andthe average was 1.77 ± 0.37, which was within the rangeof 1.4–1.8 for published estimates of c E for tree species(Chen et al. 1997). The dominant trees generally had thelargest values, followed by co-dominant and suppressedtrees. c E was generally larger at higher levels, which perhapsreflects differences in vegetation types and growthconditions (Chen et al. 2006). Duncan’s multiple comparisontest was applied separately to the same forest layer andsame tree height of Table 2. For each forest layer, therewere significant differences between T and M and no significantdifferences between M and L. In the same treeheight, DT was significantly different from CT and ST, butthere were no significant differences among M orL. According to the test results, we then separated thelayers into two categories: dominant (tree heightC25 m)and non-dominant (tree height \25 m, includingco-dominant and suppressed). c E value of each site wasdetermined according to the percentages of dominant, nondominantPinus koraiensis within a site. Ultimately, theaverage c E value of 64 points ranged from 1.70 to 1.90.LAI estimation modelsThe average effective LAI taken by hemispherical photographsin early November was 1.15 ± 0.10, and the ‘‘true’’LAI (after three corrections of a, X E , c E ) was 2.79 ± 0.56.Figure 3 presents regression analysis plots of effective LAIversus ‘‘true’’ LAI for each period and for the entire litterfallseason. Considering fit, complex, theoretical and ecologicalsignificance of the models, we selected the powerfunction models. The R 2 value changed between 0.5465and 0.6351 for all periods. Clearly, the ‘‘true’’ LAI and12327

J For ResFig. 1 Representativehemispherical photographbefore (a) and after (b) use ofPhotoshop to remove stems andbranchesa Original photob Photo after masking of stemsand branchesLAI (branches and stems removed)1.510.50.5 1 1.5Effective LAIFig. 2 Comparison of LAI calculated before and after masking ofstems and brancheseffective LAI changed over time. LAI was maximal onJuly 1 and minimal on November 1. LAI decreasedfrom July 1 to mid-September, albeit slowly. From mid-September to early October, LAI decreased rapidly—theaverage ‘‘true’’ LAI decreased from 5.28 to 3.20, indicatingthat the rate of needle/leaf fall was maximal from mid-September to early October. In July, the ‘‘true’’ LAI westudied ranged from 3.90 to 10.52, with an average of6.69. Figure 3h shows the entire falling period ofeffective LAI versus ‘‘true’’ LAI (R 2 = 0.8351). Themaximum ‘‘true’’ LAI for the entire litter-fall season was10.52, the minimum was 1.79, and the average was 4.79.From July to November, effective LAI underestimated‘‘true’’ LAI by 52.61, 53.65, 57.65, 57.68, 44.80, and58.90 %, respectively, and for the entire litter-fall season,by 53.16 %.Figure 4 shows LAI dynamics of ‘‘true’’ LAI, effectiveLAI, corrected LAI for each period, and corrected LAI forthe entire litter-fall season. Data for each LAI type indicatedthat LAI is dynamic over time. The maximum ‘‘true’’LAI was 6.69 ± 1.37 and the minimum was 2.79 ± 0.56,reflecting significant seasonal dynamics.The ‘‘true’’ LAI dropped significantly after mid-September (see Fig. 3c, d). Effective LAI was smallerby comparison, with a maximum of 3.17 ± 0.43 andminimum of 1.15 ± 0.22, and this LAI exhibited lesserseasonal dynamics. We applied Duncan’s multiple comparisontests to these different LAIs for each period. LAIvalues corrected via the correction model for each periodwere consistent with the corresponding ‘‘true’’ LAI values.There were no significant differences between LAI valuescorrected for the entire litter-fall season versus the ‘‘true’’LAI calculated for July and August, whereas the valuescalculated after September were significantly different, andthe LAI was underestimated in September and Novemberand overestimated in October. There were significant differencesbetween effective LAI and ‘‘true’’ LAI in eachperiod.Corrected LAIs for each period were very close to the‘‘true’’ LAIs, but there were so many models (Fig. 3a–g)and the calibration process was so complicated that themodels were not suitable for calibrating seasonaldynamics of LAI. On the other hand, there was a singlecorrection model (Fig. 3h) for the entire season, and thecalibration process was simple; however, there were somedifferences between the corrected LAIs and ‘‘true’’ LAIs.Upon consideration of advantages and disadvantages ofthe correction models for each period and for the entireseason, we combined the two methods. First, we combinedthe data for July and August and carried out apaired-samples t test at the 95 % significance level todetermine whether there were significant differencesbetween ‘‘true’’ LAI and effective LAI. The idea was that,if there were no significant differences, we would combinethe data of July, August, and early September, andcarry out a paired-samples t test again; if there weresignificant differences, we would combine the data of Julyand early September, and make paired-samples t test. By12328

J For ResTable 2 Needle-to-shoot area ratio (c E )ofPinus koraiensis in the Xiaoxing’an MountainsTop (T) Middle (M) Low (L) AverageDominant (D) 2.37 ± 0.48 aA 1.69 ± 0.20 aB 1.64 ± 0.19 aB 1.77 ± 0.37Co-dominant (C) 1.92 ± 0.40 bA 1.69 ± 0.25 aB 1.59 ± 0.23 aBSuppressed (S) 1.81 ± 0.37 bA 1.59 ± 0.20 aB 1.63 ± 0.16 aABShoot samples (243 in total) were taken from three layers: one dominant (D), one co-dominant (M) and one suppressed (S), at three heights: top(T), middle (M) and bottom (L), forming nine classes: DT, DM, DL, CT, CM, CL, ST, SM, and SLDifferent lowercase letters in the same column and different uppercase letters in the same row indicate a significant difference (P \ 0.05);different lowercase letters within each period indicate a significant difference (P \ 0.05)True LAI1210864201210864206543210July 1 August 1y = 1.6745x 1.1918 a1210y = 1.8243x 1.1405R 2 = 0.59298 R 2 = 0.54776420September 1y = 1.9635x 1.2149R 2 = 0.6351October 1y = 1.6208x 1.1795R 2 = 0.5877November 165g4321y = 2.1837x 1.6923R 2 = 0.546500 1 2 3 4 5ce1210864206543210121086420Effective LAISeptember 15y = 1.9849x 1.1717R 2 = 0.5795July1-November 1y = 2.0499x 1.0462R 2 = 0.8351October 15y = 1.7344x 1.1691R 2 = 0.58390 1 2 3 4 5Fig. 3 Regression analysis between effective LAI (evaluated fromhemispherical photographs) and ‘‘true’’ LAI in each period and duringthe whole deciduous season, as indicated in the panelsLAI9876543210a a aba a acTrue LAICorrected LAI through each period modelCorrected LAI through entire seasonal modelEffective LAIa abca ab7-1 8-1 9-1 9-15 10-1 10-15 11-1Datecab bca b bca abFig. 4 Seasonal dynamics of LAI from July to November in theXiaoxing’an Mountainscbdfhanalogy, we continued this iteration until we had clearlyput data with no significant differences together and significantdifferences apart. The regression model was madebetween the data of ‘‘true’’ LAI and effective LAI with nosignificant differences.After combining the two methods, we classified thecalibration models into two types: (1) maximum LAI period(July, August) and the period of maximal leaf fall(October), and (2) the periods when leaves started falling(September) and when all deciduous leaves had fallen(November). Figure 5 shows the results of this classificationas well as the calibration models, and Fig. 6 shows theseasonal dynamics of ‘‘true’’ LAI and corrected LAI fromthe classified models. From Figs. 5 and 6, we can see that‘‘true’’ LAI and corrected LAI from the classified modelswere in good agreement.DiscussionMeasurement of main parametersLarge changes of SLA have happened among differentspecies, especially between leaves and needles, and haveindicated that it was more reasonable to measure LAI byseparating tree species than mixing them for litter collectionmethod.a values changed in a wide range when effective LAIwas below 1.1, which was mainly because of the characteristicsof the forest type. When effective LAI was large,coupled with large leaf area, more leaves appeared in thecanopy, and many branches and stems were blocked out nomatter whether they were more or less, so the differencebetween effective LAI and computer-adjusted LAI wassmall. But when effective LAI was small, few branches andstems were blocked out, and the change of branches andstems area will cause the large variation of a. However, awas consistent with the results of Chen (1996), who hasproved that a for tree species is within the range of0.03–0.32. This method for estimating a is simple and alsoa good choice for estimating LAI in regions where treecutting is forbidden.12329

J For ResFig. 5 Results of the modelclassification process and thecalibration modelsTrue LAIJuly + August + OctoberSeptember +November12108y = 1.6627x 1.2053R 2 = 0.909y = 2.3491x 1.0011R 2 = 0.869264200 1 2 3 4 5 0 1 2 3 4 5Effective LAILAI86420True LAICorrected LAI through classified model7-1 8-1 9-1 9-15 10-1 10-15 11-1DateFig. 6 Seasonal dynamics of ‘‘true’’ LAI as well as the LAI valuesafter correction using the classified modelsOur study area is in the mixed broadleaved–Korean pineforest, which itself has spatial heterogeneity to which bothbiotic and abiotic environments and both characteristics oftree species and response of tree species to different disturbanceagents may belong, causing the large variability(0.46–0.89) of the clumping index in different sites. c E hada significant effect on retrieved LAI.In this study, almost no deciduous leaves existed onNovember 1, the great majority being needles of Pinuskoraiensis, Picea spp. and Abies nephrolepis, but only c E forPinus koraiensis was used for correction. This is mainlybecause, firstly, it was difficult to distinguish different needleson the hemispherical photographs, and, in our study,Picea spp. and Abies nephrolepis needles do not have shootslike Pinus koraiensis, so we estimated that their within-shootclumping was smaller than that of Pinus koraiensis. And thetotal basal area (m 2 /ha) of both Picea spp. and Abies nephrolepisaccount for only 13.5 % of the total of tree conifers,and the summed LAI from litter collection of the two accountfor 13 % of the total of three conifers. According to theseinformation, we estimated that error in resultant LAI causedby using the c E value of Pinus koraiensis for all three coniferswas no more than 6 %. For these reasons, we did notdifferentiate the differences of the three conifers on thehemispherical photographs in November. However, the c Evalue of each site was determined according to the percentagesof dominant, co-dominant, and suppressed treeswithin a site and is better than a unique c E used for allsites, though the range of c E was not so large in our studyarea.In a study of a coniferous forest, Chen (1996) proposedthat optical measurements combined with shoot sampleanalysis can produce more accurate LA1 values thandestructive sampling results. Though there were also someother weaknesses, such as only one sample being taken tomeasure SLA for deciduous species, the exposure influencewas ignored when analyzing hemispherical photography,and correction for optical LAI was still an effective way ofestimating LAI under non-destructive forest conditions.LAI estimationOptical methods are widely used to estimate LAI and itsdynamic changes because they are easier and quicker tocarry out. However, the hemispherical photography techniquetends to underestimate LAI. In our study, averageeffective LAI was 53.16 % lower than average ‘‘true’’ LAIfor the entire season, which was consistent with resultspublished by Sommer and Lang (1994) and Brenner et al.(1995), who estimated that LAI obtained from hemisphericalphotographs was up to 50 % lower comparedwith destructive harvest results. So, it is necessary to validateand improve indirect optical methods.Many studies have been reported on LAI estimation indeciduous and evergreen coniferous forest. For evergreenconiferous forest, Eq. 7 was used to obtain more accurateLAI from effective LAI, which has been widely accepted(Chen et al. 1997; Thimonier et al. 2010). And litter collectionwas the most common method for estimating LAIin deciduous forest (Neumann et al. 1989; Cutini et al.1998). However, few studies on more reasonable andaccurate LAI estimation as well as its seasonal dynamics inmixed conifer and broadleaved forest have been reported.In this study, the forest after November could be consideredas evergreen coniferous forest and from July toNovember as deciduous forest. We obtained the ‘‘true’’LAI by combining the methods of the two types of forest,which should be objective and reasonable.12330

J For ResDifferent period models were merged into two classifiedmodels not only because of the data themselves but also onbiological and ecological bases. This is possible because ofthe relationship of ‘‘true’’ LAI and effective LAI. All theleaves/needles grew to their peaks in July and August, andwith clumping in the canopy, effective LAI had somedegree of underestimation. In September, the extent ofunderestimation increased, because most leaves/needlesbegan to fall, and the increase might be caused by theincrease of the gap fraction. But the extent of underestimationreduced in October, when the decrease of clumpingplayed a more important role than the increase of gapfraction due to most leaves/needles having fallen. InNovember, almost all leaves had fallen, and the increasedextent of underestimation was mainly caused by theclumping of needles. So the effective LAI in July, August,and October had a similar lower extent of underestimation(\53.65 %), while in September and November it had asimilar higher extent of underestimation ([57.65 %). Andthe models in the months with similar extents of underestimationcould be linked together. Overall, the seasonalchange characteristics of LAI could exist not only in ourstudy area but in other similar mixed forests even indeciduous forest types.The two classified models are able to fully explain theseasonal variations of litter fall period, greatly reducingworkload. Objectively, the classified models could not onlybe used in our study area and similar forest types but themethod proposed in our study could also provide a referencefor related studies and subsequent research, because it can beused in other forest types to detect changes in the relationshipbetween indirect LAI and direct LAI, and to explore thechange regulation of different models over the whole year.Acknowledgments This work was financially supported by theMinistry of Science and Technology of China (No. 2011BAD37B01),the National Natural Science Foundation of China (No. 30770350),and the Program for Changjiang Scholars and Innovative ResearchTeam in University (IRT1054). We thank Mr. J.M. Chen for providingus with the technology to collect and analyze the data.ReferencesBehera SK, Srivastava P, Pathre UV, Tuli R (2010) An indirectmethod of estimating leaf area index in Jatropha curcas L. usingLAI-2000 Plant Canopy Analyzer. Agric For Meteorol150:307–311Bonan GB (1993) Importance of leaf area index and forest type whenestimating photosynthesis in boreal forests. Remote Sens Environ43:303–314Brenner A, Cueto Romero M, Garcia H (1995) A comparison of directand indirect methods for measuring leaf and surface areas ofindividual bushes. Plant Cell Environ 18:1332–1340Campbell GS (1986) Extinction coefficients for radiation in plantcanopies calculated using an ellipsoidal inclination angle distribution.Agric For Meteorol 36:317–321Chen JM (1996) Optically-based methods for measuring seasonalvariation of leaf area index in boreal conifer stands. Agric ForMeteorol 80:135–163Chen JM, Black TA (1992) Defining leaf area index for non-flatleaves. Plant Cell Environ 15:421–429Chen JM, Cihlar J (1995) Plant canopy gap-size analysis theory forimproving optical measurements of leaf-area index. Appl Optics34:6211–6222Chen JM, Cihlar J (2002) Quantifying the effect of canopyarchitecture on optical measurements of leaf area index usingtwo gap size analysis methods. IEEE Trans Geosci Remote Sens33:777–787Chen JM, Black TA, Adams RS (1991) Evaluation of hemisphericalphotography for determining plant area index and geometry of aforest stand. Agric For Meteorol 56:129–143Chen JM, Rich PM, Gower ST, Norman JM, Plummer S (1997) Leafarea index of boreal forests: theory, techniques, and measurements.J Geophys Res 102:29429–29443Chen JM, Govind A, Sonnentag O, Zhang Y, Barr A, Amiro B (2006)Leaf area index measurements at Fluxnet-Canada forest sites.Agric For Meteorol 140:257–268Cutini A, Matteucci G, Mugnozza GS (1998) Estimation of leaf areaindex with the Li-Cor LAI 2000 in deciduous forests. For EcolManag 105:55–65Eriksson H, Eklundh L, Hall K, Lindroth A (2005) Estimating LAI indeciduous forest stands. Agric For Meteorol 129:27–37Fukushima Y, Hiura T, Tanabe S (1998) Accuracy of the MacArthur-Horn method for estimating a foliage profile. Agric For Meteorol92:203–210Gazarini LC, Araujo MCC, Borralho N, Pereira JS (1990) Plant areaindex in Eucalyptus globulus plantations determined indirectlyby a light interception method. Tree Physiol 7:107–113Gonsamo A, Pellikka P (2009) The computation of foliage clumpingindex using hemispherical photography. Agric For Meteorol149:1781–1787Gower ST, Kucharik CJ, Norman JM (1999) Direct and indirectestimation of leaf area index, f (APAR), and net primary productionof terrestrial ecosystems. Remote Sens Environ 70:29–51Ishihara MI, Hiura T (2011) Modeling leaf area index from littercollection and tree data in a deciduous broadleaf forest. AgricFor Meteorol 151:1016–1022Jin GZ, Xie XC, Tian YY, Kim JH (2006) The pattern of seed rain inthe broadleaved-Korean pine mixed forest of XiaoxinganMountains, China. J Korean For Soc 95:621–627Johnson JD (1984) A rapid technique for estimating total surface areaof pine needles. For Sci 30:913–921Jonckheere I, Fleck S, Nackaerts K, Muys B, Coppin P, Weiss M,Baret F (2004) Review of methods for in situ leaf area indexdetermination: part I. Theories, sensors and hemisphericalphotography. Agric For Meteorol 121:19–35Kalácska M, Calvo-Alvarado JC, Sánchez-Azofeifa GA (2005)Calibration and assessment of seasonal changes in leaf areaindex of a tropical dry forest in different stages of succession.Tree Physiol 25:733–744Kucharik CJ, Norman JM, Gower ST (1998) Measurements of brancharea and adjusting leaf area index indirect measurements. AgricForest Meteorol 91:69–88Lang ARG, Xiang Y (1986) Estimation of leaf area index fromtransmission of direct sunlight in discontinuous canopies. AgricFor Meteorol 37:229–243Leblanc SG (2002) Correction to the plant canopy gap-size analysistheory used by the tracing radiation and architecture of canopiesinstrument. Appl Optics 41:7667–7670Leblanc SG, Chen JM (2001) A practical scheme for correctingmultiple scattering effects on optical LAI measurements. AgricFor Meteorol 110:125–13912331

J For ResLeblanc SG, Chen JM, Fernandes R, Deering DW, Conley A (2005)Methodology comparison for canopy structure parametersextraction from digital hemispherical photography in borealforests. Agric For Meteorol 129:187–207Levy PE, Jarvis PG (1999) Direct and indirect measurements of LAIin millet and fallow vegetation in HAPEX-Sahel. Agric ForMeteorol 97:199–212Mussche S, Samson R, Nachtergale L, De Schrijver A, Lemeur R,Lust N (2001) A comparison of optical and direct methods formonitoring the seasonal dynamics of leaf area index indeciduous forests. Silva Fenn 35:373–384Neumann HH, Den Hartog G, Shaw RH (1989) Leaf area measurementsbased on hemispheric photographs and leaf-litter collectionin a deciduous forest during autumn leaf-fall. Agric ForMeteorol 45:325–345Nilson T (1971) A theoretical analysis of the frequency of gaps inplant stands. Agric Meteorol 8:25–38Oker-Blom P, Smolander H (1988) The ratio of shoot silhouette areato total needle area in Scots pine. For Sci 34:894–906Pielou EC (1962) Runs of one species with respect to another intransects through plant populations. Biometrics 18:579–593Running SW, Gower ST (1991) FOREST-BGC, a general model offorest ecosystem processes for regional applications. II. Dynamiccarbon allocation and nitrogen budgets. Tree Physiol 9:147–160Ryu Y, Sonnentag O, Nilson T, Vargas R, Kobayashi H, Wenk R,Baldocchi DD (2010) How to quantify tree leaf area index in anopen savanna ecosystem: a multi-instrument and multi-modelapproach. Agric For Meteorol 150:63–76Sommer K, Lang A (1994) Comparative analysis of two indirectmethods of measuring leaf area index as applied to minimal andspur pruned grape vines. Funct Plant Biol 21:197–206Sonnentag O, Talbot J, Chen JM, Roulet NT (2007) Using direct andindirect measurements of leaf area index to characterize theshrub canopy in an ombrotrophic peatland. Agric For Meteorol144:200–212Sprintsin M, Cohen S, Maseyk K, Rotenberg E, Grunsweig J, KarnieliA, Berliner P, Yakir D (2011) Long term and seasonal courses ofleaf area index in a semi-arid forest plantation. Agric ForMeteorol 151:565–575Thimonier A, Sedivy I, Schleppi P (2010) Estimating leaf area indexin different types of mature forest stands in Switzerland: acomparison of methods. Eur J For Res 129:543–562Walter JMN, Himmler GG (1996) Spatial heterogeneity of a Scotspine canopy: an assessment by hemispherical photography. CanJ For Res 26:1610–1619Wang YS, Miller DR (1987) Notes: calibration of the hemisphericalphotographic technique to measure leaf area index distributionsin hardwood forests. For Sci 33:210–216Weiss M, Baret F, Smith GJ, Jonckheere I, Coppin P (2004) Reviewof methods for in situ leaf area index (LAI) determination: partII. Estimation of LAI, errors and sampling. Agric For Meteorol121:37–53Whitford KR, Colquhoun IJ, Lang ARG, Harper BM (1995)Measuring leaf area index in a sparse eucalypt forest: acomparison of estimates from direct measurement, hemisphericalphotography, sunlight transmittance and allometric regression.Agric For Meteorol 74:237–249Zou J, Yan G, Zhu L, Zhang W (2009) Woody-to-total area ratiodetermination with a multispectral canopy imager. Tree Physiol29:1069–108012332

OecologiaDOI 10.1007/s00442-012-2348-2COMMUNITY ECOLOGY - ORIGINAL RESEARCHEffects of local biotic neighbors and habitat heterogeneity on treeand shrub seedling survival in an old-growth temperate forestXuejiao Bai • Simon A. Queenborough • Xugao Wang •Jian Zhang • Buhang Li • Zuoqiang Yuan • Dingliang Xing •Fei Lin • Ji Ye • Zhanqing HaoReceived: 22 March 2011 / Accepted: 23 April 2012Ó Springer-Verlag 2012Abstract Seedling dynamics play a crucial role in determiningspecies distributions and coexistence. Exploringcauses of variation in seedling dynamics can therefore providekey insights into the factors affecting these phenomena.We examined the relative importance of biotic neighborhoodprocesses and habitat heterogeneity using survival data for5,827 seedlings in 39 tree and shrub species over 2 years froman old-growth temperate forest in northeastern China. Wefound significant negative density-dependence effects onsurvival of tree seedlings, and limited effects of habitatCommunicated by Walt Carson.Electronic supplementary material The online version of thisarticle (doi:10.1007/s00442-012-2348-2) contains supplementarymaterial, which is available to authorized users.X. Bai X. Wang B. Li Z. Yuan D. Xing F. Lin J. Ye Z. Hao (&)State Key Laboratory of Forest and Soil Ecology,Institute of Applied Ecology, Chinese Academy of Sciences,Shenyang 110016, People’s Republic of Chinae-mail: hzq@iae.ac.cnX. Bai B. Li D. XingGraduate University of Chinese Academy of Science,Beijing 100049, ChinaS. A. QueenboroughNational Centre for Ecological Analysis and Synthesis,Santa Barbara, CA 93101, USAPresent Address:S. A. QueenboroughDepartment of Evolution, Ecology and Organismal Biology,The Ohio State University, Columbus, OH 43210, USAJ. ZhangDepartment of Renewable Resources, University of Alberta,Edmonton, AB T6G 2R3, Canadaheterogeneity (edaphic and topographic variables) on survivalof shrub seedlings. The importance of negative densitydependence on young tree seedling survival was replaced byhabitat in tree seedlings C4 years old. As expected, negativedensity dependence was more apparent in gravity-dispersedspecies compared to wind-dispersed and animal-dispersedspecies. Moreover, we found that a community compensatorytrend existed for trees. Therefore, although negative densitydependence was not as pervasive as in other forest communities,it is an important mechanism for the maintenance ofcommunity diversity in this temperate forest. We concludethat both negative density dependence and habitat heterogeneitydrive seedling survival, but their relative importancevaries with seedling age classes and species traits.Keywords Negative density dependence Nichepartitioning Seedling dynamics Communitycompensatory trend Generalized linear mixed modelsIntroductionModels of species coexistence in diverse communitiesgenerally emphasize density-dependent survival, nichepartitioning, or ecological equivalence (Chesson 2000;Nakashizuka 2001; Wright 2002). These three mechanismshave each received some support in studies of tree communities.For example, recent research has identified strongdensity dependence in temperate forests (HilleRisLambersand Clark 2003; HilleRisLambers et al. 2002). Furthermore,density-dependent mechanisms acting within localbiotic neighborhoods can increase the survival of locallyrare species in subtropical and tropical forests (Chen et al.2010; Comita and Hubbell 2009; Hubbell et al. 2001;Queenborough et al. 2007b; Uriarte et al. 2004b).12333

OecologiaSimilarly, important dimensions of plant trait variationwithin tree communities are well known (Kraft et al. 2008;Martínez-Vilalta et al. 2010; Wright et al. 2007), and abioticniche partitioning may be caused by trait-related differentialpatterns of distribution and demography(Cavender-Bares et al. 2004; Comita et al. 2007b; Harmset al. 2001; McMahon et al. 2011; Queenborough et al.2007a; Russo et al. 2005; Tateno and Takeda 2003). Thenull model for evaluating either negative density dependence(NDD) or habitat niche partitioning is the ecologicalequivalence of coexisting species. In this case, dispersallimitation and stochastic ecological drift therefore governlocal community dynamics (Hubbell 2001).Although researchers are now generally convinced thatNDD and niche differentiation are not mutually exclusive(Queenborough et al. 2009), these processes are usuallytested separately. For tree communities, studies of seedlingsurvival as a function of the local biotic neighborhood haveencountered evidence of NDD (Chen et al. 2010; Comitaand Hubbell 2009; Gilbert et al. 2001; Packer and Clay2000; Queenborough et al. 2007b). Other studies haveexamined survival and growth as a function of the localabiotic environment (Russo et al. 2005; Tsujino et al.2006). However, few studies have so far investigated therelative importance of both biotic and abiotic drivers ofseedling survival in a single analysis (Comita et al. 2009;Paine et al. 2011; Queenborough et al. 2009; Shibata et al.2010; Streng et al. 1989). In this study, we aimed todetermine the relative importance of biotic and abioticdrivers of seedling survival for a relatively diverse temperatetree community in northeastern China.As well as individual-level effects, NDD can also haveconsequences at the community level. The neighbourhoodsof common species will be more likely to contain a conspecific,and these species may experience stronger NDDbecause of their higher frequency relative to less commonspecies. Rare species, therefore, have an advantage in whatConnell et al. (1984) termed a community compensatorytrend (CCT). This is related to, but contrasts with, recentevidence showing that rare species may be more sensitive tothe presence of conspecifics than common species (Comitaet al. 2010). A CCT has been found in several tropical andsubtropical sites (Chen et al. 2010; Queenborough et al.2007b; Webb and Peart 1999), although in other sites it hasnot (He et al. 1997; Welden et al. 1991). No study has yettested for a CCT in temperate forest.The current paucity of community-level seedling studiesfrom temperate forests is a major limitation in understandingwhether different mechanisms of coexistenceoperate in temperate compared to tropical forests. Giventhe incredible difference in diversity between these foresttypes, exploring potential drivers of this difference is animportant ecological question (Kraft et al. 2011). Such dataare critical for understanding the early stages of communityassembly that shape temperate forest diversity patterns, andmay also reveal similarities or differences between temperateand tropical forest community assembly.One potential cause of the plethora of hypotheses formechanisms of species coexistence is that the mechanismmay change throughout an organism’s life history. Differentfactors become important at different times and have differenteffects (Comita et al. 2007a; Grubb 1977). Furthermore,survival generally increases with size (Uriarte et al. 2004a;Winkler et al. 2005), because larger individuals may be moreresistant and resilient to biotic and abiotic stresses. In addition,differences among species in terms of functional traitssuch as dispersal agent will also affect how individualsrespond to these pressures (Ramaswami and Sukumar 2011;Streng et al. 1989). For instance, animal-dispersed (especiallybirds) species may escape from high mortality caused byNDD because some animals, such as frugivorous birds coulddisperse seeds over several tens or hundreds of meters (Iidaand Nakashizuka 1998; Murray 1988). Comparatively,gravity-dispersed species may be more affected by NDD dueto high seedling density close to the adults. Understanding thecomplex effects of this heterogeneity in the forest ecosystemis essential to clarify the potential drivers of species coexistencein the early stages of community assembly.Our study, therefore, addresses these issues by examiningthe relative importance of biotic neighbors and habitatheterogeneity for seedling survival over 2 years, usinga dataset of 5,827 seedlings in 39 tree and shrub species inan old-growth temperate forest. In particular, we sought toanswer the following questions:1. Do local biotic neighbors or habitat heterogeneity havea greater effect on the survival of tree and shrubseedlings in this temperate forest? Specifically, does anincrease in conspecific neighbors decrease the probabilityof survival to a greater or lesser extent than adecrease in, for example, soil nutrients?2. Do the effects of biotic neighbors and habitat (edaphicand topographic) variables on survival differ amongseedling age classes, dispersal-mode groups or species?Specifically, do older seedlings ([1 year) sufferlower density-dependence than new recruits?3. Is there evidence of a community compensatory trend,and is this present for both tree and shrub species?Materials and methodsStudy siteThe study is located in Changbai Nature Reserve (42°23 0 N,128°05 0 E) in northeastern China. The reserve was12334

Oecologiaestablished in 1960 and joined the World BiosphereReserve Network under the Man and the Biosphere Projectin 1980 (Shao et al. 1994; Stone 2006). The reserve isabout 200,000 ha with elevation ranging from 740 to2,691 m. There are five typical vertical vegetation zones:aspen-white birch (Populus davidiana and Betulaplatyphlla) forest, broad-leaved Korean pine (Pinus koraiensis)mixed forest, spruce-fir (Picea jezoensis andAbies nephrolepis) forest, subalpine birch (B. ermanii)forest, and alpine tundra (Yang et al. 1985).Our study site is situated in the old-growth broad-leavedKorean pine mixed forest, which is the dominant vegetationtype in northeastern China. It has high biodiversity,complex stand structure, and unique species composition(Wang et al. 1980). The climate in the forest region ischaracterized by low temperatures, high precipitation, andstrong winds, with the prevailing direction of west-southwest(Yang et al. 1985). Mean annual temperature is 3.3 °C(-16.5 °C in January and 20.5 °C in August). Mean annualprecipitation is 672 mm, most of which occurs betweenJune and September (480–500 mm) (Yang et al. 1985).Mean age of overstory trees is about 300 years.In 2004, a 25-ha (500 9 500 m) forest plot was established,which was chosen in the core zone of the reserve inorder to avoid human disturbances (Hao et al. 2007).Within the plot, all individuals with diameter at breastheight (DBH, 1.3 m above the ground) C1 cm were mapped,tagged, and identified. In the first census in 2004, therewere 38,902 living stems, belonging to 52 species, 32genera, and 18 families. Mean stand density of living treeswas 1,556 trees per ha and mean basal area of living treeswas 43.2 m 2 per ha (Hao et al. 2007, 2008).Seedling censusWe set up 150 seed traps within the plot to monitor longtermdynamics of seed production and dispersal (see Zhanget al. 2008). Around each seed trap, we set up four seedlingplots (5 9 5 m) in August and September 2006 to monitorlong-term seedling dynamics (ESM S1, n = 600). Allseedling plots were spaced at least 8 m apart. The totalsampling area of 600 seedling plots is 1.5 ha. In eachseedling plot, all tree seedlings with DBH \1 cm, shruband liana seedlings with DBH \1 cm and height C30 cmwere tagged, mappedn and identified. The age of each treeseedling was estimated by counting annual bud scale scars.All seedling plots were recensused in August and September2007, and August and September 2008.Biotic neighborhood variablesWe calculated the seedling and adult neighbors of eachfocal seedling. We calculated the density of conspecificand heterospecific seedling neighbors within 0.5 m of focalseedlings. Seedlings within 0.5 m of the seedling plot edgewere not included as focal seedlings to eliminate theboundary effects (ESM S2). The effect of adult neighborswas based on adult size modified by the distance betweenthe neighbor and the focal seedling. Using the 2004 plotcensus data (trees and shrubs with DBH C1 cm), we calculatedthe sum of the basal area (BA) of conspecific andheterospecific adults within 20 m from the focal seedlingdivided by the distance of each adult from the focal seedling(Canham et al. 2004):A ¼ XN iBA i =DISTANCE iwhere i is an individual adult. To eliminate boundaryeffects, only those seedlings with a distance greater than orequal to 20 m from the 25-ha plot edges were included inour analyses.Habitat variablesWe defined a seedling’s habitat variables in terms of soilproperties and topography. For the soil properties, wesampled soils within the plot every 30 m on a regular grid.At each grid point, two additional sample points at 2, 5n or15 m were selected in a random compass direction fromthe grid point. In total, 967 soil samples were sampled inthe 25-ha plot (Yuan et al. 2011). Eight soil properties wererecorded: pH, organic matter content, available nitrogen(N), total N, available phosphorus (P), total P, availablepotassium (K)n and total K. We interpolated the soilvariables to 5 9 5 m grid by kriging, and converted thecalculated values for each seedling plot to z scores (Johnet al. 2007). To reduce multi-collinearity and the number ofvariables describing soil factors, we performed a principalcomponents analysis (PCA) on these eight soil chemicalvariables. The first two components produced by the PCAaccounted for 86.7 % of the variance in the eight soilvariables (ESM S3). PCA axis 1 was associated with highorganic matter, available N and total N, P. PCA axis 2 wasassociated with low available N, P and total N, P.Three topographic factors were identified: slope, aspectnand elevation. Elevation was measured by total station atthe four corners of a 20 9 20 m grid in the 25-ha plot. Wethen calculated the elevation to a 5 9 5 m grid using kriginginterpolating methods. Slope, aspect and elevationvalues were calculated for each seedling plot. Slope wasdefined as the single average angle from the horizontal ofthe entire quadrat. Aspect referred to the direction to whichthe slope faced. Elevation was defined as the mean elevationof the four corners. Plot convexity was not usedbecause it was highly correlated with elevation.12335

OecologiaStatistical analysisWe modeled the survival probability of individual seedlingsfrom 2006 to 2008 as a function of biotic neighborhoodand habitat variables (Table 1), using generalizedlinear mixed-effects model (GLMM). The GLMM in thispaper was essentially a logistic regression, with theresponse variable as a logit transformation of seedlingstate: 1 (alive) or 0 (dead). All the values of the continuousexplanatory variables were standardized by subtracting themean value of the variable (across all individuals in theanalysis) and dividing by 1 standard deviation.To test the relative importance of biotic and habitatvariables, we compared the following four candidatemodels: (1) a null model only with random effects, (2) abiotic model in which the fixed effects of seedling and adultneighbors were added to the null model, (3) a habitat modelin which the fixed effects of soil and topography were addedto the null model, and (4) a biotic ? habitat model in whichthe fixed effects of all variables were added to the nullmodel. Models were compared using Akaike’s InformationCriterion (AIC), and models with a difference between AICvalues of less than 2 were judged equally valid (Burnhamand Anderson 2002). We examined four subsets of the data:(1) the community level (all tree and shrub seedlingspooled), (2) different age-classes (tree seedlings only), (3)dispersal mode, and (4) species-level analyses for commonspecies [those that occurred in C40 seedling plots, and had[100 (trees) or [130 (shrubs) seedlings in 2006].Finally, we tested whether there was a CCT for both treeand shrub species. We examined the probability of seedlingsurvival as a function of species population size (abundanceor basal area of adults C1 cm DBH in the 25-ha plot)using GLMM (Chen et al. 2010).Seedlings within the same plot are likely to have moresimilar probabilities of survival than those of seedlings indifferent plots, even when considering local biotic andhabitat conditions. To account for this spatial autocorrelationin survival, we included seedling plot as a randomeffect in the GLMMs. We tested for spatial autocorrelationin the residuals of the most likely models by plottingvariograms (see ESM S4). Furthermore, because speciesalso have inherently different survival probabilities, inmodels in which all species were pooled, we also includedspecies as a random effect.We validated the most likely models and CCT modelsusing the area under the receiver operating characteristic(ROC) curve (a standard assessment of model discriminationability for logistic regressions) (See ESM S5), andROC C 0.7 are considered acceptable fits (Hosmer andLemeshow 2000). To measure the partial effect of eachvariable on the odds of survival, we calculated odds ratiosfor each coefficient (the exponential of the estimate of eachcoefficient). Odds ratios [1 indicate positive effects onsurvival, while ratios \1 indicate negative effects.All analyses were carried out in the statistical environmentR (v.2.13.1), using the ‘sqldf’ (Grothendieck 2010)and ‘lme4’ (Bates et al. 2010) packages.ResultsCommunity levelOf the original 5,827 seedlings in 2006, trees and shrubsaccounted for 44.3 and 55.7 % of individuals, respectively.For tree seedlings, 1,385 had died by 2008. For shrubseedlings, 586 had died by 2008.Table 1 Parameters included inmodels of seedling survivalVariablesDataRange Mean MedianTree seedling age 1–21 2.47 1Seedling effect (density of seedlings per m 2 within 0.5 m radii)Conspecific (Ncon) 0–17.825 0.641 0Heterospecific (Nhet) 0–20.372 0.729 0Adult effect (sum of adult basal area m 2 divided by distance within 20 m)Conspecific (Acon) 0–1.328 0.029 0Heterospecific (Ahet) 0.148–13.268 0.521 0.500Soil propertiesSoil PC1 -3.107 to 4.671 -0.113 -1.053Soil PC2 -3.113 to 2.746 -0.085 -0.110Topographic variablesElevation (m) 794.863–808.812 803.517 804.469Slope (%) 0.145–19.075 2.977 2.479Aspect 0.708–359.132 217.073 266.55012336

OecologiaThere were clear differences in the explanatory power ofeach of the four models (null, biotic, habitat, biotic ?habitat) on seedling survival for trees and shrubs (Table 2).The biotic neighborhood model was the best-fit model fortrees, while the null and habitat models were the mostlikely models for shrubs.Odds ratios for the parameters of the most likely modelfor tree seedlings showed a small significant negative effectof neighboring conspecific adults (odds ratio = 0.80,P = 0.007; Fig. 1). There was no negative effect of seedlingneighbors, but there was a slight significant positiveeffect of neighboring heterospecific seedlings (oddsratio = 1.20, P = 0.002; Fig. 1). In shrubs, while thehabitat model (along with the null model) was the mostlikely, the habitat variables themselves had no significanteffect.Age-class levelThe relative importance of biotic neighborhood and habitatvaried greatly with the age classes of tree seedlings(Table 2). For 1-year tree seedlings, the best-fit model wasthe biotic model. For tree seedlings of 2–3 years old, theequally most likely models were the null, biotic and full ones.However, for seedlings in the C4 year age class, only thehabitat model was the most likely. We present coefficientestimates for the biotic model for 1-year tree seedlings, thefullest model for tree seedlings aged 2–3 years, and thehabitat model for tree seedlings C4 years old (Fig. 2).For tree seedlings in the 1-year age class, the only significanteffect was that of heterospecific neighbors (oddsratio = 1.29, P = 0.003; Fig. 2), consistent with the treecommunity level result (Fig. 1). For tree seedlings in the 2-to 3-year age class, conspecific adult neighborhood and soilPC axis 1 showed a significant negative and positive effect,respectively, on focal seedling survival (odds ratio = 0.74,P = 0.020; odds ratio = 2.08, P = 0.014; Fig. 2). For treeseedlings in the C4 year age class, the habitat modelshowed that soil PC axis 1 and elevation had significantpositive effects (odds ratio = 3.79, P = 0.001; oddsratio = 4.22, P = 0.001; Fig. 2). Tree seedlings in theC4 year age class had a greater probability of survival inareas of high organic content, available N and total N, P,and at higher elevation.Table 2 AIC and 4AIC values of individual-level seedling survival modelsInstancesCandidate modelsNull Biotic Habitat Biotic ? habitatAIC 4AIC AIC 4AIC AIC 4AIC AIC 4AICCommunityTree seedlings 2,636.654 9.303 2,627.351 0 2,640.811 13.460 2,630.350 2.999Shrub seedlings 2,912.121 1.095 2,917.487 6.461 2,911.026 0 2,916.515 5.489Age class (tree seedlings)1 year 1,281.274 7.323 1,273.951 0 1,287.720 13.769 1,281.955 8.0042–3 years 707.808 0.607 707.201 0 710.030 2.829 707.425 0.223Above 4 years 489.962 13.056 496.770 19.864 476.906 0 483.601 6.695Dispersal modeWind 3,059.077 1.999 3,061.238 4.160 3,057.078 0 3,058.805 1.727Gravity 2,368.715 16.197 2,352.518 0 2,370.621 18.103 2,354.378 1.860Animal 144.242 11.884 132.358 0 148.264 15.906 137.311 4.953Tree speciesTilia amurensis 128.546 6.836 121.710 0 134.046 12.336 130.749 9.039Fraxinus mandshurica 960.703 0 965.680 4.978 969.719 9.016 974.688 13.985Acer mono 766.449 0 768.009 1.560 771.876 5.427 774.185 7.736Pinus koraiensis 16.902 0 24.470 7.569 22.495 5.593 26.055 9.153Acer psedo-sieboldianum 237.217 0 242.887 5.670 244.156 6.939 249.171 11.954Shrub speciesPhiladelphus schrenkii 1,078.213 0 1,081.080 2.867 1,078.478 0.265 1,081.672 3.459Sorbaria sorbifolia 358.267 0 365.404 7.137 365.951 7.685 372.953 14.686Ribes mandschuricum 203.550 0 207.155 3.605 206.956 3.406 210.159 6.609Spiraea chamaedryfolia 251.911 0 254.888 2.977 256.179 4.268 257.750 5.839Corylus mandshurica 173.603 0 179.144 5.541 175.814 2.211 180.800 7.197The most likely models are shown in bold. See Figs. 2, 3, and 4 for odds ratios of the fullest most likely models37123

OecologiaModel parameterNconNhetAconAhetSoil PC1Soil PC2ElevationSlopeAspectDispersal-mode levelThe factors driving seedling survival varied with dispersalmodegroups (Table 2). For wind-dispersed species, thebest models were the null, habitat, and full ones, withsignificant positive effects of soil PC axis 1 and elevation(Fig. 3). For gravity-dispersed species, the biotic and fullmodels were the most likely, showing significant negativeeffects of both neighboring conspecific seedlings andadults. In animal-dispersed species, the best-fit model wasthe biotic model which included a significant positiveeffect of heterospecific seedlings (Fig. 3).Species level(a) Tree seedlings(b) Shrub seedlings0 1 2 3 4 5 0 1 2 3 4 5Odds ratioFig. 1 Odds ratios of a tree and b shrub seedling survival for thefullest most likely models in Table 2. Circles show odds ratios foreach parameter, with 95 % confidence limits (CL) indicated byhorizontal lines. Odds ratios significantly different from 1 (95 % CLdo not overlap 1) are indicated by filled circles. See Table 1 forvariable abbreviationsTen species had sufficient seedlings to be analyzed separately:five tree and five shrub species (Table 3). Sevenspecies (three tree and four shrub) showed no neighborhoodeffects at all, with the fullest best-fit model being thenull model (Table 2). The three remaining species wereconsistent with the community-level analyses above. InTilia amurensis and Acer mono, both trees, the fullest mostlikely model included the biotic neighbourhood, and inPhiladelphus schrenkii, a shrub, the fullest most likelymodel was the habitat model (Table 2).However, within these models, the effects of bioticneighborhood and habitat were limited (Fig. 4). There wereno significant effects of neighbors on the survival of Tiliaamurensis seedlings, however, there was a significantnegative effect of conspecific neighboring adults on survivalof Acer mono (odds ratio = 0.75, P = 0.029). SoilPC axis 2 showed a significant negative effect on the survivalof Philadelphus schrenki seedlings (oddsratio = 0.73, P = 0.014).Community compensatory trend (CCT)We found a negative relationship between tree seedlingsurvival and basal area (C1 cm DBH) in the 25-ha plot(estimate ± SD =-1.31 ± 0.17, P \ 0.001), indicativeof a CCT (Table 4). However, the effect of species abundanceon survival was not significant. For shrub seedlings,we found no significant effect on survival for either speciesabundance or basal area.DiscussionTo test the relative strength of NDD and habitat nichepartitioning as mechanisms of species coexistence, weexamined the effects of biotic and abiotic neighbourhoodson 2-year survival of 5,827 focal seedlings in 39 species oftemperate tree and shrub in an old-growth temperate forestin northeastern China. Overall, we found significant butlimited effects of both sets of variables. Trees were moreaffected by biotic variables than shrubs, and younger treeseedlings more so than older tree seedlings. There weredifferences in the importance of biotic versus abioticneighbourhoods depending on the seed dispersalFig. 2 Odds ratios of treeseedling survival for threedifferent age classes for thefullest most likely models inTable 2. See Table 1 forvariable abbreviationsModel parameterNconNhetAconAhetSoil PC1Soil PC2Elevation(a) 1 year(b) 2-3 years(c) 4 yearsSlopeAspect0 1 2 3 4 50 1 2 3 4 5 0 1 2 3 4 5Odds ratio12338

OecologiaTable 3 Traits for tree species with C100 focal seedlings and shrub species with C130 focal seedlings in 2006 in the analysisSpeciesGrowthformCanopylayerShade toleranceDispersalmodeNumber of seedlingsin 2006Numberof adultsTilia amurensis Tree Overstory Shade tolerant Gravity 377 2,644Fraxinus mandshurica Tree Overstory Midtolerant Wind 783 694Acer mono Tree Midstory Shade tolerant Wind 717 6,569Pinus koraiensis Tree Overstory Midtolerant Animal 155 2,450Acer psedo-sieboldianum Tree Midstory Shade tolerant Wind 198 4,891Philadelphus schrenkii Shrub Understory Shade tolerant Gravity 1,323 466Sorbaria sorbifolia Shrub Understory Shade tolerant Wind 313 4Ribes mandschuricum Shrub Understory Shade tolerant Gravity 240 1Spiraea chamaedryfolia Shrub Understory Light demanding Wind 215 0Corylus mandshurica Shrub Understory Shade tolerant Gravity 138 7,700Fig. 3 Odds ratios of seedlingsurvival for three dispersalmodegroups for the fullest mostlikely models in Table 2. SeeTable 1 for variableabbreviationsModel parameterNconNhetAconAhetSoil PC1Soil PC2Elevation(a) Wind(b) Gravity(c) AnimalSlopeAspect0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5Odds ratioTable 4 Parameter estimates in GLMMs used to test for effects of population size on seedling survivalCategories Population size (abundance) Population size (basal area)Estimate Standard error Pr([|z|) R 2 (%) Estimate Standard error Pr([|z|) R 2 (%)Tree seedlings -0.212 0.491 0.665 0.3 21.306 0.169 8.97e-15 84.5Shrub seedlings -0.132 0.085 0.12 26.3 -0.050 -0.097 0.604 0.3Significant estimate at P = 0.05 level is shown in boldmechanism. Curiously, analyzing the common speciesindividually, most showed no impact of neighborhoodvariables at all. These results are explored below.Biotic and habitat drivers of survival of treeand shrub seedlingsA prediction of most mechanisms of coexistence is thatperformance is limited more by individuals of one’s ownspecies than other species (Chesson 2000; Connell 1971;Janzen 1970; Volkov et al. 2005). We found a significantnegative effect of conspecific adults on focal seedlingsurvival for all tree seedlings pooled, young tree seedlings(2–3 years) and gravity-dispersed species. This confirmsprevious local-scale findings in temperate forest (HilleRisLambersand Clark 2003; HilleRisLambers et al. 2002),at the larger community-scale, commonly found in subtropicaland tropical forests (Chen et al. 2010; Comita andHubbell 2009; Comita et al. 2009; Metz et al. 2010;Queenborough et al. 2007b). In contrast, we found apositive effect of heterospecific seedlings on survival of alltree seedlings pooled, new recruits (tree seedlings of 1 yearold) and animal-dispersed species. These apparently contradictoryeffects are in line with theory: seedling survivalis limited by conspecifics, particularly large adult treeswhich exert strong asymmetric competition for light andsoil nutrients and also a source of enemies (e.g., insects andpathogens) (Wright 2002). However, survival of all12339

OecologiaFig. 4 Odds ratios of seedlingsurvival of three commonspecies for the fullest mostlikely models in Table 2 (sevenspecies are not shown becausetheir fullest most likely modelsare null models). See Table 1for variable abbreviationsModel parameterNconNhetAconAhetSoil PC1Soil PC2Elevation(a)Tilia amurensis(b) Acer mono(c) PhiladelphusschrenkiiSlopeAspect0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5Odds ratioseedlings is often elevated in areas that are good forseedling growth and survival (Comita and Hubbell 2009).Documenting widespread NDD at these scales in thistemperate forest may be somewhat surprising, given themuch lower densities of seedlings in temperate forestcompared to tropical forest. For example, seedling densitiesin our study are an order of magnitude lower (0.89ind m -2 ) than in subtropical and tropical forests [3.2ind m -2 in Gutian (China) (seedlings with DBH \1 cm;Chen et al. 2010), 2.45 ind m -2 in Panama (seedlings withDBH \1 and C20 cm tall; Comita and Hubbell 2009) andabout 5 ind m -2 in Gunung Palung (Indonesian) (seedlingswith DBH \1 and 5–50 cm tall; Webb et al. 2006)].Therefore, we might expect that seedling–seedling interactionsin temperate forests are weaker than in tropicalforests or even nonexistent, and therefore that mechanismsof species coexistence may also be different (Kraft et al.2011). Indeed, seedling–seedling interactions do seem tobe weaker in our study—we found few significant negativeeffects of conspecific neighboring seedlings, for example,whereas studies from tropical sites frequently report NDDfrom seedling neighbors (Comita and Hubbell 2009;Comita et al. 2009; Metz et al. 2010; Queenborough et al.2007b; Webb et al. 2006), although in other sites this NDDis less obvious (Paine et al. 2008; Svenning et al. 2008).However, NDD from large adults still had a significanteffect and could further contribute to the maintenance ofdiversity in this temperate forest.Shrub seedlings showed no such effects of the bioticneighborhood. This is likely because shrub seedlings werelarger and more robust in our seedling plots (C30 cm inheight, compared to tree seedlings for which there was nominimum cut-off). This interpretation is consistent withresults of tree seedling survival for different age classeswhere the effects of habitat become important as seedlingsbecame older and larger. In contrast to tree seedlings,habitat effects were highlighted in model comparisons forshrub seedlings. However, examination of the coefficientsshowed no significant effects of edaphic or topographicvariables. This result may be partly due to the small spatialvariation in habitat factors in our temperate forest plot.Topography is relatively gentle and the maximum differencein elevation is less than 18 m. Therefore, these habitatfactors have low spatial heterogeneity, and their effect onseedling survival may be weak. Furthermore, in order tocompensate for low light levels in the understory, manyshrub species usually regenerate by vegetative growth(Wilson 1995). Thus, shrub ‘‘seedlings’’ are often supportedexternally and the effects of competitors from thebiotic neighborhood as well as habitat effects that maydecrease survival are not so important. Consequently, otherfactors such as snow cover or physical damage mayinfluence the long-term survival of shrub seedlings.Tree seedling survival of different age classesThe effects of biotic neighborhood tended to be moreimportant for young tree seedlings of 1 year old and2–3 years old, and with increasing age classes, habitatbecame the dominant factor (Fig. 2). This change may bebecause of improved tolerance to fungal infection (Masakiand Nakashizuka 2002), herbivory, pathogens, or abioticstressors through lignification of the stems. Consequently,biotic neighbors were less important with increasing ageclass. Soil PC axis 1 and elevation showed significantpositive effect on survival, which demonstrated that largertree seedlings favored well-drained and suitable soilconditions.Seedling survival of dispersal-mode groupsand common speciesFor those species dispersed by gravity, we found convincingevidence of NDD: seedlings suffered significantnegative effects of both conspecific seedling and adultpresence. This is likely because seedling density around12340

Oecologiaconspecific adults is higher as a result of limited dispersalability. However, for wind-dispersed species, soil PC axis 1and elevation showed significant positive effects on seedlingsurvival. This finding indicates that these species canpotentially escape from mortality caused by NDD as wellas increase the chance of arriving at good recruitment sites(e.g., canopy gaps or suitable soil conditions) whereseedling survival is high. Animal-dispersed species werealso less affected by NDD and their recruitment sites maybe more random, with neighboring heterospecific seedlingsas a significant positive predictor of seedling survival.Most (7/10) of the species for which sample sizes permittedus to examine individually, the null model was thefullest most likely model. This result is more likely becausesample sizes were still too small to find any significanteffect rather than stochasticity dominating forest dynamicsin this system. In three species we did find significanteffects of neighborhood.For the gravity-dispersed species Tilia amurensis, wefound no significant negative effect of conspecifics. Thiscontradiction with gravity-dispersed seedlings treated as agroup may be because the seedling density of Tilia amurensisis low (0.04 ind/m 2 ). We found a significant negativeeffect of conspecific adult trees on the wind-dispersedspecies Acer mono. This may be related to its high treeabundance and wide distribution.For the shrub species Philadelphus schrenkii, habitateffects were more important with a significant negativeeffect of soil PC axis 2. This result may indicate thatPhiladelphus schrenkii is adapted to suitable soil conditionsin the forest understory layer and a signal of nichepartitioning.Community compensatory trendFor tree seedlings, a significant negative correlationbetween seedling survival and tree species’ basal area wasfound, which explained 84.5 % of the variation amongspecies seedling survival (Table 4). This result indicatesthat NDD may be occurring at both the local and communitylevels. However, the CCT was detected with basalarea, but not abundance. This means that tree species withlarger stems may support more natural enemies or occupy alarger portion of canopy or root space than would smallerstems. Hence, species basal area, a proxy for biomass, maybe a better index for examining the existence of a CCT.CaveatsOne major limitation of our study is that it occurred in asingle large plot over two census intervals. As such, conclusionsfrom this short-term period of a single study areaare tentative.ConclusionOur study found evidence of temporal and spatial variation inseedling survival in an old-growth temperate forest, resultingfrom both local biotic neighborhood and habitat heterogeneity.We found convincing evidence of widespread NDDamong tree seedlings. Habitat factors were more importantfor shrubs. The strength and importance of these processesalso varied with dispersal modes and species, lending credenceto the ideas that mechanisms of coexistence may varywith ontogeny. Biotic factors were important predictors ofseedling survival in young trees whereas habitat factors weresignificant for older trees and shrubs. It seems likely that bothNDD and habitat heterogeneity are important mechanismsthat maintain the coexistence of numerous species in bothtemperate and tropical forests.Acknowledgments We thank Drs. Liza S. Comita, Fangliang He,Xiangcheng Mi, Luxiang Lin, and Lei Chen, ChiaHao ChangYang forhelpful suggestions on data analysis. We thank Drs. Walter P. Carson,Erin Kurten, C. E. Timothy Paine and an anonymous reviewer for criticalcomments for the manuscript. We also thank Liwei Wang for his fieldwork and data collection. This work is supported by the National NaturalScience Foundation of China (No. 31061160188 and No.31011120470), the Knowledge Innovation Program of the ChineseAcademy of Sciences (No. KZCX2-EW-401 and No. KSCX2-EW-Z-5).The publication of this manuscript has been approved by all co-authors.ReferencesBates D, Maechler M, Bolker B (2010) lme4: Linear mixed-effectsmodels using S4 classes. R Package Version: 0.999375-37 ednBurnham KP, Anderson DR (2002) Model selection and multimodelinference: a practical information-theoretic approach. Springer,New YorkCanham CD, LePage PT, Coates KD (2004) A neighborhood analysisof canopy tree competition: effects of shading versus crowding.Can J For Res 34:778–787Cavender-Bares J, Kitajima K, Bazzaz FA (2004) Multiple traitassociations in relation to habitat differentiation among 17Floridian oak species. Ecol Monogr 74:635–662Chen L, Mi XC, Comita LS, Zhang LW, Ren HB, Ma KP (2010)Community-level consequences of density dependence andhabitat association in a subtropical broad-leaved forest. EcolLett 13:695–704Chesson P (2000) Mechanisms of maintenance of species diversity.Annu Rev Ecol Syst 31:343–366Comita LS, Hubbell SP (2009) Local neighborhood and species’shade tolerance influence survival in a diverse seedling bank.Ecology 90:328–334Comita LS, Aguilar S, Perez R, Lao S, Hubbell SP (2007a) Patterns ofwoody plant species abundance and diversity in the seedlinglayer of a tropical forest. J Veg Sci 18:163–174Comita LS, Condit R, Hubbell SP (2007b) Developmental changes inhabitat associations of tropical trees. J Ecol 95:482–492Comita LS, Uriarte M, Thompson J, Jonckheere I, Canham CD,Zimmerman JK (2009) Abiotic and biotic drivers of seedlingsurvival in a hurricane-impacted tropical forest. J Ecol97:1346–135912341

OecologiaComita LS, Muller-Landau HC, Aguilar S, Hubbell SP (2010)Asymmetric density dependence shapes species abundances in atropical tree community. Science 329:330–332Connell JH (1971) On the role of natural enemies in preventingcompetitive exclusion in some marine animals and in rain foresttrees. In: den Boer PJ, Gradwell GR (eds) Dynamics ofPopulations. Centre for Agricultural Publishing and Documentation,Wageningen, pp 298–312Connell JH, Tracey JG, Webb LJ (1984) Compensatory recruitment,growth, and mortality as factors maintaining rain forest treediversity. Ecol Monogr 54:141–164Gilbert GS, Harms KE, Hamill DN, Hubbell SP (2001) Effects ofseedling size, El Nino drought, seedling density, and distance tonearest conspecific adult on 6-year survival of Ocotea whiteiseedlings in Panama. Oecologia 127:509–516Grothendieck G (2010) sqldf: Perform SQL Selects on R DataFrames. 0.3–5 ednGrubb PJ (1977) Maintenance of species-richness in plant communities-importanceof regeneration niche. Biol Rev Camb PhilosSoc 52:107–145Hao ZQ, Zhang J, Song B, Ye J, Li BH (2007) Vertical structure andspatial associations of dominant tree species in an old-growthtemperate forest. For Ecol Manag 252:1–11Hao ZQ, Li BH, Zhang J, Wang XG, Ye J, Yao XL (2008) BroadleavedKorean pine (Pinus koraiensis) mixed forest plot inChangbaishan (CBS) of China: community composition andstructure. Acta Phytoecol Sinica 32:238–250Harms KE, Condit R, Hubbell SP, Foster RB (2001) Habitatassociations of trees and shrubs in a 50-ha neotropical forestplot. J Ecol 89:947–959He FL, Legendre P, LaFrankie JV (1997) Distribution patterns of treespecies in a Malaysian tropical rain forest. J Veg Sci 8:105–114HilleRisLambers J, Clark JS (2003) Effects of dispersal, shrubs, anddensity-dependent mortality on seed and seedling distributions intemperate forests. Can J For Res 33:783–795HilleRisLambers J, Clark JS, Beckage B (2002) Density-dependentmortality and the latitudinal gradient in species diversity. Nature417:732–735Hosmer DW, Lemeshow S (2000) Applied logistic regression, 2ndedn. Wiley, New YorkHubbell SP (2001) The unified neural theory of biodiversity andbiogeography. Princeton University Press, PrincetonHubbell SP, Ahumada JA, Condit R, Foster RB (2001) Localneighborhood effects on long-term survival of individual trees ina neotropical forest. Ecol Res 16:859–875Iida S, Nakashizuka T (1998) Spatial and temporal dispersal ofKalopanax pictus seeds in a temperate deciduous forest, centralJapan. Plant Ecol 135:243–248Janzen DH (1970) Herbivores and the number of tree species intropical forests. Am Nat 104:501–528John R et al (2007) Soil nutrients influence spatial distributions oftropical tree species. Proc Nat Acad Sci USA 104:864–869Kraft NJB, Valencia R, Ackerly DD (2008) Functional traits andniche-based tree community assembly in an Amazonian forest.Science 322:580–582Kraft NJB et al (2011) Disentangling the drivers of beta diversityalong latitudinal and elevational gradients. Science 333:1755–1758Martínez-Vilalta J, Mencuccini M, Vayreda J, Retana J (2010)Interspecific variation in functional traits, not climatic differencesamong species ranges, determines demographic ratesacross 44 temperate and Mediterranean tree species. J Ecol98:1462–1475Masaki T, Nakashizuka A (2002) Seedling demography of Swidacontroversa: effect of light and distance to conspecifics. Ecology83:3497–3507McMahon SM, Metcalf CJE, Woodall CW (2011) High-dimensionalcoexistence of temperate tree species: functional traits, demographicrates, life-history stages, and their physical context.PLoS ONE 6:e16253Metz MR, Sousa WP, Valencia R (2010) Widespread densitydependentseedling mortality promotes species coexistence in ahighly diverse Amazonian rain forest. Ecology 91:3675–3685Murray KG (1988) Avan seed dispersal of 3 Neotropical gapdependentplants. Ecol Monogr 58:271–298Nakashizuka T (2001) Species coexistence in temperate, mixeddeciduous forests. Trends Ecol Evol 16:205–210Packer A, Clay K (2000) Soil pathogens and spatial patterns ofseedling mortality in a temperate tree. Nature 404:278–281Paine CET, Harms KE, Schnitzer SA, Carson WP (2008) Weakcompetition among tropical tree seedlings: implications forspecies coexistence. Biotropica 40:432–440Paine CET et al (2011) Phylogenetic density dependence andenvironmental filtering predict seedling mortality in a tropicalforest. Ecol Lett 15:34–41Queenborough S, Burslem D, Garwood N, Valencia R (2007a)Habitat niche partitioning by 16 species of Myristicaceae inAmazonian Ecuador. Plant Ecol 192:193–207Queenborough SA, Burslem D, Garwood NC, Valencia R (2007b)Neighborhood and community interactions determine the spatialpattern of tropical tree seedling survival. Ecology 88:2248–2258Queenborough SA, Burslem DFRP, Garwood NC, Valencia R (2009)Taxonomic scale-dependence of habitat niche partitioning andbiotic neighbourhood on survival of tropical tree seedlings. ProcR Soc Lond B 276:4197–4205Ramaswami G, Sukumar R (2011) Woody plant seedling distributionunder invasive Lantana camara thickets in a dry-forest plot inMudumalai, southern India. J Trop Ecol 27:365–373Russo SE, Davies SJ, King DA, Tan S (2005) Soil-relatedperformance variation and distributions of tree species in aBornean rain forest. J Ecol 93:879–889Shao GF, Schall P, Weishampel JF (1994) Dynamic simulations ofmixed broadleaved-Pinus koraiensis forests in the ChangbaishanBiosphere Reserve of China. For Ecol Manag 70:169–181Shibata M et al (2010) Effects of abiotic and biotic factors andstochasticity on tree regeneration in a temperate forest community.Ecoscience 17:137–145Stone R (2006) Ecology—a threatened nature reserve breaks downAsian borders. Science 313:1379–1380Streng DR, Glitzenstein JS, Harcombe PA (1989) Woody seedlingdynamics in an east texas floodplain forest. Ecol Monogr59:177–204Svenning JC, Fabbro T, Wright SJ (2008) Seedling interactions in atropical forest in Panama. Oecologia 155:143–150Tateno R, Takeda H (2003) Forest structure and tree speciesdistribution in relation to topography-mediated heterogeneityof soil nitrogen and light at the forest floor. Ecol Res 18:559–571Tsujino R, Takafumi H, Agetsuma N, Yumoto T (2006) Variation intree growth, mortality and recruitment among topographicpositions in a warm temperate forest. J Veg Sci 17:281–290Uriarte M, Canham CD, Thompson J, Zimmerman JK (2004a) Aneighborhood analysis of tree growth and survival in a hurricanedriventropical forest. Ecol Monogr 74:591–614Uriarte M, Condit R, Canham CD, Hubbell SP (2004b) A spatiallyexplicit model of sapling growth in a tropical forest: does theidentity of neighbours matter? J Ecol 92:348–360Volkov I, Banavar JR, He F, Hubbell SP, Maritan A (2005) Densitydependence explains tree species abundance and diversity intropical forests. Nature 438:658–661Wang Z, Xu ZB, Li X (1980) The main forest types and their featuresof community structure in northern slope of Changbai Mountain.Res For Ecosyst 1:25–4212342

OecologiaWebb CO, Peart DR (1999) Seedling density dependence promotescoexistence of Bornean rain forest trees. Ecology 80:2006–2017Webb CO, Gilbert GS, Donoghue MJ (2006) Phylodiversity-dependentseedling mortality, size structure, and disease in a borneanrain forest. Ecology 87:S123–S131Welden CW, Hewett SW, Hubbell SP, Foster RB (1991) saplingsurvival, growth, and recruitment: relationship to canopy heightin a neotropical forest. Ecology 72:35–50Wilson BF (1995) Shrub stems: form and function. In: Gerter BL (ed)Plant stems: physiology and functional morphology. Academic,New York, pp 91–102Winkler M, Hulber K, Hietz P (2005) Effect of canopy position ongermination and seedling survival of epiphytic bromeliads in aMexican humid montane forest. Ann Bot 95:1039–1047Wright J (2002) Plant diversity in tropical forests: a review ofmechanisms of species coexistence. Oecologia 130:1–14Wright IJ et al (2007) Relationships among ecologically importantdimensions of plant trait variation in seven neotropical forests.Ann Bot 99:1003–1015Yang H, Li D, Wang B, Han J (1985) Distribution patterns ofdominant tree species on northern slope of Changbai mountain.Res For Ecosyst 5:1–14Yuan ZQ et al (2011) Scale specific determinants of tree diversity in anold growth temperate forest in China. Basic Appl Ecol 12:488–495Zhang J, Hao ZQ, Li BH, Ye J, Wang XG, Yao XL (2008)Composition and seasonal dynamics of seed rain in broad-leavedKorean pine (Pinus koraiensis) mixed forest, Changbai Mountain.Acta Ecol Sinica 28:2445–245412343

Downloaded from rspb.royalsocietypublishing.org on September 6, 2012Testing the independent species’arrangement assertion made by theoriesof stochastic geometry of biodiversityThorsten Wiegand 1, *, Andreas Huth 1 , Stephan Getzin 1 ,Xugao Wang 2, *, Zhanqing Hao 2 , C. V. Savitri Gunatilleke 3and I. A. U. Nimal Gunatilleke 3Proc. R. Soc. B (2012) 279, 3312–3320doi:10.1098/rspb.2012.0376Published online 16 May 20121 UFZ Helmholtz Centre for Environmental Research, Department of Ecological Modelling,PF 500136, Leipzig 04301, Germany2 State Key Laboratory of Forest and Soil Ecology, Institute of Applied Ecology,Chinese Academy of Sciences, Shenyang 110164, People’s Republic of China3 Faculty of Science, Department of Botany, University of Peradeniya, Peradeniya 20400, Sri LankaThe assertion that the spatial location of different species is independent of each other is fundamental inmajor ecological theories such as neutral theory that describes a stochastic geometry of biodiversity. However,this assertion has rarely been tested. Here we use techniques of spatial point pattern analysis to conduct acomprehensive test of the independence assertion by analysing data from three large forest plots with differentspecies richness: a species-rich tropical forest at Barro Colorado Island (Panama), a tropical forest inSinharaja (Sri Lanka), and a temperate forest in Changbaishan (China). We hypothesize that stochasticdilution effects owing to increasing species richness overpower signals of species associations, thereby yieldingapproximate species independence. Indeed, the proportion of species pairs showing: (i) no significant interspecificassociation increased with species richness, (ii) segregation decreased with species richness, and (iii)small-scale interspecific interaction decreased with species richness. This suggests that independence mayindeed be a good approximation in the limit of very species-rich communities. Our findings are a steptowards a better understanding of factors governing species-rich communities and we propose a hypothesisto explain why species placement in species-rich communities approximates independence.Keywords: environmental heterogeneity; forests; neutral theory; point pattern analysis;spatial pattern; species interactions1. INTRODUCTIONOne of the principal goals of ecology is to understand theprocesses and mechanisms that control the distribution,abundance and coexistence of species [1,2]. Classicaltheoretical ecology predicts that two species competingfor the same resources cannot stably coexist (the exclusionprinciple by Gause [3]). However, for reasons thatare poorly understood the number of competing speciesoften exceeds the number of limiting resources [4]. Thebest-known examples are the high diversity of species intropical forests and coral reefs [5] and the paradox ofthe plankton [4,6]. In tropical forests, several hundredsof tree species can be found within small areas [7] and,for example, up to 300 tree species per hectare havebeen recorded in the Amazonia [8].For more than 90 years or so, major efforts have beenmade in field and theoretical ecology to resolve this paradox[5,9–13], and about 10 years ago investigations of this issuehad gained considerable momentum and new directionssince Stephen Hubbell revived neutral theory in community* Authors for correspondence (thorsten.wiegand@ufz.de;wxg_7980@163.com).Electronic supplementary material is available at http://dx.doi.org/10.1098/rspb.2012.0376 or via http://rspb.royalsocietypublishing.org.ecology [5]. Surprisingly, neutral models have been remarkablysuccessful at reproducing several empirically observedmacroscopic patterns for communities with organisms atthe same trophic level such as tropical forests, grassland,shrubland, birds, groups of insects, fishes, marine Diatomeaor coral reefs, although they ‘contradict almost everythingthat ecologists have to come to understand about speciesdiversity and its maintenance in communities’ [14].Recently, McGill [15] synthesized six theories of biodiversitythat produce, based on a few underlying principles,macroscopic community patterns such as species-areacurves (SARs), species-abundance distributions (SADs)and decay of similarity of distance. He showed that thesetheories use the same three rules or assertions to describea stochastic geometry of biodiversity, namely: (i) intraspecificclustering, (ii) the species abundance distribution showstypically many rare and few common species (i.e. a hollowcurve distribution), and (iii) interspecific individuals areplaced without regard to individuals of other species.McGill [15] argued that these three rules appear sufficientfor explaining several macroscopic community patterns.Empirical evidence in support of McGill’s assertion(iii) of independent species placement is, however, lackingalthough models assuming no species interactions aresuccessful in predicting diversity patterns [5,15–17].By contrast, there is good empirical evidence that most44Received 18 February 2012Accepted 25 April 2012 3312 This journal is q 2012 The Royal Society

Downloaded from rspb.royalsocietypublishing.org on September 6, 2012Spatial associations among tree species T. Wiegand et al. 3313tropical tree species show intraspecific aggregation[16–18] and that species-rich communities containusually many rare and a few common species [5,19].The assertion that species are placed independently isalso in apparent contradiction to the vast literature in ecologydevoted to the study of species interactions [10,20].However, evidence for the importance of species interactionsstems mostly from species-poor communities[21], whereas the few studies that explored the full arrayof spatial species interactions in fully mapped species-richcommunities found that only a few species pairs showeddetectable spatial interactions [20,22–24].The question arises as to whether the effect of speciesinteractions in species-rich communities may be overpoweredby stochastic effects [20,22]. For example, Hubbelland Foster [25] noted that in species-rich forests two individualsof the same species may share only a few commonspecies among their nearest neighbours. In the BarroColorado Island (BCI) forest the 20 nearest neighboursof a given tree comprised on average 14 different treespecies [26]. Individuals of a given species are thereforeoften exposed to considerably different biotic neighbourhoods.This suggests that pairwise interactions may beweak on average, despite the existence of a few strongerinteractions [15,20,22,27]. On evolutionary timescales,such diverse and unpredictable local assemblages ofcompetitors around individual plants would not allowdirectional specialization, but species may instead convergeon similar life-history strategies that are competitivelyequivalent [26]. The hypothesis of diffuse coevolution offunctionally similar species that may produce ecologicalequivalence among species traits is a cornerstone of theneutral theory [5,26].One way of assessing the evidence for species interactionsin plants is to analyse their spatial patterns[22–24,28,29]. Because plants cannot move and mainlyinteract with their close neighbours [28], their spacingmay conserve an imprint of neighbourhood interactionsthat could be detected using point pattern analysis[21,22,24]. This approach is promising because the intraspecificspacing of plants is also closely related withpotential coexistence mechanisms [30]. For example,intraspecific clustering and interspecific segregation mayretard competitive exclusion because the relative importanceof interspecific versus intraspecific competition isreduced [29–31]. Analysis of the bivariate spatial patternsof all pairs of species allows testing if the interspecificarrangement of species is indeed independent as assumedby assertion (iii) [20,22–24]. However, such analyses arechallenging because they require complete mapping oflarge plots of a species-rich community and because of difficultiesin teasing apart two major, yet contrasting factorsthat can cause non-independence: habitat association maymediate positive or negative association and direct speciesinteractions such as competition or facilitation [21–23].In this study, we use spatial point pattern analysis[28,32,33] to test the assertion of species independencewith data from three forests of different species richnessincluding a 50 ha plot of neotropical forest at BCI,Panama (more than 300 species of trees and shrubs), a25 ha plot of tropical forest in Sinharaja, Sri Lanka(more than 200 species) and a 25 ha plot temperateforest in Changbaishan (CBS), China (52 species).Because habitat association and species interactions mayProc. R. Soc. B (2012)45occur simultaneously, we conduct one analysis studyingtheir joint effects and one analysis studying selectivelyspecies interactions. In analysis 1 we therefore analysepairwise interspecific association patterns among largetrees (greater than 10 cm diameter at breast height(dbh)) with respect to overall spatial patterning (potentiallycaused by habitat association and speciesinteractions). This analysis explores how trees of a givenspecies j are distributed within local neighbourhoods ofthe trees of a focal species i and therefore how frequentlydifferent species meet and have the opportunity to interact.In analysis 2, we approximately factor out largerscaleeffects of the environment and analyse selectivelysmall-scale interspecies interactions. This analysisexplores how trees of species j behaved when they wereclose to trees of species i and therefore if they werearranged closer or further away than expected by thelocal density of species j. For all three forests, the methodologypresented by Wiegand et al. [22] and Wang et al.[21] was used. This allows us to synthesize the resultsfor three forests with different species richness.We expect that species-rich forest communities shouldapproximate species independence more closely thanspecies-poorer communities. Our general hypothesis isthat stochastic effects dilute species associations in highlydiverse communities. This is because the density of individualspecies decreases with increasing richness, which lowersthe probability that a species j tree is among the nearestneighbours of a species i [20,22,25,26]. As a result ofthe low rate at which individuals of two species meet,the statistical tests will detect fewer significant effects.Consequently, the occurrence of species associations orinteractions in spatial patterns should decrease withincreasing species richness. With respect to overall spatialpatterning (i.e. habitat and interactions, analysis 1) it followsfrom our general hypothesis that the proportion ofspecies pairs with ‘no association’ should increase withspecies richness (specific hypothesis H1a) and that the proportionof species pairs showing segregation shoulddecrease with species richness (specific hypothesis H1b).The latter implies that the coexistence mechanism of intraspecificaggregation and interspecific segregation [30] shouldbe more important in communities with fewer species. Withrespect to species interactions (analysis 2) it follows fromour general hypothesis that the proportion of species pairswith non-significant interspecific small-scale interactionsshould increase with species richness (specific hypothesisH2). In this comparative study we found strong supportfor our hypotheses.2. METHODS(a) Study sitesWe collated data on spatial distributions of tree species fromthree forest plots. The 50 ha plot on BCI [34], Panama(9810 0 N, 79851 0 W) has a moist, lowland, tropical climate,with rainfall 2500 mm yr 21 , a strong 3.5 month dry seasonand a year-round mean daily temperature of 278C. Elevationranges from 120 to 155 m. The plot is described in detail byHubbell and Foster [35], and for details of census methodssee Condit [36]. The 25 ha plot at Sinharaja (06824 0 N,80824 0 E) is a tropical forest without a distinct dry season,and annual rainfall averaged 5016 mm yr 21 , with a rangefrom 4087 to 5907 mm. Annual daily maximum and

Downloaded from rspb.royalsocietypublishing.org on September 6, 20123314 T. Wiegand et al. Spatial associations among tree speciesminimum temperatures were 24.78C and 20.48C, respectively.Elevation ranges from 424 to 575 m above mean sealevel and includes a valley lying between two slopes. Detailsof the plot are provided in Gunatilleke et al. [37]. The25 ha plot CBS, China (42823 0 N, 128805 0 E) is located inbroadleaved Korean pine (Pinus koraiensis) mixed forest andhas a temperate continental climate with long cold wintersand warm summers, approximately 700 mm rain yr 21which mostly falls from June to September (490–500 mm).Elevation ranges from 791.8 to 809.5 m, and the meanannual temperature is 2.98C, with a mean January temperatureof 213.78C and a mean July temperature of 19.68C.The plot is described in detail by Wang et al. [21]. Additionalinformation on the plots is provided in the electronic supplementarymaterial, table A1.In the present analyses, we used data from the 1995 BCIcensus [38] and the 1996 Sinharaja census and restricted ouranalysis to the 62 and 46 species which had more than 70individuals with dbh larger than 10 cm, respectively. Forspecies with fewer individuals, stochastic effects becometoo large for the purpose of our study. To be consistentwith Wang et al. [21] we included in the analysis of theCBS plot 15 species which had more than 50 individualswith dbh greater than 10 cm.(b) Spatial pattern analysisWe used three summary statistics applicable to completelymapped bivariate point patterns, the K-function K 12 (r)[39,40], the pair correlation function g 12 (r) [32,41] and thecumulative nearest neighbour distribution function D 12 (r)that gives the probability that the nearest species 2 neighbourof an individual of species 1 is located within distance r [33].The quantity l 2 K 12 (r) can be interpreted as the number ofspecies 2 plants within distance r from species 1 plantswhere l 2 is the density of species 2 plants in the study area(¼the number of plants of pattern 2 divided by the area ofthe study plot). The pair correlation function is related tothe derivative of the K-function, i.e. dK 12 (r)/dr ¼ g 12 (r)2pr[33], and l 2 g 12 (r) can be interpreted as the density of species2 plants at distances (r2dr/2,r þ dr/2) from species 1 plants,where dr is the ‘ring width’. We also calculated an index oflocal dominance [23], L f (r), which is the average proportionof conspecific individuals within given neighbourhoods withradius r of the individuals of the focal species f.We followed the implementation of the three statisticsusing the software PROGRAMITA [22,32], which can berequested from the first author. We selected for all analysesa spatial resolution of 2 m, and for analysis 2 a ring widthof dr ¼ 6 m. This is a sufficiently fine resolution comparedwith the dimensions of the plots (1000 500 m 2 and500 500 m 2 ) and sufficiently fine to answer our questions[21,22].(c) Analysis 1. Categorization of overallspecies–species associationsTo categorize the possibly heterogeneous species–speciesassociations, we used a two-dimensional classificationscheme [21,22] based on the K-function K 12 (r) and the nearestneighbour distribution function D 12 (r) (see electronicsupplementary material, appendix A). This allowed us toquantify how the trees of a given species 2 were distributedwithin local neighbourhoods of the trees of a focal species 1,irrespective of whether spacing was caused by external effectsProc. R. Soc. B (2012)of the environment, by internal effects of species interactionsor by intraspecific clustering.To distinguish the different types of spatial associations(figure 1) from those that may arise purely by chance, we comparedthe observed bivariate point patterns with a null model inwhich the locations of the focal species remained unchanged,but trees of species 2 were distributed randomly and independentlyof the locations of species 1 (i.e. a homogeneous Poissonprocess [32]). The expectations of the summary statisticsunder the null model yield K 12 (r) ¼ pr 2 and D 12 (r) ¼(12exp(2l 2 pr 2 )). The two axes of the classification scheme[21,22] were defined as:^PðrÞ ¼ ^D 12 ðrÞ ð1 expð l 2 pr 2 Þand^MðrÞ ¼lnð ^K 12 ðrÞÞ lnðpr 2 Þ:ð2:1ÞWe subtracted the theoretical values under the null modelto move the null expectation onto the origin of the schemeand log-transformed the K-function to weight positive ornegative departures from the null model in the same way[22]. The hat symbol indicates the observed value.The two-axis scheme allows for the identification of fourfundamental types of bivariate association: type I: ‘segregation’[ ^MðrÞ , 0 and ^PðrÞ , 0] where individuals ofspecies 2 occur on average less frequently within species 1neighbourhoods than expected by chance alone; type II: ‘partialoverlap’ [i.e. ^MðrÞ . 0 and ^PðrÞ , 0], where someneighbourhoods of species 1 contain more individuals ofspecies 2 and others less; type III: ‘mixing’ [i.e. ^MðrÞ . 0and ^PðrÞ . 0] where individuals of species 2 occur on averagemore frequently within species 1 neighbourhoods; and typeIV: [^PðrÞ . 0, ^MðrÞ , 0], which can only arise if strongsecond-order effects occur. Species pairs that show for agiven neighbourhood r non-significant effects in both summarystatistics are classified as ‘no association’ type and willbe located close to the origin of the scheme. Further detailsof the scheme can be found in the electronic supplementarymaterial, appendix A.(d) Analysis 2. Small-scale species–speciesinteractionsIn analysis 1, two species may show significant effects even ifthey do not show direct interactions. This may happen if thetwo species occupy only subareas A1 and A2 of the study plot[22]. For example, imagine that species 1 and 2 both existonly in half of the landscape (A1) and assume that they areboth independently distributed in A1. Analysis 1 would indicatestrong mixing because they are more closely associatedin the data than they would be if we moved species 2 to bein random positions on the landscape. However, we canapproximately factor out this effect if we move species 2only locally. In this case the resulting large-scale patternwill look similar to the observed large-scale pattern, and itis unlikely that a significant effect will be detected [32].This null model leaves the density l 2 (x) of species 2 approximatelyunchanged, but displacement of species 2 within localneighbourhoods with radius R removes potential signals ofsmall-scale interactions at scales r , R. Thus, under approximateseparation of scales [21], we can selectively study thesmall-scale interactions by using a null model which randomizesthe data conditionally on the observed larger-scalepattern [22,32].46

Downloaded from rspb.royalsocietypublishing.org on September 6, 2012Spatial associations among tree species T. Wiegand et al. 3315(b) Dendropanax arboreus–Socratea exorrhiza(d) Garcinia intermedia–Drypetes standleyi(a)Ripley’s K classification axis M210–1–2–3type IIItype IItype Itype IV–1.0 –0.5 0nearest neighbour classification axis P(c) Ocotea whitei–Virola surinamensis( f ) Xylopia macrantha – Trichilia tuberculataNW ES(e)5004003002001000Triplaris cumingiana–Socratea exorrhiza200 400 600 800 1000Figure 1. Classification of association patterns at the BCI 1000 500 m 2 forest dynamics plot. (a) Allocation of the associationsof the 3782 species–species pairs based on the classification axes defined in equation (2.1) (black open circles)compared with those found at the Sinharaja forest (Sri Lanka; grey dots) and the CBS plot (China; red dots). The axis P ispositive (negative) if there are on average more (less) stems of species 2 at distance r L ¼ 30 m from species 1 stems thanexpected, and the axis M is positive (negative) if the probability that a stem of species 1 has its nearest species 2 neighbourwithin distance r L is larger (smaller) than expected. The blue dots locate the four examples shown in panels (b– f ). Thedashed line indicates the area of strong segregation. (b) Example of type II association with partial overlap; black circles: species1; red circles: species 2. (c) Example of type III association with mixing. (d) Example of a transition between type I and type IIassociation. (e) Example of type I association with strong segregation. ( f ) Example of a type IV association, which is onlypossible because of a strong second-order effect of pattern 1.Technically, we implemented this null model as a heterogeneousPoisson process [32] for the second species (theindividuals of the focal species remain unchanged) and weselected a value of R ¼ 30 m (for a justification, see the electronicsupplementary material, appendix A and Wang et al.[21]). The non-cumulative pair correlation function g 12 (r)is the appropriate summary statistic here because it allowsfor a more direct quantification of scale-dependent interactionsthan the cumulative K-function where, for instance,the effect of repulsion at smaller distances is only graduallydiluted out by independence at larger distances [28,32].However, for easier comparison between analyses 1 and 2we present results based on the K-function (see the electronicsupplementary material, appendix B). We tested all pairsof species, i.e. species 1 versus species 2 and species 2versus species 1 since we cannot assume that the interactionwould be symmetric.(e) Significance testsThe empirical summary statistics were compared with thosegenerated by 199 simulations of the homogeneous or heterogeneousPoisson null model. The overall fit of the null modelwas then determined with a goodness-of-fit test [42] whichreduces the scale-dependent information contained in a summarystatistic (over an appropriate distance interval) into asingle test statistic u i . The u i were calculated for the observeddata (i ¼ 0) and for the data created by the i ¼ 1, ..., 199simulations of the null model. If the rank of u 0 among allu i was larger than 195 or 190, the data showed a significantProc. R. Soc. B (2012)47departure from the null model with error rate a ¼ 0.025 or0.05, respectively [33,42].In analysis 1, we tested over the 2–50 m distance interval.A significant departure from the null model occurred if atleast one of the two summary statistics was significant witha 2.5 per cent error rate (this yields for two summary statisticsan error rate of 5%). To stabilize the variance, weused the transformation L 12 (r) ¼ (K 12 (r)/p) 0.5 2r instead ofK 12 (r) [28,33]. In analysis 2, the goodness-of-fit test wasconducted with the pair correlation function over the 2–30 m distance interval with a 5 per cent error rate. Toobtain an overview of changes in spatial association structurewith neighbourhood size r, we counted in both analyses theproportions of the different association or interaction typesseparately for each value of r (see the electronic supplementarymaterial, appendix A).3. RESULTS(a) Analysis 1. Categorization of overallspecies–species associationsFigure 1a shows the resulting association types of speciespairs for 30 m neighbourhoods for BCI (black dots), Sinharaja(grey dots) and CBS (red dots). Figure 1b–f showfor illustrative purposes the tree positions of five speciespairs that correspond to extreme cases in the BCI classificationscheme. The first observation is that the threeschemes have a similar structure, but that of Sinharajashows more extreme cases of segregation or mixingthan BCI. This observation is also valid for other

Downloaded from rspb.royalsocietypublishing.org on September 6, 20123316 T. Wiegand et al. Spatial associations among tree species(a)1.0BCI(b)Sinharaja(c)CBSproprtion of cases0.80.60.40.20 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50spatial scale, r (m) spatial scale, r (m) spatial scale, r (m)Figure 2. Assessment of overall interspecific association patterns at the three forests in dependence on the neighbourhood scale r.Illustrated are the proportion of species pairs showing the four association types and the type with no significant association. Thegrey line indicates no effect; closed red circles, segregation; closed squares, partial overlap; closed green circles, mixing; closedblack circles, type IV.(a)1.0no association(b)1.0segregationproportion of no association0.80.60.40.2proportion of segregation0.80.60.40.2050 100 150 200 250 300 350CBS Sinharaja BCItotal species richness050 100 150 200 250 300 350CBS Sinharaja BCItotal species richnessFigure 3. Overall association patterns in relation to total species richness of the three forests. (a) Proportion of species pairsshowing the no association type at 2 m (closed circles) and 10 m neighbourhoods (open circles). (b) Proportion of speciespairs showing segregation at 2 m neighbourhoods (closed circles) and 10 m neighbourhoods (open circles).neighbourhoods (figure A1 in the electronic supplementarymaterial). The scheme of the CBS forest shows (similarlyto that of the Sinharaja forest) more extreme cases of segregationthan that of the BCI forest. To quantify theproportion of species pairs with strong segregation wedefined all cases where P ,20.25 and M ,20.5(dashed lines in figure 1a) as strong segregation. Wefound that 3.2, 25.7 and 12.4 per cent of all speciespairs at BCI, Sinharaja and CBS, respectively, showedstrong segregation.All three forests showed a qualitatively similar pattern(figure 2): the proportion of the no association typedecreased with neighbourhood size, that of segregationand partial overlap increased and the proportion ofmixing peaked at neighbourhoods of approximately10 m. However, when comparing the patterns among forestswe found interesting quantitative differences relatedwith species richness. The importance of the no associationtype increased for smaller neighbourhoods withspecies richness (figure 3a). This is in agreement withour specific hypothesis H1a. Whereas the no associationtype dominates the species association patterns up toProc. R. Soc. B (2012)neighbourhoods of say 30 m at BCI, it does so at theSinharaja and CBS forests only up to 10 and 4 m, respectively.For larger neighbourhoods (i.e. greater than 30 m),the Sinharaja plot shows a low proportion of the noassociation type of approximately 14 per cent, CBS 23per cent and BCI 38 per cent (figure 2). It is also worthnoting that the importance of segregation decreases withspecies richness (our specific hypothesis H1b; figure 3b)and that the proportion of species associations showingmixing has similar patterns in all three forests withpeaks at the 10 m neighbourhood (figure 2).Only one species at the BCI forest yielded at 10 mneighbourhoods a local dominance index greater than0.2, but only because this species (Gustavia superba)showed an unusually high local density in a disturbedarea at the northern border of the plot. All other speciesat BCI showed a local dominance index of less than0.16. At Sinharaja, four species showed a local dominanceindex of between 0.2 and 0.34. When plotting the dominanceindices over relative dominance rank (electronicsupplementary material, figure A2) we find that the dominanceindices at the species-richest BCI plot are lowest,48

Downloaded from rspb.royalsocietypublishing.org on September 6, 2012Spatial associations among tree species T. Wiegand et al. 3317(a)0.25BCI(b) Sinharaja(c)proportion of species pairs0.200.0250.0250.0200.0200.150.0150.0150.0100.0100.10 0.0050.0050.050 10 20 300 10 20 30CBS0 10 20 30 40spatial scale, r (m)0 10 20 30 40spatial scale, r (m)0 10 20 30 40spatial scale, r (m)Figure 4. Analysis of species pairs for which the pair correlation function g 12 (r) showed for scales 0–30 m a significant departurefrom the null model as indicated by the goodness-of-fit test (i.e. rank .190). Shown are for each distance r the proportion of caseswhere the g 12 (r) was below the simulation envelopes (i.e. repulsion; open circles) and above the simulation envelopes (i.e. attraction;closed circles). For better clarity, the small inset graphs in (a) and(b) show the proportion with higher resolution.followed by the Sinharaja plot with lower species richnessand that the dominance indices at the species-poor CBSplot were larger when compared with that of the twotropical forests.(b) Analysis 2. Small-scale species–speciesinteractionsThe goodness-of-fit test detected for the BCI data in 5.3per cent of all cases (202 species pairs) significantlydeparts from the heterogeneous Poisson null model atscales 0–30 m. This result is similar to the 5.8 per centfound at Sinharaja (120 pairs), but quite different fromthe 30 per cent found at the CBS forest (64 pairs).Thus, detectable small-scale interspecific interactionsare rare in the spatial pattern of the two tropical forests,but more common at the temperate forest. We alsofound that almost all significant spatial interactionsbetween species pairs at BCI occurred at small scales ofless than 10 m with a sharp decline in its frequency forscales greater than 2 m (inset figure 4a). Interestingly,negative effects disappeared already at scales greaterthan 5 m which indicates that our assumption of separationof scales was met. If there were, in reality, noseparation of scales, then the frequency of significanteffects should disappear only at 30 m [21].The results of the BCI forest are qualitatively similar tothat obtained at Sinharaja (figure 4a,b): positive and negativeassociations at small scales occur at both forests withsimilar proportion (i.e. 0.02) and quickly disappear withincreasing scale. However, the temperate CBS forest yieldedsubstantially different results with negative interactionsdominating at small scales of less than 10 m (figure 4c).The K-function detected fewer significant departures fromthe null model and, owing to its cumulative nature, scaleeffects were somewhat obscured (electronic supplementarymaterial, figure A3). Comparing across the three forestsshows accordance with specific hypothesis H2, that the proportionof non-significant small-scale interactions increasedwith increasing species richness although no differenceswere found for the two species-richer tropical forests.However, the above results do not mean that speciesinteractions were always weak. We found that there weresome species with significantly more interactions thanexpected by random allocation of interactions and somewith fewer interactions (electronic supplementary material,Proc. R. Soc. B (2012)49appendix B). We also found that significant interspecificassociations and interactions did not primarily depend onthe sample sizes although, as expected, significant effectstended to be more frequent for larger sample sizes andthe correlation between the number of individuals andthe rank of the goodness-of-fit test was stronger forspecies-poorer forests (electronic supplementary material,appendix C).4. DISCUSSIONIn this study, we analysed thousands of bivariate spatialpatterns of tree and shrub species at three fully mapped25–50 ha forest plots. Our main objective was to test theassertion of independence in the interspecific arrangementof species which is fundamental in six major theoriesthat describe a stochastic geometry of biodiversity [15].We conducted two sets of analyses. In analysis 1, weexplored how frequently individuals of species pairs meetand have the opportunity to interact, and in analysis 2,we explored how trees of a species pair interacted whenthey met. Our results are in accord with our generalhypothesis that stochastic effects associated with higherspecies richness dilute species associations in highly diversecommunities, making them weak on average [15,22,26].(a) Analysis of overall patterningThe strongest interactions between plants occur betweenthose that are close together in space. Because the interspecificspacing of plants is closely related withcoexistence mechanisms [29–31], it is of basic interestto find out how individuals of different species arearranged with respect to each other [22]. If the spatialarrangement of species is independent, individuals ofspecies j co-occur within a neighbourhood of species i ata frequency no different from that expected by chancealone. This makes the identity of the nearest neighboursof individual plants less predictable if species richnessis higher.We found that larger trees (i.e. greater than 10 cm dbh)at the BCI forest, which hosts in total more than 300species of trees and shrubs, showed in more than 80 percent of all cases no detectable spatial association at a 6 mneighbourhood. However, this is likely to be a conservativeestimate because the homogeneous Poisson null model

Downloaded from rspb.royalsocietypublishing.org on September 6, 20123318 T. Wiegand et al. Spatial associations among tree speciesdoes not consider the observed clustering and because weincluded only species with more than 70 individuals. Forspecies with fewer individuals, significant effects will beeven less detectable. Thus, our analysis for the BCI datasupports the assertion of independent spatial arrangementof species, at least at small neighbourhoods of severalmetres where trees typically interact.Notably, the proportion of species pairs that showed atsmaller scale (i.e. 6 m) the no association type decreasedwith species richness from 82 per cent at BCI to 58 percent at Sinharaja and 36 per cent at the temperate CBSplot. This result is compatible with the dilution effect ofspecies richness [20,27]. At larger neighbourhoods of saygreater than 30 m, however, species associations weredominated by segregation. This is understandable becausein larger neighbourhoods stochastic effects become weakerbecause a given plant at BCI, Sinharaja and CBS has nowon average 118, 189 and 116 neighbours within 30 m,respectively, compared with 4.8, 7.6 and 4.7 neighbourswithin 6 m, respectively. Species clustering (assertion (i))now produces a detectable signal of segregation or partialoverlap and substantially reduces the cases of no association.These effects are presumably owing to first-orderhabitat effects or effects of dispersal limitation.The different proportions of the no association type atlarger neighbourhoods of greater than 40 m in the twotropical forests (38% of all species pairs at BCI versus14% at Sinharaja) may be a result of heterogeneity ofthe physical environment, which can produce negativeassociation patterns between species if the two speciesshow dissimilar habitat associations. The Sinharaja plotshows strong species–habitat association [37], whereasthe BCI plot shows a relatively low degree of species–habitat association [43,44]. However, direct comparisonof the species pairs in our scheme showed also thatthe higher habitat structuring at the Sinharaja plotincreased the strength of the different association patternssuch as segregation (electronic supplementary material,figure A1). Habitat structuring may not be the mainreason for the large proportion of segregation observedat the CBS forest because this plot has weak topographicstructuring. The signal of segregation was also enhancedby the high proportion of species pairs (more than 20%;figure 4c) that showed repulsive interactions that wereprobably caused by competitive interactions [21]. However,it is difficult here to discern between segregationpatterns that are caused by direct negative interactionbetween species or edaphic effects because soil patchinesscould have also determined the segregation pattern at theCBS forest [45].If conspecifics tend to form clusters that are segregatedfrom clusters of other species, the importance of intraspecificcompetition is increased relative to interspecific competitionand coexistence may be enhanced [10,30]. Whereas recentstudies in species-poor systems showed that such an interspecificsegregation effect may indeed lead to dominanceof intraspecific competition over interspecific competition[29], our results (electronic supplementary material,figure A2) suggest that intraspecific aggregation in speciesrichcommunities may not be strong enough to yield localdominance. This is in agreement with our specific hypothesisH1b and illustrates the dilution effect of speciesrichness with respect to the coexistence mechanism of intraspecificclustering and interspecific segregation [29–31].Proc. R. Soc. B (2012)Higher species richness means that fewer species will reachthe high local dominance required for benefiting from thiscoexistence mechanism.(b) Analysis of species interactionsThe spatial patterns of the larger trees at the three forestplots are the outcome of different processes and mechanismsduring regeneration and growth. After approximatelyfactoring out the larger-scale effects of the environment,we found that for the two tropical forests there are significantsmall-scale interactions only for a few species pairs.This suggests that the spatial patterns of larger treesshowed equilibrated spatial patterns [22]. Similarly,Volkov et al. [27] found that the collective effects of thepairwise interspecific interaction strengths of the 20most abundant species at BCI were weak comparedwith the intraspecific interactions.However, a low proportion of small-scale interspecificinteractions does not mean that certain species may notmaintain numerous interactions with other species. Thedistribution function of the number of significant interactionsper species was close to but not fully randomand contained some species with more interactions toother species than expected by chance (electronic supplementarymaterial, figure A4). To our surprise, wefound for the BCI forest (electronic supplementarymaterial, figure A4a) and the Sinharaja forest (electronicsupplementary material, figure A4b) almost the samedistribution function.When putting together the results of the three forests(figure 4), we find that the importance of negative smallscaleinterspecific interactions decreases with species richness.They are relatively unimportant at the two tropicalforests; however, the CBS forest showed significant negativeeffects in approximately one-fourth of all cases, which areprobably caused by competitive interactions (see above).These results are in agreement with the specific hypothesisH2 that the occurrence of small-scale interactions betweenspecies would be diluted by stochastic effects in species-richsystems. At the species-poorer CBS forest, significant smallscaleeffects were more likely if the two species were moreabundant (electronic supplementary material, appendixC) because more direct encounters between heterospecificindividuals will occur. This is also in accordance withrecent studies in ecological networks that suggest thestrong impact of abundance on interaction strength (e.g.the ‘abundance-asymmetry hypothesis’ [46]). Consequently,the interaction strength at the diverse BCI andSinharaja forests depended only weakly on abundances(electronic supplementary material, appendix C). Similarresults have been obtained for other species-rich plantcommunities [20,24].In conclusion, our study suggests that the independentassertion may be indeed a possible approximation in thelimit of very species-rich communities envisaged by stochasticgeometry theories of biodiversity. However, ourresults also suggest that this assertion may break downin communities with lower species richness. Whatremains for future research is to find out if the observeddepartures from independence are large enough tocause problems for the biodiversity theories summarizedby McGill [15]. The general hypothesis on speciesdilution formulated here constitutes an important50

Downloaded from rspb.royalsocietypublishing.org on September 6, 2012Spatial associations among tree species T. Wiegand et al. 3319contribution towards the understanding of the factorsgoverning species-rich communities. It is based onarguments of stochastic geometry related with the probabilityof intraspecific species encounters and suggests amechanism that can explain why species placement inspecies-rich communities cannot be distinguished fromindependence. Clearly, our support for the speciesdilution hypotheses is based only on data of three forestsand we cannot exclude the possibility that it is causedby idiosyncrasies of the different forests rather than by ageneral rule. To further test these hypotheses, it isnecessary to conduct studies with the methodologyused here for additional forest plots and plots of othervegetation types.The BCI forest dynamics research project was made possibleby National Science Foundation grants to StephenP. Hubbell: DEB-0640386, DEB-0425651,DEB-0346488,DEB-0129874, DEB-00753102, DEB-9909347, DEB-9615226, DEB-9615226, DEB-9405933, DEB-9221033,DEB-9100058, DEB-8906869, DEB-8605042, DEB-8206992, DEB-7922197, by support from the Center forTropical Forest Science, the Smithsonian Tropical ResearchInstitute, the John D. and Catherine T. MacArthurFoundation, the Mellon Foundation, the Celera Foundationand numerous private individuals and through the hardwork of over 100 people from 10 countries over the last twodecades. The Sinharaja plot research was supported by theCenter for Tropical Forest Science of the SmithsonianTropical Research Institute and the Arnold Arboretum ofHarvard University, the John D. and CatherineT. MacArthur Foundation (94-29503 and 98-55295), theNational Science Foundation, USA (0090311), and theNational Institute for Environmental Studies, Japan. All plotprojects are part of the Center for Tropical Forest Science, aglobal network of large-scale demographic tree plots. X.W.and Z.H. were supported by the Knowledge InnovationProgram of the Chinese Academy of Sciences (KSCX2-EW-Z-5, KZCX2-YW-QN402 and KZCX2-EW-401) and theNational Natural Science Foundation of China (40971286).S.G. was supported by the ERC advanced grant no. 233066to T.W. We thank two anonymous referees for constructivecriticism on earlier drafts of the manuscript.REFERENCES1 Ricklefs, R. E. 1990 Ecology. 3rd edn. New York, NY:W. H. Freeman and Company.2 Brown, J. H., Mehlman, D. H. & Stevens, G. C. 1995Spatial variation in abundance. Ecology 76, 2028–2043.(doi:10.2307/1941678)3 Gause, G. F. 1934 The struggle for existence. Baltimore,MD: Williams & Wilkins.4 Hutchinson, G. E. 1961 The paradox of the plankton.Am. Nat. 95, 137–145. (doi:10.1086/282171)5 Hubbell,S.P.2001The unified neutral theory of biodiversityand biogeography. Princeton, NJ: Princeton UniversityPress.6 Huisman, J. & Weissing, F. J. 1999 Biodiversity ofplankton by species oscillations and chaos. Nature 402,407–410. (doi:10.1038/46540)7 Losos, E. C. & Leigh, E. G. 2004 Tropical forest diversityand dynamism: findings from a large-scale plot network.Chicago, IL: University of Chicago Press.8 Gentry, A. H. 1988 Tree species richness of upperAmazonian forests. Proc. Natl Acad. Sci. USA 85,156–159. (doi:10.1073/pnas.85.1.156)9 Tilman, D. 1982 Resource competition and communitystructure. Monographs in population biology. Princeton,NJ: Princeton University Press.Proc. R. Soc. B (2012)5110 Chesson, P. 2000 General theory of competitive coexistencein spatially-varying environments. Theor. Popul.Biol. 58, 211–237. (doi:10.1006/tpbi.2000.1486)11 Wright, S. J. 2002 Plant diversity in tropical forests: areview of mechanisms of species coexistence. Oecologia130, 1–14. (doi:10.1007/s004420100809)12 Clements, F. E. 1916 Plant succession: an analysis of thedevelopment of vegetation. Publication no. 242. Washington,DC: Carnegie Institute of Washington.13 Gleason, H. A. 1926 The individualistic concept of theplant association. Bull. Torrey Bot. Club 53, 7–26.(doi:10.2307/2479933)14 Missa, O. 2005 The unified neutral theory of biodiversityand biogeography: alive and kicking. Bull. Br. Ecol. Soc.36, 12–17.15 McGill, B. J. 2010 Towards a unification of unified theoriesof biodiversity. Ecol. Lett. 13, 627–642. (doi:10.1111/j.1461-0248.2010.01449.x)16 Plotkin, J. B., Potts, M. D., Leslie, N., Manokaran, N.,LaFrankie, J. & Ashton, P. S. 2000 Species-area curves,spatial aggregation, and habitat specialization in tropicalforests. J. Theor. Biol. 207, 81–99. (doi:10.1006/jtbi.2000.2158)17 Morlon, H., Chuyong, G., Condit, R., Hubbell, S. P.,Kenfack, D., Duncan, T., Renato, V. & Green, J. L.2008 A general framework for the distance-decay of similarityin ecological communities. Ecol. Lett. 9, 914–917.18 Condit, R. et al. 2000 Spatial patterns in the distributionof tropical tree species. Science 288, 1414–1418. (doi:10.1126/science.288.5470.1414)19 McGill, B. J. et al. 2007 Species abundance distributions:moving beyond single prediction theories to integrationwithin an ecological framework. Ecol. Lett. 10,995–1015. (doi:10.1111/j.1461-0248.2007.01094.x)20 Lieberman, M. & Lieberman, D. 2007 Nearest-neighbortree species combinations in tropical forest: the role ofchance, and some consequences of high diversity. Oikos116, 377–386. (doi:10.1111/j.2006.0030-1299.15370.x)21 Wang, X., Wiegand, T., Hao, Z., Li, B., Ye, J. & Zhang, J.2010 Species associations in an old-growth temperateforest in north-eastern China. J. Ecol. 98, 674–686.(doi:10.1111/j.1365-2745.2010.01644.x)22 Wiegand, T., Gunatilleke, C. V. S. & Gunatilleke, I. A. U.N. 2007 Species associations in a heterogeneous SriLankan Dipterocarp forest. Am. Nat. 170, E77–E95.(doi:10.1086/521240)23 Wiegand, T., Gunatilleke, C. V. S., Gunatilleke, I. A. U.N. & Huth, A. 2007 How individual species structurediversity in tropical forests. Proc. Natl Acad. Sci. USA104, 19 029–19 033. (doi:10.1073/pnas.0705621104)24 Perry, G. L. W., Enright, N. J., Miller, B. P. & Lamont,B. B. 2009 Nearest-neighbour interactions in speciesrichshrublands: the roles of abundance, spatial patternsand resources. Oikos 118, 161–174. (doi:10.1111/j.1600-0706.2008.16947.x)25 Hubbell, S. P. & Foster, R. B. 1986 Biology, chance andhistory and the structure of tropical rain forest treecommunities. In Community ecology (eds J. Diamond &T. J. Case), pp. 314–329. New York, NY: Harperand Row.26 Hubbell, S. P. 2006 Neutral theory and the evolution ofecological equivalence. Ecology 87, 1387–1398. (doi:10.1890/0012-9658(2006)87[1387:NTATEO]2.0.CO;2)27 Volkov, I., Banavar, J. R., Hubbell, S. P. & Maritan, A.2009 Inferring species interactions in tropical forests.Proc. Natl Acad. Sci. USA 106, 13 854–13 859. (doi:10.1073/pnas.0903244106)28 Law, R., Illian, J., Burslem, D. F. R. P., Gratzer, G.,Gunatilleke, C. V. S. & Gunatilleke, I. A. U. N. 2009Ecological information from spatial patterns of plants:

Downloaded from rspb.royalsocietypublishing.org on September 6, 20123320 T. Wiegand et al. Spatial associations among tree speciesinsights from point process theory. J. Ecol. 97, 616–628.(doi:10.1111/j.1365-2745.2009.01510.x)29 Raventós, J., Wiegand, T. & De Luis, M. 2010 Evidencefor the spatial segregation hypothesis: a test with nineyearsurvivorship data in a Mediterranean fire-proneshrubland show that interspecific and density-dependentspatial interactions dominate. Ecology 91, 2110–2120.(doi:10.1890/09-0385.1)30 Pacala, S. W. 1997 Dynamics of plant communities. InPlant ecology (ed. M. J. Crawley), pp. 532–555. Oxford,UK: Blackwell.31 Stoll, P. & Prati, D. 2001 Intraspecific aggregation alterscompetitive interactions in experimental plant communities.Ecology 82, 319–327. (doi:10.1890/0012-9658(2001)082[0319:IAACII]2.0.CO;2)32 Wiegand, T. & Moloney, K. A. 2004 Rings, circles, andnull-models for point pattern analysis in ecology. Oikos104, 209–229. (doi:10.1111/j.0030-1299.2004.12497.x)33 Illian, J. B., Penttinen, A., Stoyan, H. & Stoyan, D. 2008Statistical analysis and modelling of spatial point patterns.Chichester, UK: Wiley and Sons.34 Hubbell, S. P., Foster, R. B., O’Brien, S. T., Harms, K. E.,Condit, R., Wechsler, B., Wright, S. J. & de Lao, S. L.1999 Light-gap disturbances, recruitment limitation,and tree diversity in a neotropical forest. Science 283,554–557. (doi:10.1126/science.283.5401.554)35 Hubbell, S. P. & Foster, R. B. 1983 Diversity of canopytrees in neotropical forest and implications for conservation.In Tropical rain forest: ecology and management(eds S. Sutton, T. Whitmore & A. Chadwick), pp.25–41. London, UK: Blackwell Scientific.36 Condit, R. 1998 Tropical forest census plots. New York, NY:Springer.37 Gunatilleke, C. V. S., Gunatilleke, I. A. U. N., Ashton,P. S., Ethugala, A. U. K., Weerasekara, N. S. &Esufali, S. 2004 Sinharaja forest dynamics plot, SriLanka. In Tropical forest diversity and dynamism: findingsfrom a large scale plot network (eds E. C. Losos & E. G.Leigh), pp. 599–608. Chicago, IL: The University ofChicago Press.38 Hubbell, S. P., Condit, R. & Foster, R. B. 2005 BarroColorado forest census plot data. See http://ctfs.arnarb.harvard.edu/webatlas/datasets/bci/.39 Ripley, B. D. 1981 Spatial statistics. New York, NY: Wiley.40 Lotwick, H. W. & Silverman, B. W. 1982 Methods foranalysing spatial processes of several types of points.J. R. Stat. Soc. Ser. B Stat. Methodol. 44, 403–413.41 Stoyan, D. & Stoyan, H. 1994 Fractals, random shapes andpoint fields. New York, NY: Wiley.42 Loosmore, N. B. & Ford, E. D. 2006 Statistical inferenceusing the G or K point pattern spatial statistics. Ecology87, 1925–1931. (doi:10.1890/0012-9658(2006)87[1925:SIUTGO]2.0.CO;2)43 Harms, K. E., Condit, R., Hubbell, S. P. & Foster, R. B.2001 Habitat associations of trees and shrubs in a 50-haneotropical forest plot. J. Ecol. 89, 947–959. (doi:10.1111/j.1365-2745.2001.00615.x)44 Comita, L. S., Condit, R. & Hubbell, S. P. 2007 Developmentalchanges in habitat associations of tropical trees.J. Ecol. 95, 482–492. (doi:10.1111/j.1365-2745.2007.01229.x)45 Wang, X., Wiegand, T., Wolf, A., Howe, R., Davies, S. J. &Hao, Z. 2011 Spatial patterns of tree species richness intwo temperate forests. J. Ecol. 99, 1382–1393. (doi:10.1111/j.1365-2745.2011.01857.x)46 Vázquez, D. P., Melian, C. J., Williams, N. M.,Bluthgen, N., Krasnov, B. R. & Poulin, R. 2007 Speciesabundance and asymmetric interaction strength in ecologicalnetworks. Oikos 116, 1120–1127. (doi:10.1111/j.0030-1299.2007.15828.x)Proc. R. Soc. B (2012)52

Local-Scale Drivers of Tree Survival in a TemperateForestXugao Wang 1 , Liza S. Comita 2,3 , Zhanqing Hao 1 *, Stuart J. Davies 4 ,JiYe 1 , Fei Lin 1 , Zuoqiang Yuan 11 State Key Laboratory of Forest and Soil Ecology, Institute of Applied Ecology, Chinese Academy of Sciences, Shenyang, Liaoning Province, People’s Republic of China,2 Department of Evolution, Ecology, and Organismal Biology, The Ohio State University, Columbus, Ohio, United States of America, 3 Smithsonian Tropical ResearchInstitute, Balboa Ancón, Republic of Panamá, 4 Center for Tropical Forest Science, Arnold Arboretum, Harvard University and Smithsonian Tropical Research Institute,Boston, Massachusetts, United States of AmericaAbstractTree survival plays a central role in forest ecosystems. Although many factors such as tree size, abiotic and bioticneighborhoods have been proposed as being important in explaining patterns of tree survival, their contributions are stillsubject to debate. We used generalized linear mixed models to examine the relative importance of tree size, local abioticconditions and the density and identity of neighbors on tree survival in an old-growth temperate forest in northeasternChina at three levels (community, guild and species). Tree size and both abiotic and biotic neighborhood variablesinfluenced tree survival under current forest conditions, but their relative importance varied dramatically within and amongthe community, guild and species levels. Of the variables tested, tree size was typically the most important predictor of treesurvival, followed by biotic and then abiotic variables. The effect of tree size on survival varied from strongly positive forsmall trees (1–20 cm dbh) and medium trees (20–40 cm dbh), to slightly negative for large trees (.40 cm dbh). Among thebiotic factors, we found strong evidence for negative density and frequency dependence in this temperate forest, asindicated by negative effects of both total basal area of neighbors and the frequency of conspecific neighbors. Among theabiotic factors tested, soil nutrients tended to be more important in affecting tree survival than topographic variables.Abiotic factors generally influenced survival for species with relatively high abundance, for individuals in smaller size classesand for shade-tolerant species. Our study demonstrates that the relative importance of variables driving patterns of treesurvival differs greatly among size classes, species guilds and abundance classes in temperate forest, which can furtherunderstanding of forest dynamics and offer important insights into forest management.Citation: Wang X, Comita LS, Hao Z, Davies SJ, Ye J, et al. (2012) Local-Scale Drivers of Tree Survival in a Temperate Forest. PLoS ONE 7(2): e29469. doi:10.1371/journal.pone.0029469Editor: Kurt O. Reinhart, USDA-ARS, United States of AmericaReceived September 22, 2011; Accepted November 29, 2011; Published February 13, 2012Copyright: ß 2012 Wang et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permitsunrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Funding: This study was supported by the Knowledge Innovation Program of the Chinese Academy of Sciences (KSCX2-EW-Z-5 and KZCX2-YW-QN402) and theNational Natural Science Foundation of China (40971286, 31011120470 and 31061160188). XW acknowledges the support of a fellowship from the Center forTropical Forest Science of the Smithsonian Tropical Research Institute and the HSBC Climate Partnership. LSC acknowledges the support of a fellowship from theNational Center for Ecological Analysis and Synthesis, a center funded by NSF (Grant #EF-0553768), the University of California, Santa Barbara, and the State ofCalifornia. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.Competing Interests: The authors have declared that no competing interests exist.* E-mail: hzq@iae.ac.cnIntroductionForests are an important and substantial part of the terrestrialbiosphere, providing a number of key ecosystem functions such ascarbon sequestration, and water and nutrient cycling. Treemortality plays a critical role in forests [1]. It can determine forestdynamics or succession, alter nutrient cycling, and furthercontribute to tree species coexistence [2–4]. Without a properunderstanding of the patterns and determinants of tree mortality,our overall understanding of forest dynamics is severely hampered.However, studies of tree mortality are hindered by the fact thattrees are long-lived and mortality rates are often low, particularlyfor mature trees, making it difficult to gather sufficient observationsand to gain insights into the factors influencing mortality andsurvival [5]. Ecologists are often forced to make overly simplisticassumptions about tree mortality processes. For instance, treemortality has been hypothesized to be random, with all treeshaving an equal probability of survival [6]. However, in variousnatural forest communities numerous studies have rejected thisrandom mortality hypothesis [7–9]. This is because trees are sessileand tree mortality is likely to be strongly affected by intrinsicproperties of individuals and by their local immediate neighborhoods.Many studies have shown that size is an important intrinsicproperty of trees that strongly affects survival. The probability ofsurvival is expected to increase as trees get larger and acquirereserves to withstand environmental stress [9–11]. In addition,larger trees have a competitive advantage over smaller trees due tothe asymmetric nature of competition for light [12]. However, treesurvival is known to vary by tree size and these relationships areknown to vary by species [3,13–19].Local neighborhood conditions broadly include two majorclasses of factors, biotic and abiotic variables. The effects of bioticand abiotic factors represent two important explanations forspecies coexistence in ecological communities: frequency (ordensity) dependence and resource niche partitioning, respectively[20–21]. In tree communities, survival is commonly observed to befrequency- or density-dependent, based on several studies thatmodeled survival as a function of the number, size and identity ofindividuals in the local biotic neighborhood [5,8–9,22–28].PLoS ONE | www.plosone.org 1 February 2012 | Volume 7 | Issue 2 | e2946953

Tree Survival in a Chinese Temperate ForestSpecifically, trees tend to have a higher probability of mortalitywhere conspecific neighbors are denser, closer, or proportionatelymore abundant than heterospecific neighbors [8,23,29–30]. Suchnegative effects of conspecific neighbors could result from strongintraspecific competition for resources or from host-specificnatural enemies, such as herbivores and pathogens that areattracted by or spread rapidly through high density patches ofsusceptible individuals. Negative conspecific density and frequencydependence are widely recognized as prominent mechanisms ofspecies coexistence, and several hypotheses, such as the Janzen-Connell hypothesis, consider their effects on community assembly[31–34]. Although there have been several studies of neighboreffects on tree survival, few have simultaneously considered theeffects of both biotic and abiotic factors, although in some casesresearchers have tried to select environmentally homogeneousareas [23,35]. However, even local environments are heterogeneousdue to variation in topography and soils [36–37]. Failing toconsider the effects of these abiotic factors on plant performancecan result in incorrect inferences about the importance of densitydependentneighborhood effects [8]. For example, a species willlikely attain higher densities in its optimal habitat due to increasedsurvival. As a result, survival could then be positively correlatedwith conspecific density, potentially masking negative impacts ofconspecific neighbors.Studies have also shown that species with different life historystrategies respond differently to intrinsic and extrinsic factors.Shade tolerance can be a key trait in determining patterns of treegrowth and survival [23,38]. In addition to being less sensitive toshading by neighbors, shade-tolerant species are less susceptible toenemy attack than light-demanding species, based on differencesin the allocation of resources to defense vs. growth [39–41]. As aresult, shade tolerance should be an important determinant oftrees’ reactions to their local biotic neighborhood [42]. In addition,some studies have found that species abundance is related tospecies’ responses to their neighbors. For example, rare tropicaltree species have been shown to be more sensitive to the density ofneighboring conspecifics [23,43]. Within a species, tree size mayinfluence the effects of biotic and abiotic variables on survival.Smaller individuals may be more sensitive to neighbor density, dueto asymmetric competition with taller individuals. Smallerindividuals may also be more likely to show effects of abiotichabitat variables on survival, since large trees are typically found inpreferred habitats due to environmental filtering occurring atearlier life stages [44–45].In this study we assessed the relative importance of tree size,abiotic conditions, and the local biotic neighborhood in drivingpatterns of tree survival in an old-growth temperate forest innortheastern China at three organizational levels (community,guild and individual species). We used data from multiple censusesof a large (25 ha), fully-mapped forest plot to address the followingquestions: (1) Does tree size have a consistent positive effect on treesurvival? (2) Are abiotic or biotic neighborhood factors moreimportant in influencing spatial patterns of tree survival? (3) Canlife history strategy or ecological guild be used to predict therelative importance of factors driving tree survival?Materials and MethodsStudy areaThe study area was the Changbai Nature Reserve, whichextends along the border of China and North Korea from 127u429to 128u179E and 41u439 to 42u269N. It is one of the largestbiosphere reserves in China and has been spared from logging andother severe human disturbances. The Changbai Nature Reservejoined the World Biosphere Reserve Network under theUNESCO Man and the Biosphere Programme in 1980. Thereserve is ca. 200,000 ha, and elevation ranges from 740 m to2,691 m at the summit of Changbai Mountain.Our study area is located in broad-leaved Korean pine (Pinuskoraiensis) mixed forest, the most common vegetation type in theregion. Mean annual precipitation is approximately 700 mm andmost of this occurs from June to September (480–500 mm). Meanannual temperature is 2.8uC, with a January mean of 213.7uC,and a July mean of 19.6uC. Mean age of overstory trees is ca. 300years. Common tree species include P. koraiensis, Tilia amurensis,Quercus mongolica, Fraxinus mandshurica, Ulmus japonica, and Acer mono[46].Field methodsA 25 ha (5006500 m) forest plot was established in the summerof 2004 in Changbaishan (CBS) Nature Reserve. The plot,representative of forests in the area, is located in the core zone ofChangbai Nature Reserve. Mean elevation in the plot is 801.5 mand ranges from 791.8 m to 809.5 m. All free-standing trees atleast 1 cm in diameter at breast height (dbh; 1.3 m above ground)were tagged, measured and identified to species, and theirgeographic coordinates were recorded following a standard fieldprotocol [47]. The first CBS plot census started in July 2004 andended in September 2004, and the second census was carried outbetween July 2009 and August 2009. The status of trees, as live ordead, was recorded in the second census. The CBS plot contains52 species with stems $1 cm dbh, belonging to 32 genera and 18families [46]. No specific permits were required for the describedfield studies.Biotic neighborhood variablesTo quantify the local biotic neighborhood, we used thefrequency of conspecific neighbors (basal area of conspecificstems/total basal area of all stems) and the total basal area of allstems $1 cm dbh within 20 m of the center of each focal tree. Wechose 20 m because tree species interactions have been shown todisappear beyond 20 m [48].Abiotic factorsElevation was measured at the corner of each 20 m620 mquadrat in the 25 ha plot. Elevation values for these 20 m620 mquadrats were interpolated to calculate the elevation of the cornersof the 5 m65 m subquadrats. Slope and aspect values were thencalculated for each subquadrat.In 2007, soils were sampled using a regular grid of points every30 m. Two additional sample points at 2, 5, or 15 m were selectedin a random compass direction from the grid point to capturevariation in soil nutrients at finer scales. In total, 967 points weresampled in the 25 ha plot. At each sample point 500 g of topsoil(0–10 cm depth) was collected, and eight soil properties (pH,organic matter, total N, available N, total P, available P, total Kand available K) were analyzed according to Lu [49]. Soil pH inwater (1:1) was measured by Beckman glass electrode. Soil organicmatter was determined colorimetrically by the dichromateoxidation method. Total N was determined colorimetrically onthe KCl extracts using the Kjeldahl method. Available N wasalkali dispelled by 1 mol NaOH L 21 . Total P was determined bymolybdenum antimony blue colorimetry after extraction usingH 2 SO 4 -HClO 4 . Available P content was extracted using a0.05 mol/LHCL-0.025 mol/LH 2 SO 4 solution. Total K wasdetermined by digesting in hydrofluoric acid and then measuredby atomic absorption spectrometer. Available K was extractedPLoS ONE | www.plosone.org 2 February 2012 | Volume 7 | Issue 2 | e2946954

Tree Survival in a Chinese Temperate Forestwith 1 mol NH4AcL-1 and then measured by atomic absorptionspectrometer.Spatial predictions for 5 m65 m quadrats were obtained usinggeostatistical methods (ordinary kriging). Because these soilvariables were strongly correlated with each other, we computedthe principal components (PCs) from the eight soil variables andused only the first two components as condensed variables becausetogether they explained 86.6% of the total variance in soilvariables (Table S1). Variable condensing can reduce thepossibility of overfitting the models [37].Data analysisWe modeled the probability of an individual tree surviving the5-year census interval as a function of initial tree size in the firstcensus (i.e. dbh) and the abiotic and biotic neighborhood factorsdescribed above, using generalized liner mixed models (GLMM)with binomial errors with a logit link function (i.e. logisticregression). GLMM is now widely used in ecology [50] anddetails on GLMM can be found in Zurr et al. [51] Biotic factorsincluded the total basal area of all neighboring trees and thefrequency of conspecifics (basal area of conspecific stems/totalbasal area of all stems) within 20 m of the focal tree. For abioticfactors, we included three topographical variables (elevation, slopeand aspect) and the two PC axes of soil variables. We assigned toeach individual tree the abiotic factors of the 565 m quadratwhere it was located. All abiotic and biotic variables, as well as logtransformedinitial tree size (dbh), entered the model as fixedeffects. For all these variables, values were standardized bysubtracting the mean value of the variables (across all individualsin the analysis) and dividing by 1 standard deviation. This allowsfor a direct comparison of the relative importance of theseexplanatory variables [52]. The mean and range of all explanatoryvariables used in the analysis are listed in Table S2. To avoid edgeeffects, we excluded all potential target trees that were within 20 mof the edge of the plot from the analyses.Tree survival was analyzed at three different scales. First, weincluded all trees in the plot and conducted a community-levelanalysis with all species. Species was included as a random effect inthe analysis. Second, we assessed the effect of species life-historytraits by dividing species into different life-history guilds andanalyzing each guild separately, with species as a random effect.Previous studies in the CBS forest found that species distributionsand interactions varied among different guilds [48,53]. Therefore,we expected that the effects of explanatory variables on treesurvival would also differ among guilds. Specifically, we classifiedspecies into three shade-tolerant guilds (shade-tolerant, midtolerantand light-demanding) [54]; we also grouped species intofour abundance classes based on their abundance in the CBS 25-ha plot (very rare: 1–100, rare: 100–1000, common: 1000–5000,and very common: .5000 individuals). In addition, we dividedindividuals into three size classes (1–20 cm, 20–40 cm and.40 cm dbh). In these analyses, species was also included as arandom effect. All explanatory variables were standardized asdescribed above. Finally, because niche theory predicts thatspecies will be affected differently by abiotic variables, weperformed species-level analyses, by separately analyzing each ofthe 20 species with .100 stems.To test the relative importance of tree size, abiotic variables andbiotic variables in affecting tree survival, we constructed fourcandidate models with different variables: (1) only tree size; (2) treesize and abiotic variables; (3) tree size and biotic variables; (4) allvariables included. We used Akaike information criterion (AIC)[55] to identify the best fit model, and then used the results fromthe best fit model to evaluate the effect and uncertainty ofindividual variables. Also, we validated the best fit models withNagelkerke’s R 2 N and Somer’s D XY as measures of modelpredictive and discriminative ability, respectively. Nagelkerke’sR 2 N is a pseudo R2 (classical R 2 is not appropriate for logisticmodels) to evaluate model predictions by calculating the square ofcorrelation coefficient between observed and predicated values[56–57]. Somers’ D XY measures the correlation between predictedsurvival probability and a binary (0–1) variable [57–58]. The D XYvalues range from 21 (where all live trees were classified as deadand vice versa) to 1 (all classifications were correct). All calculationswere carried out in R version 2.10.0 [59], using the ‘‘lme4’’package [60] with the Laplace method [50].ResultsCommunity-level analysisWe assessed the role of tree size, the biotic neighborhood (i.e.frequency of neighboring conspecific basal area and total basalarea of all neighbors), and local abiotic variables (i.e. soil propertiesand topography) on tree survival using generalized liner mixedmodels and compared the likelihood of models of increasingcomplexity to determine the best fit model. At the communitylevel,the model including all factors proved the best fit for treesurvival (as indicated by the lowest AIC; Table 1), suggesting thattree size, abiotic and biotic factors all have significant effects ontree survival in the CBS temperate forest plot. However, lowvalues of Nagelkerke’s R 2 N (0.118) and Somer’s D XY (0.452)(Table 2), measures of goodness of fit, suggest that much of thevariation in probability of mortality remained unexplained. Ofthese influential factors, tree size (i.e. dbh) had the strongest impacton survival (Figure 1), with larger trees having an increasedprobability of survival. Frequency of conspecifics and total basalarea of all neighboring trees within 20 m both had negative effectson survival. In addition, we found evidence that tree survival wasinfluenced by soil factors (Figure 1). Specifically, the probability ofsurvival was affected by soil PC1 (the first axis of a PCA using eightsoil variables), which was associated with high concentrations oftotal K and low concentrations of organic matter and total N(Table S1). In contrast, topographic factors (elevation, slope andaspect) had no significant effects on tree survival (Figure 1).Guild-level analysisWhen analyzing different life history guilds separately, we foundthat the relative importance of factors influencing tree survivalvaried among guilds. Similar to the community-level analysis, thevalues of Nagelkerke’s R 2 N and Somer’s D XY of the best fit modelfor each guild were relatively low (Table 2).Among the three shade-tolerance guilds (light-demanding, midtolerantand shade-tolerant), there were several differences in therelative importance of factors effecting tree survival. For the lightdemandinggroup, the model with the lowest AIC only includedtree size and biotic factors. That model could not be statisticallydifferentiated from the model with tree size and abiotic factors (i.e.the difference in AIC was ,2; Table 1), but none of the individualabiotic variables analyzed had a detectable effect on tree survival(Figure 2). The model with tree size and biotic factors was the bestfit for the mid-tolerant group, while the model with tree size, bioticand abiotic factors was the best fit for the shade-tolerant group.Similar to the community-level results, tree size had the strongestpositive effect on tree survival and total basal area had a significantnegative effect for all shade-tolerance groups. Frequency ofneighboring conspecifics had a marginally significant negativeeffect for all three groups (light demanding: P = 0.056; midtolerant:P = 0.063; shade-tolerant: P = 0.059). None of thePLoS ONE | www.plosone.org 3 February 2012 | Volume 7 | Issue 2 | e2946955

Tree Survival in a Chinese Temperate ForestTable 1. AIC values of the generalized linear mixed models of tree survival at the community and guild level.CandidatemodelsAllspecies Abundance Shade-tolerance Tree sizeVery rare Rare CommonVerycommonMid-tolerantSmall treesmediumtreeslargetreesSize 19753.71 580.55 3519.88 6588.36 9053.79 16175.32 2053.64 1525.81 16798.93 961.73 577.96Abiotic 19718.51 584.94 3519.56 6565.15 9029.77 16140.39 2058.27 1523.01 16769.66 961.65 586.13Biotic 19712.22 581.45 3512.13 6583.34 9017.89 16148.04 2050.55 1521.42 16758.51 964.16 570.11All factors 19698.53 585.89 3515.28 6563.68 9008.16 16130.69 2055.53 1523.50 16745.5 965.14 575.14AIC values of the most likely models are shown in bold. Size, Abiotic, Biotic and All factors represent models with only tree size, tree size and abiotic factors, tree size andbiotic factors, and all factors included, respectively. Very rare, rare, common, very common denote abundance classes with 1–100, 100–1000, 1000–5000 and .5000individuals, respectively. Small, medium, and large tree size classes include individuals with 1–20 cm, 20–40 cm, and .40 cm dbh, respectively.doi:10.1371/journal.pone.0029469.t001topographic factors had a significant effect, while soil factors (SoilPC1) significantly affected survival for the shade-tolerant group(Figure 2).We also fit models separately for four abundance classes (veryrare, rare, common, and very common). For the abundanceclass with very rare species, two best fit models were found(Table 1). Both models included the effect of tree size, and onealso included biotic factors. The model with tree size and bioticfactors was the best fit for rare species. The best fit models forcommon and very common species included all factors. Amongthese best fit models, tree size had the strongest positive effect ontree survival. Frequency of neighboring conspecific basal areaand total basal area had significant negative effects on survivalof trees of rare, common and very common species (Figure 3).For the very rare class, the parameter value for conspecificneighbor frequency was more strongly negative than for theother classes, but was not significant. Effects of soil factors werefound for common and very common classes, but topographicalfactors only contributed to survival for the very common class(Figure 3).The factors driving tree survival also varied when dividingindividuals into three tree size classes (small, medium and large).For large individuals (dbh.40 cm), the best fit model includedtree size and biotic factors, while the best model for the small sizegroup (1–20 cm dbh) included all factors (Table 1). For mediumtrees (20–40 cm dbh), the best fit model included tree size, but itwas not statistically different from the model that included tree sizeand abiotic factors (i.e. the difference in AIC was ,2; Table 1).However, none of abiotic factors analyzed had a detectable effecton tree survival for the medium size group (Figure 4). In the threesize groups, tree size did not always show strong positive effects onsurvival. For the small and medium size groups, survival increasedstrongly with increasing tree size. For the large size group,however, the relationship was slightly negative, but not significant.Total basal area of all neighboring trees had strong negative effectson tree survival for the small and large size groups, but frequencyof neighboring conspecific basal area only had a significantnegative effect on trees in the small size group (Figure 4).Topographic factors had no significant effects on tree survival forany of the three size groups, but the survival of small trees waseffected by soil factors (Soil PC1; Figure 4).Species-level analysisThere were large differences in the best fit models for twentyspecies that were analyzed individually (Table 3). Values ofNagelkerke’s R 2 N and Somer’s D XY of the best fit models for thesespecies differed greatly (Table 4). Nagelkerke’s R 2 N ranged from0.026 (Syringa reticulata) to 0.304 (Tilia mandshurica). Somer’s D XYvaried from 0.170 (S. reticulata) to 0.692 (T. amurensis), with 5 specieshigher than 0.6.The effects of tree size and abiotic and biotic factors on treesurvival varied among these species (Figure S1). Tree size had themost consistent effect on survival across species: for 18 of the 20species, tree size had the strongest positive effect on survival. Theother two species (Acer tegmentosum and Crataegus maimowiczii)showed no effect of tree size on tree survival. For C. maimowiczii,there was also no obvious effect of abiotic and biotic factors on treesurvival. In total, eight of the 20 species did not show effects ofabiotic and biotic factors on tree survival. Among the other 12species, three species were affected by only abiotic factors, fourspecies by only biotic factors, and five species by both abiotic andbiotic factors. Among the biotic factors that showed effects on treesurvival, frequency of neighboring conspecific basal area and totalbasal area of all neighbors had significant negative effects on treesurvival for 4 and 5 species, respectively. One species, Ulmuslaciniata, showed a significant positive relationship with total basalarea of all neighbors.Table 2. Predictive and discriminative measures (R 2 N and D XY ) of the generalized linear mixed models of tree survival at thecommunity and guild level.All species Abundance Shade-tolerance Tree sizeVery rare Rare Common Very commonShadetolerantLightdemandingShadetolerantMidtolerantLightdemandingSmall treesMediumtreesLargetreesR 2 N0.118 0.177 0.170 0.092 0.090 0.118 0.099 0.107 0.098 0.017 0.044D XY 0.452 0.542 0.526 0.419 0.383 0.446 0.453 0.430 0.402 0.206 0.387doi:10.1371/journal.pone.0029469.t002PLoS ONE | www.plosone.org 4 February 2012 | Volume 7 | Issue 2 | e2946956

Tree Survival in a Chinese Temperate Forestrandomly, since the probability of survival was affected by size,abiotic or biotic variables in nearly all cases. Only one out oftwenty species (C. maimowiczii) showed no relationships betweentree survival and the factors included in the study. In addition,Wang et al. [61] analyzed the spatial pattern of mortality withinthe 5-year period examined, and found that, for most species,dead individuals showed an aggregated spatial pattern atdifferent spatial scales. Together these results indicate that therandom mortality hypothesis can be largely rejected for thistemperate forest.Figure 1. Standardized parameter estimates (±2 SE) of abioticand biotic variables and size on tree survival for all species inthe Changbai temperate forest. Filled circles indicate significanteffects (P,0.05).doi:10.1371/journal.pone.0029469.g001DiscussionIn this study, we used generalized linear mixed models toconduct a comprehensive analysis of the relative importance oftree size and abiotic and biotic neighborhood variables on treesurvival in an old-growth temperate forest in northeasternChina. Our results showed that mortality did not occurTree size effectsAlthough there is no consensus on the shape of the relationshipbetween tree size and survival, empirical and theoretical studiesgenerally suggest that higher mortality rates occur in small tree sizeclasses [9–11,62]. Recently, metabolic ecology theory predictedthat tree mortality rates should decrease with tree size, and treemortality should scale with tree diameter with a constant exponent[62–63]. In the CBS forest, tree size did show a strong positiveeffect on tree survival for most species, but the estimatedrelationship between tree survival and size (i.e. log(dbh)) variedamong species and guilds (Figures 2, 3, and 4 and Figure S1).Moreover, when we classified all individuals into three size classes,we found that effects of tree size on survival turned from stronglypositive for small and medium trees, to slightly negative but notsignificant for large trees (Figure 4), which suggests that the rate atwhich survival increases with size slows down and levels off, oreven declines, at large sizes. This result is consistent with previousstudies that found trees did not continue to have higher survival asthey became larger [9,18–19]. Therefore, metabolic ecologytheory might not be applicable to our temperate forest, as hasbeen previously demonstrated for tropical forests [17]. A possiblecause is that metabolic ecology theory assumes that different sizeclasses receive and use the same amount of energy [63]. Thisassumption may be correct in even-aged forests, but does notextend to natural mixed-aged forests where equal partitioning ofenergy is unlikely [54].Size was the strongest predictor of survival at the communitylevel and for nearly all guilds and species analyzed. However,models solely including tree size were often less likely than moreFigure 2. Standardized parameter estimates (±2 SE) of abiotic and biotic variables and size on tree survival for three shadetolerantguilds in the Changbai temperate forest. Filled circles indicate significant effects (P,0.05).doi:10.1371/journal.pone.0029469.g002PLoS ONE | www.plosone.org 5 February 2012 | Volume 7 | Issue 2 | e2946957

Tree Survival in a Chinese Temperate ForestFigure 3. Standardized parameter estimates (±2 SE) of abioticand biotic variables and size on tree survival for threeabundance classes in the Changbai temperate forest. Filledcircles indicate significant effects (P,0.05). Abund1, Abund2, Abund3and Abund4 show abundance classes with ,100, 100–1000, 1000–5000and .5000 individuals, respectively.doi:10.1371/journal.pone.0029469.g003complex models with biotic and abiotic predictors. In a study ofold-growth conifer forest in the USA, Das et al. [5] also found thatneighborhood variables, such as conspecific density, improvedmodels used to predict tree mortality. Our results were alsoconsistent with results from a tropical forest: Uriarte et al. [9] foundthat tree survival in a Puerto Rican forest was size-dependent, butalso susceptible to crowding effects of neighboring trees, specieslife-history traits and infrequent disturbances (e.g. hurricanes).Abiotic and biotic neighborhood effectsThe relative importance of abiotic and biotic factors on treesurvival and species coexistence in forest communities has been thesubject of a continuous debate, but researchers are now generallyconvinced that these mechanisms are not mutually exclusive [64].In the present study, we found that the relative importance ofabiotic factors on tree survival tended to be less than that of bioticfactors, at least in terms of the variables included in this study. Thisresult may be partially caused by the small spatial variation inabiotic factors in the CBS temperate forest. Topography withinthe CBS plot is relatively gentle with a maximum difference inelevation of less than 18 m in the 25 ha plot [65]. Therefore,abiotic environmental factors have low spatial heterogeneity,which may explain the relatively small effect of abiotic factors ontree survival. This result is consistent with a recent study ofseedling survival in a temperate forest in Japan [66], which foundthat significant effects of abiotic factors on seedling survival werelimited, while effects of biotic variables were more common.However, the best fit models at the community level and formany of the guilds and species examined tended also to includeabiotic factors, indicating that abiotic factors do contribute topatterns of tree mortality. Moreover, our study only includedtopographical and soil nutrient variables, and may have missedsome other important abiotic factors, such as light (althoughvariation in light levels will be captured in part by bioticneighborhood variables). It should also be emphasized that ourstudy is constrained to the plot level (i.e. 25 ha) and that abioticfactors were measured at the 565 m scale. At larger and smallerspatial scales, the relative importance of abiotic factors maychange.Among the biotic factors tested, total basal area of neighborstended to have a negative impact on survival, suggestingcompetition for resources occurring among trees of all speciesleads to thinning [67–68]. However, the frequency of conspecificneighbor basal area also tended to have a significant negativeeffect on tree survival, indicating that neighbors of the samespecies have stronger effects than neighbors of different species,likely due to strong intraspecific competition or natural enemyeffects. Although not explicitly addressed in the present study, anumber of studies have implicated pathogens as a major driver ofnegative conspecific effects in plant communities [22,69–70]. Ourresults are consistent with most previous studies in both tropical[9,23] and temperate forests [8,28,29] that have found negativeeffects of conspecific neighbors on tree performance. These studiestended to support the idea of negative density- and frequencydependenteffects. As a result, negative density- and frequencydependenceis widely hypothesized to play an important role inpopulation dynamics (e.g. recruitment, growth and survival) offorest trees, and these processes could further contribute to speciescoexistence in these communities.Variation among ecological guilds and speciesConsiderable differences occurred among ecological guilds inthe effects of different factors on tree survival in the CBStemperate forest. For shade-tolerance, we found that survival oflight-demanding species and mid-tolerant species were influencedby the biotic, but not the abiotic factors. In contrast, the survival ofshade-tolerant species was impacted by both biotic and abioticfactors. This may be because mid-tolerant species and lightdemandingspecies are more sensitive to shading, and thus theirpatterns of mortality are primarily driven by competition for light(i.e. the biotic neighborhood). In addition, these guilds may bemore sensitive to host-specific natural enemies [42], and thereforemay be more strongly influenced by their local biotic neighborhoods[71]. Among abundance classes, we found significant effectsof abiotic variables only for the common and very commongroups. This may reflect that more common species are morestrongly impacted by soils and/or topography than less commonspecies; however, because sample sizes were smaller for the lesscommon species groups, our power to detect significant effects ofneighborhood variables was reduced relative to analyses of themore common species groups. Previous studies in tropical forestshave found that less common species suffer stronger negativeeffects of conspecific neighbors [23,43]. However, our studyshowed that negative effects of conspecific neighbors werestrongest for the very common and very rare groups (althoughPLoS ONE | www.plosone.org 6 February 2012 | Volume 7 | Issue 2 | e2946958

Tree Survival in a Chinese Temperate ForestFigure 4. Standardized parameter estimates (±2 SE) of abiotic and biotic variables and size on tree survival for three tree sizeclasses in the Changbai temperate forest. Small trees, Medium trees and Large trees show tree size classes with 1–20 cm, 20–40 cm and.40 cm dbh, respectively. Filled circles indicate significant effects (P,0.05).doi:10.1371/journal.pone.0029469.g004not statistically significant for the latter), and weaker, although stillsignificant, for the immediate abundance groups.Individuals in the three tree size classes also differed strongly intheir survival responses. For instance, small trees were morestrongly affected by abiotic factors than medium and large trees.Table 3. AIC values of the generalized linear mixed models oftree survival for 20 tree species with .100 individuals.Species Size Abiotic Biotic All factorsCorylus mandshurica 6271.48 6253.91 6247.28 6240.24Acer mono 2739.86 2740.21 2727.72 2733.22Acer pseudo-sieboldianum 1938.94 1925.24 1939.84 1929.05Acer barbinerve 2017.28 1994.27 2010.67 1993.12Tilia amurensis 793.10 775.65 790.39 779.42Pinus koraiensis 722.02 725.47 724.87 728.51Syringa reticulata 1070.24 1066.53 1069.44 1070.21Ulmus japonica 519.25 519.63 515.15 516.67Quercus mongolica 256.28 263.88 253.26 262.12Maackia amurensis 557.86 561.53 558.19 562.47Fraxinus mandshurica 175.68 180.16 172.42 176.23Acer tegmentosum 334.25 330.10 325.65 324.19Prunus padus 401.65 403.18 402.60 406.59Philadelphus schrenkii 494.82 485.27 496.82 489.04Tilia mandshurica 291.26 274.06 291.26 277.87Acer triflorum 95.53 87.18 89.18 87.91Acer mandshuricum 99.13 108.30 101.26 109.24Ulmus laciniata 69.81 74.85 66.75 73.39Crataegus maimowiczii 122.84 129.91 124.95 128.90Malus baccata 86.39 88.95 89.32 91.54AIC values of the most likely models are shown in bold. Size, Abiotic, Biotic andAll factors correspond to models with only tree size, tree size and abioticfactors, tree size and biotic factors, and all factors included, respectively.doi:10.1371/journal.pone.0029469.t003Small trees tend to be more susceptible to local environmentalconditions than larger trees [44] and can occur in suboptimal sitesas a result of seed dispersal patterns. Therefore, small trees inunfavorable habitats may grow more slowly or die off insuboptimal habitats. Thus, larger trees will be found predominantlyin their preferred habitat. This is consistent with the ideathat habitat filtering at smaller size classes results in adult treehabitat associations [72]. In addition, large trees can change localabiotic properties, such as soil nutrients, over time throughdecades of feedback via plant litter leaching, litter decompositionand root exudation, as well as associations with micro-organisms[73–75].Total neighbor basal area had negative effects on the survival ofsmall and large trees, but no effects on medium trees. Thisindicates that the basal area of neighbors influences the survival ofeven large trees, consistent with results from tropical forests [76].However, the relative importance of neighbor identity differedbetween the small and large size classes, with the frequency ofconspecifics only having a significant, negative effect on thesmaller size class. This suggests that conspecific competition, orapparent competition mediated by natural enemies [77–78], wasstronger than heterospecific competition for small trees, butbecame weaker for larger trees. This result is supported by arelated study in the CBS temperate plot that found that trees weremore regularly spaced in large size classes [53], indicating thatlarge trees have a lower probability of encountering conspecificneighbors than small trees.We expected species within the same guild to have similarresponses to intrinsic and extrinsic factors. However, our resultsdid not meet that expectation. For example, among the six shadetolerantspecies in the Aceraceae family, the relative importance ofsize, abiotic and biotic factors on survival varied dramatically.None of the influential factors had consistent positive or negativeeffects on tree survival for these six species. A. tegmentosum showedno relationship between survival and tree size. Frequency ofneighboring conspecifics had negligible effects on survival of A.pseudo-sieboldianum and A. mandshuricum. Total neighbor basal areaonly had negative effects on tree survival of A. mono. Only half ofthe six species showed a detectable relationship between survivalPLoS ONE | www.plosone.org 7 February 2012 | Volume 7 | Issue 2 | e2946959

Tree Survival in a Chinese Temperate ForestTable 4. Predictive and discriminative measures (R 2 N andD XY ) of the generalized linear mixed models of tree survivalat the species level.R 2 ND XYCorylus mandshurica 0.037 0.226Acer mono 0.035 0.268Acer pseudo-sieboldianum 0.052 0.319Acer barbinerve 0.060 0.333Tilia amurensis 0.296 0.692Pinus koraiensis 0.044 0.313Syringa reticulata 0.026 0.170Ulmus japonica 0.114 0.444Quercus mongolica 0.080 0.415Maackia amurensis 0.044 0.258Fraxinus mandshurica 0.174 0.539Acer tegmentosum 0.101 0.374Prunus padus 0.085 0.337Philadelphus schrenkii 0.083 0.268Tilia mandshurica 0.304 0.613Acer triflorum 0.279 0.673Acer mandshuricum 0.096 0.440Ulmus laciniata 0.267 0.638Crataegus maimowiczii 0.111 0.383Malus baccata 0.261 0.619doi:10.1371/journal.pone.0029469.t004and abiotic factors. None of these results conformed to the guildlevelresults for the shade-tolerant guild. The contradictionbetween the guild-level and species-level analysis may result fromthe fact that species within a particular guild (e.g. shade tolerant)often varied in terms of other characteristics that influencedsurvival patterns. For example, the six shade-tolerant species in theAceraceae family were grouped into three different abundanceguilds (two in the very common class, one in the common class andthree in the rare class). Also, tree size distributions varied greatlyamong these species [54]. All these contributed to differencesamong species. This is consistent with a recent study of seedlingsurvival in a tropical forest where considerable variation occurredamong individual species even within the same guild [79].Shortcomings of the analysisOur study presents comprehensive analyses on the relativeeffects of tree size, abiotic and biotic factors on tree survival in arelatively diverse temperate forest and demonstrates thatmultiple factors collectively determine a tree’s probability ofsurvival. However, the factors included in our study areinsufficient to fully understand and predict tree survival in theCBS temperate forest, as suggested by the relatively lowpredictive and discriminative ability of our models (Tables 2and 4). Several shortcomings likely affected our results. One isthat tree survival is a long-term and complex process that isinfluenced by many factors, including random events, and ouranalysis included only a single 5 year census interval. Data areneeded on forest dynamics over a larger temporal horizon (e.g.centuries) to capture all the relevant information about survivalprocesses in long-lived trees. Second, other factors that mayaffect tree survival are not explicitly included in the analysis,such as climate change [19,80], light [81], pathogen or insectattack [82–83], etc. Third, it is difficult to measure the effectiveneighborhood radius for each tree; hence a fixed maximumdistance of influence (i.e. 20 m) was used in the analysis. Inaddition, by grouping all heterospecific neighbors together, ouranalysis did not consider species-specific effects on neighborsurvival. Previous studies have shown that the effects of differentspecies on plant performance can be asymmetric [9,25].ConclusionsOur study suggests that intrinsic tree size, density- andfrequency-dependent effects and niche partitioning with respectto soil and topographic factors contribute to the regulation of theCBS temperate forest community, but the relative importance ofthese factors varies dramatically among guilds and species.Although the implications of this result for tree species diversityremain to be explored, they imply that attempts to understand,conserve and manage temperate forests should explicitly considerthe relative importance of intrinsic and extrinsic factors on forestdynamics. Specifically, if the CBS temperate forest is typical ofother temperate forest regions, then one prediction of metabolicecology theory, that tree mortality should scale with tree diameterwith a constant negative exponent [62–63], should be rejected.Survival did not continue increasing with tree size and theexponent varied greatly among species. Furthermore, we foundstrong evidence of negative density dependence, with thefrequency of conspecific neighbors having a particularly strongnegative impact on survival. This suggests that negative frequencydependence may play a role in the maintenance of diversity in thistemperate forest. In addition, our study demonstrates that therelative importance of local-scale variables driving patterns of treesurvival varies greatly among species, size classes, guilds andabundance classes in temperate forests. Therefore, predictivemodels and management decisions should be designed with thisvariation in mind.Supporting InformationFigure S1 Estimated effects (±2 SE) of abiotic and bioticvariables on tree survival for 20 species with .100individuals in the Changbai temperate forest. Filled circlesindicate significant effects (p,0.05).(DOC)Table S1 Factor loadings of the first two components ofthe PCA on soil variables from the Changbai temperateforest plot.(DOC)Table S2 Parameters used in models of tree survival inthe Changbaishan temperate forest, northeastern China.(DOC)AcknowledgmentsWe thank Adrian Das and Richard Condit for valuable comments on themanuscript. We thank the CBS plot census and data management teams,and especially thank Buhang Li, Xuejiao Bai, Liwei Wang, and DingliangXing, for assistance with field work and data collection.Author ContributionsConceived and designed the experiments: XW LSC. Performed theexperiments: XW LSC. Analyzed the data: XW LSC. Contributedreagents/materials/analysis tools: ZH SJD JY FL ZY. Wrote the paper:XW LSC.PLoS ONE | www.plosone.org 8 February 2012 | Volume 7 | Issue 2 | e2946960

Tree Survival in a Chinese Temperate ForestReferences1. Franklin JF, Shugart HH, Harmon ME (1987) Tree death as an ecologicalprocess. Ecoscience 37: 550–556.2. Hibbs DE (1983) Forty years of forest succession in central New England.Ecology 64: 1394–1401.3. Runkle JR (2000) Canopy tree turnover in old-growth mesic forests of easternNorth America. Ecology 81: 554–567.4. Lutz JA, Halpern CB (2006) Tree mortality during early forest development: along-term study of rates, causes, and consequences. Ecological Monographs 76:257–275.5. Das A, Battles J, van Mantgem PJ, Stephenson NL (2008) Spatial elements ofmortality risk in old-growth forests. Ecology 89: 1744–1756.6. Sterner RW, Ribic CA, Schatz GE (1986) Testing for life historical changes inspatial patterns of four tropical tree species. Journal of Ecology 74: 621–633.7. Kenkel NC (1988) Pattern of self-thinning in Jack pine: testing the randommortality hypothesis. Ecology 69: 1017–1024.8. He F, Duncan RP (2000) Density-dependent effects on tree survival in an oldgrowth Douglas fir forest. Journal of Ecology 88: 676–688.9. Uriarte M, Canham CD, Thompson J, Zimmerman JK (2004) A neighborhoodanalysis of tree growth and survival in a hurricane-driven tropical forest.Ecological Monographs 74: 591–614.10. Clark DA, Clark DB (1992) Life history diversity of canopy and emergent treesin a Neotropical rain forest. Ecological Monographs 62: 315–344.11. Blundell AG, Peart DR (2001) Growth strategies of a shade tolerant tree: theinteractive effects of herbivory and canopy gaps. Journal of Ecology 89:608–615.12. Weiner J (1990) Asymmetric competition in plant populations. Trends inEcology and Evolution 5: 360–364.13. Woods KD (2000) Dynamics in late-successional hemlock-hardwood forests overthree decades. Ecology 81: 110–126.14. Keane RE, Austin M, Field C, Huth A, Lexer MJ, et al. (2001) Tree mortality ingap models: application to climate change. Climate Change 51: 509–540.15. Lorimer CG, Dahir SE, Nordheim EV (2001) Tree mortality rates and longevityin mature and old-growth hemlock-hardwood forests. Journal of Ecology 89:960–971.16. Yao X, Titus SJ, MacDonald SE (2001) A generalized logistic model ofindividual tree mortality for aspen, white spruce, and lodgepole pine in Albertamixed wood forests. Canadian Journal of Forest Research 31: 283–291.17. Muller-Landau HC, Condit R, Chave J, Thomas SC, Bohlman SA, et al. (2006)Testing metabolic ecology theory for allometric scaling of tree size, growth andmortality in tropical forests. Ecology Letters 9: 575–588.18. Coomes DA, Allen RB (2007) Mortality and tree-size distributions in naturalmixed-age forests. Journal of Ecology 95: 27–40.19. Lines ER, Coomes DA, Purves DW (2010) Influences of forest structure, climateand species composition on tree mortality across the Eastern US. PloS One 5:e13212.20. Nakashizuka T (2001) Species coexistence in temperate, mixed deciduousforests. Trends in Ecology and Evolution 16: 205–210.21. Wright SJ (2002) Plant diversity in tropical forests: a review of mechanisms ofspecies coexistence. Oecologia 130: 1–14.22. Packer A, Clay K (2000) Soil pathogens and spatial patterns of seedling mortalityin a temperate tree. Nature 404: 278–281.23. Hubbell SP, Ahumada JA, Condit R, Foster RB (2001) Local neighborhoodeffects on long-term survival of individual trees in a Neotropical forest.Ecological Research 16: 859–875.24. HilleRisLambers J, Clark JS, Beckage B (2002) Density-dependent mortality andthe latitudinal gradient in species diversity. Nature 417: 732–735.25. Canham CD, Papaik MJ, Uriarte M, McWilliams WH, Jenkins JC, et al. (2006)Neighborhood analyses of canopy tree competition along environmentalgradients in New England forests. Ecological Applications 16: 540–554.26. Queenborough SA, Burslem D, Garwood NC, Valencia R (2007) Neighborhoodand community interactions determine the spatial pattern of tropical treeseedling survival. Ecology 88: 2248–2258.27. Comita LS, Hubbell SP (2009) Local neighborhood and species’ shade toleranceinfluence survival in a diverse seedling bank. Ecology 90: 328–334.28. Zhang J, Hao Z, Sun IF, Song B, Ye J, et al. (2009) Density dependence on treesurvival in an old-growth temperate forest in northeastern China. Annuals ofForest Sciences 66: 204.29. Zhao D, Borders B, Wilson M, Rathbun SL (2006) Modeling neighborhoodeffects on the growth and survival of individual trees in a natural temperatespecies-rich forest. Ecological Modelling 196: 90–102.30. Castagneri D, Lingua E, Vacchiano G, Nolad P, Motta R (2010) Diachronicanalysis of individual-tree mortality in a Norway spruce stand in the easternItalian Alps. Annuals of Forest Science 67: 304.31. Janzen DH (1970) Herbivores and the number of tree species in tropical forests.American Naturalist 104: 501–528.32. Connell JH (1971) On the role of natural enemies in preventing competitiveexclusion in some marine animals and in rain forest trees. In: den Boer PJ,Gradwell GR, eds. Dynamics of numbers in populations. Wageningen,Netherlands: Centre for Agricultural Publishing and Documentation. pp298–312.33. Chesson P (2000) Mechanisms of maintenance of species diversity. AnnualReview of Ecology and Systematics 31: 343–358.34. Volkov I, Banavar JR, He FL, Hubbell SP, Maritan A (2005) Densitydependence explains tree species abundance and diversity in tropical forests.Nature 438: 658–661.35. Condit R, Hubbell SP, Foster RB (1994) Density dependence in two understorytree species in a Neotropical forest. Ecology 75L: 671–680.36. Harms KE, Condit R, Hubbell SP, Foster RB (2001) Habitat associations oftrees and shrubs in a 50-ha Neotropical forest plot. Journal of Ecology 89:947–959.37. John R, Dalling JW, Harms KE, Yavitt JB, Stallard RF, et al. (2007) Soilnutrients influence spatial distributions of tropical tree species. Proceedings ofthe National Academy of Sciences 104: 864–869.38. Kunstler G, Coomes DA, Canham CD (2009) Size-dependence of growth andmortality influence the shade tolerance of trees in a lowland temperate rainforest. Journal of Ecology 97: 685–695.39. Coley PD (1980) Effects of leaf age and plant life history patterns on herbivory.Nature 284: 545–546.40. Loehle C (1988) Tree life history strategies: the role of defenses. CanadianJournal of Forest Research 18: 209–222.41. Imaji A, Seiwa K (2010) Carbon allocation to defense, storage, and growth inseedlings of two temperate broad-leaved tree species. Oecologia 162: 273–281.42. McCarthy-Neumann S, Kobe RK (2008) Tolerance of soil pathogens co-varieswith shade tolerance across species of tropical tree seedlings. Ecology 89:1883–1892.43. Comita LS, Muller-Landau HC, Aguilar S, Hubbell SP (2010) Asymmetricdensity dependence shapes species abundances in a tropical tree community.Science 329: 330–332.44. Russo SE, Davies SJ, King DA, Tan S (2005) Soil-related performance variationand distributions of tree species in a Bornean rain forest. Journal of Ecology 93:879–889.45. Comita LS, Condit R, Hubbell SP (2007) Developmental changes in habitatassociations of tropical trees. Journal of Ecology 95: 482–492.46. Hao Z, Li B, Zhang J, Wang X, Ye J, et al. (2008) Broad-leaved Korean pinemixed forest plot in Changbaishan (CBS) of China: Community compositionand structure. Acta Phytoecologica Sinica 32: 238–250. (in Chinese).47. Condit R (1998) Tropical forest census plots. Springer, Berlin.48. Wang X, Wiegand T, Hao Z, Li B, Ye J, et al. (2010) Spatial associations in anold-growth temperate forest, Northeastern China. Journal of Ecology 98:674–686.49. Lu R (1999) Analytical Methods of Soil Agrochemistry. China Agricultural Science andTechnology Press, Beijing;(in Chinese).50. Bolker BM, Brooks ME, Clark CJ, Geange SW, Poulsen JR, et al. (2009)Generalized linear mixed models: a practical guide for ecology and evolution.Trends in Ecology and Evolution 24: 127–135.51. Zurr AF, Ieno EN, Walker NJ, Saveliev AA, Smith GM (2009) Mixed effectsmodels and extensions in ecology with R. Springer, New York.52. Gelman A, Hill J Data Analysis Using Regression and Multilevel/HierarchicalModels, Cambridge University Press, Cambridge, UK.53. Wang X, Ye J, Li B, Zhang J, Lin F, et al. (2010) Spatial distributions of speciesin an old-growth temperate forest, northeastern China. Canadian Journal ofForest Research 40: 1011–1019.54. Wang X, Hao Z, Zhang J, Lian J, Li B, et al. (2009) Tree size distributions in anold-growth temperate forest. Oikos 118: 25–36.55. Burnham KP, Anderson DR (2002) Model selection and inference: A practicalinformation-theoretic approach. Springer-Verlag, New York.56. Nagelkerke N (1991) A Note on a General Definition of the Coefficient ofDetermination. Biometrika 78: 691–692.57. Harrell FE (2001) Regression modeling strategies with applications to linearmodels, logistic regression, and survival analysis. Springer, New York.58. Somers RH (1962) A new asymmetric measure of association for ordinalvariables. American Sociological Review 27: 799–811.59. R Development Core Team (2009) R: A language and environment for statisticalcomputing R Foundation for Statistical Computing, Vienna, Austria;ISBN 3-900051-07-0. Available: http://www.R-project.org..60. Bates D, Maechler M (2009) lme4: Linear mixed-effects models using S4 classes.Available: http://CRAN.R-project.org/package=lme4/.61. Wang L, Li B, Ye J, Bai X, Yuan Z, et al. (2011) Dynamics of short-term treemortality in broad-leaved Korean pine (Pinus koraiensis) mixed forest in theChangbai Mountains. Biodiversity Science 19: 260–270. (in Chinese).62. Brown JH, Gillooly JF, Allen AP, Savage VM, West GB (2004) Toward ametabolic theory of ecology. Ecology 85: 1771–1789.63. Coomes DA (2006) Challenges to the generality of WBE theory. Trends inEcology and Evolution 21: 593–596.64. Queenborough SA, Burslem DFRP, Garwood NC, Valencia R (2009)Taxonomic scale-dependence of habitat niche partitioning and biotic neighborhoodon survival of tropical tree seedlings. Proceedings of the Royal Society B276: 4197–4205.65. Wang X, Hao Z, Ye J, Zhang J, Li B, et al. (2008) Spatial pattern of diversity inan old-growth temperate forest in Northeastern China. Acta Oecologica 33:345–354.66. Shibata M, Masaki T, Tanaka H, Niiyama K, Iida S, et al. (2010) Effects ofabiotic and biotic factors and stochasticity on tree regeneration in a temperateforest community. Ecoscience 17: 137–145.PLoS ONE | www.plosone.org 9 February 2012 | Volume 7 | Issue 2 | e2946961

Tree Survival in a Chinese Temperate Forest67. White J, Harper JL (1970) Correlated changes in plant size and number in plantpopulations. Journal of Ecology 58: 467–485.68. Norberg RA (1988) Theory of growth geometry of plants and self-thinning ofplant populations: geometric similarity, elastic similarity, and different growthmodes of plant parts. American Naturalist 131: 220–256.69. Mangan SA, Schnitzer SA, Herre EA, Mack KML, Valencia MC, et al. (2010)Negative plant-soil feedback predicts tree-species relative abundance in atropical forest. Nature 466: 752–755.70. Mordecai EA (2011) Pathogen impacts on plant communities: unifying theory,concepts, and empirical work. Ecological Monographs 81: 429–441.71. McCarthy-Neumann S, Kobe RK (2010) Conspecific and heterospecific plantsoilfeedbacks influence survivorship and growth of temperate tree seedlings.Journal of Ecology 98: 408–418.72. Comita LS, Engelbrecht BMJ (2009) Seasonal and spatial variation in wateravailability drive habitat associations in a tropical forest. Ecology 90: 2755–2765.73. Finzi AC, Canham CD, van Breemen N (1998a) Canopy tree-soil interactionswithin temperate forests: species effects on pH and cations. EcologicalApplications 8: 447–454.74. Finzi AC, van Breemen N, Canham CD (1998b) Canopy tree-soil interactionswithin temperate forests: species effects on carbon and nitrogen. EcologicalApplications 8: 440–446.75. Ehrenfeld JG, Ravit B, Elgersma E (2005) Feedback in the plant-soil system.Annual Review of Environment and Resources 30: 75–115.76. Peters HA (2003) Neighbour-regulated mortality: the influence of positive andnegative density dependence on tree populations in species-rich tropical forests.Ecology Letters 6: 757–765.77. Packer A, Clay K (2002) Soil pathogens and Prunus serotina seedlings and saplinggrowth near conspecific trees. Ecology 84: 108–119.78. Reinhart K, Clay K (2009) Spatial variation in soil-borne disease dynamics of atemperate tree, Prunus serotina. Ecology 90: 2984–2993.79. Comita LS, Uriarte M, Thompson J, Jonckheere I, Canham CD, et al. (2009)Abiotic and biotic drivers of seedling survival in a hurricane-impacted tropicalforest. Journal of Ecology 97: 1346–1359.80. Slik JW (2001) El Niño droughts and their effects on tree species compositionand diversity in tropical rain forests. Oecologica 141: 114–120.81. Davies SJ (2001) Tree mortality and growth in 11 sympatric Macaranga speciesin Borneo. Ecology 82: 920–932.82. Filip GM, Goheen DJ (1982) Tree mortality caused by root pathogen complex inDeschutes national forest, Oregon. Plant Disease 66: 240–243.83. Maloney PE, Smith TF, Jensen CE, Innes J, Rizzo DM, et al. (2008) Initial treemortality and insect and pathogen response to fire and thinning restorationtreatments in an old-growth mixed-conifer forest of the Sierra Nevada,California. Canadian Journal of Forest Research 38: 3011–3020.PLoS ONE | www.plosone.org 10 February 2012 | Volume 7 | Issue 2 | e2946962

Effects of Soil Water and Nitrogen on Growth andPhotosynthetic Response of Manchurian Ash (Fraxinusmandshurica) Seedlings in Northeastern ChinaMiao Wang*, Shuai Shi, Fei Lin, Zhanqing Hao, Ping Jiang, Guanhua DaiInstitute of Applied Ecology, Chinese Academy of Sciences, Shenyang, ChinaAbstractBackground: Soil water and nitrogen (N) are considered to be the main environmental factors limiting plant growth andphotosynthetic capacity. However, less is known about the interactive effects of soil water and N on tree growth andphotosynthetic response in the temperate ecosystem.Methods/Principal Findings: We applied N and water, alone and in combination, and investigated the combined effect ofdifferent water and N regimes on growth and photosynthetic responses of Fraxinus mandshurica seedlings. The seedlingswere exposed to three water regimes including natural precipitation (CK), higher precipitation (HW) (CK +30%) and lowerprecipitation (LW) (CK 230%), and both with and without N addition for two growing seasons. We demonstrated that waterand N supply led to a significant increase in the growth and biomass production of the seedlings. LW treatment significantlydecreased biomass production and leaf N content, but they showed marked increases in N addition. N addition couldenhance the photosynthetic capability under HW and CK conditions. Leaf chlorophyll content and the initial activity ofRubisco were dramatically increased by N addition regardless of soil water condition. The positive relationships were foundbetween photosynthetic capacity, leaf N content, and SLA in response to water and N supply in the seedling. Rubiscoexpression was up-regulated by N addition with decreasing soil water content. Immunofluorescent staining showed thatthe labeling for Rubisco was relatively low in leaves of the seedlings under LW condition. The accumulation of Rubisco wasincreased in leaf tissues of LW by N addition.Conclusions/Significance: Our study has presented better understanding of the interactions between soil water and N onthe growth and photosynthetic response in F. mandschurica seedlings, which may provide novel insights on the potentialresponses of the forest ecosystem to climate change associated with increasing N deposition.Citation: Wang M, Shi S, Lin F, Hao Z, Jiang P, et al. (2012) Effects of Soil Water and Nitrogen on Growth and Photosynthetic Response of Manchurian Ash(Fraxinus mandshurica) Seedlings in Northeastern China. PLoS ONE 7(2): e30754. doi:10.1371/journal.pone.0030754Editor: Ivan Baxter, United States Department of Agriculture, Agricultural Research Service, United States of AmericaReceived December 9, 2010; Accepted December 20, 2011; Published February 8, 2012Copyright: ß 2012 Wang et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permitsunrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.Funding: This work was sponsored by the Major State Basic Research Development Program (2010CB833502), the Knowledge Innovation Project of the ChineseAcademy of Sciences (KSCX-YW-Z-1022), the Key Program of the National Natural Science Foundation of China (30590382/C011108) and the Program for theNational Natural Science Foundation of China (No. 30870376). The funders had no role in study design, data collection and analysis, decision to publish, orpreparation of the manuscript.Competing Interests: The authors have declared that no competing interests exist.* E-mail: wangmiao@iae.ac.cnIntroductionHuman activities such as fossil fuel burning, forest disturbance,and land conversion have globally elevated the atmosphericconcentration of carbon dioxide (CO 2 ) and atmospheric depositionof nitrogen (N) [1]. The atmospheric N depositions arealtering the availability of this limiting nutrient in many terrestrialecosystems [2]. Elevated N availability can affect plant growth,biodiversity, and ecosystem functioning [3,4]. Soil N availabilityhas the potential to alter plant physiology in terrestrial ecosystems[5,6]. Increases in atmospheric N deposition can affect the amountof N available to plants which influence the growth and survival ofthe seedlings [7]. Photosynthesis may be altered in responses toelevated N availability [8]. Increased N availability results inincreased photosynthesis and growth in northern hardwood trees[9]. N additions increase leaf N concentrations accompanied byhigher net photosynthetic rates in Douglas-fir [10], poplar [11],pond pine and red maple [12]. Maximum photosynthetic capacityis strongly regulated by leaf N concentration [13]. It is showed thatthere is a significant and positive correlation between photosyntheticcapacity and leaf N content [14–16]. Increases in Navailability have been shown to correspond with increased leafchlorophyll content [11,17], Rubisco (Ribulose-1, 5-bisphosphatecarboxylase/oxgenase) [10]. N addition enhances tolerance ofplants to abiotic stresses such as water deficits, salt and hightemperatures [18–20]. Despite the potential importance of Ndeposition in plant, there is still limited knowledge regarding therelationship between N application, photosynthesis and growth intemperate forest ecosystems.Soil water content is the primary limitation in photosyntheticprocesses in plants. Water availability influences leaf phenology[21] and photosynthetic rate [22]. It is well known that one of theprimary physiological consequences of drought is photosynthesisinhibition [23,24]. Inhibition of photosynthesis under drought hasPLoS ONE | www.plosone.org 1 February 2012 | Volume 7 | Issue 2 | e3075463

Soil Water and Nitrogen Affect Photosynthesisbeen attributed mainly to stomatal closure, reduced mesophyllconductance, and inhibition of Rubisco activity [25–29]. Themajor effects of water deficit on plant function include decreasedshoot growth due to decreased leaf biomass and leaf areaallocation, and increased leaf N content [30].Physiological responses of plants to either water deficit ornitrogen addition have been documented [31]. Soil N availabilitycan be affected by soil water availability via several microbialmediatedpathways, such as litter decomposition [32] and Nmineralization [33]. Appropriate N supply is recommended toimprove photosynthetic efficiency under water stress [15].However the interactions between these two factors on plantphysiological responses have received relatively little attention[34]. The overall effect of N addition and water changes on treesremains still unclear.Rubisco is a kind of special enzymes that catalyzes the initialfixation reaction of photosynthesis [35]. Rubisco is mainly locatedin the chloroplasts of the bundle sheath cells in the leaves of higherplants. The large subunits of Rubisco play an important role inphotosynthesis for CO 2 assimilating [35]. Some evidences suggestthat Rubisco functions increasingly as a storage protein in additionto its catalytic functions with increasing N area [36]. The responseof Rubisco to N supply in trees remains equivocal. The resultsshow greater concentration of Rubisco in seedling foliage at highrates of N supply [37], whereas another study found no effect offertiliser application on Rubisco concentration, Rubisco activity orphotosynthesis in 25- to 30-year-old trees [38]. Less is knownabout the relationships between photosynthetic capacity, leaf Ncontent, and the expression and activity of Rubisco in response toN and water in the seedling.Fraxinus mandschurica is the most economically important foresttree species and primarily distributed in the temperate forests ofnorthern hemisphere. However, forest decline of F. mandschuricahave been recently observed in forest areas in the northeast inChina due to logging and hunting. Protection and restoration ofthis ecologically important deciduous tree in temperate forestregions is crucial. Little information is available regarding theeffects of N depositions and water availability on the growth andphotosynthetic responses of F. mandschurica seedlings. Therefore, inthe present study we applied N and water, alone and incombination, and investigated the interactive effects of N additionand soil water on the growth and physiological function of F.mandschurica seedlings. We specifically aimed to examine potentialimpacts of increased soil N and water availability and theirinteraction on whole-plant growth, biomass allocation, photosyntheticgas exchange, specific leaf area, leaf N content andphotosynthetic pigment content in the seedlings. We also studiedthe changes in the expression and activity of Rubisco to clarifyhow N addition and water treatment affect photosyntheticfunctions of the seedlings. Better understanding of the interactionsbetween soil water and N on trees may provide critical insights onthe potential responses of the forest ecosystem to climate changeassociated with increasing atmospheric N deposition.Materials and MethodsStudy siteThis study was carried out in the Changbai Mountain NaturalReserve in northeastern China (42u249090N, 128u059450E). Thearea is situated in the temperate continental climatic zone. Altitudeabove sea level of the study site is 738 m. Mean annualtemperature is 3.6uC with monthly mean temperatures of215.6uC in January and 19.7uC in July, respectively. Meanannual precipitation is 695 mm. The period of snow cover is fromNovember to April, with a maximum depth of about 30 cm. Mostprecipitation in this area occurs from June to September (480–500 mm) [39]. The soil is classified as dark brown forest soil(Calcis-orthic Aridisol in the US Soil Taxonomy classification)with pH of 5.85, and with the top 30 cm containing an average of156.6 g kg 21 organic carbon and 7.17 g kg 21 total N. Thetemperate broad-leaved Korean pine (Pinus koraiensis) mixed forestin the study area is dominated by Pinus koraiensis Sieb. et Zucc.,Fraxinus mandschurica Rupr., Quercus mongolica Fisch. ex Ledeb. andTilia amurensis Rupr.Experimental designThe experiment was conducted in openings within a maturebroad-leaved Korean pine (P. koraiensis) mixed forest. A paired,nested design was used with precipitation as the primary factorand N addition the secondary one. The experiment involved threepairs of 21.661.6 m plots. N was added to one plot in each pair(+N), while the other plot in that pair contained no addition of N,but only that resulting from naturally occurring addition (CK).Each pair was also subjected to one of three water (precipitation)regimes: a) naturally occurring precipitation (CK); b) precipitationdeduction (LW), in which 33% of naturally occurring precipitationwas removed and diverted to c) precipitation enhancement plot(HW). These three regimes were applied to the three pairs of plots,yielding an overall experimental design as depicted in Figure 1.Each of the 6 plots was divided into nine 2.461.6 m subplots thatserved as replicates, yielding a total of 54 subplots in theexperiment.On 5 May 2006, two-year-old seedlings of F. mandshurica wereplanted individually in the 54 subplots that served as locations forexperiment replications. Precipitation was manipulated by meansof troughs (0.1661.6 m) suspended above the dry plots such thatabout 33% of the precipitation was trapped and passivelytransferred by gravity to polyvinylchloride piping and then acrossan ambient plot to a wet plot. In order to allow sunlight in, theseflumes were made of transparent plastic board. The flumesinclined to the ground level at the angle of 15u with the highestand lowest points 1.43 m and 1 m above the ground, respectively.Flumes were spaced 40 cm apart. Soil water treatments began on15 May 2006. To reduce nutrient heterogeneity, the original soilwas excavated to a depth of 0.3 m and replaced with soil collectedfrom the floor of a mature broad-leaved Korean pine (P. koraiensis)mixed forest. The soil was passed through a 4 mm sieve aftercollection.Two N levels were control (CK) without N addition and Naddition (+N) 10 g N m 22 yr 21 experimental input. The latterwas applied by use of a backpack sprayer. Ammonium nitrate wasapplied twice per year on 15 May and 15 July in 2006, 2007 and2008 as two equal applications (5 g N m 22 , i.e. 54.86 g NH 4 NO 3 )over the entire year. During each application, fertilizer wasweighed and mixed with 20 L of water. For each of the watertreatments, soil volumetric water content (v/v) was periodicallymeasured in the 0–30 cm depth range with a portable timedomain reflect meter (TDR 100 Campbell, USA). Whole seedlingdry mass, tree height and stem base diameter at the beginning ofthe experiment were 8.6460.49 g, 26.3560.80 cm and8.3360.20 mm, respectively. For assessment of water 6 N effectson the physiological/morphological characteristics of F. mandshuricaseedlings, all seedlings were grown under the same conditionswith the exception of variations in soil water and N levels.Growth parametersIn late September 2008, 36 randomly selected seedlings wereharvested (n = 6 per replicate) to determine final shoot height, rootPLoS ONE | www.plosone.org 2 February 2012 | Volume 7 | Issue 2 | e3075464

Soil Water and Nitrogen Affect PhotosynthesisFigure 1. Total seedling, leaf, root and stem biomass (panels A–D) and SLA and S/R (panel E and F) under high-water (HW), (CK),and low-water (LW) conditions in combinations with natural (dotted) or high N-supply level (hatched). Bars represent means of 6replications 6 standard deviation. Values accompanied by different letters differ significantly at p = 0.05. Abbreviations: SLA, specific leaf area; S/R, theratio of the stem and root biomass.doi:10.1371/journal.pone.0030754.g001collar diameter, and stem, leaf, and root biomass. Roots wereseparated from shoots by severing the seedling at the root collar,and were then carefully washed clean of growth media. The shootswere divided into leaf and stem components. Seedling fractionswere oven-dried separately for at least 72 h at 80uC and the drymass of each fraction was determined. Specific leaf area (SLAcm 2 g 21 ) was measured on six seedlings for each treatment, usinga LI-3000 leaf area meter (Li-Cor, Lincoln, NE).Gas exchange parametersTo characterize water- and N-induced shifts in carbonacquisition, instantaneous gas exchanges on fully expanded,exposed current-year leaves were measured under controlledoptimal conditions using an open-mode portable photosynthesissystem (LI-6400, Li-Cor, Lincoln, NE). For each treatment, threeto four leaves of three individuals per replicate were randomlyselected for sampling. For each seedling a series of fivemeasurements per leaf was averaged (after the system hadachieved a predetermined stability point), and the mean value ofthree individuals was used as the replicate for statistical analysis.PN-PAR response curves were measured at 1800, 1500, 1200,1000, 800, 500, 200, 100, 50, 20, and 0 mmol m 22 s 21 of PARunder uniform conditions (25uC, 360610 mmol (CO 2 ) mol 21 , and65–75% RH at 9:00–11:30 on two sunny days. Maximum netphotosynthetic rate (A max ) and saturation irradiance were estimatedaccording to Ellsworth (2000) [40]. All the measurements wererecorded 5 times. In addition, water use efficiency was calculatedusing instantaneous measurements. Instantaneous water useefficiency (WUE i ) was calculated and defined as A max /E, whichA max is the light-saturated net CO 2 assimilation rate and E isPLoS ONE | www.plosone.org 3 February 2012 | Volume 7 | Issue 2 | e3075465

Soil Water and Nitrogen Affect Photosynthesistranspiration rate. All of the measurements were taken between9:00 am to 11:30 pm on two fully sunny days (July 17–21, 2008)under natural conditions.Determination of N concentration per unit leaf areaOn 15 June, 17 July and 20 August 2008, two or threenonshaded leaves per seedling were harvested and washed withdeionized water. The area of the fresh leaves was measured afterpetiole removal, with an area meter (LI-3000A; Li-Cor). Theharvested leaves were dried at 70uC during 48 h and ground foranalysis. The specific leaf area (SLA) was determined as the ratioof leaf area to leaf dry mass, 20 leaves were collected andtransported to the laboratory in refrigerated bags to avoid weightloss by respiration in each treatment. The leaf area was measured,after petiole removal, with an area meter (LI-3000A; Li-Cor). Thedried leaves were ground to fine powder with a vibrating samplemill (MM-400 Retsch, Haan, Germany). The concentration of Nin the powder was determined with a CHN analyzer (Vario EL;Elementar, Hanau, Germany). The N concentration per unit leafarea (N area ) was determined as the ratio of N concentration to SLAof the leaves. Photosynthetic N-use efficiency (PNUE) wasdetermined as the ratio of A360 to N area . [41].Photosynthesis pigment contentThe fully expanded leaves from each seedling were collected,placed between layers of ice in a thermal insulated box, and takento the laboratory of the National Research Station of ChangbaiMountain forest ecology where they were analyzed immediately.The leaf disks (1 cm 2 ) were taken and homogenized in chilled 80%(v/v) acetone, and the homogenates were centrifuged at 10000 gfor 10 min at 4uC in the dark. The supernatant was used fordetermining pigment contents. The absorbance of the supernatantwas recorded at 470, 646, and 663 nm. The amounts ofchlorophyll a, b, and total chlorophyll were calculated as describedby Inskeep and Bloom [42]. Total carotenoids were calculated asdescribed by Arnon [43]. All the spectrophotometric assays wereconducted using a UV-1601 spectrophotometer (Shimadzu,Japan).Measurement of Rubisco activity and activation stateOn 18th August 2008, three non-shaded first-flush leaves perseedling were harvested from 9:00 to 11:30 am. The harvestedleaves were washed with deionized water. Leaf samples (0.1 g)were frozen in liquid nitrogen until the measurements of activityand concentration of Rubisco. The stored leaf samples werehomogenized to a fine powder in liquid nitrogen with a mortarand pestle. Subsequently, Rubisco was extracted by grinding thefine powder in a 1.0 ml extraction buffer containing 50 mMHEPES-KOH (pH 8.0), 10 mmol/L MgCl 2 , 0.5 mmol/L EDTA,1% (w/v) polyvinylpolypyrrolidone. The crude homogenate wascentrifuged at 16000 g for 15 min. The supernatant of the samplewas used in the assay of activity of Rubisco. The activity ofRubisco was determined spectrophotometrically by measuring thedisappearance rate of NADH [44]. To determine the initialactivity of Rubisco, immediately after combining the desaltedsample solution (100 mL with assay solution containing 50 mMHEPES-KOH (pH 8.0), 10 mM NaHCO 3 ,1.5 mM NADH,5 mM ATP, 1 mM EDTA, 20 mM MgCl 2 , 2.5 mM DTT,5 mM phosphocreatin, 10 units per ml of phosphoglyceric kinase,10 units per ml of glyceraldehyde-3-phosphate dehydrogenase and20 units per ml of phosphocreatine kinase at final concentration,the reaction was started by adding 60 mL of 10 mM RuDP. Thechange in the absorption of the activation state of Rubisco wascalculated as the ratio of initial activity to total activity of thisenzyme.Western blottingLeaf samples were ground in liquid N 2 with mortar and pestle.Total proteins were extracted with a buffer containing 50 mMphosphate buffer solution (pH 7.5), 2% b-mercaptoethanol,100 mM EDTA, 1% PVPP (w/v), and 1% Triton X-100 (v/v).After 15 min centrifugation (4uC, 15000 g), the upper phase wastransferred to a new centrifuge tube. Two volumes of TRISsaturated phenol (pH 8.0) were added and then the mixture wasfurther vortexed for 30 min. Proteins were precipitated by adding5 vols of ammonium sulphate-saturated methanol, and incubatedat 220uC for at least 4 h. After centrifugation as described above,the protein pellets were re-suspended and rinsed with ice-coldmethanol followed by washing with ice-cold acetone twice, andspun down at 15000 g for 10 min at 4uC after each washing.Finally the washed pellets were air-dried and recovered in the lysisbuffer containing 62.5 mM TRIS-HCl (pH 6.8), 2% SDS (v/v),10% glycerol (v/v), and 2% b-mercaptoethanol (v/v). Proteinconcentrations were quantified using the Bradford assay [45].For Western-blot analysis, an aliquot of the proteins (20 mg) wasseparated by SDS-PAGE using 12% (w/v) acrylamide gelsaccording to the method of Laemmli (1970) and electrophoreticallytransferred to nitrocellulose membranes (Millipore, Saint-Quentin, France). The protein blot was probed with a primaryantibody of the Rubisco large subunit (AS03037-200, Agrisera,Sweden) at a dilution of 1:5000 for 4 h at room temperature withagitation. The blot was washed three times in phosphate bufferedsaline with Tween-20 solution containing 50 mM TRIS-HCl(pH 8.0), 150 mM NaCl, 0.05% Tween-20 (v/v), and followed byincubation with the secondary antibody (anti-rabbit IgG horseradishperoxidase conjugated, Abcam, UK, 1:5000 dilution) for2 h at room temperature. The blots were finally washed as aboveand developed with SuperSigmal West Pico ChemiluminescentSubstrate (Pierce, USA) according to the manufacturer’s instructions.Images of the blots were obtained using a CCD imager(FluorSMax, Bio-Rad, USA). The QuantityOne software (Bio-Rad, Hercules, CA, USA) was used to determine the opticaldensity.ImmunolocalizationLeaf sections were embedded in OCT compound (SakuraFinetek CA, USA) and sections were cut using a microtome andadhered to a poly-lysine coated slide. Sections were then fixed in3% paraformaldehyde. After being rinsed with phosphate-bufferedsaline (PBS; 150 mM NaCl, 5 mM KCl, 0.8 mM KH 2 PO 4 ,3.2 mM Na 2 HPO 4 , pH 7.3), tissue sections were blocked with 1%bovine serum in PBS. Samples were washed extensively in PBSand then incubated at 4uC overnight with the polyclonal primaryrabbit anti-Rubisco (1:2000) in 0.5% BSA in PBS. After twowashings in PBS, samples were incubated with anti-rabbitsecondary antibody conjugated to Alexa 635 (1:500) (MolecularProbes, Eugene, OR) for 30 min. Nuclei were stained with DAPI(49, 69-diamidino-2-phenylindole) (Molecular Probes, Eugene,OR). Slides were viewed with a Leica TCS SP2 confocal scanningmicroscope (Leica Microsystems, Heidelberg GmbH, Mannheim,Germany). Images were composed and analysed using AdobePhotoShop 8.0.Statistical analysesAll statistical analyses were performed using SPSS 10.0 (SPSS,Chicago, Il, USA). Effects of soil water, N addition interactionbetween soil water and N addition were analyzed using a two wayPLoS ONE | www.plosone.org 4 February 2012 | Volume 7 | Issue 2 | e3075466

Soil Water and Nitrogen Affect PhotosynthesisANOVA (p,0.05). Differences between the means among soilwater or N addition treatments were compared using Duncan’smultiple range tests at ,0.05 probability levels. For relationshipsof photosynthesis rate with leaf N content and SLA analysis wasalso performed (p,0.05). All the means involved in the interactionwere compared. All data were presented as mean 6 SD.ResultsGrowth of the seedlingsWater and N addition had significant effects on seedling growth.A significant interactive effect of N addition and water treatmenton plant height and root collar diameter of F. mandshurica seedlingswas found as described in Table 1. N addition stimulated asignificant increase in the height in HW treatment and root collardiameter of the seedlings in different water treatments (p,0.05).These two parameters markedly decreased under LW conditioncompared with that of the seedlings in CK, whereas N additionameliorated the reduction. Soil water and N addition hadadvantaged effects in height growth and root collar diameter. Naddition, water regimes and their interaction significantlyinfluenced total seedling biomass, aboveground biomass (Fig. 1A–C). The significant changes were detected in total seedling biomassand above-ground biomass by N addition under HW and CKconditions (p,0.05). Even under LW treatment, N addition alsoled to a significant increase in total seedling biomass, leaf biomassand stem biomass (p,0.05). N addition increased root biomassespecially in LW treatment. The ratio of the stem and rootbiomass (S/R) showed a significant decrease in LW treatmentcompared with that in CK (Fig. 1F). But no obvious changes weredetected in S/R ratio by N addition, suggesting that N additionmight not affect biomass allocation of the seedlings.Gas exchangeThe responses of A max , stomatal conductance (g s ), WUEi and Ein leaves of F. mandshurica seedlings to N and water were shownin Fig. 2. Leaf A max significantly increased by N addition underHW and CK, whereas this was not affected by N apply in LWtreatment (Fig. 2A). The HW treatment alone induced a slightincrease in g s , while the combination of N and HW treatment ledto a dramatic enhancement of g s (Fig. 2B). Similar to the responsesof g s , leaf transpiration rate was significantly enhanced by Naddition under HW (Fig. 2C). However, N addition didn’t changethe transpiration rate under CK and LW conditions. However,there was significant change in g s in response to N addition underCK and LW conditions. A significant decrease in WUEi was foundin LW treatment, whereas N addition didn’t affect WUEi (Fig. 2D).A slight increase in WUEi but no significant difference was foundby N addition under HW and CK conditions.Leaf N content and photosynthetic nitrogen-useefficiency (PNUE)In order to examine possible relationships between photosynthesisand N availability in different soil water and N additionconditions, the effects of soil water and/or N addition on leaf Ncontent and photosynthetic nitrogen-use efficiency (PNUE) weredetermined in the leaves of seedlings. There were significantdifferences in leaf N content between soil water treatments(Fig. 3A). HW treatment induced a significant increase in leaf Ncontent, and this increase was further enhanced by N addition.Similarly, N addition also led to an elevation in leaf N contentunder CK and LW conditions (Fig. 3A). These results showed aninteractive effect of soil N and water on the leaf N content in F.mandshurica seedlings. Leaf N content displayed significant positivecorrelation with A max (r 2 = 0.79, p,0.05, Fig. 4A) and was positivewith SLA (r 2 = 0.60, p,0.05, Fig. 4C). SLA was also positivelycorrelated with A max (r 2 = 0.82, p,0.05, Fig. 4B). No significantvariations in PNUE were found under different soil moisture(Fig. 3B). N addition resulted in a marked decrease for PNUE inLW treatment (Fig. 3B).Leaf chlorophyll contentTo examine how N addition and soil water affect photosyntheticcapacity, we determined the response of leaf chlorophyll contentsunder N apply and different soil moisture conditions. Theconcentrations of leaf chlorophyll including total Chl (a+b), Chla and Chl b were significantly influenced by N supply and watertreatments in F. mandshurica (p,0.05) (Fig. 5A–C). N supply hadsignificant positive effects on Chl a, Chl b and Chl (a+b) regardlessof soil water contents. Chl a/b ratios were not significantlydifferent in N addition or water treatment (Fig. 5D). In addition,the ratio of Car/Chl was dramatically increased under LWtreatment (Fig. 5E). However, N addition enhanced the synthesisof chlorophyll, which leading to a recovery in the ratio of Car/Chlunder LW condition (Fig. 5E).Rubisco activityRubisco catalyzes CO 2 assimilation and is a major limited factorin leaf photosynthetic responses of plants. We determined theeffects of N addition and water regimes on the total activity andinitial activity of Rubisco, as well as Rubisco activation state(Fig. 6). Water treatments had no significant effects on totalRubisco activity at natural N level (Fig. 6A). Total Rubisco activitywas significantly increased at high N level in HW and CK. Inaddition, the initial activity of Rubisco was not affected by waterregimes at natural N level and it was dramatically increased athigh N level in all water treatments (Fig. 6B). Rubisco activationstate was significantly increased by N supply (Fig. 6C).Table 1. The effects of N addition on plant height and root collar diameter of F. mandshurica seedlings grown in three differentsoil water regimes and two N treatment leaves.Growth characteristicsTreatmentNatural N levelHigh N levelHW CK LW HW CK LWPlant height (m) 2.6560.05 b 2.6760.06 b 2.3560.09 c 2.8560.03 a 2.6360.08 b 2.5260.11 bRoot collar diameter (mm) 30.6360.24 b 29.2660.87 b 25.8960.89 c 35.7660.29 a 33.5660.92 a 29.0261.30 bValues are mean 6SD of six replicates. And the same letter in the same row are not significantly different between treatments at the p,0.05 level.doi:10.1371/journal.pone.0030754.t001PLoS ONE | www.plosone.org 5 February 2012 | Volume 7 | Issue 2 | e3075467

Soil Water and Nitrogen Affect PhotosynthesisFigure 2. Effects of nitrogen addition and water regime on gas exchange. Parameters include: A. A max ,B.g s , C. WUEi, D. E. Each columnrepresents means 6 SD (n = 6). Different letters indicate significant differences among treatments at p = 0.05. Abbreviations: A max : maximumphotosynthetic rate; g s , stomatal conductance; WUEi intrinsic water use efficiency; E, transpiration.doi:10.1371/journal.pone.0030754.g002Expression of RubiscoIn addition to the activity and activation state of Rubisco, soil Nand water might affect photosynthetic responses by regulatingexpression of Rubisco in the leaves of the seedlings. Therefore, wefurther analyzed the effects of N addition and water regimes on theexpression of the large subunits in Rubisco in the leaves of F.mandshurica seedlings. The leaf protein of F. mandshurica seedlingswas separated by means of SDS-PAGE, and a 55 kDa distinctband showed up by western blot analysis (Fig. 7A). The analysis ofband intensity indicated that the expression of Rubisco was downregulatedby LW treatment, while the expression of the proteinwasn’t significantly influenced in HW treatment (Fig. 7B). Theexpression level of Rubisco was higher in N addition and LWtreatment than that in LW condition, suggesting that N additionincreased the protein expression of Rubisco under LW condition.However, under HW and CK conditions, N addition didn’tinduce a change in the expression of.Immunolocalization of RubiscoWe further detected the distribution of Rubisco in the leaves ofF. mandshurica seedlings. Immunofluorescent staining showed thatRubisco was found in chloroplasts throughout the leaf chlorenchymain the seedlings (Fig. 8). Labeling for Rubisco was abundantin chloroplasts of leaf tissues of HW and CK, while the labelingwas relatively low in leaves of the seedlings under LW condition.We observed that labeling for Rubisco was also concentrated inthe chloroplasts of leaf chlorenchyma after N addition. Theaccumulation of Rubisco has increased in leaf tissues of LW by Naddition, which was similar to the result from the expression ofRubisco by Western blots.DiscussionSoil N and water content are coupled tightly to the growth ofplants. However, the interactive effects of N addition and soilwater on plant physiological responses of tree seedlings havereceived relatively little attention [34]. In this study, wedemonstrated significant interactive effects of N addition and soilwater on the whole-plant growth and photosynthetic capacity of F.mandschurica seedlings in the temperate forest ecosystem innortheastern China. We showed that N addition increasedseedling growth including plant height, total biomass andaboveground biomass under different soil water conditions. Thereduction of the seedlings growth induced by low water supply wasPLoS ONE | www.plosone.org 6 February 2012 | Volume 7 | Issue 2 | e3075468

Soil Water and Nitrogen Affect PhotosynthesisFigure 3. Effects of nitrogen addition and water regimes onleaf nitrogen (panel A) and photosynthetic nitrogen-useefficiency (PNUE) (panel B). Each column represents mean 6 SD(n = 6). Mean values sharing the same letter are not significantlydifferent among treatments (p.0.05).doi:10.1371/journal.pone.0030754.g003significantly attenuated by N addition. We observed the strongphotosynthetic responses of F. mandschurica seedlings to N additionand water regimes. There were significant differences in leaf Ncontent between soil water treatments, and an interactive effect ofsoil N and water on the leaf N content was found. Leaf N contentdisplayed significant positive correlation with A max and also withSLA. N addition changed the photosynthetic capacity of seedlingsunder high water and natural conditions. N addition led to asignificant increase in leaf chlorophyll content and the initialactivity of Rubisco regardless of soil water condition. The proteinexpression of Rubisco was increased by N addition under LWcondition. Immunofluorescent staining showed that the labelingfor Rubisco was relatively low in leaves of the seedlings under LWcondition. The accumulation of Rubisco was increased in leaftissues of LW by N addition.In the present study, we found the interactive effects of Naddition and soil water on the whole-plant growth of F. mandshuricaseedlings in the temperate forest ecosystem. N addition stimulatedthe growth of the seedlings under different soil water conditions, asreflecting by a significant increase in seedling growth parameterssuch as plant height, root collar diameter, total biomass andaboveground biomass (Table 1). Similar effects of N addition onplant growth have been reported for annual grass and wheat[46,47]. The growth response of the seedlings to N addition andsoil water suggested that N supply could amplify the positive effectsof elevated soil moisture on plant growth [48]. In addition, thegrowth of the seedlings was negatively affected by LW treatmentand this tendency was partially diminished by N addition, whichwas consistent with the previous finding in Sophora davidii seedlings[20]. These results indicated that N addition might alleviate thenegative effects of LW manipulation on whole-plant growth of theseedlings. It demonstrated that N addition might play a key role inmaintaining plant productivity under different soil water conditionsin the temperate forest ecosystem.The shifts in biomass allocation had an important impact ontree growth in the acclimation to changes of soil nutrient andwater content [49,50]. The ratio of stem and root biomass (S/R)was an indicator that represented demand-supply balance forenvironmental stresses [51]. Nutrient limitation and drought stresswere found to increase carbon translocation from the leaves to theroots, thereby decreased the S/R ratio [52,53]. Similar result waspresented in our study, as the S/R ratio decreased with decreasingsoil water content (Fig. 1F, P,0.05), which supported theassumption that reduced soil water content could lead tocarbohydrate accumulation in the roots of plants [53]. Our resultsprovided the evidence that N addition did not drive an alternationin the ratio of the aboveground and belowground biomass inseedlings. Biomass allocation for F. mandshurica seedlings might benot primarily N limited.This study added important evidence in the interactive effects ofN addition and soil water on the photosynthetic rate of F.mandshurica seedlings and the investigations conducted in thedurum wheat [46] and hybridizing species [54]. We showed astrong interaction between N and soil water on A max of theseedlings in this ecosystem. N addition significantly enhanced theeffect of HW regime on A max . The photosynthetic responses to Navailability have been well documented in hardwood tree species[55–57], which indicated that the photosynthetic rate of theseedlings might be dependent on soil N availability in thetemperate ecosystem. We further investigated the changes ofSLA and leaf N content to explain the potential mechanism in leafphotosynthesis. SLA and leaf N content were both significantlycorrelated with A max of the seedlings (Fig. 4). The results weresimilar to that found in boreal tree species and wheat [46,58]. Inaddition, N addition triggered a significant increase in the stomataldiffusive conductance to H 2 O (g s ) of the seedlings under HWcondition (Fig. 2B). It is likely that N addition accelerate thetransport of photosynthetic CO 2 in the leaves, leading to enhancedA max of the seedlings.Leaf chlorophyll content is a good indicator of photosyntheticcapacity. Low concentrations of chlorophyll limit photosyntheticpotential directly and lead to a decrease in biomass production inthe plants [59]. In this study, a strong interaction of soil N andwater on leaf chlorophyll including total chlorophyll (a+b), Chl aand Chl b were found. Total chlorophyll (a+b), Chl a and Chl bcontent per unit area were all significantly increased in response toN addition in different soil water treatments (Fig. 5). The effects ofN addition on chlorophyll were in agreement with the previousfindings [20,60,61]. We also noticed a significant increase in Car/Chl ratio under LW condition.Rubisco is a key enzyme in photosynthesis and its activity is themain limitation for photosynthetic CO 2 fixation [62,63]. IncreasedN availability may affect the photosynthesis of plants by alteringthe activity of Rubisco in the leaves [13,64,65]. Previous studiessuggested that leaf A max is associated with Rubisco activity [66] orPLoS ONE | www.plosone.org 7 February 2012 | Volume 7 | Issue 2 | e3075469

Soil Water and Nitrogen Affect PhotosynthesisFigure 4. Correlations between A max vs. leaf N (panel A), A max vs. SLA (panel B), PNUE vs. WUEi (panel C) and leaf N vs. SLA (panel D). Datapoints are means of data from all the different water regimes and N treatments.doi:10.1371/journal.pone.0030754.g004its activation state [67]. In our study, increased initial activity ofRubisco and its activation state were found in the leaves of theseedlings under N addition and different soil water conditions(Fig. 6). Total activity of Rubisco in the leaves also augmented withN addition under HW and CK conditions. A highly positivecorrelation was observed between the initial activity of Rubiscoand leaf A max of the seedlings (data not shown). These resultsindicated that the initial activity of Rubisco was more closelyinvolved in the regulation of the photosynthetic rate than totalRubisco activity or its activation state in F. mandshurica seedlings.The result was consistent with the previous findings in Pinus pinaster[68]. It was known that Rubisco activity increased linearly withleaf N in plants [69–72]. We observed that leaf N content wasincreased with N addition in the seedlings (Fig. 3), which might atleast in part explained the increase of Rubisco activity.It has been reported that the protein synthesis of Rubisco wasinfluenced by leaf N content [73,74], we therefore hypothesizedthat differences in photosynthetic response to N addition may bedue to the expression of Rubisco in the leaves in F. mandshuricaseedlings. In this study, the protein expression of Rubisco under Naddition and water regimes were determined by immunoblottingand immuno-labeled techniques. We found that the expression ofRubisco in the leaves was down-regulated under LW conditionand the tendency was reversed by N addition in the seedlings,indicating that N addition could alleviate the negative response ofRubisco to LW. A lower expression level of Rubisco andphotosynthesis down-regulation were found in seedlings only insevere drought situation [75]. Immunofluorescent staining forRubsico showed that the immunolocalization of Rubisco occurredin chloroplasts of the leaves in the seedlings. Low accumulation ofRubisco was detected in the leaves of the seedlings grown in LWcondition, while labeling for Rubisco in the chloroplasts wasincreased by N addition in LW condition. These results wereconsistent with the findings from Western blots (Fig. 7, 8). Theamount of Rubisco was usually considered to be much greaterthan required for photosynthesis under a wide range ofenvironmental conditions [76–78]. Under plentiful soil water, Naddition might not stimulate Rubisco expression and therefore nosignificant effect on leaf Rubisco content was showed by Naddition. These results provided increasing evidence that Rubiscoin the leaves might be in excess and function as an N store innormal environmental condition for F. mandshurica seedlings in thetemperate forest ecosystem. The amount of Rubisco in the leaf wasdetermined by the balance between its synthesis and degradation[79,80]. With decreasing soil water content, the balance betweenRubisco synthesis and degradation might be disrupted, and moreRubisco was degraded in response to water stress, which led to adecrease in the amount of leaf Rubisco. N addition mightsignificantly increase the expression of Rubisco in the leaves toalleviate the negative response of photosynthesis in the seedlings.Therefore, in addition to effects of the enzyme activity andactivation state, N addition might affect the photosynthetic rate ofPLoS ONE | www.plosone.org 8 February 2012 | Volume 7 | Issue 2 | e3075470

Soil Water and Nitrogen Affect PhotosynthesisFigure 5. Impacts of nitrogen addition and water regimes onchlorophyll. Parameters include: A. Chl a, B. Chl b, C. Chl (a+b), D. Chla/b, E. Car/Chl. Each column represents mean 6 SD (n = 6). Mean valuessharing the same letter are not significantly different among treatments(p.0.05). Abbreviations: Chl, chlorophyll; Car, carotinoid.doi:10.1371/journal.pone.0030754.g005the seedlings by regulating the expression of Rubisco in the leavesunder low soil moisture in the temperate forest ecosystem.In summary, this study evaluated for the first time theinteractive effects of N and soil water on the growth andphotosynthetic responses of F. mandschurica seedlings in thetemperate ecosystem in northeastern China. We demonstratedthat the growth of the seedlings was positively affected bycombined manipulations of N addition and soil water. N additionFigure 6. Total and initial Rubisco activity and Rubiscoactivation state (panels A–C) under HW, CK, and LW conditionsin combinations with natural (dotted) or high N-supply level(hatched). Bars represent means of 6 replications 6 standarddeviation. Values accompanied by different letters differ significantlyat p = 0.05.doi:10.1371/journal.pone.0030754.g006PLoS ONE | www.plosone.org 9 February 2012 | Volume 7 | Issue 2 | e3075471

Soil Water and Nitrogen Affect PhotosynthesisFigure 7. Western blot analysis of large subunits of Rubisco in leaves of F. mandshurica seedlings (panel A). The relative expression levelis shown as the ratio of the band intensities between different treatments and CK with the analysis by Quantity One software (panel B).doi:10.1371/journal.pone.0030754.g007Figure 8. Confocal microscopy to show in situ immunolocalization of Rubisco in leaves of F. mandshurica seedlings. Label appears asgreen particles and nuclei are stained with DAPI (blue). (A) HW; (B) CK; (C) LW; (D) +NHW; (E) +NCK; (F) +NLW.doi:10.1371/journal.pone.0030754.g008PLoS ONE | www.plosone.org 10 February 2012 | Volume 7 | Issue 2 | e3075472

Soil Water and Nitrogen Affect Photosynthesissignificantly enhanced the growth and biomass production of theseedlings under plentiful soil water condition and could alleviatethe negative effect of LW treatment on plant growth. Furthermore,N addition could lead to a dramatic increase in the photosyntheticcapacity under high-water and natural conditions, which wasparalleled with the shifts of leaf chlorophyll content and Rubiscoenzymatic activity. Rubisco expression was up-regulated by Naddition in LW condition, which might be implicated inmaintaining the balance of its synthesis and degradation. Ourdata provided increasing evidence that N deposition might bebeneficial to biomass production and photosynthesis in forestseedlings in the temperate ecosystem.References1. Matson P, Lohse KA, Hall SJ (2002) The globalization of nitrogen addition:consequences for terrestrial ecosystems. Ambio 31: 113–119.2. Aber JD, Goodale CL, Ollinger SV, Smith ML, Magill AH, et al. (2003) Isnitrogen deposition altering the nitrogen status of northeastern forests?BioScience 53: 375–389.3. Compton JE, Watrud LS, Porteous LA, DeGrood S (2004) Response of soilmicrobial biomass and community composition to chronic nitrogen additions atHarvard forest. Forest Ecology and Management 196: 143–158.4. Schwinning S, Starr BI, Wojcik NJ, Miller ME, Ehleringer JE, et al. (2005)Effects of nitrogen deposition on arid grassland in the Colorado Plateau colddesert. Rangeland Ecology & Management 58: 565–574.5. Vitousek PM, Howarth RW (1991) Nitrogen limitation on land and in the sea:how can it occur. Biogeochemistry 13: 87–115.6. LeBauer DS, Treseder KK (2008) Nitrogen limitation of net primaryproductivity in terrestrial ecosystems is globally distributed. Ecology 89:371–379.7. Catovsky S, Bazzaz FA (2002) Nitrogen availability influences regeneration ofconiferous and broad-leaved tree species in the understory seedling bank.Ecological Applications 12: 1056–1070.8. Correia CM, Moutinho Pereira JM, Coutinho JF, Björn LO, Torres-Pereira JMG (2005) Ultraviolet-B radiation and nitrogen affect the photosynthesisof maize: a Mediterranean field study. Eur J Agron 22: 337–347.9. Latham RE (1992) Co-occurring tree species change rank in seedlingperformance with resources varied experimentally. Ecology 73: 2129–2144.10. Mitchell A, Hinckley T (1993) Effects of foliar nitrogen concentration onphotosynthesis and water used efficiency in Douglas-fir. Tree Physiol 12:403–410.11. Ripullone F, Grassi G, Lauteri M, Amato M, Borghetti M (2003) Photosynthesis–nitrogenrelationships: interpretation of different patterns betweenPseudotsuga menziesii and Populus euroamericana in a ministand experiment.Tree Physiol 23: 137–144.12. Vaitkus MR, Ciravolo TG, McLeod KW, Mavity EM, Novak KL (1993)Growth and photosynthesis of seedlings of five bottomland tree species followingnutrient enrichment. Am Midl Nat 129: 42–51.13. Field C, Mooney HA () The photosynthesis–nitrogen relationship in wild plants.In: Givnish T, ed. On the economy of plant form and function. Cambridge:Cambridge University Press, 1986, 22–55.14. Egli P, Schmid B (1999) Relationships between leaf nitrogen and limitations ofphotosynthesis in canopies of Solidago altissima. Acta Oecol 20: 559–570.15. Shangguan ZP, Shao MA, Dyckmans J (2000) Nitrogen nutrition and waterstress effects on leaf photosynthetic gas exchange and water use efficiency inwinter wheat. Environ Exp Bot 44: 141–149.16. DaMatta FM, Loos RA, Silva EA, Loureiro ME, Ducatti C (2002) Effects of soilwater deficit and nitrogen nutrition on water relations and photosynthesis of potgrownCoffea canephora Pierre. Trees 16: 555–558.17. Chandler JW, Dale JE (1995) Nitrogen deficiency and fertilization effects onneedle growth and photosynthesis in Sitka spruce (Picea sitchensis). TreePhysiology 15: 813–817.18. Lauter DJ, Munns DN, Clarkin KL (1981) Salt response of chickpea asinfluenced by N deposition. Agron J 73: 961–966.19. Yin CY, Berninger F, Li CY (2006) Photosynthetic responses of Populusprzewalski subjected to drought stress. Photosynthetica 44: 62–68.20. Wu FZ, Bao WK, Li LF, Wu N (2008) Effects of water stress and nitrogendeposition on leaf gas exchange and fluorescence parameters of Sophora davidiiseedlings. Photosynthetica 46: 40–48.21. Penuelas J, Rutishauser T, Filella I (2009) Phenology feedbacks on climatechange. Science 324: 887–888.22. Patrick LD, Ogle K, Tissue DT (2009) A hierarchical Bayesian approach forestimation of photosynthetic parameters of C 3 plants. Plant, Cell Environ 32:1695–1709.23. Lawlor DW (1995) Photosynthesis, productivity and environment. J Exp Bot 46:1449–1461.24. Brestic M, Cornic G, Fryer MJ, Baker NR (1995) Does photorespiration protectthe photosynthetic apparatus in Frech bean leaves from photoinhibition duringdrought stree? Planta 1996: 450–457.AcknowledgmentsWe thank all of the staff from the Changbai Mountain forest ecosystemresearch station for their assistance in collecting field data. We gratefullyacknowledge anonymous reviewers for helpful comments on thismanuscript.Author ContributionsConceived and designed the experiments: MW SS FL ZH. Performed theexperiments: SS. Analyzed the data: MW. Contributed reagents/materials/analysis tools: MW GD. Wrote the paper: MW. Designed thesoftware used in analysis: PJ.25. Foyer CH, Valadier MH, Migge A, Becker TW (1998) Drought-induced effectson nitrate reductase activity and mRNA and on the coordination of nitrogen andcarbon metabolism in maize leaves. Plant Physiology 117: 283–292.26. Chaves MM, Oliveira MM (2004) Mechanisms underlying plant resilience towater deficits: prospects for water-saving agriculture. Journal of ExperimentalBotany 55: 2365–2384.27. Flexas J, Bota J, Loreto F, Cornic G, Sharkey TD (2004) Diffusive and metaboliclimitations to photosynthesis under drought and salinity in C3 plants. PlantBiology 6: 269–279.28. Ennahli S, Earl HJ (2005) Physiological limitations to photosynthetic carbonassimilation in cotton under water stress. Crop Science 45: 2374–2382.29. Grassi G, Magnani F (2005) Stomatal, mesophyll conductance and biochemicallimitations to photosynthesis as affected by drought and leaf ontogeny in ash andoak trees. Plant, Cell & Environment 28: 834–849.30. Alves AAC, Setter TL (2000) Response of Cassava to Water Deficit: Leaf AreaGrowth and Abscisic Acid. Crop Sci 40: 131–137.31. Huang Y, Rothwell JC, Edwards MJ, Chen RS (2008) Effect of PhysiologicalActivity on an NMDA-Dependent Form of Cortical Plasticity in Human. CerebCortex 18: 563–570.32. Liu P, Huang J, Han X, Sun OJ, Zhou Z (2006) Differential responses of litterdecomposition to increased soil nutrients and water between two contrastinggrassland plant species of Inner Mongolia, China. Applied Soil Ecology 34: 266–275.33. Wang C, Wan S, Xing X, Zhang L, Han X (2006) Temperature and soilmoisture interactively affected soil net N mineralization in temperate grasslandin Northern China. Soil Biology and Biochemistry 38: 1101–1110.34. McDonald AJS, Davies WJ (1996) Keeping in touch: responses of the wholeplant to deficits in water and nitrogen supply. Advances in Botanical Research22: 229–300.35. Andersson I, Backlund A (2008) Structure and function of Rubisco. PlantPhysioI Biochem 46: 275–291.36. Warren CR, Dreyer E, Adams MA (2003) Photosynthesis-Rubisco relationshipsin foliage of Pinus sylvestris in response to nitrogen supply and the proposed roleof Rubisco and amino acids as nitrogen stores. Trees 17: 359–366.37. Gezelius K (1986) Free amino acids and nitrogen during shoot development inScats pine seedlings. Physiol Plant 67: 435–441.38. Laitinen K, Luomala E-M, Kellom ki S, Vapaavuori E (2000) Carbonassimilation and nitrogen in needles of fertilized and unfertilized field-grownScots pine at natural and elevated concentrations of CO 2 . Tree Physiol 20:881–892.39. Zhang M, Guan DX, Han SJ, Wu JB, Zhang JH, et al. (2005) Climatic dynamicsof broadleaved Korean pine forest in Changbai Mountain during the last 22years. Chinese Journal of Ecology 24: 1007–1012. (in Chinese with Englishabstract).40. Ellsworth DS (2000) Seasonal CO 2 assimilation and stomatal limitations in aPinus taeda canopy. Tree Physiology 20: 435–445.41. Yamaguchi, Masashi (2007) Pneumatic tire. Publication number: US 2010/0180994 A1.42. Inskeep WP, Bloom PR (1985) Extinction Coefficients of Chlorophyll a and b inN,N-Dimethylformamide and 80% Acetone. Plant Physiol 77: 483–485.43. Arnon DI (1949) Copper enzymes in isolated chloroplasts. Polyphenoloxidae inBeta vulgaris. Plant Physiol 24: 1D15.44. Tissue DT, Thomas RB, Strain BR (1993) Long-term effects of elevated CO 2and nutrients on photosynthesis and Rubisco in loblolly pine seedlings. PlantCell and Environment 16: 859–865.45. Bradford MM (1976) Anal Biochem. 72: 248–254.46. Cabrera-Bosquet L, Molero G, Bort J, Nogues S, Araus JL (2007) The combinedeffect of constantwater deficit and nitrogen supply on WUE, NUE and Delta C-13 in durum wheat potted plants. Ann Appl Biol 151: 277–289.47. Zhou X, Zhang Y, Ji X, Alison D, Serped M (2011) Combined effects ofnitrogen deposition and water stress on growth and physiological responses oftwo annual desert plants in northwestern China. Environmental andExperimental Botany. pp 1–8.48. Reich PB, Hungate BA, Luo YQ (2006) Carbon-nitrogen interactions interrestrial ecosystems in response to rising atmospheric carbon dioxide. AnnuRev Ecol Evol Syst 37: 611–636.PLoS ONE | www.plosone.org 11 February 2012 | Volume 7 | Issue 2 | e3075473

Soil Water and Nitrogen Affect Photosynthesis49. Poorter H, Remkes C (1990) Leaf area ratio and net assimilation rate of 24 wildspecies differing in relative growth rate. Oecologia 83: 553–559.50. Reich PB, Kloeppel BD, Ellsworth DS, Waiters MB (1995) Differentphotosynthesis-nitrogen relations in deciduous hardwood and evergreenconiferous tree species. Oecologia 104: 24–30.51. Lambers H, Chapin FS, Pons TL (1998) Plant Physiological Ecology. Springer-Verlag, New York 540.52. Andrews M, Sprent JI, Raven JA, Eady PE (1999) Relationships between shootto root ratio, growth and leaf soluble protein concentration of Pisum sativum,Phaseolus vulgaris and Triticum aestivum under different nutrient deficiencies.Plant Cell Environ 22: 949–958.53. Poorter H, Nagel O (1999) The role of biomass allocation in the growth responseof plants to different levels of light, CO 2 , nutrients and water: a quantitativereview. Aust J Plant Physiol 27: 1191–1191.54. Campbell DR, Wu CA, Travers SE (2010) Photosynthetic and growth responsesof reciprocal hybrids to variation in water and nitrogen availability. AmericanJournal of Botany 97: 925–933.55. Wendler R, Millard P (1996) Impacts of water and nitrogen supplies on thephysiology, leaf demography and nitrogen dynamics of Betula pendula. TreePhysiol 16: 153–159.56. Wang YP, Leuning R (1998) A two-leaf model for canopy conductance,photosynthesis and partitioning of available energy I: Model description andcomparison with a multi-layered model. Agricultural and Forest Meteorology91: 89–111.57. Tyree MC, Seiler JR, Maier CA (2009) Short-term impacts of nutrientmanipulations on leaf gas exchange and biomass partitioning in contrasting 2-year-old Pinus taeda clones during seedling establishment. Forest Ecology andManagement 257: 1847–1858.58. Reich PB, Walters MB, Tjoelker MG, Vanderklein DW, Buschena C (1998)Photosynthesis and respiration rates depend on leaf and root morphology andnitrogen concentration in nine boreal tree species differing in relative growthrate. Functional Ecology 12: 395–405.59. Naumann JC, Young DR, Anderson JE (2008) Leaf chlorophyll fluorescence,reflectance, and physiological response to freshwater and saltwater flooding inthe evergreen shrub, Myrica cerifera. Environmental and Experimental Botany63: 402–409.60. Van den Berg AK, Perkins TD (2004) Evaluation of portable chlorophyll meterto estimate chlorophyll and nitrogen contents in sugar maple (Acer saccharumMarsh.) leaves. Forest Ecology and Management 200: 113–117.61. Shaw B, Thomas TH, Cooke DT (2002) Responses of sugar beet (Beta vulgarisL.) to drought and nutrient deficiency stress. Plant Growth Regul 37: 77–83.62. Evans JR (1986) The relationship between carbon-dioxide-limited photosyntheticrate and ribulose-1,5-bisphosphate-carboxylase content in two nuclearcytouplasmsubstitution lines of wheat, and the co-ordination of ribulosebisphosphate-carboxylationand electron-transport capacities. Planta 167:351–358.63. Makino A, Mae T, Ohira K (1984) Relation between nitrogen and Ribulose-1,5-bisphosphate carboxylase in rice leaves from emergence through senescence.Plant and Cell Physiology 25: 429–437.64. Kutik J, Lubomir N, Demmers-Derks HH, et al. (1995) Chloroplastultrastructure of sugar beet (Beta vulgaris L.) cultivated in normal and elevatedCO 2 concentrations with two contrasted nitrogen supplies. Journal ofExperimental Botany 46: 1797–1802.65. Bondada BR, Syvertsen JP (2003) Leaf chlorophyll, net gas exchange andchloroplast ultrastructure in citrus leaves of different nitrogen status. TreePhysiology 23: 553–559.66. Bota J, Medrano H, Flexas J (2004) Is photosynthesis limited by decreasedRubisco activity and RuBP content under progressive water stress? The NewPhytologist 162: 671–682.67. Manter DK, Kavanagh KL, Rose CL (2005) Growth response of Douglas-firseedlings to nitrogen fertilization: importance of Rubisco activation state andrespiration rates. Tree Physiology 25: 1015–1021.68. Warren CR, Adams MA (2001) Distribution of N, Rubisco and photosynthesisin Pinus pinaster and acclimation to light. Plant Cell Environ 24: 597–609.69. Evans JR (1983) Nitrogen and photosynthesis in the flag leaf of wheat (Triticumaestivum L). Plant physiology 72: 297–302.70. Sage RF, Pearcy RW, Seemann JR (1987) The nitrogen use efficiency of C3 andC4 plants. III. Leaf nitrogen effects on the activity of carboxylating enzymes inChenopodium album (L.) and Amaranthus retroflexus (L.). Plant Physiology 85:355–359.71. Nakano H, Makino A, Mae T (1997) The effect of elevated partial pressure ofCO 2 on the relationship between photosynthetic capacity and N content in riceleaves. Plant Physiology 115: 191–198.72. Cheng L, Fuchigami LH (2000) CO 2 assimilation in relation to nitrogen in appleleaves. Journal of Horticultural Science and Biotechnology 75: 383–387.73. Stitt M (1996) Metabolic regulation of photosynthesis. In Photosynthesis and theEnvironment NR. Baker, ed. Kluwer Academic Publishers, Dordrecht. pp151–190.74. Nakaji T, Fukami M, Dokiya Y, Izuta T (2001) Effects of high nitrogen load ongrowth, photosynthesis and nutrient status of Cryptomeria japonica and Pinusdensiflora seedlings. Trees 15: 453–461.75. Asghari R, Ebrahimzadeh H (2006) Drought stress increases the expression ofwheat leaf Ribulose 1,5-Bisphosphate carboxylase/oxyenase protein. IranianJournal of Science and Technology, Transaction 30: 1–7.76. Stitt M, Schulze ED (1994) Does Rubisco control the rate of photosynthesis andplant growth? An exercise in molecular ecophysiology. Plant Cell Environ 17:465–487.77. Eichelmann H, Laisk A (1999) Ribulose-1, 5-bisphosphate carboxylase/oxygenase content, assimilatory charge, and mesophyll conductance in leaves.Plant Physiol 119: 179–189.78. Warren CR, Adams MA (2000) Capillary electrophoresis for the determinationof major amino acids and sugars in foliage: application to the nitrogen nutritionof sclerophyllous species. J Exp Bot 51: 1147–1157.79. Mae T, Makino A, Ohira K (1983) Changes in the amounts of Ribulosebisphosphate carboxylase synthesized and degraded during the life-span of riceleaf (Oryza sativa L). Plant and Cell Physiology 24: 1079–1086.80. Makino A, Mae T, Ohira K (1988) Differences between wheat and rice in theenzymic properties of ribulose-l,5-bisphosphate carboxylase/oxygenase and therelationship to photosynthetic gas exchange. Planta 174: 30–38.PLoS ONE | www.plosone.org 12 February 2012 | Volume 7 | Issue 2 | e3075474

Oikos 121: 1145–1153, 2012doi: 10.1111/j.1600-0706.2011.19757.x© 2011 The Authors. Oikos © 2011 Nordic Society OikosSubject Editor: Martin F. Qigley. Accepted 31 August 2011What happens below the canopy? Direct and indirect influencesof the dominant species on forest vertical layersZuoqiang Yuan, Antonio Gazol, Xugao Wang, Dingliang Xing, Fei Lin, Xuejiao Bai, Yuqiang Zhao,Buhang Li and Zhanqing HaoZ. Yuan, X. Wang, D. Xing, F. Lin, X. Bai, Y. Zhao, B. Li and Z. Hao (hzq@iae.ac.cn), State Key Laboratory of Forest and Soil Ecology,Inst. of Applied Ecology, Chinese Academy of Sciences, CN-110164 Shenyang, PR China. – DX, FL, XB, YZ and BL also at: GraduateUniv. of Chinese Academy of Science, CN-100049 Beijing, PR China. – A. Gazol, Inst. of Ecology and Earth Sciences, Univ. of Tartu,Lai 40, EE-Tartu 51005, Estonia.Temperate forests are one of the most important ecosystems in the world, and thus disentangling the factors that drivediversity within these ecosystems is of major concern. However, due to the complex interactions among forests layers,topography and soil factors, discovering the drivers of diversity is often complicated. In this study, we tested three a priorihypotheses about the effect of the dominant competitor (Pinus koraiensis) on the different forest layers in a 25 ha fullmapped plot of temperate forest in the Changbai Mountain of northeastern China. Structural equation modelling (SEM)was used to study the direct and indirect interactions between four vertical forest layers (dominant competitor, canopycomposition, sub-canopy diversity and shrub diversity), topographic factors, edaphic factors to discover sub-canopy andshrub diversity drivers. Our results suggest that the dominant competitor (Pinus koraiensis) is a key factor explainingcanopy variation, and sub-canopy diversity and shrub diversity, and that this competitor can act directly (through shading)and indirectly (through the modification of the soil). Topographic heterogeneity also had significant effects on thesoil variation and the diversity of the sub-canopy and shrub layers. Finally our results indicate that the influence of canopycomposition on the diversity of the rest of forest layers is indirect and positive, suggesting that the dominant competitor isthe main factor limiting diversity.In conclusion, we have found strong evidence that the dominant species of the canopy can influence, both directly andindirectly, the diversity of the different vertical forest layers. Patterns of diversity in forests are driven by a multiplicity offactors that are inherently related.Variations in topography (Beatty 1984), soil factors(Dzwonko and Gawronski 2002), and light availability(Miller et al. 2002, Härdtle et al. 2003) are assumed to bethe most important environmental gradients in forested ecosystems.Each of these factors may contribute to the determinationof forest composition and diversity (Peterken 1996).Similarly, biotic interactions such as competition (Tuomistoet al. 2003) and species dispersal (Schwarz et al. 2003) canplay a significant role in maintaining forest diversity. In addition,these different abiotic and biotic factors can be interrelated(Augusto et al. 2003), and thus plant–environmentfeedbacks can be a key factor to maintain diversity in forestedecosystems. For example, light fluctuations are determinedby the topographic heterogeneity but also by the verticalstructure and species composition of the different forest layers(Härdtle et al. 2003). Similarly, the soil chemistry canbe modified by the presence of different tree species (Finziet al. 1998, Gómez-Aparicio and Canham 2008, Vivancoand Austin 2008), altering niche conditions and thus speciescomposition. Therefore, the forces regulating forest diversitycan be the result of complex interactions among thesedifferent abiotic and biotic factors, influencing diversityboth: directly and indirectly.Canopy composition in temperate forests is usuallycomposed by a few numbers of species (Peterken 1996),resulting from the influence of competitive interactions duringtrees life cycle (Coates et al. 2009). Because competitionacts asymmetrically on a vertical scale (from canopy toshrub layers), a dominant competitor (sensu Goldberg 1990)may be a major factor driving diversity in the rest of the forestlayers. Similarly, the existence of competitive hierarchiesamong the different tree species (Coates et al. 2009) makespossible that that trees in each vertical layer can be influencedby trees present above them. However, niche differentiationand competition in forested ecosystems not only occurs vertically(shading; Härdtle et al. 2003) it has an importanthorizontal component (soil variability; John et al. 2007).The presence of different tree species can modify the soilconditions (Augusto et al. 2003), and thus the soil–canopyinteractions can play an important role on determining forestspecies composition (Finzi et al. 1998, Gómez-Aparicioand Canham 2008, Vivanco and Austin 2008). However,751145

these interactions can also be dependent on the influenceof topographic factors that can modify microsite conditions(Gazol and Ibáñez 2009). Therefore, there exists evidenceof the importance of vertical and horizontal competition, aswell as niche differentiation, for determining species diversityin forested ecosystems. However, to our knowledge, nostudies have tried to evaluate the relative contribution ofdirect effects (e.g. shading) and indirect effects (e.g. modificationof soil conditions) of tree species composition onforest diversity in a single model.One of the main problems regarding the study of multiplerelationships (i.e. direct and indirect) in plant communitiesis the selection of suitable methods allowing complexhypothesis testing (Scheiner and Willis 2005, Grace et al.2010). Recently, structural equation modelling (SEMs;Bollen 1989, Kline 2005) has attracted the interest of ecologistsdue to the possibility of this technique to link theoreticalconcepts and statistical techniques (Grace 2006). Itallows to the understanding of complex multiple interactionsamong factors, making possible to understand theirdirect and indirect (trough the modification of anotherfactor) effects on each other. Although SEMs have their rootsin scientific disciplines other than biological sciences, theefforts of several researchers have favoured their applicabilityto ecological problems (Iriondo et al. 2003, Scheiner andWillig 2005, Grace 2006). Despite the increasing applicationof SEM to ecological questions, to our knowledge, noone has yet tried to study the direct and indirect effects ofthe canopy layer on the diversity of the lower layers, thuspresenting an ideal opportunity to test these hypothesesthrough the use of structural equation models.Our objective in the present study is to model the directand indirect relationships between forest vertical layers, withemphasis on the importance of the dominant competitor.To achieve this objective, a 25-ha plot was established in amixed-broadleaved temperate forest in northeast China. Wehypothesized that the dominant competitor (Pinus koraiensis)would directly and indirectly modify the diversity of thedifferent forest layers. We expected: 1) a negative directeffect of Pinus koraiensis on canopy abundance and diversity(hereafter ‘canopy composition’) and sub-canopy and shrubrichness due to shading, 2) an indirect negative effect ofP. koraiensis on the above mentioned forest layers due to themodification of soil chemistry (nitrogen and water content)and 3) a reciprocal positive relationship between canopycomposition and soil conditions reflecting the complexcanopy-soil interactions in forested ecosystems. To test thesehypotheses, we constructed a SEM relating the topographicstructure of the study site, forest layers, and variations in soilnitrogen and moisture.Material and methodsStudy siteThe study area is located in the Changbai Mountain NaturalReserve, along the border of China and North Korea, extendingfrom 41°43¢ to 42°26¢N and 127°42¢ to 128°17¢E. Thearea has been part of the World Biosphere Reserve Networkunder the Man and the Biosphere Project since 1980; it isthe largest protected temperate forest in the world (Stone2006). Canopy composition is dominated by Pinus koreinsisaccompanied by species as Quercus mongolica. Pinus koreinsisis a common tree in the temperate humid region of China(Editorial board of Flora of China 1984). It prefers zoneswith moist conditions, deep and fertile soils, and slightlyacidic substrates. A detailed description of the study site canbe found in Hao et al. (2007).A 25-ha (500 500 m) plot was established in thesummer of 2004 in a broad-leaved mixed Korean pineforest (Hao et al. 2007, Wang et al. 2008). The area has atemperate continental climate with long cold winters andwarm summers. Mean annual precipitation is approximately700 mm, most of which occurs from June to September(480–500 mm). Mean annual temperature is 2.8°C, witha January mean of –13.7°C, and a July mean of 19.6°C(Yang et al. 1985). The terrain of the plot is relatively gentle,varying from 791.8 m to 809.5 m above sea level. The soilis classified as dark brown forest soil (Mollisols according toAmerican Soil Taxonomy Series 1999). Canopy compositionis dominated by Korean pine P. koraiensis, but Quercusmongolica and Tilia amurensis are also common. Between thesub-canopy and shrub layers several species of the Aceraceaefamily (e.g. Acer mono, A. tegumentosum and A. barvinerve),Corylus mandschurica and Syringa reticulata are frequentspecies.Data sampling procedureInside the study plot, all plant individuals with diameter atbreast height (DBH) 1 cm were tagged, identified, measured,and their geographic coordinates were recorded followinga standard field protocol (Condit 1998). The totalnumber of living individuals in the first census of 2004 was38 902, belonging to 52 species (Hao et al. 2007, Wanget al. 2008, Zhang et al. 2010). Species were classified in fourvertical layers: dominant competitor, canopy, sub-canopyand shrub (Supplementary material Appendix 1). After that,the study site was divided into a grid of 20 20 m, groupingspecies composition into the 625 plots of the grid. Twotopographic attributes, elevation and slope, were calculatedfor each 20 20 m quadrat in the plot. Following Harmset al. (2000), the elevation of each quadrat was obtainedfrom the mean elevation at the four corners of a quadrat.Each quadrat was divided into four triangular planes, eachformed by joining three corners of the quadrat. The averageangle of the four triangular planes that deviated from thehorizontal plane and the north direction provided the slopeof each quadrat.In October 2007, soil conditions were sampled followingstandard protocols (John et al. 2007) and using a 30 msampling grid. The volumetric soil water content (%) wasmeasured at each sample location at a depth of 20 cm usinga TDR probe. Additionally, three cores were taken at 10 cmdepth from within a 0.2 m area around the sample locationusing a 5-cm diameter cylinder. The three sub samples weremixed thoroughly to ensure that the sample was representativeof the surrounding area. Eight soil properties weredetermined according to Lu (1999), standard methods wereused to determine the following soil conditions (Yuan et al.2011): soil pH, soil carbon, available N, P and K, and total114676

N, P and K. Finally, geostatistical methods were used toobtain predicted values of soil conditions for each plot in the20 20 m grid.Data analysisStructural equation modelling (Bollen 1989, Kline 2005)with latent variables were used to analyze direct and indirectrelationships of: 1) forest layers; 2) soil conditions 3) elevationand 4) topographic heterogeneity. The first step is thecreation of a theoretical model relating the different factorsmentioned above on the basis of previous knowledge of thesystem and theory (Grace 2006). Our theoretical modelimplies the use of three latent exogenous variables and fourendogenous variables (Fig. 1). The three exogenous variableswere position, topographic heterogeneity, and dominantcompetitor, and were created with three indicators (elevation,slope and P. koreinsis abundance, respectively). The fourendogenous latent variables were: canopy composition, soilvariation, sub-canopy diversity and shrub diversity. Canopylayer has two indicators: species evenness (calculated as theShannon diversity index divided by the natural logarithm ofspecies richness) and number of steams per plot (hereafter,canopy density). Soil variation has also two indicators: soiltotal nitrogen content and soil moisture. We used only nitrogenand moisture to reflect soil conditions because pH variationwas relatively low and phosphorus and potassium werestrongly correlated with nitrogen. Sub-canopy and shrubdiversity were created with one indicator each: sub-canopyand shrub richness, respectively. Classification of species inthe different vertical layers can be seen in the Supplementarymaterial Appendix 1.Our theoretical model specifies a reciprocal interaction(feedback loop) between canopy composition and soil conditions(Fig. 1), and thus it is a nonrecursive SEM. Reciprocalinteractions are static representations of temporal dynamics,and they imply the indirect effect of one variable on itself(Grace 2006). The usage of non-recursive models in ecologyis not common (but see Johnson et al. 1991), and theircalculation requires several important considerations (Grace2006). The most important one is that several direct effectshave to be omitted from the model in order to make itscalculation possible. Our theoretical model hypothesizedthat the dominant competitor (P. koraiensis) can have adirect negative influence on canopy composition and subcanopydiversity due to shading (Härdtle et al. 2003). Similarly,we hypothesized that canopy composition can have thesame negative influence on sub-canopy and shrub diversity,reflecting competitive hierarchies (Coates et al. 2009). But wealso hypothesize that the canopy composition, and thus thedominant competitor, can influence sub-canopy and shrubdiversity indirectly through the modification of the soil conditions(Finzi et al. 1998, Augusto et al. 2003, Vivanco andAustin 2008). However, the soil variability can also influencecanopy composition (Tuomisto et al. 2003), and thus wecreate a reciprocal interaction between soil variability andcanopy composition. Due to the vertical asymmetry of theforest layers, sub-canopy diversity can influence shrub diversity.Since topography can be the main factor determiningsoil variation (Augusto et al. 2003), we expect direct effectsof position and topography on soil heterogeneity. We alsohypothesize that the gradient in elevation can influence canopycomposition and that the variation in topography can increaseshrub diversity (Gazol and Ibáñez 2009). Moreover, due tothe gentle topography of the study site (Hao et al. 2007) theelevation can serve as a proxy to control the spatial autocorrelationof the data, which in turn can influence the interpretationof structural equation results (Grace et al. 2010).The package ‘lavaan’ (ver. 0.4-9; Rosseel 2011) in theR statistical language (R development core Team 2009)evennessdensityPinuskoraiensisdominantcompetitorcanopycompositionsp. richnesssub-canopydiversityelevationpositionshrubdiversitysp. richnessslopetopographysoilvariabilityNitrogenmoistureFigure 1. Hypothetical model constructed to represent theoretical direct and indirect relationships of factors driving sub-canopy andshrub diversity. Solid arrows indicate positive effects, while dashed ones indicate negative relationships. Five out of nine indicators(i.e. variables into rectangular boxes) were assumed to be measured without error: elevation, slope, Pinus koraiensis, sub-canopy richnessand shrub richness.771147

was used to analyze the structural equation models. Toachieve normality assumptions several variables were transformedbefore the analyses. Specifically, P. koraiensis abundanceand canopy density were square-root transformed( ( x 10.5)), and slope, sub-canopy and shrub richness werelog transformed (log(x 11)). Since we use several latent variablesthat have only one indicator, those indicators were consideredto be measured without error (i.e. fixing their error at0). The maximum likelihood method was used to calculatethe model. The overall fit of the model was assessed usingthe c 2 statistic and associated probability, the RMSEA index,and the comparative fit index. These different indexes can beobtained using the command: ‘fit.measurements TRUE’when fitting the SEM. The significance of the differentcausal relationships hypothesized in the model was judgedaccording to their associated probability (p 0.05).One of the major problems with these analyses is the inflationof the type I error due to the presence of spatial autocorrelationin the data (Legendre 1993). Therefore, we studiedthe spatial autocorrelation of the sub-canopy and shrub richnessand their residuals after adjusting the structural equationmodels. Residual values were obtained by performinga linear regression of the two layers’ richness against the restof the variables (the shrub richness model also includes thesub-canopy richness as a predictor). Note that the results ofthe SEM and the linear model will vary, since it is not possibleto use latent variables in regression analyses. To studyspatial autocorrelation we used the Moran correlogram,which is a graphical distribution of the Moran’s I statisticagainst distance classes (Legendre and Legendre 1998).Correlograms were constructed using distance classes withincrements of 20 m up to a maximum distance of 400 m.The significance of the Moran’s I statistic at each distanceclasses was tested against 999 permutations. The global significanceof the entire correlogram (p 0.0025 at least forone distance class) was based on a Bonferroni correction(Legendre and Legendre 1998). The SAM program (Rangelet al. 2010) was used to analyze the spatial autocorrelation ofsub-canopy and shrub richness and their residual variation.Additionally, a conditional autoregressive model (CAR) ofsub-canopy and shrub richness (Rangel et al. 2010), usingthe rest of the observed variables as predictors, was used todiscover the fraction of variance explained when controllingfor spatial autocorrelation. CAR explicitly incorporates in themodel the spatial relationships between pairs of sites (Rangelet al. 2010). We assumed that distant locations affect eachother less than nearby locations, and thus the weight in theCAR model was defined to decrease with increasing distance(1/distance).ResultsWe used 50 species in the analyses. Seven species (Ulmusjaponica, Quercus mongolica, Tilia amurensis, Abies holophylla,Populus koreiensis, Populus ussuriensis and Tilia mandshurica)were considered to be part of the canopy (excluding Pinuskoereinsis, which was considered the dominant competitor,18 were classified as subcanopy species, and 24 as the shrublayer species (Supplementary material Appendix 1). Values ofrichness and number of individuals were variable across thestudy site (Table 1). The study site is topographically gentle,as is indicated by the low variation in elevation and slopevalues (Table 2). However, soil conditions show a highervariation, mostly in soil moisture content (Table 2).The final model fit partially with the relationships consideredin our theoretical model (Fig. 2). The c 2 statistic(17.490; p 0.231 with 14 degrees of freedom), comparativefit index (0.996) and the RMSEA index (0.020; p 0.976)all indicated that the model shall be accepted (i.e. lower devianceof the empirical and predicted covariance matrix). Inthis model, the dominant competitor had direct negativeeffects on canopy composition and sub-canopy diversity, andalso indirectly on shrub diversity (Fig. 2, Table 3). Canopycomposition did not have significant direct effects either onsub-canopy or shrub diversity, but it influences positively soilvariation. The results showed that soil variation had positivedirect influences on canopy composition, sub-canopy diversityand shrub diversity, but it was only significant in the caseof sub-canopy diversity. Therefore, canopy composition hada positive indirect effect on sub-canopy composition troughthe modification of the soil conditions. The total effect ofcanopy composition on sub-canopy and shrub diversity waspositive (Table 4), as well as the total effect of soil variationon the diversity of these two layers. Similarly, the dominantcompetitor had negative indirect effects on sub-canopy andshrub diversity through the modification of the canopy compositionand the soil variation (Fig. 2, Table 3). The totaleffect of the dominant competitor on the three forest layerswas negative (Table 4). Soil variability was also directly influencedby the position and topography, both having positiveeffects on it. Finally, the model shows that position andtopography had direct influences on sub-canopy and shrubdiversity respectively. All the coefficients associated with thedifferent arrows of the models are shown in Table 3. Finally,the model indicates that 9% of the variation in canopy composition,26% of the variance in sub-canopy diversity, 10%of the variance in shrub diversity and 10% of soil variationwas explained by the model.The analyses of spatial autocorrelation of the sub-canopyand shrub diversity, and their residual variation after a linearregression against the different variables, showed that a greatamount of spatial structure was accounted for by the predictorvariables (Fig. 3). Spatial autocorrelation, as calculated bythe Moran I index, was found to be significant for 19 out of20 distance classes in the sub-canopy diversity correlogram,and for 15 out of 20 distance classes in the shrub diversitycorrelogram. The two correlograms were statistically significant(p 0.0025 at least for one distance class). Conversely,the analyses of the residuals showed that only six out ofTable 1. Descriptive statistics of the forest vertical layers (dominantcompetitor, canopy, sub-canopy and shrub) studied in the 625 quadratsof the Changbai plot. The number of species in each layer, mean( SD) number of individuals per plot, and mean ( SD) speciesrichness of each layer are shown.Vertical layerNo. of No. of individualsspecies (mean SD)Species richness(mean SD)Dominant competitor 1 3.9 2.6 2Canopy layer 7 8.2 4.0 2.6 0.78Sub-canopy layer 18 25.7 11.7 4.5 1.4Shrub layer 24 20.1 10.1 2.9 1.1114878

Table 2. Descriptive statistics of the different environmental variablesstudied in the 625 quadrats of the Changbai plot. The units ofmeasurement, mean ( SD) value and the range of values for eachvariable are shown.Variable Units Mean SD RangeElevation m 803.2 3.5 7932809Slope degrees 3.2 2.4 0.3216.1Soil water content % 40.0 6.7 13.1260.4Soil nitrogen content mg kg 21 6.4 1.8 2.5212.420 and four out of 20 distance classes showed significantspatial autocorrelation. Only the correlogram of sub-canopydiversity was statistically significant. The conditional autoregressivemodels (CAR) showed that sub-canopy richness wassignificantly explained by slope, Pinus abundance, canopyevenness, total nitrogen, and soil moisture (Table 5). Similarly,shrub richness was significantly explained by elevation,canopy evenness, sub-canopy richness, and total nitrogen.The results were consistent were those obtained in the SEMand there were no important differences between the regressioncoefficients in the ordinal least square (OLS) and theCAR models, indicating that spatial autocorrelation doesnot strongly modify the results.DiscussionOur study provides insight into the importance of nichedifferentiation and inter-specific competition (Chesson 2000,Levine and HilleRisLambers 2009) for forest diversity. Byusing structural equation modelling, we were able to discoverthe direct and indirect causal relationships and multipleinteractions that drive diversity. We provide evidencethat forested ecosystems are of complex nature and that theeffects of the different forest layers on each other are notonly the result of direct interactions. Our results suggest thatthe dominant competitor (i.e. dominant tree) is a key factordetermining variations in the diversity of forest vertical layers,both directly and indirectly. Pinus koreinsis lowers diversityprobably by reducing light values (Canham et al. 1999,Härdtle et al. 2003) and indirectly modifying soil conditions(Finzi et al. 1998). Our results also suggest that competitiveinteractions in forested ecosystems are restricted to the influenceof the dominant competitor, indicating that facilitativeinteractions can also play an important role in determiningforest diversity.The direct negative influence of Pinus abundance oncanopy composition (evenness and density) and diversity ofthe sub-canopy layer can be attributed to shading (Canhamet al. 1999, Härdtle et al. 2003, Coates et al. 2009). Alongthese lines, Zhou (2003) reported that the Leaf Area Index,and thus shading, was lower in mixed-broadleaved foreststhan in the pure Pinus forests of the Cangbai Mountain. Thepresence of different deciduous species such as Quercusmongolica, Ulmus japonica and Tilia amurensis, can increasethe values of light in the understory, thus enabling the presenceof more diverse sub-canopy and shrub layers. Althoughnot significant, the positive effect of canopy compositionon sub-canopy and shrub diversity partially supports thehypothesis of higher values of light, and thus higher diversity(Härdtle et al. 2003), under a mixed- than under purecanopystands. Along the same lines, the positive influenceof sub-canopy diversity on shrub diversity seems to indicatethat these suitable places, with lower abundance of Pinusindividuals and thus higher light values, favour both subcanopyand shrub diversity, giving importance to the environmentalfilters and microsite conditions (Beatty 1984,Keddy 1992). This direct influence of Pinus on the rest offorest layers, excluding shrub diversity, indicates that verticalevennessdensityPinuskoraiensisdominantcompetitorcanopycompositionR 2 =0.09sp. richnessR 2 =0.26sub-canopydiversityelevationpositionR 2 =0.10shrubdiversityslopetopographysoilvariabilityR 2 =0.10sp. richnessNitrogenmoistureFigure 2. Final model showing direct and indirect relationships of factors driving sub-canopy and shrub diversity. Solid arrows indicatepositive effects, while dashed ones indicate negative relationships. Width of the arrows is proportional to the standardized effect of the pathrepresented (thin arrows represent non significant effects). The coefficient associated with each arrow can be seen in Table 2. The fractionof variance explained in each endogenous variable is shown.791149

Table 3. SEM results for the model in Fig. 2. Unstadardized and standardized coefficients as well as standard error are shown for eachpathway. The probability indicates the statistical significance of the pathwat (p 0.05).Pathway Unstandardized coefficient Standardized coefficient Standard error p-valueposition * ® canopy composition 0.004 0.120 0.003 0.101position ® soil variability 0.130 0.393 0.023 0.001position ® shrub diversity 0.005 0.159 0.001 0.001topography ® soil variability 2.329 0.430 0.346 0.001topography ® sub-canopy diversity 0.054 0.100 0.020 0.008dominant competitor ® canopy composition 20.043 20.224 0.013 0.001dominant competitor ® sub-canopy diversity 20.043 20.241 0.007 0.001canopy composition ® soil variability 4.936 0.525 1.921 0.010canopy composition ® sub-canopy diversity 0.139 0.149 0.089 0.118canopy composition ® shrub diversity 0.119 0.122 0.085 0.160soil variability ® canopy composition 0.012 0.116 0.015 0.420soil variability ® sub-canopy diversity 0.029 0.296 0.004 0.001soil variability ® shrub diversity 0.004 0.034 0.005 0.468sub-canopy diversity ® shrub diversity 0.178 0.170 0.049 0.001position§ « position 12.388 1.000 0.701 0.001topography « topography 0.046 1.000 0.003 0.001dominant competitor « dominant competitor 0.418 1.000 0.024 0.001position « topography 20.324 20.430 0.033 0.001position « dominant competitor 0.226 0.099 0.091 0.013topography « dominant competitor 20.026 20.184 0.006 0.001canopy « canopy 0.014 0.915 0.008 0.077soil « soil 1.222 0.905 0.328 0.001canopy « soil 20.071 20.545 0.049 0.148sub-canopy diversity « sub-canopy diversity 0.010 0.742 0.001 0.001shrub diversity « shrub diversity 0.013 0.904 0.001 0.001position ® elevation 1.000 - - -topography ® slope 1.000 - - -dominant competitor ® Pinus koraiensis 1.000 - - -canopy composition ®canopy evenness 1.000 0.654 - -canopy composition ® canopy density 1.114 0.215 0.573 0.052soil variability ® nitrogen content 1.000 0.654 - -soil variability ® water content 5.542 0.938 0.454 0.001sub-canopy diversity ® sub-canopy richness 1.000 - - -shrub diversity ® shrub richness 1.000 - - -elevation « elevation 0 - - -slope « slope 0 - - -Pinus koraiensis « Pinuskoraiensis 0 - - -canopy evenness « canopy evenness 0.020 0.572 0.008 0.010canopy density « canopy density 0.391 0.954 0.024 0.001nitrogen content « nitrogen content 1.805 0.572 0.138 0.000water content « water content 5.637 0.120 2.831 0.046sub-canopy richness « sub-canopy richness 0 - - -shrub richness « shrub richness 0 - - -* single-headed arrows indicate direction of causation.§ double-headed arrows error terms or correlation between latent variables.inter-specific competition occurs in this forest, and that itlowers diversity (Chesson 2000). Moreover, the resultssupport the idea of the dominant competitor (Goldberg1990), and that this competition occurs in a vertical scale.Indirect influences of Pinus, also negative, on the diversityof the sub-canopy and shrub layers are driven through themodification of soil conditions. Although the most importantinfluences of the dominant competitor occurs directly,indirect relationships are also significant proving that treespecies composition can modify soil conditions and thusforest diversity (Augusto et al. 2003). A higher abundance ofPinus trees, accompanied by a low frequency of other trees,can create a relatively pure Pinus litter layer (Li et al. 2007).This pattern can result in lower values of pH and soil organicmatter content than in mixed broad-leaved and pine forests(Wu and Han 1990, Qin et al. 2001). These changes can alsoinfluence the water retention properties of the soil. Furthermore,the lower light penetration under a pure Pinus canopy,combined with the higher acidity of its leaves, can lower biologicalactivity and thus reduce nitrogen values (Qin et al.2001, Li et al. 2007). These results are in accordance withother studies that have shown that canopy composition caninfluence soil biogeochemical cycles (Vivanco and Austin2008) and species composition (Finzi et al. 1998, Augustoet al. 2003, Gómez-Aparicio and Canham 2008), thusemphasizing its importance as a diversity driver in forested115080

Table 4. Standardized total, direct and indirect effects of the differentvariables on sub-canopy and shrub diversity according to thestructural equation model shown in Fig. 2.Sub-canopyShrubVariables Total Direct Indirect Total Direct IndirectDominant 20.314 20.241 20.073 20.087 0.000 20.087competitorPosition 0.170 0.000 0.170 0.225 0.159 0.067Topography 0.244 0.100 0.143 0.063 0.000 0.063Canopy 0.324 0.149 0.175 0.204 0.122 0.082compositionSoil 0.334 0.296 0.038 0.108 0.034 0.074Sub-canopydiversity- - - 0.170 0.170 0.000ecosystems. However, the negative influence of the dominantcompetitor on the rest of forest layers can be seen as apositive influence of the rest of canopy species in sub-canopyand shrub diversity. The lower abundance of Pinus facilitatesthe presence of a different canopy, enabling higher values ofdiversity in the rest of forest vertical layers. Moreover, thereciprocal interaction between canopy composition and soilconditions, although not significant, indicates that the presenceof a canopy with more deciduous trees creates betterconditions for itself (i.e. positive relationship). The canopyeffect mainly occurs through the modification of the soilconditions, emphasizing the horizontal component of nichediversification and inter-specific interactions in forestedecosystems (John et al. 2007, Coates et al. 2009).It is notable that position and topographic heterogeneityhave a significant influence on the diversity of the twolayers. In our opinion, this indicates the important influenceof factors other than pure topographic heterogeneity.The topographic structure of the study site is relatively gentle(20 m of variation) and the slope values are relatively low.The influence of topography on diversity found in this studycould serve as a proxy for processes related to the spatialconfiguration of species distributions, such as dispersal(Legendre 1993, Laliberté et al. 2009). Similarly, althoughmost of the spatial variation in sub-canopy and shrubdiversity was explained by the different factors studied, theremaining spatial autocorrelation could indicate the influenceof spatially stochastic processes (Bell 2001, Hubbell2001). Along these lines, several studies have reported(a)0.4Moran's I0–0.4sub-canopy richnessresidual variation20 60 100 140 180 220 260 300 340 380(b)0.4Moran's I0–0.4shrub richness residual variation20 60 100 140 180 220 260 300 340 380Distance (m)Figure 3. Moran’s I spatial autocorrelograms of richness in the sub-canopy (a) and shrub layers (b). Solid lines are the correlograms of themeasured variables and dashed lines of the residuals after fitting a linear model. Solid circles indicate significant spatial autocorrelation atthe considered distance class while empty ones indicate lack of spatial autocorrelation.811151

Table 5. Results of the ordinary least square (OLS) and conditionalautoregressive models (CAR) of the sub-canopy and shrub richnessagainst the different predictor variables.Sub-canopyShrubVariables OLS CAR p-value OLS CAR p-valueIntercept 20.180 0.171 0.901 25.091 24.826 0.001Elevation 0.001 0.001 0.778 0.007 0.007 0.001Slope 0.066 0.063 0.010 0.029 0.023 0.421Pinus 20.047 20.044 0.001 0.005 0.002 0.767koraiensiscanopy 0.055 0.054 0.013 0.052 0.055 0.029evennesscanopy 0.005 0.005 0.433 20.003 20.004 0.588densityNitrogen 0.008 0.008 0.005 0.009 0.009 0.011contentwater 0.003 0.003 0.001 20.001 20.001 0.170contentSub-canopyrichness† 2 2 2 0.177 0.169 0.001† sub-canopy richness was used as predictor variable in the shrubrichness regression model.spatial autocorrelated patterns in other forested ecosystems,interpreting them as the influence of dispersal processes(Miller et al. 2002, Schwarz et al. 2003, Laliberté et al.2009). Therefore, we hypothesize that diversity of the differentforest layers is mainly driven by niche diversificationand inter-specific competition, but other stochastic processessuch as dispersal can play a secondary role (Tuomisto et al.2003). Otherwise, the effect of topography and position canbe reflecting the influence of processes acting at scales largerthan those reflected with our study grain (Levin 1992). Dueto the long life of the tree species that comprises the forestcanopy (Editorial board of Flora of China 1984), the spatialconfiguration of position and topography can be reflectingpast forest changes not captured in our study.To summarize, we have found strong evidence thatthe dominant species of the canopy can influence, bothdirectly and indirectly, the diversity of the different verticalforest layers. Similarly, we demonstrate that positiveand negative interactions occurs simultaneously in forestedecosystems, and that their final outcome depends on complexrelationships. Patterns of diversity in forests are drivenby a multiplicity of factors that are inherently related. Themodel presented shows that a modification of one of thefactors creates a cascade effect that can influence the diversityof the different forest layers. However, the dominantcompetitor appears as a key factor, modifying light and soilconditions, and thus driving vertical diversity. In absenceof perturbations, the increase of the dominant competitorwill lower diversity due to the displacement of othercanopy trees. The use of structural equation mode is advantageousbecause it provides a means of direct hypothesistesting and is also a useful tool for modelling and understandingcomplex interactions that are of great importancein forested ecosystems.Acknowledgements – This study was sponsored by the KnowledgeInnovation Program of the Chinese Academy of Sciences (KZCX2-EW-401 and KSCX2-EW-Z-5), the National Natural ScienceFoundation of China (40971286,31061160188) and the NationalKey Technologies R&D Program of China (2008BAC39B02). AGhas the financial support of the ERMOS programme (co-foundedby the Marie Curie Actions) – ERMOS14. We are grateful toFangliang He for providing constructive comments on manuscriptof this paper. We would also like to thank Anne Bjorkman,Erin Kurten and Ryan Chisholm for their helpful assistance withEnglish language and grammatical editing of the manuscript.ReferencesAugusto, L. et al. 2003. Effects of tree species on understoryvegetation and environmental conditions in temperate forests.– Ann. For. Sci. 60: 823–831.Beatty, S.W. 1984. Influence of microtopography and canopy specieson spatial patterns of forest understory plants. – Ecology65: 1406–1419.Bell, G. 2001. Neutral macroecology. – Science 293: 2413–2417.Bollen, K. A. 1989. Structural equations with latent variables.– Wiley.Canham, C. D. et al. 1999. Measurement and modeling of spatiallyexplicit variation in light transmission through interior cedar–hemlock forests of British Columbia. – Can. J. For. Res. 29:1775–1783.Chesson, P. 2000. Mechanisms of maintenance of species diversity.– Annu. Rev. Ecol. Syst. 31: 344–366.Coates, K. D. et al. 2009. Above- versus below-ground competitiveeffects and responses of a guild of temperate tree species.– J. Ecol. 97: 118–130.Condit, R. 1998. Tropical forest census plots: methods and resultsfrom Barro Colorado Island, Panama and a comparison withother plots. – Springer.Dzwonko, Z. and Gawronski, S. 2002. Effect of litter removal onspecies richness and acidification of a mixed oak–pine woodland.– Biol. Conserv. 106: 389–398.Editorial Board for Flora of China, 1984. Flora of China. – SciencePress, Beijing, in Chinese.Finzi, A. C. et al. 1998. Canopy tree–soil interactions withintemperate forests: species effects on pH and cations. – Ecol.Appl. 8: 447–454.Gazol, A. and Ibáñez, R. 2009. Different response to environmentalfactors and spatial variables of two attributes (cover anddiversity) of the understorey layers. – For. Ecol. Manage. 258:1267–1274.Goldberg, D. E. 1990. Components of resource competition in plantcommunities. –In: Grace, J. B. and Tilman, D. (eds), Perspectivesin plant competition. Academic Press, pp. 27–65.Gómez-Aparicio, L. and Canham, C. 2008. Neighbourhoodmodels of the effects of invasive tree species on ecosystempro cesses. – Ecol. Monogr. 78: 69–86.Grace, J. B. 2006. Structural equation modeling and naturalsystems. – Cambridge Univ. Press.Grace, J. B. et al. 2010. On the specification of structural equationmodels for ecological systems. – Ecol. Monogr. 80: 67–87.Hao, Z. Q. et al. 2007. Vertical structure and spatial associationsof dominant tree species in an old growth temperate forest.– For. Ecol. Manage. 252: 1–11.Härdtle, W. et al. 2003. The effects of light and soil conditions onthe species richness of the ground vegetation of deciduous forestsin northern Germany (Schleswig-Holstein). – For. Ecol.Manage. 182: 327–338.Harms, K. E. et al. 2000. Pervasive density-dependent recruitmentenhances seedling diversity in a tropical forest. – Nature 404:493–495.Hubbell, S. P. 2001. The unified neutral theory of biodiversity andbiogeography. – Princeton Univ. Press.Iriondo, J. M. et al. 2003. Structural equation modelling: an alternativefor assessing causal relationships in threatened plantpopulations. – Biol. Conserv. 113: 367–377.115282

John, R. et al. 2007. Soil nutrients influence spatial distributionsof tropical tree species. – Proc. Natl Acad. Sci. USA 104:864–869.Johnson, M. L. et al. 1991. Ecosystem modeling with LISREL: anew approach for measuring direct and indirect effects. – Ecol.Appl. 1: 383–398.Keddy, P. A. 1992. Assembly and response rules: two goals forpredictive community ecology. – J. Veg. Sci. 3: 157–164.Kline, R. B. 2005. Principles and practice of structural equationmodelling, 2nd ed. – Guilford Press.Laliberté, E. et al. 2009. Assessing the scale-specific importanceof niches and other spatial processes on beta diversity: a casestudy from a temperate forest. – Oecologia 159: 377–388.Legendre, P. 1993. Spatial autocorrelation: trouble or new paradigm?– Ecology 74: 1659–1673.Legendre, P. and Legendre, L. 1998. Numerical ecology. – Elsevier.Levin, S. A. 1992. The problem of pattern and scale in ecology:the Robert H. MacArthur Award Lecture. – Ecology 73:1943–1967.Levine, J. M. and HilleRisLambers, J. 2009. The importance ofniches for the maintenance of species diversity. – Nature 461:254–257.Li, X. et al. 2007. Foliar decomposition in a broadleaf-mixedKorean pine (Pinus koraiensis Sieb. Et Zucc) plantation forest:the impact of initial litter quality and the decomposition ofthree kinds of organic matter fraction on mass loss and nutrientrelease rates. – Plant Soil 295: 151–167.Lu, R. K. 1999. Analytical methods of soil agrochemistry. – ChinaAgric. Sci. and Technol. Press, in Chinese.Miller, T. F. et al. 2002. Old-growth northern hardwood forest:spatial autocorrelation and patterns of understory vegetation.– Ecol. Mongr 72: 487–503.Peterken, G. F. 1996. Natural woodland: ecology and conservationin northern temperate regions. – Cambridge Univ. Press.Qin, C. Y. et al. 2001. Study on growth of Pinus koraiensis broadleafmixed forest and its silvicultural techniques. – Jilin. For.Sci. Tech. 30: 18–20, in Chinese.Rangel, T. F. L. V. B. 2010. SAM: a comprehensive applicationfor spatial analysis in macroecology. – Ecography 33:46–50.Rosseel, Y. 2011. lavaan: Latent variable analysis. R packageversion 0.4–9. Available at: http://CRAN.R-project.org/package lavaan.Schwarz, P. A. et al. 2003. Factors controlling spatial variation oftree species abundance in a forested landscape. – Ecology 84:1862–1878.Scheiner, S. M. and Willig, M. R. 2005. Developing unified theoriesin ecology as exemplified with diversity gradients. – Am.Nat. 166: 458–469.Stone, R. 2006. A threatened nature reserve breaks down Asianborders. – Science 313: 1379–1380.Tuomisto, H. et al. 2003. Dispersal, environment, and floristicvariation of western Amazonian forests. – Science 299:241–244.Vivanco, L. and Austin, A. T. 2008. Tree species identity altersforest litter decomposition through long-term plant andsoil interactions in Patagonia, Argentina. – J. Ecol. 96:727–736.Wang, X. G. et al. 2008. Tree size distributions in an old-growthtemperate forest. – Okios 98: 674–686.Wu, Y. G. and Han, J. X. 1990. The habitat regionalization of Pinuskoraiensis forests and the zonal communities in northeastChina. China. – J. Nat. Res. 5: 335–342, in Chinese.Yang, H. et al. 1985. Distribution patterns of dominant tree specieson northern slope of Changbai Mountain. – For. Ecol. Res. 5:1–14, in Chinese.Yuan, Z. et al. 2011. Scale specific determinants of tree diversityin an old growth temperate forest in China. – Basic Appl. Ecol.12: 488–495.Zhang, J. et al. 2010. Spatial patterns and associations of six congenericspecies in an old-growth temperate forest. – Acta.Oecol. 36: 29–38.Zhou, Y. 2003. Measurement of LAI in Changbai mountains naturereserve and its result. – Res. Sci. 25: 38–42, in Chinese.Supplementary material (available as Appendix O19757 atwww.oikosoffice.lu.se/appendix). Appendix 1.831153

Journal of Vegetation Science 23 (2012) 271–279Seed rain dynamics reveals strong dispersal limitation,different reproductive strategies and responses toclimate in a temperate forest in northeast ChinaBuhang Li, Zhanqing Hao, Yue Bin, Jian Zhang & Miao WangKeywordsAutoregressive model; Changbai mountain;Climate; Mast; Seed dispersal; The timing ofreleasing seedsAbbreviationsNUM = number of seeds; WEI = weight ofseeds; T = Temperature; P = Precipitation.Received 6 April 2011Accepted 11 August 2011Co-ordinating Editor: Ingolf KühnHao, Z. (corresponding author, hzq@iae.ac.cn)Wang, M. (wangmiao@iae.ac.cn) &Li,B.(libuhang320@163.com): State Key Laboratoryof Forest and Soil Ecology, Institute of AppliedEcology, Chinese Academy of Sciences,Shenyang, 110164, ChinaLi, B.: Graduate University of ChineseAcademy of Sciences, Beijing, 100049, ChinaBin, Y. (byicymoon@163.com): Department ofEcology, School of Life Science/State KeyLaboratory of Biocontrol, Sun Yat-senUniversityZhang, J. (jzhang04@163.com): Department ofRenewable Resources, University of Alberta,751 General Service Building, Edmonton,Alberta, T6G 2H1, CanadaAbstractQuestions: Species co-existence may be achieved by limited dispersal due tothe reduced chance of inter-specific competition. But the timing of seed releasealso provides clues for species co-existence because asynchronous reproductioncan alleviate inter-specific competition for dispersal agents, while synchronousreproduction can facilitate overall seed dispersal. How strong is seed dispersallimitation, do co-existing species release seeds synchronously or asynchronously,and what is the relationship between seed production and main meteorologicalmeasures?Location: A 25-ha plot in a temperate forest, Changbai Mountain, NortheastChina.Methods: We calculated Jaccard coefficients between seed rain compositionand neighbouring adult tree composition, analysed long-term seed rain dynamicsfor both the whole community and different species, and regressed seed raindensity with meteorological measures using autoregressive models.Results: The Jaccard coefficient dropped sharply as neighbourhood radiusincreased to about 10 m, indicating severe dispersal limitation. Both synchronyand asynchrony in seed release were observed: most species released matureseeds synchronously from August to November, with small segregations in time;one species released seeds in Jun. Bimodal and unimodal seasonal dynamics ofseed rain were observed and some species released seeds beyond the main fruitingseasons. The seasonal dynamics of seed release might be driven by differentstrategies of seed dispersal. Seed rain density is significantly positively related totemperature and precipitation, with a 2-mo time lag.Conclusions: Both dispersal limitation and timing of seed release by co-existingspecies may contribute to maintenance of diversity of this forest, but variationsin temperature and precipitation considerably alter seed rain density.IntroductionSeed dispersal can have an impact on species composition(Matthiessen & Hillebrand 2006), forest diversity (Janzen1970; Harms et al. 2000) and dynamics of plant communities(Nathan & Ne’eman 2004; Swamy et al. 2011). Seeddispersal also determines the spatial ranges of populationregeneration (Nathan & Muller-Landau 2000; Levine &Murrell 2003). Previous studies of seed dispersal show thatseed density declines rapidly with distance from the maternaltree (Muller-Landau et al. 2008). Such a phenomenonis called limited dispersal, which might be related to maintenanceof diversity in species-rich plant communities(Dalling & Hubbell 2002; Wright 2002). This is because thefailure of seeds of superior competitors to arrive at suitablemicrosites provides an opportunity for less competitivespecies to take their place, thus slowing competitive exclusion(Seidler & Plotkin 2006). Knowledge of the degree ofdispersal limitation in a forest is valuable for understandingthe mechanisms of species co-existence.Not only limited dispersal, but also timing of seed sheddingmay contribute to species co-existence. In old-growthJournal of Vegetation ScienceDoi: 10.1111/j.1654-1103.2011.01344.x © 2011 International Association for Vegetation Science 27184

Seed rain dynamics in temperate forestB. Li et al.forests both climax species and pioneer species can co-exist(Wright et al. 2003; Hao et al. 2008) because pioneer speciesgrow rapidly and occupy gaps as soon as they appear(Wright et al. 2003). Similarly, the time of seedling emergenceoften determines subsequent plant performance andsuccess (Harper 1977), especially under competitive situations(Dyer et al. 2000). Moreover, timing of seed dispersalgreatly influences the time of seedling emergence,although in some cases dormancy occurs when environmentalconditions are unfavourable. If the inferior speciesgerminates first before the superior species reaches itsreproductive season, the inferior species will produce olderseedlings that are capable of competing with the youngerseedlings of the superior species, leading to species co-existencein a forest. The asynchrony in reproduction can alsoreduce inter-specific competition for dispersal agents andfurther increase species co-existence (Wheelwright 1985).On the other hand, co-existing species may also benefitfrom synchronous reproduction when the effectiveness ofseed dispersal agents varies seasonally or when large, synchronousfruit displays enhance dispersal (Poulin et al.1999). Analysing the seed rain dynamics separately perspecies can help to identify such relationships among species.However, previous studies have focused on seed rainof a single tree or stand and do not address this aspect ofstand development (Levine & Murrell 2003). Therefore,studies of large-scale, long-term, community-level seedrain dynamics have recently been undertaken (Wrightet al. 2005; Muller-Landau et al. 2008; Du et al. 2010).Mixed forest, dominated by broad-leaved species andKorean pine (Pinus koraiensis), is widespread in northeastChina and is well known for its high biodiversity, complexstand structure and unique species composition (Hao et al.2008). There have been some short-term seed rain compositionanalyses (Jin et al. 2006; Zhang et al. 2008), butthese have not addressed the larger-scale problems mentionedabove. Furthermore, long-term study of seed rain isessential, especially at this time when climate change hasbeen found to have fundamental impacts on forest diversityand dynamics (Liu et al. 2004; Boisvenue et al. 2006).In this paper, we aim to examine levels of dispersal limitation,how climate influences seed rain density andwhether species differ in their timing of seed release byanalysing species composition, temporal dynamics andspatial distribution of seed rain based on 4 yr of data onseed rain collected from 150 seed traps set up in a 25-haplot in broad-leaved–Korean pine mixed forest.MethodsStudy siteThe study site is located in the Changbai Mountain NatureReserve (a UNESCO World Heritage Site), established alongthe border between China and North Korea (127°42′–128°17′E, 41°43′–42°26′N). Our plot was established in 2004 ina stand of about 300 yr old of broad-leaved–Korean pinemixed forest in Changbai Mountain. The reserve coversabout 200 000 ha and elevation ranges from 740 to 2691 ma.s.l. The soil is classified as dark brown forest soil. Climateis characterized as moist temperate monsoon, with meanannual temperature of 3.3 °C( 16.5 °CinJanand20.5 °Cin Aug), and mean annual precipitation of 671.9 mm . yr 1 ,most of which occurs between Jun and Aug.Theplotcoversanareaof25ha(500m9 500 m), setup in the core zone of the reserve in order to avoid humandisturbance. The elevation of this plot ranges from 791.8 to809.5 m a.s.l. During a survey in 2004, 38902 living stems(1 cm DBH), belonging to 52 species, 32 genera and 18families were censored. The average stand density of livingtrees was 1556 trees ha 1 , and the average basal area of livingtrees was 43.2 m 2 ha 1 (Hao et al. 2007). Based onexisting documents, these species were classified into overstorey,mid-storey and understorey species (Liu 1955;Wang et al. 2008). Species were defined as rare if the numberof individuals in a single hectare was less than one,otherwise they were considered as common.Seed trapsWe set up 150 seed traps in our plot in Jun 2005(Fig. 1). A combination of typical grids and circular plotswas used for arrangement of seed traps so as to limit themaximum distance between trees and the nearest seedtrap. The traps were arranged so that seeds travelled amaximum distance of 31 m to arrive at the nearest seedtrap. This arrangement of seed traps helps to avoid collectionof seeds from only confined and aggregated areaswithin a small region, as found in some other forestdynamic plots, e.g. Barro Colorado Island (BCI) (Muller-Landau et al. 2008). Although the number of seed trapsis very limited for a background area of 25 ha, the densityof seed traps is 1.6 times of that used in BCI, where188 seed traps were arranged along small trails in a 50-ha forest plot. Each trap used in the present study was1 m above the ground, 0.75 9 0.75 m 2 square, made offine, flexible mesh of 1 mm and supported on four PVCtubes. The area of each trap was 0.56 m 2 . Here wedefined all dispersal units as seeds. Seeds were collectedtwice a month from May to Dec and once a month fromJan to Apr, starting in May 2006. The number of seeds(NUM) and weight of seed production (WEI) were usedin autoregressive models to study the relationshipbetween seed production and the methodological factors,precipitation (P) or temperature (T). Dispersal modeswere assigned to the investigated species based on publishedreports (e.g. Liu 1955). Seeds were judged to beJournal of Vegetation Science272 Doi: 10.1111/j.1654-1103.2011.01344.x © 2011 International Association for Vegetation Science85

B. Li et al. Seed rain dynamics in temperate forestexclude non-significant variables. All statistical tests andmodelling were done using R 2.9.2 (R Development CoreTeam, 2010, Vienna, Austria).Seed–adult tree spatial analysisFig. 1. Arrangement of the 150 seed traps in the 25-ha plot. The pointsrepresent for seed traps.mature if they were plump, and otherwise considered asimmature.Meteorological factor monitoringThe meteorological data were taken from surface climatedata collected from 2006 and 2009 at the MeteorologicalObservatory of Changbai Mountain Forest Ecosystem Station,Chinese Academy of Sciences (42.24°N, 128. 6°E,738 m a.s.l.). The topography is relatively flat within severaldozen kilometers around the station. P is recordedonce a day, T is recorded at 08:00, 14:00 and 20:00 everyday. The monthly averages of P and T were used in subsequentanalyses of the relationship between meteorologicalmeasures and seed production.To control the autocorrelation of seed rain density, weused autoregressive models to study the relation betweenseed rain density (NUM and WEI) and meteorologicalmeasures (P and T). Because P and T were highly correlated(r = 0.746,P = 9.315e-09, df = 41), they were neverused together in the same autoregressive model. Becausewe observed clear time lags in the response of seed productionto P and T, we assessed different monthly time lags.Our autoregressive models were as follows:S i ¼ c þ b X S i¼k þ u X M i k ð1Þwhere S i is the seed rain density at time i; b and u are vectorsof coefficients; k is the time lag; and M is the meteorologicalmeasure (P and T). Two measures of seed raindensity were used: NUM and WEI. k was set to 1, 2 or 3. Astep-wise search procedure was run on the models so as toThe Jaccard similarity coefficient between seed compositionand the composition of adult trees in the neighbourhoodwas calculated separately for all species combined,species dispersed by gravity, species dispersed by wind,overstorey species, mid-storey species and understoreyspecies within radii of 1–20 m in steps of 1 m. The Jaccardsimilarity index was not calculated for gravity- -and-animaldispersed species, because animal behaviour is difficultto predict and we do not know to what degree they areinvolved in seed dispersal. The average Jaccard similaritycoefficients for 150 sets of seed trap data and their neighbourhoodwere separately plotted against radii for everyspecies set. When calculating the similarity index withneighbourhood radii larger than 10 m, 24 seed traps wereexcluded because their neighbourhoods were partiallyoutside the plot. Pearson’s correlation coefficient was usedto test the correlation between radii and Jaccard similarityindex.ResultsSpecies compositionIn the 150 seed traps, 308276 seeds/fruits were collected,belonging to 25 species, 16 genera and 11 families, whichonly accounted for 48.1% of the total species that werecensored in 2004 (Table 1). The number of seeds collecteddiffered so strikingly among species, with 79.6% of thetotal seeds collected being from Tilia amurensis and Fraxinusmandshurica. The collected seeds were highly variable intaxonomic identity among years (Table 1). Not only seedsof common species were collected, but also seeds from tenrare species (Table 1). The collected species were alsodiverse in dispersal mode and growth form: a much largerproportion of wind-dispersed and overstorey and midstoreyspecies were collected in comparison with those ofother groups (Table 2).Seed rain dynamics and the correlation withmeteorological dataThe inter-annual variation in seed collection was strongwithin the forest (Table 1; Fig. 2), yet seasonal dynamicswas similar among years (Fig. 2). Relatively heavy seedrain began in Jun and ended in November, during whichtime there were two main peaks of NUM, the first inAug and the second in Oct (Fig. 2a). When seed rain wasJournal of Vegetation ScienceDoi: 10.1111/j.1654-1103.2011.01344.x © 2011 International Association for Vegetation Science 27386

Seed rain dynamics in temperate forestB. Li et al.Table 1. Species composition of the seed rain collected in an old-growth temperate forest during 4 yr (May 2006–Apr 2010). Family, abundance class andnumber of seeds collected for each species in each year is reported.Species Family Abundance class Number of seeds1st year05/06-05/072nd year05/07-05/083rd year05/08-05/094th year05/09-04/10 TotalAcer barbinerve Aceraceae Common 213 93 31 2116 2453Acer ginnala Aceraceae Rare 12 0 8 11 31Acer mandshuricum Aceraceae Common 41 12 131 197 381Acer mono Aceraceae Common 1910 1791 144 7137 10982Acer pseudo-sieboldianum Aceraceae Common 1341 1480 509 8149 11479Acer tegmentosum Aceraceae Common 1 4 1 316 322Acer triflorum Aceraceae Common 20 28 11 139 198Betula platyphylla Betulaceae Rare 3054 0 2100 1800 6954Corylus mandshurica Betulaceae Common 2 0 59 10 71Quercus mongolica Fagaceae Common 1065 399 1929 601 3994Maackia amurensis Leguminosae Common 78 18 521 12 629Fraxinus mandshurica Oleaceae Common 9176 18476 263 32564 60479Syringa reticulata Oleaceae Common 0 0 2 19 21Abies holophylla Pinaceae Rare 0 1 0 0 1Abies nephrolepis Pinaceae Rare 0 1 0 0 1Pinus koraiensis Pinaceae Common 39 9 608 151 807Crataegus maximowiczii Rosaceae Rare 0 0 1 0 1Malus baccata Rosaceae Rare 68 3 0 35 106Prunus padus Rosaceae Rare 0 0 0 2 2Phellodendron amurense Rutaceae Rare 0 7 0 0 7Populus koreana Salicaceae Rare 0 0 140 6600 6740Populus ussuriensis Salicaceae Rare 0 0 0 10 10Tilia amurensis Tiliaceae Common 96042 120 51145 37441 184748Tilia mandshurica Tiliaceae Common 239 0 0 117 356Ulmus japonica Ulmaceae Common 1284 79 16140 0 17503Total 114585 22521 73743 97427 308276Table 2. Number of species collected in the seed rain and number of allspecies with DBH 1 cm in the plot, grouped by dispersal mode andgrowth form.Grouping Type No. of speciesin seed rainDispersal mode Gravity & animal 6 20Wind 16 21Gravity 3 11Growth form Overstorey 9 9Mid-storey 12 19Understorey 4 24No. of specieswith DBH 1 cmrepresented as WEI, only one peak could be observed, atabout the time of the second peak of NUM (Fig. 2b).At species level, we investigated the seed rain dynamicsof only eight species because only a small number of seedswere collected from the remaining 17 species. Two maintypes of seasonal dynamics were observed. The first wascharacterized by two peaks (bimodal), which were madeup of immature and mature seeds, respectively; there werefour species belonging to this type, e.g. Acer mono (Fig. 3a,App. S1). The second had a single narrow peak (unimodal);there were four species in this group, e.g. Ulmus japonica(Fig. 3b, App. S1). Besides these two main types, fivespecies released a few mature seeds beyond the main fruitingseason; e.g. A. barbinerve (Fig. 3c App. S1). This featurewas not mutually exclusive with the above unimodal andbimodal types. Among the species, we found both synchronyand asynchrony in seed release; three species hadNUM peaks synchronously around October; two speciespeaked around September; another two had peaks inAugust; and one reached a peak in Jun (U. japonica;Fig. 3b, d). The widths of the peaks differed among species(Fig. 3d); the peaks of two species were wide, from Aug toOct (Fig. 3d), while the other six species had narrow peaks,only lasting 1 mo (Fig. 3d).The autoregressive models yielded very good fits to theobserved data (Table 3; Fig. 2). Both NUM and WEIshowed significant relationships with T and P. The autocorrelationof NUM was constantly significant, whereasthat of WEI was rarely significant. The P i-2 was the bestpredictive meteorological measure for both NUM and WEI(Table 3).Journal of Vegetation Science274 Doi: 10.1111/j.1654-1103.2011.01344.x © 2011 International Association for Vegetation Science87

B. Li et al. Seed rain dynamics in temperate forest5000040000(a)NUM T, k = 1 P, k = 1 T, k = 2 P, k = 2 T, k = 3 P, k = 3Number300002000010000025002000(b)WEI T, k = 1 P, k = 1 T, k = 2 P, k = 2 T, k = 3 P, k = 3Weight (g)1500100020062007200820095000JulSepNovJanMarMayJulSepNovJanMarMayJulSepNovJanMarMayJulSepNovFig. 2. Observed and predicted dynamics of seed rain density measured as (a) number of seeds and (b) weight of seeds. The predicted seed rain dynamicswere obtained from the autoregressive models using eq. 1. The abbreviations in the figure are the components in eq. 1. NUM, number of seeds; WEI,weight of seeds; T, temperature; P, precipitation; k represents monthly lag used in the autoregressive models.TimeFig. 3. Species-specific seed rain dynamics. Three types of seed raindynamics (a, b, c) and the synchrony and asynchrony of seed dispersalamong eight species (d). In panel d, different symbols and colours denotedifferent species.Spatial distributionBoth the abundance and species richness of seeds capturedin specific seed traps differed substantially, indicating highheterogeneity in seed dispersal across space. The number ofspecies captured in each seed trap is possibly related to theadult trees around the corresponding seed trap because theJaccard coefficient between seed composition and treecomposition within several meters was very close to one(Fig. 4). The Jaccard coefficients of all types of species classifiedby either dispersal mode or the vertical layer to whichthey belong became steady 10 m from the seed traps(Fig. 4). When the species were classified by dispersalmode, the Jaccard coefficient for wind-dispersed speciesdropped from 0.991 to 0.675 as the distance increased from0 to 10 m, while that for gravity-dispersed species was from0.998 to 0.338, indicating that a larger proportion of winddispersedspecies can disperse seeds to the seed traps within10 m than gravity-dispersed species (Fig. 4). The heightand crown of a species is also associated with the dispersalefficiency. The Jaccard coefficient of the overstorey specieswas largest among species of different growth forms, followedby that of the mid-storey, and then shrub specieswithin 10 m (Fig. 4). However, between 10 and 20 m, themid-storey species had the largest Jaccard coefficient, followedby overstorey species and shrub species (Fig. 4). Astrong correlation (r < 0.8, P < 10 6 ) between the meanvalue of the Jaccard similarity index and the neighborhoodradius was observed for all groups of species.DiscussionSpecies composition and dispersal limitationLimited seed dispersal has been cited as a mechanism forspecies co-existence in species-rich communities, byJournal of Vegetation ScienceDoi: 10.1111/j.1654-1103.2011.01344.x © 2011 International Association for Vegetation Science 27588

Seed rain dynamics in temperate forestB. Li et al.Table 3. The response variables, predictive variables retained after the step-wise procedure and other related information for the autoregressive models.S in eq 1: seedrain densityk ineq(1)M in eq 1:meteorological measureVariables retainedafter step-wise procedureR 2PNUM 1 T NUM i 1 ;T i 1 0.417 1.01E-05NUM 2 T NUM i 1 ;T i 1 0.416 1.39E-05NUM 3 T NUM i 1 ;T i 2 ,NUM i 3 0.491 4.61E-06NUM 1 P NUM i 1 ;P i 1 0.365 5.42E-05NUM 2 P NUM i 1 ;P i 2 0.534 1.87E-07NUM 3 P NUM i 1 ;P i 2 0.534 2.75E-07WEI 1 T T i 1 0.281 0.000613WEI 2 T T i 2 0.332 4.84E-05WEI 3 T T i 2 0.323 7.73E-05WEI 1 P WEI i 1 ;P i 1 0.238 0.001867WEI 2 P P i 2 0.435 1.64E-06WEI 3 P P i 2 0.433 2.40E-06The abbreviations in the table are components in eq 1. NUM, number of seeds; WEI, weight of seeds; T, temperature; P, precipitation; k represents monthlylag used in the autoregressive models.Fig. 4. Similarity in species composition between species in the seed rainand species occurring within different radii (1–20 m) from each seed trap.Both patterns for all species and species grouped according to dispersalmode and growth form are shown. Similarity was measured with theJaccard coefficient.slowing down the process of competitive exclusion (Seidler& Plotkin 2006). Our research implies strong dispersal limitationin the studied forest. In our study, no seed was collectedfor 52% of the species. A similar phenomenon wasobserved in the BCI 50-ha plot, where no single seed wastrapped for 10% of the species during 10 yr (Hubbell et al.1999). Hubbell et al. (1999) also showed that most seedtraps captured about 50 (16.7%) species and that morethan 50% of species were distributed among no more thansix seed traps (Hubbell et al. 1999). We also observed similaraggregated distribution of species among seed traps.Besides dispersal limitation, the relatively short time ofseed collection may also contribute to the small number ofspecies collected because seed production is influenced bymasting years (Kelly & Sork 2002). Yet, dispersal limitationappears to be stronger in our plot than in BCI, given thatthe seed traps in our plot are so arranged that the largestdistance from a seed trap for all trees in the plot is only31 m (Fig. 1).It is widely recognized that seeds from forests that aredispersed by wind can disperse over a longer distance thanthose dispersed in other ways (Greene & Johnson 1995;Fragoso 1997; Qian 2009). In the study, we foundevidence for longer dispersal distance of wind-dispersedspecies: we collected a larger proportion (16/21) of winddispersedspecies than those dispersed by gravity (3/11)or gravity and animals combined (6/20) (Table 2); theJaccard coefficient of species dispersed by wind was consistentlylarger than for species dispersed by gravity, indicatingthat a larger proportion of wind-dispersed seeds in theneighbourhood arrive at seed traps than those dispersed bygravity. However, because our seed traps were erected 1 mabove the surface, the chances of collecting wind-dispersedand animal-dispersed seeds may differ and receives furtherinvestigation.In addition, our study indicated that dispersal distance isalso related to growth form of the species. In primary seeddispersal stages, tree height and large crown size facilitateseed dispersal. With the inverse modelling method, Muller-Landauet al. (2008) found that tree height is positivelycorrelated with dispersal distance; Tackenberg et al. (2003)further confirmed the effect of tree height on dispersal distance.Shrubs are much shorter and have smaller crownareas. This may explain why they have the smallest Jaccardcoefficients among growth forms. However, furtherwork is necessary because the seed trap arrangement couldbe even finer in this plot, although this arrangement isalready superior to similar research in other forest dynamicsplots (e.g. BCI).Dispersal distance can have an impact on subsequentseedling recruitment processes. For example, it is very difficultfor seeds near parent trees to thrive because of possiblespecies-specific pests and pathogens (Janzen 1970;Connell 1971) and other forms of inhibition. Therefore,Journal of Vegetation Science276 Doi: 10.1111/j.1654-1103.2011.01344.x © 2011 International Association for Vegetation Science89

B. Li et al. Seed rain dynamics in temperate forestabundant seed does not guarantee population growth. Acombination of knowledge on seedling dynamics and seedrain dynamics is important to understand the dynamics ofthis broad-leaved–Korean pine mixed forest.Seasonal dynamics and strategies of dispersing seedsStrong seasonal dynamics was observed in the plot, similarto that observed in other forests (Urbanska & Fattorini2000; Du et al. 2010). Among species, both synchrony andasynchrony in the timing of seed release were observed.U. japonica dispersed seeds in Jun, in contrast to the otherspecies which released seeds from Aug to Nov. The relativeearly seed dispersal of U. japonica may lead to early seedgermination and subsequent improved plant performanceand establishment success. However, the difference intiming is only about 1 mo; therefore synchrony in seeddispersal was the most significant phenomenon found inour study. A synchronous trend has also been observed inother phenologies, such as flowering (Sun et al. 2007) andfruiting (Poulin et al. 1999). The temporally aggregatedreproduction may be favoured by selection when the effectivenessof seed dispersal agents varies seasonally or whenlarge, synchronous fruit displays enhance seed dispersal(Poulin et al. 1999).Not only comparisons of seed rain dynamics amongspecies, but also species seed rain dynamics alone provideinsight into how different species adapt to the environment.Species seed rain dynamics can be classified into twomain groups, possibly suggesting different strategies thatare adjusted to the environment. The first group has a unimodalcurve in seed number along the time axis, becausethese species release nearly all their seeds in a relativelyshort time. The time of seed shedding being appropriate,this strategy can maximize the chance of seeds finding suitableplaces for germination. U. japonica and P. koraiensis aregood examples of such species, but having different dispersalmodes, they demonstrate two distinctly differentways of maximizing establishment. U. japonica, whichisdispersed by wind, sheds seeds from late May to early Jun(Fig. 3b) when broad-leaved trees are still leafless. Thusseeds of U. japonica can disperse over long distances due tostrong winds and relatively few obstructions (here, mainlythe branches in space) in the dispersal path. The case ofP. koraiensis reflects a phenomenon called predator satiation,in which large intermittent seed production canreduce losses to predators (Janzen 1971). This predictionhas been tested with a natural experiment in a Malaysianrain forest, where the percentage of seeds consumed byboth insects and primates dropped with the magnitude ofthe general flowering event (Sun et al. 2007). By sheddingabundant seeds in a relatively short time, the suddenincrease in available food for animals exceeds the amountof food they can consume temporally and causes animals tomove and store seeds (Janzen 1969; Smith 1970; Howe &Smallwood 1982). Some of these seeds may germinate andsurvive much better because they escape from the Janzen-Connell effect (Janzen 1970; Connell 1971). This methodof dispersal is far better than shedding seeds gradually andevenly across the year and providing a continuous foodsource for animals if there are enough animals for dispersal.Bimodal dispersal curve of seed number along the timeaxis is another type of seasonal dynamics. Species in thisgroup disperse seeds in both the rainy season (Aug) andthe dry season (Oct), leading to two peaks on the seasonaldynamics curve. During the rainy season, the dispersedseeds are mainly immature and often damaged, whileseeds are mainly fully mature in the second (dry) period.Fruiting consumes resources and energy, and if resourceslimit reproduction, smaller, immature seeds are likely tofall during heavy rain and strong winds in the rainy season.Thus, seeds left on the trees may gain more resourcesand energy, which enhance their future rates of germinationand survival. In the dry season, the dispersed seeds donot germinate immediately due to low temperatures andlack of moisture (Du et al. 2010), but remain dormantuntil the following spring, when seedlings have a relativelylonger time for growth before the hard winter due to thetime lag between seed fall and germination. If most matureseeds were dispersed in the rainy season (late summer)and germinated immediately, the seedlings would be relativelyweak when winter arrived, which would reduceseedling establishment.Some seed dispersal beyond the main dispersal seasonsis found in species with both unimodal and bimodalseasonal dynamics of seed rain, e.g. A. barbinerve andA. pseudo-sieboldianum. Such curves may indicate the existenceof canopy seed banks. In unfavourable and unpredictableenvironments, shedding seeds beyond the maindispersal seasons helps to reduce the risk of becominglocally extinct and maintains the population (Günster1994). This strategy is common in fire-prone, nutrientpoorand seasonally dry woody vegetation (Lamont &Enright 2000), e.g. Agriophyllum squarrosum in sand dunes(Liu et al. 2006) and some Mediterannean pines in postfireconditions (Tapias et al. 2001). Changbai Mountaindoes not experience such conditions, so the canopy seedbanks are less obvious.Climate and seed productionOur autoregressive models revealed that both T and P arepositively associated with seed production and that bothfactors limit seed production in temperate forests. However,it did not confirm that global warming leads toincrease in seed production, because 22 yr of long-term,Journal of Vegetation ScienceDoi: 10.1111/j.1654-1103.2011.01344.x © 2011 International Association for Vegetation Science 27790

Seed rain dynamics in temperate forestB. Li et al.meteorological data revealed that annual average T isincreasing and is accompanied by strong fluctuation in P inthe Changbai Mountain region (Zhang et al. 2005). In yearswith high T and low P, seed production may decline. Furthermore,the analysis of seed production and climate variablespresented should be considered as preliminary. Cain& Shelton (2000) showed that the climate variables havedifferent effects among flower-development, pollinationand fertilization stages. A more detailed analysis may providemore information on how plants respond to climate.Inter-annual variations in seed rainInter-annual variations in seed rain were also significant inour studied plot (Table 1). This phenomenon was alsoobserved in research from Salonen (1987) and Koeniget al. (1994). Masting at 2-yr intervals in a major cause ofinter-annual differences in seed rain (Kelly & Sork 2002),but may not be significant in this study. In our research,for all species combined, 22521 seeds/fruits were collectedin the second year, less than 20% of the correspondingnumber for the first year. However, there were no greatdifferences between the figures in the first and fourth yearor the third and fourth year (Table 1; Fig. 2); for specificspecies, this trend was found e.g. in A. ginnal and A. pseudosieboldianum(Table 1).In general, we found strong dispersal limitation, a significantresponse of seed production to climate, differences intiming and both synchrony and asynchrony in seed dispersalamong species in our studied plot. Because of strongvariation in seed production among years and ongoing climatechange, long-term monitoring of seed rain is essentialto obtain a comprehensive understanding of the dynamicsof seed rain in this forest.AcknowledgementsThis study was supported by the Knowledge InnovationProgram of the Chinese Academy of Sciences (KZCX2-EW-401 and KSCX2-EW-Z-5), the National NaturalScience Foundation of China (31061160188,30870400,40971286) and the National Key Technologies R&DProgram of China (2008BAC39B02).ReferencesBoisvenue, C. L., Running, S. W. & Steven, W. 2006. Impacts ofclimate change on natural forest productivity – evidencesince the middle of the 20th century. Global Change Biology12: 862–882.Cain, M. D. & Shelton, M. G. 2000. Revisiting the relationshipbetween common weather variables and loblolly–shortleafpine seed crops in natural stands. New Forests 19: 187–204.Connell, J.H. 1971. On the role of natural enemies in preventingcompetitive exclusion in some marine animalsand in rain forest trees. In: den Boer, P.J. & Gradwell,G.R. (eds.) Dynamics of populations, pp. 298–312. Centerfor Agricultural Publication and Documentation, Wageningen,NL.Dalling, J.W. & Hubbell, S.P. 2002. Seed size, growth rate andgap microsite conditions as determinants of recruitment successfor pioneer species. Journal of Ecology 90: 557–568.Du, Y., Mi, X., Liu, X., Chen, L. & Ma, K. 2010. Seed dispersalphenology and dispersal syndromes in a subtropical forestunderstory in East China. Forest Ecology and Management 258:1147–1152.Dyer, A., Fenech, A. & Rice, K. J. 2000. Accelerated seedlingemergence in interspecific competitive neighbourhoods.Ecology Letters 3: 523–529.Fragoso, J.M.V. 1997. Tapir-generated seed shadows: scaledependentpatchiness in the Amazon rain forest. Journal ofEcology 85: 519–529.Greene, D.F. & Johnson, EA. 1995. Long-distance wind dispersalof tree seeds. Canadian Journal of Botany 73: 1036–1045.Günster, A. 1994. Seed bank dynamics–longevity, viability andpredation of seed of serotinous plants in the central NamibDesert. Journal of Arid Environments 28: 195–205.Hao, Z., Zhang, J., Song, B., Ye, J. & Li, B. 2007. Vertical structureand spatial associations of dominant tree species in anold-growth temperate forest. Forest Ecology and Management252: 1–11.Hao, Z.-Q., Li, B.-H., Zhang, J., Wang, X.-G., Ye, J. & Yao, X.-L.2008. Broad-leaved Korean pine (Pinus koraiensis) mixedforest plot in Changbaishan (CBS) of China: communitycomposition and structure. Journal of Plant Ecology 32:238–250.Harms, K.E., Wright, S.J., Calderon, O., Hernandez, A. & Herre,E.A. 2000. Pervasive density-dependent recruitmentenhances seedling diversity in a tropical forest. Nature 404:493–495.Harper, J.L. 1977. Population Biology of Plants. Academic Press,London, UK.Howe, H.F. & Smallwood, J. 1982. Ecology of seed dispersal.Annual Review of Ecology and Systematics 13: 201–228.Hubbell, S.P., Foster, R.B., O’Brien, S.T., Harm, K.E., Condit, R.,Wechsler, B., Wright, S.J. & Loo de Lao, S. 1999. Light-gapdisturbances, recruitment limitation, and tree diversity in aneotropical forest. Science 283: 554–557.Janzen, D.H. 1969. Seed-eaters vs seed size, number, toxicityand dispersal. Evolution 23: 1–27.Janzen, D.H. 1970. Herbivores and the number of trees speciesin tropical forest. American Naturalist 104: 501–528.Janzen, D.H. 1971. Seed predation by animals. Annual Review ofEcology and Systematics 2: 465–92.Jin, G., Xie, X., Tian, Y. & Kim, J.H. 2006. The pattern of seedrain in the broadleaved–Korean pine mixed forest of Xiaoxing’anMountains, China. Journal of Korean Forest Society 95:621–627.Journal of Vegetation Science278 Doi: 10.1111/j.1654-1103.2011.01344.x © 2011 International Association for Vegetation Science91

B. Li et al. Seed rain dynamics in temperate forestKelly, D. & Sork, V.L. 2002. Mast seeding in perennial plants:why, how, where? Annual Review of Ecology and Systematics33: 427–47.Koenig, W.D., Mumme, R.L., Carmen, W.J. & Stanback, M.T.1994. Acorn production by oaks in central coastal California:variation within and among years. Ecology 75: 99–109.Lamont, B. B. & Enright, N. J. 2000. Adaptive advantages ofaerial seed banks. Plant Species Biology 15: 157–166.Levine, J.M. & Murrell, D.J. 2003. The community-level consequencesof seed dispersal patterns. Annual Review of EcologyEvolution and Systematics 34: 549–574.Liu, S. 1955. Woody plant photography records in Northeast China.Science Press, Beijing, CN.Liu, C., Westman, C. J., Berg, B., Kutsch, W., Wang, G.Z., Man,R. & Ilvesniemi, H. 2004. Variation in litterfall–climate relationshipsbetween coniferous and broadleaf forests in Eurasia.Global Ecology and Biogeography 13: 105–114.Liu, Z., Yan, Q., Baskin, C.C. & Ma, J. 2006. Burial of canopystoredseeds in the annual psammophyte Agriophyllumsquarrosum Moq. (Chenopodiaceae) and its ecological significance.Plant and Soil 288: 71–80.Matthiessen, B. & Hillebrand, H. 2006. Dispersal frequencyaffects local biomass production by controlling local diversity.Ecology Letters 9: 652–662.Muller-Landau, H.C., Wright, S.J., Caldero′n, O., Condit, R. &Hubbell, S.P. 2008. Interspecific variation in primary seeddispersal in a tropical forest. Journal of Ecology 96: 653–667.Nathan, R. & Muller-Landau, H.C. 2000. Spatial patterns of seeddispersal, their determinants and consequences for recruitment.Trends in Ecology and Evolution 15: 278–285.Nathan, R. & Ne’eman, G. 2004. Spatiotemporal dynamics ofrecruitment in Aleppo pine (Pinus halepensis Miller). PlantEcology 171: 123–137.Poulin, R., Wright, S. J., Lefebvre, G.E. & Calderón, O. 1999.Interspecific synchrony and asynchrony in the fruitingphenologies of congeneric bird-dispersed plants in Panama.Journal of Tropical Ecology 15: 213–227.Qian, H. 2009. Beta diversity in relation to dispersal ability forvascular plants in North America. Global Ecology and Biogeography18: 327–332.Salonen, V. 1987. Relationship between the seed rain and theestablishment of vegetation in two areas abandoned afterpeat harvesting. Holarctic Ecology 10: 171–174.Seidler, T.G. & Plotkin, J.B. 2006. Seed dispersal and spatialpattern in tropical trees. PLoS Biology 4: e344.Smith, C.C. 1970. The coevolution of pine squirrels (Tamiasciurus)and conifers. Ecological Monographs 40: 349–371.Sun, I.-F., Chen, Y.-Y., Hubble, P., Wright, S. J. & Noor, N.S.M.2007. Seed predation during general flowering events ofvarying magnitude in a Malaysian rain forest. Journal ofEcology 95: 818–827.Swamy, V., Terborgh, J., Dexter, K.G., Best, B.D., Alvarez, P. &Cornejo, F. 2011. Are all seeds equal? Spatially explicitcomparisons of seed fall and sapling recruitment in a tropicalforest Ecology Letters 14: 195–201.Tackenberg, O., Poschlod, P. & Bonn, S. 2003. Assessment ofwind dispersal potential in plant species. Ecological Monographs73: 191–205.Tapias, R., Gil, L., Fuentes-Utrilla, P. & Pardos, J. A. 2001. Canopyseed banks in Mediterranean pines of south-easternSpain: a comparison between Pinus halepensis Mill., P. pinasterAit., P. nigra Arn. and P. pinea L. Journal of Ecology 89: 629–638.Urbanska, K.W. & Fattorini, M. 2000. Seed rain in high altituderestoration plots in Switzerland. Restoration Ecology 8: 74–79.Wang, X., Hao, Z., Zhang, J., Lian, J., Li, B., Ye, J. & Yao, X.2008. Tree size distributions in an old-growth temperateforest. Oikos 118: 25–36.Wheelwright, N. T. 1985. Competition for dispersers, and thetiming of flowering and fruiting in a guild of tropical trees.Oikos 44: 465–477.Wright, S.J. 2002. Plant diversity in tropical forests: a review ofmechanisms of species coexistence. Oecologia 130: 1–14.Wright, S.J., Muller-Landau, H.C., Condit, R. & Hubbell, S.P.2003. Gap-dependent recruitment, realized vital rates, andsize distributions of tropical trees. Ecology 84: 3174–3185.Wright, S.J., Muller-Landau, H.C., Calderon, O. & Hernandéz,A. 2005. Annual and spatial variation in seedfall and seedlingrecruitment in a neotropical forest. Ecology 86: 848–860.Zhang, M., Guan, D., Han, S., Wu, J., Zhang, J., Jin, M., Xu, H.,He, X. & Dai, G. 2005. Climate dynamics of broadleaved–Korean pine forest in Changbai Mountain during the last 22years. Chinese Journal of Ecology 24: 1007–1012.Zhang, J., Hao, Z., Li, B., Ye, J., Wang, X. & Yao, X. 2008. Compositionand seasonal dynamics of seed rain in broad-leaved–Korean pine (Pinus koraiensis) mixed forest in ChangbaiMountain, China. Acta Ecologica Sinica 28: 2445–2454.Supporting InformationAdditional Supporting Information may be found in theonline version of this article:Appendix S1. The Species-specific seed raindynamics of all eight species.Please note: Wiley-Blackwell are not responsible forthe content or functionality of any supporting materialssupplied by the authors. Any queries (other than missingmaterial) should be directed to the corresponding authorfor the article.Journal of Vegetation ScienceDoi: 10.1111/j.1654-1103.2011.01344.x © 2011 International Association for Vegetation Science 27992

Journal of Ecology 2011, 99, 1382–1393doi: 10.1111/j.1365-2745.2011.01857.xSpatial patterns of tree species richness in twotemperate forestsXugao Wang 1 , Thorsten Wiegand 2 , Amy Wolf 3 , Robert Howe 3 , Stuart J. Davies 4 andZhanqing Hao 1 *1 State Key Laboratory of Forest and Soil Ecology, Institute of Applied Ecology, Chinese Academy of Sciences,Shenyang 110164, China; 2 Department of Ecological Modelling, UFZ Helmholtz Centre for EnvironmentalResearch-UFZ, D-04301 Leipzig, Germany; 3 Department of Natural and Applied Sciences, University of Wisconsin-Green Bay, Green Bay, WI 54311, USA; and 4 Center for Tropical Forest Science, Arnold Arboretum, HarvardUniversity and Smithsonian Tropical Research Institute, Boston, MA 02131, USASummary1. The relative contribution of external vs. internal clustering mechanisms for determining communitystructure and its manifestations has been the subject of a continuous debate, but few attemptshave been made to examine their single and joint effects in a compound process model.2. In this study, we tested four apriorihypotheses on the relative importance of habitat heterogeneity(topography and soil) and internal clustering mechanisms such as dispersal limitation on the species–arearelationship (SAR) in two fully mapped 25-ha plots of temperate forests in theChangbaishan (CBS) Nature Reserve, China, and the Chequamegon-Nicolet National Forest inWisconsin, USA.3. We used the distance decay curve to test the generality of the results obtained for the SAR. Tofind out if the relative importance of internal and external clustering mechanisms changed with lifestage, we conducted separate analyses for small, large and all trees.4. Model selection favoured the most complex hypothesis that assumed an influence of both habitatheterogeneity and internal clustering on SAR and the distance decay curve. For the CBS plot, whichshows weak topographical structuring, models were consistent with data only if soil factors wereincluded into assessment of habitat heterogeneity. At the Wabikon plot, we could not test soil variables,but inclusion of topographical variables substantially improved the fit of the distance decaycurve.5. In general, the results of the SAR agreed with those of the distance decay curve, but the latterwas sensitive to positive habitat-mediated species associations. The SAR, but not distance decay,distinguished among competing hypotheses for the community of large trees at CBS, where speciesexhibited only weak clustering.6. Contrary to our expectations, we did not find substantial differences in the relative importanceof internal and external clustering mechanisms with life stage.7. Synthesis. Our analysis of spatial community structure for two relatively diverse temperate forestsrevealed that the factors governing spatial community structure may not substantially differfrom those in highly diverse tropical forests. This result adds to our understanding of the ecologicalprocesses underlying the spatial diversity structure in natural forest communities.Key-words: aggregation, Changbaishan, determinants of plant community diversity andstructure, habitat heterogeneity, Possion processes, temperate forest, Thomas processes,WabikonIntroduction*Correspondence author. E-mail: hzq@iae.ac.cnThe increase in number of species (species richness) withincreasing sampling area is one of the most important attributesof biological communities (Holt et al. 1999; He & Legendre2002). This pattern, called the species–area relationship(SAR), quantifies basic aspects of biodiversity in a simple way,allowing comparisons among different study areas and ecologicalsystems. Herein, we focus on tree communities that arecompletely mapped within similar-sized (25-ha) study areas.Ó 2011 The Authors. Journal of Ecology Ó 2011 British Ecological Society93

Detecting underlying mechanisms controlling species–area relationships 1383In this case, one can calculate ‘local’ species-area relationshipsin which the area of sampling plots (A) within the 25 ha is consecutivelyincreased (Connor & McCoy 1979). Although spatialpatterns of species richness vary widely among naturalcommunities, they show basic similarities that suggest generalunderlying mechanisms (He & Legendre 2002). Indeed, thelocal SAR has played an important role in recent debatesabout whether ecological communities are dispersal assembledor niche assembled, (Hubbell 1997, 2001) because it can beused to test neutral models (McGill, Maurer & Weiser 2006).One major factor that influences the shape of the localSAR is spatial aggregation. If a species is more aggregatedat a given spatial scale, the probability of presence in a randomlyselected area (corresponding to this scale) becomessmaller. Consequently, the SAR values for a given area ofthe sampling plots will decrease if more species are aggregatedat this scale (Plotkin et al. 2000; He & Legendre 2002;Tjørve et al. 2008). However, aggregated distribution patternsin species may be broadly attributed to two major, yetcontrasting, factors: (i) external effects of the environment,such as habitat heterogeneity and (ii) internal processes ofpopulation and community dynamics. One of the mostprominent examples of an internal clustering mechanism isdispersal limitation (Hubbell 2001) that builds one cornerstoneof neutral theory (Hubbell 2001), but many other factorssuch as non-random seed deposition (Howe 1989),facilitation (Kikvidze et al. 2005), succession (Felinks & Wiegand2007) and gap dynamics (Nagel, Svoboda & Diaci2006) can contribute to clustered patterns in homogeneousenvironments. However, the relative contributions of externalvs. internal clustering mechanisms for generating patternsof species richness in real communities are difficult to quantify(Wiegand, Gunatilleke & Gunatilleke 2007a; Wang et al.2010a). This issue has been the subject of a continuousdebate around the question of whether ecological communitiesare dispersal assembled or niche assembled (Hubbell2001; McGill, Maurer & Weiser 2006), but researchers arenow generally convinced that these assembly mechanismsare mutually complementary rather than mutually exclusive(He & Legendre 2002; Shen et al. 2009). However, only afew investigations have attempted to examine the single andjoint effects of habitat heterogeneity and dispersal limitationin a compound process model (Shen et al. 2009). Newinsight can be expected when comparing results of such analysesamong plant communities with contrasting characteristics(e.g. tropical vs. temperate forest) (Wang et al. 2010a,b).The theory of spatial point processes (Møller & Waagepetersen2003; Illian et al. 2008; Waagepetersen & Guan 2009)offers an opportunity to test if an observed local SAR is predominantlyshaped by habitat heterogeneity and ⁄ or by internalclustering. Habitat models are fitted to the distributiondata of individual species to quantify the influence of environmentalfactors on the species distribution, and fitting clusterpoint process models allows quantifying spatial clusteringwithout and with consideration of the underlying habitatmodel (Shen et al. 2009; Waagepetersen & Guan 2009).Combination of these elements allows testing of four apriorihypotheses on single and joint effects of habitat heterogeneityand internal clustering: (i) random placement (no habitat association,no clustering), (ii) habitat heterogeneity (no internalclustering), (iii) internal clustering (no habitat association) and(iv) joint effects of habitat heterogeneity and internal clustering(Shen et al. 2009). Comparison of the observed SAR with thatof simulated communities corresponding to the differenthypotheses helps identify the hypothesis that is most consistentwith the data.An important uncertainty of this approach (and tests ofneutral theories in general; McGill, Maurer & Weiser 2006)is whether the SAR is sensitive enough to separate competinghypotheses of the underlying mechanisms or not. It hasbeen suggested that the SAR may have low discriminatorypower in distinguishing between niche assembly and dispersalassembly (e.g. Chave 2004; Purves & Pacala 2005),and McGill, Maurer & Weiser (2006) argued that additionalpredictions should be evaluated. One promising alternativesummary statistic of community structure is the decay ofsimilarity with distance curve (Chave & Leigh 2002; Conditet al. 2002; Morlon et al. 2008; (McGill 2010). As the distancedecay curve evaluates other aspects of spatial communitystructure than the SAR (Morlon et al. 2008), it is notclear aprioriif both will favour the same hypothesis. Forexample, Morlon et al. (2008) showed that one hypothesisfor explaining the structure of three tropical forest communitiesyielded accurate predictions for the SAR, but not for alldistance decay curves.Previous studies have found that species often showdifferent ecological habitat associations (Webb & Peart 2000;Comita, Condit & Hubbell 2007) and different degrees of spatialclustering with life stage (Wiegand et al. 2007b; Wang et al.2010b). Do such species-specific variations even out on thecommunity level or do they result in community-wide shifts inthe relative importance of internal and external clustering withlife stage? For example, strong influence of habitat associationfor small trees, but a loss of strong habitat associations forlarge (adult) trees could be interpreted as support for neutraltheories in structuring canopy tree communities (e.g. Hubbell2001).In this study, we test four apriorihypotheses on the relativeimportance of habitat heterogeneity and internal clusteringmechanisms on spatial community structure of two 25-ha fullymapped plots of temperate forests in the Changbaishan (CBS)Nature Reserve, north-eastern China and the Wabikon plot inthe Chequamegon-Nicolet National Forest of north-easternWisconsin, USA. This analysis allows us to address three specificobjectives. Firstly, we explore the relative importance ofhabitat heterogeneity and internal clustering in explaining theobservedSAR.Secondly,weusethedistancedecaycurveasan additional summary statistic to test the generality of theresults obtained for the SAR. Lastly, we analyse assemblagesof small trees [

1384 X. Wang et al.Materials and methodsSTUDY AREAS AND FIELD METHODSTwo temperate forest tree communities were chosen for this study.The first is represented by a 25-ha (500 · 500 m), fully censused temperateforest plot (42°23¢N, 128°05¢E) in the Changbaishan (CBS)Nature Reserve, north-eastern China. The Reserve, located along theborder of China and North Korea, is one of the largest biospherereserves in China and has been spared from logging and other severehuman disturbances. Mean elevation in the CBS temperate forest plotis 801.5 m a.s.l., and elevation ranges from 791.8 to 809.5 m. All freestandingtrees at least one centimetre in d.b.h. were mapped andidentified to species, and their geographical coordinates wererecorded following a standard field protocol (Condit 1998) by scientistsfrom the Institute of Applied Ecology of the Chinese Academyof Science. The first census in 2004 yielded 38,902 living individuals(d.b.h. ‡1 cm) belonging to 52 species, 32 genera and 18 families. Themain tree species included Pinus koraiensis, Tilia amurensis, Quercusmongolica, Fraxinus mandshurica, Ulmus japonica and Acer mono.Unlike tropical rain forests without obvious dominant species, eightspecies were recorded with more than 1000 individuals, togetheraccounting for 83.4% of the total individuals in the plot. Mean standdensity was 1556 living trees per hectare. Mean basal area was43.2 m 2 ha )1 (Hao et al. 2008; Wang et al.2009).The second data set derives from the 25.2-ha (300 · 840 m) WabikonForest Dynamics plot (45°33¢N, 88°48¢W) established in 2008 byscientists at the University of Wisconsin-Green Bay using the samemethods described by Condit (1998). Living trees of at least 1 cmd.b.h. numbered 48 858, belonging to 36 species, 28 genera and 17families. The Wabikon plot is located within the Chequamegon-NicoletNational Forest in north-eastern Wisconsin, USA, c. 10 km eastof Crandon. The glacially formed topography consists of hummockyoutwash features, including an esker running through part of the site.Elevations range from 488.3 to 514.2 m, with a mean of 498.1 m.Mesic northern hardwoods occupy most of the plot, dominated bysugar maple (A. saccharum), basswood (T. americana), white ash(F. americana) and ironwood ⁄ eastern hop hornbeam (Ostrya virginiana).Like the CBS plot, a relatively small number of speciescomprised the majority of individuals; 10 species were represented by1000 or more individuals, together comprising 95.1% of all liveindividuals. The eight most abundant species at the Wabikon plotrepresented 90.1% of all individuals. Mean stand density was 1939living trees per hectare. Mean basal area was 32.0 m 2 ha )1 .To examine the effect of habitat heterogeneity on species–area relationshipsof the CBS and Wabikon plots, we evaluated three topographicalvariables (elevation, slope and aspect) and, for the CBSplot, eight soil properties (pH, organic matter, total N, total P, totalK and available N, available P and available K). The plots weredivided into grid systems using a 5 · 5 m quadrat size, and the meanvalues for these environmental variables (topographical and soil variables)were then calculated at the 5-m scale using geostatistical methods.Overall tree density in each quadrat (5 · 5 m) was alsocalculated and used as a comprehensive bioenvironmental index forthis analysis.POINT PATTERN ANALYSISWe used recent advances in the theory of spatial point processes(Møller & Waagepetersen 2003; Illian et al. 2008; Waagepetersen &Guan 2009) to fit corresponding species-specific point process modelsfor each of the four alternative hypotheses. For each species, we subsequentlygenerated 100 realizations of a fitted point process model(i.e. simulated distribution patterns) and independent superpositionof the simulated distribution patterns for each hypothesis results in100 simulated communities. For each simulated community, we calculatedSAR and the distance decay curve and compared them withthe observed patterns. The algorithms of the four point processeshave been described in detail by Møller & Waagepetersen (2003),Illian et al. (2008) and others. Herein, we only summarize the basicframework of the four processes.Homogeneous ⁄ inhomogeneous Poisson processThe random placement hypotheses can be represented by a homogenousPoisson process in which the points are: (i) independently scatteredand (ii) the intensity k of the process (i.e. the mean point densityin a unit area) is constant (Stoyan & Stoyan 1994). The habitat heterogeneityhypothesis can be represented by an inhomogeneous Poissonprocess in which condition: (i) holds, but where the intensity of theprocess depends on location x (i.e. the probability k(x)dx of a pointoccurring in an infinitesimally small disc of centre x and area dxdepends on location x). The intensity k(x) may be influenced by environmentalfactors. In general, statistical habitat models or species distributionmodels (Elith & Leathwick 2009) may be used forparametric estimation of the intensity function. The most obviousparametric model to fit the intensity function for a heterogeneousPoisson process is the loglinear model (Waagepetersen 2007).Homogeneous ⁄ inhomogeneous Thomas processConsidering the unrealistic independence assumption (i) of the twoPoisson process for real data, two kinds of cluster processes are usedto model clustered spatial distribution patterns. To represent theinternal clustering hypothesis we used the homogeneous Thomascluster processes (Thomas 1949). It generates a number of randomlyand independently distributed clusters, where the cluster centres followa homogeneous Poisson process with intensity q, and the numberof points per cluster follows a Poisson distribution with meanl = k ⁄ q. The location of the points in a given cluster, relative to thecluster centre, has a bivariate Gaussian distribution with variance r 2(Stoyan & Stoyan 1994). The K-function of the Thomas process canbe calculated analytically (Stoyan & Stoyan 1994; Wiegand et al.2007b), which allows fitting the parameters of this process to the datafor each species. Realizations of the fitted process can be easily simulated(Stoyan & Stoyan 1994). Note that fitting a Thomas process toan inhomogeneous pattern may formally produce a good fit. Thus,some of the effects of environmental heterogeneity may already beaccounted for by a homogeneous Thomas process.The inhomogeneous Thomas process represents the most complexhypothesis, where habitat heterogeneity and internal clustering occursimultaneously. The inhomogeneous Thomas process results fromthinning a homogeneous Thomas process with intensity function k(x)(Waagepetersen 2007). If k(x) is known, the parameters of the correspondinghomogeneous Thomas process can be fitted using the inhomogeneousK-function (Baddeley, Møller & Waagepetersen 2000;Waagepetersen 2007). This process provides a simple phenomenologicaldescription of clustering that explicitly includes the effect of environmentalheterogeneity.Model fittingWe used a two-step approach proposed by Waagepetersen & Guan(2009) to estimate the parameters of our four point processes. ThereÓ 2011 The Authors. Journal of Ecology Ó 2011 British Ecological Society, Journal of Ecology, 99, 1382–139395

Detecting underlying mechanisms controlling species–area relationships 1385are basically two sets of parameters: regression parameters for the differentcovariates to estimate intensity functions k(x) (inhomogeneousPoisson and inhomogeneous Thomas process) and the clusteringparameters q and r of the homogeneous and inhomogeneous Thomasprocess. We used minimum contrast estimation to estimate theclustering parameters (Stoyan & Stoyan 1994). A Poisson likelihoodfunction corresponding to the proposed intensity function was maximizedto estimate the regression parameters both in case of the inhomogeneousPoisson process and the inhomogeneous Thomasprocess. Although this is not a maximum likelihood estimation, onecan show that the estimated regression parameters are still consistentand asymptotically normal (Waagepetersen 2007; Waagepetersen &Guan 2009).We used the soil variables together with three topographicalparameters and the bioenvironmental index (total tree density in a5 · 5 m quadrat) as environmental variables to determine the intensityfunction k(x). To reduce the risk of over-fitting, we computedthe principal components (PCs) from the eight soil variables andused only the first two components as condensed variables becausetogether they explained 97.2% of total variance in soil variables(John et al. 2007). The intensity function is then fitted using maximumlikelihood estimation to models of the loglinear formkðxÞ ¼expðb 0 þ b 1 m 1 ðxÞþ::: þ b n m n ðxÞÞ with coefficients b i and thevariables m i (x). At CBS we have n = 6 variables and at Wabikonn = 4 variables. To account for the problem (common to both inhomogeneousPoisson and inhomogeneous Thomas processes) thatsome random variation is bound to be picked up by covariates eventhough these covariates in reality do not influence the spatial patternof trees, we performed stepwise model reduction using Wald-tests(Waagepetersen & Guan 2009). Otherwise, too much variation wouldbe attributed to the covariates.Model selectionWe generated, for each hypothesis, 100 communities by superposingrealizations of the fitted point process models for each species.The species–area relationships were constructed by randomlythrowing quadrats with increasing sizes in these simulated communities(Shen et al. 2009). To calculate the distance decay curve, wedivided the plot into 20 · 20 m quadrats and calculated the similarityamong these quadrats using the Jaccard index of similarity(Legendre & Legendre 1998). The predicted summary statistics (i.e.SAR and distance decay curve) for the four models were computedby averaging the simulated patterns of 100 simulated communities,and 95% simulation envelopes were constructed for each predictedsummary statistic.Finally, the observed summary statistics from the original dataobtained in the CBS and Wabikon plots were compared with thesummary statistics predicted by the four point process models. Amodel is considered satisfying, if the observed summary statistic fallswithin the simulation envelopes of the predicted summary statistic.To select the hypothesis that received most support from the data, weused a type of Akaike’s Information Criterion (AIC) in which the loglikelihood was approximated by the sum of squared residuals foreither the SAR or the distance decay summary statistics (Webster &McBratney 1989; Shen et al. 2009). Due to the stepwise model reductionused for estimation of the intensity function, the number ofparameters for the inhomogeneous Poisson and Thomas process maydiffer among species. However, as we calculated the AIC on the communitylevel (superposing realizations of the point process models forindividual species), we counted all covariates that were used at a givensite and model at least once.In addition, we illustrated the observed and simulated spatial distributionmaps of U. japonica in the CBS plot. The distribution functionG(r) of the nearest-neighbour distances r was then calculated for eachsimulated distribution pattern to evaluate the goodness of fit of eachmodel (Ripley 1988; Møller & Waagepetersen 2003). All calculationswere carried out in R version 2.10.0 (R Development Core Team2009), using the ‘spatstat’ package (Baddeley & Turner 2005).ResultsSMALL AND LARGE TREES TOGETHERThe species–area pattern in the two temperate forest plots(CBS and Wabikon) showed a similarly increasing tendency ofspecies richness with increased sampling area (Figs 1a and 2a)although the final increase at areas larger than 12 ha was somewhatsteeper at Wabikon. Note that all four hypotheses provideseemingly accurate results near the two ends of thespecies–area relationships, i.e.near0and25 haintheCBSplotor 0 and 25.2 ha in the Wabikon plot. However, this was anartefact because the total species richness and plot area werefixed regardless of what models were applied.The observed distribution of U. japonica at the CBS plot(Fig. 3e) together with realizations of the different point processmodels (Fig. 3a–d) illustrates a typical result (see Fig. S1Supporting Information). It is clear from visualization of thepatterns that the two point processes without internal clustering(i.e. hypothesis i and ii; Fig. 3a,b) miss important aspectsof the observed spatial structure. The homogenous Thomasprocess (i.e. hypothesis iii) reproduces the small-scale clusteringbetter (Fig. 3c); however, it produces unrealistically largegaps and does not effectively reproduce the spatial variation intree density. The realization of the inhomogeneous Thomasprocess with soil factors (hypothesis iv; Fig. 3d) overcomes theshortcomings of the homogeneous Thomas process, and producedpatterns that agreed well with the observed distributionof U. japonica.The above observations for a single species generally holdfor the entire communities. In most cases, the SAR was able todistinguish among the four hypotheses. Hypotheses i and iiwere clearly rejected both on the basis of the simulation envelopes(Figs 1b and 2b) and on the basis of formal model selectionusing AIC (Table 1). Both hypotheses tended toconsiderably overestimate species richness. Interestingly, whensoil variables were not considered, hypothesis iii (i.e. the internalclustering hypothesis) received the most support for bothplots (Table 1), but yielded not fully satisfying fits (Figs 1band 2b). The more complex model for the Wabikon forest(hypothesis iv) received almost the same support as model iii(DAIC = 3.4), but internal clustering (hypothesis iii) slightlyunderestimated species richness at larger areas (>7 ha)whereas consideration of the joined effect of internal clusteringand habitat association (hypothesis iv) slightly overestimatedspecies richness (Fig. 2b). Inclusion of the two soil variablesinto hypothesis iv for the CBS plot produced an excellent fit ofthe observed SAR (Fig. 1b) and was clearly the most parsimoniousmodel (Table 1).Ó 2011 The Authors. Journal of Ecology Ó 2011 British Ecological Society, Journal of Ecology, 99, 1382–139396

1386 X. Wang et al.(a)All CBS(b)(c)Large CBS(d)(e)Small CBS(f)Number of speciesNumber of speciesNumber of speciesArea (ha)Area (ha)Fig. 1. The observed (dots) and predicted (coloured lines) species–area relationships for the data from the CBS plot. Small and large treestogether (a, b), only for large trees (‡10 cm d.b.h.; c, d), and only for small trees (

Detecting underlying mechanisms controlling species–area relationships 1387WabikonAllLargeSmallNumber of speciesNumber of speciesNumber of speciesArea (ha)Area (ha)Fig. 2. The observed (dots) and predicted (coloured lines) species–area relationships for the data from the Wabikon plot. Conventions as inFig. 1.plots was substantially lower (34 at CBS and 23 at Wabikon).Unsurprisingly, the SAR results obtained for separate analysisof small trees did not differ substantially from those obtainedforalltreestogether(Figs 1and2).Separate analyses of the distance decay function for smalland large trees generally supported the results of the analysisfor all trees; hypothesis iv was selected in all cases. Again,hypothesis iv with and without soil factors received the samesupport for separate analysis of small and large trees at theCBS plot (DAIC

1388 X. Wang et al.Table 1. Comparison of Akaike’s Information Criterion (AIC) among the four process models for the different analyses and forest plotsPoisson processThomas processHomogeneousrandom placementInhomogeneous habitatheterogeneityHomogeneous internalclusteringInhomogeneoushabitat heterogeneityand internal clusteringSARAll CBS 260.9 229.2#, 212.1$ 117.8 136.2#, 94.7Wabikon 288.1 210.9 152.3 155.7Large CBS 221.1 183.4#, 144.9$ 139.4 145.8#, 82.5$Wabikon 242.2 160.7 79.9 18.9Small CBS 257.2 211.7#, 182.1$ 81.6 109.6#, 45.4$Wabikon 288.4 228.2 110.6 159.2Distance decay curveAll CBS 153.4 112.8#, 107.8$ 44.0 41.1#, 39.1Wabikon 171.9 140..7 95.7 40.0Large CBS 34.1 11.6#, 9.1$ 9.4 11.2#, 11.1$Wabikon 51.5 39.4 26.8 18.2Small CBS 95.5 62.9#, 58.3$ 17.1 15.2#, 14.5$Wabikon 115.3 89.2 22.6 17.7#, without soil data, $, with soil data.of 23 species, slope: 7 ⁄ 23, aspect: 9 ⁄ 23) compared with theCBS plot (elevation: 7 ⁄ 32, slope: 5 ⁄ 32, aspect: 3 ⁄ 32, PCA1:9 ⁄ 32, PCA2: 12 ⁄ 32). As expected, the coefficients for analyseswith all individuals were in close accordance with those forsmall individuals, but coefficients for large trees were oftenquite different (Tables S2 and S3).DiscussionOur results have shown that the combined effects of habitatheterogeneity (including soil factors) and internal clusteringlead to good approximations of the observedspecies–area relationships and distance decay curves in tworepresentative temperate forests, one in north-eastern Chinaand the other in north-central United States. The species–area relationship proved to be a sensitive summary statisticthat can detect subtle differences among competing pointprocess models and yielded results that, in general, agreedwith those of the distance decay curve. Contrary to ourexpectations, the most complex model including habitatassociation and internal clustering received, in most cases,the most support. Thus, we did not find substantial differencesin the relative importance of habitat association andinternal clustering with life stage (this would be given ifdifferent models would be selected for different life stages).Exceptions were the SAR of small trees at Wabikon,where the homogeneous Thomas process was clearlyfavoured and two cases (SAR of all trees at Wabikon anddistance decay for all trees at CBS) where several modelsincluding the most complex one received similar support(Table 1). Comparison of our results from these two relativelydiverse temperate forests with spatial patterns oftrees in highly diverse tropical and subtropical forestssuggests that community structure may be governed by thesame underlying mechanisms.RELATIVE IMPORTANCE OF INTERNAL AND EXTERNALAGGREGATION MECHANISMSIn our analyses, the random placement hypothesis yielded thepoorest model for predicting spatial community structure. Inparticular, this model overestimated species richness for mostof the sampling range and did not account for the observeddecline in similarity with distance. This result is consistent withprevious studies that have identified spatial clustering as thedominant pattern of species from temperate forests to tropicalforests (He, Legendre & LaFrankie 1997; Condit et al. 2000;Morlon et al. 2008; Wang et al. 2010b). However, the randomplacement hypothesis performed relatively well at the CBSplot and for the analysis of all trees yielded a maximal overestimationof species richness of c. 28% at the 0.9-ha area: (36 specieswere predicted and 28 observed). At Wabikon, themaximal overestimation yielded 64% at the 0.6-ha area (23species were predicted and 14 observed). This result and thesteeper increase of the SAR for larger areas at Wabikon suggeststhat the CBS plot shows a better mixing of species withinthe plot compared with the Wabikon plot. Indeed, some speciesat the Wabikon plot show very patchy distribution patternsthat result in hotspots of species richness (and number ofindividuals), related in part to the elevation pattern (Fig. S2).These trends are much weaker at the CBS forest (Fig. S3).Consequently, the consideration of habitat heterogeneity (i.e.hypothesis ii) improved the SAR prediction at Wabikon considerably(the maximum error dropped from nine to four species),but only moderately at the CBS plot. The Wabikon plotexperienced localized logging during the 1900s, as evidencedby historical records and air photos dating to 1938. These disturbances,in addition to pronounced variation in elevation,have contributed to spatial heterogeneity of trees in parts ofthe plot. Nevertheless, we found for analyses of all trees thatthe distribution pattern of approximately two-thirds of allÓ 2011 The Authors. Journal of Ecology Ó 2011 British Ecological Society, Journal of Ecology, 99, 1382–139399

Detecting underlying mechanisms controlling species–area relationships 13890 100 200 300 400 5000 100 200 300 400 500(a) Homogeneous Poisson (b) Inhomogeneous Poisson0 100 200 300 400 5000 100 200 300 400 5000 100 200 300 400 500(c) Homogeneous Thomas (d) Inhomogeneous Thomas0 100 200 300 400 500 0 100 200 300 400 5000 100 200 300 400 5000 100 200 300 400 500Real distribution0 100 200 300 400 500Fig. 3. The observed distribution of Ulmus japonica in the CBS plot and its distribution generated from the homogeneous Poisson process, inhomogeneousPoisson process with soil factors, homogeneous Thomas process and inhomogeneous Thomas process with soil factors.(e)species showed a significant relationship with environmentalvariables (Table S1). Thus, habitat heterogeneity cannot beneglected at our two study plots, but it is not sufficient to producethe observed patterns in SAR and distance decay. Thisresult suggests a need to consider additional clustering mechanisms.Recently, both theoretical and empirical studies haveemphasized the importance of dispersal limitation in controllingspecies distributions (Condit et al. 2002; Seidler &Plotkin 2006; Wiegand, Martinez & Huth 2009). However,the effect of dispersal limitation on spatial patterns is rarelystraightforward and many other biological interactions andprocesses (other than habitat association) may contribute tothe spatial pattern of a species. Seidler & Plotkin (2006)found that the extent and scale of conspecific spatial clusteringin the 50-ha Pasoh Forest Plot was correlated with themode of seed dispersal. Thus, dispersal limitation is likely toleave a signal on the species clustering process and thus onspecies–area patterns and distance decay curves. Our resultsindicate that homogeneous Thomas processes mimicking thespatial clustering of species (i.e. hypothesis iii) reproduce theSAR fairly well (see also Plotkin et al. 2000; Morlon et al.2008), but failed in reproducing observed distance decaycurves. This is because the homogeneous Thomas processcan produce good fits to aggregation caused by environmentalheterogeneity, but it cannot describe positive dependencyin the distribution pattern among species mediated by habitatassociation. As a consequence, it produces communitiesÓ 2011 The Authors. Journal of Ecology Ó 2011 British Ecological Society, Journal of Ecology, 99, 1382–1393100

1390 X. Wang et al.(a)CBS-all(b)Wabikon-all(c)CBS-large(d)Wabikon-large(e)CBS-small(f)Wabikon-smallDistance d (m)Distance d (m)Fig. 4. The observed (dots) and predicted (coloured lines) decay of similarity with distance curve for the CBS and the Wabikon plot. We used theJaccard index to quantify the similarity of 20 · 20 m quadrats that were distance d away. Colour conventions as in Fig. 1.that are too well mixed (i.e. similarity decays too quickly;Fig. 4a,b). However, the inhomogeneous Thomas process(hypothesis iv), which places clusters only at suitable areas,can generate the positive association among species necessaryto produce the observed decay in similarity with distance. Inparticular, the inhomogeneous Thomas process is able toaccount for the patchy diversity hotspots at Wabikon forest.In other words, this model is better able to simulate spatialpatterns for communities that show strong spatial structuring(Figs S2 and S3). Thus, at the local spatial scale studiedhere, both habitat association and internal clustering arerequired to explain the observed patterns in spatial communitystructure.Our results on the relative importance of internal vs. externalaggregation mechanisms agree with the findings of Shenet al. (2009) and Morlon et al. (2008) for highly diverse subtropicaland tropical forests. This is a further indication thatthe effects of the environment and internal clustering mechanismscontribute in a complementary way to the assembly ofspecies rich communities (He & Legendre 2002; Shen et al.2009).However, it is important to emphasize that species–area patternsare dramatically dependent on spatial scales (Palmer &White 1994; Whittaker, Willis & Field 2001). Our study wasconfined to the local community scale, where habitat typesusually do not change dramatically, but rather more gradually(as depicted by the habitat models for the intensity function).Beyond the local scale, however, the relative effects of differentprocesses may change. For example, dispersal limitation mayhave a dominant effect on the species–area relationships at thelocal community scale, whereas habitat heterogeneity maybecome more pronounced at the regional scale (Kallimaniset al. 2008).SAR VS. DISTANCE DECAY CURVEOur results support the analysis of more than onesummary statistic for understanding spatial patterns ofÓ 2011 The Authors. Journal of Ecology Ó 2011 British Ecological Society, Journal of Ecology, 99, 1382–1393101

Detecting underlying mechanisms controlling species–area relationships 1391community structure (McGill, Maurer & Weiser 2006). Ingeneral, the distance decay curve supported the resultsfound for the SAR, but there was a somewhat curiousreversal of the AIC model ranking between the CBS andWabikon plots (Table 1). At the Wabikon plot, the distancedecay curve favoured for all trees and small treesthe more complex model iv that included the influence ofboth habitat association and internal clustering, whereasthe SAR favoured the simpler model iii that included internalclustering only (although the AIC difference for all treeswas small). At the CBS plot, the distance decay curve couldnot discriminate between models with or without habitatassociations, whereas the SAR clearly favoured the morecomplex model iv with soil variables included. The distancedecay curve of hypothesis iii for all trees at Wabikon(Fig. 4b) clearly underestimated similarity, but this discrepancywas only slight at CBS (Fig. 4a). As discussed above,this may be due to the inability of hypothesis iii to generatepositive associations among species caused by shared habitats.Consequently, hypothesis iv, which included habitatassociation, improved the fit of the distance decay curve atWabikon considerably. This effect is also visible at the CBSplot, but habitat associations appear to be weaker and influentialover a smaller range of distances (100–200 m). Thus,the distance decay curve proved to be sensitive to violationof the independence assumption of hypothesis iii due to habitatassociation, but the SAR was not as sensitive. Thisimportant result shows that the independence assumptionunderlying most theories of stochastic geometry of biodiversity(McGill 2010) may indeed not be valid. If local diversityis patchy with marked diversity hotspots as shown by theWabikon plot, positive association among species (mediatedby habitat heterogeneity) is required to account for theobserved distance decay curve. The CBS plot, however, wasmore strongly characterized by interspecific segregation patterns(Wang et al. 2010a) that resulted in a more even distributionof local diversity. Consequently, the error of notconsidering habitat heterogeneity was less severe for the distancedecay curve.The distance decay curve was not always able to distinguishamong competing hypotheses. For example, the distancedecaycurveforlargetreesattheCBSplotwasnotable to distinguish among hypotheses ii, iii and iv with andwithout soil factors (Table 1), unlike the SAR. The reasonfor this is probably that the community of large trees atCBS was only weakly clustered, an attribute for which thedistance decay curve is especially sensitive. Morlon et al.(2008) have demonstrated that distance decay is closelyrelated to the pair correlation function, which describesdetails of clustering of individual species (Wiegand, Martinez& Huth 2009), whereas the SAR is related to the sphericalcontact distribution H S (r), the probability that there isno tree of species i within distance r from the centre of acircular sampling plot with radius r (Illian et al. 2008). Notethat the spherical contact distribution basically describes theempty space between the clusters (Illian et al. 2008), and istherefore less sensitive to details of smaller-scale clusteringas long as a simplified point process model represents thegaps.ENVIRONMENTAL VARIABLESOur analysis revealed some interesting differences betweentemperate forests and species rich tropical and subtropical forests(Shen et al. 2009). For example, we found that, withoutincluding soil factors, consideration of environmental heterogeneitydid not improve the predicted SAR based on internalclustering, unlike the results of Shen et al. (2009) for the tropicalforest plot on Barro Colorado Island (BCI). Interestingly,this result applies to a separate analysis of small and large treecommunities. Thus, soil factors may have a stronger effect onspecies–area patterns in the CBS forests compared with theBCI plot. The flat topography of the CBS plot (elevation variesonly 18 m; Wang et al. 2008) may explain the weak impact oftopographical variables compared with that at BCI, where elevationwithin the plot varies 40 m.The stronger effect of topography at BCI could also berelated to the specific hydrological conditions, where slopes arewetter than plateaus and experienced a shorter drought duringthe dry season (Daws et al. 2002; Leigh et al. 2004). In addition,it should be emphasized that we interpreted the effect ofsoil factors on these spatial patterns as directional responses ofspecies to variations in soil properties in the study. However,tree species in forests may both affect and respond to soil propertiesthrough litterfall inputsandeffectsonmicrobialcommunitiesand decomposition rates, etc. (Boerner & Koslowsky1989; Finzi, Canham & Van Breemen 1998). In other words,soil variables are not necessarily abiotic factors generated byprocesses extrinsic to population and community dynamics.Although some studies have argued that biotic feedback effectsare less likely to influence spatial variation in soil nutrientavailability in species rich tropical forests (Powers, Kalicin &Newman 2004), their influence on soils in temperate forests isstill poorly known.At both of our temperate forest plots, the intensity functionfor most species was positively related with the bioenvironmentalindex (overall tree density in each 5 · 5mquadrat;Tables S2 and S3). This suggests that some environmental constraint,common for most species, was not captured by theenvironmental variables used in our analysis.SIMPLIFICATIONS OF THE APPROACHThe point process models that combined habitat heterogeneityand internal clustering provided relatively good, but not alwaysperfect, fits to the SAR. This is to a large extent due to thepower of the Thomas processes to represent clustering (see alsoPlotkin et al. 2000; Morlon et al. 2008). However, the predictionsfor the distance decay curve may be improved using pointprocess models that are able to capture more complex distributionpatterns and underlying processes. For example, Wiegandet al. (2007b); Wiegand, Martinez and Huth (2009) showedthat species may often cluster at several critical scales. Considerationof only one scale of clustering did not severely reduceÓ 2011 The Authors. Journal of Ecology Ó 2011 British Ecological Society, Journal of Ecology, 99, 1382–1393102

1392 X. Wang et al.the power of our point process models to describe the observedspatial community structure. However, the analysis of Morlonet al. (2008) for the tropical forest at the Korup National Park(Cameroon) showed that hypothesis iii overestimated similarity,which is the opposite of our results. They attributed this tothe inability of the Thomas process to describe the complexmulti-clustered spatial patterns at this plot (Morlon et al.2008). The fit with the homogeneous Thomas process probablycaptured only the larger scale of clustering and produced patternsthat lacked the small-scale clustering necessary for producinglower similarities among subplots.An interesting result of our study is that the assumption ofindependence among the patterns of individual species led to agood fit of models with the observed species–area patterns(Plotkin et al. 2000). Herein, we found that inclusion of positive,habitat-mediated species associations were important tofit the distance decay curve (but not SAR) at the Wabikonplot, where diversity hotspots are present. The segregation patternsthat dominated the intraspecific species relationships oflarger trees (>10 cm d.b.h.) at the CBS plot (Wang et al.2010a) are probably also produced by the impact of environmentalheterogeneity, because the point process model thatassumed homogeneous clustering (without habitat heterogeneity)failed to reproduce both the SAR and the distance decaycurve. However, intraspecific interactions of large trees thatwere detected for 1 ⁄ 3 of all pairs of large species at the CBSplot (after removing large-scale habitat effects; Wang et al.2010a) cancelled out and did not compromise the fit of theSAR and the species decay curve. Thus, smaller-scale speciesinteractions may not be major factors structuring spatial communitystructure.Finally, the point process models used here are static, anddo not incorporate the effects of temporal processes or site history(Ripley 1988; Plotkin et al. 2000; Møller & Waagepetersen2003). In all, more sophisticated dynamic models arerequired to further explain the underlying mechanisms controllingthe relationship between species richness and samplingarea at these spatial scales.AcknowledgementsWe thank the CBS and Wabikon plot census and data management teams, thelatter including many student assistants from the University of Wisconsin-Green Bay. In particular, we thank Buhang Li, Ji Ye, Fei Lin, Zuoqiang Yuan,Xuejiao Bai, Liwei Wang and Dingliang Xing for their field work and data collectionin the CBS plot. At the Wabikon plot, Gary Fewless, Kathryn Corioand Juniper Sundance were instrumental in collecting and managing the data.This study was sponsored by the Knowledge Innovation Program of the ChineseAcademy of Sciences (KZCX2-YW-QN402, KSCX2-EW-Z-5 andKZCX2-EW-401), the National Natural Science Foundation of China(40971286) and the National Key Technologies R&D Program of China(2008BAC39B02). The ERC advanced grant 233066 supported travelling ofX.W. The Center for Tropical Forest Science of the Smithsonian TropicalResearch Institute and the HSBC Climate Partnership provided support forX.W. and for establishment of the Wabikon Plot. Primary funding for the WabikonPlot was provided by the Cofrin Center for Biodiversity with additionalsupport from the Department of Natural and Applied Sciences at the Universityof Wisconsin-Green Bay. We are grateful for permission and importantlogistic contributions by the U.S. Forest Service, particularly Chequamegon-Nicolet National Forest ecologists Linda Parker and Steven Janke. Other significantcontributors to the project include Richard Condit, Steve Dhein andKimberlee McKeefry.ReferencesBaddeley, A.J., Møller, J. & Waagepetersen, R. (2000) Non- and semiparametricestimation of interaction in inhomogeneous point patterns. StatisticaNeerlandica, 54, 329–350.Baddeley, A. & Turner, R. (2005) SPATSTAT: an R package for analyzing spatialpoint patterns. Journal of Statistical Software, 12, 1–42.Boerner, R.E.J. & Koslowsky, S.D. (1989) Microsite variations in soil chemistryand nitrogen mineralization in a beech-maple forest. Soil Biology and Biochemistry,21, 795–801.Chave, J. (2004) Neutral theory and community ecology. Ecology Letters, 7,241–253.Chave, J. & Leigh, E.G. (2002) A spatially explicit neutral model of beta-diversityin tropical forests. Theoretical Population Biology, 62, 153–168.Comita, L.S., Condit, R. & Hubbell, S.P. (2007) Developmental changes inhabitat associations of tropical trees. Journal of Ecology, 95, 482–492.Condit, R. (1998) Tropical Forest Census Plots. Springer, Berlin.Condit, R., Ashton, P., Baker, P., Bunyavejchewin, S., Gunatilleke, S., Gunatilleke,N. et al. (2000) Spatial patterns in the distribution of common andrare tropical tree species: a test from large plots in six different forests. Science,288, 1414–1418.Condit, R., Pitman, N., Leigh, E.G., Chave, J., Terborgh, J., Foster, R.B. et al.(2002) Beta-diversity in tropical forest trees. Science, 295, 666–669.Connor, E.F. & McCoy, E.D. (1979) The statistics and biology of the species–area relationship. American Naturalist, 113, 791–833.Daws, M.I., Mullins, C.E., Burslem, D., Paton, S.R. & Dalling, J.W. (2002)Topographic position affects the water regime in a semideciduous tropicalforest in Panama. Plant and Soil, 238, 79–90.Elith, J. & Leathwick, J. (2009) Species distribution models: ecological explanationand prediction across space and time. Annual Review of Ecology, Evolutionand Systematics, 40, 677–697.Felinks, B. & Wiegand, T. (2007) Exploring spatiotemporal patterns in earlystages of primary succession on former lignite mining sites. Journal of VegetationScience, 19, 267–276.Finzi, A.C., Canham, C.D. & Van Breemen, N. (1998) Canopy tree-soil interactionswithin temperate forests: species effects on pH and cations. EcologicalApplications, 8, 447–454.Hao, Z., Li, B., Zhang, J., Wang, X., Ye, J. & Yao, X. (2008) Broad-leavedKorean pine mixed forest plot in Changbaishan (CBS) of China: communitycomposition and structure. Acta Phytoecologica Sinica, 32, 238–250. (in Chinese).He, F. & Legendre, P. (2002) Species diversity patterns derived from species–area models. Ecology, 83, 1185–1198.He, F., Legendre, P. & LaFrankie, J.V. (1997) Distribution patterns of tree speciesin a Malaysian tropical rain forest. Journal of Vegetation Science, 8,105–114.Holt, R.D., Lawton, J.H., Polis, G.A. & Martinez, N.D. (1999) Tropical rankand the species–area relationship. Ecology, 80, 1495–1504.Howe, H.F. (1989) Scatter and clump dispersal and seedling demography:hypothesis and implications. Oecologia, 79, 417–426.Hubbell, S.P. (1997) A unified theory of biogeography and relative speciesabundance and its application to tropical rain forests and coral reefs. CoralReefs, 16,S9–S21.Hubbell, S.P. (2001) The Unified Neutral Theory of Biodiversity and Biogeography.Princeton University Press, Princeton, NJ.Illian, J., Penttinen, A., Stoyan, H. & Stoyan, D. (2008) Statistical Analysis andModeling of Spatial Point Patterns. Wiley, Chichester.John, R., Dalling, J.W., Harms, K.E., Yavitt, J.B., Stallard, R.F., Mirabello,M., Hubbell, S.P., Valencia, R., Navarrete, H., Vallejo, M. & Foster, R.B.(2007) Soil nutrients influence spatial distributions of tropical tree species.Proceedings of the National Academy of Sciences, 104, 864–869.Kallimanis, A.S., Mazaris, A.D., Tzanopoulos, J., Halley, J.M., Pantis, J.D. &Sgardelis, S.P. (2008) How does habitat diversity affect the species-area relationship?Global Ecology and Biogeography, 17, 532–538.Kikvidze, Z., Pugnaire, F., Brooker, R., Choler, P., Lortie, C., Michalet, R. &Callaway, R. (2005) Linking patterns and processes in alpine plant communities:a global study. Ecology, 86, 1395–1400.Legendre, P. & Legendre, L. (1998) Numerical Ecology, 2nd English edn. ElsevierScience BV, Amsterdam.Leigh, E.G.J., Lao, S., Condit, R., Hubbell, S.P., Foster, R.B. & Perez, R.(2004) Barro Colorado island forest dynamics plot, Panama. Tropical ForestDiversity and Dynamism: Findings from a Large-Scale Plot Network (eds E.Losos & E.G.J. Leigh), pp. 451–463. University of Chicago Press, Chicago.McGill, B.J. (2010) Towards a unification of unified theories of biodiversity.Ecology Letters, 13, 627–642.Ó 2011 The Authors. Journal of Ecology Ó 2011 British Ecological Society, Journal of Ecology, 99, 1382–1393103

Detecting underlying mechanisms controlling species–area relationships 1393McGill, B.J., Maurer, B.A. & Weiser, M.D (2006) Empirical evaluation of neutraltheory. Ecology, 87, 1411–1423.Møller, J. & Waagepetersen, R. (2003) Statistical Inference and Simulation forSpatial Point Processes. Chapman & Hall CRC Press, New York.Morlon, H., Chuyong, G., Condit, R., Hubbell, S., Kenfack, D., Thomas, D.,Valencia, R. & Green, J.L. (2008) A general framework for the distancedecayof similarity in ecological communities, theory. Ecology Letters, 11,904–91775:1.Nagel, T., Svoboda, M. & Diaci, J. (2006) Regeneration patterns after intermediatewind disturbance in an old-growth Fagus-Abies forest in southeasternSlovenia. Forest Ecology and management, 226, 268–278.Palmer, M.W. & White, P.S. (1994) Scale dependence and the species-area relationship.American Naturalist, 144, 717–740.Plotkin, J.B., Potts, M.D., Leslie, N., Manokaran, N., Lafrankie, J. & Ashton,P.S. (2000) Species–area curves, spatial aggregation, and habitat specializationin tropical forests. Journal of Theoretical Biology, 20,81–99.Powers, J.S., Kalicin, M.H. & Newman, M.E. (2004) Tree species do not influencelocal soil chemistry in a species rich Costa Rica rain forest. Journal ofTropical Ecology, 20, 587–590.Purves, D.W. & Pacala, S.W. (2005) Ecological drift in niche-structured communities:neutral pattern does not imply neutral process. Biotic Interactionsin the Tropics (eds D. Burslem, M. Pinard & S. Hartley). pp. 108–138, CambridgeUniversity Press, New York.R Development Core Team. (2009) R: A language and environment for statisticalcomputing. R Foundation for Statistical Computing, Vienna, Austria.ISBN 3-900051-07-0. URL: http://www.R-project.org.Ripley, B.D. (1988) Statistical Inference for Spatial Processes. Cambridge UniversityPress, Cambridge.Seidler, T.G. & Plotkin, J.B. (2006) Seed dispersal and spatial pattern in tropicaltrees. PLoS Biology, 4, 2132–2137.Shen, G., Yu, M., Hu, X., Mi, X., Ren, H., Sun, I. & Ma, K. (2009) Species–area relationships explained by the joint effects of dispersal limitation andhabitat heterogeneity. Ecology, 90, 3033–3041.Stoyan, D. & Stoyan, H. (1994) Fractals, Random Shapes and Point Fields:Methods of Geometrical Statistics. Wiley, Chichester, UK.Thomas, M. (1949) A generalization of Poisson’s binomial limit for use in ecology.Biometrika, 36,18–25.Tjørve, E., Kunin, W.E., Polce, C. & Tjørve, K.M.C. (2008) Species–area relationship:separating the effects of species abundance and spatial distribution.Journal of Ecology, 96, 1141–1151.Waagepetersen, R. (2007) An estimating function approach to inference forinhomogeneous Neyman-Scott processes. Biometrics, 63, 252–258.Waagepetersen, R. & Guan, Y. (2009) Two-step estimation for inhomogeneousspatial point processes. Journal of the Royal Statistical Society, Series B, 71,685–702.Wang, X., Hao, Z., Ye, J., Zhang, J., Li, B. & Yao, X. (2008) Spatial pattern ofdiversity in an old-growth temperate forest in Northeastern China. ActaOecologica, 33, 345–354.Wang, X., Hao, Z., Zhang, J., Lian, J., Li, B., Ye, J. & Yao, X. (2009) Tree sizedistributions in an old-growth temperate forest. Oikos, 118, 25–36.Wang, X., Wiegand, T., Hao, Z., Li, B., Ye, J. & Lin, F. (2010a) Species associationsin an old-growth temperate forest in north-eastern China. Journal ofEcology, 98, 674–686.Wang, X., Ye, J., Li, B., Zhang, J., Lin, F. & Hao, Z. (2010b) Spatial distributionsof species in an old-growth temperate forest, northeastern China.Canadian Journal of Forest Research, 40, 1011–1019.Webb, C.O. & Peart, D.R. (2000) Habitat associations of trees and seedlings ina Bornean rain forest. Journal of Ecology, 88, 464–478.Webster, R. & McBratney, A.B. (1989) On the Akaike information criterionfor choosing models for variograms of soil properties. Journal of Soil Science,40, 493–496.Whittaker, R.J., Willis, K.J. & Field, R. (2001) Scale and species richness:towards a general hierarchical theory of species diversity. Journal of Biogeography,28, 453–470.Wiegand, T., Gunatilleke, C.V.S. & Gunatilleke, I.A.U.N. (2007a) Speciesassociations in a heterogeneous Sri Lankan dipterocarp forest. AmericanNaturalist, 170, E77–E95.Wiegand, T., Martinez, I. & Huth, A. (2009) Recruitment in tropical tree species:revealing complex spatial patterns. American Naturalist, 174, E106–E140.Wiegand, T., Gunatilleke, C.V.S., Gunatilleke, I.A.U.N. & Okuda, T. (2007b)Analyzing the spatial structure of a Sri Lankan tree species with multiplescales of clustering. Ecology, 88, 3088–3102.Received 16 February 2011; accepted 24 May 2011Handling Editor: Kyle HarmsSupporting InformationAdditional Supporting Information may be found in the online versionof this article:Table S1. Number of cases for which the intensity function k(x)couldnot be estimated (number of individuals

Author's personal copyBasic and Applied Ecology 12 (2011) 488–495Scale specific determinants of tree diversity in an old growth temperateforest in ChinaZuoqiang Yuan a,b , Antonio Gazol c , Xugao Wang a , Fei Lin a,b ,JiYe a,b , Xuejiao Bai a,b ,Buhang Li a,b , Zhanqing Hao a,∗a State Key Laboratory of Forest and Soil Ecology, Institute of Applied Ecology, Chinese Academy of Sciences, Shenyang 110164, PR Chinab Graduate University of Chinese Academy of Science, Beijing 100049, PR Chinac Institute of Ecology and Earth Sciences, University of Tartu, Lai 40, Tartu 51005, EstoniaReceived 8 March 2011; accepted 20 July 2011AbstractThe major processes generating pattern in plant community composition depend upon the spatial scale and resolution ofobservation, therefore understanding the role played by spatial scale on species patterns is of major concern. In this study, weinvestigate how well environmental (topography and soil variables) and spatial variables explain variation in species compositionin a 25-ha temperate forest in northeastern China. We used new variation partitioning approaches to discover the spatial scale(using multi-scale spatial PCNM variables) at which environmental heterogeneity and other spatially structured processesinfluence tree community composition. We also test the effect of changing grain of the study (i.e. quadrat size) on the variationpartitioning results. Our results indicate that (1) species composition in the Changbai mixed broadleaf-conifer forest wascontrolled mainly by spatially structured soil variation at broad scales, while at finer scales most of the explained variationwas of a spatial nature, pointing to the importance of biotic processes. (2) These results held at all sampling grains. However,reducing quadrat size progressively reduced both spatially and environmentally explained variance. This probably partly reflectsincreasing stochasticity in species abundances, and the smaller proportion of quadrats occupied by each species, when quadratsize is reduced. The results suggest that environmental heterogeneity (i.e. niche processes) and other biotic processes such asdispersal work together, but at different spatial scales, to build up diversity patterns.ZusammenfassungDie ausschlaggebenden Prozesse, welche die Muster in der Zusammensetzung von Pflanzengemeinschaften beeinflussen, sindvon der räumlichen Skala und der Auflösung der Beobachtung abhängig. Deshalb ist es von großem Interesse zu verstehen, welcheRolle die räumliche Skala auf die Verteilungsmuster der Arten hat. In dieser Studie untersuchen wir, wie gut die Umweltvariablen(Topographie und Bodenvariablen) und die räumlichen Variablen die Variation in der Artenzusammensetzung in einem 25 hagroßen gemäßigten Wald im Nordosten Chinas erklären. Wir benutzten neue variable Einteilungsverfahren um die räumlicheSkala festzustellen (durch Nutzung von räumlichen Multi-Skala PCNM Variablen) auf der Umweltheterogenität oder andereräumlich strukturierte Prozesse die Gemeinschaftszusammensetzung der Bäume beeinflussen. Wir testeten auch den Effekt vonVeränderungen in der Genauigkeit der Untersuchung (z. B. die Größe der Quadrate) auf die Variation der Einteilungsergebnisse.Unsere Ergebnisse weisen darauf hin, dass (1) die Artenzusammensetzung im gemischten Laub-Nadelwald in Changbai aufgroßen Skalen vor allem durch räumlich strukturierte Bodenvariation kontrolliert wurde. Auf kleineren Skalen war der größte∗ Corresponding author. Tel.: +86 2483970209; fax: +86 24 8397 0300.E-mail address: hzq@iae.ac.cn (Z. Hao).1439-1791/$ – see front matter © 2011 Gesellschaft für Ökologie. Published by Elsevier GmbH. All rights reserved.doi:10.1016/j.baae.2011.07.008105

Author's personal copyZ. Yuan et al. / Basic and Applied Ecology 12 (2011) 488–495 489Anteil der erklärten Variation von räumlicher Natur und wies auf die Wichtigkeit der biotischen Prozesse hin. (2) Diese Ergebnissegalten für alle Untersuchungsskalen. Wurde jedoch die Fläche der Quadrate progressiv reduziert, verringerte sich sowohl dieerklärte räumliche Varianz als auch die Umweltvarianz. Dies spiegelt wahrscheinlich die zunehmende Stochastizität in derArtenzusammensetzung und den geringeren Anteil der Quadrate wider, die von jeder Art besetzt sind, wenn die Quadratgrößeverringert wird. Die Ergebnisse lassen vermuten, dass die Umweltheterogenität (z. B. Nischenprozesse) und andere biotischeProzesse, wie zum Beispiel die Ausbreitung, zusammen wirken, wenn auch auf unterschiedlichen räumlichen Skalen, um dieDiversitätsmuster zu erzeugen.© 2011 Gesellschaft für Ökologie. Published by Elsevier GmbH. All rights reserved.Keywords: Changbai; PCNM variables; Sampling grain; Soil factors; Spatial scale; Variation partitioningIntroductionDisentangling the mechanisms responsible for the maintenanceof high levels of diversity has been a major goal inecology (Hubbell 2001; Legendre, Borcard, & Peres-Neto2005; Legendre et al. 2009). Currently, ‘niche assembly’and ‘dispersal assembly’ are often cited as importantdeterministic and stochastic processes governing communitycomposition (Hubbell 2001). The niche assemblyview assumes that species coexist by partitioning limitingresources through niche differentiation (Silvertown 2004),whereby species differ in their requirements for a potentiallylimiting resource such as water (Engelbrecht et al. 2007), soil(Tuomisto 2006), or light (Montgomery & Chazdon 2002).Neutral theory represents an extreme dispersal assembly scenario,under which species have equal competitive abilities,and thus species patterns are only generated by dispersal limitation,without invoking species differences in life historytraits (Hubbell 2001).Assessing the relative importance of different factors inexplaining -diversity (i.e. species turnover) requires thedetermination of the relative contribution of environmentalheterogeneity and other spatially structured processes(Borcard & Legendre 2002), and also discovering the spatialscales at which they work (Gilbert & Lechowicz 2004). First,environmental heterogeneity is inherently spatially autocorrelated(Legendre 1993), and thus the use of spatial variableshelps to account for its variation (Borcard, Legendre, &Drapeau 1992; Legendre et al. 2009), and also the spatialscale at which they influence -diversity (Laliberté, Paquette,Legendre, & Bouchard 2009). Second, assuming that dispersalis limited in space (Hubbell 2001), and environmentalheterogeneity has been well sampled (Borcard et al. 1992),the existence of spatially structured variations of -diversityindependent of the environmental fraction can indicate theimportance of dispersal processes (Legendre et al. 2009;Laliberté, Paquette, Legendre, & Bouchard 2009). Variationpartitioning is considered to be one of the most promisingapproaches to distinguish between environmental control ofcommunity composition and the influence of other spatialprocesses (Borcard et al. 1992; Karst, Gilbert, & Lechowicz2005; Smith & Lundholm 2010). Using modern techniques torepresent spatial structures such as Principal Coordinates ofNeighbour Matrices (hereafter PCNM; Borcard & Legendre2002; Dray, Legendre, & Peres-Neto 2006), makes it possibleto include complex spatial structures in variation partitioninganalysis. Moreover, recently developed methods allowdissection of spatial structures in community composition(Laliberté et al. 2009).Spatial scale is a function of both grain (sampling unitsize) and the extent (study area) (Dray et al. 2006). The spatialscale of a study can considerably influence our ability toquantify diversity patterns and their determinants (Lalibertéet al. 2009). A low resolution (i.e. large grain size) canobscure or mitigate important species patterns and environmentalvariations, whereas a high resolution (i.e. small grainsize) can increase the noise in the data and reduce the fractionexplained, making it difficult to draw strong conclusions.Therefore, comparing the results obtained using a constantstudy extent but variable sampling grain can help to choosethe most balanced sampling design, and to interpret the resultsobtained.In the present study, we test the hypothesis that the relativeinfluence of environmental heterogeneity (i.e. nichepartitioning) and other spatially structured biotic processesin the spatial variation of -diversity depends on the spatialscale considered. Specifically, we hypothesize thatenvironmental factors are of major importance at broadscale,while spatially structured biotic processes dominateat fine-scale. To test our hypotheses we (1) quantified theinfluence that the modification of the sampling grain (quadratsize) has on -diversity and (2) assessed the spatial scale(represented by spatial PCNM variables) at which environmentalheterogeneity and other factors generating spatialstructure influence tree diversity patterns at each samplingresolution.Materials and methodsStudy siteThe study area is located at Changbai Mountain NaturalReserve along the border of China and North Korea, extendingfrom 41 ◦ 43 ′ to 42 ◦ 26 ′ N and 127 ◦ 42 ′ to 128 ◦ 17 ′ E(Fig. 1).It is the largest protected temperate forest in the world (Stone2006). The reserve is about 200,000 ha and elevation rangesfrom 740 to 2691 m at the summit of Changbai Mountain106

Author's personal copy490 Z. Yuan et al. / Basic and Applied Ecology 12 (2011) 488–495Fig. 1. Contour map of the 25-ha (500 m × 500 m) Changbai plot.on the Chinese side. There are four main vegetation typesalong the elevational gradient: mixed broadleaf-conifer forest,spruce-fir forest, subalpine birch forest, and tundra (Yang& Li 1985).A 25-ha (500 m × 500 m) plot was established in the summerof 2004 in a broad-leaved Korean pine mixed forest (Hao,Zhang, Song, Ye, & Li 2007). The area has a temperatecontinental climate with long cold winters and warm summers.Mean annual precipitation is approximately 700 mm,most of which occurs from June to September (480–500 mm).Mean annual temperature is 2.8 ◦ C, with a January mean of−13.7 ◦ C, and a July mean of 19.6 ◦ C(Yang & Li 1985).All trees with diameter at breast height (DBH) ≥1cmwere tagged, identified, measured, and their geographic coordinateswere recorded following a standard field protocol(Condit 1998). After that, tree community composition datawere gridded at three different grain sizes for the differentanalyses (10 × 10 m; 20 × 20 m and 50 × 50 m).Environmental data surveyIn October 2007, we sampled soils following the protocolof John, Dalling, Harms, Yavitt, and Stallard (2007) with twoexceptions. First, we used a regular grid of 30 m instead of50 m; second, we took two additional samples at every gridpoint at either 2, 5, or 15 m in a random compass directionfrom the grid point to capture variation in soil nutrients at finerscales. We thus sampled at 967 points (Five sample pointscould not be sampled because of large roots and stumps) (seeAppendix A: Fig. 1).The volumetric soil water content (%) was measured ateach sample location at a depth of 20 cm using a TDR probe.Recognisable plant fragments and leaves were then removedfrom the soil surface and three cores were taken to 10 cmdepth from within a 0.2-m area around the sample locationusing a 5-cm diameter cylinder. The three cores were mixedbefore further processing.Samples were air-dried at ambient laboratory temperatureand then sieved through a 2-mm mesh to remove rootsand stones. Eight soil properties were determined mainlyaccording to Lu (1999). Soil pH in water (1:1 soil:solutionratio) was determined using a Beckman glass electrode. Soilorganic carbon was determined colorimetrically followingdichromate oxidation. Available N was alkali dispelled by1 mol NaOH L −1 . Total N was estimated colorimetrically onthe KCl extracts, using the Kjeldahl method. Available Pwas extracted following the method of Mehlich 1 (Nelson,Mehlich, & Winters 1953). Total P was determined by molybdatecolorimetry, after digestion in H 2 SO 4 –HClO 4 ; AvailableK was extracted with 1 mol L −1 NH 4 Ac with detection byatomic absorption spectrometery (AAS). Total K was determinedby digestion in hydrofluoric acid with detection byatomic absorption spectrometery.Geostatistical methods were used to generate soil predictionsfor each tree sampling quadrat at each of the threesampling resolutions. Kriging maps for each soil variablewere performed using the adjusted parameters of the theoreticalsemivariogram. For detailed descriptions of the approachsee Gallardo (2003).Spatial dataWe used the PCNM approach to introduce space as anexplanatory variable. This approach is based on the calculationof a principal coordinate analysis (PCoA) from atruncated matrix of Euclidean distances among samplingunits (Borcard & Legendre 2002). The resulting eigenvectorsassociated with positive eigenvalues take the form oforthogonal sine waves of decreasing wavelength and can beused to model spatial structures at different scales. For each107

Author's personal copyZ. Yuan et al. / Basic and Applied Ecology 12 (2011) 488–495 49110 × 10 m, 20 × 20 m, and 50 × 50 m quadrat, PCNMs wererespectively computed across the 2500, 625 and 100 points ofthe spatial grid (Borcard & Legendre 2002). First, the x and ycoordinates from each plot were centered (by subtracting themean x and y values from each coordinate) and a Euclideandistance matrix between sampling sites was created. Second,this distance matrix was truncated using the largest distancein a minimum spanning tree (i.e. largest distance that keeps allthe points connected). The threshold distance used to truncatethe distance matrix defines the finest spatial scales at whichspatial patterns can be detected (Borcard & Legendre 2002).The threshold distances used were 50, 20 and 10 m in the50 × 50 m, 20 × 20 m and 10 × 10 m spatial resolution analyses,respectively. Then, a PCoA was applied to the truncateddistance matrix and the eigenvectors associated with positiveeigenvalues were retained as PCNM variables. Finally, 54,324 and 1274 spatial PCNM variables were created in the50 × 50 m, 20 × 20 m and 10 × 10 m analyses, respectively.Statistical proceduresWe assessed the relative weight of environmental heterogeneityand spatial variables using variation partitioningwith Redundancy Analysis (RDA) (Borcard et al. 1992;Smith & Lundholm 2010). The analyses were performedfor each sampling resolution (50 m × 50 m, 20 m × 20 mand 10 m × 10 m). In each situation, the analyses involvethe use of one response matrix (species composition) andthree explanatory data matrices: topography, soil factorsand spatial variables (UTM-coordinates and a set of forwardselected PCNM variables). The UTM-coordinates wereincluded because they represent linear trends that require alarge number of PCNM variables to be modeled (Lalibertéet al. 2009). Prior to the analyses, species composition wasHellinger transformed to reduce the weight of the most abundantspecies (Legendre & Gallagher 2001). Since strongcollinearity was detected between several soil variables, onlyfive (water content, pH, available N, P and K) were used in theanalyses (Blanchet, Legendre, & Borcard 2008). After that,a forward selection was run on soil and topographic explanatoryvariables to select subsets that explained significant(P ≤ 0.05 after 999 random permutations) effects on variationin community composition (Borcard & Legendre 2002).Forward selection of PCNM variables was performed consideringtwo criteria (the total fraction of variance explainedusing all the PCNMs together and the significance, P < 0.05,of each individual PCNM variable; Blanchet et al. 2008).In a further approach, we also tested the contribution of thetopography and soil factors on four spatial sub-models: thebroad scale linear trend (x- and y-UTM coordinates), PCNMscale (all PCNM variables selected), medium (medium-scalePCNM variables) and fine-scale (fine-scale PCNM variables)models. The fitted values of each spatial fraction were usedas response variable to discover the scale-specific contributionof the topographic and soil variables. As pointed out byTable 1. Total number of trees with DBH ≥1 cm and species richnessfound in the Changbai plot according to the different grain sizeused (50 m × 50 m, 20 m × 20 m and 10 m × 10 m).Spatial grainsizeTotal number of treesSpecies richnessMean ± SD Range Mean ± SD Range50 m × 50 m 369.2 ± 57.3 268–549 20.2 ± 3.1 13–2920 m × 20 m 59.1 ± 14.9 27–126 11.5 ± 2.3 5–1910 m × 10 m 14.8 ± 5.9 0–46 6.2 ± 1.8 0–13Laliberté et al. (2009), the terms of “broad”, “medium” or“fine”-scale have no absolute meaning but they are relativeto the sampling extent and grain.All the analyses were performed using the R language (RDevelopment Core Team 2009). Geostatistical analyses wereperformed by using the package “geoR” (Ribeiro & Diggle2001); PCNM variables were created and the forward selectionwas performed using “PCNM” (Blanchet et al. 2008) and“packfor” (Dray 2005) packages respectively; Variation partitioninganalyses were preformed using the “vegan” package(Oksanen, Kindt, Legendre, & O ′ Hara 2010).ResultsGeneralA total number of 36,919 presences of 52 species werefound in the 25 ha area studied in the Changbai Mountain.The most abundant species was represented by 7705 individualsin the entire area, and the least abundant seven specieswere each represented by a single individual. The four mostabundant species, in descending order of abundance, wereCorylus mandshurica, Acer mono, Acer pseudosieboldianumand Acer barbinerve. Species richness and the total number oftrees found in each quadrat varied depending on the samplinggrain (Table 1). The topography in this study area was relativelylow, but the soil factors were spatially autocorrelatedand displayed considerable variations according to differentgrain (quadrat size) (Table 2; see Appendix A: Fig. 2).Multi-scale variation of species compositionThe variation partitioning results across differing spatialgrain sizes (different quadrat sizes) showed that the topographic(a+d+f+g),soil (b+d+e+g)andspatial fractions(c+e+f+g) explained in species composition decreasesas the size of the sampling units decreases (Table 3;Figs. 2 and 3).The number of topographic and soil variables selected bythe stepwise procedure was different depending on the spatialgrain size (Table 3). Elevation and slope were selectedin all the analyses while convexity was only selected in the108

Author's personal copy492 Z. Yuan et al. / Basic and Applied Ecology 12 (2011) 488–495Table 2. Descriptive statistics of the topographic and soil variables used as predictor variables in the variation partitioning results of speciescomposition in the Changbai mountain plot. The Mean value (±Standard Deviation) of each variable, at each spatial scale studied (grain sizeof 50 m × 50; 20 m × 20 and 10 m × 10 m), is shown.Grain size 50 m × 50 m 20 m × 20 m 10 m × 10 mMean ± SD Range Mean ± SD Range Mean ± SD RangeElevation (m) 803.2 ± 3.5 794.5–808.8 803.2 ± 3.5 793.1–809.4 803.2 ± 3.5 793.0–809.3Slope (degrees) 3.2 ± 2.0 0.6–10.8 3.2 ± 2.4 0.3–16.1 3.2 ± 2.0 0.5–12.1Convexity (m) 0.0 ± 0.1 −0.4–0.3 −0.0 ± 0.3 −1.5–1.3 −0.0 ± 0.2 −0.7–0.6Soil moisture (%) 40.1 ± 6.67 28.7–57.6 40.0 ± 6.86 14.1–60.7 40.1 ± 6.62 27.3–61.5pH 5.45 ± 0.15 5.13–5.87 5.45 ± 0.20 4.48–6.89 5.45 ± 0.16 4.93–6.05Organic matter (g kg −1 ) 162.3 ± 54.9 81.6–314.5 162.9 ± 56.4 62.6–431.7 163.5 ± 53.8 625.2–351.7Available N (mg kg −1 ) 502.2 ± 71.1 369.2–666.3 502.4 ± 75.9 28.2–753.9 503.3 ± 71.1 313.7–698.6Available P (mg kg −1 ) 8.44 ± 1.45 5.13–11.97 8.46 ± 1.74 3.23–24.44 8.45 ± 1.40 4.73–14.69Available K(mg kg −1 ) 261.0 ± 44.8 161.3–337.3 258.6 ± 49.1 102.6–407.3 260.4 ± 44.5 156.3–340.0Total N (g kg −1 ) 6.35 ± 1.67 3.22–9.59 6.37 ± 1.78 2.54–12.41 6.38 ± 1.68 2.66–11.55Total P(g kg −1 ) 1.26 ± 0.35 0.68–2.01 1.26 ± 0.36 0.58–2.50 1.26 ± 0.35 0.58–2.11Total K (g kg −1 ) 16.53 ± 1.20 13.21–19.21 16.53 ± 1.46 7.83–21.22 16.51 ± 1.29 10.13–20.0110 m × 10 m spatial grain analysis. The five soil variables(water content, pH, available N, P and K) were selected in the20 m × 20 m and 10 m × 10 m spatial grain analysis, while inthe 50 m × 50 m spatial grain analysis, available P was notselected. Spatial variables selected also depended on the resolution.In the 50 m × 50 m analysis 22 spatial variables wereselected (x- and y-UTM coordinates, 8 medium-scale and12 fine-scale PCNMs); 119 spatial variables were selectedin the 20 m × 20 m analysis (x- and y-UTM coordinates, 18medium-scale and 99 fine-scale PCNMs); Finally, In the10 m × 10 m analysis 86 spatial variables were retained (xandy-UTM coordinates, 25 medium-scale and 59 fine-scalePCNMs). In the 50 m × 50 m spatial grain analyses, mediumscalePCNMs represent patterns ranging from around 150 mto 300 m. The smallest scale represented by the mediumscalevariables in the 20 m × 20 m spatial grain analyses wasaround 120 m. Finally, in the 10 m × 10 m grain size analysesmedium-scale patterns ranged from around 100 to 300 m.The fraction of variation in species composition explainedby the different variables considered was dependent on thesampling resolution (Figs. 2 and 3). 55% of the variation inspecies composition was explained in the 50 m × 50 m spatialgrain analysis, 39% of the total variation was explainedin the 20 m × 20 m analysis and only 17% of species compositionvariation in the 10 m × 10 m analysis. In general, thespatial variables explained a higher fraction of variance thanthe topographic and soil ones, and most of the effects of thetopographic variation were encompassed by correlated variationin soil conditions, indicating that soil variables weremore important than the topographic variables (Fig. 3).The division of the total spatial fraction (c+e+f+g) atdifferent spatial scales showed that the importance of thecoordinates against the PCNM variables decreased as thesampling grain decreased (from 50 m × 50 m to 10 m × 10 m;Fig. 2). Most of the spatially structured fraction of variationexplained by the coordinates was shared with the topographicand soil variables. Only 21.9% of this fraction was pure inthe 50 m × 50 m grain analysis, 24.7% in the 20 m × 20 manalysis, and 19.4% in the 10 m × 10 m spatial grain analysis.The division of the PCNM spatially explained fractioninto medium- and fine-scale patterns showed that these twofractions varied depending on the grain size considered,Table 3. Results of the forward selection procedure applied to select the topographic and soil variables that explain a significant fraction ofvariation in species composition (P < 0.05) in each grain size.Grain size 50 m × 50 m 20 m × 20 m 10 m × 10 mVariable R 2 P R 2 P R 2 PElevation 0.081 0.001 0.034 0.001 0.013 0.001Slope 0.115 0.001 0.048 0.001 0.025 0.001Convexity – – – – 0.001 0.001Water content 0.296 0.001 0.14 0.001 0.064 0.001pH 0.034 0.001 0.012 0.001 0.005 0.001Available N 0.047 0.001 0.022 0.001 0.010 0.001Available P – – 0.002 0.045 0.001 0.002Available K 0.016 0.005 0.006 0.001 0.004 0.001109

Author's personal copyZ. Yuan et al. / Basic and Applied Ecology 12 (2011) 488–495 493Fig. 2. Variation partitioning results of species composition against the topographic, soil and spatial variables at three different grain sizes:(A) 50 m × 50 m; (B) 20 m × 20 m; (C) 10 m × 10 m. The Venn diagrams indicate the fractions of variation explained by the topographic,soil and spatial variables. (a) = pure topographic, (b) = pure soil, (c) = pure spatial, (d) = shared between topography and space but not bysoil, (e) = shared between soil and space but not by topography, (f) = shared between topography and soil but not by space, and (g) = sharedbetween topography, soil and space. The pie charts indicate the division of the total spatial fraction (c+e+f+g)into different spatial scales.The fraction of variation explained by each scale is expressed as a percentage of the total spatial fraction. The fraction of spatial variationattributable to the PCNM variables is divided into two scales. The terms “medium” and “fine” have no meaning, rather are relative to thesampling grain and extent. Values in bold are statistically significant after 999 permutations.although the fine-scale PCNM fraction was higher in general(Figs. 2 and 3). The analyses also showed that the fractionexplained by the topographic and soil variables contributedless to these two PCNM fractions (medium- and fine-scale)than to the coordinate’s fraction (Figs. 2 and 3).DiscussionFig. 3. Fractions of variation explained (expressed in percentages)in species composition by the topographic, soil and spatial variables(A). Relative contributions (as a percentage of the total spatialfraction) of the quadrat coordinates, PCNM-broad and PCNM-finescale variables (B). The values are shown for the different spatialgrain sizes used in the analyses (50 m × 50 m, 20 m × 20 m and10 m × 10 m).The results clearly show that the selection of the spatialgrain (i.e. quadrat size) strongly influences the abilityto explain variations in species composition, emphasizingthe importance of considering the spatial scale in ecologicalstudies (Dungan et al. 2002). The variation in the amountof variance explained in species composition across spatialscales can be due to three main reasons. First, the reductionof spatial grain size, maintaining the extent constant, impliesan increase in the number of samples, increasing the variancein species abundances but also the noise in the data. Thefact is that several species can be present in a small numberof plots, making it difficult to model their distribution withordination techniques. Second, although we used an extensivesampling design to account for soil heterogeneity, theuse of geostatistical techniques implies a potential error insoil quantification that can increase with the reduction of thesampling grain. Finally, the reduction of the spatial grain sizecan facilitate the quantification of fine-scale spatial patterns,which can potentially represent the influence of biotic processes(Gilbert & Lechowicz 2004; Laliberté et al. 2009) suchas dispersal (Hubbell 2001; Kolb & Diekmann 2004).110

Author's personal copy494 Z. Yuan et al. / Basic and Applied Ecology 12 (2011) 488–495Regarding the spatial grain size, the results indicate thatthe topographic and soil heterogeneity are mainly structuredat broad scales (x- and y-UTM coordinates and medium-scalePCNMs) indicating that community patterns are partiallycreated by a broad- and medium-scale spatially structuredenvironmental heterogeneity (Gilbert & Lechowicz 2004;Laliberté et al. 2009). Our results also show that soil factorsexplain a higher fraction of variation in species compositionthan the topographic variables, pointing to the importance ofsoil factors in forested ecosystems (Tuomisto 2006). It canbe due to the gentle topography of the Changbai mountain,but also because the influence of the topographic heterogeneityis partially included in the soil fraction. Overall, ourresults clearly indicate the importance of environmental heterogeneityat broad spatial scales and supports niche theory(Silvertown 2004).Previous studies have reported that environmental heterogeneitywas less important than other factors in structuringdiversity in the Changbai forests (Wang et al. 2008; Zhang,Zhao, & Gadow 2010). The two earlier studies did not modeltree community composition, but instead focused on speciesabundance and diversity measures. Also, Wang et al. (2008)did not include soil parameters, rather focused on topographyalone. The results of the current study show that soil variationwas a much stronger predictor of variation in communitycomposition than topography. Similarly, studies carried outin different forests have reported that environmental heterogeneityis not the main source of variation in species patterns(Palmer 1991). Our results point to the scale-dependenceof such conclusions, since in reducing the sampling grainthe proportion of spatial relative to environmental variationincreases. One interpretation for this pattern is that it reflectsthe influence of unmeasured factors such as light variability(Montgomery & Chazdon 2002) or leaf litter coverage (Gazol& Ibáňez 2010) that can be spatially structured and thusinduce spatial dependence in species composition (Borcardet al. 1992; Legendre et al. 2009).However, it can be also argued that the high fraction ofvariance explained by the spatial variables alone, and theirhigher importance at small scales, can indicate the importanceof biotic processes in explaining diversity variation(Gilbert & Lechowicz 2004; Laliberté et al. 2009), emphasizingthe importance of dispersal (Kolb & Diekmann 2004).These results are coincident with other studies that have foundthat environmental heterogeneity dominates species compositionat broader scales, while the possible influence of bioticstochastic processes increases at small scales (Laliberté et al.2009; Gazol & Ibáňez 2010).The high proportion of unaccounted variation may beexplained by a variety of factors, such as non-spatiallystructured biological or environmental factors that were notmeasured in this study, or historical effect (Graae, Økland,Petersen, Jensen, & Fritzbøger 2004). Another feasibleexplanation may be random processes, with a theoreticalconnection to the neutral theory of macroecology. This explanationassumes that the dynamics of populations are primarilydriven by ecological drift, i.e. random mortality, and dispersal,with or without limitation, and are not habitat dependent(Legendre et al. 2009). Usually dispersal processes are spatiallystructured (Hubbell 2001), and thus their influence isgenerally included in the spatial PCNM fraction of variation,while the ecological drift is included in the unexplainedfraction (Legendre et al. 2009).ConclusionAnswering questions about how -diversity is maintainedin a community depends on our ability to isolate the contributionsof the different process that affect diversity. Here,we highlight the effects of topographic variables and biologicallyimportant soil factors in structuring plant communities,as well as non habitat-related spatially structured processesusing the PCNM method in a natural temperate forest. Ourresults indicate that the spatial resolution of study (i.e. grainsize) strongly influences our ability to detect the spatialstructure of plant diversity and the factors that influence it.Similarly, the influence of environmental heterogeneity (i.e.niche theory) and the importance of stochastic processes suchas dispersal (i.e. neutral theory) are not mutually exclusive,but our results indicate that they can act at different spatialscales.AcknowledgementsThis study was sponsored by the Knowledge InnovationProgram of the Chinese Academy of Sciences (KZCX2-EW-401 and KSCX2-EW-Z-5), the National Key TechnologiesR&D Program of China (2008BAC39B02), and NationalNatural Science Foundation of China (31061160188 and40971286). Antonio Gazol has the financial support ofthe ERMOS programme (Co-founded by the Marie CurieActions)—ERMOS14. We thank Pierre Legendre for technicalhelp in variation partitioning analysis. We also thankTom McRae for his assistance with English language andgrammatical editing of the manuscript.Appendix A. Supplementary dataSupplementary data associated with this articlecan be found, in the online version, atdoi:10.1016/j.baae.2011.07.008.ReferencesBlanchet, F. G., Legendre, P., & Borcard, D. (2008). Forward selectionof explanatory variables. Ecology, 89, 2623–2632.Borcard, D., & Legendre, P. (2002). All-scale spatial analysis ofecological data by means of principal coordinates of neighbourmatrices. Ecological Modelling, 153, 51–68.111

Author's personal copyZ. Yuan et al. / Basic and Applied Ecology 12 (2011) 488–495 495Borcard, D., Legendre, P., & Drapeau, P. (1992). Partialling outthe spatial component of ecological variation. Ecology, 73,1045–1055.Condit, R. (1998). Tropical forest census plots: Methods and resultsfrom Barro Colorado island, Panama and a comparison withother plots. Berlin: Springer.Dray, S. (2005). Pack for: Forward selected with multivariate Y bypermutation under reduced model. Lyon: Laboratoire Biométrieet Biologie Évolutive.Dray, S., Legendre, P., & Peres-Neto, P. (2006). Spatial modelling:A comprehensive framework for principal coordinate analysis ofneighbour matrices (PCNM). Ecology Modelling, 196, 483–493.Dungan, J. L., Perry, J. N., Dale, M. R. T., Legendre, P., Citron-Pousty, S., Fortin, M. J., et al. (2002). A balanced view of scalein spatial statistical analysis. Ecography, 25, 626–640.Engelbrecht, B. M. J., Comita, L. S., Condit, R., Kursar, T. A.,Tyree, M. T., et al. (2007). Drought sensitivity shapes speciesdistribution patterns in tropical forests. Nature, 447, 80–82.Gallardo, A. (2003). Spatial variability of soil properties in a floodplainforest in Northwest Spain. Ecosystems, 6, 564–576.Gazol, A., & Ibáňez, R. (2010). Variation of plant diversity in atemperate unmanaged forest in northern Spain: Behind the environmentaland spatial explanation. Plant Ecology, 207, 1–11.Gilbert, B., & Lechowicz, M. J. (2004). Neutrality, niches, anddispersal in a temperate forest understory. Proceedings of theNational Academy Sciences, 101, 7651–7656.Graae, B. J., Økland, R. H., Petersen, P. M., Jensen, K., & Fritzbøger,B. (2004). Influence of historical, geographical and environmentalvariables on understorey composition and richness in Danishforests. Journal of Vegetation Science, 15, 465–474.Hao, Z. Q., Zhang, J., Song, B., Ye, J., & Li, B. H. (2007). Verticalstructure and spatial associations of dominant tree species in anold growth temperate forest. Forest Ecology and Management,252, 1–11.Hubbell, S. P. (2001). The unified neutral theory of biodiversity andbiogeography. Princeton: Princeton University Press.John, R., Dalling, J. W., Harms, K. E., Yavitt, J. B., Stallard, R. F.,et al. (2007). Soil nutrients influence spatial distributions of tropicaltree species. Proceedings of the National Academy Sciences,104, 864–869.Karst, J., Gilbert, B., & Lechowicz, M. J. (2005). Fern communityassembly: The roles of chance and the environment at local andintermediate scales. Ecology, 8, 2473–2486.Kolb, A., & Diekmann, M. (2004). Effects of environment, habitatconfiguration and forest continuity on the distribution offorest plant species. Journal of Vegetation Science, 15, 199–208.Laliberté, E., Paquette, A., Legendre, P., & Bouchard, A. (2009).Assessing the scale-specific importance of niches and otherspatial processes on diversity: A case study from a temperateforest. Oecologia, 159, 377–388.Legendre, P. (1993). Spatial autocorrelation: Trouble or newparadigm? Ecology, 74, 1659–1673.Legendre, P., Borcard, D., & Peres-Neto, P. (2005). Analyzing betadiversity: Partitioning the spatial variation of community compositiondata. Ecological Monographs, 75, 435–450.Legendre, P., & Gallagher, E. D. (2001). Ecologically meaningfultransformations for ordination of species data. Oecologia, 129,271–280.Legendre, P., Mi, X. C., Ren, H. B., Ma, K. P., Yu, M. J., et al.(2009). Partitioning beta diversity in a subtropical broad-leavedforest of China. Ecology, 90, 663–674.Lu, R. K. (1999). Analytical methods of soil agrochemistry. BeijingChina Agricultural Science and Technology Press. (in Chinese)Montgomery, R. A., & Chazdon, R. L. (2002). Light gradient partitioningby tropical tree seedlings in the absence of canopy gaps.Oecologia, 131, 165–174.Nelson, W. L., Mehlich, A., & Winters, E. (1953). The development,evaluation, and use of soil tests for phosphorus availability.Agronomy, 4, 153–188.Oksanen, J., Kindt, R., Legendre, P., & O ′ Hara, R. B. (2010). Vegan:Community Ecology Package. Version 1.17-2.Palmer, M. W. (1991). Patterns of species richness among NorthCarolina hardwood forests: Tests of two hypotheses. Journal ofVegetation Science, 2, 361–366.R Development Core Team (2009). R: A language and environmentfor statistical computing. Vienna, Austria.Ribeiro, P. J., Jr., & Diggle, P. J. (2001). geoR: A package forgeostatistical analysis. RNews, 1, 15–18.Silvertown, J. (2004). Plant coexistence and the niche. Trends inEcology & Evolution, 19, 605–611.Smith, T. W., & Lundholm, J. T. (2010). Variation partitioning as atool to distinguish between niche and neutral processes. Ecography,33, 648–655.Stone, R. (2006). A threatened nature reserve breaks down Asianborders. Science, 313, 1379–1380.Tuomisto, H. (2006). Edaphic niche differentiation among Polybotryaferns in western Amazonia: Implications for coexistenceand speciation. Ecography, 29, 273–284.Wang, X. G., Hao, Z. Q., Ye, J., Zhang, J., Li, B. H., & Yao, X. L.(2008). Spatial pattern of diversity in an old-growth temperateforest in Northeastern China. Acta Oecologica, 33, 345–354.Yang, H., & Li, F. (1985). Distribution patterns of dominant treespecies on northern slope of Changbai Mountain. Research ForestEcosystem, 5, 1–14 (in Chinese).Zhang, C. Y., Zhao, X. H., & Gadow, K. (2010). Partitioningtemperate plant community structure at different scales. ActaOecologica, 36, 306–313.112

Spatial distributions of species in an old-growthtemperate forest, northeastern China1011Xugao Wang, Ji Ye, Buhang Li, Jian Zhang, Fei Lin, and Zhanqing HaoAbstract: Studying spatial distributions of species can provide important insights into processes and mechanisms thatmaintain species richness. We used the relative neighborhood density U based on the average density of conspecific speciesin circular neighborhoods around each species to quantify spatial distributions of species with ‡10 individuals in afully mapped 25 ha temperate plot at Changbaishan, northeastern China. Our results show that spatial aggregation is adominant pattern of species in the Changbaishan temperate forests. However, the percentage of significantly aggregatedspecies decreases with spatial scale, especially for rare species. Rare species are more aggregated than intermediate andcommon species. The aggregation intensity declines with increasing size class (diameter at breast height), i.e., species becomemore regularly spaced as species grow, which is consistent with the predictions of self-thinning and Janzen–Connellspacing effects. Species functional traits (canopy layer, seed dispersal ability, shade tolerant, etc) also havea significant effecton the spatial distributions of species. Our results partially conform to the prediction that better dispersal reduces aggregation.Consequently, dispersal limitation, self-thinning, Janzen–Connell spacing effects, and habitat heterogeneity mayprimarily contribute to spatial distributions of species in the temperate forests.Résumé : L’étude de la distribution spatiale des espèces peut fournir d’importantes informations sur les processus et lesmécanismes qui maintiennent la richesse en espèces. Nous avons utilisé la densité relative des voisins U basée sur la densitémoyenne des individus de la même espèce présents dans un rayon autour de chaque espèce de façon à quantifier ladistribution spatiale des espèces comportant plus de 10 individus sur une superficie cartographiée de 25 ha dans une forêttempérée de Changbaishan, au nord-est de la Chine. Nos résultats montrent que le regroupement spatial est un patron dominantdes espèces dans les forêts tempérées de Changbaishan. Cependant, le pourcentage d’espèces significativement regroupéesdiminue avec l’échelle spatiale, particulièrement pour les espèces rares. Les espèces rares sont davantageregroupées que les espèces communes et intermédiaires. L’intensité de regroupement diminue avec l’augmentation de laclasse de diamètre à hauteur de poitrine, de telle sorte que l’espacement entre les individus d’une espèce devient plus régulierà mesure qu’ils grossissent, ce qui est cohérent avec les prédictions de l’autoéclaircie et les effets d’espacement deJanzen–Connell. Les caractéristiques fonctionnelles des espèces (strate de la canopée, capacité de dispersion des graines,tolérance à l’ombre, etc.) ont aussi des effets significatifs sur la distribution spatiale des espèces. Nos résultats supportentpartiellement l’hypothèse qu’une meilleure dispersion diminue le regroupement. En conséquence, les contraintes de dispersion,l’autoéclaircie, les effets d’espacement de Janzen–Connell et l’hétérogénéité de l’habitat peuvent contribuer de façonimportante à la distribution spatiale des espèces dans les forêts tempérées.[Traduit par la Rédaction]IntroductionA major objective of ecological research is to infer theunderlying processes or mechanisms by analyzing the spatialdistributions of species. Although substantial efforts havebeen made in explaining the observed distributions of species,there are still important challenges, in large part becauseidentical spatial distributions of species may begenerated by several different processes. For example, aggregationdistributions in species, a widespread pattern observedin nature (He et al. 1997; Condit et al. 2000; Hao etal. 2007), may be broadly attributed to two major, yet contrasting,effects of habitat heterogeneity and dispersal limitation,but the relative contributions of these two effects aredifficult to quantify. These two effects represent two majorbiodiversity theories, niche versus neutral, in biodiversitystudies and are fundamental to understanding community assemblages(Hubbell 2001; Chase 2005; Gaston and Chown2005). Models based on neutral theory assume that speciesare functionally identical and drift randomly in abundanceuntil they vanish and can form patterns of distribution andabundance similar to those found in nature (Mouquet andLoreau 2003; Chave 2004; Purves and Pacala 2005). However,ample evidence indicates that species are not equivalent;species-specific differences in their functional traitsand ecological strategies affect the spatial distributions anddynamics of species (Peters 2003; Lortie et al. 2004; Stolland Newbery 2005; Condit et al. 2006; Murrell 2009). Forexample, seed dispersal ability affects distributions of species,with well-dispersed species being less aggregated thanReceived 18 August 2009. Accepted 1 March 2010. Published on the NRC Research Press Web site at cjfr.nrc.ca on 7 May 2010.X. Wang, J. Ye, B. Li, J. Zhang, F. Lin, and Z. Hao. 1 Institute of Applied Ecology, Chinese Academy of Science, P.O. Box 417,Shenyang, 110016, China.1 Corresponding author (e-mail: hzq@iae.ac.cn).Can. J. For. Res. 40: 1011–1019 (2010) doi:10.1139/X10-056 Published by NRC Research Press113

1012 Can. J. For. Res. Vol. 40, 2010Fig. 1. Location and contour map of the 25 ha (500 m 500 m) Changbaishan temperate plot.poorly dispersed species (Condit et al. 2000; Li et al. 2009).Additionally, as the classical Janzen–Connell spacing effectpredicts, spatial distributions of adult trees would becomemore regular than those of juveniles because of the differentialattack rates between adults and juveniles by distance/frequency-responsivepathogens or herbivores (Janzen 1970;Connell 1971). However, lack of effective quantitativemethods and high quality data of species distributions atproper spatial and temporal scales has also contributedmuch confusion and controversy.Current knowledge on spatial distributions of species andunderlying mechanisms is mostly derived from tropical rainforests (but see Li et al. (2009) for subtropical forests). Consideringthe complexity of these hyperdiverse forests, it maybe more realistic to expect these underlying mechanisms todiffer among species, just as the resulting spatial distributionsof species differ (He et al. 1997). In addition, thereare some questions that cannot sufficiently be addressed bystudying tropical forests alone. An important question iswhether these theories or hypotheses that have been developedto explain the hyperdiverse tropical forests may alsobe applicable to other forests, such as temperate forests.Although there is a long tradition of analyzing spatial distributionsof species in temperate forests (e.g., Kenkel 1988;Duncan 1991), most of these studies have focused on fewdominant or overstory species in these forests (e.g., Hao etal. 2007) and have been conducted at small scales (£1ha)(e.g., Kubota 2006; Gravel et al. 2008; Mori and Komiyama2008). Consequently, the community-wide analysis of spatialdistributions of species, the species–habitat association,diversity pattern, and the underlying mechanisms of speciescoexistence are not well understood in temperate forests (butsee Canham et al. 2006; Papaik and Canham 2006).Motivated by these challenges and for the ultimate understandingof the mechanisms of species coexistence, the ChineseAcademy of Sciences, in collaboration with the Centerfor Tropical Forest Science of the Smithsonian Tropical ResearchInstitute, has recently initiated an ambitious largescale,long-term Forest Dynamics and Diversity Plots network.The China network has been designed to establishfive 20–25 ha plots along the latitudinal gradient from northto south China. This study reports the spatial distributions ofspecies in the 25 ha Changbaishan (CBS) plot. The CBSplot is the northernmost plot of the China Forest BiodiversityMonitoring Network (www.cfbiodiv.org), which is alsothe largest forest plot in a temperate region.The data from the temperate CBS plot provide a uniqueopportunity to address how the temperate forest communityis spatially structured and the implication of the spatialstructure in maintenance of the species assemblage. The specificobjectives are (i) to analyze the spatial distributions ofconspecific species in the CBS plot and compare spatial distributionsin hyperdiverse tropical forests with those in temperateforests, (ii) to investigate the change in spatialdistributions of species with spatial scale, (iii) to examinewhether the Janzen–Connell spacing effect, i.e., species becomemore regularly spaced as trees grow, is also in operationin this temperate forest, and (iv) to test whetherdifferent functional groups (abundance, canopy, shade tolerance,dispersal mode, etc.) have an effect on the spatial distributionsof species. It is expected that this study willprovide important insights into the possible mechanismsthat structure and maintain the assemblage of the tree speciesof the temperate forests and also serve as a foundationfor subsequent studies.Materials and methodsStudy areaThe study site is located in the Changbai Nature Reserve,which was established along the border of China and NorthKorea extending from 127842’ to 128817’E and from 41843’to 42826’N. The reserve was first established in 1960 and isone of the largest biosphere reserves in China. It has beenspared from logging and other severe human disturbancesdue to its remoteness and inaccessibility to the general publicbefore establishment of the reserve. Changbai Nature Reservejoined the World Biosphere Reserve Network underthe UNESCO Man and the Biosphere Programme in 1980.The reserve is about 200 000 ha in size with an elevationranging from 740 to 2691 m at the summit of ChangbaiMountain on the Chinese side. Changbai Mountain is thehighest mountain in northeastern China and is the head ofthree large rivers (the Songhua, Yalu, and Tumen) in thenortheastern provinces. Topographic features differ on thefour slopes of the mountain, with the northern slope being114Published by NRC Research Press

Wang et al. 1013Table 1. Functional traits for species with ‡10 individuals in the Changbaishan plot.Species Family No. of individuals Canopy layer Shade tolerance Dispersal modeAcer barbinerve Aceraceae 3910 Understory Shade tolerant WindAcer ginnala Aceraceae 108 Understory Shade tolerant WindAcer mandshuricum Aceraceae 251 Midstory Shade tolerant WindAcer mono Aceraceae 6609 Midstory Shade tolerant WindAcer pseudo-sieboldianum Aceraceae 5984 Midstory Shade tolerant WindAcanthopanax senticosus Araliaceae 35 Understory Shade tolerant GravityAcer tegmentosum Aceraceae 846 Midstory Shade tolerant WindAcer triflorum Aceraceae 276 Midstory Shade tolerant WindAcer tsckonoskii Aceraceae 39 Understory Shade tolerant WindAralia elata Araliaceae 12 Understory Shade tolerant GravityBetula costata Betulaceae 16 Overstory Light demanding WindBetula platyphylla Betulaceae 96 Midstory Light demanding WindCerasus maximowiczii Rosaceae 18 Understory Light demanding AnimalCorylus mandshurica Betulaceae 7833 Understory Shade tolerant GravityCrataegus maximouwiczii Rosaceae 121 Understory Shade tolerant GravityEuonymus alatus Celastraceae 38 Understory Shade tolerant GravityEuonymus macropterus Celastraceae 10 Understory Shade tolerant GravityEuonymus pauciflorus Celastraceae 37 Understory Shade tolerant GravityFraxinus mandshurica Oleaceae 681 Overstory Midtolerant WindFraxinus rhynchophylla Oleaceae 10 Midstory Shade tolerant WindLonicera monantha Caprifoliaceae 27 Understory Shade tolerant GravityMaackia amurensis Leguminosae 753 Midstory Midtolerant GravityMalus baccata Rosaceae 106 Midstory Shade tolerant GravityPhellodendron amurense Rutaceae 60 Midstory Light demanding GravityPhiladelphus schrenkii Saxifragaceae 470 Understory Shade tolerant GravityPinus koraiensis Pinaceae 2468 Overstory Midtolerant AnimalPopulus davidiana Salicaceae 27 Midstory Light demanding WindPopulus ussuriensis Salicaceae 30 Overstory Light demanding WindPrunus padus Rosaceae 515 Midstory Shade tolerant GravityPyrus ussuriensis Rosaceae 74 Midstory Light demanding GravityQuercus mongolica Fagaceae 926 Overstory Light demanding AnimalRhamnus davarica Rhamnaceae 26 Understory Shade tolerant GravityRhamnus ussuriensis Rhamnaceae 118 Midstory Shade tolerant GravitySambucus williamsii Caprifoliaceae 19 Understory Light demanding GravitySorbus alnifolia Rosaceae 19 Understory Shade tolerant GravitySyringa reticulata Oleaceae 1598 Midstory Light demanding WindTilia amurensis Tiliaceae 2927 Overstory Shade tolerant GravityTilia mandshurica Tiliaceae 410 Overstory Shade tolerant GravityUlmus japonica Ulmaceae 1109 Overstory Midtolerant WindUlmus laciniata Ulmaceae 192 Midstory Midtolerant WindViburnum burejaeticum Caprifoliaceae 23 Understory Shade tolerant GravityViburnum sargenti Caprifoliaceae 43 Understory Shade tolerant Gravityrelatively moderate (average slope

1014 Can. J. For. Res. Vol. 40, 2010Table 2. Spatial distributions of species in the Changbaishan plot as measured by U.Aggregated RandomAnnulus (m) Abundant (n = 13) Intermediate (n = 12) Rare (n = 17) Total (n = 42) Abundant (n = 13) Intermediate (n = 12) Rare (n = 17) Total (n = 42)0–10 12 (92.3%) 12 (100%) 14 (82.4%) 38 (90.5%) 1 (7.7%) 0 (0%) 3 (17.6%) 4 (9.5%)10–20 13 (100%) 12 (100%) 7 (41.2%) 32 (76.2%) 0 (0%) 0 (0%) 10 (58.8%) 10 (23.8%)20–30 13 (100%) 12 (100%) 2 (11.8%) 27 (64.3%) 0 (0%) 0 (0%) 15 (88.2%) 15 (35.7%)30–40 12 (92.3%) 11 (91.6%) 3 (17.6%) 26 (61.9%) 1 (7.7%) 1 (8.3%) 14 (82.4%) 16 (38.1%)40–50 11 (84.6%) 11 (91.6) 4 (23.5%) 26 (61.9%) 2 (15.4%) 1 (8.3%) 13 (76.5%) 16 (38.1%)Note: Species with 1, the species is consideredaggregated, whereas U x1,x2 < 1 indicates regular distribution.We used Monte Carlo simulation to test whether aspecies is not significantly from random distribution. Fourhundred and ninety-nine distributions were simulated byrandomly labeling all species in the plot while keeping theabundance of each species the same as the observed. If theobserved U falls within the 2.5th and 97.5th quartiles, thenull hypothesis cannot be rejected. Otherwise, we wouldconclude that the species in the CBS plot is significantly differentfrom random distribution.Because U values in nearby distance classes were highlycorrelated with one another, we used U 0–10 , the mean conspecificdensity within 10 m of an individual, as a simplemeasure of the intensity of aggregation of a species (Conditet al. 2000) to compare spatial distributions of species in differentguilds (Table 1). We chose 10 m because direct interactionsamong species only occur within a limited distanceof 10 m (Wang et al. 2010). First, we divided species intothree abundance classes: rare (with abundance 10 individuals using U 0–10 as dependentvariables and abundance, maximum DBH, mean DBH,canopy, shade tolerance, and dispersal mode as independentvariables.ResultsOf the 42 species studied in the full CBS plot, 17 areclassified as rare, 12 as intermediate, and 13 as abundant.At the

Wang et al. 1015Fig. 2. Examples of species distribution patterns in the Changbaishan plot. Panels on the left show the relationship between U and scale andpanels on the right show the corresponding distribution patterns together with contour lines for six species. The lines with points are for U;the other lines are the simulation envelopes generated from 499 Monte Carlo simulations under the null hypothesis of complete spatialrandomness.(76.2%) are significantly aggregated at 10–20 m, and 27(64.3%) are significantly aggregated at 20–30 m (Table 2).The U generally declines with distance (Fig. 2): U 10–20

1016 Can. J. For. Res. Vol. 40, 2010Table 3. Spatial distributions across DBH classes for species with ‡10 individualsin the Changbaishan plot.DBH class (cm) Median U 0–10 species aggregated speciesTotal no. of No. of significantly1–5 7.79 33 305–10 5.45 21 2010–20 3.83 20 1620–30 7.54 14 1030–40 2.81 8 440–50 1.28 7 3>50 1.10 6 1Fig. 3. Relationship between aggregation index (U 0–10) and abundanceof species with abundances ‡10 in the Changbaishan plot.Fig. 4. Relationship between aggregation index (U 0–10 ) and DBH ofTilia amurensis. DBH classes: 1, 1–5 cm; 2, 5–10 cm; 3, 10–20 cm;4, 20–30 cm; 5, 30–40 cm; 6, 40–50 cm; 7, >50 cm.than understory species. In contrast, midstory species are notsignificantly different from overstory species and understoryspecies.The average U 0–10 of animal-dispersed species (2.5, SE =1.3) is smaller than that of gravity-dispersed species (17.6,SE = 3.9) and of wind-dispersed species (15.4, SE = 4.8).Results of a t test show that the aggregation intensity of animal-dispersedspecies is significantly different from that ofgravity-dispersed and wind-dispersed species. Animal-dispersedspecies are less aggregated than gravity-dispersedspecies and wind-dispersed species, whereas gravity-dispersedspecies are not significantly different from wind-dispersedspecies.The average U 0–10 of light-demanding species (11.4, SE =3.4) is smaller than that of shade-tolerant species (19.3,SE = 4.1), while the average U 0–10 of midtolerant species isthe smallest (4.2, SE = 2.1). Results of a t test show that theaggregation intensity of midtolerant species is significantlydifferent from that of light-demanding and shade-tolerantspecies. Midtolerant species are less aggregated than lightdemandingspecies and shade-tolerant species, whereaslight-demanding species are not significantly different fromshade-tolerant species.The results of multiple regression for U 0–10 show that theregression model is significant (P = 0.004). The standardizedcoefficients indicate that abundance has the largest effecton spatial aggregation followed by shade tolerance,mean DBH, dispersal mode, maximum DBH, and canopy(Table 4). The effects of mean DBH and canopy on spatialaggregation are negative, i.e., aggregation intensity declineswith these factors.DiscussionThe 42 species with ‡10 individuals comprised 99.9% ofall trees in the fully mapped 25 ha plot at CBS. Most specieswere aggregated, but the proportion of aggregated speciesdecreased with spatial scale in the temperate forests:aggegation is 90.5% at 0–10 m, 76.2% at 10–20 m and64.3% at 20–30 m. However, no clear decrease with spatialscale was found in tropical forests and subtropical forests.For example, Condit et al. (2000) found that >97.8% weresignificantly aggregated at the corresponding scales in tropicalforests, and Li et al. (2009) showed that aggregation was>96.1%, slightly lower than that in tropical forests. All ofthese suggest that the aggregation percentage of species in118Published by NRC Research Press

Wang et al. 1017Table 4. Multiple regression of aggregation intensity (U 0–10 ) with abundance maximumDBH, mean DBH, canopy, shade tolerance, and dispersal mode showing the estimatedcoefficients, standard errors, and standardized coefficients.Unstandardized coefficientEstimate SE Standardized (beta) coefficientConstant 27.727 24.33Abundance –6.084 1.565 –0.644Maximum DBH 0.036 0.259 0.08Mean DBH –0.196 0.388 –0.195Canopy –0.551 8.412 –0.022Shade tolerance 6.125 4.068 0.274Dispersal mode 4.05 4.908 0.137Note: The standardized coefficients are partial regression coefficients that indicate the relativeeffects of each variable on U 0–10 .natural forest communities may increase with increasingspecies richness.Rare species tended to be more aggregated than abundantones, which was consistent with that found in other forests(Condit et al. 2000; Davis et al. 2005; Li et al. 2009). However,not all species respond in a similar way. For example,Populus ussuriensis Komarov, one of the rare species with30 individuals, is expected to have a high U 0–10 value; however,the U 0–10 only equals 1.8, much less than the medianU 0–10 (24.4) of rare species. In contrast, the abundant speciesPrunus padus L. with 515 individuals has a relativelyhigh U 0–10 of 20.4 (Fig. 2E), which is more than the medianU 0–10 (1.9) of abundant species. One of the most importantreasons is that spatial distributions of species can arise frommany ecological processes, such as competition, stochasticrecruitment, dispersal limitation, habitat heterogeneity, disturbance,etc. (e.g., Cale et al. 1989; Rees et al. 1996; Tuomistoet al. 2003; Wiegand et al. 2007).The functional traits of species (e.g., size class, canopylayer, shade tolerance, and dispersal mode) were importantfactors in affecting spatial distributions of species in theCBS temperate forests. Species aggregation generally decreasedwith increasing size class (DBH) in the CBS plot.The finding that smaller individuals of a species were moreaggregated than larger individuals may be largely due toself-thinning. However, pathogens or herbivores may alsoplay an important role as spacing mechanisms in reducingaggregation in temperate forests (e.g., Seiwa et al. 2008). Inthe CBS temperate forests, previous studies showed that theseedlings or saplings near adult trees were often eaten byherbivores (e.g., Zhao and Zhang 2005). There are a numberof studies that support the notion of less aggregation withincreasing DBH (e.g., He et al. 1997; Condit et al. 2000;Getzin et al. 2008; Seiwa et al. 2008). For example, He etal. (1997) studied the spatial distributions of the 18 mostabundant species in the Pasoh Forest, Malaysia, and found adecrease in aggregation with increasing size class (DBH).Similarly, Li et al. (2009) observed a clear trend that aggregationis weaker in larger diameter classes. However, Conditet al. (2000) examined the spatial distributions of species insix different tropical forest plots and found that species atthe smaller diameter class were more aggregated at four ofthe six plots, whereas the pattern was reversed at the othertwo plots: most species became more aggregated at the largesize. According to these contrasting results, Murrell (2009)pointed out that although there was ample evidence for a reductionin aggregation with an increase in DBH, it was entirelypossible for adult trees to be more aggregated thanjuveniles when adult recruitment rates were low and dispersalwas poor even in the absence of any environmental heterogeneitysuch as slope or elevation.Dispersal limitation is commonly regarded as one of theimportant mechanisms to explain species aggregation, especiallyin hyperrich tropical forests (Hubbell 1979; Condit etal. 2000; Plotkin et al. 2000). In the CBS temperate forests,the aggregation distributions of species also indicated dispersallimitation. These species occurred in small-scaleclumps that did not correspond to topography (Figs. 2F and2H). They had relatively high U 0–10 values but these decreasedrapidly with distance (Figs. 2E and 2G). Some studiesindicated that the extent and scale of conspecific spatialaggregation were dependent on the mode of seed dispersal(Condit et al. 2000; Seidler and Plotkin 2006; Li et al.2009). Species with high dispersal ability were assumed tobe better dispersed than species with low dispersal ability,thus causing a less aggregated pattern for these species withhigh dispersal ability. Our study showed that species dispersedby animals were better dispersed than wind- andgravity-dispersed species. In addition, overstory species usuallyhave well-dispersed seeds relative to understory speciesand thus are expected to be less clumped than understoryspecies. Here, the prediction that better dispersal reduces aggregationwas borne out. Overstory species tended to be lessaggregated than understory species in the temperate forests.However, in tropical forests, there was a significant differencein aggregation intensity between overstory and understoryspecies at one plot, but at another plot, there may benot (Condit et al. 2000).Shade tolerance may also be expected to have a significanteffect on species distribution pattern. Previous studiesshowed that shade-tolerant species tended to have a steeplydescending monotonic diameter distribution with a largenumber of suppressed small trees (Leak 1975; Hett andLoucks 1976; Lorimer 1980; Wang et al. 2009), whereasmidtolerant species had almost unimodal distributions withfew suppressed small trees (Lorimer and Krug 1983; Wanget al. 2009). As we showed above that smaller trees weremore aggregated than larger trees, shade-tolerant specieswere thus expected to be more aggregated than midtolerantspecies. In addition, light-demanding species tended to be119Published by NRC Research Press

1018 Can. J. For. Res. Vol. 40, 2010localized in some gaps created by small-scale disturbances(e.g., windthrow), thus causing more aggregation than formidtolerant species. Our studies were consistent with the expectationthat midtolerant species were less clumped thanshade-tolerant species and light-demanding species.In addition, spatial heterogeneity, caused by topography,edaphic, or other environmental factors, has been widely consideredas an important factor in affecting spatial distributionsof species (e.g., Harms et al. 2001; John et al. 2007).Although the terrain of the CBS plot is relatively gentle, J.Ye et al. (unpublished analysis) found that nearly 60% of 35species studied showed significant habitat association (habitattype was classified based on topography). For example, thetwo species, Tilia mandshurica Rupr. & Maxim. and Ulmuslaciniata (Trautv.) Mayr favor the slope habitat (Figs. 2Gand 2L). Species differ in their ability to adapt to different environmentalconditions, which may result in the different distributionpattern of species in relation to environment.ConclusionsOur study provides unique and comprehensive analyseson the spatial distributions of species in a megaplot of atemperate forest, northeastern China. The results show thatmost species studied in the CBS temperate forests are aggregated,but the proportion of aggregated species decreaseswith distance. Analogous analyses in the species-rich tropicalor subtropical forests also show that spatial aggregationis the dominant pattern of species but no clear decreasewith distance (Condit et al. 2000; Li et al. 2009). In addition,species abundance has significant effects on the spatialaggregation pattern of species in the CBS temperate forests.For example, rare species are more aggregated than intermediateand common species. The aggregation intensity decreaseswith increasing DBH, i.e., species become moreregularly spaced as species grow, which is consistent withthe predictions of self-thinning and Janzen–Connell spacingeffects. Species functional traits (canopy layer, seed dispersalability, shade-tolerance, etc.) also have a significant effecton the spatial distributions of species. Consequently,dispersal limitation, self-thinning, Janzen–Connell spacingeffects, and habitat heterogeneity may be the primary contributingfactors in the spatial distributions of species in thetemperate forests.AcknowledgementsThis study was sponsored by the National Natural ScienceFoundation of China (40971286 and 30870400), the KnowledgeInnovation Program of the Chinese Academy of Sciences(KZCX2-YW-QN402), and the National KeyTechnologies R&D Program of China (2008BAC39B02).We thank Richard Condit for providing R programs for conductingpart of our analysis. We thank Micheal Papaik forchecking the English and providing useful comments. Wealso thank the Associate Editor and anonymous reviewersfor constructive criticism on the manuscript.ReferencesCale, W.G., Henebry, G.M., and Yeakley, J.A. 1989. Inferring processfrom pattern in natural communities. Bioscience, 39(9):600–606. doi:10.2307/1311089.Canham, C.D., Papaik, M.J., Uriarte, M., McWilliams, W.H.,Jenkins, J.C., and Twery, M.J. 2006. Neighborhood analyses ofcanopy tree competition along environmental gradients in NewEngland forests. Ecol. Appl. 16(2): 540–554. doi:10.1890/1051-0761(2006)016[0540:NAOCTC]2.0.CO;2. PMID:16711043.Chase, J. 2005. Towards a really unified theory for metacommunities.Funct. Ecol. 19(1): 182–186. doi:10.1111/j.0269-8463.2005.00937.x.Chave, J. 2004. Neutral theory and community ecology. Ecol. Lett.7(3): 241–253. doi:10.1111/j.1461-0248.2003.00566.x.Condit, R. 1998 Tropical forest census plots. Springer, Berlin, andLandes, Georgetown, Tex.Condit, R., Ashton, P.S., Baker, P., Bunyavejchewin, S., Gunatilleke,S., Gunatilleke, N., Hubbell, S.P., Foster, R.B., , Itoh, A., LaFrankie,J.V., Lee, H.S., Losos, E., Manokaran, N., Sukumar, R.,and Yamakura, T. 2000. Spatial patterns in the distribution oftree species. Science, 288: 1414–1418. doi:10.1126/science.288.5470.1414. PMID:10827950.Condit, R., Ashton, P., Bunyavejchewin, S., Dattaraja, H.S.,Davies, S., Esufali, S., Ewango, C., Foster, R., Gunatilleke,I.A.U.N., Gunatilleke, C.V.S., Hall, P., Harms, K.E., Hart, T.,Hernandez, C., Hubbell, S.P., Itoh, A., Kiratiprayoon, S.,Lafrankie, J., de Lao, S.L., Makana, J.R., Noor, M.N., Kassim,A.R., Russo, S., Sukumar, R., Samper, C., Suresh, H.S., Tan,S., Thomas, S., Valencia, R., Vallejo, M., Villa, G., and Zillio,T. 2006. The importance of demographic niches to tree diversity.Science, 313(5783): 98–101. doi:10.1126/science.1124712.PMID:16763113.Connell, J.H. 1971. On the role of natural enemies in preventingcompetitive exclusion in some marine animals and in rain foresttrees. In Dynamics of Populations: Proceedings of the AdvancedStudy Institute on Dynamics of Numbers in Populations, 7–18September 1970, Oosterbeek, The Netherlands. Edited by P.J.den Boer and G.R. Gradwell. Center for Agricultural Publicationsand Documentation, Wageningen, The Netherlands.pp. 298–312.Davis, M.A., Curran, C., Tietmeyer, A., and Miller, A. 2005. Dynamictree aggregation patterns in a species-poor temperatewoodland disturbed by fire. J. Veg. Sci. 16(2): 167–174. doi:10.1111/j.1654-1103.2005.tb02352.x.Duncan, R.P. 1991. Competition and the coexistence of species in amixed podocarp stand. J. Ecol. 79(4): 1073–1084. doi:10.2307/2261099.Gaston, K.J., and Chown, S.L. 2005. Neutrality and the niche.Funct. Ecol. 19(1): 1–6. doi:10.1111/j.0269-8463.2005.00948.x.Getzin, S., Wiegand, T., Wiegand, K., and He, F. 2008. Heterogeneityinfluences spatial patterns and demographics in foreststands. J. Ecol. 96(4): 807–820. doi:10.1111/j.1365-2745.2008.01377.x.Gravel, D., Beaudet, M., and Messier, C. 2008. Partitioning the factorsof spatial variation in regeneration density of shade-toleranttree species. Ecology, 89(10): 2879–2888. doi:10.1890/07-1596.1. PMID:18959325.Hao, Z., Zhang, J., Song, B., Ye, J., and Li, B. 2007. Vertical structureand spatial associations of dominant tree species in an oldgrowthtemperate forest. For. Ecol. Manag. 252(1–3): 1–11.doi:10.1016/j.foreco.2007.06.026.Hao, Z., Li, B., Zhang, J., Wang, X., Ye, J., and Yao, X. 2008.Broad-leaved Korean pine mixed forest plot in Changbaishan(CBS) of China: community composition and structure. ActaPhytoecol. Sin. 32: 238–250. [In Chinese.]Harms, K.E., Condit, R., Hubbell, S.P., and Foster, R.B. 2001. Habitatassociations of trees and shrubs in a 50-ha neotropical for-120Published by NRC Research Press

Wang et al. 1019est plot. J. Ecol. 89(6): 947–959. doi:10.1111/j.1365-2745.2001.00615.x.He, F., Legendre, P., and LaFrankie, J.V. 1997. Distribution patternsof tree species in a Malaysian tropical rain forest. J. Veg.Sci. 8(1): 105–114. doi:10.2307/3237248.Hett, J.M., and Loucks, O.L. 1976. Age structure models of balsamfir and eastern hemlock. J. Ecol. 64(3): 1029–1044. doi:10.2307/2258822.Hubbell, S.P. 1979. Tree dispersion, abundance, and diversity in atropical dry forest. Science, 203(4387): 1299–1309. doi:10.1126/science.203.4387.1299. PMID:17780463.Hubbell, S.P. 2001. The unified neutral theory of biodiversity andbiogeography. Princeton University Press, Princeton, N.J.Illian, J., Penttinen, A., Stoyan, H., and Stoyan, D. 2008. Statisticalanalysis and modeling of spatial point patterns. Wiley, Chichester,U.K.Janzen, D.H. 1970. Herbivores and the number of tree species intropical forests. Am. Nat. 104(940): 501–528. doi:10.1086/282687.John, R., Dalling, J.W., Harms, K.E., Yavitt, J.B., Stallard, R.F.,Mirabello, M., Hubbell, S.P., Valencia, R., Navarrete, H.,Vallejo, M., and Foster, R.B. 2007. Soil nutrients influence spatialdistributions of tropical tree species. Proc. Natl. Acad. Sci.U.S.A. 104(3): 864–869. doi:10.1073/pnas.0604666104. PMID:17215353.Kenkel, N.C. 1988. Pattern of self-thinning in jack pine: testing therandom mortality hypothesis. Ecology, 69(4): 1017–1024.doi:10.2307/1941257.Kubota, Y. 2006. Spatial pattern and regeneration dynamics in atemperate Abies–Tsuga forest in southwestern Japan. J. For.Res. 11(3): 191–201. doi:10.1007/s10310-006-0205-z.Leak, W.B. 1975. Age distribution in virgin red spruce and northernhardwoods. Ecology, 56(6): 1451–1454. doi:10.2307/1934714.Li, L., Huang, Z., Ye, W., Cao, H., Wei, S., Wang, Z., Lian, J.,Sun, I.-F., Ma, K., and He, F. 2009. Spatial distributions of treespecies in a subtropical forest of China. Oikos, 118(4): 495–502.doi:10.1111/j.1600-0706.2009.16753.x.Lorimer, C.G. 1980. Age structure and disturbance history of asouthern Appalachian virgin forest. Ecology, 61(5): 1169–1184.doi:10.2307/1936836.Lorimer, C.G., and Krug, A.G. 1983. Diameter distribution in evenagedof shade-tolerant and midtolerant tree species. Am. Midl.Nat. 109(2): 331–345. doi:10.2307/2425414.Lortie, C.J., Brooker, R.W., Choler, P., Kikvidze, Z., Michalet, R.,Pugnaire, F.I., and Callaway, R.M. 2004. Rethinking plant communitytheory. Oikos, 107(2): 433–438. doi:10.1111/j.0030-1299.2004.13250.x.Mori, A.S., and Komiyama, A. 2008. Differential survival amonglife stages contributes to co-dominance of Abies mariesii andAbies veitchii in a sub-alpine old-growth forest. J. Veg. Sci.19(2): 239–244. doi:10.3170/2008-8-18364.Mouquet, N., and Loreau, M. 2003. Community patterns in source–sink metacommunities. Am. Nat. 162(5): 544–557. doi:10.1086/378857. PMID:14618534.Murrell, D.J. 2009. On the emergent spatial structure of size-structuredpopulations: when does self-thinning lead to a reduction inclustering? J. Ecol. 97(2): 256–266. doi:10.1111/j.1365-2745.2008.01475.x.Papaik, M.J., and Canham, C.D. 2006. Multi-model analysis of treecompetition along environmental gradients in southern New Englandforests. Ecol. Appl. 16(5): 1880–1892. doi:10.1890/1051-0761(2006)016[1880:MAOTCA]2.0.CO;2. PMID:17069379.Peters, H.A. 2003. Neighbour-regulated mortality: the influence ofpositive and negative density dependence on tree populations inspecies-rich tropical forests. Ecol. Lett. 6(8): 757–765. doi:10.1046/j.1461-0248.2003.00492.x.Plotkin, J.B., Potts, M.D., Leslie, N., Manokaran, N., Lafrankie, J.,and Ashton, P.S. 2000. Species-area curves, spatial aggregation,and habitat specialization in tropical forests. J. Theor. Biol.207(1): 81–99. doi:10.1006/jtbi.2000.2158. PMID:11027481.Purves, D.W., and Pacala, S.W. 2005. Ecological drift in nichestructuredcommunities: neutral pattern does not imply neutralprocess. In Biotic interactions in the tropics. Edited by D. Burslem,M. Pinard, and S. Hartley. Cambridge University Press,Cambridge, U.K.Rees, M., Grubb, P.J., and Kelly, D. 1996. Quantifying the impactof competition and spatial heterogeneity on the structure and dynamicsof a four-species guild of winter annuals. Am. Nat.147(1): 1–32. doi:10.1086/285837.Ripley, B.D. 1981. Spatial statistics. Wiley, New York.Seidler, T.G., and Plotkin, J.B. 2006. Seed dispersal and spatialpattern in tropical trees. PLoS Biol. 4(11: e344): 2132–2137.doi:10.1371/journal.pbio.0040344.Seiwa, K., Miwa, Y., Sahashi, N., Kanno, H., Tomita, M., Ueno,N., and Yamazaki, M. 2008. Pathogen attack and spatial patternsof juvenile mortality and growth in a temperate tree, Prunusgrayana. Can. J. For. Res. 38(9): 2445–2454. doi:10.1139/X08-084.Stoll, P., and Newbery, D.M. 2005. Evidence of species-specificneighborhood effects in the Dipterocarpaceae of a Bornean rainforest. Ecology, 86(11): 3048–3062. doi:10.1890/04-1540.Tuomisto, H., Ruokolainen, K., and Yli-Halla, M. 2003. Dispersal,environment, and floristic variation of western Amazonian forests.Science, 299(5604): 241–244. doi:10.1126/science.1078037. PMID:12522248.Wang, X., Hao, Z., Ye, J., Zhang, J., Li, B., and Yao, X. 2008a.Spatial pattern of diversity in an old-growth temperate forest innortheastern China. Acta Oecol. 33(3): 345–354. doi:10.1016/j.actao.2008.01.005.Wang, X., Hao, Z., Ye, J., Zhang, J., Li, B., and Yao, X. 2008b.Spatial variation of species diversity across scales in an oldgrowthtemperate forest of China. Ecol. Res. 23(4): 709–717.doi:10.1007/s11284-007-0430-8.Wang, X., Hao, Z., Zhang, J., Lian, J., Li, B., Ye, J., and Yao, X.2009. Tree size distributions in an old-growth temperate forest.Oikos, 118(1): 25–36. doi:10.1111/j.0030-1299.2008.16598.x.Wang, X., Wiegand, T., Hao, Z., Li, B., Ye, J., and Lin, F. 2010.Species associations in an old-growth temperate forest in northeasternChina. J. Ecol. 98: 674–686. doi:10.1111/j.1365-2745.2010.01644.x.Wiegand, T., and Moloney, K.A. 2004. Rings, circles, and nullmodelsfor point pattern analysis in ecology. Oikos, 104: 209–229. doi:10.1111/j.0030-1299.2004.12497.x.Wiegand, T., Gunatilleke, S., Gunatilleke, N., and Okuda, T. 2007.Analyzing the spatial structure of a Sri Lankan tree species withmultiple scales of clustering. Ecology, 88(12): 3088–3102.doi:10.1890/06-1350.1. PMID:18229843.Zhao, Y., and Zhang, L. 2005. Investigation of pests for wild treesin Changbai Mountains. J. Northeast For. Univ. 33: 107–109. [InChinese.]121Published by NRC Research Press

Journal of Ecology 2010, 98, 674–686doi: 10.1111/j.1365-2745.2010.01644.xSpecies associations in an old-growth temperateforest in north-eastern ChinaXugao Wang 1 , Thorsten Wiegand 2 , Zhanqing Hao 1* , Buhang Li 1 ,JiYe 1 and Fei Lin 11 Institute of Applied Ecology, Chinese Academy of Sciences, PO Box 417, Shenyang 110016, China; and 2 Departmentof Ecological Modelling, UFZ Helmholtz Centre for Environmental Research-UFZ, PF 500136, D-04301 Leipzig,GermanySummary1. Studying the spatial pattern of plants may provide significant insights into processes and mechanismsthat maintain species richness. We used data from a fully mapped 25-ha temperate forest plotat Changbaishan (CBS), north-eastern China, to conduct a community-wide assessment of the typeand frequency of intra- and interspecific spatial association patterns. We analysed complex scaleeffects in the patterning of large trees of 15 common species. First, we tested for overall spatialpatterning at 6, 30 and 50 m neighbourhoods and classified the types of bivariate associationpatterns at these spatial scales (analysis 1). We then explored small-scale (0–20 m) associationpatterns conditioning on the larger-scale pattern (analysis 2) and tested for positive large-scale(50–250 m) association patterns (analysis 3).2. Analysis 1 provided ample evidence for non-random spatial patterning, and the type and frequencyof spatial association patterns changed with scale. Trees of most species pairs co-occurredless than expected by chance and positive associations were rare in local neighbourhoods. Analysis2 revealed a separation of scales in which significant small-scale interactions faded away at distancesof 10–15 m. One third of all species pairs showed significant and mostly negative bivariatesmall-scale association, which occurred more often than expected by chance between species sharingattributes such as family, fruit type and habitat association. This suggests the occurrence of competitiveinteractions. Analysis 3 showed that only 8% of all species pairs co-occurred at large scales.3. Comparison of our results with an analogous study conducted in the species-rich tropical forestat Sinharaja, Sri Lanka, revealed several structural similarities including the dominance of segregationand partial overlap in the overall patterning (analysis 1) and the separation of scales (analysis2). However, species pairs at CBS showed considerably more significant negative small-scale associations(31% vs. 6% at Sinharaja).4. Synthesis. The techniques presented here allow for a detailed analysis of the complex spatialassociations in species-rich forests and have the potential to reveal indicative patterns that mayallow researchers to discriminate among competing hypotheses of community assemblage anddynamics. However, this will require comparative studies involving a large number of plots.Key-words: Changbaishan (CBS), coexistence, pair correlation function, point pattern analysis,spatial segregation, species association, temperate forestIntroductionA central aim in ecology is to understand processes and mechanismsthat control the distribution and abundance of species(Ricklefs 1990). The importance of spatial patterns in thisrespect has been increasingly recognized (e.g. Watt 1947; Pacala1997; Tilman & Kareiva 1997; Amarasekare 2003), and investigationof the effect of spatial interactions on population andcommunity dynamics is a focus of current ecological research*Correspondence author. E-mail: hzq@iae.ac.cn(Murrell, Purves & Law 2001; McIntire & Fajardo 2009). Spatialpatterns and processes are especially important for plantsbecause they cannot move from one environment to anotherand they interact mainly with their close neighbours (i.e. the‘plant’s eye view’ of the community; Turkington & Harper1979; Purves & Law 2002). Although substantial gains havebeen made in understanding processes and mechanisms thatcontrol the distribution and abundance of plant species, thereare still important challenges, mainly because similar patternscan be explained by substantially different theories (e.g. Chave2004; Gilbert & Lechowicz 2004; Bell 2005). For example, theÓ 2010 The Authors. Journal compilation Ó 2010 British Ecological Society122

Analysing complex spatial association patterns 675observed patterning of plants may be explained by environmentalniches and trade-offs among species in dispersal and competitiveability (e.g. Tilman 2004). However, models based onneutral theories, where ecological drift is the only process occurringbesides stochastic but limited dispersal and speciation (Bell2001; Hubbell 2001; Chave 2004), can also form patterns ofdistribution and abundance similar to those found in nature(Mouquet & Loreau 2003; Chave 2004; Purves & Pacala 2005).One possible reason for the current inability to decidebetween competing theories is that the patterns used, basicallythe species abundance distribution and the species–area relationship,have low discriminatory power (e.g. Chave 2004; Purves& Pacala 2005). McGill et al. (2006) pointed to a need toevaluate additional and independent predictions, such as spatialpatterns of the location of plants within a community(Hubbell et al. 2001; Wiegand, Gunatilleke & Gunatilleke2007a; Wiegand et al. 2007b). These point patterns are theresults of all mechanisms and processes that have affected thefate of the plants during their life and constitute an ‘ecologicalarchive’ which may contain encoded information on the underlyingprocesses (Hubbell et al. 2001; Wiegand et al. 2003;Grimm et al. 2005; McIntire & Fajardo 2009; Wiegand, Huth&Martínez 2009). The observed spatial patterns may be usedfor model selection in dynamic and spatially explicit simulationmodels to infer processes (e.g. Wiegand et al. 2003; Grimmet al. 2005). However, a prerequisite of this research programto decide between competing theories (Levin 1992; Grimmet al. 2005) is identification of indicative patterns in the spatialdata. (Note that we use ‘indicative pattern’ to denote patternsthat have the potential to reveal underlying biological mechanisms.This differs from the spatial point pattern literature inwhich ‘pattern’ denotes the spatial pattern of the coordinates ofecological objects within an observation window.) Promisingcandidates for indicative patterns that could reveal informationon community structure include the interspecific spatial associationsbetween pairs of species. However, the scale-dependentnature of spatial processes and mechanisms involved in spatialpattern generation makes examination of these relationships adifficult undertaking. It cannot be expected that all spatial processesand mechanisms act on the same characteristic spatialscales (Wiegand, Huth & Martínez 2009), therefore properanalysis of spatial association patterns must take into accountthe multi-scale nature of the patterns. For example, two speciesmay be segregated at a larger scale (i.e. they occupy largely disjointpatches), but if plants of the two species are direct neighbours,they may show a positive association. This positive,small-scale association may be overlooked if the analysis is notconducted conditionally on the otherwise overpowering largescaledistribution pattern. Only a few attempts have been madeto characterize the spatial association between species withinspecies-rich plant communities (but see Kubota, Kubo & Shimatani2007; Lieberman & Lieberman 2007; Wiegand, Gunatilleke& Gunatilleke 2007a; Wiegand et al. 2007b; Perry et al.2009; Illian, Moller & Waagepetersen 2009). However, suchstudies are of prime importance for advancing our understandingof spatial processes and mechanisms on coexistence in species-richplant communities.The spatial segregation hypothesis is an important exampleof a hypothesis related to species coexistence that involvesspatial association between pairs of species (Pacala 1997;Pacala & Levin 1997). Intraspecific aggregation leads to interspecificsegregation whereby an average plant in the communitywill compete mostly locally with con-specific plants. As aresult, competitively superior species become suppressed,which prevents (or retards) the elimination of competitivelyinferior species (Stoll & Prati 2001), thereby promoting speciesdiversity (Kareiva 1990; Tilman 1994; Rees, Grubb & Kelly1996; Pacala 1997; Stoll & Weiner 2000). Motivated by the segregationhypothesis, we will study for many species pairs howoften plants of the second species are located within a specifiedneighbourhood around plants of the first species. Comparisonof such spatial structure of several plant communities with contrastingcharacteristics (e.g. tropical vs. temperate forest) mayfinally allow us to identify indicative patterns in the data to distinguishamong competing theories of species coexistence(Wiegand et al. 2003; Grimm et al. 2005; McIntire & Fajardo2009; Wiegand, Huth & Martínez 2009). In this study, we takea step in this direction.We focused here on detailed point pattern analysis based onmaps of the locations of all larger trees in a given forest plot(Kenkel 1988; Duncan 1991; Hoshino, Nishimura & Yamamoto2001; Kubota, Kubo & Shimatani 2007; Wiegand, Gunatilleke& Gunatilleke 2007a; Wiegand et al. 2007b). Weanalysed point pattern data from the completely mapped 25-ha Changbaishan (CBS) plot located in the Changbai NatureReserve, China. The large size of the CBS plot provided theunique opportunity to investigate community-wide interspecificassociations for a temperate forest (we included 15 species).To compare the results of the CBS plot with those of the25-haplotoftropicalforestatSinharaja,SriLanka(Wiegand,Gunatilleke & Gunatilleke 2007a), we used the methodologyproposed by Wiegand, Gunatilleke & Gunatilleke (2007a), butconducted additional analyses to relate our results to the speciescharacteristics.Our overall goal was an assessment of the type and frequencyof intra- and interspecific spatial association patternsin the CBS forest. Because of the complex scale effects, wedivided this goal into three analyses. First, we tested for overallnon-random spatial patterning in uni- and bivariate patternsand classified the types of bivariate association patterns at variousspatial scales (analysis 1). Under the segregation hypothesis,we expected mostly aggregated univariate patterns andsegregation or partial overlap between species pairs in localneighbourhoods. In analysis 2, we tested selectively for smallscaleassociation patterns (0–30 m) conditional on theobserved large-scale (i.e. > 30 m) variations in the intensity(that may be caused by dispersal and ⁄ or association tosmoothly varying environmental factors). We expected a highproportion of negative associations (caused by competitiveinteractions) among species that share ecological attributes.Finally, we used a null model that tests selectively for positivelarge-scale association patterns (50–250 m; analysis 3). Speciesbelonging to the same successional stage should occupy thesame larger-scale patches.Ó 2010 The Authors. Journal compilation Ó 2010 British Ecological Society, Journal of Ecology, 98, 674–686123

676 X. Wang et al.Materials and methodsSTUDY AREAThe study area is located in the Changbai Nature Reserve, whichextends along the border of China and North Korea from 127°42¢ to128°17¢ E and 41°43¢ to 42°26¢ N (Fig. 1). It is one of the largest biospherereserves in China and has been spared from logging and othersevere human disturbances since it was established in 1960. TheChangbai Nature Reserve joined the World Biosphere Reserve Networkunder the UNESCO Man and the Biosphere Programme in1980. The CBS plot is located in the core zone of the Changbai NatureReserve and is representative of broad-leaved Korean pine mixedforest, the most common vegetation type in species composition andecosystem structure for the region.The 500 · 500 m CBS plot was established in summer 2004 and isthe northernmost plot of the China Forest Biodiversity MonitoringNetwork (http://www.cfbiodiv.org) initiated by the Chinese Academyof Sciences (CAS) in collaboration with the Center for TropicalForest Science (CTFS). Mean elevation in the plot is 801.5 m; the elevationranges from 791.8 to 809.5 m. The soil is classified as darkbrown forest soil. Mean annual precipitation is approximately700 mm; most of this occurs from June to September (480–500 mm).Mean annual temperature is 2.8 °C, with a January mean of)13.7 °C, and a July mean of 19.6 °C (Yang & Li 1985). All trees atleast 1 cm in diameter at breast height (d.b.h.) were mapped and identifiedto species; their geographic coordinates were recorded followinga standard field protocol (Condit 1998). The total number of livingindividuals in the first census of 2004 was 38 902, consisting of 52species, 32 genera and 18 families (Hao et al. 2008).FOREST CHARACTERISTICSThe main tree species at CBS include Pinus koraiensis, Tilia amurensis,Quercus mongolica, Fraxinus mandshurica, Ulmus japonica andAcer mono. Unlike tropical rain forests without obvious dominantspecies, there were eight species with more than 1000 individuals,which accounted for 83.4% of the total individuals in the plot. Meanstand density was 1556 living trees per ha. Mean basal area was43.2 m 2 ha )1 (Hao et al. 2008; Wang et al. 2009). In this study, thespatial pattern of 10 527 non-juvenile trees with d.b.h. ‡ 10 cm wasanalysed. To obtain a sufficiently large sample size for the pointpattern analyses, we included 15 species each with more than 50 suchtrees (Table 1), making up 97.7% (10 313 trees) of the total. Theecological characteristics of these species are also shown in Table 1.Most species in the studied forest were clustered at some spatial scale(Hao et al. 2007; Wang et al. 2008b) and about 60% of 35species studied showed significant habitat association (J. Ye, unpubl.data).Forest vegetation at the CBS plot is largely the result of naturalsuccession (Zhao 1981), which is now at the late-succession stage. Themean age of the canopy trees is about 300 years. One common coexistencehypothesis is niche differentiation in temperate forests (Daiet al. 2004), resulting from heterogeneous environmental conditions(e.g. differences in topographical and soil factors). However, Wanget al. (2008a) showed that pure topographical factors were not sufficientto explain the species diversity pattern at the CBS plot. Anothercoexistence hypothesis is that natural disturbance (e.g. wind) createscanopy openings and consequently initiates regeneration, maintainingearly successional species by local secondary succession, thusenabling the coexistence of tree species (Zang et al. 1998). In addition,species competition may be important for the maintenance of speciesdiversity at small scales (Wang et al. 2008a).Statistical analysesPOINT PATTERN ANALYSIS: SUMMARY STATISTICSTo quantify the spatial patterns found at the CBS forest ‘from theplant’s eye view’, we used recent techniques of spatial point patternanalysis (Ripley 1981; Stoyan & Stoyan 1994; Diggle 2003; Illianet al. 2008) and summary statistics such as the pair-correlation function(Stoyan & Stoyan 1994), Ripley’s (1981) K-function and the distributionfunction of nearest neighbour distances (Diggle 2003). Thebivariate pair-correlation function g 12 (r) can be estimated using thequantity k 2 g 12 (r), which is the mean density of trees of species 2 at distancer away from trees of the focal species 1, whereby k 2 is the meandensity of trees of species 2 in the whole study area. Ripley’s (1981)K-function K (r) is the cumulative version of the pair-correlationfunction, i.e. the quantity k 2 K 12 (r) is the average number of trees ofspecies 2 within distance r from trees of the focal species 1. The univariateK- and pair-correlation functions follow intuitively (Wiegand &Moloney 2004), but the focal point is not counted. To describe additionalcharacteristics of the spatial patterns, we used the bivariate distributionfunction D 12 (y) which gives the fraction of trees of the focalspecies 1 that have their nearest species 2 neighbour within distancey (Diggle 2003; Illian et al. 2008). Note that D(y) is often referred toas G(y) in the literature (e.g. Diggle 2003), but we have adopted thenotation of the recent textbook by Illian et al. (2008).ChinaY-coordinates (m)0 100 200 300 400 5007967988008028048068088068048048020 100 200 300 400 500X-coordinates (m)NFig. 1. The location and contour map of the 25-ha (500 · 500 m) Changbai temperate plot.Ó 2010 The Authors. Journal compilation Ó 2010 British Ecological Society, Journal of Ecology, 98, 674–686124

Analysing complex spatial association patterns 677Table 1. Species properties. Habitat associations (J. Ye, unpubl. data) include the topographic types: low-plateau (slope < 7°, elevation< 804.0 m), high-plateau (slope < 7°, elevation‡ 804.0 m) and slope (slope ‡ 7°). ‘+’ indicates significant positive association; ‘)’ indicatessignificant negative association; N indicates no significant associationSpeciesSymbolNo.treesFamilyCanopylayerShade-tolerantFruittypeDispersalmodeHabitatassociationAcer mandshuricum acma 67 Aceraceae Midstorey Shade-tolerant Samara Wind NAcer mono acmo 1677 Aceraceae Midstorey Shade-tolerant Samara Wind Low plateau+Acer pseudo-sieboldianum acps 1036 Aceraceae Midstorey Shade-tolerant Samara Wind Low plateau)Acer tegmentosum acte 115 Aceraceae Midstorey Shade-tolerant Samara Wind High plateau+Acer triflorum actr 102 Aceraceae Midstorey Shade-tolerant Samara Wind High plateau)Betula platyphylla bepl 91 Betulaceae Midstorey Light-demanding Tiny winged nut Wind NFraxinus mandshurica frma 647 Oleaceae Canopy Midtolerant Samara Wind Low plateau+Maackia amurensis maam 365 Leguminosae Midstorey Midtolerant Legume Gravity NPhellodendron amurense pham 56 Rutaceae Midstorey Light-demanding Drupe Gravity Low plateau+Pinus koraiensis piko 2443 Pinaceae Canopy Midtolerant Cone Gravity Slope)or animalPrunus padus prpa 77 Rosaceae Midstorey Shade-tolerant Drupe Gravity Low plateau+Quercus mongolica qumo 769 Fagaceae Canopy Light-demanding Nut Gravity Nor animalTilia amurensis tiam 2333 Tiliaceae Canopy Shade-tolerant Nut Gravity NTilia mandshurica tima 131 Tiliaceae Canopy Shade-tolerant Nut Gravity Slope+Ulmus japonica ulja 404 Ulmaceae Canopy Midtolerant Samara Wind NThe g-, K- andD- statistics are usually interpreted for homogeneouspatterns to indicate interactions among pairs of points. In thiscase, they reflect properties of a ‘typical tree’ of the pattern (Illianet al. 2008). However, the patterns at our study site are certainly notall homogeneous, which means that a typical tree of a pattern maynot exist. Instead, we interpreted the g- andK-functions as averagestaken over all trees of the focal pattern and designed our analyses andnull models so as to account for potential heterogeneities.POINT PATTERN ANALYSIS: TESTING SIGNIFICANCEOF PATTERNS AGAINST A NULL MODELFor a given species or species pair, we contrasted the observed summarystatistics to that expected under an appropriate null model. Weused a Monte-Carlo approach to test for significant departures fromthe null models. Each of the n = 199 simulations of a point processunderlying the null model generates a summary statistic [e.g. a paircorrelationfunction g 12 (r)]; simulation envelopes with a = 0.05 werecalculated from the 5th highest and lowest values of g 12 (r) in the 199simulations (Stoyan & Stoyan 1994). Significant departure from thenull model occurred at scale r if the test statistic was outside the simulationenvelopes. This approach allowed us to assess scale effectsapproximately for illustrative purposes and to determine the type ofsignificant effect.However, to avoid problems due to simultaneous inference (e.g.Loosmore & Ford 2006), we evaluated the overall ability of a givennull model to describe the data by means of a goodness-of-fit test(GoF; Diggle 2003; Illian et al. 2008). This test reduces the scaledependentinformation contained in the summary statistics into a singletest statistic, u i , which represents the total squared deviationbetween the observed pattern and the theoretical results across thedistances of interest (i.e. a Cramer–von Mises type statistic as e.g.used in Plotkin et al. 2000). The u i values were calculated for theobserved data (i = 0) and for the data created by the (i = 1...199)simulations of the null model, and the rank of u 0 among all u i valueswas determined. If the rank of u 0 was larger than 190, the data showeda significant departure from the null model (across the distances ofinterest) with error rate a =0.05.ANALYSIS 1: DETECTION OF OVERALL NON-RANDOMPATTERNINGUnivariate caseTo detect overall departure from randomness, we confronted ourdata with the null model of complete spatial randomness (CSR; Wiegand& Moloney 2004). To quantify departures from the null model,we used the pair-correlation function. Note that this test is sensitiveto effects from heterogeneous environment, dispersal and tree interactions.Bivariate caseOur basic question was conceptually simple: we wanted to knowhow the trees of a given species 2 were distributed within localneighbourhoods of the trees of a focal species 1. Did they occur onaverage more (or less) frequently within the neighbourhoods thanexpected by chance alone, and was this association homogeneousor heterogeneous? In the heterogeneous case this distribution variessubstantially among trees of the focal species, e.g. some species 1trees may have many species 2 neighbours but other species 1 treeshave few species 2 neighbours. To distinguish the various types ofspatial associations from those that may arise purely by chance, wecompared the observed bivariate point patterns with a null modelin which the locations of the focal species remained unchanged, buttrees of species 2 were distributed randomly and independently ofthe locations of species 1 (i.e. CSR). Clearly, testing against CSR isoften not very informative (Wiegand & Moloney 2004); however,we used this test to quantify and categorize the overall bivariatespatial associations based on a scheme developed by Wiegand,Gunatilleke & Gunatilleke (2007a; see section below). The schemeuses the bivariate K 12 (r) andD 12 (y) as test statistics and distinguishesfour significant types of spatial associations that may occurbetween two (possibly heterogeneous) patterns (see section below).A ‘not significant’ type arises if neither K 12 (r) norD 12 (y) showsignificantdepartures from the CSR null model (as measured by theGoF tests).Ó 2010 The Authors. Journal compilation Ó 2010 British Ecological Society, Journal of Ecology, 98, 674–686125

678 X. Wang et al.Scheme to characterize bivariate associationsThe spatial association between two species can be characterized bythe distribution function P 12 (n, r) that gives the probability of findingn trees of species 2 within neighbourhoods of radius r around trees ofspecies 1. If the point configurations between pairs of trees of the twospecies are the same all over the study plot except for stochastic variation(i.e. homogeneous patterns), we do not need the full distributionP 12 (n, r) to describe the association between the two species. In thiscase the mean of P 12 (n, r)withrespectton suffices, which is given byk 2 K 12 (r). However, we cannot expect that a typical bivariate pointconfiguration exists at the CBS plot because the patterns of severalspecies show heterogeneities (Table 1; J. Ye, unpubl. data). Wetherefore need a second characteristic of P 12 (n, r) in order to characterizethe bivariate spatial association patterns more fully. This isbecause the same value of the mean (i.e. k 2 K 12 (r)) may arise for substantiallydifferent situations, e.g. if (i) all trees of species 1 have moreor less the same number of neighbours of species 2 (i.e. a homogenouspattern) or if (ii) a few trees of species 1 have many species 2 neighboursbut many trees of species 1 have no species 2 neighbours (anextremely heterogeneous pattern). Wiegand, Gunatilleke & Gunatilleke(2007a) selected the value of the distribution P 12 (n, r)atn =0as an additional summary statistic. P 12 (n =0,r) is the probabilitythat a tree of species 1 has within distance r no neighbour of species 2,i.e. P 12 (n =0,r) =1)D 12 (r). Because the summary statistics K 12 (r)and D 12 (r) express fundamentally different properties of bivariatepoint patterns (Illian et al. 2008), they are a good choice for classifyingdifferent types of bivariate associations. The expectations of thetwo summary statistics under the null model yield D exp12 ¼1 e k2pr2 and K exp 12 (r) =pr 2 where the ‘exp’ superscript indicates‘expected by the null model of no spatial patterning’. The two axes ofthe scheme are defined as^PðrÞ ¼ ^D 12 ðrÞ ð1 e k2pr2 Þ^MðrÞ ¼ lnð ^K 12 ðrÞÞ lnðpr 2 Þeqn 1whereby the hat symbol indicates the observed value. We subtractedthe theoretical values under the null model to move nullassociation onto the origin of the scheme (i.e. no departure fromthe null model) and log transformed the K-function in order toweight positive or negative departures from the null model in thesame way (Wiegand, Gunatilleke & Gunatilleke 2007a).The two-axis scheme allows for four fundamental types of bivariateassociation. In the case of ‘segregation’ (type I), both the averagenumber of neighbours within distance r and the proportion of nearestneighbours within distance r are smaller than expected [i.e. ^MðrÞ 0 and ^PðrÞ > 0]. In the case of ‘partial overlap’(type II), the mean number of trees of species 2 within neighbourhoodsof radius r around trees of species 1 is larger than would beexpected according to the null model [i.e. ^MðrÞ > 0], and the probabilitythat a tree of species 1 has no neighbour of species 2 is smallerthan expected [i.e. ^D 12 ðrÞ < 0]. This configuration is only possiblefor heterogeneous patterns if some trees of species 1 are surroundedat the given neighbourhood r by many trees of species 2 but others aresurrounded by few (or no) trees of species 2. Finally, in the case oppositeto partial overlap (type IV), trees of species 1 are highly clusteredand trees of species 2 overlap the cluster of species 1. As a result, themean number of species 2 neighbours is smaller than expected[ ^MðrÞ < 0], but the probability to have the nearest neighbour of species2 within distance r is larger than expected [i.e. ^D 12 ðrÞ > 0]. Thisis because a few trees of species 2 are in fact the nearest neighbour ofmost trees of the highly clustered species 1. Type IV associations willrarely occur (Wiegand, Gunatilleke & Gunatilleke 2007a).ANALYSIS 2: ISOLATION OF SMALL-SCALE EFFECTSAcknowledging the multi-scale nature of the spatial association patterns,we selectively studied the small-scale association pattern byusing a null model which randomizes the data conditionally on theobserved large-scale pattern. In practice, this can be done by displacingthe known locations of trees randomly within a neighbourhoodwith radius R (i.e. a heterogeneous Poisson null model; Wiegand,Gunatilleke & Gunatilleke 2007a; Wiegand et al. 2007b). Thisdisplacement removes potential patterns for distances r < R, butitleaves the larger-scale patterns untouched. Contrasting the observedpattern to realizations of this null model will therefore detect onlysmall-scale effects.While this analysis can be conducted for any displacement distanceR, it is desirable to use a distance which is likely to separate biologicaleffects. In general, it is expected that direct interactions among largertrees only occur within a limited spatial separation (e.g. < 30 m).For example, Hubbell et al. (2001) found that the neighbourhoodeffects of conspecific density on survival disappeared within approximately12–15 m of the focal plant. Several other studies using individual-basedanalyses of local neighbourhood effects on tree growth andsurvival confirmed this result (e.g. Uriarte et al. 2004, 2005; Stoll &Newbery 2005) suggesting that direct plant–plant interactions in forestsmay fade away at larger scales. We therefore used a separationdistance of R = 30 m (Wiegand, Gunatilleke & Gunatilleke 2007a).An interesting question is whether separation of scalesoccurred: this can be tested in a simple way. Because the heterogeneousPoisson process conditions on the spatial structure forscales larger than 30 m, it is only able to indicate significanteffects at scales smaller than 30 m. In cases without separationof scales, we expect therefore that the frequency of significanteffects, taken over all pairs of species, should fade awaysmoothly at 30 m. However, if small-scale effects operate onlyover a short range (i.e. r > 30 m), the frequency of significanteffects should disappear well below the threshold of 30 m.Univariate caseThe null model was a heterogeneous Poisson process based on a nonparametricintensity estimate using the Epanechnikov kernel with abandwidth of R = 30 m (see e.g. Wiegand, Gunatilleke & Gunatilleke2007a). This null model is basically equivalent to randomly displacingthe tree locations of the species of interest within a radius of30 m. We used the univariate pair-correlation function g(r) as test statisticand studied species associations with a spatial resolution of 2 mup to 40 m. This is a sufficiently fine resolution to capture detailedvariation in the pair-correlation function over the range of scales ofinterest, but it is coarse enough to yield feasible computation time forthe high number of analyses required. We selected 40 m as the maximalscale of analysis in order to check that the heterogeneous Poissonprocess does indeed only depict significant effects at scales smallerthan 30 m. Thus, this analysis has several types of ‘spatial resolutions’:the kernel bandwidth, the grain at which we estimate g(r), andthe maximal scale up to which we estimate g(r).Bivariate caseThe null model was analogous to that of the univariate case, butwe left the locations of species 1 untouched and distributed theÓ 2010 The Authors. Journal compilation Ó 2010 British Ecological Society, Journal of Ecology, 98, 674–686126

Analysing complex spatial association patterns 679trees of species 2 using the heterogeneous Poisson process describedabove. We used the bivariate pair-correlation function g 12 (r) asteststatistic. Again we studied species associations with a spatial resolutionof 2 m up to 40 m. Because the association may be asymmetric,we tested all pairs, i.e. both species 1 versus species 2 andspecies 2 versus species 1.LARGE-SCALE ASSOCIATIONS (ANALYSIS 3)Approximately 60% of species at the CBS plot showed significanthabitat association (J. Ye, unpubl. data); it is thereforeinteresting to find out whether andhowoftentwospeciesshareroughly the same areas within the 25-ha study plot. An appropriatenull model for this question leaves the trees of species 1unchanged and has the trees of species 2 distributed in accordancewith the large-scale pattern of species 1. We implementedthis null model as a heterogeneous Poisson processwhereby the locations of the trees of species 2 were randomizedin accordance with the intensity function of species 1. We usedan Epanechnikov kernel with a bandwidth of 50 m to estimatethe intensity function of species 1 (Stoyan & Stoyan 1994; seeWiegand, Gunatilleke & Gunatilleke 2007a). Note that thebandwidth of 50 m was selected to capture effects of largerscale association.SETTINGS FOR ANALYSES AND GOF TESTFor analyses 1 and 2, we used a spatial resolution dw = 2 m; weused 5 m for analysis 3. Note that the scale r = 0 m for the paircorrelation function refers to effects ranging from 0 m to half ofthe spatial resolution, and scale r refers to effects ranging withinscales r)dw ⁄ 2 and r + dw ⁄ 2. For analysis 2, we conducted theGoF test as in Wiegand et al. (2007b) over the distance intervalof 0–20 m because we expected significant effects to occur primarilyover this range of scales. For analysis 3, we conductedthe GoF test over the full range of scales (i.e. 50–250 m) atwhichweexpectedthenullmodeltobemet.Weretaineddatasets for further analysis only if the observed P value of the GoFtest was smaller than 0.05 (analyses 2) or larger than 0.05 (analysis3). Note that analysis 3 assumes a specific association, thusdata sets with P ‡ 0.05 are cases where the null model was met,that is, indicating significant large-scale association. All analyseswere performed with the Programita software (Wiegand & Moloney2004). Details on edge correction and estimators of theg 12 (r), g(r), K(r) andK 12 (r) can be found in Wiegand & Moloney(2004). Following Diggle (2003), we calculated the D 12 functionwithout edge correction.ResultsANALYSIS 1: DETECTION OF OVERALL NON-RANDOMPATTERNINGUnivariate caseThe 15 species studied at the CBS plot showed stemmaps with diverse spatial patterns (Fig. 2). Use of theCSR null model revealed that all species showed significantclustering at some or all scales, thus confirming ourexpectation.Bivariate caseThe various types of non-random effects in interspecificassociation were not equally distributed among the 210species pairs analysed, and their relative frequency dependedon scale (Fig. 3; Fig. S1 in Supporting Information). Themost notable result is that roughly half of the species pairsshowed segregation at all three spatial scales analysed (113,115 and 110 cases at scales 6, 30 and 50 m, respectively).Thus, trees of different species co-occur less often thanexpected by chance. Non-significant associations (i.e. nodeparture from the null model detected by the two test statisticsK 12 and D 12 ) were relatively frequent in small neighbourhoods(60 cases at 6 m), but stabilized for largerneighbourhoods at a frequency of approximately 20% (38and 34 cases at scales of 30 and 50 m, respectively). Interestingly,partial overlap was rare in the 6-m neighbourhood(nine cases; Fig. 3a), but more frequent in the 30 and 50-mneighbourhoods (42 and 53 cases, respectively; Fig. 3b,c).Mixing occurred in 28 cases at the 6-m scale but only 13 and10 times at the 30 and 50-m scale respectively. The dominanceof segregation and partial overlap confirms our expectationbased on the segregation hypothesis. Figure S1 indicates thatmost changes in the relative frequency of the different bivariateassociation types occur at scales smaller than 20 m.ANALYSIS 2: ISOLATION OF SMALL-SCALE EFFECTSUnivariate caseThe GoF test revealed for nine of the 15 species (60%) significantdepartures from the heterogeneous Poisson null model atthe scale of 0–20 m. Seven species (47%) showed small-scaleaggregation (Fig. 4a,b), the two species Acer mono (Fig. 4c)and Quercus mongolica (Fig. 4d) showed significant regularityat scale r = 0–2 m, and the other six species followed the nullmodel.In order to roughly estimate the effect of scale on speciesspatial pattern (i.e. regularity and aggregation), we counted thenumber of species (using only species where the rank of GoFtest was > 190) for each scale r where the pair-correlationfunction was above or below the fifth-highest or fifth-lowestvalue of the pair-correlation function in the 199 Monte Carlosimulations. The frequency of aggregation peaked at a scalebetween 0 and 4 m, and aggregation effects disappeared atscales r > 10 m (Fig. 5a). Figure 5a reveals repulsion at intermediatescales larger than 10 m. Thus, the clusters have a tendencyto be regularly distributed. Most significant effectsalready disappeared at 10 m, confirming that separation ofscales occurred. Note that strong aggregation at small scalessuggests repulsion at larger scales when tested with the heterogeneousPoisson null model. The regularity effects at largerÓ 2010 The Authors. Journal compilation Ó 2010 British Ecological Society, Journal of Ecology, 98, 674–686127

680 X. Wang et al.(b) Acer triflorum -Tilia mandshuricum0 100 200 300 400 5000 100 200 300 400 5000 100 200 300 400 500 0 100 200 300 400 500Classification axis M(d) Acer mandshuricum -Prunus padus0 100 200 300 400 500210–1–2–3(a)TypeIIType I(c) Betula platyphylla -Phellodendron amurense0 100 200 300 400 500 0 100 200 300 400 500Type IIIType IV–4–1 –0.5 0 0.5Classification axis P(e) Acer mandshuricum-Quercus mongolica0 100 200 300 400 500Fig. 2. Classification of large-scale associations within the CBS plot at a 30-m scale. Diagram (a) shows the allocation of the large-scale associationof the 210 species pairs based on the classification axes defined in eqn (1). Diagrams b, c, d and e represent examples for type II associationwith partial overlap, type III association with mixing, type I association with segregation and type IV association, respectively. Solidcircle = species 1, open circle = species 2.Classification axis M2 (a) 6 m(b) 30 m (c) 50 m10–1–2–3Partial overlapSegregation–4–1.0 –0.5 0.0Classification axis PMixing–1.0 –0.5 0.0Classification axis P–1.0 –0.5 0.0Classification axis PFig. 3. Overall assessment of non-random interspecific patterns at the neighbourhood scale r = 6, 30 and 50 m. The circle symbols indicate the‘not significant’ associations (i.e. no departure from the null model detected by K 12 and D 12 ). Other conventions are as in Fig. 2a.scales (e.g. > 20 m) depicted in Fig. 5a may be primarily dueto this effect.Bivariate caseAtotalof15· 14 = 210 bivariate point pattern analyses forall pairs of the n = 15 species were performed (Fig. 6;Appendix S1, Table S1). The GoF test revealed significantdepartures from the null model for 65 species pairs (31% of allcases); in 55 cases, the small-scale association was negative(repulsion), thus confirming our expectations, and in the other10 cases, the small-scale association was positive (attraction).However, note that approximately five of the 10 cases of significantattraction may arise by chance because this analysis has aÓ 2010 The Authors. Journal compilation Ó 2010 British Ecological Society, Journal of Ecology, 98, 674–686128

Analysing complex spatial association patterns 6814030(a)Betula platyphylla8060(b)Tilia mandshuricaFig. 4. Examples for univariate analyses.Shown are the pair-correlation functions ofthe data in dependence on scale r (solidsquares), the expected g function under theheterogeneous Poisson null model (opensquares; average of the simulation of nullmodel) and the simulation envelopes (solidlines) being the fifth-lowest and fifth-highestg(r) values of the 199 simulations of the nullmodel. The horizontal lines give the expectedg function for random patterns.Univariat pair-correlation function2010000 10 20 30 40 0 10 20 30 4021.510.5(c)Acer mono40201.50.50 10 20 30 40 0 10 20 30 40Scale (m)21(d)Quercus mongolica5%typeIerrorrate.TherankoftheGoFtestcorrelatedstrongly and positively with the product of the number of treesof the two species (r SP = 0.52; P < 0.01), moderately andpositively with the number of trees of the focal species(r SP =0.38;P < 0.01), and negatively with clustering of thefocal species (r SP = )0.39; P < 0.01; Appendix S1,Table S2). As expected, we found that negative associationsoccurred more frequently among species with the same attributefamily (which was in our case the same as genus), fruittype and habitat associations (Appendix S1, Table S3). Positiveassociations occurred less frequently among species withthesameattributefamilyandmorefrequentlyforthosespeciesoccupying the same canopy layer (Appendix S1, Table S3).We also counted, for each scale r, the number of speciesfor which the pair-correlation function was above or belowthe simulation envelopes (using only species where therank of GoF test was > 190). Repulsion occurred morefrequently than attraction; pair frequencies for repulsionand attraction peaked at the 0 and 2-m scales with 49 and9 pairs, respectively (Fig. 5b). Significant effects were rareat scales r > 10 m, again demonstrating separation ofscales. If no separation of scales was present, significanteffects would only gradually disappear at the neighbourhoodradius of R = 30 m used for the heterogeneousPoisson null model.LARGE-SCALE ASSOCIATIONS (ANALYSIS 3)The GoF test revealed that 17 species pairs (8.1%) followedthe null model at the distance interval 50–250 m (compared to3.3% at Sinharaja), i.e. they showed significant large-scaleassociation (see Appendix S2, Fig. S2). However, significanteffectswereonlysymmetricforonespeciespair(Ulmus japonica–Maackiaamurensis;see Appendix S2). Note that the specificnull model tested here is also met if species 2 is onlydistributed inside a subarea of the area broadly occupied byspecies 1 (see Appendix S3). Thus, large-scale associationswere mostly of the partial overlap type.Twelve of the 15 species studied at the CBS plot showed significantlarge-scale association with at least one other species(compared with 42 of 46 species at Sinharaja). In the role ofspecies 1, the shade-tolerant canopy species Tilia amurensisshowed significant association with five species followed by theshade-tolerant midstorey species Acer mono with four associations.The shade-tolerant midstorey species Acer tegmentosumshowed, in the role of species 2, four associations to other species.The species without significant large-scale association toany other species were Pinus koraiensis, Tilia mandshurica andPhellodendron amurense. Relating significant large-scale associationto species properties, however, did not reveal such clearrules as for analysis 2. The reason for this is probably that lar-Fig. 5. Assessment of scale effects in spatialassociation patterns. The figures show thenumber of species (univariate) and speciespairs (bivariate) where the observed pair-correlationfunction for a given scale is outsidethe Monte Carlo simulation envelopes, beingthe fifth-lowest and fifth-highest values ofthe 199 simulated g(r).Number of species(a) Univariate analysis (b) Bivariate analysis10608642AggregationRegularityRandom00 10 20 30 40Scale (m)Number of species pairs40200AttractionRepulsion0 10 20 30 40Scale (m)Ó 2010 The Authors. Journal compilation Ó 2010 British Ecological Society, Journal of Ecology, 98, 674–686129

682 X. Wang et al.1.5(a)Pinus koraiensis-3(b) Ulmus japonica-1.5(c) Quercus mongolica-Tilia amurensisMaackia amurensisTilia amurensis211Bivariate pair-correlation function10.50.5000 10 20 30 40 0 10 20 30 40 0 10 20 30 401.5 (d) Acer pseudo-sieboldianum- 1.5 (e) Acer pseudo-sieboldianum- 2 (f) Maackia amurensis-Tilia amurensisFraxinus mandshuricaAcer tegmentosum1.51110.50.50.50 10 20 30 4000 10 20 30 4000 10 20 30 40Scale (m)Fig. 6. Examples of significant small-scale associations in bivariate patterns. Shown are the bivariate g 12 pair-correlation function of the data independence on scale r (solid diamond), the expected g 12 function under the heterogeneous Poisson null model (open diamond) and the simulationenvelopes (solid lines) being the fifth-lowest and fifth-highest values of the Monte Carlo simulations of the null modes. The horizontal lines givethe expected g 12 function for independent patterns.ger-scale associations may dependinacomplexwayonfactorssuch as habitat association, dispersal characteristics, historicevents and successional dynamics.DiscussionIn this study, we conducted a comprehensive spatial patternanalysis to assess species associations among large trees of15 common species at the fully mapped 25 ha CBS temperateforest plot in north-eastern China. These species comprised97.7% of all trees at the study plot with d.b.h. largerthan 10 cm. Our study revealed marked differences to findingsof a similar study of a hyper-diverse tropical forest(Wiegand, Gunatilleke & Gunatilleke 2007a). Most importantly,we found that one third of all species pairs at theCBS plot showed significant and mostly negative small-scaleassociations. Negative associations occurred more frequentlythan expected by chance between species which shared oneor more attributes such as family, fruit type or habitat association.This suggests that the spatial pattern of larger treesshowed signatures of competitive effects between speciessharing ecological attributes. Analogous analyses of datafrom tropical forests at Sinharaja (Sri Lanka) and BarroColorado Island (Panama) revealed significant small-scaleassociations (Wiegand, Gunatilleke & Gunatilleke 2007a;Table2;T.Wiegand,unpubl.data),afigurewhichisthesame as the error rate of the analysis. However, we cannotexclude the possibility that these differences may owe asmuch to differences in other factors such as heterogeneity,range of conditions, species richness or patterning of abioticconditions. Further comparisons across sites will help toassess the effects of these possibilities.Despite substantial differences, our analyses revealed interestingsimilarities between the spatial structure of the Sinharajaand the CBS forests (Table 2). For example, we found thatmost species co-occurred in small neighbourhoods less oftenthan expected by chance, and only 8% of all species pairsshared roughly the same areas of the plot. At the Sinharaja forest,essentially the same results were found (Wiegand, Gunatilleke& Gunatilleke 2007a; Table 2). Thus, individuals ofdifferent species show a clear tendency to ‘avoid’ each other,both at small and larger scales.THE SEGREGATION HYPOTHESISSpatial segregation is a mechanism known from theoreticalmodels to promote coexistence (Pacala 1997; Pacala & Levin1997; Chesson 2000). The underlying mechanism is that spatialsegregation among species decreases the probability of interspecificencounters, with the effect that the importance of interspecificcompetition decreases relative to intraspecificcompetition and competitive exclusion is retarded. We foundthat all 15 species were significantly aggregated and that spatialsegregation and partial overlap were the most common bivariateassociation types at the CBS plot. Thus, trees of differentspecies have a tendency to avoid each other. This observationis in line with the segregation hypothesis. Suppression of competitivelysuperior species by this spatial configuration preventsÓ 2010 The Authors. Journal compilation Ó 2010 British Ecological Society, Journal of Ecology, 98, 674–686130

Analysing complex spatial association patterns 683Table 2. Comparison of results between the temperate forest at Changbaishan (CBS) and the tropical Sinharaja forest. Results for Sinharaja aretaken from Wiegand, Gunatilleke & Gunatilleke (2007a) or were obtained after reanalysis of the raw data. Given are the percentages of cases in agiven category. At the CBS plot, we analysed 15 species and 210 species pairs; at Sinharaja 46 species and 2070 species pairs(a) Bivariate association type (analysis 1)*Study site Segregation (type I) Partial overlap (type II) Mixing (type III) Type IV No sig. effectCBS 54.8% 20% 6.2% 0.9% 18.1%Sinharaja 50.2% 34% 5.4% 0.1% 10.3%(b) Univariate analysis 2Small-scale aggregation Small-scale regularity No sig. effectCBS 47% 13% 40%Sinharaja 46% 6% 48%(c) Bivariate analysis 2Small-scale attraction Small-scale repulsion No sig. effectCBS 26% 5% 69%Sinharaja 3% 3% 94%*The ‘no significant effect’ type showed for the two axes M(r) and P(r) no significant departure from the null model, and the four typesI to IV are defined by positive or negative departures from the null model in the M(r) and ⁄ or P(r) axis (see Fig. 2a).(or retards) the elimination of competitively inferior species(Stoll & Prati 2001). This effect can tip the balance fromcompetitive exclusion to coexistence and promote speciesdiversity (Kareiva 1990; Tilman 1994; Rees, Grubb & Kelly1996; Pacala 1997; Stoll & Weiner 2000).The observed segregation may also explain why only 31%of all species pairs showed significant small-scale associations(analysis 2). This hypothesis is supported by our finding thatsignificant small-scale effects were more likely if the two specieswere more abundant. This result is somewhat to be expectedbecause the power of the GoF test (type-II error rate) is a functionof abundance (e.g. Plotkin et al. 2000). This in turn isbecause fewer direct encounters between a species pair at agiven spatial scale r will weaken the power of the pair-correlationfunction to detect significant effects. However, this is notonly a statistical issue but may also have additional biologicalrelevance. Recent studies on species abundance and interactionstrength in ecological networks suggest a strong dependency ofinteraction strength on abundance, i.e. interactions tend to bestronger if individuals of two species have more direct encounters(e.g. the ‘abundance-asymmetry hypothesis’; Vázquezet al. 2007). Interestingly, the interaction strength (measuredas the rank of the GoF test) at the tropical forest in Sinharajadepended only weakly on abundances (Wiegand, Gunatilleke& Gunatilleke 2007a). Similar results were obtained from nearest-neighbourpair analyses by Lieberman & Lieberman (2007)in the tropical rainforest at La Selva, Costa Rica, and by Perryet al. (2009) in shrublands of the Eneabba sand plain, Australia.INTERPRETATION OF SIGNIFICANT SPECIESASSOCIATION PATTERNSPrevious studies showed that species within the same guildshould compete more strongly and show mutual negative correlationscompared with correlations outside the guild (Wilson& Roxburgh 2001; Zaal, Liana & David 2005). We found thatnegative associations between species of the same familyoccurred more frequently than expected by chance and thatpositive associations occurred less frequently. For example,the five Aceraceae species that share almost all ecological propertieslisted in Table 1 show only negative (or no) small-scaleassociations. Negative associations were also more frequentbetween species that shared the same fruit type and habitatassociation, indicating competitive associations. Positive associationswere more frequent for species sharing the same canopylayer. The latter may be explained by successionaldynamics.SEPARATION OF SCALESOur analyses revealed, in line with the study at Sinharaja and astudy of individual species area relationships (Wiegand et al.2007b), separation of scales in the spatial association structureof the forest community. The observed separation of scales isin need of explanation and is a promising candidate for anindicative pattern. Note that this finding did not require anybiological assumptions, because conditioning of the null modelto preserve the large-scale spatial patterns selectively is a purelytechnical issue. Our analysis revealed that there is a separationof scales but does not provide direct evidence on the processesand mechanisms involved. We found a pattern in the data, butsubsequent research needs to derive specific hypotheses on processesand mechanisms that could be tested in the field. Onehypothesis is that the patterns at scales > 30 m would be drivenby limited dispersal and habitat conditions, which typicallyvary along environmental gradients, which, in turn, are oftenrelated to topographical features (Harms et al. 2001; Valenciaet al. 2004; Gunatilleke et al. 2006; John et al. 2007), and thatthe patterns at scales smaller than 10–15 m are driven by directintraspecific interactions. Variation in the local light environmentis another potential factor that may explain theseÓ 2010 The Authors. Journal compilation Ó 2010 British Ecological Society, Journal of Ecology, 98, 674–686131

684 X. Wang et al.fine-scale patterns. However, while local light environment iscertainly important for recruits, it may be less important forthe larger trees analysed here. Clearly, in small neighbourhoodsone would expect repulsion because there are physicallimits to how close individual stems can be based on the sizesoftheircrowns(Wiegandet al. 2007b). However, our findingthat most of the significant negative small-scale associationsoccurred within guilds suggests involvement of interactionsother than those due to physical tree size.POINT PATTERN ANALYSIS AND POINT PROCESSMODELLINGTo answer our ecological questions, we used conceptually simplenonparametric techniques in the spirit of an exploratoryanalysis (Law et al. 2009). This is an important initial step toreveal basic indicative patterns of complex multivariate datasets such as the one used in our study. A promising next step toelucidate species interactions in more detail is the use of recentdevelopments in point-process modelling (e.g. Grabarnik &Sa¨rkka¨ 2009; Illian, Moller & Waagepetersen 2009). Suchmodels assume parametric interaction structures whereby thefitted parameters contain information on the strength anddirection of the interactions. Point-process models also allowadjustments for the effects of the remaining species on the associationof species pairs (Illian, Moller & Waagepetersen 2009).This is not possible with our approach, but may not be a severelimitation given the strong segregation between species foundin our study forest. However, point-process models that considerall pairwise interactions quickly become intractable forspecies-rich communities. Therefore, Grabarnik & Särkkä(2009) and Illian, Moller & Waagepetersen (2009) used hierarchicalmodels to simplify the interactions structure based onbiological arguments (e.g. small trees do not influence the patternof large trees). Future developmentofpoint-processmodelswill allow more detailed investigation of the spatialorganization of species-rich communities.ConclusionsOur study is unique in that it presents comprehensive analyseson the spatial association structure of most of the tree communityin a megaplot of a temperate forest. Clearly, there is a longtradition of spatial pattern analyses in temperate forests (e.g.Kenkel 1988; Duncan 1991), but most of these studies havefocused on a few dominant or canopy species in these forests(Hao et al. 2007; but see Kubota, Kubo & Shimatani 2007).Because of the analogy in sampling and analyses with megaplotsof tropical forests, our study is a first step towards moredetailed investigations of similarities and differences betweentropical and temperate forests in their complex spatial associationstructures. We found for large trees both indications forcompetitive interactions due to niche overlap and avoidanceeffects due to spatial segregation patterns. Analogous analysesin the species-rich tropical forests at Sinharaja showed almostno evidence for competitive interactions, but strong segregationpatterns (Wiegand, Gunatilleke & Gunatilleke 2007a). Wehypothesize that the balance between these two effects mayvary with species richness and that species-poorer forestsshould show stronger (or more frequent) significant interspecificassociation patterns than species-rich forests. We took afirst step by revealing patterns in the complex spatial data, butcomparative studies among a larger number of plots will benecessary in order to prove our hypotheses and to determinewhether these observations are indeed indicative patterns. Weare confident that analysis of spatial patterns may eventuallyprovide the additional information required in order to decidebetween competing hypotheses of community assemblage anddynamics, and to stimulate new theoretical developments.AcknowledgementsThis study was sponsored by the Knowledge Innovation Program of theChinese Academy of Sciences (KZCX2-YW-QN402), the National NaturalScience Foundation of China (40971286 and 30870400) and theNational Key Technologies R&D Program of China (2008BAC39B02).X.W. and T.W. revised the manuscript during a workshop funded by theCenter for Tropical Forest Science (CTFS). T.W. was supported by theERC advanced grant 233066. We thank Egbert Leigh, Micheal Papaikand Justin Calabrese for checking the English and providing useful comments.We also thank anonymous referees for constructive criticism onearlier drafts of the manuscript.ReferencesAmarasekare, P. (2003) Competitive coexistence in spatially structured environments:a synthesis. Ecology Letters, 6, 1109–1122.Bell, G. (2001) Neutral macroecology. Science, 293, 2413–2418.Bell, G. (2005) The co-distribution of species in relation to the neutral theory ofcommunity ecology. Ecology, 86, 1757–1770.Chave, J. (2004) Neutral theory and community ecology. Ecology Letters, 7,241–253.Chesson, P. (2000) General theory of competitive coexistence in spatiallyvarying environments. Theoretical Population Biology, 58, 211–237.Condit, R. (1998) Tropical Forest Census Plots. Springer, Berlin.Dai, L., Gu, H., Shao, G. & Wang, Q. (2004) The Broad-Leaved Korean PineMixed Forest on Changbai Mountain of China. Liaoning Science andTechnology Publishing House, Shenyang (in Chinese).Diggle, P.J. (2003) Statistical Analysis of Spatial Point Patterns. Arnold, London.Duncan, R.P. (1991) Competition and the coexistence of species in a mixedpodocarp stand. Journal of Ecology, 79, 1073–1084.Gilbert, B. & Lechowicz, M.J. (2004) Neutrality, niches, and dispersal in a temperateforest understorey. Proceedings of the National Academy of Sciencesof the United States of America, 101, 7651–7656.Grabarnik, P. & Särkkä, A. (2009) Modelling the spatial structure of foreststands by multivariate processes with hierarchical interactions. EcologicalModelling, 220, 1232–1240.Grimm, V., Revilla, E., Berger, U., Jeltsch, F., Mooij, W.M., Railsback, S.F.,Thulke, H.H., Weiner, J., Wiegand, T. & DeAngelis, D.L. (2005) Patternorientedmodeling of agent-based complex systems: lessons from ecology.Science, 310, 987–991.Gunatilleke, C.V.S., Gunatilleke, I.A.U.N., Esufali, S., Harms, K.E., Ashton,P.M.S., Burslem, D.F.R.P. & Ashton, P.S. (2006) Species-habitat associationsin a Sri Lankan dipterocarp forest. Journal of Tropical Ecology, 22,371–384.Hao, Z., Zhang, J., Song, B., Ye, J. & Li, B. (2007) Vertical structure and spatialassociations of dominant tree species in an old-growth temperate forest. ForestEcology and Management, 252,1–11.Hao, Z., Li, B., Zhang, J., Wang, X., Ye, J. & Yao, X. (2008) Broad-leavedKorean pine mixed forest plot in Changbaishan (CBS) of China: communitycomposition and structure. Acta Phytoecologica Sinica, 32, 238–250 (in Chinese).Harms, K.E., Condit, R., Hubbell, S.P. & Foster, R.B. (2001) Habitat associationsof trees and shrubs in a 50-ha neotropical forest plot. Journal ofEcology, 89, 47–959.Ó 2010 The Authors. Journal compilation Ó 2010 British Ecological Society, Journal of Ecology, 98, 674–686132

Analysing complex spatial association patterns 685Hoshino, D., Nishimura, N. & Yamamoto, S. (2001) Age, size structure andspatial pattern of major tree species in an old-growth Chamaecyparis obtusaforest, central Japan. Forest Ecology and Management, 152,31–43.Hubbell, S.P. (2001) The Unified Neutral Theory of Biodiversity and Biogeography.Princeton University Press, Princeton, NJ, USA.Hubbell, S.P., Ahumada, J.A., Condit, R. & Foster, R.B. (2001) Local neighborhoodeffects on long-term survival of individual trees in a neotropical forest.Ecological Research, 16, 859–875.Illian, J.B., Moller, J. & Waagepetersen, R.P. (2009) Hierarchical spatial pointprocess analysis for a plant community with high biodiversity. Environmentaland Ecological Statistics, 16, 389–405.Illian, J., Penttinen, A., Stoyan, H. & Stoyan, D. (2008) Statistical Analysis andModeling of Spatial Point Patterns. Wiley, Chichester, UK.John, R., Dalling, J.W., Harms, K.E., Yavitt, J.B., Stallard, R.F., Mirabello,M., Hubbell, S.P., Valencia, R., Navarette, H., Vallejo, M. & Foster, R.B.(2007) Soil nutrients influence spatial distributions of tropical tree species.Proceedings of the National Academy of Sciences of the United States ofAmerica, 104, 864–869.Kareiva, P. (1990) Population dynamics in spatially complex environments:theory and data. Philosophical Transaction of the Royal Society of London B,330, 175–190.Kenkel, N.C. (1988) Pattern of self-thinning in jack pine: testing the randommortality hypothesis. Ecology, 69, 1017–1024.Kubota, Y., Kubo, H. & Shimatani, K. (2007) Spatial pattern dynamics over10 years in a conifer ⁄ broadleaved forest, northern Japan. Plant Ecology,190, 143–157.Law, R., Illian, J., Burslem, D.F.R.P., Gratzer, G., Gunatilleke, C.V.S. & Gunatilleke,I.A.U.N. (2009) Ecological information from spatial patterns ofplants: insights from point process theory. Journal of Ecology, 97, 616–628.Levin, S.A. (1992) The problem of pattern and scale in ecology. Ecology, 73,1943.Lieberman, M. & Lieberman, D. (2007) Nearest-neighbor tree species combinationsin tropical forest: the role of chance, and some consequences of highdiversity. Oikos, 116, 377–386.Loosmore, N.B. & Ford, E.D. (2006) Statistical inference using the G or Kpoint pattern spatial statistics. Ecology, 87, 1925–1931.McGill, B.J., Enquist, B.J., Weiher, E. & Westoby, M. (2006) Rebuilding communityecology from functional traits. Trends in Ecology and Evolution, 21,178–185.McIntire, E.J.B. & Fajardo, A. (2009) Beyond description: the active and effectiveway to infer processes from spatial patterns. Ecology, 90,46–56.Mouquet, N. & Loreau, M. (2003) Community patterns in source-sink metacommunities.American Naturalist, 162, 544–577.Murrell, D.J., Purves, D.W. & Law, R. (2001) Uniting pattern and process inplant ecology. Trends in Ecology and Evolution, 16, 529–530.Pacala, S.W. (1997) Dynamics of plant communities. Plant Ecology (ed. M.J.Crawley), pp. 532–555. Blackwell Science, Oxford.Pacala, S.W. & Levin, S. (1997) Biologically generated spatial pattern and thecoexistence of competing species. Spatial Ecology (eds D. Tilman & P. Kareiva),pp. 204–232. Princeton University Press, Princeton, NJ.Perry, G.L.W., Enright, N.J., Miller, B.P. & Lamont, B.B. (2009) Nearestneighbourinteractions in species-rich shrublands: the roles of abundance,spatial patterns and resources. Oikos, 118, 161–174.Plotkin, J.B., Potts, M.D., Leslie, N., Manokaran, N., LaFrankie, J. & Ashton,P.S. (2000) Species-area curves, spatial aggregation, and habitat specializationin tropical forests. Journal of Theoretical Biology, 207, 81–99.Purves, D.W. & Law, R. (2002) Fine-scale spatial structure in a grassland community:quantifying the plant’s-eye view. Journal of Ecology, 90,121.Purves, D.W. & Pacala, S.W. (2005) Ecological drift in niche-structured communities:neutral pattern does not imply neutral process. Biotic Interactionsin the Tropics: Their Role in the Maintenance of Species Diversity. (eds D.Burslem, M. Pinard & S. Hartley), pp. 107–138. Cambridge University Press,Cambridge, UK.Rees, M., Grubb, P.J. & Kelly, D. (1996) Quantifying the impact of competitionand spatial heterogeneity on the structure and dynamics of a four-speciesguild of winter annuals. American Naturalist, 147,1–32.Ricklefs, R.E.. (1990) Ecology. W. H. Freeman and Company, New York.Ripley, B.D. (1981) Spatial Statistics. Wiley, New York.Stoll, P. & Newbery, D.M. (2005) Evidence of species-specific neighborhoodeffects in the dipterocarpaceae of a Bornean rain forest. Ecology, 86, 3048–3062.Stoll, P. & Prati, D. (2001) Intraspecific aggregation alters competitive interactionsin experimental plant communities. Ecology, 82, 319–327.Stoll, P. & Weiner, J. (2000) A neighborhood view of interactions amongplants. The Geometry of Ecological Interactions: Simplifying Spatial Complexity(eds U. Dieckmann, R. Law & J. Metz), pp. 11–27, Cambridge UniversityPress, Cambridge, UK.Stoyan, D. & Stoyan, H. (1994) Fractals, Random Shapes and Point Fields:Methods of Geometrical Statistics. Wiley, Chichester, UK.Tilman, D. (1994) Competition and biodiversity in spatially structured habitats.Ecology, 75, 2–16.Tilman, D. (2004) Niche tradeoffs, neutrality, and community structure: a stochastictheory of resource competition, invasion, and community assembly.Proceedings of the National Academy of Sciences of the United States ofAmerica, 101, 10854–10861.Tilman, D. & Kareiva, P. (1997) Spatial Ecology: The Role of Space in PopulationDynamics and Interspecific Interactions. Princeton University Press,Princeton, NJ.Turkington, R.A. & Harper, J.L. (1979) The growth, distribution and neighborrelationships of Trifolium repens in a permanent pasture. I. Ordination, patternand contact. Journal of Ecology, 67, 201–208.Uriarte, M., Condit, R., Canham, C.D. & Hubbell, S.P. (2004) A spatiallyexplicit model of sapling growth in a tropical forest: does the identity ofneighbours matter? Journal of Ecology, 92, 348–360.Uriarte, M., Hubbell, S.P., John, R., Condit, R. & Canham, C.D. (2005)Neighborhood effects on sapling growth and survival in a neotropical forestand the ecological equivalence hypothesis. Biotic Interactions in the Tropics:Their Role in the Maintenance of Species Diversity (eds D. Burslem, M. Pinard& S. Hartley), pp. 89–106. Cambridge University Press, Cambridge,UK.Valencia, R., Foster, R.B., Villa, G., Condit, R., Svenning, J.C., Hernandez,C., Romoleroux, K., Losos, E., Magard, E. & Balslev, H. (2004) Tree speciesdistributions and local habitat variation in the Amazon: large forest plot ineastern Ecuador. Journal of Ecology, 92, 214–229.Vázquez, D.P., Melian, C.J., Williams, N.M., Bluthgen, N., Krasnov, B.R. &Poulin, R. (2007) Species abundance and asymmetric interaction strength inecological networks. Oikos, 116, 1120–1127.Wang, X., Hao, Z., Ye, J., Zhang, J., Li, B. & Yao, X. (2008a) Spatial patternof diversity in an old-growth temperate forest in Northeastern China. ActaOecologica, 33, 345–354.Wang, X., Hao, Z., Ye, J., Zhang, J., Li, B. & Yao, X. (2008b) Spatial variationof species diversity across scales in an old-growth temperate forest of China.Ecological Research, 23, 709–717.Wang, X., Hao, Z., Zhang, J., Lian, J., Li, B., Ye, J. & Yao, X. (2009) Tree sizedistributions in an old-growth temperate forest. Oikos, 118,25–36.Watt, A.S. (1947) Pattern and process in the plant community. Journal of Ecology,35, 1–22.Wiegand, T., Gunatilleke, C.V.S. & Gunatilleke, I.A.U.N. (2007a) Speciesassociations in a heterogeneous Sri Lankan dipterocarp forest. AmericanNaturalist, 170,77–95.Wiegand, T., Huth, A. & Martıńez, I. (2009) Recruitment in tropical tree species:revealing complex spatial patterns. American Naturalist, 174, E106–E140.Wiegand, T. & Moloney, K.A. (2004) Rings, circles, and null-models for pointspattern analysis in ecology. Oikos, 104, 209–229.Wiegand, T., Jeltsch, F., Hanski, I. & Grimm, V. (2003) Using pattern-orientedmodeling for revealing hidden information: a key for reconciling ecologicaltheory and application. Oikos, 100, 209–222.Wiegand, T., Gunatilleke, C.V.S., Gunatilleke, I.A.U.N. & Huth, A. (2007b)How individual species structure diversity in tropical forests. Proceedings ofthe National Academy of Sciences of the United States of America, 104,19029–19033.Wilson, J.B. & Roxburgh, S.H. (2001) Intrinsic guild structure: determinationfrom competition experiments. Oikos, 92, 189–192.Yang, H. & Li, D. (1985) Distribution patterns of dominant tree species onnorthern slope of Changbai Mountain. Research of Forest Ecosystem, 5, 1–14 (in Chinese).Zaal, K., Liana, K. & David, K. (2005) Small-scale guild proportitions andniche complementarity in a Caucasian subalpine hay meadow. Journal ofVegetation Science, 16, 565–570.Zang, R., Liu, T., Guo, Z. & Gao, W. (1998) Gap disturbance regime in abroadleaved-Korean pine forest in the Changbai mountain natural reserve.Acta Phytoecologica Sinica, 22, 135–142 (in Chinese).Zhao, D. (1981) Preliminary study of the effects of volcano on vegetationdevelopment and succession. Forest Ecosystem Research, 2, 81–87 (inChinese).Received 9 September 2009; accepted 1 February 2010Handling Editor: Kyle HarmsÓ 2010 The Authors. Journal compilation Ó 2010 British Ecological Society, Journal of Ecology, 98, 674–686133

686 X. Wang et al.Supporting InformationAdditional Supporting Information may be found in the online versionof this article:Appendix S1. Additional information on analysis 2.Appendix S2. Association matrix for large-scale association.Appendix S3. Why is the null model 3 met if species 2 is onlydistributed inside a subarea of the area occupied by species 1?As a service to our authors and readers, this journal provides supportinginformation supplied by the authors. Such materials may be reorganizedfor online delivery, but are not copy-edited or typeset.Technical support issues arising from supporting information (otherthan missing files) should be addressed to the authors.Ó 2010 The Authors. Journal compilation Ó 2010 British Ecological Society, Journal of Ecology, 98, 674–686134

Acta Oecologica 36 (2010) 29–38Contents lists available at ScienceDirectActa Oecologicajournal homepage: www.elsevier.com/locate/actoecOriginal articleSpatial patterns and associations of six congeneric speciesin an old-growth temperate forestJian Zhang a,b , Bo Song c , Bu-Hang Li a,b ,JiYe a , Xu-Gao Wang a , Zhan-Qing Hao a, *a Institute of Applied Ecology, Chinese Academy of Sciences, Shenyang 110016, PR Chinab Department of Renewable Resources, University of Alberta, Edmonton, Alberta T6G 2H1, Canadac Belle W. Baruch Institute of Coastal Ecology and Forest Science, Clemson University, P.O. Box 596, Georgetown, SC 29442, USAarticleinfoabstractArticle history:Received 12 January 2009Accepted 28 September 2009Published online 15 October 2009Keywords:Comparative ecologyPoint pattern analysisPair-correlation functionMapleAcerAnalyses of the spatial patterns of pairs of sympatric congeneric species present unique opportunitiesand challenges in explaining species coexistence. In this study, we compared the population structureand spatial patterns of six coexisting species of Acer (maple) in a large-scale 25-ha old-growth temperateforest plot in northern China with the aim of gaining insight into the coexistence of these congeners. Asecond-order spatial point pattern analysis based on the pair-correlation function was used to evaluatespatial patterns and examine intra- and interspecies associations among them. The abundance variedfrom 251 to 6609 individuals, but these species showed similar, distinct L-shaped size-class distributions.All six species were aggregated at small scales. The degree of aggregation of the three species with lowabundance was in fact higher than that for the others at small scales. Aggregated patterns were alsofound for small and large trees at small scales. For spatial associations among the congeners, thegoodness-of-fit (GoF) test revealed significant associations for only 10 of 30 species pairs, indicating noclear evidence for interspecific competition within them. In comparing the relationships among differentsize classes, we found no significant relationship for most of the analyzed species pairs. The GoF testdetected significant associations for 6 of 20 species pairs among large trees of different species, 6 of 30species pairs among small trees of different species, and 11 of 30 species pairs between large and smalltrees of different species. Based on a comparison of population structure, spatial patterns and associations,we infer that for these congeners there exists no significant difference in seed dispersal,competitive ability, or the requirement for specific topographic and light environments. Other factors(e.g., seed production and other habitat requirements) may contribute to the coexistence of thesecongeners and the trade-off between species commonness and rarity.Ó 2009 Elsevier Masson SAS. All rights reserved.1. IntroductionAnalysis of the spatial patterns of tree species is important ingaining a better understanding of the underlying ecologicalprocesses that controlled the development of the observed patterns(Druckenbrod et al., 2005; He et al., 1997; Wiegand et al., 2007). Thespatial patterns of trees and their interactions provide criticalinformation about community structure and species coexistence,and significantly determine reproduction, growth, mortality,dispersal, resource use, gap creation, and understory development(Condit et al., 2000; Wiegand et al., 2007).* Corresponding author. Institute of Applied Ecology, Chinese Academy ofSciences, Wenhua road, No. 72, Shenyang 110016, PR China. Tel.: þ86 24 83970209;fax: þ86 24 83970300.E-mail address: hzq@iae.ac.cn (Z.-Q. Hao).Analysis of the spatial patterns of pairs of sympatric congenericspecies presents unique opportunities and challenges in explainingspecies coexistence (Swenson et al., 2006; Helmus et al., 2007;Mooney et al., 2008; Valiente-Banuet and Verdu, 2008). Because oftheir status as descendants of a common ancestor, sympatriccongeners are more likely to be similar in many phenotypic andecological traits and may utilize a similar set of resources ina similar way (Ackerly and Donoghue, 1998; Mooney et al., 2008).Thus, they may experience relatively intense interspecific competitionthat limits their coexistence (Mooney et al., 2008; Webb et al.,2002); however, in both tropical plant communities with diversespecies and temperate plant communities with relatively fewspecies, many congeners are found to coexist (e.g., Davies et al.,1998; Queenborough et al., 2007; Tanaka et al., 2008). It is thenimportant to ask the following questions: (1) Why are some speciesless abundant and/or less widely distributed than other congeners?What regulates species commonness and rarity? and (2) How do1146-609X/$ – see front matter Ó 2009 Elsevier Masson SAS. All rights reserved.doi:10.1016/j.actao.2009.09.005135

30J. Zhang et al. / Acta Oecologica 36 (2010) 29–38these congeners coexist in the same communities? These questionshave long fascinated ecologists and evolutionary biologists (Brownet al., 1996).Comparative studies of congeneric species have revealed thatless widely distributed species may differ from their widespreadcongeners in terms of particular habitat requirements (Queenboroughet al., 2007; Tanaka et al., 2008), inferior dispersal abilities(Cowling, 2001; Simon and Hay, 2003), and lower investment inreproduction (Byers and Meagher, 1997; Cowling, 2001; Simon andHay, 2003). However, previous studies have reported inconsistentresults. For example, Moora and Jõgar (2006) and Walck et al.(1999) found that a common species was a superior competitor,while Rabinowitz (1981) reported the opposite finding, and Snyderet al. (1994) concluded that the competitive ability of a rarecongener was intermediate between that of two common congeners.Bevill and Louda (1999) reviewed 38 comparative studies ofrelated rare species and common species in which a total of 71response variables were compared, and found no significantdifference among them.There are many factors that might explain the conflicting resultsreported in previous studies (Bevill and Louda, 1999). Indeed, twoimportant factors have been ignored in many previous studies.First, the problem of insufficient sample size makes it difficult totest for general relationships between rarity and competitive ability(Lloyd et al., 2002). Second, many previous studies comparedcongeneric species that occupied different habitats (e.g., Debusscheand Thompson, 2003; Lloyd et al., 2002), which may limit theability to explain the causes for their coexistence (Münzbergová,2005). A large set of quantitative data based on the same or similarhabitat is needed for comparative studies. For example, Davies et al.(1998) compared 11 species of Macaranga in a 52-ha permanentplot of tropical rain forest in Borneo, and concluded that heterogeneityof resource availability (including light, establishmentmicrosites, and soil textural properties) resulted in differences intree distribution, which is important for their coexistence. Queenboroughet al. (2007) compared 16 species of Myristicaceae in a 50-ha permanent plot of tropical rain forest in Amazonian Ecuador,and found no evidence that partitioning of physical habitats couldexplain the coexistence of all 16 closely related species.In the present study, we compared the population structure andspatial patterns of six co-occurring species of Acer (maple) ina large-scale 25-ha plot of old-growth temperate forest with theaim of gaining insight into their coexistence. A second-order spatialpoint pattern analysis based on the pair-correlation function wasperformed to evaluate the spatial patterns and examine intra- andinterspecies associations among these species. We addressed thefollowing specific questions: (1) Are there significant differences inpopulation structure and spatial patterns among these congenericspecies in the forest? (2) How do the spatial patterns and associationschange with scale, size class, and species abundance? and (3)Does strong competition exist for resources and space among thesecongeneric species? Are there significant differences in competitiveability among them?2. Materials and methods2.1. Study siteThe study site is located in the Changbai Mountain NaturalReserve, northeast China (42 23 0 N, 128 05 0 E). The reserve, establishedin 1960, was added to the World Biosphere Reserve Networkin 1980 as part of the Man and the Biosphere Project. The presentstudy site is located in broad-leaved Korean pine (Pinus koraiensis)mixed forest, the most common vegetation type in the region interms of species composition and community structure. The standsoil is classified as dark brown forest soil. Mean annualprecipitation is approximately 700 mm, with most occurring fromJune to September (480–500 mm). Mean annual temperature is2.8 C, with a January mean of 13.7 C and a July mean of 19.6 C(Yang et al., 1985). The mean age of overstory trees is about 300years. The main canopy species include Korean pine, Tilia amurensis,Quercus mongolica, Fraxinus mandshurica, and Ulmus japonica(Yang et al., 1985).A 25 ha (500 500 m) forest plot was established in the forestduring the summer of 2004. The plot was chosen in the core zone ofthe reserve to avoid the influence of human activities. Following thefield protocol of the Center for Tropical Forest Science (CTFS) of theSmithsonian Institution, all stems, including branches, 1 cmdiameter at breast height (DBH, 1.3 m above the ground) wereidentified, measured, mapped, and monitored (Condit, 1998). Theterrain at the site is relatively gentle; elevation ranges from 791.8 to809.5 m (mean, 801.5 m). The total number of living individualscounted in the first census (2004) was 38902, consisting of 52species, 32 genera, and 18 families. Abundance varied by over fourorders of magnitude, from only one individual (Actinidia kolomikta,Rosa davurica, and Sorbus pohuashanensis) to 7834 individuals ofCorylus mandshurica. Mean stand density was 1556 living trees perhectare. Mean basal area was 43.2 m 2 per ha (Hao et al., 2008).2.2. Study speciesThe genus Acer (maple) of the family Aceraceae containsapproximately 200 species, most of which are native to Asia, butseveral are native to North America, Europe, and northern Africa(van Gelderen et al., 1994; Xu, 1996). The genus is well representedin East Asia, especially in China. There are about 154 Acer speciesfound in China, which accounts for more than 75% of known species(Xu, 1996). The different species vary from shade-tolerant understoryshrubs and small trees to pioneers with high light requirements.In northeastern China, Acer is a conspicuous component ofseveral forest types, due to both the large number of species andhigh stem densities. In particular, in broad-leaved Korean pinemixed forest, the typical forest type of northeastern China, thereexist nine Acer species. In our plot, Acer is the largest genus, witheight species in the plot; the second-largest genus includes onlythree species. The total abundance of Acer is 18,024, which accountsfor 46.33% of the total abundance of all species in the plot. Six mainAcer species were chosen to compare their population structureand spatial patterns in this study because two others have very fewindividuals in the plot. The selected species were Acer barbinerve(ACBA), Acer mandshuricum (ACMA), Acer mono (ACMO), Acerpseudosieboldianum (ACPS), Acer tegmentosum (ACTE), and Acertriflorum (ACTR).As congeneric species, the six selected species have a number ofcommon traits (Editorial Board for Flora of China, 1984; Fu, 1995).All are shade-tolerant and can regenerate and grow under theclosed canopy. Flowering occurs in May–June, and fruit maturationis usually in September, with seed dispersal shortly after maturity.The fruit is a double samara, usually one-seeded, endospermabsent. The seeds are shaped to spin as they fall, thereby travelinga considerable distance on the wind. They germinate early in thefollowing spring, immediately after snowmelt.There exist differences in growth type, morphological traits (e.g.,foliage form and seed size), and life history traits among the sixspecies (Fig. 1). ACBA is an understory species, while the others aremidstory species. For geographic distribution, ACMO is a widelydistributed species, whereas the others are largely restricted toNortheast China, Korea, and East Russia. Foliage form is significantlydifferent in ACMO compared with the other five species,making it the main trait used for species identification (Fig. 1).136

J. Zhang et al. / Acta Oecologica 36 (2010) 29–38 31Fig. 1. Morphological traits of six Acer species in the forest. See Table 1 for species codes.The fruits of ACTE and ACTR are the largest, followed by those ofACBA and ACMA. The angles of wing spread differ among the sixspecies; however, seed sizes are similar (Xu, 1996).2.3. Data analysisSpatial point pattern analysis, such as Ripley’s K function and thepair-correlation function, are commonly applied to detect thespatial arrangement of individuals within communities and togenerate hypotheses regarding the underlying processes controllingthe observed patterns (Ripley, 1981; Stoyan and Stoyan, 1994;Wiegand et al., 2007). In the present study, we used the paircorrelationfunction g(r) to determine whether the distribution ofthese congeners was random, aggregated, or regular; the spatialscales at which these patterns occurred; and the spatial associationsamong them. The pair-correlation function (Stoyan andStoyan, 1994), which is the derivative of Ripley’s popular K function(Ripley, 1981), replaces the circles used in calculating Ripley’s Kfunction with rings, and uses the mean number of neighbors ina ring of radius r and ring width around an individual, thus isolatingspecific distance classes (Wiegand and Moloney, 2004). Withincreasing radius, analyses using Ripley’s K function, which includeall information in the circle, fail to distinguish between effects atlarge scales and those at small scales, as this approach is a cumulativemeasure, with each progressively larger scale including theinformation from all smaller scales (Condit et al., 2000; Wiegandand Moloney, 2004).In contrast, the pair-correlation function, which characterizespatterns based on the frequency of points co-occurring at a givendistance, can be used to easily and intuitively analyze the spatialpatterns derived from ecological processes (Wiegand et al., 2007;Wiegand and Moloney, 2004). Similar to Ripley’s K function, thepair-correlation function g(r) includes both univariate and bivariatestatistics. The univariate statistics are used to analyze the spatialpattern of a single object, while the bivariate statistics are used toanalyze the spatial association of two objects. For a univariate pointpattern, g(r) > 1 indicates that the points are aggregated atdistances r, while g(r) < 1 indicates they are regularly dispersed. Fora bivariate point pattern, g(r) > 1 indicates a positive interactionbetween the two patterns at a given distance r, while g(r) < 1indicates a spatial segregation or repulsion between them atdistance r.In our study, the univariate statistic was used to analyze thespatial patterns of the six congeneric species and the patterns ofthese species in different size classes, and the bivariate statistic wasused to analyze intra- and interspecies spatial associations amongdifferent species and size classes. Two size classes were established:DBH < 5 cm and 5 cm. The species ACBA wasn’t dividedinto these size classes because it is a shrub species with few individualsat DBH 5 cm. Heterogeneous Poisson null models,accounting for possible environmental heterogeneity, were used toreveal significant uni- and bivariate second-order interactions. Inheterogeneous Poisson point process, the occurrence of any point isindependent of that of others, but the points are distributed inaccordance with an intensity function that varies with location(Stoyan and Stoyan, 1994; Wiegand and Moloney, 2004; Wiegandet al., 2007). For the univariate statistic, we constructed theintensity function based on the distribution of the six species andselected for all patterns a bandwidth h ¼ 30 m and a spatial resolutionof 2 m. For the bivariate statistic, we kept the locations of thetrees of the first species fixed and randomized the locations of thetrees of the second species using a heterogeneous Poisson pointprocess as the null model. The intensity function was constructedbased on the pattern of the second species, and a bandwidthh ¼ 30 m and a spatial resolution of 2 m were selected for allanalyses. For the analyses of the species pairs and the pairs betweenthe same size classes, we tested all pairs, species 1 vs species 2 andspecies 2 vs species 2, because the interaction may be nonsymmetric.For the analyses of the species pairs between larger andsmaller size classes, we hypothesized that larger size classes137

32J. Zhang et al. / Acta Oecologica 36 (2010) 29–38Table 1Population structure of six Acer species in a 25-ha old-growth temperate forest in Northern China.Species Species codes Growth type Abundance (Living) Abundance (Dead) Basal area (m 2 /ha) Mean DBH (cm)WTB WB WTB WB WTB WBAcer barbinerve ACBA Understory 3911 11762 1471 0.0812 0.1696 2.3 1.94Acer mandshuricum ACMA Midstory 251 255 8 0.0896 0.0914 7.07 7.08Acer mono ACMO Midstory 6609 6834 319 2.6855 2.7028 7.45 7.3Acer pseudosieboldianum ACPS Midstory 5984 8144 295 1.0953 1.2394 6.14 5.55Acer tegmentosum ACTE Midstory 846 1233 92 0.1075 0.1131 4.62 3.76Acer triflorum ACTR Midstory 276 278 17 0.1123 0.1125 8.67 8.64Note: ‘‘WTB’’ means the abundance without branches, and ‘‘WB’’ means the abundance with branches.suppress the recruitment and growth of smaller ones, whereassmaller size classes do not affect larger ones (Cipriotti and Aguiar,2005). Therefore, we kept the locations of the larger classes fixedand randomized the locations of the smaller classes usinga heterogeneous Poisson null model.For all analyses, 99% simulation envelopes were calculated fromthe highest and lowest values of g(r) taken from 99 Monte Carlosimulations of the null model. However, the approach is invalid fortesting an observed pattern against spatial model because the type Ierror may occur if the value of g(r) is close to a simulation envelope(Loosmore and Ford, 2006; Wiegand et al., 2007). A goodness-of-fit(GoF) test was used to provide expected type I error rates (Loosmoreand Ford, 2006). Details can be found in work by Diggle(2003) and Loosmore and Ford (2006). In our study, we selecteddistance intervals of 0–20 m to assess departures from the nullmodel for application of the GoF test. We retained the results forFig. 2. Size (DBH) class of six Acer species in the study plot. See Table 1 for species codes.138

J. Zhang et al. / Acta Oecologica 36 (2010) 29–38 33further analysis only when the observed p value of the GoF test wassmaller than 0.05. All analyses were performed using the gridbasedestimators in the Programita software package (Wiegand andMoloney, 2004).3. Results3.1. Population structureThe abundances of the six selected Acer species ranged from 251individuals of ACMA to 6609 of ACMO (Table 1). The abundances ofACMO, ACPS, and ACBA were much greater than those of the otherthree species. In counts of the number of branches, the understoryspecies ACBA has the most branches, while ACMO, ACMA, and ACTRhave few branches. The basal areas of ACMO and ACPS are significantlylarger than those of the other four species.Although the six species showed great variability in abundance,they had similar, distinct L-shaped size-class distributions, withmore individuals in smaller size classes (Fig. 2). Comparing the sizedistributions of pairs with similar abundances, both the species pairwith high abundance (ACMO and ACPS) and the pair with lowabundance (ACMA and ACTR) showed different size distributions:there are few individuals of ACMO and ACMA in larger size classes,but relatively many large individuals of ACPS and ACTR.3.2. Spatial patternsThe six species showed significant aggregation (Fig. 3 and Fig. 4).The GoF test also revealed a significant departure from theFig. 3. Spatial distribution of six Acer species in relation to the topography of the study plot. See Table 1 for species codes.139

34J. Zhang et al. / Acta Oecologica 36 (2010) 29–383ACBA12ACMA15ACMA (DBH

J. Zhang et al. / Acta Oecologica 36 (2010) 29–38 35(attraction) for two pairs (ACMO vs. ACTR and ACTR vs. ACMO) andnegative associations (repulsion) for eight pairs at small scales(Table 2). Six of eight pairs with negative associations were foundsymmetrically in ACBA and ACPS, ACMA and ACPS, and ACTE andACMO. ACBA and ACPS were associated negatively at scales of0–2 m, ACMA and ACPS at scales of 0–10 m, and ACTE and ACMO atscales of 0–28 m. ACBA showed a significant negative associationwith ACMO at scales of 0–2 m, and ACTR showed a significantnegative association with ACPS at scales of 0–16 m.We also analyzed the spatial associations among the individualsof each species in two size classes (Fig. 5). When comparing thespatial associations among large trees (DBH 5 cm) of differentspecies, the GoF test detected significant negative associations for 6of 20 species pairs at small scales. There was no species pair withpositive association at any scale. All six species pairs with negativeassociations were found symmetrically among ACPS and 3 otherspecies (ACMO, ACTE and ACMA). When comparing the spatialassociations among small trees (DBH < 5 cm) of different species,the GoF test detected significant positive associations for 5 of 30species pairs and negative association for only 1 species pair (ACBSvs. ACPS) at small scales. When the spatial associations among largeand small trees of these species were analyzed, the GoF test detectedsignificant positive associations for 3 of 30 species pairs and negativeassociations for 8 of 30 species pairs at small scales. Large treesof ACPS and ACTE showed significant positive associations withtheir small trees at scales of 0–20 m, while large trees of otherspecies didn’t show significant association with their small trees.4. DiscussionAggregated spatial distributions are commonly observed innaturally regenerated forests (Condit et al., 2000; He et al., 1997).The present study provides further evidence of clumping for sixcongeneric species in an old-growth temperate forest. The degreeof aggregation varied with species, size class, and spatial scale. Allsix congeners were aggregated at small scales. This result is inaccordance with the findings of many other studies (e.g., Conditet al., 2000; He et al., 1997; Henriques and Desousa, 1989). Aggregatedpatterns were also found for small and large trees of species.Interestingly, in all analyses of univariate statistics, the degree ofaggregation in the distribution of species with low abundance wasnot lower than that for species with high abundance at small scales(even being higher in some cases), consistent with the finding thatrare species are more aggregated than common species in tropicalforests (Hubbell, 1979; Condit et al., 2000).Aggregated distributions may result from limited seed dispersal(Grubb, 1977) or habitat heterogeneity (Harms et al., 2000;Queenborough et al., 2007); for the genus Acer, limited seeddispersal is considered the most likely explanation (Shibata et al.,2008; Zhang et al., 2008a). In the present study, all six of the closelyrelated congener species disperse their seeds by wind, and seedsize is consistent among the different species. Shibata et al. (2008)analyzed the effect of population density on the reproduction ofACMO in a temperate forest in Japan, and found that many matureseeds fall around their mother trees and that the percentage offalling seeds being immature or empty increased significantly withincreased distance from the parent tree, indicating limited seeddispersal for the species. Zhang et al. (2008a) analyzed the relationshipbetween seed dispersal and parent tree based on seed raindata collected over 2 years in the present study plot, and found thatthe mature seeds of many species, including several Acer species,were not dispersed far from their parent trees.Habitat variables such as topography, soil nutrients, andheterogeneous light distribution make a limited contribution to thespatial distribution of Acer species. In our plot, the spatial distributionof many species showed no significant relation with topographicalvariables (unpublished data). Tanaka et al. (2008)estimated the trade-off between gap dependency and shadetolerance in three coexisting Acer species in a temperate deciduousforest, and found that all species showed high seedling and saplingTable 2Analyses of the spatial associations of six Acer species in the study plot.Species A Species B p value Scales (m)0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40ACBA ACMA 0.01 r r r r r r r þ þ þ þ r r r r r r r r r rACBA ACMO 0.01 – – r r r r r r r r r r r r r r r r r r rACBA ACPS 0.01 – – r þ þ þ þ þ r r r r r r r r r r r r rACBA ACTE 0.01 r r r þ þ þ þ r r r r r r r r r r r r r rACBA ACTR 0.01 r r r r r r r r r r r r r r r r r r r r rACMA ACBA 0.03 r r r r r r þ þ þ þ þ þ r r r r r r r r rACMA ACPS 0.02 – – – – – – – r r r r r r r r r r r r r rACMA ACTE 0.01 r r r r r r r r r r – – r r r r – – – – rACMO ACBA 0.01 – r r r r r r r r r r r r r r r r r r r rACMO ACMA 0.01 r r r r r r r r r r r r r r r r r r r r rACMO ACPS 0.01 r r r – – – – – – – – r r r r r r r r r rACMO ACTE 0.01 – – – – – – – – – – – – – – – r r r r r rACMO ACTR 0.01 þ þ þ r r r r r r r r r r r r r r r r r rACPS ACBA 0.01 – – r r þ þ þ r r r r r r r r r r r r r rACPS ACMA 0.01 – – – – – – r r r r r r r r r r r r r r rACPS ACMO 0.01 r r r r – – – – – – – r r r r r r r r r rACPS ACTE 0.01 r r r r r r r r r r r r r r r r r r r r rACPS ACTR 0.01 r r r r r – – – r r r r r r r r r r r r rACTE ACBA 0.01 r r r þ þ þ þ þ r r þ r r r r þ þ r r r rACTE ACMO 0.01 – – – – – – – – – – – – – – – r r r r r rACTR ACMA 0.01 r r r r r r r r r r r þ þ þ þ r r r þ þ þACTR ACMO 0.03 þ þ þ þ r r r r r r r r þ þ þ r r r r r rACTR ACPS 0.01 – – – – – – – – – r r r r r r r r r r r rNote: The bivariate statistic of the pair-correlation function was used to analyze the spatial associations among six species under the heterogeneous Poison null model. Theintensity function was constructed based on the pattern of species B with the bandwidth h ¼ 30 m. The ring width for estimation of the pair-correlation function was 4 m; cellsize was 2 2m‘‘þ’’ stands for positive association, ‘‘r’’ stands for no spatial association, and ‘‘–’’ for negative association. Scale ¼ 0 means that the points are within the cells.Monte Carlo confidence was constructed at the 99% confidence level (99 simulations). The p values were calculated based on the goodness-of-fit (GoF) test. Only species pairswith the p value < 0.05 are showed in the table. See Table 1 for species codes.141

36J. Zhang et al. / Acta Oecologica 36 (2010) 29–381.2ACMO (DBH≥5 cm) vs. ACPS (DBH≥5 cm)1.6ACPS (DBH≥5 cm) vs. ACTE (DBH≥5 cm)1.01.4g12(r)0.81.20.61.01.5ACMO (DBH

J. Zhang et al. / Acta Oecologica 36 (2010) 29–38 37Comparative studies of congeneric species help to minimize theconfounding effect of phylogenetic differences between species(Bevill and Louda, 1999; Lloyd et al., 2002; Swenson et al., 2006),and may reveal those factors that limit species rarity. A species maybe relatively rare for various reasons, including inferior dispersalabilities (Cowling, 2001; Simon and Hay, 2003), low investment inreproduction (Münzbergová, 2005; Murray et al., 2002), lowpollinator effectiveness (Rymer et al., 2005), low competitive ability(Lloyd et al., 2002; Moora and Jõgar, 2006; Murray et al., 2002), andspecialized habitat requirements (Queenborough et al., 2007).In the present study, the abundance of six coexisting Acerspecies varied from 251 to 6609 individuals in the study plot. Basedon a comparison of population structure, spatial patterns, andspatial associations, we infer no significant difference among thesecongeners in terms of seed dispersal ability, competitive ability, andrequirements for topographical and light environments. Otherfactors (e.g., seed production and other habitat requirements) maycontribute to the trade-off between commonness and rarity. Forexample, based on the data obtained from 150 seed traps monitoredat the present study site over a period of 2 years, Zhang et al.(2008a) found that the common species ACMO and ACPS producedsignificantly more seeds than did relatively rare congeners.Although several comparative studies of other Acer species havebeen performed (e.g., Ackerly and Donoghue, 1998; Shibata et al.,2008; Tanaka et al., 2008), these results cannot be directly relatedto our study species because the genus Acer exhibits great variabilityin life history traits, regeneration ecology, seed size, leaf size,and canopy structure (Ackerly and Donoghue, 1998). To the best ofour knowledge, there exists little information regarding the six Acerspecies considered in the present study; consequently, we areunable to obtain a clear understanding of the coexistence of thesespecies in the forest.In summary, our comparative analysis of six sympatric Acerspecies at the forest scale revealed both similarities and differencesin population structure and spatial patterns, which contribute totheir coexistence. Although the influences of some factors (e.g.,competition, habitat heterogeneity, seed production, and seeddispersal) on the coexistence of these congeneric species have beenclarified, there exists no evidence that these factors can explain thecoexistence of all of these closely related species or explain whysome species are relatively rare and others common. Explaining thecoexistence of closely related species remains a long-standingchallenge in ecology (Mooney et al., 2008; Münzbergová, 2005).AcknowledgementsThis project was supported by National Natural Science Foundationof China (30870400 and 30700093) and National Key TechnologiesR&D Program of China (2008BAC39B02). This work was conductedwhile Jian Zhang was visiting Department of Forestry and NaturalResource, Clemson University in 2008.ReferencesAckerly, D.D., Donoghue, M.J., 1998. Leaf size, sapling allometry, and Corner’s rules:phylogeny and correlated evolution in maples (Acer). American Naturalist 152,767–791.Bevill, R.L., Louda, S.M., 1999. Comparisons of related rare and common species inthe study of plant rarity. Conservation Biology 13, 493–498.Brown, J.H., Stevens, G.C., Kaufman, D.M., 1996. The geographic range: size, shape,boundaries, and internal structure. Annual Review of Ecology and Systematics27, 597–623.Byers, D.L., Meagher, T.R., 1997. A comparison of demographic characteristics ina rare and a common species of Eupatorium. Ecological Applications 7, 519–530.Cipriotti, P.A., Aguiar, M.R., 2005. Effects of grazing on patch structure in a semi-aridtwo-phase vegetation mosaic. Journal of Vegetation Science 16, 57–66.Condit, R., 1998. Tropical Forest Census Plots: Methods and Results from BarroColorado Island, Panama and a Comparison with Other Plots. Springer, Berlin.Condit, R., Ashton, P.S., Baker, P., Bunyavejchewin, S., Gunatilleke, S., Gunatilleke, N.,Hubbell, S.P., Foster, R.B., Itoh, A., LaFrankie, J.V., Lee, H.S., Losos, E.,Manokaran, N., Sukumar, R., Yamakura, T., 2000. Spatial patterns in the distributionof tropical tree species. Science 288, 1414–1418.Cowling, R.M., 2001. Endemism. In: Levin, S.A. (Ed.), Encyclopedia of Biodiversity.Academic Press, San Diego, pp. 497–507.Davies, S.J., Palmiotto, P.A., Ashton, P.S., Lee, H.S., Lafrankie, J.V., 1998. Comparativeecology of 11 sympatric species of Macaranga in Borneo: tree distribution inrelation to horizontal and vertical resource heterogeneity. Journal of Ecology 86,662–673.Debussche, M., Thompson, J.D., 2003. Habitat differentiation between two closelyrelated Mediterranean plant species, the endemic Cyclamen balearicum and thewidespread C. repandum. Acta Oecologica-International Journal of Ecology 24,35–45.Diggle, P.J., 2003. Statistical Analysis of Point Patterns, second ed. Arnold, London.Druckenbrod, D.L., Shugart, H.H., Davies, I., 2005. Spatial pattern and process inforest stands within the Virginia piedmont. Journal of Vegetation Science 16,37–48.Editorial Board for Flora of China, 1984. Flora of China. Science Press, Beijing.Fu, P.Y., 1995. Keys to the Plants of Northeast China. Science Press, Beijing.van Gelderen, D.M., de Jong, P.C., Oterdoom, H.J., 1994. Maples of the World. TimberPress, Portland.Grubb, P.J., 1977. Maintenance of species-richness in plant communities: theimportance of the regeneration niche. Biological Reviews of the CambridgePhilosophical Society 52, 107–145.Hao, Z.Q., Li, B.H., Zhang, J., Wang, X.G., Ye, J., Yao, X.L., 2008. Broad-leavedKorean pine (Pinus koraiensis) mixed forest plot in Changbaishan (CBS) ofChina: community composition and structure. Journal of Plant Ecology 32,238–250.Harms, K.E., Wright, S.J., Calderon, O., Hernandez, A., Herre, E.A., 2000. Pervasivedensity-dependent recruitment enhances seedling diversity in a tropical forest.Nature 404, 493–495.He, F.L., Legendre, P., LaFrankie, J.V., 1997. Distribution patterns of tree species ina Malaysian tropical rain forest. Journal of Vegetation Science 8, 105–114.Helmus, M.R., Savage, K., Diebel, M.W., Maxted, J.T., Ives, A.R., 2007. Separating thedeterminants of phylogenetic community structure. Ecology Letters 10,917–925.Henriques, R.P.B., Desousa, E., 1989. Population structure, dispersion and microhabitatregeneration of Carapa guianensis in northeastern Brazil. Biotropica 21,204–209.Hubbell, S.P., 1979. Tree dispersion, abundance, and diversity in a tropical dry forest.Science 203, 1299–1309.Lloyd, K.M., Lee, W.G., Wilson, J.B., 2002. Competitive abilities of rare and commonplants: comparisons using Acaena (Rosaceae) and Chionochloa (Poaceae) fromNew Zealand. Conservation Biology 16, 975–985.Loosmore, N.B., Ford, E.D., 2006. Statistical inference using the G or K point patternspatial statistics. Ecology 87, 1925–1931.Mooney, K.A., Jones, P., Agrawal, A.A., 2008. Coexisting congeners: demography,competition, and interactions with cardenolides for two milkweed-feedingaphids. Oikos 117, 450–458.Moora, M., Jõgar, U., 2006. Competitive responses of the rare Viola elatior and thecommon Viola mirabilis. Plant Ecology 184, 105–110.Münzbergová, Z., 2005. Determinants of species rarity: population growth rates ofspecies sharing the same habitat. American Journal of Botany 92, 1987–1994.Murray, B.R., Thrall, P.H., Gill, A.M., Nicotra, A.B., 2002. How plant life-history andecological traits relate to species rarity and commonness at varying spatialscales. Austral Ecology 27, 291–310.Queenborough, S.A., Burslem, D., Garwood, N.C., Valencia, R., 2007. Habitat nichepartitioning by 16 species of Myristicaceae in Amazonian Ecuador. PlantEcology 192, 193–207.Rabinowitz, D., 1981. Seven forms of rarity. In: Synge, H. (Ed.), The Biological Aspectsof Rare Plant Conservation. John Wiley, New York, pp. 205–217.Ripley, B.D., 1981. Spatial Statistics. Wiley, New York, pp. 252.Rymer, P.D., Whelan, R.J., Ayre, D.J., Weston, P.H., Russell, K.G., 2005. Reproductivesuccess and pollinator effectiveness differ in common and rare Persooniaspecies (Proteaceae). Biological Conservation 123, 521–532.Shibata, M., Kikuchi, S., Tanaka, H., Sueyoshi, M., Yoshimaru, H., Niiyama, K., 2008.Effects of population density, sex morph, and tree size on reproduction ina heterodichogamous maple, Acer mono, in a temperate forest of Japan.Ecological Research. doi:10.1007/s11284-11008-10474-11284.Simon, M.F., Hay, J.D., 2003. Comparison of a common and rare species of Mimosa(Mimosaceae) in Central Brazil. Austral Ecology 28, 315–326.Snyder, K.M., Baskin, J.M., Baskin, C.C., 1994. Comparative ecology of the narrowendemic Echinacea tennesseensis and two geographically widespread congeners:relative competitive ability and growth characteristics. InternationalJournal of Plant Sciences 155, 57–65.Stoyan, D., Stoyan, H., 1994. Fractals, Random Shapes and Point Fields. Wiley,Chichester.Swenson, N.G., Enquist, B.J., Pither, J., Thompson, J., Zimmerman, J., 2006. Theproblem and promise of scale dependency in community phylogenetics.Ecology 87, 2418–2424.Tanaka, H., Shibata, M., Masaki, T., Iida, S., Niiyama, K., Abe, S., Kominami, Y.,Nakashizuka, T., 2008. Comparative demography of three coexisting Acerspecies in gaps and under closed canopy. Journal of Vegetation Science 19127–U127.143

38J. Zhang et al. / Acta Oecologica 36 (2010) 29–38Valiente-Banuet, A., Verdu, M., 2008. Temporal shifts from facilitation to competitionoccur between closely related taxa. Journal of Ecology 96, 489–494.Walck, J.L., Baskin, J.M., Baskin, C.C., 1999. Relative competitive abilities and growthcharacteristics of a narrowly endemic and a geographically widespread Solidagospecies (Asteraceae). American Journal of Botany 86, 820–828.Webb, C.O., Ackerly, D.D., McPeek, M.A., Donoghue, M.J., 2002. Phylogenies andcommunity ecology. Annual Review of Ecology and Systematics 33, 475–505.Wiegand, T., Gunatilleke, S., Gunatilleke, N., 2007. Species associations in a heterogeneousSri Lankan dipterocarp forest. American Naturalist 170, E67–E95.Wiegand, T., Moloney, K.A., 2004. Rings, circles, and null-models for point patternanalysis in ecology. Oikos 104, 209–229.Xu, T.Z., 1996. Phytogeorgraphy of the family Aceraceae. Acta Botanica Yunnanica18, 43–50.Yang, H.X., Li, D.J., Wang, B.N., Han, J.X., 1985. Distribution patterns of dominant treespecies on northern slope of Changbai Mountain. Research of Forest Ecosystem5, 1–14.Zhang, J., Hao, Z.Q., Li, B.H., Ye, J., Wang, X.G., Yao, X.L., 2008a. Composition andseasonal dynamics of seed rain in Broad-leaved Korean pine mixed forest,Changbai Mountain. Acta Ecologica Sinica 28, 2245–2254.Zhang, J., Hao, Z.Q., Song, B., Li, B.H., Wang, X.G., Ye, J., 2008b. Fine-scale species cooccurrencepatterns in an old-growth temperate forest. Forest Ecology andManagement 257 (10), 2115–2120.144

Forest Ecology and Management 257 (2009) 2115–2120Contents lists available at ScienceDirectForest Ecology and Managementjournal homepage: www.elsevier.com/locate/forecoFine-scale species co-occurrence patterns in an old-growth temperate forestJian Zhang a,b , Zhanqing Hao a, *, Bo Song c , Buhang Li a,b , Xugao Wang a ,JiYe a,ba Institute of Applied Ecology, Chinese Academy of Sciences, 72 Wenhua Road, Shenyang 110016, PR Chinab Graduate University of Chinese Academy of Sciences, Beijing 100049, PR Chinac Belle W. Baruch Institute of Coastal Ecology and Forest Science, Clemson University, Georgetown, SC 29442, USAARTICLEINFOABSTRACTArticle history:Received 24 December 2008Received in revised form 12 February 2009Accepted 13 February 2009Keywords:Assembly ruleCommunity structureInterspecific competitionSpatial patternTemperate forestThe pattern of species co-occurrence is instrumental for understanding community assembly rules. Inthis paper, we analyzed the co-occurrence of tree species in a 25-ha old-growth temperate forest plot inNortheastern China. The analysis was conducted at seven scales from 5 m 5 m up to 100 m 100 m inorder to determine the patterns of co-occurrence at different spatial scales. Our analyses were conductedfor all species, species with larger abundances, species with larger sizes, and five phylogenetic-basedspecies groups. Our results showed that at smaller scales, the co-occurrence patterns of all species,species with larger abundances, and species with larger sizes were significantly higher than expected bychance, suggesting that strong interspecies competition exists in the community. At larger scales, therewas no significant difference compared to randomized matrices. The result indicated that plant assemblyrules are only found at small spatial scales. However, when co-occurrence metrics were restricted tophylogenetic groups, we could not find any clear evidence of interspecific competition within thesegroups. In conclusion, we found that competition is an important assembly rule at small scales ingoverning tree communities of our temperate forest, although it is not the only process involved. Theimportance of other processes should also be taken into account to explain species co-occurrencepatterns.ß 2009 Elsevier B.V. All rights reserved.1. IntroductionSpecies co-occurrence analyses are increasingly applied toevaluate whether communities are random assemblages ofspecies or the result of deterministic mechanisms, such ascompetition (Gotelli and McCabe, 2002; Jenkins, 2006; Adams,2007) or other processes (Chen and Bradshaw, 1999; North et al.,2004; Ward and Beggs, 2007). Over three decades ago, Diamond(1975) described ‘‘checkerboard’’ distributions of avian species inthe Bismarck Archipelago that never co-occurred, and predictedthat competing species among assemblages should co-occur lessthan expected by chance. That study sparked a heated controversyin that the significance, or even the existence, of assembly ruleswas questioned (Gotelli and Graves, 1996; Adams, 2007). Connorand Simberloff (1979) argued that assembly rules could not beinferred from observed patterns by comparing the patterns withthose generated by Monte Carlo null models. Since then, contraryconclusions have been drawn from different communities. Gotelliand McCabe (2002) conducted a meta-analysis of published datamatrices, and found that nonrandom co-occurrence patterns areprevalent. Using a database of 45 species and 4540 geographic* Corresponding author. Tel.: +86 24 83970209; fax: +86 24 83970300.E-mail address: hzq@iae.ac.cn (Z. Hao).sites, Adams (2007) analyzed the patterns of co-occurrence byvirtue of a null model derived from competitive interactions, andfound that patterns of co-occurrence were significantly nonrandomat both regional and continental scales, providing strongevidence for competitive-based community assembly. Nonrandomspecies co-occurrence patterns can vary with nichedifferentiation (Hofer et al., 2004), spatial scale (Gotelli andEllison, 2002), temporal scale (Badano et al., 2005), andassemblage diversity (Badano et al., 2005; Mouillot et al.,2005). However, recent work on ectoparasites in fish (Jacksonet al., 1992; Gotelli and Rohde, 2002), birds (Feeley, 2003), andzooplankton (Jenkins, 2006) found little support for nonrandomspecies co-occurrence patterns.Scale is an important factor for studying species co-occurrence(Bycroft et al., 1993; Chen and Bradshaw, 1999; Jenkins, 2006;Sanders et al., 2007). Co-occurrence can exhibit one pattern at onescale, but a different pattern at another scale. Although there aremany studies comparing of co-occurrence patterns between localand regional scales (e.g., Gotelli and Ellison, 2002; Jenkins, 2006;Sanders et al., 2007), little has been mentioned of co-occurrencepatterns at fine neighborhood scales. Wilson et al. (1992) arguedthat evidence for plant community structure can be found mainly,perhaps only, at a small spatial scale. The analyses of fine-scale cooccurrencepatterns can give us important information aboutassembly mechanisms (Bycroft et al., 1993).0378-1127/$ – see front matter ß 2009 Elsevier B.V. All rights reserved.doi:10.1016/j.foreco.2009.02.016145

2116J. Zhang et al. / Forest Ecology and Management 257 (2009) 2115–2120In this paper, we analyzed fine-scale species co-occurrence in a25-ha old-growth temperate forest, Northeastern China in order tounderstand the importance of competition in structuring the treecommunity. Co-occurrence analyses were conducted for allspecies, species with larger abundances, species with larger sizes,and phylogenetic-based species groups. Three questions wereaddressed in this paper: (1) what are species co-occurrencepatterns in the old-growth temperate forest? (2) Does scale play akey role for species co-occurrence in the forest? And how doassembly rules change with scales? (3) Are the co-occurrencepatterns consistent with co-occurrence patterns structured bycompetition?2. Methods2.1. Study siteThe study site is located in the Changbai Mountain NatureReserve, which was established along the border of China andNorth Korea extending from 127842 0 to 128817 0 E and 41843 0 to42826 0 N. The reserve was established in 1960 and joined the WorldBiosphere Reserve Network under the Man and the BiosphereProject in 1980. The reserve is about 200,000 ha in size andelevation ranges from 740 m at the lowest part to 2691 m at thesummit of Changbai Mountain on the Chinese side. There are fivetypical vertical vegetation zones, namely aspen-white birch(Populus davidiana and Betula platyphylla) forest, broad-leavedKorean pine (Pinus koraiensis) mixed forest, spruce-fir (Piceajezoensis and Abies nephrolepis) forest, subalpine birch (Betulaermanii) forest and the tundra (Yang et al., 1985; Hao et al., 2007).Our study site is located in the broad-leaved Korean pine mixedforest, a conifer broad-leaved mixed temperate forest, which is themost common vegetation type in northeastern China. It is wellknown for its high biodiversity, complex stand structure, andunique species composition. The soil is classified as dark brownforest soil. The climate is characterized by the moist temperatemonsoon. Mean annual temperature is 3.3 8C (January 16.5 8Cand August 20.5 8C). Mean annual precipitation is671.9 mm year 1 , most of which occurs between June and August.The stand age is about 300 years. Vertical structure of the forest canbe identified (Hao et al., 2007). The canopy layer is >20 m tall, andthe main tree species are Korean pine, Amur linden (Tiliaamurensis), Mongolian oak (Quercus mongolica), Manchurian ash(Fraxinus mandshurica), and Japanese elm (Ulmus japonica). Mainspecies of the midstory layer (10–20 m tall) include Mono maple(Acer mono), Purplebloom maple (A. pseudo-siebodianum), Manchustriplemaple (A. tegmentosum), Amur Maackia (Maackiaamurensis). Understory layer is below 10 m, includes ManchurianHazelnut (Corylus mandshurica), Barbedvein maple (A. barbinerve),European Bird Cherry (Prunus padus) and so on.A 25-ha (500 m 500 m) plot was established in 2004. The plotwas chosen in the core zone of the reserve in order to avoid humandisturbances (Hao et al., 2007). All individuals with DBH 1cmwere stem-mapped, and identified to species. There are 38,902living stems, belonging to 52 species, 32 genura, and 18 families.Mean stand density of living trees was 1556 trees ha 1 , and meanbasal area of living trees was 43.2 m 2 ha 1 (Hao et al., 2007; Zhanget al., 2007).2.2. Data analysisWe divided our study plot into 10,000, 2500, 625, 256, 100, 49,and 25 quadrats at seven scales: 5 m 5m, 10m 10 m,20 m 20 m, 30 m 30 m, 50 m 50 m, 70 m 70 m, and100 m 100 m, respectively. Presence–absence data matriceswere attained from the abundance data sets based on the quadratsat the seven scales. In each presence–absence matrix, each rowrepresents a species and each column represents the presence (1)or absence (0) of a species in a sample (or site). Fifty-two specieswere included for the co-occurrence analyses of all species.Similarly, the plot data were transformed to presence–absencematrices for the species with restricted species abundance classes(100, 500, and 1000 stems, respectively), the species withlarger size classes (DBH 10 cm, 20 cm, and 30 cm, respectively),and the species group based on phylogeny in order to analyze theco-occurrence in each case. The species with high abundanceswere chosen to avoid the ‘‘dilution effect’’ (Gotelli and Graves,1996). The species of different size classes were chosen in order totest if competition changes with size classes. Phylogenetic data of52 species in our plot were obtained by Phylomatic(R20050610.new, http://www.phylodiversity.net/phylomatic/),which is based on the work of the Angiosperm Phylogeny Group(Stevens, 2001). Five species groups were chosen by phylogeneticbasedclassification, and related co-occurrence indices werecalculated for each species group.Analyses of presence–absence matrices with ‘‘null model’’randomization tests and co-occurrence indices have been thestandard method for measuring co-occurrences (e.g., Gotelli andEllison, 2002; Bell, 2005; Jenkins, 2006; Adams, 2007; Burns,2007), although there remains disagreement about what nullmodel and index should be used. In this paper, species cooccurrenceswere calculated by using four indices: the checkerboardscore of matrix (C score), the number of checkerboardspecies pairs (CHECKER), the number of species combinations(COMBO), and the variance ratio (V ratio). Their measures andperformance were described in detail in Gotelli (2000). If anassemblage is structured by competition, observed communitiesshould contain a larger C score, more checkerboard pairs, fewerspecies combinations, and lower V ratio than expected by chance(Gotelli and Ellison, 2002). We used a fixed-equiprobable nullmodel (SIM2 of Gotelli (2000)) to generate the randomlyconstructed assemblages. In this null model, row sums are fixedin order that each species occurs with the same frequency in therandomly constructed assemblages as in the observed assemblage,while all columns are equiprobable (Gotelli and Entsminger, 2006).Gotelli (2000) found that fixed-equiprobable null model seemsmost appropriate for analyzing ‘‘sample lists’’, particularly whencomparing standardized samples that have been collected in areasof homogenous habitat which is largely the condition of our 25-haplot (unpublished data). In the study plot, the terrain is relativelygentle. The elevation ranges from 791.8 m to 809.5 m, with a meanelevation of 801.5 m. The sequential swap algorithm was used forrandomization.To make the results comparable, we calculated a standardizedeffect size (SES) for each matrix (Gurevitch et al., 1992). The SES iscalculated as: (observed index mean of simulated index)/standard deviation of simulated index. Assuming a normaldistribution of SES, a 95% confidence interval of the SES valuesshould locate between 2.0 and 2.0. For the C score and theCHECKER, the values larger than 2.0 indicate nonrandom speciessegregation, and the values lower than 2.0 indicate nonrandomspecies aggregation. In contrast, for the COMBO and the V ratio, thevalues larger than 2.0 indicate nonrandom species aggregation,and the values lower than 2.0 indicate nonrandom speciessegregation. The mean and standard deviations of these indiceswere calculated from 5000 null randomization matrices. Cooccurrenceanalyses and associated randomization tests werecalculated from 5000 null matrices by using null model softwareEcoSim 7.72 (Gotelli and Entsminger, 2006).To explore whether the competition exits among different guildgroups and within them and how they change with scales, we usedthe ‘‘Guild structure’’ module in EcoSim for patterns among the146

J. Zhang et al. / Forest Ecology and Management 257 (2009) 2115–2120 2117guilds as a group (Gotelli and Entsminger, 2006). The same fivephylogenetic groups were used for the analyses at seven scales. Themethod tests whether the mean co-occurrence index among guildgroups is larger or smaller than expected by chance and tests thevariance of the co-occurrence index among guilds groups. Meantime,the favored states hypothesis was tested in this module (Foxand Brown, 1993). This hypothesis assumes that the distribution ofspecies among guilds, or even among communities, is moreuniform than predicted by a random distribution. If communitiesare formed by sequentially adding species in different functionalgroups, there should be a significantly larger number of favoredstates than expected by chance (Gotelli and Entsminger, 2006). Themean and standard deviations of the same indices with cooccurrenceanalyses were calculated from 5000 null matrices, andthe SES values were calculated for the related comparison.3. Results3.1. Co-occurrence patterns of the communityCommunity-wide co-occurrence patterns varied with scales(Fig. 1). At 5 m and 10 m scales, the observed C score of all specieswere significantly higher than expected by chance, suggesting anegative pattern of species co-occurrence. The positive cooccurrencepatterns of all species were captured at 30 m and50 m. The checkerboard patterns (the index CHECKER) cannot befound at all observed scales. For the co-occurrence index COMBO(the numbers of species combinations), the observed values of allspecies were significantly lower than expected by chance at scales5–30 m, indicating a strong negative association, which are likelyto be the result of interspecific competition. The V ratio of allspecies measures the variability in the number of species per site.There is no significant value found at small scales, whereas valueslarger than expected by chance occurred at scales of 20–70 m. Inaddition, the highest values of C score and COMBO for all speciesoccurred at a scale of 10 m, suggesting that the strongestinterspecific competition exists at the scale.Nonrandom co-occurrence patterns were found at small scales,and there is no significant difference compared to randomizedmatrices at large scales, when we measured the co-occurrence byrestricting to higher species abundance and larger tree sizes (Figs. 1and 2). To the species with higher abundances, the observed Cscore and COMBO showed significantly higher values thanexpected by chance at small scales, which were similar with theco-occurrence pattern of all species. And there were the highest Cscores for three abundance levels at scale 10 m, and the highestCOMBO for abundance 100 and 500 at scale 10 m, suggestingthat strong interspecific competition occurs at the scale. The V ratiodid not show evidence of nonrandom co-occurrence patterns. Forthe species at three size classes, similar co-occurrence patternswere found in the four co-occurrence indices (Fig. 2).3.2. Co-occurrence patterns of different phylogenetic groupsThe co-occurrence patterns for the five phylogenetic groupsdiffered largely with the co-occurrence patterns of the community(Fig. 3). We could not find any clear evidence for interspecificcompetition within these groups. The hypothesis that phylogeneticgroups may experience greater interspecific competition becauseof their ecological similarity cannot be validated. For the fourindices, the co-occurrence of two groups showed positiveassociation at 5 m and 10 m scales, which are likely to be theresult of interspecific facilitation. For the indices C score and Vratio, the co-occurrence of one group showed positive associationat 5–50 m.For the four co-occurrence indices, the unusually largevariances were captured at all scales, when we analyzed the cooccurrencepatterns among the five phylogenetic groups (Table 1).The large variances suggested that there were significantdifferences in their levels of co-occurrence. The same results alsowere found when we analyzed the co-occurrence within eachgroup (Fig. 3). In the favored states analyses, SES of the four indicesdid not show significant difference compared with that expectedby chance.Fig. 1. The co-occurrence patterns of all species and species with larger abundances (100, 500, and 1000 individuals, respectively) at seven spatial scales in a 25-ha oldgrowthtemperate forest plot in northeastern China. Standardized effect sizes of four co-occurrence indices were calculated with null model software EcoSim 7.72. For thecheckerboard score of the matrix (C score) and the number of checkerboard species pairs (CHECKER), the values larger than 2.0 indicate nonrandom species segregation, andthe values lower than 2.0 indicate nonrandom species aggregation. For the number of species combinations (COMBO) and the variance ratio (V ratio), the values larger than2.0 indicate nonrandom species aggregation, and the values lower than 2.0 indicate nonrandom species segregation. Bold lines with diamond symbols represent all species,while broken lines with sold square, asterisk, triangle point-up symbols represent the species with abundances 100, 500, and 1000, respectively.147

2118J. Zhang et al. / Forest Ecology and Management 257 (2009) 2115–2120Fig. 2. The co-occurrence patterns of species at three size classes, DBH 10 cm (bold lines with square), DBH 20 cm (broken lines with asterisk), and DBH 30 cm (boldlines with diamond), respectively. Other explanations are the same as in Fig. 1.Fig. 3. The co-occurrence patterns of five species groups by phylogenetic-based classification. Other explanations are the same as in Fig. 1.Table 1The co-occurrence patterns among five phylogenetic groups in the study plot. The patterns were analyzed using the ‘‘Guild structure’’ module in EcoSim. Standardized effectsizes were used in order to test whether the mean co-occurrence index among guild groups is larger or smaller than expected by chance, and the variance of the co-occurrenceindex among guild groups. An unusually large variance would mean that the guild groups differ significantly from one another in their levels of co-occurrence, while anunusually small variance would mean that guilds are strikingly similar to one another in the level of co-occurrence observed. If communities are formed by sequentiallyadding species in different functional groups, there should be a significantly larger number of favored states than expected by chance. Other explanations are the same as inFig. 1.Scales SES of C score SES of CHECKER SES of COMBO SES of V ratioVariance Favored states Variance Favored states Variance Favored states Variance Favored states5m 5m 6,117,606,000 1.04 157.03 1.01 714.75 1.03 2.81 1.0410 m 10 m 923,182,100 1.02 116.73 1.04 2170.44 1.04 2.93 1.0420 m 20 m 3,519,734 0.63 77.98 0.61 776.74 0.62 1.97 0.6030 m 30 m 91,478.58 0.41 58.50 0.42 371.39 0.42 2.38 0.4150 m 50 m 5,766.67 0.39 31.86 0.40 182.63 0.39 1.59 0.3870 m 70 m 579.57 0.29 19.95 0.30 58.12 0.30 0.88 0.31100 m 100 m 50.45 0.29 12.69 0.29 14.47 0.29 0.66 0.28148

J. Zhang et al. / Forest Ecology and Management 257 (2009) 2115–2120 21194. DiscussionAccording to our analyses, the community exhibited lessspecies co-occurrence patterns than expected by chance at smallspatial scales, which partly supported the predictions of Diamond’s(1975) assembly rules model. The same result has been reported inseveral other studies (e.g., Watkins and Wilson, 1992; Gotelli andMcCabe, 2002; Swenson et al., 2006; Adams, 2007; Sanders et al.,2007). Our analysis also showed that at larger spatial scales, thecommunity did not show significant co-occurrence patterns.Similar results were obtained when the co-occurrence patternswere analyzed both for species of higher abundance and trees oflarger size.A major effort of community ecology is to document nonrandompatterns of coexisting species and to infer about underlyingprocesses or assembly rules that may have given rise to theobserved patterns (Gotelli and Graves, 1996; Sanders et al., 2007).For a long time, nonrandom co-occurrence patterns wereinterpreted to result from interspecific competition for essentiallythe same set of mineral resources, light and water (Diamond, 1975;Kelt and Brown, 1999; Tilman, 2007; Hanski, 2008). Diamond(1975) emphasized that interspecific competition was the mostimportant determinant of observed species combinations andcould lead to consistent rules regarding species organization anddistribution. In our plot, strong interspecific competition may existat small scales, evidenced by the nonrandom co-occurrencepatterns (Figs. 1 and 2), consistent with the results of previousresearches in the forest (Sun and Zhao, 1997; Hao et al., 2007;Zhang et al., 2007).Random co-occurrence patterns were found at larger scales, ineither community level or species with larger abundances or sizeclasses, suggesting that there may be no dominant processes thatinfluence species distribution of the plot. At larger scales, habitatheterogeneity was very small in our plot relative to many otherforest plots (Hao et al., 2008). Compared with the results at smallscales, we can conclude that spatial scale plays an important role inshaping community assembly rules, and co-occurrence patternschange with spatial scale (Bycroft et al., 1993; Chen and Bradshaw,1999; Adams, 2007). In accord with Watkins and Wilson’s (1992)argument, nonrandom co-occurrence in temperate forests can befound mainly at small spatial scales.Co-occurrence patterns based on phylogenetic groups differedlargely from the co-occurrence patterns of the community. There isno clear evidence for interspecific competition within thephylogenetic groups in the forest. Generally, phylogeneticallyrelated species with ecologically similar characters are likely tocoexist less frequently than expected by chance because of limitedresources (Kelt and Brown, 1999; Webb et al., 2002). Interspecificcompetition should be much stronger among these species,especially at small scales. However, we found that co-occurrencepatterns of three groups showed significant positive associations atsmaller scales, which are likely to be the result of interspecificfacilitation. Similar results were also found in other communities(Wilson and Lee, 1994; Veech, 2006; Valiente-Banuet and Verdú,2008). Similar habitat requirement and limited seed dispersal,which lead to clumped distribution of species, perhaps are twomain factors caused the patterns (Helmus et al., 2007). There is noevidence for interspecific competition at large scales in thecommunity. Maybe there are enough resources provided for thecoexistence of these ecologically similar species. Meantime, wefind that there are significant differences among these phylogeneticgroups, suggesting some difference in ecological andevolutionary processes.Additionally, several alternative explanations rather thancompetition for nonrandom co-occurrence patterns are possible.These include habitat heterogeneity (Ward and Beggs, 2007),evolutionary history (Gotelli and McCabe, 2002), and evenstochastic processes (Ulrich, 2004; Bell, 2005). In particular,habitat heterogeneity is an important factor for shaping nonrandompatterns at small scales (Ward and Beggs, 2007). We tried toremove this factor by restricting the analyses to one forest plotwith homogeneous habitat. However, it is still possible that thedifferences of micro-habitat at small scales influenced speciesdistributions. For example, some tree species in our forest, such asT. mandshurica, U. laciniata, A. tegmentosum, and A. mandshuricum,showed significantly positive or negative spatial associations withtopographic conditions (Hao et al., 2008). Further studies areneeded to link species co-occurrence patterns and habit heterogeneity.Ulrich (2004) and Bell (2005) both found that the neutralmodel could generate nonrandom co-occurrence patterns predictedby Diamond’s (1975) assembly rules model, which is basedon niche differentiation and interspecific competition. The role ofstochastic processes or chance for shaping species distributionremains to be explored in our forest.In summary, we used a stem-mapped database from a 25-haold-growth temperate forest plot to understand species cooccurrenceat multiple spatial scales. Our results showed thatthe co-occurrence patterns of the community are closely related tospatial scales. At smaller scales, the co-occurrence patterns of thecommunity were significantly higher than expected by chance,suggesting that strong interspecies competition may exist in thecommunity. However, at larger scales, there was no significantdifference compared to randomized matrices, which indicated thatplant assembly rules are only captured at small spatial scales.When co-occurrence metrics were restricted to phylogeneticgroups, we could not find any clear evidence of negativenonrandom co-occurrence patterns within these groups. Finally,we can conclude that competition should be an importantassembly rule at small scales in structuring tree communities,although it is not the only process involved. The other processes,including habitat heterogeneity, evolutionary history, and stochasticprocesses, should also be taken into account to explaincommunity assembly rules.AcknowledgementsWe thank Drs. Fangliang He, Michael Papaik, and Hong Qian forconstructive comments which greatly improved our study. We alsothank University of Alberta and Clemson University for researchopportunity. This project was supported by National NaturalScience Foundation of China (30870400 and 30700093), and theKnowledge Innovation Project of the Chinese Academy of Sciences(No. KZCX2-YW-430).ReferencesAdams, D.C., 2007. Organization of Plethodon salamander communities: guild-basedcommunity assembly. Ecology 88, 1292–1299.Badano, E.I., Regidor, H.A., Núñez, H.A., Acosta, R., Gianoli, E., 2005. Species richnessand structure of ant communities in a dynamic archipelago: effects of islandarea and age. J. Biogeogr. 32, 221–227.Bell, G., 2005. The co-distribution of species in relation to the neutral theory ofcommunity ecology. Ecology 86, 1757–1770.Burns, K.C., 2007. Patterns in the assembly of an island plant community. J.Biogeogr. 34, 760–768.Bycroft, C.M., Nicolaou, N., Smith, B., Wilson, J.B., 1993. Community structure (nichelimitation and guild proportionality. In relation to the effect of spatial scale, in aNothofagus forest sampled with a circular transect. New Zeal. J. Ecol. 17, 95–101.Chen, J., Bradshaw, G.A., 1999. Forest structure in space: a case study of an oldgrowth spruce-fir forest in Changbaishan Natural Reserve, P.R. China. For. Ecol.Manage. 120, 219–233.Connor, E.F., Simberloff, D., 1979. The assembly of species communities: chance orcompetition? Ecology 60, 1132–1140.Diamond, J.M., 1975. Assembly of species communities. In: Cody, M.L., Diamond,J.M. (Eds.), Ecology and Evolution of Communities. Harvard University Press,Cambridge, pp. 342–444.149

2120J. Zhang et al. / Forest Ecology and Management 257 (2009) 2115–2120Feeley, K., 2003. Analysis of avian communities in Lake Guri, Venezuela, usingmultiple assembly rule models. Oecologia 137, 104–113.Fox, B.J., Brown, J.H., 1993. Assembly rules for functional groups in North Americandesert rodent. Oikos 67, 358–370.Gotelli, N.J., 2000. Null model analysis of species co-occurrence patterns. Ecology81, 2606–2621.Gotelli, N.J., Ellison, A.M., 2002. Assembly rules for New England ant assemblages.Oikos 99, 591–599.Gotelli, N.J., Entsminger, G.L., 2006. EcoSim: null models software for ecology. Version7. Acquired Intelligence Inc. and Kesey-Bear. Jericho, VT05465. http://garyentsminger.com/ecosim.htm.Gotelli, N.J., Graves, G.R., 1996. Null Models in Ecology. Smithsonian InstitutionPress, Washington.Gotelli, N.J., McCabe, D.J., 2002. Species co-occurrence: a meta-analysis of J.M.Diamond’s assembly rules model. Ecology 83, 2091–2096.Gotelli, N.J., Rohde, K., 2002. Co-occurrence of ectoparasites of marine fishes: a nullmodel analysis. Ecol. Lett. 5, 86–94.Gurevitch, J., Morrow, L.L., Wallace, A., Walsh, J.S., 1992. A meta-analysis of fieldexperiments on competition. Am. Nat. 140, 539–572.Hao, Z.Q., Li, B.H., Zhang, J., Wang, X.G., Ye, J., Yao, X.L., 2008. Broad-leaved Koreanpine (Pinus koraiensis) mixed forest plot in Changbaishan of China: communitycomposition and structure. J. Plant Ecol. 32, 238–250 (in Chinese).Hao, Z.Q., Zhang, J., Song, B., Ye, J., Li, B.H., 2007. Vertical structure and spatialassociations of dominant tree species in an old-growth temperate forest. For.Ecol. Manage. 252, 1–11.Hanski, I., 2008. Spatial patterns of coexistence of competing species in patchyhabitat. Theor. Ecol. 1, 29–43.Helmus, M.R., Savage, K., Diebel, M.W., Maxted, J.T., Ives, A.R., 2007. Separating thedeterminants of phylogenetic community structure. Ecol. Lett. 10, 917–925.Hofer, U., Bersier, L.-F., Borcard, D., 2004. Relating niche and spatial overlap at thecommunity level. Oikos 106, 366–376.Jackson, D.A., Somers, K.M., Harvey, H.H., 1992. Null models and fish communities:evidence of nonrandom patterns. Am. Nat. 139, 930–951.Jenkins, D.G., 2006. In search of quorum effects in metacommunity structure:species co-occurrence analyses. Ecology 87, 1523–1531.Kelt, D.A., Brown, J.H., 1999. Community structure and assembly rules: confrontingconceptual and statistical issues with data on desert rodents. In: Weiher,E., Keddy, P.A. (Eds.), Ecological Assembly Rules: Perspectives, Advances,Retreats. Cambridge University Press, Cambridge, pp. 75–107.Mouillot, D., George-Nascimento, M., Poulin, R., 2005. Richness, structure andfunctioning in metazoan parasite communities. Oikos 10, 447–460.North, M., Chen, J., Oakley, B., Song, B., Rudnicki, M., Gray, A., Innes, J., 2004. Foreststand structure and pattern of old-growth western hemlock/Douglas-fir andmixed-conifer forests. For. Sci. 50, 299–311.Sanders, N.J., Gotelli, N.J., Wittman, S.E., Ratchford, J.S., Ellison, A.M., Jules, E.S., 2007.Assembly rules of ground-foraging ant assemblages are contingent on disturbance,habitat, and spatial scale. J. Biogeogr. 34, 1632–1641.Stevens, P.F., 2001 onwards. Angiosperm Phylogeny Website. Version 6, May 2005.URL: http://www.mobot.org/MOBOT/research/APweb/.Sun, W.Z., Zhao, S.D., 1997. Distribution patterns of main tree species in Tiliabroadleaf Korean pine forest on northern slope of Changbai Mountains. Chin.J. Appl. Ecol. 8, 119–122 (in Chinese).Swenson, N.G., Enquist, B.J., Pither, J., Thompson, J., Zimmerman, J., 2006. Theproblem and promise of scale dependency in community phylogenetics. Ecology87, 2418–2424.Tilman, D., 2007. Interspecific competition and multispecies coexistence. In: May,R., McLean, A. (Eds.), Theoretical Ecology: Principles and Applications. 3rd edn.Oxford University Press, New York, pp. 84–97.Ulrich, W., 2004. Species co-occurrences and neutral models: reassessing J.M.Diamond’s assembly rules. Oikos 107, 603–609.Valiente-Banuet, A., Verdú, M., 2008. Temporal shifts from facilitation to competitionoccur between closely related taxa. J. Ecol. 96, 489–494.Veech, J.A., 2006. A probability-based analysis of temporal and spatial co-occurrencein grassland birds. J. Biogeogr. 33, 2145–2153.Ward, D., Beggs, J., 2007. Coexistence, habitat patterns and the assembly of antcommunities in the Yasawa islands Fiji. Acta Oecol. 32, 215–223.Watkins, A.J., Wilson, J.B., 1992. Fine-scale community structure of lawns. J. Ecol. 80,15–24.Webb, C.O., Ackerly, D.D., McPeek, M.A., Donoghue, M.J., 2002. Phylogenies andcommunity ecology. Annu. Rev. Ecol. Syst. 33, 475–505.Wilson, J.B., Lee, W.G., 1994. Niche overlap of congeners: a test using plantaltitudinal distribution. Oikos 69, 469–475.Wilson, J.B., Roxburgh, S.H., Watkins, A.J., 1992. Limitation to plant speciescoexistence at a point: a study in a New Zealand lawn. J. Veg. Sci. 3, 711–714.Yang, H.X., Li, D.J., Wang, B.N., Han, J.X., 1985. Distribution patterns of dominant treespecies on northern slope of Changbai Mountain. Res. For. Ecosyst. 5, 1–14 (inChinese).Zhang, J., Hao, Z.Q., Song, B., Ye, J., Li, B.H., Yao, X.L., 2007. Spatial distributionpatterns and associations of Korean pine (Pinus koraiensis) and Tilia amurensis inbroad-leaved Korean pine mixed forest in Changbai Mountain. Chin. J. Appl.Ecol. 18, 1681–1687 (in Chinese).150

vol. 180, no. 1 the american naturalist july 2012E-ArticleThe Contribution of Rare Species to Community PhylogeneticDiversity across a Global Network of Forest PlotsXiangcheng Mi, 1 Nathan G. Swenson, 2 Renato Valencia, 3 W. John Kress, 4 David L. Erickson, 4Álvaro J. Pérez, 3 Haibao Ren, 1 Sheng-Hsin Su, 5 Nimal Gunatilleke, 6 Savi Gunatilleke, 6Zhanqing Hao, 7 Wanhui Ye, 8 Min Cao, 9 H. S. Suresh, 10 H. S. Dattaraja, 10 R. Sukumar, 10 andKeping Ma 1, *1. State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, 20 Nanxincun,Xiangshan, Beijing 100093; 2. Department of Plant Biology, Michigan State University, East Lansing, Michigan 48824; 3. Laboratoriode Ecología de Plantas, Escuela de Ciencias Biológicas, Pontificia Universidad Católica del Ecuador, Apartado 17-01-2184, Quito,Ecuador; 4. Department of Botany, National Museum of Natural History, Smithsonian Institution, Washington, DC 20013; 5. TaiwanForestry Research Institute, Taipei 10066; 6. Department of Botany, Faculty of Science, University of Peradeniya, Peradeniya 20400, SriLanka; 7. Institute of Applied Ecology, Chinese Academy of Sciences, Shenyang 110016; 8. South China Botanical Garden, ChineseAcademy of Sciences, Guangzhou 510650; 9. Key Laboratory of Tropical Forest Ecology, Xishuangbanna Tropical Botanical Garden,Chinese Academy of Sciences, Kunming 650223; 10. Center for Ecological Sciences, Indian Institute of Science, Bangalore 560012, IndiaSubmitted August 3, 2011; Accepted February 11, 2012; Electronically published May 23, 2012Online enhancements: appendixes, zip file. Dryad data: http://dx.doi.org/10.5061/dryad.p4n8rg64.abstract: Niche differentiation has been proposed as an explanationfor rarity in species assemblages. To test this hypothesis requiresquantifying the ecological similarity of species. This similaritycan potentially be estimated by using phylogenetic relatedness. Inthis study, we predicted that if niche differentiation does explain theco-occurrence of rare and common species, then rare species shouldcontribute greatly to the overall community phylogenetic diversity(PD), abundance will have phylogenetic signal, and common andrare species will be phylogenetically dissimilar. We tested these predictionsby developing a novel method that integrates species rankabundance distributions with phylogenetic trees and trend analyses,to examine the relative contribution of individual species to theoverall community PD. We then supplement this approach withanalyses of phylogenetic signal in abundances and measures of phylogeneticsimilarity within and between rare and common speciesgroups. We applied this analytical approach to 15 long-term temperateand tropical forest dynamics plots from around the world.We show that the niche differentiation hypothesis is supported insix of the nine gap-dominated forests but is rejected in the six disturbance-dominatedand three gap-dominated forests. We also showthat the three metrics utilized in this study each provide unique butcorroborating information regarding the phylogenetic distribution ofrarity in communities.Keywords: rare species, community phylogenetic diversity, speciesabundance distribution, phylogenetic relatedness, niche differentiation,community assembly.* Corresponding author; e-mail: kpma@ibcas.ac.cn.Am. Nat. 2012. Vol. 180, pp. E17–E30. 2012 by The University of Chicago.0003-0147/2012/18001-53229$15.00. All rights reserved.DOI: 10.1086/665999IntroductionA central goal in community ecology is to determine themechanisms underlying the relative abundances of species.Addressing this question is particularly challenging andinteresting in diverse communities where a large proportionof the species are relatively rare. Thus, basic researchinto species abundances, particularly in diverse communities,requires a consideration of the forces underlyingspecies rarity. Two opposing families of hypotheses havebeen proposed to explain rarity in species assemblages.One family, which focuses on niche differentiation, stressesthe importance of specialization and spatiotemporal resourcepartitioning. For example, the niche position hypothesis(Gaston 1994; Kunin 1997) and similar hypotheses,such as Hanski’s (1982) core-satellite specieshypothesis and Grime’s (1998) classification of dominant,subordinate, and transient species, postulate that rare speciesutilize spatially scarce resources that are underutilizedby common species. Similar hypotheses that focus on temporalpartitioning of the environment, such as the storageeffect, have also been proposed to explain the recruitmentof rare species (Chesson 2000; Kelly and Bowler 2002).Contrary to this family of niche differentiation hypothesesare neutral models (e.g., Hubbell 2001). Neutral modelsassume that species are interchangeable and species abundancesare not best explained by niche differences butrather may be better explained by dispersal limitation, the151

E18The American Naturaliststructure of the metacommunity, and demographic stochasticity.A niche differentiation hypothesis and a neutralmodel make opposing predictions regarding the similarityof rare and common species. In particular, a niche differentiationhypothesis predicts that rare species will bedissimilar from common species and other rare species,whereas a neutral model predicts no significant dissimilaritybetween rare and common species or between rarespecies.The main challenge in testing the validity of the abovehypotheses lies in quantifying the ecological similarity betweenrare and common species (Keddy 1989; Clark et al.2007). For years ecologists have utilized evolutionary relatednessas a proxy for ecological similarity (Jarvinen1982; Webb et al. 2002; Cadotte et al. 2008; Wiens et al.2010). The rationale for this approach has been that commondescent from a recent ancestor should result in closelyrelated species being, on average, more similar to one anotherthan they are to a distantly related species. Of course,divergent evolution between closely related species is oftendemonstrated, for example, during an adaptive radiation,and in these cases the assumption that closely related speciesare similar will inevitably break down (Losos 2008).That said, in instances where the analyses incorporate taxafrom a broad sampling of the tree of life, the assumptionmay be more reasonable (Wiens et al. 2010). For example,two palm species in a rain forest tree community are likelyto be much more ecologically similar to one another thanto a species in the common shrub genus Piper (Piperaceae).Aside from the potential utility of the assumption thatclosely related species are likely to be ecologically similar,is the reality that quantifying the multidimensional ecologicalsimilarity of hundreds or thousands of species indiverse assemblages is often unreasonable. A further rationalefor using phylogenetic relatedness in ecologicalstudies comes from recent research that has compared andcontrasted results from trait and phylogenetic analyses.This work has reported that phylogenetic measures cancapture additional information not contained in the smallset of traits that ecologists often measure (Cadotte et al.2008, 2009; Burns and Strauss 2011). In other words, thephylogeny may contain information pertaining to importantand unmeasured traits that have phylogenetic signals(Cadotte et al. 2009).Phylogenetic analyses of rarity in communities can beutilized to address several key questions. Perhaps the mostfundamental question involves determining the degree towhich rare species are closely or distantly related to commonspecies. Phylogenetic assessments of a niche differentiationhypothesis would therefore test the degree ofphylogenetic relatedness between common and rare species.Indeed, recent phylogenetically based work has aimedto test this expectation. For example, studies from a diverseMexican tree community and a cactus yeast communityhave found that common species are distantly related torare species (Anderson et al. 2004; Kelly et al. 2008). However,these results are not totally consistent with the predictionsof a niche differentiation hypothesis. Specifically,rare species are expected to utilize locally or temporallyscarce resources, and thus rare species are expected to benot only dissimilar to common species but also dissimilarto one another. This dissimilarity of rare species to commonspecies and to one another is therefore expected toresult in rare species disproportionately contributing tothe overall phylogenetic diversity (PD) in a community.Thus, the question becomes, do rare species disproportionatelycontribute to the community PD as expected bya niche differentiation hypothesis, or is their contributionrandom as expected by a neutral model?A second important question regarding the relationshipbetween relatedness and rarity is simply to determinewhether there is phylogenetic signal (Blomberg et al. 2003;Swenson and Enquist 2009) in the relative abundances ofspecies in communities. A niche differentiation hypothesiswould predict that there is phylogenetic signal in abundancesin communities because of the expected distinctivenessbetween common and rare species.A third important question regarding phylogeny andrarity is what is the degree of phylogenetic dissimilaritywithin and between groups of common and rare species?A niche differentiation hypothesis predicts that rare specieswould be dissimilar to common species. Further commonspecies should also be closely related to one another, butrare species should be distantly related to one another.In this study, our central objective is to test the nichedifferentiation hypothesis by addressing these three centralquestions: (1) what is the contribution of rare species tothe community PD? (2) Is there phylogenetic signal inrank abundance across forest plots? (3) What is the levelof phylogenetic similarity within and between commonand rare species in forest plots? We utilize 15 forest dynamicsplots (FDPs; 16–52 ha in size) located across temperate,subtropical, and tropical forests of Asia and America(fig. A1; tables A1, A2, available online) as our studysystems. The results of these tests will be used to indentifythe degree to which a niche differentiation hypothesis canbe invoked to explain the distribution of rarity across variedtree communities.Material and MethodsStudy Sites and DataWe analyzed 15 FDPs from around the world (fig. A1).The majority of these FDPs exist within extensive primaryforests either without anthropogenic disturbance or with152

Rare Species to Community PhylogenyE19only slight anthropogenic disturbance (tables A1, A2). TheFushan, Luquillo, Palanan, and Dinghushan FDPs sufferfrom hurricane damage to varying degrees, and the Luquilloplot also has a history of anthropogenic disturbance(Losos and Leigh 2004). The Mudumalai plot in India hasbeen influenced by logging operations since the early twentiethcentury, and it experiences frequent and widespreadground fires. The La Planada FDP was logged lightly decadesago, and the Lambir FDP is frequently affected bylandslides generated by continuous heavy rain. Althoughmost of the Barro Colorado Island (BCI) plot is primaryforest, about 2 ha of the plot is secondary forest. The FDPscover 16–52 ha within which the locations of all trees withdiameters ≥1 cm at 1.3 m above the ground are mapped,measured, and identified to the species level when possible.In this study we define rare species within the FDPsfollowing Hubbell and Foster (1986), who state that rarespecies have an average density of fewer than one individualper hectare. The abundance data for the plots wereavailable from either the published standard books of theplots (table A2) or from the website of the SmithsonianInstitution Global Earth Observatories (http://www.sigeo.si.edu/). The 15 FDPs vary widely with respect to bothspecies richness and their proportion of rare species (tableA2). In total, the data set we utilized represents 5,637species and 2,167 rare species in a wide variety of forests.Community Phylogenetic AnalysesFor the community phylogenetic analyses a phylogenetictree was constructed for each plot, using the plant phylogenydatabase Phylomatic (Webb and Donoghue 2005).In the Yasuní plot, 26 species (14 of which are rare speciesaccording to the criteria of Hubbell and Foster [1986])were identified only to the family level and were excludedfrom the analysis. Phylomatic used the latest AngiospermPhylogeny Group classification (Angiosperm PhylogenyGroup III 2009) as the phylogenetic tree backbone. Thephylogenetic branch lengths were calibrated by implementingthe BLADJ algorithm in Phylocom 4.1 (Webb etal. 2008) with estimated molecular and known fossil dates(Wikstrom et al. 2001). These branch lengths are taken asrough estimates, but they provide a substantial advantageover using nodal distances where all branch lengths aretreated as equivalent (Webb 2000). To explore the potentialimpact of using a Phylomatic-generated phylogeny insteadof a molecular phylogeny, we performed an additional setof analyses, using highly resolved community phylogeniesfor the BCI and Luquillo plots (Kress et al. 2009, 2010).While there were a few differences in the results, our resultsare generally robust to differences in the phylogeny estimates(data not shown).A Framework for Quantifying the Contributionof Rare Species to Community PDThe first goal of this study is to quantify the contributionof rare species to the overall community PD. We start bydefining community PD and standardized community PD(stdPD). Community PD is defined as the sum of thebranch lengths throughout a phylogenetic tree connectingall taxa in a community or sample (Faith 1992). We definestdPD as the difference between the observed PD of subsetsof a community or sample and the mean PD value of 999null communities or samples generated by shuffling thenames of species across the tips of the phylogeny.We now integrate the stdPD metric with species rankabundance to quantify the contribution of rare species tooverall community PD. Specifically, we first quantify thecumulative PD by adding increasingly rare species to thesample starting with the second-most abundant species,then the third-most abundant species, and so on. Next westandardize the cumulative PD values by quantifyingstdPD. The stdPD value of the most abundant species willalways be undefined, because PD cannot be calculatedfrom a sample of one species, and the cumulative stdPDvalue upon adding the rarest species is 0, because all speciesare included in the PD calculation, assuring that the meannull value equals the observed value.Trends in the species rank abundance–stdPD (SAPD)curve can now be used to reveal the contribution of anindividual species to the overall community PD whileweighing that observation by that randomly expected. Specifically,a decrease in the stdPD from one species to thenext in the rank abundance distribution shows that theadded species was more closely related than expected bychance to the species that are more abundant in the community(i.e., it adds little to the community PD). Thus, ifrare species contribute little to the overall community PD,we would expect a decreasing trend toward the right endof the SAPD. For example, in the hypothetical communityrepresented in figure 1b, adding the rare species E and Cto a community containing the common species B and Dresults in a decreasing trend of stdPD in species rank 4and 5 (fig. 1b). Conversely, an increase in the stdPD fromone species to the next in the rank abundance distributionshows that the added species was more distantly relatedthan expected by chance (i.e., it adds greatly to the communityPD). Thus, if rare species contribute greatly to theoverall community PD, we would expect an increasingtrend toward the right end of the SAPD. For example, inthe hypothetical community in figure 1c, the addition ofthe rare species D and G to a community containing thedistantly related common species A and B leads to anincreasing trend of stdPD from species rank 3 to 4 (fig.1c). Finally, if rare species contribute a random amount153

Figure 1: Highly simplified illustration of how the species abundance rank–standardized phylogenetic diversity (SAPD) curves are generatedand interpreted. a, Hypothetical phylogenetic tree of the species pool; b, standardized community PD (stdPD) increases along species rankwhen rare species are distantly related to common species; c, stdPD decreases along species rank when rare species are closely related tocommon species; d, stdPD randomly fluctuates along species rank when rare species are randomly distributed in the phylogenetic tree; e,reverse stdPD decreases along species rank from a peak value when rare species are distantly related to common species; and f, reverseSAPD curves are concave because of closely related rare species. Note that the species numbers in the hypothetical communities of thisfigure are lower than the species numbers in the species pool. The stdPD values for the SAPD curves of the rarest species in this figure arenot 0, and so the SAPD curves in this figure are slightly different from those generated when all species are included in the analysis, suchas in figure 2 and figure C1, available in a zip file. The filled circles represent common species and the open circles represent rare species.The size of the circle indicates species abundance. The solid lines are SAPD curves and the dashed lines are reverse SAPD curves.E20154

Rare Species to Community PhylogenyE21to the overall community PD, then the SAPD would fluctuaterandomly. For example, in the hypothetical communityin figure 1d, the rare species B and D were closelyrelated to the more abundant species A and E, whereas Ewas distantly related to the common species. This causesa fluctuation of stdPD values in the SAPD curve.Note that the scenario of decreasing stdPD against speciesrank from common species to rare species shown infigure 1c may not only be explained by the high relatednessbetween rare and common species, but also be interpretedas the high relatedness between rare species (fig. 1e, 1f ).Fortunately, these two situations of decreasing stdPDagainst species rank can be distinguished by reversing theSAPD curves where the curve is quantified beginning withthe second-rarest species. The reverse SAPD along reverseabundance rank (i.e., starting from the second-rarest species)will decrease from a peak value if rare species aremore closely related to common species than chance wouldpredict (fig. 1e). For example, stdPD decreases along reversespecies rank because of high relatedness between rarespecies E and C and common species B, D, and F (fig.1e). A second scenario is that the reverse SAPD wouldfirst exhibit a downward trend and then display an increasingtrend along reverse species rank if the rare speciesare more closely related (fig. 1f ). For example, in the highlysimplified hypothetical community in figure 1f, the reverseSAPD curve first decreases from species rank 5 to 4 becauseof close relatedness between rare species G and H. Thenthe reverse SAPD curve increases from species rank 4 to2 because of distant relatedness between common speciesB and D and rare species F, G, and H. Therefore, the reverseSAPD curve is not simply the complement of the SAPDcurve, it also contains useful information of its own.Statistical analyses of the SAPD curves require the quantificationof trends in stdPD values along the rank abundanceaxis and a single SAPD curve may present multipletrends in different subseries. Thus, it is necessary to identifysignificant trends across all subseries. In other words,it is necessary to identify regions of the SAPD curve thatare significantly increasing or decreasing. The potentialbreakpoints between subseries were analyzed using piecewiseregression (Muggeo 2003, 2011; Toms and Lesperance2003; Betts et al. 2007):stdPD p b b i b (i a )I(i 1 a ), (1)i 0 1 j j jj≥2where stdPD i is the value of stdPD for ith species rank; iis the corresponding species abundance rank from 2 ton 1; n is the species richness in a community; a j is thejth breakpoint (i.e., the breakpoint between subseries); theslopes of the lines are b 1 , b 1 b 2 , and so on; b j is thedifference in slope values; and I is an indicator variable.We built piecewise linear regression models with one ormore breakpoints in a SAPD curve by using Muggeo’s(2003) method. The significance of the breakpoints in thepiecewise regressions was evaluated by using Zeileis et al.’s(2003) method. We also determined the weight of evidencefor breakpoints in relation to linear models with j breakpointsin relation to model with j 1 breakpoints by usingthe Akaike Information Criterion (AIC; Burnham and Anderson2002). The relative likelihood of each model inrelation to the best model was determined based on evidenceratios (ERs) derived from AIC values (Burnham andAnderson 2002).To quantify the trend of an identified subseries in theSAPD curve, we used a Mann-Kendall trend test to calculatethe significance of the decreasing or increasingtrends of stdPD values along species abundance rank. AMann-Kendall test, commonly known as the Kendall’s taustatistic, has been applied widely to test for randomnessagainst trends in climatological and hydrological time series(Hirsch and Slack 1984; Yu et al. 1993; Douglas et al.2000). In this test, the null hypothesis H 0 states that stdPD 1 ,stdPD 2 , ..., stdPD n are a sample of n independent andidentically distributed random variables (Yu et al. 1993).The alternative hypothesis H 1 of a two-sided test is thatthe distributions of stdPD k and stdPD j are not identicalfor all k, where j ≤ n and k ( j. The test statistic S iscalculated using equations (2) and (3):n1 n jkp1 jpk1S p sgn(stdPD stdPD k), (2)1 if (stdPDj stdPD k) 1 0sgn(stdPDj stdPD k) p 0 if (stdPDj stdPD k) p 0,{ 1 if (stdPD j stdPD k ) ! 0(3)where S has a mean of 0 and variance of S is computedby Var (S) p [n(n 1)(2n 5) tt(t 1)(2t 5)]/18and is asymptotically normal (Hirsch and Slack 1984),where t is the extent of any given tie and tdenotes thesummation over all ties. The standard normal variable zis computed by using the following equation (Douglas etal. 2000):{S 1if S 1 01/2(Var (S))z p 0 if S p 0. (4)S 1if S ! 0(Var (S)) 1/2A positive value of z indicates an increasing trend in theSAPD curve meaning that rare species in the communityare more distantly related to common species than expectedby chance, while a negative value of z indicates adecreasing trend in SAPD curve, meaning rare species in155

E22The American Naturalista community are more closely related to common speciesthan expected by chance.The SAPD curve is autocorrelated because of its cumulativenature, which may inflate type I error rates andresult in the overestimation of significant increasing ordecreasing trends. In order to account for the autocorrelationof stdPD values between neighboring speciesranks, we randomly shuffled species abundance across thetips of the phylogeny 999 times and constructed 999 nullSAPD curves. Then we calculated the probability that theobserved z value of a subseries in the observed SAPD curvewas greater or lower than z value of the subseries in nullSAPD curves.In this study we have proposed a new approach, theSAPD curve, to assess the contribution of rare species tocommunity PD. Several other indices frequently used inphylogenetic community ecology, such as the mean pairwisephylogenetic distance (MPD), the net relatedness index(NRI; Webb 2000; Webb et al. 2002), phylogeneticspecies variability (PSV), phylogenetic species richness(PSR; Helmus et al. 2007), and evolutionary distinctiveness(ED; Redding and Mooers 2006; Isaac et al. 2007; Cadotteet al. 2010), may also have the potential to assess thecontribution of rare species. That said, our choice of metrics,PD, stdPD, and SAPD, was conditional on the biologicalquestion being tested. For example, frequently usedindices such as MPD, NRI, and PSV have their merit inmeasuring average relatedness of community across theentire phylogenetic tree, but the central focus in this studywas on the relative contribution of an individual speciesor a range of species on the rank abundance distribution(app. B, available online), which is less interpretable whenusing MPD, PSV, PSR, or NRI. We did find that the integrationof the evolutionary distinctiveness (ED) metricwith the species rank abundance distribution—that is, speciesabundance rank–standardized ED—had a similar abilityto assess relative contribution of individual species tocommunity PD when compared with the SAPD; these resultsare presented in appendix B.Phylogenetic Signal Analysis of Species AbundanceThe SAPD curve analyses above consider the contributionof rare species to the overall community PD. Quantifyingthe phylogenetic signal in species rank abundance providesa complementary analysis that addresses the degree towhich closely related species have similar rank abundances.We quantified the phylogenetic signal in rank abundancedata of all species and of only rare species in each plot,using the K statistic proposed by Blomberg et al. (2003).The abundance data were log 10 transformed before analysisto homogenize variances. The K statistic provides a comparisonof the observed and expected level of phylogeneticsignal under the assumption of Brownian motion traitevolution given a phylogenetic tree (Blomberg et al. 2003).If K 1 1, then the abundance data have more phylogeneticsignal than expected from Brownian motion, whereas ifK ! 1, then the abundance data have less phylogeneticsignal than expected. We assessed the significance of thephylogenetic signal by randomly shuffling species abundanceamong species 999 times and calculating 95% confidenceintervals.Phylogenetic Similarity within and betweenCommon and Rare SpeciesThe last aim of this study was to calculate the phylogeneticsimilarity within and between groups of common and rarespecies in each forest plot. To accomplish this, we firstcalculated the phylogenetic dispersion within the group ofcommon species and within the group of rare species,using the nearest taxon index (NTI). The NTI is a standardizedeffect size (SES) of the mean nearest phylogeneticneighbor distance (MNND) between species in a community(Webb et al. 2002; Swenson et al. 2007) or, in thiscase, the mean nearest neighbor distance between commonspecies or between rare species. We calculated the phylogeneticsimilarity between the groups of common andrare species using a phylogenetic beta diversity metricbased on nearest neighbor distances (D nn ; Swenson 2011;Swenson et al. 2011):nmip1min dirjp1min djcD p , (5)nn2where min d ir is the nearest phylogenetic neighbor fromcommon species i to rare species, min d jc is the nearestphylogenetic neighbor from rare species j in common species,and m and n are the number of common and rarespecies, respectively. The SES of D nn and MNND, SES(D nn ), and NTI, respectively, were calculated by implementinga null model analysis. The null model shuffledthe names of taxa across the tips of the phylogeny 999times. The SES (D nn ) and NTI were calculated as the observedD nn or MNND minus the mean value of the 999randomizations divided by the standard deviation of the999 null values. The NTI and SES (D nn ) consider only thenearest phylogenetic neighbors and indicate the “terminal”phylogenetic structure. The NTI and SES (D nn ) thereforecomplement the SAPD curve and the standardized EDmetrics in this study, which are more heavily influencedby the “basal” phylogenetic structure.Most statistical analyses were performed in the free softwareR, version 2.9.0 (R Development Core Team 2009),Blomberg’s K was computed using the R package “picante”(Kembel et al. 2010), Mann-Kendall tests were performed156

Rare Species to Community PhylogenyE23Table 1: Estimates of the breakpoints in the species abundance rank–standardized phylogenetic diversity(SAPD) curves for the 15 plots in this studyForest dynamics plot,breakpoints 95% CI supF P AIC L (#10 3 ) AIC w (#10 3 ) ER (w)Changbaishan:45.5 43.6, 47.4 16.3 !.001 .630 .608 180,000Gutianshan:91.7 85.9, 97.5 95.9 !.001 1.947 1.824 180,000141.4 134.6, 148.2 10.6 !.001 1.814 115.6Dinghushan:40.9 39.3, 42.5 99.3 !.001 2.706 2.621 180,00057.0 54.8, 59.2 74.1 !.001 2.426 180,000Xishuangbanna:259 252.7, 265.3 862.2 !.001 6.740 6.020 180,000Sinharaja:185.4 181.4, 189.4 49.9 !.001 2.715 2.636 180,000Lambir:950.4 943.2, 957.5 268.4 !.001 15.637 15.184 180,000984.4 977.9, 990.9 230.9 !.0011,075.0 1,067.0, 1,082.0 243.1 !.001 14.965 180,000Pasoh:546 530.6, 561.4 348.8 !.001 10.680 10.178 180,000Yasuní:499.7 492, 507.4 3,143.3 !.001 16.417 14.332 180,000BCI:138.9 135.2, 142.6 88.2 !.001 3.820 3.747 180,000208.6 205.1, 212.1 51.7 !.001 3.329 180,000Huai Kha Khaeng:59.5 57.3, 61.72 75.1 !.001 3.689 3.577 180,000149.3 146.3, 152.3 .6 .527 3.547 180,000223.2 220.7, 225.6 33.5 !.001 3.090 180,000La Planada:81.1 77.7, 84.6 71.5 !.001 3.003150.1 146.1, 154.1 23.0 !.001 2.673 180,000Fushan:8.0 3.9, 12.1 5.3 .006 1.352 1.337 1,900.7Palanan:122.2 115.7, 128.7 50.1 !.001 3.893 3.842 180,000198.0 191.1, 204.9 38.4 !.001 3.631 180,000Mudumalai:50.1 43.5, 56.7 9.9 !.001 .761 .746 1,299.9Luquillo:124.9 122, 127.8 30.3 !.001 1.763 1.716 180,000Note: supF is a statistic used to test the significance of every potential breakpoint (Andrews 2003); AIC L is the Akaike InformationCriterion value of the linear model; AIC w is the AIC value of the piecewise regression model with one, two, or three breakpoints;and ER (w) is the evidence ratio, to be interpreted as the evidence against the linear model or the j-breakpoints piecewise regressionmodel against the j 1 breakpoint piecewise regression model. Blank cells indicate no converged j-breakpoints piecewise regressionmodel. BCI, Barro Colorado Island.using the R package “Kendall” (McLeod 2011), piecewiseregressions were implemented using the R package “segmented”(Muggeo 2011), and the significance of breakpointswere evaluated using the R package “strucchange”(Zeileis et al. 2011). Finally, the NTI and SES of D nn werecalculated using the software “phylocom” (Webb et al.2008).ResultsWe first partitioned the SAPD curve for each forest plotinto several subseries using piecewise regressions to dissectthe presence of multiple trends (table 1). For example, rarespecies (with fewer than one individual per hectare; speciesrank, 188–305) in BCI were decomposed into two sub-157

Table 2: Mann-Kendall trend test of common species and rare species in the forestdynamics plot communitiesForest dynamics plot Common species Rare speciesChangbaishan .308 (.285) (2–34) a .118 (.703) (35–51) a.848 (.057) (2–45) b .733 (.985)* (46–51) b.226 (.378) (2–102) a .629 (.968) (103–158) aGutianshan.369 (.347) (2–92) b .765 (.994)** (93–141) b.456 (.346) (142–158) bXishuangbanna .746 (.020)* (2–238) a .542 (.927) (239–467) a.786 (.010)* (2–259) b .494 (.900) (260–467) bSinharaja .267 (.282) (2–161) a .418 (.880) (162–205) a.340 (.202) (2–185) b .544 (.960) (186–205) b.455 (.152) (2–781) a .408 (.823) (782–1,191) a.575 (.053) (2–950) b .482 (.904) (950–984) bLambir.940 (.002)** (985–1,075) b.735 (.985)* (1,075–1,191) bPasoh .352 (.261) (2–533) a .747 (.988)* (534–812) a.332 (.249) (2–546) b .744 (.990)* (547–812) bYasuní .738 (.021) (2–678) a .634 (.940) (679–1,092) a.907 (.001)*** (2–500) .649 (.976)* (501–1,092) b.402 (.392) (2–187) a .712 (.127) (187–304) aBCI.679 (.083) (1–139) b .827 (.023)* (210–304) b.747 (.995)** (140–209) b.695 (.066) (2–96) a .056 (.518) (97–276) aHuai Kha Khaeng.878 (.005)** (2–60) b .661 (.978)* (61–149) b.902 (.012)* (150–223) b.774 (.996)** (224–276) b.091 (.626) (2–155) a .667 (.122) (156–219) aLa Planada.614 (.090) (2–81) b .708 (.097) (151–219) b.767 (.992)* (82–150) b.067 (.541) (2–102) a .087 (.536) (103–209) aDinghushan .822 (.015)* (2–41) b .213 (.699) (58–209) b.294 (.762) (42–57) bFushan .526 (.095) (2–77) a .182 (.738) (78–109) a.500 (.781) (2–8) b .619 (.037) (9–109) bPalanan.334 (.861) (2–235) a .040 (.568) (236–321) a.346 (.265) (2–122) b .404 (.313) (199–321) b.584 (.939) (123–198) bMudumalai .281 (.761) (2–22) a .165 (.675) (23–66) a.381 (.992)* (2–50) b .662 (.162) (51–66) bLuquillo .285 (.358) (2–82) a .011 (.601) (83–136) a.493 (.087) (2–125) b .600 (.932) (126–136) bNote: The first number in each cell is the z value from the Mann-Kendall trend test, the numberin the first set of parentheses indicates the probability that an observed z value in a subseries is greaterthan that in null species rank abundance–standardized phylogenetic diversity (SAPD) curves, and therange in the second set of parentheses indicates the species rank range. BCI, Barro Colorado Island.aSpecies rank of common species and rare species, according to Hubbell and Foster (1986), whodefined rare species as having fewer than one individual per hectare.bPartitioning of the SAPD curve into several subseries by piecewise regression in order to identifymultiple trends in a single SAPD curve.* P ! .05.** P ! .01.*** P ! .001.E24158

Rare Species to Community PhylogenyE25Figure 2: Three typical species abundance rank–standardized phylogeneticdiversity (SAPD) curves for three forest dynamics plots. SAPDcurves here describe the following: a, Yasuní, with a significant upwardtrend for rare species, indicating little relatedness between commonand rare species; b, Barro Colorado Island (BCI), with a significantdownward trend of rare species, indicating close relatedness betweencommon and rare species; and c, Dinghushan, with random fluctuationof rare species along species abundance rank, indicating random relatednessbetween common and rare species. The solid line representscommon species in the curves, while the dashed line stands for rarespecies (!1 individual/ha). Dashed vertical lines and shared zones indicatebreakpoints and associated 95% CIs, respectively. The dottedlines are the segmented linear regression curves.series by piecewise regression: the first part of the rarespecies region of the rank abundance distribution (speciesrank, 188–209) exhibited a significant increasing trend,whereas the second part of rare species region of the rankabundance distribution (species rank, 210–305) showed asignificant decreasing trend (tables 1, 2). The significanceof all breakpoints between neighboring subseries was assessedusing piecewise regression, the supF statistic, andthe evidence ratio (table 1). All 15 SAPD curves werepartitioned into two to four subseries by piecewise regression(fig. 2; fig. C1, available in a zip file).Next we quantified the significance of the trends for thegroups of common and rare species in the forest plots,following the description of rarity used by Hubbell andFoster (1986), using a Mann-Kendall trend test to accountfor the autocorrelated nature of stdPD (table 2). Ultimately,we found no general result regarding the contributionof rare species to community PD in the 15 forestdynamic plots (FDPs; table 2). In particular, our observedSAPD curves for the 15 FDPs revealed that the contributionof rare species to community PD can range fromsignificantly less than expected to significantly greater thanexpected (fig. 2; table 2; app. C, available in a zip file):the contribution of rare species to community PD in sixFDPs (Changbaishan, Gutianshan, Xishuangbanna, Pasoah,Lambir, and Yasuní) was significantly greater thanexpected, whereas the contribution of rare species to communityPD in BCI was significantly less than expected, thecontribution of rare species in seven FDPs (Palanan, Dinghushan,Luquillo, Mudumalai, La Planada, Sinharaja, andFushan) was not significantly different from expected, andrare species of Huai Kha Khaeng (HKK) showed multipletrends. We assigned rare species from the Xishuangbannaplot to the group supporting niche differentiation becausemost of the rare species (ranging from 259 to 290 andfrom 353 to 468) showed an increasing trend and therewas a decreasing trend for rare species (species rank, 291–352) that was caused by close relatedness among rare species(see the reverse stdPD between species rank 291–352in fig. E1d, available in a zip file).Our second central question in this study was whetherthere is phylogenetic signal in abundance in the 15 FDPs.Species abundance in 11 FDPs (all except Luquillo, LaPlanada, Dinghushan, and Mudumalai) exhibited significantphylogenetic signals, indicating that closely relatedspecies tended to have similar abundances (table 4). Inthe Luquillo, La Planada, Dinghushan, and MudumalaiFDPs, abundance was randomly distributed with respectto phylogeny (table 4).Our third and final question concerned the phylogeneticsimilarity within and between common and rare speciesin each of the 15 FDPs. To answer this question, we quantifiedthe phylgenetic dispersion within and between commonspecies and rare species groups in each FDP, usingthe nearest neighbor metrics NTI and D nn . We found thatcommon species were phylogenetically clustered in nineFDPs (Changbaishan, Xishuangbanna, Sinharaja, Pasoh,Yasuní, BCI, Huai Kha Khaeng, Luquillo, and Lambir; tables3, 5) and were phylogenetically overdispersed in sixFDPs (Gutianshan, Dinghushan, La Planada, Fushan, Palanan,and Mudumalai; tables 3, 5). Similarly, we alsofound that rare species were phylogenetically overdispersed159

E26The American NaturalistTable 3: Standardized phylogenetic diversity (stdPD) for common species and rare speciesForest dynamics plot stdPD of common species stdPD of rare speciesChangbaishan 151 (.225) (1–34) a 222 (.816) (35–52) a436 (.003)** (1–45) b 329 (.957) (46–51) b187 (.224) (1–102) a 291 (.834) (103–159) aGutianshan264 (.163) (1–92) b 238 (.795) (93–141) b25 (.477) (142–159) b89 (.040) (1–102) a 189 (.708) (103–210) aDinghushan517 (.042) (1–41) b 51 (.542) (58–210) b306 (.913) (42–57) bXishuangbanna 722 (.026) (1–238) a 630 (.958) (239–468) a921 (.002)** (1–259) b 724 (.988)* (260–468) bSinharaja 528 (.031) (1–161) a 607 (.976)* (162–205) a409 (.017)* (1–185) b 61 (.557) (186–205) b377 (.111) (1–781) a 131 (.629) (782–1,192) aLambir457 (.067) (1–950) b 731 (.984)* (951–984 and 1,076–1,192) b698 (.004)** (985–1,075) bPasoh 600 (.027) (1–533) a 665 (.968) (534–813) a528 (.033) (1–546) b 706 (.979)* (547–813) bYasuní 1,337 (.002)** (1–678) a 948 (.987)* (679–1,093) a1,768 (.002)** (1–500) b 1,280 (.998)** (501–1,093) b59 (.590) (1–187) a 434 (.092) (188–305) aBCI373 (.119) (1–139) b 605 (.024)* (210–305) b811 (.992)** (140–209) b470 (.047) (1–96) a 216 (.773) (97–278) aHuai Kha Khaeng 553 (.013)* (1–60) b 750 (.999)*** (61–149 and 224–278) b343 (.112) (150–223) b240 (.780) (1–155) a 509 (.099) (156–220) aLa Planada605 (.070) (1–81) b 543 (.064) (151–219) b1,055 (.997)** (82–150) bFushan 80 (.359) (1–77) a 110 (.361) (78–110) a374 (.951) (1–8) b 296 (.034) (8–110) b132 (.715) (1–235) a 152 (.284) (236–323) aPalanan213 (.194) (1–122) b 293 (.146) (199–323) b159 (.727) (123–198) bMudumalai 21 (.528) (1–22) a 165 (.101) (23–67) a152 (.953) (1–50) b 28 (.415) (51–67) bLuquillo 128 (.348) (1–82) a 162 (.701) (83–137) a356 (.050) (2–125) b 347 (.935) (126–137) bNote: The first number in each cell is the stdPD value, the number in the first set of parentheses indicates theprobability of the observed PD being greater than that of the null communities, and the range in the second set ofparentheses indicates the species rank. BCI, Barro Colorado Island.aSpecies rank of common species and rare species, according to Hubbell and Foster (1986), who define rare speciesas having fewer than one individual per hectare.bPartitioning of the species abundance rank–stdPD curve into several subseries by piecewise regression in order toidentify multiple trends in a single species abundance rank–stdPD curve.* P ! .05.** P ! .01.*** P ! .001.in 11 FDPs (Changbaishan, Gutianshan, Dinghushan,Pasoh, HKK, Xishuangbanna, Fushan, Yasuní, Palanan,Sinharaja, and Lambir; tables 3, 5) and phylogeneticallyclustered in three FDPs (BCI, La Planada, and Mudumalai).The SES of D nn of all plots except Luquillo waslower than expected (table 5), indicating that rare andcommon species are phylogenetically dissimilar.DiscussionDespite the range of hypotheses that have been proposedto explain the assembly of common and rare species intocommunities, the question of whether spatial or temporalniche differentiation of species promotes their co-occurrenceremains an open one (Keddy 1989; Clark et al. 2007).160

Rare Species to Community PhylogenyE27Taking a phylogenetic approach, we asked three centralquestions. First, we asked what is the contribution of rarespecies to the overall PD of a community. It is predictedthat if niche differentiation facilitates the co-occurrence ofrare species with common species, then rare species areexpected to contribute more to the community PD thanwould be expected by chance. Second, we asked whetherthere is phylogenetic signal in the abundance of species incommunities. If there is phylogenetic signal, then rare speciestend to be on average distantly related to commonspecies. Finally, we asked about the phylogenetic similaritywithin and between the common and rare species in acommunity. A niche differentiation hypothesis predictsthat rare species should be phylogenetically distinct fromcommon species, thereby permitting their co-occurrence.Thus, we predicted that if niche differentiation is occurring,there should be phylogenetic signal in species abundanceand rare species should be distantly related to commonspecies and to each other.The degree to which rare species contribute to the overallPD of a community was expected to reflect phylogeneticdissimilarity between common and rare species and alsobetween rare species themselves. The contribution of rarespecies to the overall community PD was quantified usinga SAPD curve. In six of the forest plots (Changbaishan,Gutianshan, Xishuangbanna, Pasoh, Lambir, and Yasuní),we found a significant increasing trend at the end of theSAPD curve, indicating that rare species contribute morethan expected to overall community PD. This suggests thatrare species are generally phylogenetically distinct fromcommon species and from each other, providing partialsupport for the niche differentiation hypothesis. In theBCI forest plot we found the opposite result, with rarespecies contributing significantly less than expected tooverall community PD, as indicated by a decreasing trendin the SAPD curve. This result shows that rare species inBCI forest areas typically come from clades that also containcommon species and therefore are likely to be lessdifferentiated in their niches. Finally, in the Dinghushan,Palanan, Luquillo, Fushan, Sinharaja, La Planada, and Mudumalaiforests, the contribution of rare species to overallcommunity PD was not different from random, while inthe Huai Kha Khaeng (HKK) forest plot there was a mixtureof significantly increasing and decreasing trends inthe SAPD. Thus, in nine of the 15 forest plots, we failedto support the prediction of the niche differentiationhypothesis.We found that among gap-dominated FDPs, six (Changbaishan,Gutianshan, Xishuangbanna, Pasoh, Lambir, andYasuní) of nine FDPs showed evidence supporting theniche differentiation hypothesis. Conversely, all FDPs withdominant disturbance (Dinghushan, Palanan, Luquillo,Fushan, HKK, and Mudumalai) rejected the niche differentiationhypothesis. That the niche differentiation hypothesiswas not supported for nine of the forests may beexplained by different types of disturbance in these plots,such as frequent typhoons in Fushan and Palanan andwidespread fires and browsing of elephants in Mudumalai.It is likely that disturbance selects for more closely relatedspecies with similar resistances to disturbances (Warwickand Clarke 1998; Abellán et al. 2006; Helmus et al. 2010),which supports an environmental filtering hypothesis(Webb 2000; Vamosi et al. 2009).Our second main question was whether there is phylogeneticsignal in the relative abundances of species inthe 15 forest plots studied. The results of the phylogeneticsignal analyses generally corroborated the results of theSAPD analyses. In particular, we expected that phylogeneticsignal in abundance would cause an increasing trendin the SAPD curve. In each of the forest plots with asignificant increasing trend in the SAPD curve, we detectedphyogenetic signal in abundance (table 4; figs. 2, C1, availablein a zip file). Thus, phylogenetic signal in relativeabundance does influence the shapes of the SAPD curves.Our last set of analyses quantified the phylogenetic similaritywithin and between groups of common and rarespecies in each of the 15 forest plots. The SES values inthe between-group similarity analyses (SES (D nn )) wereTable 4: Phylogenetic signals inspecies abundance dataForest dynamics plotValueChangbaishan .295 (.01)Gutianshan .185 (.001)Xishuangbanna .227 (.032)Sinharaja .176 (.001)Lambir .104 (.001)Pasoh .103 (.001)Yasuní .074 (.001)BCI .099 (.001)Huai Kha Khaeng .199 (.001)La Planada .116 (.051)Dinghushan .136 (.059)Fushan .217 (.001)Palanan .128 (.021)Mudumalai .150 (.375)Luquillo .144 (.132)Note: The data are K values, which indicatethe comparison between the observedand the expected phylogenetic signalunder the assumption of Brownian motion,followed by the numbers in parentheses,which indicate the probability that the observedphylogenetic signal is greater thanthe null expectation generated by randomlyshuffling the species abundances across thetips of the phylogeny. BCI, Barro ColoradoIsland.161

E28The American Naturalistalmost always negative, indicating that common and rarespecies tended to be phylogenetically dissimilar (table 5),but many of these results were not statistically significant.These results therefore only weakly corroborate the resultsof the above analyses, showing that on average, rare speciesare phylogenetically distinct from common species in theforests studied. This further supports the prediction ofniche differentiation hypotheses, where common and rarespecies are expected to be dissimilar. It is also interestingto note that the results from the within-common-groupanalyses generally had positive NTI values and rare specieshad negative values (table 5). Thus, common species aretypically phylogenetically clustered and rare species arephylogenetically overdispersed in the forests studied. Thissupports the prediction of niche differentiation hypotheses,where rare species are not only distinct from commonspecies but also distinct from one another. Further, thisdemonstrates why measuring phylogenetic signal in relativeabundances alone does not provide a complete pictureof how commonness and rarity relate to phylogeny.When the three lines of evidence—the SAPD curve, thephylogenetic signal analyses, and the phylogenetic similarityanalyses—are considered together, we believe thatthe SAPD curve analyses provide the best window into thecontribution of rare species to community PD and theirphylogenetic similarity with respect to common species.This is because the SAPD curve can be used to detect finescaleshifts in how abundance relates to phylogeny. Thislevel of resolution is difficult to achieve by using measuresof phylogenetic signal or phylogenetic similarity that effectivelyaverage over the entire phylogeny. For example,the less than expected contribution of rare species to communityPD in the BCI plot according to the SAPD curvesseems to contradict the evidence from the phylogeneticsignal results for the plots. Upon closer examination, wesee that rare species in the BCI plot had less PD thanexpected by chance (table 3) and rare species were phylogeneticallyclustered (table 5). Thus, by combining thesimilarity and signal analyses, we can link these resultsback to those from the SAPD curve. In other words, phylogeneticsignal is likely detected in this plot because ofspecies ranking from 1 to 131 and species ranking from132 to 210 being distantly related to one another (fig. 2b)and rare species being phylogenetically clustered (table 5)such that they cumulatively contribute little to the communityPD.ConclusionsIn summary, rare species were found to be distantly relatedto common species and to have significantly higher cumulativePD than expected by chance in six of the ninegap-dominated forests, supporting the predictions madeby niche differentiation hypotheses. We therefore inferredTable 5: Phylogenetic similarity within and between common and rarespecies for 15 forest dynamics plotsNTIForest dynamics plot SES of D nn Common species Rare speciesChangbaishan 1.620 .082 .948Gutianshan 1.052 .016 1.083Dinghushan .217 .509 .373Xishuangbanna .169 1.009 1.102Sinharaja .624 1.514 1.199Lambir 2.872** .645 .579Pasoh .182 1.903* 1.749*Yasuní 3.304*** 2.358*** 1.096***BCI .659 .453 .226Huai Kha Khaeng 3.323*** .941 .059La Planada 1.174 .971 1.092Fushan .367 .284 .215Palanan .534 .976 .051Mudumalai 1.633 .749 1.293Luquillo .325 1.018 .354Note: Nearest taxon index (NTI) of common and rare species and standardizedeffect size of mean nearest neighbor distance (SES of D nn ) between common speciesand rare species. BCI, Barro Colorado Island.* P ! .05.** P ! .01.*** P ! .001.162

Rare Species to Community PhylogenyE29that in these six forests, rare species may have spatially ortemporally divergent niches that permit their co-occurrencewith common species in these forests. In contrast,rare species in six disturbance-dominated forests and threegap-dominated forests were found to be closely or randomlyrelated to common species and have less cumulativePD or ED, meaning that niche differentiation hypothesesare rejected and the environmental filtering hypothesis issupported. Along with these biological inferences, we havepresented a novel methodology for examining the contributionof increasingly rare or increasingly common speciesto community PD that provides finer-scale insights thatcannot be achieved by using metrics that average over allspecies in a community.AcknowledgmentsThe data analyses reported in this study were supportedby the Key Innovation Project of C.A.S. (KSCX2-EW-Z-5) and the Natural Science Foundation of China projects(31061160188 and 31011120470). We are grateful to C.Cannon, L. Comita, R. Condit, M. Helmus, F. Slik, H. Su,and two reviewers for their insightful comments and suggestionson the earlier form of the manuscript. Data collectionwas funded by many organizations, principally, theNational Science Foundation; the Andrew W. MellonFoundation; the Peninsula Community Foundation; theSmithsonian Institution; the Smithsonian Tropical ResearchInstitute; the Arnold Arboretum (Harvard University);the Indian Institute of Science; the Forest ResearchInstitute of Malaysia; the Royal Thai Forest Department,the National Institute of Environmental Studies (Japan);the Taiwan Forestry Research Institute and Forestry Bureau;SENESCYT (Secretaría Nacional de Educación Superior,Ciencia, Tecnología e Innovación), PUCE (PontificiaUniversidad Católica de Ecuador), and Donacionesdel Impuesto a la Renta (Ecuador); and the John D. andCatherine T. MacArthur Foundation. We thank the hundredsof field-workers who measured and mapped the treesanalyzed in this study.Literature CitedAbellán, P., D. T. Bilton, A. Millán,D.Sánchez-Fernández, and P.M. Ramsay. 2006. Can taxonomic distinctness assess anthropogenicimpacts in inland waters? a case study from a Mediterraneanriver basin. Freshwater Biology 51:1744–1756.Anderson, T. M., M. A. Lachance, and W. T. Starmer. 2004. Therelationship of phylogeny to community structure: the cactus yeastcommunity. American Naturalist 164:709–721.Andrews, D. W. K. 1993. Tests for parameter instability and structuralchange with unknown change point. Econometrica 61:821–856.Angiosperm Phylogeny Group III. 2009. An update of the AngiospermPhylogeny Group classification for the orders and familiesof flowering plants: APG III. Botanical Journal of the LinneanSociety 161:105–121.Betts, M. G., G. J. Forbes, and A. W. Diamond. 2007. Thresholds insongbird occurrence in relation to landscape structure. ConservationBiology 21:1046–1058.Blomberg, S. P., T. Garland Jr., and A. R. Ives. 2003. Testing forphylogenetic signal in comparative data: behavioral traits are morelabile. Evolution 57:717–745.Burnham, K. P., and D. R. Anderson. 2002. Model selection andmultimodel inference: an information theoretic approach.Springer, New York.Burns, J. H., and S. Y. Strauss. 2011. More closely related species aremore ecologically similar in an experimental test. Proceedings ofthe National Academy of Sciences of the USA 108:5302–5307.Cadotte, M. W., B. J. Cardinalec, and T. H. Oakley. 2008. Evolutionaryhistory and the effect of biodiversity on plant productivity. Proceedingsof the National Academy of Sciences of the USA 105:17012–17017.Cadotte, M. W., J. Cavender-Bares, D. Tilman, and T. H. Oakley.2009. Using phylogenetic, functional and trait diversity to understandpatterns of plant community productivity. PLoS ONE 4:e5695.Cadotte, M. W., T. J. Davies, J. Regetz, S. W. Kembel, E. Cleland,and T. H. Oakley. 2010. Phylogenetic diversity metrics for ecologicalcommunities: integrating species richness, abundance and evolutionaryhistory. Ecology Letters 13:96–105.Chesson, P. L. 2000. Mechanisms of maintenance of species diversity.Annual Review of Ecology, Evolution and Systematics 31:343–366.Clark, J. S., M. Dietze, S. Chakraborty, P. K. Agarwal, I. Ibanez, S.LaDeau, and M. Wolosin. 2007. Resolving the biodiversity paradox.Ecology Letters 10:647–659.Douglas, E. M., R. M. Vogel, and C. N. Kroll. 2000. Trends in floodsand low flows in the United States: impact of spatial correlation.Journal of Hydrology 240:90–105.Faith, D. P. 1992. Conservation evaluation and phylogenetic diversity.Biological Conservation 61:1–10.Gaston, K. J. 1994. Rarity. Chapman & Hall, London.Grime, J. P. 1998. Benefits of plant diversity to ecosystems: immediate,filter and founder effects. Journal of Ecology 86:902–910.Hanski, I. 1982. Dynamics of regional distribution: the core andsatellite species hypothesis. Oikos 38:210–221.Helmus, M. R., T. J. Bland, C. K. Williams, and A. R. Ives. 2007.Phylogenetic measures of biodiversity. American Naturalist 169:E68–E83.Helmus, M. R., W. Keller, M. J. Paterson, N. D. Yan, C. H. Cannon,and J. A. Rusak. 2010. Communities contain closely related speciesduring ecosystem disturbance. Ecology Letters 13:162–174.Hirsch, R. M., and J. R. Slack. 1984. A nonparametric trend test forseasonal data with serial dependence. Water Resources Research20:727–732.Hubbell, S. P. 2001. The unified neutral theory of biodiversity andbiogeography. Princeton University Press, Princeton, NJ.Hubbell, S. P., and R. B. Foster. 1986. Commonness and rarity in aNeotropical forest: implications for tropical tree conservation.Pages 205–231 in M. E. Soule, ed. Conservation biology: the scienceof scarcity and diversity. Sinauer, Sunderland, MA.Isaac, N. J. B., S. T. Turvey, B. Collen, C. Waterman, and J. E. M.163

E30The American NaturalistBaillie. 2007. Mammals on the EDGE: conservation priorities basedon threat and phylogeny. PLoS ONE 2:e296.Jarvinen, O. 1982. Species-to-genus ratios in biogeography: a historicalnote. Journal of Biogeography 9:363–370.Keddy, P. A. 1989. Competitition. Chapman & Hall, London.Kelly, C. K., and M. G. Bowler. 2002. Coexistence and relative abundancein forest trees. Nature 417:437–440.Kelly, C. K., M. G. Bowler, O. Pybus, and P. H. Harvey. 2008. Phylogeny,niches, and relative abundance in natural communities.Ecology 89:962–970.Kembel, S. W., P. D. Cowan, M. R. Helmus, W. K. Cornwell, H.Morlon, D. D. Ackerly, S. P. Blomberg, et al. 2010. Picante: R toolsfor integrating phylogenies and ecology. Bioinformatics 26:1463–1464.Kress, W. J., D. L. Erickson, F. A. Jones, N. G. Swenson, R. Perez,O. Sanjur, and E. Bermingham. 2009. Plant DNA barcodes and acommunity phylogeny of a tropical forest dynamics plot in Panama.Proceedings of the National Academy of Sciences of the USA106:18621–18626.Kress, W. J., D. L. Erickson, N. G. Swenson, J. Thompson, M. Uriarte,and J. K. Zimmerman. 2010. Improvements in the application ofDNA barcodes in building a community phylogeny for tropicaltrees in a Puerto Rican forest dynamics plot. PLoS ONE 5:e15409.Kunin, W. E. 1997. The biology of rarity: causes and consequencesof rare-common differences. Chapman & Hall, London.Losos, E. C., and E. G. J. Leigh. 2004. Tropical forest diversity anddynamism: findings from a large-scale plot network. University ofChicago Press, Chicago.Losos, J. B. 2008. Phylogenetic niche conservatism, phylogenetic signaland the relationship between phylogenetic relatedness and ecologicalsimilarity among species. Ecology Letters 11:995–1003.McLeod, A. I. 2011. Kendall rank correlation and Mann-Kendalltrend test. Ver. 2.2. http://cran.r-project.org/.Muggeo, V. M. R. 2003. Estimating regression models with unknownbreak-points. Statistics in Medicine 22:3055–3071.———. 2011. Segmented relationships in regression models. R package,ver. 0.2–8.2. http://www.r-project.org.R Develoment Core Team. 2009. R: a language and environment forstatistical computing. R Foundation for Statistical Computing, Vienna.http://www.r-project.org.Redding, D. W., and A. O. Mooers. 2006. Incorporating evolutionarymeasures into conservation prioritization. Conservation Biology20:1670–1678.Swenson, N. G. 2011. The role of evolutionary processes in producingbiodiversity patterns, and the interrelationships between taxonomic,functional and phylogenetic biodiversity. American Journalof Botany 98:472–480.Swenson, N. G., P. Anglada-Cordero, and J. A. Barone. 2011. Deterministictropical tree community turnover: evidence from patternsof functional beta diversity along an elevational gradient.Proceedings of the Royal Society B: Biological Sciences 278:877–884.Swenson, N. G., and B. J. Enquist. 2009. Opposing assembly mechanismsin a Neotropical dry forest: implications for phylogeneticand functional community ecology. Ecology 90:2161–2170.Swenson, N. G., B. J. Enquist, J. Thompson, and J. K. Zimmerman.2007. The influence of spatial and size scale on phylogenetic relatednessin tropical forest communities. Ecology 88:1770–1780.Toms, J. D., and M. L. Lesperance. 2003. Piecewise regression: a toolfor identifying ecological thresholds. Ecology 84:2034–2041.Vamosi, S. M., S. B. Heard, J. C. Vamosi, and C. O. Webb. 2009.Emerging patterns in the comparative analysis of phylogeneticcommunity structure. Molecular Ecology 18:572–592.Warwick, R. M., and K. R. Clarke. 1998. Taxonomic distinctness andenvironmental assessment. Journal of Applied Ecology 35:532–543.Webb, C. O. 2000. Exploring the phylogenetic structure of ecologicalcommunities: an example for rain forest trees. American Naturalist156:145–155.Webb, C. O., D. D. Ackerly, and S. W. Kembel. 2008. Phylocom:software for the analysis of phylogenetic community structure andtrait evolution. Bioinformatics 24:2098–2100.Webb, C. O., D. D. Ackerly, M. A. McPeek, and M. J. Donoghue.2002. Phylogenies and community ecology. Annual Review of Ecologyand Systematics 33:475–505.Webb, C. O., and M. J. Donoghue. 2005. Phylomatic: tree assemblyfor applied phylogenetics. Molecular Ecology Notes 5:181–183.Wiens, J. J., D. D. Ackerly, A. P. Allen, B. L. Anacker, L. B. Buckley,H. V. Cornell, E. I. Damschen, et al. 2010. Niche conservatism asan emerging principle in ecology and conservation biology. EcologyLetters 13:1310–1314.Wikstrom, N., V. Savolainen, and M. W. Chase. 2001. Evolution ofthe angiosperms: calibrating the family tree. Proceedings of theRoyal Society B: Biological Sciences 268:2211–2220.Yu, Y. S., S. Zou, and D. Whittemore. 1993. Non-parametric trendanalysis of water quality data of rivers in Kansas. Journal of Hydrology150:61–80.Zeileis, A., C. Kleiber, W. Krämer, and K. Hornik. 2003. Testing anddating of structural changes in practice. Computational Statisticsand Data Analysis 44:109–123.Zeileis, A., F. Leisch, K. Hornik, and C. Kleiber. 2011. strucchange:an R package for testing for structural change in linear regressionmodels. R package, ver. 1.4–6. http://www.r-project.org.Associate Editor: Stephen B. HeardEditor: Mark A. McPeek164

Covariation in Plant Functional Traits and Soil Fertilitywithin Two Species-Rich ForestsXiaojuan Liu 1,5 , Nathan G. Swenson 2 , S. Joseph Wright 3 , Liwen Zhang 4 , Kai Song 1 , Yanjun Du 1 ,Jinlong Zhang 1 , Xiangcheng Mi 1 , Haibao Ren 1 , Keping Ma 1 *1 State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, The Chinese Academy of Sciences, Xiangshan, Beijing, China, 2 Department of PlantBiology, Michigan State University, East Lansing, Michigan, United States of America, 3 Smithsonian Tropical Research Institute, Balboa, Panama, 4 Yantai Institute ofCoastal Zone Research, The Chinese Academy of Sciences, Yantai, Shandong, China, 5 Graduate University of The Chinese Academy of Sciences, Beijing, ChinaAbstractThe distribution of plant species along environmental gradients is expected to be predictable based on organismal function.Plant functional trait research has shown that trait values generally vary predictably along broad-scale climatic and soilgradients. This work has also demonstrated that at any one point along these gradients there is a large amount ofinterspecific trait variation. The present research proposes that this variation may be explained by the local-scale sorting oftraits along soil fertility and acidity axes. Specifically, we predicted that trait values associated with high resource acquisitionand growth rates would be found on soils that are more fertile and less acidic. We tested the expected relationships at thespecies-level and quadrat-level (20620 m) using two large forest plots in Panama and China that contain over 450 speciescombined. Predicted relationships between leaf area and wood density and soil fertility were supported in some instances,but the majority of the predicted relationships were rejected. Alternative resource axes, such as light gradients, thereforelikely play a larger role in determining the interspecific variability in plant functional traits in the two forests studied.Citation: Liu X, Swenson NG, Wright SJ, Zhang L, Song K, et al. (2012) Covariation in Plant Functional Traits and Soil Fertility within Two Species-Rich Forests. PLoSONE 7(4): e34767. doi:10.1371/journal.pone.0034767Editor: Kurt O. Reinhart, USDA-ARS, United States of AmericaReceived November 4, 2011; Accepted March 5, 2012; Published April 3, 2012Copyright: ß 2012 Liu et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricteduse, distribution, and reproduction in any medium, provided the original author and source are credited.Funding: This work was supported by a key innovation project of the Chinese Academy of Sciences (KZCX2-YW-430) and a research grant from the Center forTropical Forest Science (CTFS). The F. H. Levinson Fund supported functional traits measurements at Barro Colorado Island (BCI). The funders had no role in studydesign, data collection and analysis, decision to publish, or preparation of the manuscript.Competing Interests: The authors have declared that no competing interests exist.* E-mail: kpma@ibcas.ac.cnIntroductionThe distribution of species and communities along environmentalgradients is a central focus in ecology. The distribution ofspecies is expected to be determined by the distribution ofresources. The functional strategy of a species will dictate itsresource use and therefore its location along a resource axis orresource axes. Thus function should vary predictably along thesegradients. This has lead to a tradition in plant ecology ofpredicting and analyzing the geographic distribution of functionalstrategies [1,2].The relationship between plant function traits and environmentalgradients has been quantified for a number of plant traitsusing large-scale datasets. Evidence from these broad-scalefunctional trait analyses suggest that the mean functional traitvalue of an assemblage changes predictably along environmentalgradients. For example, leaf and wood traits, seed mass andmaximum height have been shown to vary predictably with meanannual temperature [3–10]. Additional studies have also examinedthe relationship between leaf and wood traits with soil nutrientlevels. Leaf economic traits related to resource acquisition such asspecific leaf area, leaf nitrogen content and leaf phosphorus arepositively correlated with soil nutrient content and theserelationships were stronger than those with climatic gradients[11]. Wood density, which is negatively correlated with volumetricgrowth rates, is negatively correlated with nitrogen and phosphoruslevels across the Amazon Basin [12]. A running theme in manyof these papers is there is a trade-off between the structuralallocation and demographic rates based on the resourceavailability. Specifically, species that favor high resource environmentsshould have higher growth and mortality rates wherebiomass is allocated to producing a large amount of small seedsthat germinate quickly, structurally cheap leaves that have highspecific leaf areas but photosynthesize at a high rate, andstructurally cheap wood that permits rapid volumetric growthinto the canopy. In contrast, species that favor low resourceenvironments should be characterized by ecological strategies thatincrease structural investment at the cost of decreased resourceacquisition and demographic rates. While many of the abovestudies have supported the expected relationships betweenenvironmental gradients and plant traits across broad gradients,this work has also demonstrated that a tremendous level of interspecificvariation occurs within locations along the gradient [13].The large inter-specific trait variation within sites in globaldatasets could be the result of trait – environmental gradientrelationships on local scales and how different ecological strategiesrelated to resource acquisition and demographic rates sort outalong important resource axes. For example, given the previousresearch showing strong and consistent relationships between planttraits and soil nutrients on global scales, it is expected that localscale plant trait distributions should also vary predictably alonglocal scale soil nutrient gradients. In particular, we predict thatPLoS ONE | www.plosone.org 1 April 2012 | Volume 7 | Issue 4 | e34767165

Traits, Soil Fertility and Distributionindividual species with plant traits associated with high rates ofresource acquisition and growth such as high values of specific leafarea, maximum height and leaf area and low values of seed massand wood density are predicted to occur on soils with high nutrientcontent. Conversely, species with low values of specific leaf area,maximum height and leaf area and high values of seed mass andwood density are expected to be located in soils with low nutrientlevels.Here, we integrate tree distribution and soil nutrient data withfive plant functional traits – specific leaf area, maximum height,leaf area, seed mass and wood density to test the predictedrelationships among local-scale gradients in soil nutrient levels. Inparticular, we quantify: (1) the correlation between species meantrait values and their mean position on soil nutrient gradients and(2) the correlation between the mean trait value in 20620 mquadrats and the soil nutrient level in that quadrat. The analysesare performed separately in two forest inventory plots. The twoforest plots were chosen for two important reasons. First, theyshare similar forest inventory, trait collection and soil nutrientmapping protocols making a comparative study feasible. Second,the forests are vastly different in their topographic heterogeneitythereby allowing us to determine whether the degree of localhabitat heterogeneity influences the strength of trait-soil relationships.We first test the above predictions using species-level dataand then ask whether the species-level relationships scale-up to thequadrat-level where the mean trait value within a quadrat can bepredicted based on the soil nutrient levels in that quadrat.Materials and MethodsResearch SitesThe datasets used in this study were compiled from twopermanent large forest dynamics plots in tropical and subtropicalforests. The Barro Colorado Island (BCI) 50-ha forest dynamicsplot (9u109N, 79u519W) is located on well-weathered kaoliniticOxisols on Barro Colorado Island, Panama (Fig. 1), and it ischaracterized as a lowland semideciduous moist forest. In the10006500 m rectangular area, the plot spans an altitudinal rangeof 120 to 160 m and the slope ranges from 7u to 20u. Dailymaximum and minimum temperatures average 30.8uC and23.4uC, respectively. Annual rainfall averages 2600 mm, with just10% of the annual total falling during a 4-mo dry season. The BCIplot was first censused in 1981/82 [14]. All trees with a diameterat breast height (dbh) $1 cm were measured, identified andmapped. A second census was performed in 1985 and censuseshave been repeated every 5 years thereafter. Here we use the 2005census, which includes 208,387 individual trees belonging to 299species.The Gutianshan (GTS) 24-ha permanent plot (29u159N,118u079E) is located in the old-growth forest of GutianshanNational Nature Reserve, Kaihua County, Zhejiang Province,Southeast China (Fig. 1), and it is characterized as a subtropicalevergreen broad-leaved forest. The GTS forest plot containsapproximately 140,000 individual trees (dbh$1 cm), representing49 families, 103 genera and 159 species in the plot. It wasestablished in the summer of 2005, following the same protocol asfor BCI [15,16]. In the 6006400 m rectangular area, the plotspans an altitudinal range of 446.3 to 714.9 m and the sloperanges from 13u to 62u. The mean annual temperature in theGutianshan Reserve is 15.3uC. The hottest month is July (meantemperature of 27.91uC), and the coldest is January (meantemperature of 4.31uC). Annual precipitation averages1963.7 mm, with a dry period between October and February.The major soils can be classified into four types: red soil, redyellowsoil, yellow-red soil and marsh soil [15].Plant Functional TraitsWe measured leaf area (LA), specific leaf area (SLA), wooddensity (WD), seed mass (SM) and maximum height (H max ) forspecies at both sites. The trait collection protocols for BCI aredescribed in Wright et al. [17], and the GTS collection protocolsfollowed Cornelissen et al. [18] with the exception of WD whichfollowed the protocols of Wright et al. [17]. Below we brieflydescribe the collection methods and sample sizes for the GTS plot.Leaf area and SLA were measured using at least ten matureleaves collected from the tallest portion of the canopies of 5–10 ofthe largest individuals of each species. The SLA was calculated asmean of fresh leaf area divided by the leaf dry mass without thepetiole. The LA was measured as the mean leaf surface areawithout petioles for each species. The SM was calculated bycollecting 30 to 200 mature, fresh seeds from more than fiveindividual trees of each species in or near the plot. We removedappendages and oven dried seeds for 48 h at 80uC. The SM valueis the mean value over all seeds of each species. The WD wascalculated by collecting wood samples from 5 to 10 individuals foreach species in the area surrounding the plot using methodsdescribed in Wright et al. [17]. The H max values for GTS wereestimated using values reported in the Flora of China [19] and theFlora Reipublicae Popularis Sinicae [20].Phylogenetic TreesTwo phylogenetic trees were utilized in this study. Specifically,we utilized a phylogenetic tree from Kress et al. [21] for the BCIplot. This phylogeny was constructed using a DNA supermatrixcomposed of three sequence regions - rbcL, matK, and trnHpsbA.The supermatrix and the software RAxML [22] were usedto construct a maximum likelihood phylogenetic tree. Weconstructed a phylogenetic tree for the GTS forest plot speciesfollowing the same methodology as Kress et al. [21]. Figures ofboth phylogenetic trees are available in the supplemental material(see Figure S1& Figure S2).Soil FertilitySoil samples were collected and analyzed following the protocolsestablished by the Center for Tropical Forest Science (CTFS) inboth plots (http://ctfs.si.edu/datasets/bci/soilmaps/BCIsoil.html)[23]. However, the sampling design and intensity differed. At BCI,the plot was divided into a 50650 m grid, grid intersections werebasal collection points, and additional collection points weremarked at 2 m, 8 m or 20 m along a random compass directionfrom each basal point. Thus, 300 points were sampled in the 50-ha plot. At GTS, the grid was 30630 m, the additional collectionpoints were 2, 5 and 15 m along a random compass direction fromeach basal point, and a total of 892 samples were collected insidethe 24-ha plot.John et al. [23] describe the methods used to process BCI soilsamples. At GTS, a 300–400 g topsoil sample was taken from 0–10 cm depth and air-dried. The soil was then sieved with a 2 mmmesh screen. The sieved soil was used to extract available cations.50 g of the 2 mm-filtered soil was filtered again with 0.15 mmmesh screen for analyses of total C, N, and P. Additional samplesfrom a depth of 15 cm were taken using two polyethene pipes witha diameter of 5 cm. One of these samples was used for extractingNH 4 + and NO 3 2 (using 2.0 M KCl on 2 g soil) and measuringgravimetric moisture content and pH value (soil: water was 1:5).The other sample was sealed and left in the original state for 26–31days in order to measure N mineralization rates. Available cationsPLoS ONE | www.plosone.org 2 April 2012 | Volume 7 | Issue 4 | e34767166

Traits, Soil Fertility and DistributionFigure 1. A map of the geographic location of the 50-ha BCI plot, Barro Colorado Island, Panama and the 24-ha GTS plot,Gutianshan National Nature Reserve, China.doi:10.1371/journal.pone.0034767.g001were extracted using Mehlich-III extractant solution and elementalanalysis was done by Atomic Emission-Inductively CoupledPlasma (AE-ICP) spectrometry. We analyzed NH 4 + and NO 32with a Continuous Flow Analyzer in the Key Laboratory of Plant-Soil Interactions, China Agriculture University. We used Walkley-Black method to estimate total C and used the Kjeldahl NitrogenDetermination method to measure total N. Total P was measuredby UV-Spectrometer in the State Key Laboratory of Vegetationand Environmental Change, Institute of Botany, CAS. Finally, weobtained the values of 13 soil nutrients (Fe, Mn, Zn, Cu, K, P, Ca,Mg, B, Al, N, pH, Nmin) for both plots by ordinary kriging.Additional detailed information regarding soil data collection canbe found in John et al. [23] for BCI and Liwen et al. [24] for GTS.For each forest plot, we used a principal components analysis(PCA) to extract orthogonal axes of soil fertility and acidity fromthe 13 measured soil nutrients and to reduce informationredundancy. We used the significant PCA axes to characterizesoil fertility and acidity for all subsequent analyses.Statistical AnalysesOur datasets included mean trait values (T) for each species S(T S ), soil fertility (F) for each 20620 m quadrat Q (F Q ) and thenumber of individuals of each species in each quadrat (N SQ ). Weused these measured values to calculate mean soil fertilities for eachspecies (F S ) and mean trait values for each quadrat (T Q ) as follows:F S ~S Q N SQ |F QSQ N SQT Q ~S S N SQ |T SSS N SQWe first performed a species-level analysis to test our predictionsby calculating a Pearson correlation of trait values and mean soilproperties for species calculated by weighting the PCA scores ofeach 20620 m quadrat by the number of individuals of species inthat same quadrat (eqn. 1) The LA and SM values for species inboth forest plots were log transformed to satisfy the normalityassumption. Next we used phylogenetically independent contrasts(PICs) [25–27] to evaluate relationships between measured valuesð1Þð2Þof T S and calculated values of F S . This second analysis was used tofactor out the bias of phylogenetic non-independence and toevaluate the hypothesis that evolutionary changes in trait valueswere associated with the spatial distribution of species with respectto soil fertility. PIC regressions were forced through the origin [28]and significance was evaluated after removing extreme outliers(absolute value of studentized residual.5) and contrasts withundue leverage (leverage.0.2).We performed a third correlation analysis to evaluate therelationship between calculated T Q (eqn. 2) and measured/estimated F Q . The LA and SM values from the BCI plot were logtransformed to satisfy the normality assumption. This quadratlevelanalysis was used to test whether quadrat-level traitdistributions shift in a predictable direction along the soil fertilityand acidity gradients within each forest plot. This analysis wasthen repeated using torus translation simulations [29]. Theprocedure included two steps: 1) we moved the true soil map by20-m increments two-dimensionally, but kept the above trees mapstill; 2) We recalculated the Pearson correlation between T Q andF Q based on 20620 m quadrats for each simulation andcompared the observed and simulated correlation coefficients. Ifthe rank of the r-value from the true quadrat was higher than97.5% or lower than 2.5% of the ranks of the simulated r-values(two-tailed test), it was considered that T Q and F Q was significantlycorrelated. The torus translations maintained the observed spatialdistribution of soil fertility and tree distributions, but break theirobserved dependence by shifting the observed soil fertilitydistribution on a torus relative to fixed tree distributions.We performed the three correlation analyses for five plant traits(SM, LA, SLA, WD, H max ) and the soil properties of the first twoprincipal component axes (see Results: Soil Properties). Wetherefore used the false discovery rate (FDR) approach to adjustp-values [30,31]. Except the torus translation simulation, all otherreported p-values refer to the adjusted p-values. All analyses wereperformed in R 2.13.0 (R core team, 2011).ResultsSoil PropertiesThe significant soil PCA axes combined to explain more than70% the variation of soil nutrients at each site (Table 1). For GTS,PLoS ONE | www.plosone.org 3 April 2012 | Volume 7 | Issue 4 | e34767167

Traits, Soil Fertility and Distributionthe first axis explained 42.2% of the overall variation andrepresented a general soil fertility axis with negative loadings onmost key limiting elements. The second axis explained 17.8% ofthe overall variation and represented an acidity index with arelatively large negative loading on pH (20.461) and positiveloadings on Fe (0.579) and B (0.610). The third axis explained 12%of the overall variation (Table 1).For BCI, the PC1 axis explained 55.3% of the overall variation.Similar to the GTS analysis, this axis was also generally indicativeof a soil fertility index with most elements decreasing. Againsimilar to GTS, the PC2 axis for BCI explained more than 12% ofthe overall variation and was an acidity index with low pH(20.470) and N (20.595) and high Al (0.348) and Fe (0.336). ThePC3 axis explained about 11% of the overall variation andcaptured a large correlation between Al (0.600) and P (0.692)(Table 1). This is also similar to the third axis at GTS, which hadlarge loadings of the same sign for Al and P (Table 1).For both plots, the PC1 axis was negatively related to soilfertility as shown by the negative loadings of most essentialnutrients. Therefore, a negative correlation between a mean traitvalue and the soil fertility PC1 axis meant that trait values werelarger on more fertile soils. The PC2 axis was positively related tosoil acidity as shown by the negative loading of pH on PC2 (orlower pH for larger values of PC2) (Table 1). Therefore, a negativecorrelation with the soil acidity PC2 axis meant that larger traitvalues occurred on less acidic soils. Given the similarities betweenthe loadings of nutrients on the first two PCA axes in both plotsand because of their interpretability as general fertility and acidityaxes, the following will focus on these first two axes.Species-level Relationships between Trait Values and SoilPropertiesSpecies functional trait values (T S ) were unrelated to calculatedmean species soil properties (F S ) at BCI (Table 2). For GTS, afterTable 1. Principal component analyses for 13 soil fertility forthe GTS plot and the BCI plot.GTSBCIPC1 PC2 PC3 PC1 PC2 PC3Al 20.012 0.172 20.396 0.167 0.348 0.600B 20.050 0.610 0.109 20.339 20.131 0.111Ca 20.391 20.031 0.150 20.355 0.049 20.041Cu 20.321 0.031 0.334 20.305 0.218 0.115Fe 0.039 0.579 0.001 20.278 0.336 0.133K 20.348 0.086 20.075 20.353 0.009 0.013Mg 20.409 20.002 0.064 20.331 0.044 20.028Mn 20.324 20.075 0.374 20.254 0.245 0.206Zn 20.359 0.165 20.096 20.338 20.006 20.048N 20.324 0.027 20.088 20.127 20.595 0.243Nmin 20.206 20.083 20.519 20.269 0.102 20.109P 20.234 20.024 20.511 0.038 20.228 0.692pH 20.125 20.461 0.015 20.240 20.470 20.005Eigenvalue 5.483 2.316 1.566 7.187 1.582 1.44% explained 42.2 17.8 12 55.3 12.2 11.1Entries are component loadings; eigenvalues and percentage of variationexplained for the three significant principal components for each site.Significant loadings are in boldface type.doi:10.1371/journal.pone.0034767.t001the false discovery rate (FDR) correction to p-values, LA wasnegatively related to the soil fertility axis PC1 (r = 20.220;p = 0.008) and LA (r = 20.275; p,0.001), SLA (r = 20.221;p = 0.008) and WD (r = 0.269; p,0.001) were significantly relatedto the soil acidity axis PC2 (Table 2). We repeated these analysesfor individual soil variables (see Table S1& Table S2). In sum, inboth plots several significant relationships were uncoveredbetween individual soil variables and all traits except for H max atGTS and SM at BCI.Phylogenetically Independent Species-level Relationshipsbetween Trait Values and Soil PropertiesA second set of species level analyses were performed usingphylogenetically independent contrasts (PICs) to account for theevolutionary non-independence of species. In the GTS plot, fivesignificant relationships were found between trait contrasts and soilcontrasts (F S from eqn. 1) (Table 3). In particular, significantpositive relationships were found between SLA and the soil fertilityaxis PC1 (r = 0.302; p,0.001), SM and the soil acidity axis PC2(r = 0.197; p = 0.028) and WD and PC2 (r = 0.286; p,0.001).Negative relationships were found between LA and the soil fertilityaxis PC1 (r = 20.176; p = 0.036) and SLA and the soil acidity axisPC2 (r = 20.245; p = 0.007).In the BCI plot, five significant relationships were also foundbetween trait contrasts and soil contrasts (Table 3). The soilfertility axis PC1 was positively correlated with SM (r = 0.352;Table 2. Pearson correlation coefficients between fivefunctional traits and the calculated scores of the twosignificant principal components of soil fertility and acidity forboth GTS and BCI at the species-level (Leaf area and Seedmass of both plots are log 10 transformed).GTSBCIPC1 PC2 PC1 PC2Leaf area r 20.220 20.275 20.067 20.074n 157 157 283 283p 0.003 0.000 0.131 0.107p-adj 0.008 0.000 0.330 0.330Specific leaf area r 0.048 20.221 20.032 20.028n 157 157 284 284p 0.275 0.003 0.296 0.319p-adj 0.324 0.008 0.399 0.399Seed mass r 20.105 0.065 0.014 20.086n 141 141 171 171p 0.108 0.222 0.428 0.132p-adj 0.216 0.324 0.476 0.330Wood density r 0.047 0.269 20.048 20.034n 157 157 262 262p 0.279 0.000 0.220 0.292p-adj 0.324 0.000 0.399 0.399Maximum height r 20.044 0.010 20.000 20.118n 157 157 283 283p 0.292 0.451 0.500 0.023p-adj 0.324 0.451 0.500 0.230Significant correlations are in boldface type (p-value,0.05 after the FalseDiscovery Rate adjustment).doi:10.1371/journal.pone.0034767.t002PLoS ONE | www.plosone.org 4 April 2012 | Volume 7 | Issue 4 | e34767168

Traits, Soil Fertility and DistributionTable 3. Phylogenetically independent contrasts (PICs)between five functional traits and the calculated scores of thetwo significant principal components of soil fertility andacidity for both GTS and BCI.GTSp,0.001) and WD (r = 0.167; p = 0.010). The soil acidity axis PC2was positively correlated with LA (r = 0.268; p,0.001) and SM(r = 0.274; p,0.001) and negatively correlated with WD(r = 20.202; p = 0.003).Similar to the non-phylogenetic analyses, we repeated allanalyses using individual soil nutrients (see Table S3 & TableS4). Similar to the above non-phylogenetic results, severalsignificant relationships were uncovered in both plots betweenindividual soil variables and all traits except for SM at GTS.Relationships between Quadrat-Level Mean Trait Valuesand Soil PropertiesA second goal of this study was to test whether our predictionsregarding species-level trait – soil relationships scale-up to thequadrat-level. Fifteen of the 20 possible relationships weresignificant after the FDR correction to probability levels(Table 4). In the GTS plot, there was a negative relationshipbetween the soil fertility axis PC1 and LA (Fig. 2; r = 20.394;p,0.001), SM (r = 20.450; p,0.001) and WD (r = 20.179;p,0.001) and positive relationship with SLA (r = 0.224;p,0.001) and H max (r = 0.186; p,0.001). Negative relationshipswere found between the soil acidity axis PC2 and LA (r = 20.458;p,0.001), SM (r = 20.130; p = 0.001) and H max (r = 20.115;p = 0.003), and a positive relationships with WD (r = 0.325;p,0.001). For the distribution pattern of LA and soil PC1 valuessee Fig. 2. The spatial distribution of the other traits and soil PC2BCIPC1 PC2 PC1 PC2Leaf area r 20.176 0.130 0.032 0.268n 144 145 253 247p 0.018 0.057 0.356 ,0.001p-adj 0.036 0.076 0.445 ,0.001Specific leaf area r 0.302 20.245 0.077 0.084n 143 144 254 252p ,0.001 0.002 0.101 0.087p-adj ,0.001 0.007 0.144 0.144Seed mass r 0.055 0.197 0.352 0.274n 131 132 145 148p 0.263 0.011 ,0.001 ,0.001p-adj 0.263 0.028 ,0.001 ,0.001Wood density r 0.077 0.286 0.167 20.202n 143 145 232 227p 0.178 ,0.001 0.005 0.001p-adj 0.198 ,0.001 0.010 0.003Maximum height r 0.130 0.130 0.011 0.008n 143 142 151 150p 0.061 0.061 0.449 0.463p-adj 0.076 0.076 0.463 0.463Significant correlations are in boldface type (p-value,0.05 after the FalseDiscovery Rate adjustment).doi:10.1371/journal.pone.0034767.t003values could be found in Figure S3. In the BCI plot, the soilfertility axis PC1 was negatively correlated with H max (r = 20.283;p,0.001) and positively correlated with SLA (r = 0.258; p,0.001)and SM (r = 0.113; p,0.001).The soil acidity axis PC2 axis waspositively correlated with LA (r = 0.242; p,0.001) and H max(r = 0.057, p = 0.037), but negatively correlated with WD(r = 20.206; p,0.001). The spatial distribution of all traits andsoil PC1 and PC2 values are shown in Figure S4. As with thespecies-level analyses, all analyses were conducted on individualsoil nutrients (see Table S5 & Table S6). At the quadrat-level alltraits were correlated with at least one individual soil variable ineach forest plot.Torus Translation SimulationsAs there is substantial spatial auto-correlation in speciesdistributions and soil nutrient levels, we re-analyzed all of thequadrat-level trait-soil relationships using a torus translationapproach. In Table 5, we provide the rank of observed Pearsonr-values in the distribution of the randomized r-values for the 10predictions for both BCI and GTS. The rank value could be usedto calculate the significance of the observed correlations. Inparticular, low ranks or p-values indicated stronger than expectednegative correlation and high ranks or p-values indicated astronger than expected positive correlation. In the GTS plot, theobserved significant Pearson correlation r-values between LA andSM and the soil fertility axis PC1 were still significant in the torussimulation (see Table 5; p.0.975 and p = 1.000). The observedrelationship between the soil acidity axis PC2 and LA and WDwere also significant (see Table 5; p = 1.000 and p,0.025). In theBCI plot, none of the observed r-values were significant afterimplementing the torus translation simulations and the falsediscovery rate (FDR) correction to p-values (Table 5).DiscussionThe distribution of plant species and communities along broadscaleenvironmental gradients is expected to be determined by thesorting of species along these gradients on the basis of theirfunction (e.g. [4,8–12,32,33]). If a similar sorting of species byfunction occurs on local scales, then this may explain thesubstantial level of interspecific variation within local sites [13].An expected mechanism underlying these trends is that specieswith ‘fast’ leaf, seed and wood economies that have faster resourceacquisition and demographic rates should prefer resource richends of the gradient and species with ‘slow’ economies that haveslower resource acquisition and demographic rates should preferthe resource poor ends of the gradient. The present analysis testedthese mechanistic predictions in two forest plots. While thedistribution of some plant traits showed significant relationshipswith local soil gradients, the majority of the expected relationshipswere not supported. In the following we discuss the results in detailfor each forest plot.Relationships between Functional Traits and SoilResource Axes in the Subtropical Gutianshan (GTS) forestplotOur species-level correlation analyses of the Gutianshan (GTS)forest plot in China supported three of our ten predictions whenconsidering the results of both the species-level analyses and thephylogenetically independent contrasts (PICs) (Table 6). Specifically,leaf area was positively correlated with soil fertility, specificleaf area was negatively correlated with soil acidity (pH) and wooddensity was positively correlated with soil acidity (pH). Leaf areaand specific leaf area (SLA) are known to be correlated with highPLoS ONE | www.plosone.org 5 April 2012 | Volume 7 | Issue 4 | e34767169

Traits, Soil Fertility and DistributionFigure 2. Maps of the quadrat trait and soil fertility patterns. a) The observed leaf area pattern for the GTS plot. b) The soil PC1 values patternfor the GTS plot. The color scale on the right of each map indicates the trait and soil PC1 values. The lines are elevation contour lines at 10-m intervals.See Figure S3 for the complete maps of other traits and the soil PC2 values for GTS plot and Figure S4 for maps of all traits and the soil PC1 and PC2values for BCI plot.doi:10.1371/journal.pone.0034767.g002rates of resource acquisition and growth [11,17,34,35]. Forexample, species with high SLA values have low structuralinvestment and relatively high photosynthetic and respirationrates, whereas species with low SLA values tend to invest more onleaf structures and have relatively low photosynthetic andrespiration rates [2,18,36]. It was therefore expected that plantsTable 4. Pearson correlation coefficients between fivefunctional traits and the calculated scores of the twosignificant principal components of soil fertility and acidity forboth GTS and BCI at the quadrat-level (Leaf area and Seedmass of BCI plot are log 10 transformed).Table 5. Torus translation simulation of the Pearsoncorrelation between traits and the calculated scores of thetwo significant principal components of soil fertility andacidity shifting at 20 m-scale at the quadrat-level for both GTSand BCI.GTSBCIGTSBCIPC1 PC2 PC1 PC2Leaf area r 20.394 20.458 0.035 0.242n 598 598 1248 1248p ,.001 ,.001 0.108 ,.001p-adj ,.001 ,.001 0.135 ,.001Specific leaf area r 0.244 20.025 0.258 20.001n 598 598 1248 1248p ,.001 0.271 ,.001 0.486p-adj ,.001 0.271 ,.001 0.486Seed mass r 20.450 20.130 0.113 0.039n 598 598 1248 1248p ,.001 0.001 ,.001 0.084p-adj ,.001 0.001 ,.001 0.120Wood density r 20.179 0.325 0.017 20.206n 598 598 1248 1248p ,.001 ,.001 0.274 ,.001p-adj ,.001 ,.001 0.304 ,.001Maximum height r 0.186 20.115 20.283 0.057n 598 598 1248 1248p ,.001 0.002 ,.001 0.022p-adj ,.001 0.003 ,.001 0.037Significant correlations are in boldface type (p-value,0.05 after the FalseDiscovery Rate adjustment).doi:10.1371/journal.pone.0034767.t004PC1 PC2 PC1 PC2Leaf area r 599 600 399 163n 600 600 1250 1250p 0.998 1.000 0.319 0.130p-adj 0.993 1.000 0.495 0.325Specific leaf area r 31 282 49 485n 600 600 1250 1250p 0.052 0.47 0.039 0.388p-adj 0.081 0.47 0.195 0.495Seed mass r 600 536 362 509n 600 600 1250 1250p 1.000 0.893 0.289 0.407p-adj 1.000 0.881 0.495 0.495Wood density r 576 2 561 1176n 600 600 1250 1250p 0.960 0.003 0.449 0.941p-adj 0.920 0.008 0.495 0.803Maximum height r 38 561 1218 618n 600 600 1250 1250p 0.063 0.935 0.975 0.495p-adj 0.081 0.919 0.805 0.495Significant correlations are in boldface type (p-value,0.025 or p-value.0.975after the False Discover y Rate adjustment).doi:10.1371/journal.pone.0034767.t005PLoS ONE | www.plosone.org 6 April 2012 | Volume 7 | Issue 4 | e34767170

Traits, Soil Fertility and Distributionwith larger SLA were found in high nutrient soils, whereas thereverse was expected to occur in low nutrient-supply soil. Thetrade-off between wood density and species growth and mortalityrates has been shown in previous studies [37–40]. Speices inshaded or arid sites gernerally have smaller vessels and thickerfiber walls, thereby increasing their wood density. In the GTS plot,light wooded species tended to be found on fertile soils suggestingthat high resource environments favoured species that allocate lessbiomass per unit volume and that have higher growth andmortality rates.Relationships between Functional Traits and SoilResource Axes in the Tropical Barro Colorado Island (BCI)forest plotThe species-level correlation analyses of the Barro ColoradoIsland (BCI) forest plot in Panama found no support for ourpredictions regarding species traits and soil fertility or acidity inspecies-level analyses (Table 6), while the PIC analyses providedsupport for all but three of our predictions. Thus we could onlysupport the predictions that seed mass would be positively relatedto soil acidity and negatively related to soil fertility and wooddensity would be negatively related to soil fertility (Table 6). In thelow resource environments, large seeds could provide morereserves for individuals early in their life cycle. Small seeds, onthe other hand, have the potential advantage of greater dispersalability and rapid growth in high resource environments [41–44].Quadrat-Level Trait-Soil Relationships in the Two ForestPlotsA secondary goal of the present study was to determine whetherour predictions regarding species-level trait relationships with soilfertility and acidity gradients would scale-up to the quadrat-level.Although we found many significant relationships, the majority ofthese were non-significant once we accounted for spatialautocorrelation (Table 6). For example, in the GTS forest plot,the relationships between most traits and soil fertility weresignificant, but after controlling for spatial autocorrelation via atorus translation analysis, only four were still significant and onlythree of our ten predictions were still supported. This finding wasconsistent with our findings at species-level. From this, we caninfer that for these few trait-soil relationships, it may be that theobserved local relationships scale-up to generate the regional scalerelationships reported elsewhere.The quadrat-level results from BCI yielded no support for ourpredictions once we accounted for spatial autocorrelation. Thenon-significant co-variation between traits and soil at the quadratlevelmay be based on very weak relationships at species-level inthe BCI forest plot. A possible explanation for the BCI results isthat the location for this forest plot was chosen to be ashomogeneous as possible and a large proportion of trees thereoccur in shaded environments [45]. Therefore, the most importantfactors influencing the sorting of plant traits may be light levels orother factors rather than soil nutrients. Thus, this level ofhomogeneity also highlights one weakness of our study. Specifically,while the forest plots being analyzed used standardize treeinventory protocols, they were not set up to standardize the level ofenvironmental heterogeneity. Future comparative research intothe relationship between traits and soil nutrient gradients shouldtherefore seek to standardize the level of soil nutrient heterogeneityat the plot-level.Ultimately, the relationship between traits and soil fertilitymight be moderated by additional environmental parameters notpresently analyzed and likely by the difference in the breadth ofvarious resource axes. This may explain the lack of strong trait –environment relationships in this study. Besides, a potential reasonfor the different results for the two plots is the difference in terms ofseasonality and associated harshness of the abiotic environment.In summary, plant functional trait research has shown that planttraits vary predictably along broad-scale climatic and soilgradients. The present research predicted that this variation mightbe explained partly by local-scale soil fertility and acidity gradients.Although we found leaf area and wood density had a consistentand predictable relationship with soil fertility both at species andquadrat-level for GTS, we failed to find support for most predictedrelationships between plant traits and soil fertility and acidity axes.Table 6. A summary table of whether the predicted correlation results for both the GTS and BCI forest plots were supported in thisstudy.Species-level(Table 2) Species PICs Quadrat-level Torus Translation Simulation(Table 2) (Table 3) (Table 4) (Table 5)Predicted Correlation GTS BCI GTS BCI GTS BCI GTS BCINegative LA & PC1 + NS + NS + NS + NSNegative LA & PC2 + NS NS 2 + 2 + NSNegative SLA & PC1 NS NS 2 NS 2 2 NS NSNegative SLA & PC2 + NS + NS NS NS NS NSPositive SM & PC1 NS NS NS + 2 + 2 NSPositive SM & PC2 NS NS + + 2 NS NS NSPositive WD & PC1 NS NS NS + 2 NS NS NSPositive WD & PC2 + NS + 2 + 2 + NSNegative H max & PC1 NS NS NS NS 2 + NS NSNegative H max & PC2 NS NS NS NS + 2 NS NSThe correlations were calculated between the soil PC axes and the species-level or quadrat-level trait value. The table depicts whether the prediction was significant andsupported the prediction (+), was significant and did not support the prediction (2) or was non-significant (NS). Phylogenetically independent contrasts (PICs) and torustranslation simulations were utilized to correct for evolutionary non-independence in the species-level analyses and spatial auto-correlation in the quadrat-levelanalyses respectively. LA: leaf area; SLA: specific leaf area; SM: seed mass; WD: wood density; H max : maximum height.doi:10.1371/journal.pone.0034767.t006PLoS ONE | www.plosone.org 7 April 2012 | Volume 7 | Issue 4 | e34767171

Traits, Soil Fertility and DistributionIn particular, the limited evidence for species-level associationsbetween traits and soil fertility and acidity failed to scale up to thequadrat-level for both the GTS and BCI plots. The general lack ofsupport for the predictions at the BCI forest plot may be due tolimited heterogeneity in soil nutrients in this particular forest, butthe same cannot be said for the GTS plot as it is quiteheterogeneous with rugged terrain. In both plots it is clear thatsoil nutrients are not the only determinant of plant traitdistributions and alternative resource axes, such as light, will haveto be considered in future work. Ultimately, while some of ourpredictions regarding local-scale trait distributions, soil fertility andsoil acidity were supported other factors likely play a larger role indetermining the large interspecific variation in trait values in theforests studied.Supporting InformationFigure S1 The phylogenetic tree constructed using DNAbarcodes of the species in the GTS plot (See details ontree construction in the text).(TIF)Figure S2 The phylogenetic tree constructed using DNAbarcodes of the species in the BCI plot (See details ontree construction in the text).(TIF)Figure S3 Maps of the quadrat trait and soil fertilitypatterns for the GTS plot. Map a), b), c) and d) are theobserved SLA, seed mass, wood density and maximum heightpatterns; and map e) is the soil PC2 values for the GTS plot. Thecolor scale on the right of each map indicates the trait and soil PC2values. The lines are elevation contour lines at 10-m intervals.(TIF)Figure S4 Maps of the quadrat trait and soil fertilitypatterns for the BCI plot. Map a), b), c), d) and e) are theobserved leaf area, SLA, seed mass, wood density and maximumheight pattern for the BCI plot; and maps f) and g) are the soil PC1and PC2 values for the BCI plot. The color scale on the right ofeach map indicates the trait and soil PC1 and PC2 values. Thelines are elevation contour lines at 5-m intervals.(TIF)Table S1 Pearson correlation coefficients between fivefunctional traits and 13 soil nutrients for the GTS plot atReferences1. Schimper AFW (1989) Planzengeographie auf Physiolo- gischer Grundlage. GFischer, Bonn, Germany.2. Wright IJ, Westoby M (2002) Leaves at low versus high rainfall: coordination ofstructure, lifespan and physiology. New Phytologist 155: 403–416.3. Dolph GE, Dilcher DL (1980) Variation in Leaf Size with Respect to Climate inCosta-Rica. Biotropica 12: 91–99.4. Reich PB, Walters MB, Ellsworth DS (1997) From tropics to tundra: Globalconvergence in plant functioning. Proceedings of the National Academy ofSciences of the United States of America 94: 13730–13734.5. Wiemann MC, Williamson GB (2002) Geographic variation in wood specificgravity: Effects of latitude, temperature, and precipitation. Wood and FiberScience 34: 96–107.6. Reich PB, Oleksyn J (2004) Global patterns of plant leaf N and P in relation totemperature and latitude. Proceedings of the National Academy of Sciences ofthe United States of America 101: 11001–11006.7. Moles AT, Ackerly DD, Webb CO, Tweddle JC, Dickie JB, et al. (2005) Factorsthat shape seed mass evolution. Proceedings of the National Academy ofSciences of the United States of America 102: 10540–10544.8. Swenson NG, Enquist BJ (2007) Ecological and evolutionary determinants of akey plant functional trait: Wood density and its community-wide variation acrosslatitude and elevation. American Journal of Botany 94: 451–459.9. Chave J, Coomes D, Jansen S, Lewis SL, Swenson NG, et al. (2009) Towards aworldwide wood economics spectrum. Ecology Letters 12: 351–366.the species-level (leaf area and seed mass are log 10 -transformed).(DOCX)Table S2 Pearson correlation coefficients between fivefunctional traits and 13 soil nutrients for the BCI plot atthe species-level (leaf area and seed mass are log 10 -transformed).(DOCX)Table S3 Phylogenetically independent contrasts (PICs)between five functional traits and 13 soil nutrients forthe GTS plot at the species-level.(DOCX)Table S4 Phylogenetically independent contrasts (PICs)between five functional traits and 13 soil nutrients forthe BCI plot at the species-level.(DOCX)Table S5 Pearson correlation coefficients between fivefunctional traits and 13 soil nutrients for the GTS plot atthe quadrat-level.(DOCX)Table S6 Pearson correlation coefficients between fivefunctional traits and 13 soil nutrients for BCI plot at thequadrat-level (leaf area and seed mass are log 10 transformed).(DOCX)AcknowledgmentsWe thank Fang Teng, Chen Shengwen, Lai Zhenxi and Jiang Zhiyong forspecies identification and data collection at Gutianshan. The AdministrationBureau of the Gutianshan National Nature Reserve and theSmithsonian Tropical Research Institute provided local support. TheBiodiversity Committee of the Chinese Academy of Sciences supported theGTS plot. The National Science Foundation, the MacArthur Foundationand the Smithsonian Tropical Research Institute supported the BCI plot.Author ContributionsConceived and designed the experiments: SJW KM. Performed theexperiments: XL SJW LZ KS YD XM HR. Analyzed the data: XL NGSSJW. Contributed reagents/materials/analysis tools: JZ. Wrote the paper:XL NGS SJW KM.10. Moles AT, Warton DI, Warman L, Swenson NG, Laffan SW, et al. (2009)Global patterns in plant height. Journal of Ecology 97: 923–932.11. Ordonez JC, van Bodegom PM, Witte JPM, Wright IJ, Reich PB, et al. (2009) Aglobal study of relationships between leaf traits, climate and soil measures ofnutrient fertility. Global Ecology and Biogeography 18: 137–149.12. Quesada CA, Lloyd J, Schwarz M, Baker TR, Phillips OL, et al. (2009) Regionaland large-scale patterns in Amazon forest structure and function are mediatedby variations in soil physical and chemical properties. Biogeosciences discussions6: 3993–4057.13. Westoby M, Wright IJ (2006) Land-plant ecology on the basis of functionaltraits. Trends in Ecology & Evolution 21: 261–268.14. Condit R (1998) Tropical forest census plots. Springer-Verlag Berlin, Germany.15. Lai JS, Mi XC, Ren HB, Ma KP (2009) Species-habitat associations change in asubtropical forest of China. Journal of Vegetation Science 20: 415–423.16. Legendre P, Mi XC, Ren HB, Ma KP, Yu MJ, et al. (2009) Partitioning betadiversity in a subtropical broad-leaved forest of China. Ecology 90: 663–674.17. Wright SJ, Kitajima K, Kraft NJB, Reich PB, Wright IJ, et al. (2010) Functionaltraits and the growth-mortality trade-off in tropical trees. Ecology 91: 3664–3674.18. Cornelissen JHC, Lavorel S, Garnier E, Diaz S, Buchmann N, et al. (2003) Ahandbook of protocols for standardised and easy measurement of plantfunctional traits worldwide. Australian Journal of Botany 51: 335–380.19. Wu ZY, Raven PH, eds. Flora of China. Beijing: Science Press & St Louis:Missouri Botanical Garden Press.PLoS ONE | www.plosone.org 8 April 2012 | Volume 7 | Issue 4 | e34767172

Traits, Soil Fertility and Distribution20. Editorial Committee for Flora Reipublicae Popularis Sinicae (1961–2004) FloraReipublicae Popularis Sinicae. Beijing: Science Press.21. Kress WJ, Erickson DL, Jones FA, Swenson NG, Perez R, et al. (2009) PlantDNA barcodes and a community phylogeny of a tropical forest dynamics plot inPanama. Proceedings of the National Academy of Sciences of the United Statesof America 106: 18621–18626.22. Stamatakis A (2006) RAxML-VI-HPC: Maximum likelihood-based phylogeneticanalyses with thousands of taxa and mixed models. Bioinformatics 22:2688–2690.23. John R, Dalling JW, Harms KE, Yavitt JB, Stallard RF, et al. (2007) Soilnutrients influence spatial distributions of tropical tree species. Proceedings ofthe National Academy of Sciences of the United States of America 104:864–869.24. Zhang LW, Mi XC, Shao HB, Ma KP (2011) Strong plant-soil associations in aheterogeneous subtropical broad-leaved forest. Plant Soilin press.25. Felsenstein J (1985) Phylogenies and the Comparative Method. AmericanNaturalist 125: 1–15.26. Harvey PH, Pagel M (1991) The comparative method in evolutionary biology.Oxford: Oxford University Press.27. Bolmgren K, Cowan PD (2008) Time - size tradeoffs: a phylogeneticcomparative study of flowering time, plant height and seed mass in a northtemperateflora. Oikos 117: 424–429.28. Garland T, Harvey PH, Ives AR (1992) Procedures for the Analysis ofComparative Data Using Phylogenetically Independent Contrasts. SystematicBiology 41: 18–32.29. Harms KE, Condit R, Hubbell SP, Foster RB (2001) Habitat associations oftrees and shrubs in a 50-ha neotropical forest plot. Journal of Ecology 89:947–959.30. Benjamini Y, Hochberg Y (1995) Controlling the False Discovery Rate - aPractical and Powerful Approach to Multiple Testing. Journal of the RoyalStatistical Society Series B-Methodological 57: 289–300.31. Garcia LV (2004) Escaping the Bonferroni iron claw in ecological studies. Oikos105: 657–663.32. Wright IJ, Reich PB, Westoby M, Ackerly DD, Baruch Z, et al. (2004) Theworldwide leaf economics spectrum. Nature 428: 821–827.33. Swenson NG, Weiser MD (2010) Plant geography upon the basis of functionaltraits: an example from eastern North American trees. Ecology 91: 2234–2241.34. Givnish TJ (1978) On the adaptive significance of compound leaves, withparticular reference to tropical trees. Tropical trees as living systems. In: PB.Tomlinson, MH. Zimmermann, eds. Cambridge: Cambridge University Press,Cambridge. pp 351–380.35. Poorter L, Bongers L, Bongers F (2006) Architecture of 54 moist-forest treespecies: Traits, trade-offs, and functional groups. Ecology 87: 1289–1301.36. Sterck FJ, Poorter L, Schieving F (2006) Leaf traits determine the growthsurvivaltrade-off across rain forest tree species. American Naturalist 167:758–765.37. Muller-Landau HC (2004) Interspecific and inter-site variation in wood specificgravity of tropical trees. Biotropica 36: 20–32.38. Nascimento HEM, Laurance WF, Condit R, Laurance SG, D’Angelo S, et al.(2005) Demographic and life-history correlates for Amazonian trees. Journal ofVegetation Science 16: 625–634.39. King DA, Davies SJ, Tan S, Noor NSM (2006) The role of wood density andstem support costs in the growth and mortality of tropical trees. Journal ofEcology 94: 670–680.40. van Gelder HA, Poorter L, Sterck FJ (2006) Wood mechanics, allometry, andlife-history variation in a tropical rain forest tree community. New Phytologist171: 367–378.41. Westoby M, Leishman M, Lord J (1996) Comparative ecology of seed size anddispersal. Philosophical Transactions of the Royal Society of London Series B-Biological Sciences 351: 1309–1317.42. Leishman MR, Masters GJ, Clarke IP, Brown VK (2000) Seed bank dynamics:the role of fungal pathogens and climate change. Functional Ecology 14:293–299.43. Azcarate FM, Sanchez AM, Arqueros L, Peco B (2002) Abundance and habitatsegregation in Mediterranean grassland species: the importance of seed weight.Journal of Vegetation Science 13: 159–166.44. Muller-Landau HC (2010) The tolerance-fecundity trade-off and the maintenanceof diversity in seed size. Proceedings of the National Academy of Sciencesof the United States of America 107: 4242–4247.45. Hubbell SP, Foster RB (1983) Diversity of canopy trees in a neotropical forestand implications for conservation. In: Tropical Rain Forest: Ecology andManagement SL. Sutton, TC. Whitmore, AC. Chadwick, eds. Oxford:Blackwell Scientific Publications. pp 25–41.PLoS ONE | www.plosone.org 9 April 2012 | Volume 7 | Issue 4 | e34767173

Author's personal copyTrees (2012) 26:283–290DOI 10.1007/s00468-011-0590-6ORIGINAL PAPERAge and radial growth pattern of four tree species in a subtropicalforest of ChinaPei Xing • Qi-Bin Zhang • Patrick J. BakerReceived: 13 December 2010 / Revised: 21 March 2011 / Accepted: 1 July 2011 / Published online: 16 July 2011Ó Springer-Verlag 2011Abstract Subtropical forests are usually composed ofmany tree species. Knowledge of the age and radial growthvariation of the dominant tree species is useful for understandingforest dynamics and community structure andfunction. The aims of this study are to explore whetherthere are identifiable annual growth rings in the main treespecies and to examine the growth characteristics withinand among the species in Mount Gutian subtropical forestof China. The results showed that four out of eight treespecies from which samples were collected had visible andcross-datable rings. There were no stable relationshipsbetween the age and diameter for these subtropical trees.Significant differences existed in radial growth rate withinand among the four species, suggesting a high spatialheterogeneity in the mixed-species subtropical forest. Thecommon pattern in age distribution of multiple speciessuggests a stand-wide disturbance occurring around the1960s. It is interesting to note that the growth rate at thesame age intervals was different for trees younger than40 years of age and older than 40 years of age, suggestinga change in climate or forest structure in the two timeCommunicated by A. Braeuning.P. Xing Q.-B. Zhang (&)State Key Laboratory of Vegetation and Environmental Change,Institute of Botany, Chinese Academy of Sciences,20 Nanxincun, Xiangshan, Beijing 100093, Chinae-mail: qbzhang@ibcas.ac.cnP. XingGraduate University of the Chinese Academy of Sciences,19A Yuquanlu, Beijing 100049, ChinaP. J. BakerSchool of Biological Sciences, Monash University,Bld 18, Melbourne, VIC 3800, Australiaperiods. The results obtained from this study help understandthe growth dynamics in other subtropical forestshaving these tree species.Keywords Subtropical forests Tree rings Growth dynamics Age intervalIntroductionTree species coexisting in a forest have different ecophysiologicalcharacteristics and usually show differentgrowth patterns (Stewart 1986). Knowledge about the ageand radial growth pattern in dominant tree species is ofimportance for understanding forest dynamics and structure.In Mount Gutian of eastern China, a 24-ha permanentplot of subtropical broad-leaved forest was established tostudy the forest dynamics in detail (Ma 2008). Althoughdendrometers were applied in the permanent plot to monitorthe growth rate of dominant tree species, informationabout the age and long-term growth dynamics is sparse.The lack of data inhibited reliable assessment of the variationin tree growth along the lifespan and among species.Such assessment is essential for forest sustainablemanagement.Tree-ring analysis is an effective tool to identify the ageof trees for evaluating forest dynamics and reconstructingthe stand development patterns (Baker et al. 2005; Bergeron2000; Dang et al. 2010). Tree-ring analysis has been used toobtain insight into canopy disturbance, lifetime growthpatterns and historical growth rates of trees in temperateforests (Landis and Peart 2005; Lorimer and Frelich 1989;Lusk and Smith 1998). Seasonal radial growth and annualring development have been confirmed in some tropical andsubtropical tree species (Schongart et al. 2006; Grau et al.123174

Author's personal copy284 Trees (2012) 26:283–2902003; Fichtler et al. 2003; Worbes 1995; Worbes and Junk1989). To date, most tree-ring research in China focuses onreconstruction of past climate or tree-line dynamics in aridand semiarid areas of western China (Li et al. 2008; Wanget al. 2006; Liu et al. 2004; Zhang et al. 2003), whereasmuch less studies have been carried out on subtropical trees.In forestry, there is a general need to estimate the age oftrees from their diameter. Such estimation is usually basedon empirical relationship between the tree’s age anddiameter. The age of trees is determined by counting thenumber of rings in the stem, if the rings are formedannually and clearly visible. Many researchers and managershave used stem diameter as proxy measures of treeage for convenience. The reliability of the estimationdepends on the stability of the relationship among individualtrees (Burley et al. 2007; Liu and Hong 1999). Oneof the aims of this study is to examine if there are clearlyvisible rings in the main tree species in Mount Gutiansubtropical forest and if there are stable relationshipsbetween the age and diameter. The second aim of this studyis to compare growth rates in different size classes and ageintervals among the main tree species that have clearlyidentifiable rings. Such information is useful for understandingthe ecological characteristics of the dominant treespecies in the subtropical forest and therefore helps toevaluate forest dynamics.Materials and methodsStudy site and samplingThis study was conducted at Mount Gutian National NatureReserve in Zhejiang Province, eastern China (29°10 0 19 00 –29°17 0 41 00 N, 118°03 0 50 00 –118°11 0 12 00 E). The area is characterizedby rugged terrain with elevation at the study siteranging from 450 to 750 m above sea level. A total of 1991vascular plant species, belonging to 244 families and 897genera, have been recorded within the entire Mount Gutian(Chen and Feng 2002). Within the 24-ha permanent plot,there are 159 tree species which belong to 49 families(Legendre et al. 2009). The dominant vegetation type in theregion is subtropical evergreen mixed (broadleaved andconiferous) forest dominated by Schima superba, Castanopsiseyrei, Cyclobalanopsis glauca, Pinus massonianaand Quercus serrata (Zhu et al. 2008; Yu et al. 2001). Theclimate is sub-tropical with distinct temperature and precipitationseasonality. According to the meteorologicaldata at the nearby Tunxi weather station from 1953 to2002, the mean annual number of frost-free days is 250.The mean monthly temperature ranges from 4°C in Januaryto 28°C in July with an annual mean of 16°C and meanannual precipitation of 1723 mm (Fig. 1).Fig. 1 Monthly rainfall (gray bars, left axis) and temperature (blackline, right axis) averaged from 1953 to 2002 for Mt. Gutian, ChinaIncrement core samples were collected at breast heightfrom trees that have no sign of obvious rot and damage inthe stem. The trees selected for sampling were from eightmain species growing in an area adjacent to the 24-hapermanent forest plot. These eight tree species wereP. massoniana, S. superba, Cunninghamia lanceolata,Q. serrata, Daphniphyllum oldhami, C. eyrei, C. fargesiiand C. tibetana. The perimeters at breast height of thesampled trees were measured. At least 26 trees of differentdiameter class from each species were selected for sampling(Table 1).Tree-ring analysisIn laboratory, the increment core samples were mounted,air dried and sanded with sandpapers of progressively finergrit (up to 600 grits) to make the rings visible. Ring widthsof the samples that have clearly visible rings were measuredto a precision of 0.001 mm under a stereomicroscopewith a Velmex incremental measuring device. The ringwidthseries were cross-dated and quality checked usingthe program COFECHA (Holmes 1983) to insure that eachring was assigned to the correct calendar year of its formation.Standard tree-ring chronology of each species wasdeveloped from the cross-dated ring-width series using theprogram ARSTAN (Cook 1985). Negative exponentialcurves or linear regression lines of negative slope or horizontallines were used to remove the age-related growthtrends (Fritts 1976). Tree-ring samples of less than20 years old were not included into chronology development,and the early section of each chronology with lessthan five sample replications was truncated.The age of the trees was determined by counting thenumber of rings from the outermost ring to the pith. Whenthe pith was not obtained in the core samples, we estimatedthe age of the ring closest to the pith according to its shapeof curvature (Brienen and Zuidema 2006). It is worth123175

Author's personal copyTrees (2012) 26:283–290 285Table 1 Number of increment core samples of the eight tree species from the Mt. Gutian subtropical forest of ChinaSpecies Total Diameter classes distribution (cores number)0–10 cm 10–20 cm 20–30 cm 30–40 cm 40–50 cm 50–60 cmPinus massoniana 31 6 11 8 4 2 0Schima superba 30 9 8 7 4 0 2Cunninghamia lanceolata 30 4 10 9 6 0 1Quercus serrata 29 11 14 4 0 0 0Daphniphyllum oldhami 27 6 13 7 1 0 0Castanopsis eyrei 30 7 10 4 6 3 0Castanopsis fargesii 26 7 7 5 4 3 0Castanopsis tibetana 31 7 6 13 5 0 0noting that the age of the trees obtained thereafter is the ageof the stem at the breast height, because the age does notinclude the time that the trees grow from the ground to thesampling height.The cumulative radial growth curves of each tree werecalculated over its lifetime and averaged to obtain meangrowth curves for each species. Multiplying the radialgrowth values by two, we obtained the cumulative stemdiameter of each species which reflect an approximate age–diameter relationship. The size class of different tree specieswas set to be 10 cm in diameter. The variations inradial growth rates were analyzed for each species bycalculating the median, minimum and maximum growthrates for each diameter class. Species were tested foroverall differences in growth rates in each size class usingKruskal–Wallis tests and Dunn tests for a posteriori pairwisecomparison (Kruskal and Wallis 1952). To examinethe growth rate at the same age intervals in young and oldtrees, we compared the growth rates between trees youngerthan 40 years and older than 40 years of age in each10-year age interval. The difference of growth ratesbetween the young (\40 years of age) and old ([40 years)trees was tested by Mann–Whitney test for each ageinterval (Whitney 1997).ResultsTree-ring characteristics of the eight tree speciesExamination of the tree-ring samples of the eight treespecies showed that clearly visible and cross-datable ringswere found in four species, i.e., P. massoniana, S. superba,C. lanceolata and Q. serrata (Fig. 2a–d). Tree rings in therest of the four species were either difficult to distinguishor to cross-date (Fig. 2e–h). For D. oldhami and C. fargesiitrees, the latewood is too narrow and light to be certain if itforms a true ring or is simply a false ring, and this difficultyincreases especially in the portion close to the bark.For C. eyrei and C. tibetana trees, the latewood is too faintto insure if it is the boundary with the earlywood of thefollowing year. Tree-ring widths in the four species thathave clearly visible rings were cross-dated for each speciesand the mean inter-serial correlation coefficients rangedfrom 0.336 to 0.382, indicating a high quality of crossdatingamong the tree-ring series. Tree rings of thesespecies were studied previously in other regions (Songet al. 2011; Shao et al. 2009; Kuang et al. 2008; Fujihara1996). There have not been any publications on dendrochronologicalstudies of the four species that do not showclearly visible rings.Tree-ring chronologies for the four species under studyare shown in Fig. 3. The correlation coefficients amongthese chronologies were low, ranging from -0.337 to0.276 (for the period 1915–2007), suggesting that thehabitats in this mixed subtropical forest are heterogeneous.As shown in Table 2, Cunninghamia trees showed thehighest mean ring width, reflecting its fast radial growthrelative to other species. Schima exhibited the highestmean sensitivity, suggesting that its ring widths had highinter-annual variability and was sensitive to yearly environmentalchanges. The first-order autocorrelations rangedfrom 0.530 to 0.769 for the four chronologies, indicatingthat the chronologies contained considerable low-frequencyvariance related with growth condition and treephysiology (Fritts 1976). In common period analyses(1951–2000), the expressed population signal ranged from0.741 to 0.872 for the four chronologies, indicating thatthese four species were useful for dendroecologicalanalyses.Radial growth dynamics within speciesThe cumulative radial growth in relation to tree’s ageshows that there is great variation in age for a certaindiameter (Fig. 4). For instance, the range of age for a20-cm DBH tree was 60–112 years for Pinus, 35–95 yearsfor Schima, 45–110 years for Cunninghamia and123176

Author's personal copy286 Trees (2012) 26:283–290Fig. 2 Photographs of the tree rings of the eight tree species in Mt. Gutian subtropical forest of China. a Pinus massoniana, b Schima superba,c Cunninghamia lanceolata, d Quercus serrata, e Daphniphyllum oldhami, f Castanopsis eyrei, g Castanopsis fargesii, h Castanopsis tibetanagrowth) over the lifetime of the tree. Schima had a complexgrowth pattern. The young trees showed high, lineargrowth rates, whereas the older trees showed fast earlygrowth, followed by a distinct reduction in growth rate andthen a later increase in growth. The dashed lines in Fig. 5indicate constant diameter growth of 1 mm per year, whichis clearly exceeded by most individuals of Pinus andCunninghamia.Radial growth dynamics among speciesFig. 3 Ring-width chronologies of the four species in Mt. Gutiansubtropical forest of China40–105 years for Quercus. This wide range of age for agiven tree size derives from the high variability of individualgrowth trajectories within the species (Fig. 5). MostPinus and Cunninghamia showed fast early growth withdecreasing growth rates in larger trees. In contrast, Quercusgrowth trajectories were relatively linear (i.e., constantSpecies differed strongly in mean age–size relations(Fig. 6). Trees around 5 cm in diameter had comparablemean ages for Cunninghamia, Schima and Pinus trees, butthe ages differed strongly among species at larger diameter.For instance, mean ages at 20 cm in diameter varied from55 years in Cunninghamia trees to more than 85 years forSchima and Pinus trees. The average age–size relations ofSchima and Pinus trees were very similar. Quercus treestended to have slow and constant growth over their entirelife span and showed the highest average ages at anydiameter. Cunninghamia owned the highest average radialgrowth rate among the four species.The four species under study showed different patternsin growth rates (Fig. 7). Schima had the highest mediangrowth rates of the four species, followed by123177

Author's personal copyTrees (2012) 26:283–290 287Table 2 Statistics for standardtree-ring chronologies of thefour species under study inMt. Gutian subtropical forestof ChinaSpeciesPinusmassonianaSchimasuperbaCunninghamialanceolataQuercusserrataMean ring width (mm) 1.02 1.32 1.6 0.88Mean inter-serial correlation 0.382 0.354 0.336 0.364Mean sensitivity 0.168 0.181 0.142 0.168First-order autocorrelation 0.571 0.769 0.530 0.664Expressed population signal 0.795 0.872 0.741 0.814Fig. 4 The relationship between the diameter at breast height (DBH)and tree age of the four species in Mt. Gutian subtropical forest ofChinaFig. 5 Cumulative radial growth in relation to tree age for the fourspecies under study in Mt. Gutian subtropical forest of China; eachline represents one individual tree. The dashed lines indicate constantdiameter growth of 1 mm per yearCunninghamia. InSchima and Cunninghamia, the highestmedian growth rates were found in the second and first sizeclasses, respectively. Concerning Pinus and Quercus, theontogenetic growth pattern was similar, i.e., the growth ratehardly changed with size. The minimum growth rates ofthe four species were almost about the same values(\0.3 mm/year), but the maximum varied much with differentspecies or size classes. Cunninghamia had an especiallylarge gap between the maximum and minimumgrowth rates.Radial growth in the same age intervals in youngand old treesBesides the age, there are other factors that affect the radialgrowth of the four species under study (Fig. 8). ForQuercus, the growth rate of young trees (\40 years of age)was higher than that of old trees ([40 years of age) in thesame age intervals 0–10, 10–20, 20–30 and 30–40 years ofage. This pattern was opposite in Pinus. The pattern ofFig. 6 Mean cumulative stem diameter growth curves for the fourspecies under study in Mt. Gutian subtropical forest of Chinagrowth in Schima was similar to Quercus except the first10 years. For Cunninghamia, the young trees grew morerapidly than the old ones at the initial lifetime, and much123178

Author's personal copy288 Trees (2012) 26:283–290Discussion and conclusionsFig. 7 Diameter growth rates (minimum, median and maximum) ofthe four species under study in Mt. Gutian subtropical forest of China.Median growth rates differed among species in each of the sizecategories (Kruskal–Wallis tests, p \ 0.001). Different superscriptletters under the bars indicate significant (p \ 0.05) differencesamong species in that size class using the Dunn tests for a posterioricomparison between groupsFig. 8 Radial growth rate (medium, minimum and maximum) fordifferent age intervals in trees younger than 40 years of age andgreater than 40 years of age. Triangles in the age class axis indicatethat the medians of the radial growth rate in young and old trees aresignificantly different (p \ 0.05) based on the Mann–Whitney testslower than the old trees in the age intervals 10–20 and20–30 years. The Mann–Whitney test showed that therewas significant difference (p \ 0.05) between the mediansof the same age interval in young and old trees except thefourth age interval of Pinus and the first age interval ofCunninghamia.The results of our study showed that not all tree species inthe Mount Gutian subtropical forest have clearly visiblerings. For the four species that have distinct annual growthrings, the age–diameter relationship has a great variationamong individuals, suggesting that it is unstable and shouldbe used cautiously. Additional information about thegrowth condition is needed to obtain a reliable estimationof the tree’s age (Baker 2003). This is similar to the resultsobtained in other studies (Baker and Wilson 2003; Burleyet al. 2007; Worbes et al. 2003).Although our samples were collected from trees of allsize classes, the age distribution showed that P. massoniana,S. superba and Q. serrata mainly have two agecohorts, one older than 80 years of age and the otheryounger than 40 years of age (Fig. 4). This phenomenonsuggests that a major disturbance occurred about 40 yearsago, i.e., in the 1960s. It was documented that, in order togenerate fuel for the local steel industry, many old growthforests in the region experienced various degrees of loggingaround 1960, before Mount Gutian became a reservein 1975 (Song et al. 2011; Zhang et al. 2002). The abovethree are all pioneer tree species, which play an importantrole in the regeneration process after canopy opening. Thisdisturbance was also evidenced by the growth release inliving trees in the mid-1960s (Fig. 3).The forest might also experience other kinds of disturbancessuch as typhoon or forest fire. It was reported thatthe congregation of some P. massoniana individuals inapproximately 1-ha area of the northern 24-ha permanentplot was a result of a forest fire in the 1920s (Zhu et al.2008). Information about lifetime growth patterns from alarger number of trees of different species would help inthe evaluation of historical disturbances and the acquaintanceof community succession (Gutierrez et al. 2004;Splechtna et al. 2005).Where species richness is particularly high, as in tropicaland subtropical mountain ecosystems, diverse treespecies with different phenological rhythmicity and differentmicroclimates result in heterogeneities of treegrowth for individuals (Hu and Yu 2008; Bräuning et al.2008). Radial growth patterns of individual trees examinedin our study varied within and among species. The differencesin radial growth of trees suggest that the habitat ofthe subtropical forest under study is highly heterogeneousin space and such heterogeneity may contribute to themulti-species coexistence of the forest.It is interesting to note that the growth rate at 10-year ageintervals in trees younger than 40 years of age is differentfrom that at the same age intervals in trees older than40 years of age. In other words, the age-related growthtrend in the early 40 years is different in young trees and old123179

Author's personal copyTrees (2012) 26:283–290 289trees. This phenomenon suggests that environmental factors(such as climate) and/or forest structure (such as forestdensity and canopy structure) might have changed in therecent 40 years. Quercus and Schima trees showed that theradial growth rates of young trees were faster than that ofthe initial 40 years of old trees, indicating that the growthconditions in the recent 40 years were more favorable to thegrowth of young trees than the earlier time. Conversely, thelow growth rate in young Pinus and Cunninghamia treesrelative to the same age intervals (except the 1–10 year ageinterval in Cunninghamia) in old trees suggested a poorercondition for the growth of young Pinus and Cunninghamiatrees in the recent 40 years. The changes in growth rates ofyoung trees would alter their progress in basal area coverageand possibly their course reaching the canopy as well,thus eventually affecting the forest structure (Rozendaalet al. 2010; Landis and Peart 2005).In conclusion, our study demonstrated significant differencesin radial growth rate not only within and amongtree species, but also among different size classes and ageintervals. High variability of radial growth patterns highlightsthe importance of tree-ring data in studies of habitatheterogeneity and points to uncertainties in diameter-basedinference of forest age. Common growth patterns in multiplespecies could provide information about stand-widedisturbances and such knowledge is essential for forestmanagement. This study further demonstrated that the fourtree species growing in this subtropical forest are suitablefor dendrochronological studies.Acknowledgments This study was supported by the ChineseAcademy of Sciences Projects No. KSCX-YW-Z-1022 and KZCX2-YW-430, the National Natural Sciences Foundation of China ProjectNo. 40631002, and a grant from the Office for InternationalEngagement, Monash University, Australia.ReferencesBaker PJ (2003) Tree age estimation for the tropics: a test from thesouthern Appalachians. Ecol Appl 13:1718–1732. doi:10.1890/02-5025Baker PJ, Wilson JS (2003) Coexistence of tropical tree species.Nature 422:581–582. doi:10.1038/422581aBaker PJ, Bunyavejchewin S, Oliver CD, Ashton PS (2005)Disturbance history and historical stand dynamics of a seasonaltropical forest in Western Thailand. Ecol Monogr 75:317–343.doi:10.1890/04-0488Bergeron Y (2000) Species and stand dynamics in the mixed woods ofQuebec’s southern boreal forest. Ecology 81:1500–1516. doi:10.1890/0012-9658(2000)081[1500:SASDIT]2.0.CO;2Bräuning A, Homeier J, Cueva E, Beck E, Günter S (2008) Growthdynamics of trees in tropical mountain ecosystems. In: Gradientsin a tropical mountain ecosystem of Ecuador. Springer, Berlin,pp 291–302Brienen RJW, Zuidema PA (2006) Lifetime growth patterns and agesof Bolivian rain forest trees obtained by tree ring analysis. J Ecol94:481–493. doi:10.1111/j.1365-2745.2005.01080.xBurley AL, Phillips S, Ooi MKJ (2007) Can age be predicted fromdiameter for the obligate seeder Allocasuarina littoralis (Casuarinaceae)by using dendrochronological techniques? Aust J Bot55:433–438. doi:10.1071/BT06160Chen J, Feng Z (2002) Study on geographical compositions of seedplant flora in Gutianshan Mountain of Zhejiang Province. J EastChina Normal Univ (Nat Sci) 48:104–111 (in Chinese)Cook ER (1985) A time series analysis approach to tree-ringstandardization. Ph.D. Dissertation, University of Arizona,TucsonDang H, Zhang Y, Zhang K, Jiang M, Zhang Q (2010) Age structureand regeneration of subalpine fir (Abies fargesii) forests acrossan altitudinal range in the Qinling Mountains, China. Forest EcolManag 259:547–554. doi:10.1016/j.foreco.2009.11.011Fichtler E, Clark DA, Worbes M (2003) Age and long-term growth oftrees in an old-growth tropical rain forest, based on analyses oftree rings and 14 C 1 . Biotropica 35:306–317. doi:10.1646/03027Fritts HC (1976) Tree rings and climate. Academic Press, New YorkFujihara M (1996) Development of secondary pine forests after pinewilt disease in western Japan. J Veg Sci 7:729–738. doi:10.2307/3236384Grau HR, Easdale TA, Paolini L (2003) Subtropical dendroecology—dating disturbances and forest dynamics in northwestern Argentinamontane ecosystems. Forest Ecol Manag 177:131–143. doi:10.1016/S0378-1127(02)00316-XGutierrez AG, Armesto JJ, Aravena JC (2004) Disturbance andregeneration dynamics of an old-growth North Patagonian rainforest in Chiloe Island, Chile. J Ecol 92:598–608. doi:10.1111/j.0022-0477.2004.00891.xHolmes RL (1983) Computer-assisted quality control in tree-ringdating and measurement. Tree Ring Bull 43:69–78Hu Z, Yu M (2008) Study on successions sequence of evergreenbroad-leaved forest in Gutian Mountain of Zhejiang, EasternChina: species diversity. Front Biol China 3:45–49. doi:10.1007/s11515-008-0011-4Kruskal WH, Wallis WA (1952) Use of ranks in one-criterionvariance analysis. J Am Stat Assoc 47:583–621Kuang YW, Sun FF, Wen DZ, Zhou GY, Zhao P (2008) Tree-ring growthpatterns of Masson pine (Pinus massoniana L.) during the recentdecades in the acidification Pearl River Delta of China. Forest EcolManag 255:3534–3540. doi:10.1016/j.foreco.2008.02.036Landis RM, Peart DR (2005) Early performance predicts canopyattainment across life histories in subalpine forest trees. Ecology86:63–72. doi:10.1890/03-0848Legendre P, Mi X, Ren H, Ma K, Yu M, Sun I-F, He F (2009)Partitioning beta diversity in a subtropical broad-leaved forest ofChina. Ecology 90:663–674. doi:10.1890/07-1880.1Li J, Cook ER, D’Arrigo R, Chen F, Gou X, Peng J, Huang J (2008)Common tree growth anomalies over the northeastern TibetanPlateau during the last six centuries: implications for regionalmoisture change. Glob Change Biol 14:2096–2107. doi:10.1111/j.1365-2486.2008.01603.xLiu J, Hong W (1999) Time series model of individual age anddiameter in Castanopsis kawakamii population. Acta PhytoecolSin 23:283–288 (in Chinese)Liu Y, Ma L, Leavitt SW, Cai Q, Liu W (2004) A preliminaryseasonal precipitation reconstruction from tree-ring stable carbonisotopes at Mt. Helan, China, since AD 1804. Global PlanetChange 41:229–239. doi:10.1016/j.gloplacha.2004.01.009Lorimer CG, Frelich LE (1989) A methodology for estimating canopydisturbance frequency and intensity in dense temperate forests.Can J For Res 19:651–663. doi:10.1139/x89-102Lusk CH, Smith B (1998) Life history differences and tree speciescoexistence in an old-growth New Zealand rain forest. Ecology79:795–806. doi:10.1890/0012-9658(1998)079[0795:LHDATS]2.0.CO;2123180

Author's personal copy290 Trees (2012) 26:283–290Ma K (2008) Large scale permanent plots: important platform forlong term research on biodiversity in forest ecosystem. J PlantEcol (Chinese Version) 32:237Rozendaal DMA, Brienen RJW, Soliz-Gamboa CC, Zuidema PA (2010)Tropical tree rings reveal preferential survival of fast-growingjuveniles and increased juvenile growth rates over time. NewPhytol 185:759–769. doi:10.1111/j.1469-8137.2009.03109.xSchongart J, Orthmann B, Hennenberg KJ, Porembski S, Worbes M(2006) Climate–growth relationships of tropical tree species inWest Africa and their potential for climate reconstruction.Global Change Biol 12:1139–1150. doi:10.1111/j.1365-2486.2006.01154.xShao Q, Huang L, Liu J, Yang H, Chen Z (2009) Dynamic analysis oncarbon accumulation of a plantation in Qianyanzhou based ontree ring data. J Geogr Sci 19:691–706. doi:10.1007/s11442-009-0691-ySong K, Yu Q, Shang K, Yang T, Da L (2011) The spatio-temporalpattern of historical disturbances of an evergreen broadleavedforest in East China: a dendroecological analysis. Plant Ecol(Early view). doi:10.1007/s11258-011-9907-1Splechtna BE, Gratzer G, Black BA (2005) Disturbance history of aEuropean old-growth mixed-species forest—a spatial dendroecologicalanalysis. J Veg Sci 16:511–522. doi:10.1111/j.1654-1103.2005.tb02391.xStewart GH (1986) Population dynamics of a montane conifer forest,western Cascade Range, Oregon, USA. Ecology 67:534–544.doi:10.2307/1938596Wang T, Zhang QB, Ma KP (2006) Treeline dynamics in relation toclimatic variability in the central Tianshan Mountains,northwestern China. Glob Ecol Biogeogr 15:406–415. doi:10.1111/j.1466-822x.2006.00233.xWhitney J (1997) Testing for differences with the nonparametricMann–Whitney U test. J Wound Ostomy Cont Nurs 24:12Worbes M (1995) How to measure growth dynamics in tropicaltrees—a review. Iawa J 16:337–351Worbes M, Junk WJ (1989) Dating tropical trees by means of 14 Cfrom bomb tests. Ecology 70:503–507. doi:10.2307/1937554Worbes M, Staschel R, Roloff A, Junk WJ (2003) Tree ring analysisreveals age structure, dynamics and wood production of a naturalforest stand in Cameroon. For Ecol Manag 173:105–123. doi:10.1016/S0378-1127(01)00814-3Yu M, Hu Z, Yu J, Ding B, Fang T (2001) Forest vegetation types inGutianshan Natural Reserve in Zhejiang. J Zhejiang Univ (AgricLife Sci) 27:375–380 (in Chinese)Zhang Y, Liu P, Zhu Q (2002) Study on rare and endangered flora inGutianshan Reserve of Zhejiang. J Zhejiang For Sci Technol22:0005–0008 (in Chinese)Zhang Q-B, Cheng G, Yao T, Kang X, Huang J (2003) A 2,326-yeartree-ring record of climate variability on the northeasternQinghai-Tibetan Plateau. Geophys Res Lett 30:1739–1742. doi:10.1029/2003gl017425Zhu Y, Zhao G, Zhang L, Shen G, Mi X, Ren H, Yu M, Chen J, ChenS, Fang T, Ma K (2008) Community composition and structureof Gutianshan forest dynamic plot in a mid-subtropical evergreenbroad-leaved forest, East China. J Plant Ecol (ChineseVersion) 32:262–273123181

ArticleEcologyFebruary 2012 Vol.57 No.6: 623630doi: 10.1007/s11434-011-4869-1SPECIAL TOPICS:Comparison of phylobetadiversity indices based on community datafrom Gutianshan forest plotFENG Gang 1,2 , ZHANG JinLong 1,2 , PEI NanCai 3 , RAO MiDe 4 , MI XiangCheng 1* ,REN HaiBao 1 & MA KePing 11State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, Beijing 100093, China;2Graduate University of Chinese Academy of Sciences, Beijing 100049, China;3Research Institute of Tropical Forestry, Chinese Academy of Forestry, Guangzhou 510520, China;4College of Chemistry and Life Sciences, Zhejiang Normal University, Jinhua 321004, ChinaReceived June 13, 2011; accepted September 28, 2011; published online December 8, 2011Phylobetadiversity incorporates phylogenetic information and beta diversity, and can account for the ecological similarities betweencommunities with a phylogenetic perspective. Although different phylobetadiversity indices reflect differences in differentcharacteristics between communities, the results of different phylobetadiversity indices are not comparable. In this study we examinedphylobetadiversity indices for a 24-hm 2 plot in the Gutianshan National Nature Reserve. It was found the abundanceweightedD pw was almost identical to Rao’s D of Rao’s quadratic entropy. PhyloSor had a similar ecological meaning and algorithmto UniFrac. Although D nn was different in definition from UniFrac and PhyloSor, they were all strongly correlated. Theeffect of species abundance on phylobetadiversity was not significant when scales were relatively small, but was significant atlarger scales. These contrasts likely resulted from reductions in evenness in communities as scales increased. P ST and Rao’s Hbetter reflected the distance-decay changes caused by spatial and habitat variation than other indices at larger scales, whereasAW-D nn and D nn better reflected these changes at small scales.Scale, abundance, correlation, spatial pattern, distance decayCitation:Feng G, Zhang J L, Pei N C, et al. Comparison of phylobetadiversity indices based on community data from Gutianshan forest plot. Chinese Sci Bull,2012, 57: 623630, doi: 10.1007/s11434-011-4869-1Beta diversity is generally defined as the change in communitycomposition along environmental gradients [1,2]. Becauseit directly links local diversity (alpha diversity) withregional diversity (gamma diversity), beta diversity is currentlyan important topic in community ecology [1,2].Studies on the relationships between beta diversity and speciescharacteristics, habitat gradients and limitations to seeddispersal can explain the mechanisms shaping patterns ofbeta diversity and test ecological hypotheses regarding theeffects of regional diversity on local diversity [3,4].Recently, the consideration of phylogenetic relationshipsamong the species making up a community has providednew insights into community ecology [5]. Based on species*Corresponding author (email: mixiangcheng@ibcas.ac.cn)beta diversity, phylobetadiversity is defined as the phylogeneticdistance among species or individual organisms ofdifferent communities [6]. Species beta diversity can beused to describe the dissimilarity of species compositionamong communities, but it is uninformative about the dissimilarityof phylogenetic relationships among communities[7,8]. For example, for 4 forest communities A, B, C and Dlocated at different latitudes, all dissimilarity indices of speciesbeta diversity between the forests of 1 means the speciescomposition is completely different. However, thephylobetadiversity among these 4 communities may be dissimilarto different extents according to the relatedness ofthe communities.Phylobetadiversity has been of considerable interest becauseof the potential new insights into community ecology© The Author(s) 2011. This article is published with open access at Springerlink.com csb.scichina.com www.springer.com/scp182

624 Feng G, et al. Chin Sci Bull February (2012) Vol.57 No.6it provides. A variety of definitions of phylobetadiversityand associated phylobetadiversity indices have been putforward. For example, D pw is the mean pairwise phylogeneticdistance of species or individuals among communities[9]; PhyloSor is defined as the proportion of branch lengthof shared species relative to total branch length of all speciesin two communities [10]; and PCD is defined as theextent to which the variance of a randomly selected trait inone community can be predicted by the value of the sametrait in another community [11]. Presently, about 12 phylobetadiversityindices have been proposed. These indicescan reflect dissimilarities among communities from differentaspects, but the results using different indices are unlikelyto be comparable.The main ecological processes structuring the patterns ofphylobetadiversity include niche-based deterministic anddispersal-based neutral models [6]. Niche-based deterministicmodels assume that phylobetadiveristy patterns are determinedby habitat heterogeneity and interspecific tradeoffsin resource utilization, whereas dispersal-based neutralmodels hypothesize that phylobetadiversity patterns areshaped by spatial community dynamics, such as dispersallimitation [6]. Which of these indices can better reflect thechanges in phylogenetic community structure along spatialand habitat gradients? So far, no studies have compared thecharacteristics of different phylobetadiversity indices.This study aimed to compare properties of different phylobetadiversityindices using community data and environmentaldata from a 24-hm 2 forest dynamic plot in the GutianshanNational Nature Reserve. The Gutianshan Reserveis ideal to conduct such a comparative study. Previous studieson subtropical evergreen broad-leaved forests in the Gutianshanreserve found that both niche and spatial processeshave similar effects (30% vs. 29%) on species beta diversity[12]. Moreover, the soil nutrients and species distributionhave been mapped in the reserve at high resolution.Based on the data from the Gutianshan plot, we aimed toaddress 4 questions: (1) Do phylobetadiversity indices, integratedwith phylogenetic information, differ from speciesbeta diversity indices? (2) Are phylobetadiversity indicescorrelated with each other, and do they have similar ecologicalmeaning? (3) Does the effect of abundanceweightingon indices of phylobetadiversity vary with scale?(4) Can these indices reflect the phylobetadiversity alongspatial and habitat gradients?1 Materials and methods1.1 Study siteThe Gutianshan subtropical evergreen broad-leaved forestdynamic plot is located in the Gutianshan National NaturalReserve in Kaihua County, Zhejiang Province, China(29°10′19.4″~29°17′41.4″N, 118°03′49.7″~118°11′12.2″E).The total area of the reserve is 8107 hm 2 and its topographyis complex. The climate belongs to the middle subtropicalmonsoon climate zone, where the mean annual temperatureis 15.3°C, mean hottest month temperature is 28.9°C andmean coldest month temperature is 4.1°C. With 140-d precipitation,the annual precipitation is 1963.7 mm. The areaexperiences, on average, 1747.5 sunshine hours and a 250-dfrost-free season every year [13]. The Gutianshan forest plotis 600 m long and 400 m wide, and the altitude above searanges from 446.3 to 714.9 m. The plot was established in2005 following the Center for Tropical Forest Science censusprotocol [14].1.2 Reconstruction of community phylogenyWe reconstructed community phylogenies following Kresset al. [15] by sequencing 3 chloroplast DNA regions (rbcLa,matK and trnH-psbA) of 156 woody species growing in theplot. Total DNA was extracted from leaf tissue of plantsamples using the CTAB method [16,17]. The 3 chloroplastDNA regions were amplified and sequenced, and the nucleotidesequences were aligned using MUSCLE [18]. Thethree regions were assembled into a supermatrix using Rpackage phylotools [19]. Three-partition GTR + GAMMAmodels were applied to the 3 regions separately usingRAxML [20] and a community phylogeny was constructedusing maximum likelihood analysis. A bootstrap analysiswith 1000 replicates was conducted to assess the percentagesupport for each node. Finally, an ultrametric tree was obtainedusing the non-parametric rate smoothing approach inthe r8s software package [15,21,22].1.3 Data analysisCommunity data was obtained from the first survey of theGutianshan 24 hm 2 forest plot. The survey covered 140700woody plant individuals with diameter at breast height (dbh)≥1 cm that belonged to 49 families, 104 genera and 159species. The most abundant species were Castanopsis eyreiand Schima superba. We divided the 24-hm 2 plot into 60020 m × 20 m, 150 40 m × 40 m, and 24 100 m × 100 m separatesamples.First, we conducted a Spearman correlation analysis ofthe 12 phylobetadiversity indices at the same spatial scale.Then we tested for correlations at 4 spatial scales betweenthe 3 pairs of indices (D pw vs. AW-D pw , D nn vs. AW-D nn ,and ∏ ST vs. P ST ) in which each pair differed by whether ornot it was abundance-weighted. Finally, we ran a PartialMantel test with the phylobetadiversity indices values andspatial distance and/or environmental distance to comparethe extent that the variation of phylobetadiversity caused byspatial and environmental distance could be explained bythe different indices.The environmental variables included four topographic183

Feng G, et al. Chin Sci Bull February (2012) Vol.57 No.6 625factors (mean elevation, convexity, slope, and aspect) [12]and 20 soil factors (total C, total N, total P, extractable Fe,extractable Mn, extractable Zn, extractable Cu, extractableK, extractable P, extractable Ca, extractable Mg, extractableNa, extractable B, extractable Si, extractable Al, extractableN, pH, N mineralization rate, bulk density, and soil moisture).Topographic factors can reflect soil moisture and nutrientsindirectly. Because species in different conditions ofsoil moisture and nutrition have different competitive abilities,soil moisture and nutrition can affect the distribution ofplants directly, so we used topographic and soil factors torepresent ecological niche processes and spatial distance torepresent spatial processes. According to geostatisticalmethods [23], we conducted soil sampling in 30 m × 30 mgrids at different scales and standardized the measured factorswith different methods depending on their attributes:pH values of every sample were standardized, the other soilfactors were log-transformed, aspects of samples weretransformed with sin(aspect) and cos(aspect) values to representthe extent of aspects facing south and east, and theother topographic factors were standardized with the samemethod as soil pH values.We conducted principal component analysis (PCA) todetermine the extent of variables that contributed to theprincipal components. The first principal component, inwhich mean elevation was the most important contributingvariable, explained 92.9% of the information. The loadingmatrix of the environmental factors of the PCA is presentedin Table S1. Thus we used the values for the first axis ofevery sample to represent habitat factors when calculatingenvironmental distance between samples. Finally, we appliedvariance partitioning to phylobetadiversity values withspatial distance and/or environmental distance.1.4 Phylobetadiversity indices(i) D pw . D pw calculates the mean pairwise phylogeneticdistance of different species or individuals among communities[9]. The algorithms areDnk1 nk2 i1 ik2 j1jk1PW,nk n1 k 2nkn1 k2f 1 i ikfi2 j1j jk1AW Dpw , species i species j,2where ikis the mean pairwise phylogenetic distance2between species i in community k 1 and all species in communityk 2 ; nk 1represents the number of species in communityk 1 ; f i is the relative abundance of species i in communityk 1 [24]; and AW-D pw is the abundance-weightedD pw .(ii) D nn . D nn is defined as the mean phylogenetic distancebetween a species in community A and its most-related species in community B [9]. The formulas areSASB min mini1 iB j1jADnn,S SA BSASBf min min1 i iB fij1j jAAW- Dnn2,species i species j,where S A is the number of species in community A;min jArepresents the phylogenetic distance between speciesj in community B and its closest relative species incommunity A; f i is the relative abundance of species i incommunity A [24,25]; and AW-D nn is the abundanceweightedD nn .(iii) PhyloSor. PhyloSor (Phylogenetic Sørensen index)is the proportion of branch length between shared species tototal branch length of all species in two communities. Asindicated in the name of this index, PhyloSor is a derivativeof the Sørensen index (Sor), which is the proportion ofshared species in relation to the total number of species inthe communities. The larger the Sørensen value, the moresimilar the species composition is among communities; alarger PhyloSor value indicates closer community relationships.PhyloSor and Sørensen values are calculated by theformulas:2SijSorij ,Si Sj2BLijPhyloSorij ,BLi BLjwhere S ij is the number of species shared by two communities;S i is the total number of species in community i [26];BL ij represents branch length between species shared by twocommunities; and BL i is the branch length between all speciesof community i [10].(iv) UniFrac. UniFrac (unique fraction) is defined asthe percentage of branch length between species unique toone community [27]. The Jaccard index [1], which is similarto UniFrac, is an index of species beta diversity. Formulasfor these two indices arebcJaccard ,a b cBCUniFrac ,A BCwhere a is number of shared species between communities;b is number of species unique to community 1; c is numberof species unique to community 2; A represents branchlength between species shared by the two communities; Brepresents branch length of species unique to community 1;and C represents branch length of species unique to community2. UniFrac and PhyloSor have similar ecologicalmeanings; the only difference is that the former is an indexof dissimilarity between communities, whereas the latter isan index of similarity.184

626 Feng G, et al. Chin Sci Bull February (2012) Vol.57 No.6(v) Rao’s D. Rao’s D has the same meaning as AW-D pw . It is computed asD t x x kl ij ki lji jwhere x ki is relative abundance of species i in community k;and t ij represents phylogenetic distance between species iand j [28].(vi) Rao’s H. Rao’s H is an index of phylogenetic diversitywithout consideration of the effect of phylogeneticdiversity within communities. The formula isH D ( D D)/2,kl kl kk llwhere D kl is the mean pairwise phylogenetic distance betweencommunities; and D kk and D ll are the mean pairwisephylogenetic distances within communities k and i, respectively.(vii) ST and P ST . ST is calculated as the mean pairwisephylogenetic distance among communities minus themean pairwise phylogenetic distance within communities. ST only uses (0, 1) data, whereas P ST contains speciesabundance information [8]. The indices are calculated byP PTSST ,PDT DP PT SST ,PDTPwhere T P is total phylogenetic diversity among all communitiesthat can be understood as phylogenetic γ diversity; SPrepresents the mean value of phylogenetic diversity withincommunities that can be understood as phylogenetic alphadiversity; and D T P and D S P are abundance-weighted phylogeneticgamma and alpha diversity, respectively.(viii) PCD. PCD (phylogenetic community dissimilarity)can be broken up into a non-phylogenetic component(PCD c ), which reflects shared species among communities,and a phylogenetic component (PCD p ), which reflects relatednessof different species among communities. Interpretationof this phylogenetic beta diversity required the assumptionthat the evolutionary process of a non-selectedtrait is random, which is Brownian motion. Given the phylogeneticrelationship between two communities, PCD representsthe extent that variance of the trait in community 1can be predicted by the same trait value in community 2:n1PSV1|2 n2PSV2|11PCD ,nPSV n PSV Dn ( nC)1 1 2 2 1 2 poolwhere n 1 means species number in community 1; PSV 1|2represents the variance of the trait among species in community1 given the trait variance in community 2 [29]; PSV 1is the variance of the trait among species in community 1;,and Dn (1n2C pool) is used to remove the deviationcaused by n 1 , n 2 [11].2 Results2.1 Correlations between indicesWe investigated correlations between phylobetadiversityindices caused by the integration of phylogenetic informationwith species beta diversity using Spearman correlationanalysis between UniFrac and PhyloSor, UniFrac andJaccard [1], and PhyloSor and Sørensen [26]. These threepairs of indices have similar ecological meaning. Results atdifferent spatial scales are shown in Table 1. BecauseSørensen and Jaccard are the species beta diversity of PhyloSorand UniFrac, correlations between the phylobetadiversityindices change after integration of phylogenetic information(Table 1).We calculated 12 phylobetadiversity indices (if weightedand unweighted abundances are included), among whichfive pairs were highly correlated (Figure 1, Tables 2 and 3).The highly correlated pairs were AW-D pw and Rao’s D,PhyloSor and UniFrac, D nn and PhyloSor, D nn and UniFrac,and Rao’s H and P ST .Since the indices PCD and PCDp account for the covarianceof phylogenetic distance between species, we comparedthem with UniFrac and ∏ ST , which also take the covarianceof phylogenetic distance between species into consideration(Table 3 for the spatial scale of 100 m × 100 m,and see Tables S2 and S3 for all other spatial scales).2.2 Effect of abundance-weighting on phylobetadiversityWe conducted Spearman correlation analysis on three pairsof indices that differed in being either abundance-weightedor not at 4 spatial scales, namely 10 m × 10 m, 20 m × 20 m,40 m × 40 m, and 100 m × 100 m. The difference betweeneach pair of indices was relatively small and became largerwith increased spatial scale (Figure 2). This could be explainedby the change of species evenness with increasedscale. Sampling of species abundance distribution at differentscales supported this hypothesis (Figure 3).Table 1 Spearman rank correlation coefficients between UniFrac andPhyloSor, UniFrac and Jaccard and PhyloSor and Sørensen at three spatialscalesIndices 20 m × 20 m 40 m × 40 m 100 m × 100 mUniFrac and PhyloSor −1 −1 −1UniFrac and Jaccard 0.81 0.82 0.81PhyloSor and Sørensen 0.81 0.82 0.81185

Feng G, et al. Chin Sci Bull February (2012) Vol.57 No.6 627Figure 1 Spearman rank correlation coefficients between phylobetadiversity indices at a spatial scale of 100 m × 100 m.Table 2 Spearman rank correlation coefficients between phylobetadiversity indices at a spatial scale of 100 m × 100 m; high correlation coefficients werebolded.0.26 0.47 0.34 0.31 −0.32 0.32 1 0.47 0.41 P ST0.04 0.36 0.6 0.6 −0.55 0.55 0.42 0.36 ∏ ST−0.05 1 0.11 0.17 −0.11 0.11 0.49 Rao’s D0.25 0.49 0.34 0.32 −0.32 0.32 Rao’s H−0.08 0.11 0.98 0.66 −1 UniFrac0.08 −0.11 −0.98 −0.66 PhyloSor−0.02 0.17 0.66 AW-D nn−0.01 0.11 D nn−0.05 AW-D pwD pwTable 3 Spearman rank correlation coefficients between phylobetadiversityindices at a spatial scale of 100 m × 100 m0.27 0.57 0.24 ∏ ST0.86 0.43 UniFrac0.53 PCD pPCD2.3 Ability of indices to reflect variation in communitycomposition along habitat and spatial gradientsTo find which index can best reflect the variation of phylobetadiversityalong habitat and spatial gradients, we useda Partial Mantel test to partition phylobetadiversity withrespect to habitat and/or spatial gradients (Table 4 for 100 m× 100 m spatial scale, and see Tables S4 and S5 for all otherscales).Figure 2 Effect of spatial scale on correlation coefficients betweenabundance-weighted and non-weighted phylobetadiversity indices.186

628 Feng G, et al. Chin Sci Bull February (2012) Vol.57 No.6Figure 3Absolute abundances of abundance rankings at different spatial scales.Table 4 Results of a variation partitioning between phylobetadiversityvalues and habitat and spatial gradients at the spatial scale of 100 m × 100 mIndex a + b + c a + b b + c a cD pw 0.025 −0.004 0.022 0.003 0.029AW-D pw 0.065 0.002 0.067 −0.002 0.063D nn 0.225 0.097 0.206 0.018 0.128AW-D nn 0.303 0.106 0.29 0.013 0.197UniFrac 0.233 0.091 0.219 0.014 0.142PhyloSor 0.236 0.091 0.222 0.014 0.145Rao’s D 0.065 0.002 0.067 −0.002 0.063Rao’s H 0.321 0.053 0.323 −0.002 0.268∏ ST 0.216 0.121 0.18 0.037 0.096P ST 0.328 0.056 0.33 −0.002 0.272PCD 0.018 0.021 0.005 0.013 −0.002PCD p 0.002 0.001 −0.003 0.005 03 DiscussionBased on species beta diversity, phylobetadiversity can providenew insights into species coexistence from phylogeneticrelationships among species [6]. Niche conservationduring evolutionary history plays a critical role in determiningspecies distribution and offers a basis for phylobetadiversitystudies [30,31]. Phylobetadiversity can beused to explore mechanisms of biodiversity maintenanceand may have a higher utility than species beta diversity asa conservation criterion for management decisions.3.1 Correlations between indicesCorrelation analysis of phylobetadiversity indices can helpus to understand better the ecological meaning of phylobetadiversityand avoid confusion. Both AW-D pw andRao’s D represent abundance-weighted mean pairwise phylogeneticdistance and results of the two indices are identical,so we can choose one of these indices in future studies;PhyloSor and UniFrac are derived from the Sørensen andJaccard indices and have similar ecological meanings, i.e.,the proportion of branch length shared between species tototal branch length of all species in two communities. Theonly difference between the two indices is that PhyloSor isan index of similarity, whereas UniFrac is an index of dissimilarity.Rao’s H and P ST also have similar ecologicalmeanings and are highly correlated; what needs furtherstudy is the finding that correlations between D nn and PhyloSor,D nn and UniFrac are also highly correlated. D nn representsthe mean nearest phylogenetic distance between twocommunities, whereas PhyloSor and UniFrac are the proportionof branch length between species shared relative tototal branch length of all species in two communities. All ofthese indices represent the difference between the terminalsof phylogenetic trees of species from two different communities,which may explain their strong correlation.3.2 Effect of abundance-weighting on indicesAbundance, which reflects differences in species evennessbetween communities, is an important concept in communityecology. Lozupone et al. [32] found that abundanceweightedindices are suitable in studies that investigatechanges in species abundance when the mechanisms may becorrelated with subtle environmental changes. On the otherhand, abundance-unweighted indices are used mainly todiscuss factors limiting species presence.The correlation analysis of the three pairs of indices atfour spatial scales showed a decrease in strength of correlationwith increasing spatial scales. In other words, when thescale is small, effect of abundance is not significant and theeffect becomes more obvious as scale increases. Thus if thescale applied in an investigation is relatively small, both ofeach of the pairs might not need to be calculated.3.3 Performance of indices that reflect habitat andspatial gradientsIt has been shown widely that similarities between commu-187

Feng G, et al. Chin Sci Bull February (2012) Vol.57 No.6 629nities decrease as spatial distances between the communitiesincrease [33]. There are two main mechanisms that explainthis phenomenon: (1) habitat characteristics change withincreased separation distance, and affect community compositionand can be interpreted by the niche hypothesis [34];and (2) according to neutral theory [35], although the habitatbetween communities may be similar, similarity betweencommunities will decrease with increased spatial distancebecause of the limiting dispersal abilities of organisms.Phylobetadiversity, as the measurement of phylogeneticdistance between communities, should also decay with distance.Results of variance partitioning of phylobetadiversitybetween habitat and spatial distance indicate that P ST andRao’s H can best reflect distance decay along habitat andspatial gradients in subtropical forest communities at thespatial scale of 100 m × 100 m. At this scale, habitat inconjunction with spatial distance explained 32.77% and32.1% of the variance of Rao’s H and P ST , respectively;however, as the spatial scale decreased to 20 m × 20 m and40 m × 40 m, distance decay along spatial and habitat gradientswas best reflected by AW-D nn and D nn . At a scale of20 m × 20 m the variance of AW-D nn and D nn explained byhabitat factors and spatial distance was 13.4% and 9.375%,respectively, and at the 40 m × 40 m scale 18.63% and13.47% were explained by the same variables, respectively.Therefore we recommend using P ST and Rao’s H when thescales are relatively large, and AW-D nn and D nn when scalesare smaller, for studies on effects of different factors onphylobetadiversity.We found that AW-D pw and Rao’s D, PhyloSor andUniFrac, D nn and PhyloSor, D nn and UniFrac, Rao’s H andP ST are strongly correlated. D pw , AW-D pw , Rao’s D, Rao’sH, P ST and ∏ ST are based on the mean pairwise phylogeneticdistance, which reflects branch differences close to thephylogenetic tree root between communities and can moreeffectively reflect the habitat difference than other indices.Calculations of D nn , AW-D nn , UniFrac and PhyloSor aremeasures of differences in phylogenetic tree terminals,which are differences in evolutionary distinctiveness [36]and may be used to reflect differences in resource utilizationstrategies. For these indices, abundance-weighting at smallscales has little effect, but the effect of abundance increasesat larger scales. At large spatial scales, habitat and spaceexplain P ST and Rao’s H best, but at smaller spatial scalesAW-D nn and D nn are better explained by habitat and space.Phylobetadiversity provides a new perspective on relationshipsbetween communities; therefore, studies combiningspecies beta diversity and phylobetadiversity may bemore comprehensive. This study analyzed correlationsamong phylobetadiversity indices, effects of abundanceweightingon phylobetadiversity indices at different spatialscales, and the potential of the indices to reflect spatial andhabitat gradients. The findings might be helpful to researchersinterested in phylogenetic ecology.We thank Dr. Nathan Swenson for his insightful suggestions on this manuscript,Liu Xiaojuan, Man Xingxing and Song Kai for their help with dataprocessing. Dr. Liwen Zhang supplied the soil data and a number ofteachers and workmates also assisted in the Gutianshan 24 ha plot. Wealso thank Christine Verhille, University of British Columbia, for her assistancewith English language and grammatical editing of the manuscript.This work was supported by Key Innovation Project of Chinese Academy ofSciences (KZCX2-EW-Z-5) and National Natural Science Foundation ofChina (31170401).1 Whittaker R H. Vegetation of the Siskiyou Mountains, Oregon andCalifornia. Ecol Monogr, 1960, 30: 280–3382 Cody M L. Towards a theory of continental species diversities: birddistributions over Mediterranean habitat gradients. In: Cody M L,Diamond J M, eds. Ecol Evolcomm. Harvard: Harvard UniversityPress, 1975, 214–2573 Koleff P, Gaston K J, Lennon J J. Measuring beta diversity for presence-absencedata. J Anim Ecol, 2003, 72: 367–3824 Chen S B , Ouyang Z Y, Xu W H, et al. The research progress of betadiversity (in Chinese). Biodiver Sci, 2010, 18: 323–3355 Webb C O. Exploring the phylogenetic structure of ecological communities:An example for rain forest trees. Am Nat, 2000, 156: 145–1556 Graham C H, Fine P V A. Phylogenetic beta diversity: Linking ecologicaland evolutionary processes across space in time. Ecol Lett,2008, 11: 1265–12777 Chave J, Chust G, Thébaud C. The importance of phylogeneticstructure in biodiversity studies. In: Storch D, Marquet, Brown J H,eds. Scaling Biodiversity, Institute Editions, Santa Fe, 2007. 151–1678 Hardy O J, Senterre B. Characterizing the phylogenetic structure ofcommunities by an additive partitioning of phylogenetic diversity. JAnim Ecol, 2007, 95: 493–5069 Webb C O, Ackerly D D, Kembel S W. Phylocom: Software for theanalysis of phylogenetic community structure and trait evolution. Bioinformatics,2008, 24: 2098–210010 Bryant J A, Lamanna C, Morlon H, et al. Microbes on mountainsides:Contrasting elevational patterns of bacterial and plant diversity. ProcNatl Acad Sci USA, 2008, 105: 11505–1151111 Ives A R, Helmus M R. Phylogenetic metrics of community similarity.Am Nat, 2010, 176: E128–E14212 Pierre L, Mi X C, Ren H B, et al. Partitioning beta diversity in a subtropicalbroad-leaved forest of China. Ecology, 2009, 90: 663-67413 Lou L H, Jin S H. Spermatophyta flora of Gutianshan Nature Reservein Zhejiang (in Chinese). J Beijing Forest Univ, 2000, 22: 33–3914 Chen B, Mi X C, Fang T, et al. Gutianshan Forest Dynamic Plot:Tree Species And Their Distribution Patterns (in Chinese). Beijing:China Forestry Publishing House, 200915 Kress W J, Erickson D L, Jones F A, et al. Plant DNA barcodes and acommunity phylogeny of a tropical forest dynamics plot in Panama.Proc Natl Acad Sci USA, 2009, 106: 18621–1862616 Doyle J J, Doyle J L. A rapid DNA isolation procedure for smallquantities of fresh leaf tissue. Phytochem Bull, 1987, 19: 11–1517 Khanuja S P S, Shasany A K, Darokar M P, et al. Rapid isolation ofDNA from dry and fresh samples of plants producing large amountsof secondary metabolites and essential oils. Plant Mol Biol Rep, 1999,17: 1–718 Edgar R C. MUSCLE: Multiple seq uence alignment with high accuracyand high throughput. Nucleic Acids Res, 2004, 32: 1792–179719 Zhang J L, Mi X C, Pei N C. Phylotools: Phylogenetic tools forecologists. R package version 0.0.7.4. 20101920 Alexandros S. RAxML-VI-HPC: Maximum likelihood-based phylogeneticanalyses with thousands of taxa and mixed models. Bioinformatics,2006, 22: 2688–269021 Sanderson MJ. r8s: Inferring absolute rates of molecular evolutionand divergence times in the absence of a molecular clock. Bioinformatics,2003, 19: 301–30222 Lemey P, Rambaut A, Drummond A J, et al. Bayesian Phylogeographyfinds its roots. PLoS Comput Biol, 2009, 9: e100052023 Zhang L W, Mi X C, Shao H B, et al. Strong plant-soil associations188

630 Feng G, et al. Chin Sci Bull February (2012) Vol.57 No.6in a heterogeneous subtropical broad-leaved forest. Plant Soil, 2011,doi: 10.1007/s11104-001-0839-224 Swenson N G, Anglada-Cordero P, Barone J A. Deterministic tropicaltree community turnover: Evidence from patterns of functionalbeta diversity along an elevational gradient. Proc R Soc London SerB, 2011, 278: 877–88425 Ricotta C, Burrascano S. Testing for differences in beta diversity withasymmetric dissimilarities. Ecol Indic, 2009, 9: 719–72426 Krebs C J. Ecological Methodology. Menlo Park, CA: Addison-Wesley, 199927 Lozupone C, Hamady M, Knight R. UniFrac—An online tool forcomparing microbial community diversity in a phylogenetic context.BMC Bioinformatics, 2006, 7: 371–38428 Rao C R. Diversity and dissimilarity coefficients: A unified approach.Theor Popul Biol, 1982, 21: 244329 Helmus M R, Bland T J, Williams C K, et al. Phylogenetic measuresof biodiversity. Am Nat, 2007, 169: E68–E8330 Donoghue M J. A phylogenetic perspective on the distribution ofplant diversity. Proc Natl Acad Sci USA, 2008, 105: 11549–1155531 Cavender-Bares J, Ackerly D D, Baum D A, et al. Phylogeneticoverdispersion in Floridian oak communities. Am Nat, 2004, 163:823–84332 Lozupone C A, Hamady M, Kelley S T, et al. Quantitative and qualitativebeta diversity measures lead to different insights into factorsthat structure microbial communities. Appl Environ Microbiol, 2007,73: 1576–158533 Soininen J, McDonald R, Hillebrand H. The distance decay of similarityin ecological communities. Ecography, 2007, 30: 3–1234 Nekola J C, White P S. The distance decay of similarity in biogeographyand ecology. J Biogeogr, 1999, 26: 867–87835 Hubbell S P. The unified neutral theory of biodiversity and biogeography.Princeton: Princeton University Press, 200136 Redding D W, Mooers A O. Incorporating evolutionary measures intoconservation prioitization. Conserv Biol, 2006, 20: 1670–1678Open Access This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproductionin any medium, provided the original author(s) and source are credited.Supporting InformationTable S1 Principal component analysis of environmental factors at the spatial scale of 100 m × 100 mTable S2 Spearman rank correlation coefficients between phylobetadiversity indices at spatial scale of 40 m × 40 mTable S3 Spearman rank correlation coefficients between phylobetadiversity indices at spatial scale of 20 m × 20 mTable S4 Results of Mantel test between phylobetadiversity values and habitat and spatial gradients at a spatial scale of 40 m × 40 mTable S5 Results of Mantel test between phylobetadiversity values and habitat and spatial gradients at a spatial scale of 20 m × 20 mThe supporting information is available online at csb.scichina.com and www.springerlink.com. The supporting materialsare published as submitted, without typesetting or editing. The responsibility for scientific accuracy and content remains entirelywith the authors.189

Acta Oecologica 39 (2012) 87e93Contents lists available at SciVerse ScienceDirectActa Oecologicajournal homepage: www.elsevier.com/locate/actoecOriginal articleThe effects of ice storm on seed rain and seed limitation in an evergreenbroad-leaved forest in east ChinaYanjun Du a , Xiangcheng Mi a , Xiaojuan Liu a,b , Keping Ma a, *a State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, No. 20 Nanxincun, Xiangshan, Beijing 100093, Chinab Graduate University of Chinese Academy of Sciences, Beijing 100049, ChinaarticleinfoabstractArticle history:Received 1 June 2011Accepted 17 January 2012Available online 7 February 2012Keywords:Seed dispersalGutianshan plotSeed reproductionDispersal limitationSource limitationDispersal modeExtreme climatic events almost universally play a major role in influencing the composition andstructure of plant and animal communities, and thus could influence seed production, seed dispersal andseedling recruitment. We explored the effects of ice storm damage on seed rain and seed limitation ina 24-ha permanent forest plot in an evergreen broad-leaved forest in east China. We compared seedproduction before and after the storm in 2008. We evaluated the following hypotheses: 1) seedproduction after the ice storm was less than that before the storm; 2) seed limitation after the storm wasmore severe than before the storm. The results showed that seeds from one species, Eurya muricata,dominated the seed rain after the storm, accounting for more than half of the total seeds. Post-ice stormseed production of species other than E. muricata was only one fifth of that before the storm. Seedproduction in the second year after the ice storm recovered to pre-storm levels. The results indicate largeinter-specific variation in response to the ice storm. Disturbance caused by the ice storm greatlyincreased seed diversity. The Jaccard similarity of species before and after the ice storm was 58%. Therewas no significant difference in seed limitation or dispersal limitation before and after the storm, butthere was a significant difference in source limitation. Neither seed limitation nor dispersal limitationwas correlated with dispersal modes. Only source limitation for rodent dispersed species increased afterthe ice storm.Ó 2012 Elsevier Masson SAS. All rights reserved.1. IntroductionExtreme climatic events play a major role in influencing thecomposition and structure of plant and animal communities, andthus could influence seed production, seed dispersal and seedlingrecruitment (Wright et al., 1999). In early 2008, the longest coldspell for more than 40 years brought ice storms to a broad band ofsubtropical China, causing massive mechanical damage to nativebroad-leaved forests (Stone, 2008). The main culprit seems to havebeen freezing rain e supercooled liquid water droplets that freezeon contact with a surface

88Y. Du et al. / Acta Oecologica 39 (2012) 87e93are caused by the increased litter inputs after the storm, whichmay kill seedlings already present, inhibit germination bycovering seeds, and prevent seedling establishment (Uriarte et al.,2005). Ice storm damage to tree crowns could also decrease and/or delay seed production in the short term, as self-preservationagainst extreme storm conditions consumes stored reserves. Inthis case, seed rain will contribute little to new recruitment, andseed and seedling banks would be the major sources of regeneration.Ice storm events could also lead to changes in patterns offruit production, which would likely affect dispersers’ abundanceand behavior due to the loss of food resources. Moreover,increased light penetration through the damaged canopy to thesoil surface may kill seeds before germination or desiccateexisting seedlings.Conversely, positive effects may be due to increased lightpenetration, which would stimulate seed germination from theseed bank and the growth of seedlings present in the seedling bank,while increased litter inputs could also enhance nutrient supply(Lomascolo and Aide, 2001; Uriarte et al., 2005). In addition, icestorms may be expected to lead to weaker negative densitydependenteffects of seedling survival, as extra light could permitseedlings to resist pathogens and herbivores and thereby increasesurvival, which has significant implications for both seedlingsurvival and species coexistence. Understanding how ice stormsinfluence seed production and seedling recruitment is essential tothe prediction of the long-term effects of natural disturbance onsubtropical forest dynamics and species richness.Many previous studies of ice storms in forests have concentratedon the susceptibility of tree species to ice damage (Siccama et al.,1976; Seischab et al., 1993). Others have focused on the influenceof elevation (Downs, 1938; Nicholas and Zedaker, 1989), slope(Rhoades, 1995; Rebertus et al., 1997), and topography (Rebertuset al., 1997; Mou and Warrillow, 2000) on ice deposition, and onthe response of forest understory seedling/sapling density to majorice storms (Darwin et al., 2004). To our knowledge, no previousstudies have investigated the effects of ice storms on seedproduction.Because seedling data is largely unavailable, we focused here onthe seed rain phase. The major objective of our study was to explorethe effect of ice storm damage on seedfall, and investigate whetherthe degree of seed limitation is correlated with dispersal modes.We evaluated the following hypotheses:1) Seed production after the storm is less than that before thestorm, as self-preservation against extreme storm conditionsconsumes stored reserves;2) Seed limitation after the storm is more severe than before thestorm, as a consequence of decreased seed production;2. Material and methods2.1. Study siteThe study was conducted in a 24-ha permanent forest plot(29 15.101 0 -29 15.344 0 N, 118 07.010 0 e118 07.400 0 E) in GutianshanNational Nature Reserve (GNNR), Kaihua County, Zhejiang Provincein eastern China. GNNR covers a total area of approximately8107 ha. Mountains with steep slopes characterize the topography.The substrate consists mainly of granite. The mean annualtemperature is 15.3 C. The hottest month is July with a meantemperature of 27.9 C, and the coldest is January with a meantemperature of 4.3 C. The mean annual precipitation is 1787 mmwith seasonal distribution over the year (Yu et al., 2001). A total of1991 vascular plant species, belonging to 244 families and 897genera, have been recorded within the entire GNNR (Chen andFeng, 2002; Hu et al., 2003).2.2. Methods2.2.1. Description of plotIn 2005, the Gutianshan permanent plot (GTS) covering 24-ha(400 600 m, horizontal distance) was established within theevergreen broad-leaved forest in GNNR. Data were collectedfollowing the standard census protocol of the CTFS (Center forTropical Forest Science) network (Condit, 1998). The elevationrange between the highest and lowest points in the plot was 269 m(from 446 m to 715 m). The soil was a subtropical red soil (equivalentto Ultisols in US soil taxonomy) (Zhang et al., 2011). The firsttree census was conducted in 2005. All woody stems 1 cm in DBHwere mapped, measured, identified, and tagged (Legendre et al.,2009). There were a total of 140,700 individuals belonging to 159species in the plot. The dominant species were Castanopsis eyrei(Fagaceae), Schima superba (Theaceae) and Pinus massoniana(Pinaceae) (Zhang et al., 2011).2.2.2. Seed collectionSeed rain has been censused weekly since June 2006, using 130seed traps set along 2.3 km of trails within the plot (Du et al., 2009).As the ice storm destroyed all seed traps in February 2008, we reestablishedall seed traps in March and April of 2008 in theiroriginal sites. We resumed weekly collections in May 2008. Eachseed trap consisted of a square, 0.5 m 2 PVC frame supportinga shallow, open-topped, 1-mm nylon mesh bag suspended 0.8 mabove the ground on four PVC posts. All seeds, fruits, seed-bearingfruit fragments, flowers, capsules, and other reproductive parts ofplants that fell into the traps were identified to species and recorded.Fruits were categorized as aborted, immature, damaged,fragments and mature (Wright and Calderón, 1995; Wright et al.,1999). Because the seed traps were located above the ground,they captured fruits and falling seeds directly from trees, as well asthose spat or defecated by birds, bats and arboreal mammals; theydid not, however, record secondary dispersal by rodents and otherterrestrial animals (Muller-Landau et al., 2008). All data presentedrefer to seed number, based upon either a count of actual seeds perfruit, or calculated based upon the mean number of seeds per fruit.Dispersal mode was assigned to each species based on fruitmorphology (Du et al., 2009). Each plant species was assigned toone main dispersal mode. Fruit morphology was the basis forclassifying species as dispersed by wind, animals, or explosion(Table 1).2.2.3. Data analysis2.2.3.1. Calculation of recruitment limitation. We calculatedmeasures of seed limitation, source limitation and dispersal limitation(Clark et al., 1998; Nathan and Muller-Landau, 2000; Muller-Landau et al., 2002) for periods of 20 months before and after theice storm using data on seed arrival into the 130 seed traps in theplot. Seed limitation refers to the failure of seeds to arrive at allsuitable sites (Eriksson and Ehrlen, 1992; Turnbull et al., 2000).Fundamental seed limitation is defined as 1 ða=nÞ, where a is thenumber of seed traps reached by seeds and n is the total number ofseed traps. Thus, seed limitation can arise from limited seednumbers and limited dispersal of available seeds (Muller-Landauet al., 2002). Source limitation is the failure of seeds to reach sitesdue simply to insufficient seed numbers: there are not enoughseeds to go around, even if all seeds are uniformly distributedamong sites (Clark et al., 1998). Source limitation (stochastic) can bedefined as expð ðs=nÞÞ, where s is total seed number arriving alltraps. Non-uniform distribution of seeds is nearly ubiquitous,191

Y. Du et al. / Acta Oecologica 39 (2012) 87e93 89Table 1Demographic characteristics of 62 species whose seeds were collected during the 40-month period (T, tree; S, shrub). Dispersers are from Du et al. (2009), and Teng Fang(unpublished data). Pre-/post-seed density is the density of seeds 20 months before and 20 months after the ice storm. Pre-/post-seed limitation, pre-/post-source limitation,and pre-/post-dispersal limitation are the limitation 20 months before and 20 months after the ice storm. IV is the abbreviation of important value. Abundance is the totalindividual number in 24 ha plot. Table was sorted by the pre-seed density.SpeciesLifehistoryDispersersIVrankPre-seeddensity(m 2 )Post-seeddensity(m 2 )Pre-seedlimitationPost-seedlimitationPre-sourcelimitationPost-sourcelimitationPre-dispersallimitationSchima superba T Wind 2 930.59 49.97 0.00 0.02 0.00 0.00 0.00 0.02Castanopsis eyrei T Rodents 1 55.34 4.42 0.27 0.65 0.00 0.11 0.27 0.61Fraxinus insularis T Wind 33 43.06 1.22 0.18 0.83 0.00 0.54 0.18 0.63Ilex micrococca T Birds 92 26.12 6.83 0.98 0.98 0.00 0.03 0.98 0.98Eurya muricata S Mammals, birds 4 24.86 243.94 0.88 0.74 0.00 0.00 0.88 0.74Quercus serrata T Rodents 25 16.02 12.31 0.73 0.71 0.00 0.00 0.73 0.71Toxicodendron succedaneum T or S Animals 38 15.91 12.86 0.07 0.42 0.00 0.00 0.07 0.42Pinus massoniana T Wind 23 10.97 26.29 0.07 0.01 0.00 0.00 0.07 0.01Rhododendron ovatum S Wind 3 10.54 15.78 0.97 0.98 0.01 0.00 0.97 0.98Ternstroemia gymnanthera T or S Birds 9 8.89 12.35 0.52 0.34 0.01 0.00 0.52 0.34Daphniphyllum oldhamii T or S Mammals, birds 8 7.82 2.49 0.45 0.67 0.02 0.29 0.44 0.54Vaccinium bracteatum T or S Mammals, birds 19 6.91 0.37 0.96 0.98 0.03 0.83 0.96 0.86Eurya rubiginosa S Birds 12 5.80 1.20 0.99 0.99 0.06 0.55 0.99 0.98Vaccinium carlesii S Mammals, birds 14 5.62 0.00 0.95 1.00 0.06 1.00 0.94 1.00Cyclobalanopsis gracilis T or S Rodents 125 5.46 0.15 0.93 0.98 0.07 0.93 0.93 0.69Alniphyllum fortunei T Wind 47 4.38 1.29 0.95 0.99 0.11 0.52 0.95 0.98Machilus thunbergii T Animals 22 3.02 0.18 0.89 0.96 0.22 0.91 0.86 0.56Cyclobalanopsis glauca T Animals 21 2.08 0.08 0.85 0.97 0.35 0.96 0.76 0.18Itea oblonga T or S wind 27 1.80 0.00 0.99 1.00 0.41 1.00 0.99 1.00Neolitsea aurata T Mammals 5 1.72 0.77 0.88 0.84 0.42 0.68 0.79 0.49Adinandra millettii T or S Animals 35 1.69 44.00 0.99 0.96 0.43 0.00 0.99 0.96Euonymus oblongifolius T or S Animals 114 1.69 0.00 0.98 1.00 0.43 1.00 0.96 1.00Corylopsis glandulifera S Explosive, Wind 34 1.58 0.55 0.94 0.93 0.45 0.76 0.89 0.71Rhododendron latoucheae S Wind 13 1.38 14.85 0.99 0.98 0.50 0.00 0.98 0.98Loropetalum chinense T or S Explosive, weight 10 1.08 1.58 0.83 0.83 0.58 0.45 0.59 0.69Lithocarpus glaber T Rodents 20 1.08 0.06 0.95 0.99 0.58 0.97 0.87 0.75Myrica rubra T Mammals, birds 24 0.74 2.97 0.95 0.94 0.69 0.23 0.83 0.92Idesia polycarpa T Animals 99 0.72 0.00 0.99 1.00 0.70 1.00 0.97 1.00Vaccinium mandarinorum T or S Mammals, birds 11 0.40 3.80 0.99 0.98 0.82 0.15 0.96 0.97Camellia chekiangoleosa T or S Mammals 7 0.22 0.05 0.98 0.99 0.90 0.98 0.77 0.66Albizia kalkora T Animals 44 0.17 0.14 0.93 0.99 0.92 0.93 0.15 0.88Dalbergia hupeana T Wind 48 0.15 0.05 0.98 0.99 0.93 0.98 0.69 0.66Distylium myricoides T or S Explosive 36 0.14 4.88 0.99 0.87 0.93 0.09 0.88 0.86Castanopsis fargesii T Rodents 29 0.11 0.00 0.99 1.00 0.95 1.00 0.85 1.00Camellia fraterna T or S Mammals 15 0.08 0.03 0.96 0.98 0.96 0.98 0.02 0.01Styrax dasyanthus S Animals 65 0.06 0.05 0.98 0.98 0.97 0.98 0.49 0.33Michelia skinneriana T Birds 70 0.06 0.00 0.99 1.00 0.97 1.00 0.75 1.00Malus leiocalyca T or S Rodents 56 0.05 0.02 0.99 0.99 0.98 0.99 0.66 0.00Euscaphis japonica T or S Birds 67 0.05 0.03 0.99 0.99 0.98 0.98 0.66 0.50Diospyros glaucifolia T Mammals 59 0.05 0.03 0.99 0.98 0.98 0.98 0.66 0.01Elaeocarpus decipiens T Animals 42 0.02 0.00 0.99 1.00 0.99 1.00 0.00 1.00Sloanea sinensis T Animals 101 0.02 0.02 0.99 0.99 0.99 0.99 0.00 0.00Nyssa sinensis T Animals 123 0.02 0.00 0.99 1.00 0.99 1.00 0.00 1.00Rhaphiolepis indica S Mammals, birds 17 0.02 0.46 0.99 0.94 0.99 0.79 0.00 0.70Rhododendron simsii S Animals 30 0.02 0.00 0.99 1.00 0.99 1.00 0.00 1.00Litsea coreana T Animals 32 0.00 0.06 1.00 0.99 1.00 0.97 1.00 0.75Wikstroemia monnula S Animals 81 0.00 0.18 1.00 0.99 1.00 0.91 1.00 0.91Symplocos anomala T or S Animals 61 0.00 0.17 1.00 0.99 1.00 0.92 1.00 0.91Syzygium buxifolium T or S Mammals, birds 16 0.00 5.23 1.00 0.82 1.00 0.07 1.00 0.80Photinia glabra T Animals 26 0.00 0.12 1.00 0.99 1.00 0.94 1.00 0.87Meliosma oldhamii T Animals 31 0.00 0.02 1.00 0.99 1.00 0.99 1.00 0.00Ilex elmerrilliana T Birds 39 0.00 0.12 1.00 0.99 1.00 0.94 1.00 0.87Osmanthus cooperi T or S Animals 60 0.00 0.14 1.00 0.98 1.00 0.93 1.00 0.65Platycarya strobilacea T or S Wind 73 0.00 0.69 1.00 0.99 1.00 0.71 1.00 0.97Magnolia cylindrica T Animals 108 0.00 0.06 1.00 0.99 1.00 0.97 1.00 0.75Rhododendron mariesii S Wind 50 0.00 5.54 1.00 0.99 1.00 0.06 1.00 0.99Manglietia yuyuanensis T Animals 102 0.00 0.12 1.00 0.98 1.00 0.94 1.00 0.74Symplocos sumuntia T Animals 87 0.00 0.26 1.00 0.98 1.00 0.88 1.00 0.87Lindera glauca T or S Animals 90 0.00 0.18 1.00 0.98 1.00 0.91 1.00 0.83Cleyera japonica T or S Birds 53 0.00 0.12 1.00 0.99 1.00 0.94 1.00 0.87Styrax odoratissimus T Animals 37 0.00 0.06 1.00 0.98 1.00 0.97 1.00 0.49Acer cordatum T or S Wind 41 0.00 0.48 1.00 0.88 1.00 0.79 1.00 0.42Post-dispersallimitationbecause seeds are dispersed limited distances from their sourcesand are often dispersed in clumps and because adult trees arethemselves clumped, increasing the number of sites that are veryfar from any sources (Ribbens et al., 1994). Clark et al. (1998)quantify dispersal limitation as1a=n1 source limitationTherefore, dispersal limitation depends not only on dispersalbut also on tree abundance and distribution, and on seed192

90Y. Du et al. / Acta Oecologica 39 (2012) 87e93production (Terborgh et al., 2011). Source limitation depends onseed production per tree and adult abundance, while seed limitationdepends on all of these factors.We used R 2.11 for all data analysis (R Development Core Team,2009). We used seed rain data from June 2006 to Jan 2008 as preicestorm data, and from May 2008 to Dec 2009 as post-ice stormdata. We used Bartlett’s test to compare the total seed number 20months before and after the storm, and found the variances werenot the same. We therefore used a two-sample Wilcoxon test tocompare the difference between the two time periods (monthlysums were the replicates). In order to investigate annual variationin seed production before and after the ice storm, we divided ourdata into different time periods. As there are only twenty months ofseed collection data before the storm, we defined 2006.06 to2007.05 as the 1st year, 2007.02 to 2008.01 as the 2nd year,2008.5e2009.4 as the 3rd year, and 2009.05e2010.04 as the 4thyear. Thus, there are two years of seed rain data both before andafter the storm. We compared the seed production before and afterthe ice storm using the Wilcoxon test. We calculated Simpsondiversity of seeds in traps before and after the storm, and useda paired T test to ascertain whether Simpson diversity increased.The Jaccard similarity of species before and after the storm was alsocompared. Jaccard similarity is calculated as J ¼ c/(a þ b c), wherea is the number of species in seed traps before the storm, b is thenumber of species in seed traps after the ice storm, and c is thenumber of species common to both a and b. In order to see whetherrare species were more likely to fruit after the storm, we useda linear regression analysis. The response variable was the increasein seed number after the ice storm, while the explanatory variablewas the species abundance per hectare. We used Kendall’s rankcorrelation to test whether the change in seed limitation (includingsource limitation and dispersal limitation) is correlated with seedmass. We also used a KruskaleWallis test to determine whetherchanges in seed limitation are influenced by dispersal modes. In ourresults, we identify statistical significance at P 0.05.3. Results3.1. Seed productionThere were 77,541 seeds in total over the twenty months beforeice storm, belonging to 45 species, 37 genera, and 22 families, whileafter the storm there were only 30,995 seeds over the twentymonths, belonging to 53 species, 41 genera, and 23 families. Seedrain differed in species abundance and composition before andafter the ice storm (Fig. 1). The total number of seeds trappedduring the twenty months after the ice storm was significantlylower than before the storm (Wilcoxon test, P ¼ 0.025). The poststormseed collection was less than half of that before the icestorm. Seeds from one species, Eurya muricata, dominated the seedrain after the storm, accounting for more than half of the totalseeds. This means that post-storm seed quantity of species otherthan E. muricata, was only 20% of pre-storm quantity. Seed numbersone year after the ice storm decreased significantly compared toseed numbers in the previous two years (monthly sums were thereplicates; the 3rd year versus the 1st year, P ¼ 0.0029, the 3rd yearversus the 2nd year, P ¼ 0.0193; Fig. 2). In order to explore whetherthe seed production decrease was due to the ice storm rather thaninterannual variation in seed production, we performed a regressionbetween the seed density changes and number of individualdeaths after the storm (dbh 8 cm), and found the regression wassignificant (P ¼ 0.0071; R 2 ¼ 0.212). However, seed production inthe second year after the ice storm recovered to the pre-storm level(monthly sums were replicates; the 4th year versus the 1st year,P ¼ 0.4095; the 4th year versus the 2nd year, P ¼ 0.7125; Fig. 2). Inorder to determine whether the effects of the ice storm weregeneral among seed traps, we also performed analyses with thesums of seeds at each trap as the replicates. Seed number in thethird year was significantly different from the first year (the pairedWilcoxon test; P ¼ 0.000) and the second year (P ¼ 0.000). Therewas no significant difference between seed number in the fourthyear and in the second year (the paired Wilcoxon test; P ¼ 0.6877),which indicates that seedfall had recovered to the pre-storm level.Seed production showed large inter-specific variation inresponse to the ice storm (Table 1; Fig. 3). Some species, such asVaccinium carlesii, Itea oblonga, and Euonymus oblongifolius, did notproduce fruits in the two fruiting seasons following the storm,while some species, such as Fraxinus insularis, Alniphyllum fortunei,and Eurya rubiginosa, showed significantly decreased seedproduction. In contrast, some species, such as Rhododendron mariesii,Platycarya strobilacea, and Magnolia cylindrical, did notproduce fruits/seeds before the storm, but did produce fruits afterthe storm, and some species, such as E. muricata, P. massoniana,Adinandra millettii, and Rhododendron latoucheae, produced someseeds before the storm but produced even more seeds after thestorm. The top six species whose seed production increased notablyfollowing the disturbance were E. muricata, P. massoniana, A. millettii,R. latoucheae, R. mariesii, and Rhododendron ovatum. All ofthese are shrubs or low stature trees except P. massoniana. Five ofthe six are from the Theaceae and Ericaceae families. For E. muricata,the seed number increased from 1616 seeds in all 130 seedtraps to 15,856, about ten times more than before the ice storm. Thefour dominant species in the plot, C. eyrei, S. superba, P. massoniana,and E. muricata, all behaved differently (Fig. 3). Seedfall of E. muricatadecreased in the first year after the ice storm, and increaseddramatically in the second year after the storm (Fig. 3). For P.massoniana, seed production increased in the two years after theice storm. S. superba stopped fruiting in the first year after theNumber of species3027242118151296302006.062006.072006.082006.092006.102006.112006.122007.012007.022007.032007.042007.052007.062007.072007.082007.092007.102007.112007.122008.012008.062008.072008.082008.092008.102008.112008.122009.012009.022009.032009.042009.052009.062009.072009.082009.092008.05MonthNumber of speciesNumber of seedsFig. 1. Variation of seed production and the number of species from June, 2006 to December, 2009.2009.102009.112009.1235000300002500020000150001000050000Number of Seeds193

Y. Du et al. / Acta Oecologica 39 (2012) 87e93 91Seed number800006000040000200000691021953445641 2 3 4Year40986Fig. 2. Temporal pattern of total seed number. Value near each point is seed numbereach year.storm, but began to fruit in the second year (Fig. 3). There wasa significant decrease in seed production for Fagaceae species(Paired T test, P ¼ 0.03). For the six species in Fagaceae, one speciesstopped fruiting and the other five produced far fewer seeds afterthe ice storm. For instance, seed production of the most abundantspecies in the plot, C. eyrei, had little seed production in the twoyears after the storm (Fig. 3).There was negative relationship between species abundanceper hectare and increase in seed number after the ice storm (Fig. 4).Species richness increased from 45 species before the ice storm to53 species after storm. There was a significant increase in Simpsondiversity of seeds in traps twenty months after the storm(t ¼ 9.01, df ¼ 129, P < 0.0001), and the mean increase inSimpson diversity was 0.22. This means that the disturbancecaused by ice storm greatly increased seed diversity. The Jaccardsimilarity of species before and after ice storm was 58%. Theoverlap of common species was moderately strong, with about 30species in common.3.2. Seed limitationWe calculated and compared the seed limitation, source limitation,and dispersal limitation 20 months before and 20 monthsafter the ice storm, and found there were no significant differencesfor seed limitation and its two components (Table 1; seed limitationpaired T test: t ¼ 1.08, df ¼ 61, P ¼ 0.28; source limitation paired Ttest: t ¼ 0.69, df ¼ 61, P ¼ 0.49; dispersal limitation paired T test:t ¼ 0.66, df ¼ 61, P ¼ 0.51). There were no significant differences inseed limitation (Wilcox test, W ¼ 1385.5, P ¼ 0.18) and dispersallimitation (Wilcox test, W ¼ 1178, P ¼ 0.87) before and after thestorm, but there was a significant difference for source limitation(W ¼ 1521, P ¼ 0.02).There were also no significant differences between dispersalmodes for seed limitation (df ¼ 4, P ¼ 0.26), and between dispersalmodes and dispersal limitation (F ¼ 0.91, P ¼ 0.46), but there wasa marginal significant effect of dispersal mode on source limitation(df ¼ 4, F ¼ 2.1344, P ¼ 0.09). Source limitation of rodent dispersedspecies increased after the ice storm.4. Discussion4.1. Seed productionAt the community level, the total seed production decreaseddramatically after the ice storm. However, seed production showedlarge inter-specific variation in response to the ice storm. Manyspecies stopped fruiting or produced only a few fruits in the nextfruiting season, including the dominant species in the plot, S.superba, and all Fagaceae species. The results of our study areconsistent with the results of a study on pre- and post-Hurricanefruit availability in a Puerto Rican forest, in which fruit productionreached its lowest point in October 1989, just after HurricaneHugo (Wunderle, 1999). Whigham et al. (1991) have also analyzedSeed number1600012000800040000768Eurya muricata8484641 2 3 4Year15840Seed number2000150010005000446Pinus massoniana35917081 2 3 4Year1793Schima superbaCastanopsis eyreiSeed number6000045000300001500005986210553058121 2 3 4YearSeed number4000300020001000014634562831 2 3 4Year4Fig. 3. Temporal pattern of seedfall density for four dominant species. Value near each point is seed number each year.194

92Y. Du et al. / Acta Oecologica 39 (2012) 87e93Fig. 4. Regression relationship between species abundance per hectare and increase inseed number after the ice storm.the impact of hurricanes on forests in the northeastern YucatanPeninsula, and found the storm destroyed all arboreal fruit andflowers. Without a control location that didn’t have an ice storm, itis very difficult to determine whether the decrease in seed numberwas due only to the ice storm. However, the 2008 ice stormdestroyed a non-negligible portion of potential parent trees, asdescribed in an inventory of tree damage after the ice storm in theGTS plot in 2008 (Man et al., 2011). Those results showed that theice storm caused severe damage to one third of trees (dbh 8 cm)and relatively slight damage to another third of trees (dbh 8 cm).The significant correlation between seed density changes and thenumber of individual deaths after the storm suggests that the icestorm caused a decrease in seed production. However, we cannotexclude other factors, like interannual variation in seed production,which could also give rise to the low seed abundance pattern in2008. Seed production in the second year after the ice stormrecovered to the pre-storm level. This is in agreement with thetrends in fruit production after disturbance seen in the Puerto Ricanforest mentioned above. Fruit production reached its lowest pointin October 1989, just after Hurricane Hugo, but the number ofspecies bearing fruit increased and peaked in the second year afterthe hurricane (Wunderle, 1999).On the contrary, some species did not produce seeds before thestorm but fruited after the storm, while some species producedmore seeds after the storm than before. For E. muricata, the seednumber increased dramatically. E. muricata dominated the seedrain after the ice storm, accounting for more than half of the totalseeds. This pioneer shrub, which suffered little damage during theice storm because of its low stature and also had high growth ratesas a consequence of light availability, probably reproduced soonafter the storm. The opening of the canopy by ice storm damageapparently favored increased seed production by surviving understoryshrubs, a pattern also noted by Brokaw et al. (2004) inincreased seed production for shrubs after a hurricane. Somespecies may put more resources into seed production aftera disturbance in anticipation of future death. For P. massoniana,seed production increased steadily after the storm. There are twopotential explanations for this trend. First, P. massoniana isa dominant species in the forest, and its large stature prevents theadult trees from breaking or dying due to storm damage. Second, P.massoniana usually grows along ridges; adults could recover andfruit immediately after the ice storm due to beneficial light conditions.Seedfall of S. superba stopped in the first year after the storm,but began again in the second year, while C. eyrei fruited little in thetwo years following the storm. This may be because seed productionof tree species that depend on scatter-hoarding animals fordispersal may be more likely to be influenced by weather. Therewere six species whose seed production increased greatlyfollowing the disturbance; nearly all of them were shrubs or lowstature trees (with the exception of P. massoniana). Five of the sixspecies were from the Theaceae and Ericaceae families. A lowstature may prevent adults from being severely damaged, and thesespecies can grow quickly in open canopy gaps and reproduce morein the next year.The 2008 ice storm increased the size and prevalence of canopygaps and dramatically altered the light environment in the GTS plot(Man et al., 2011). Episodic recruitment into gaps after the stormcould alter density-dependent interactions among seedlings. Theice storm disturbance may lead to weaker negative densitydependenteffects of seedling survival, as extra light could permitseedlings to resist pathogens and thereby increase survival, whichhas significant implications for species coexistence. The storm canalso lead to a shift in species composition of the seedling bank bypermitting the growth of seedlings that were inhibited by low lightin the understory (Uriarte et al., 2005). In addition, extensivedamage to abundant species may provide an opportunity for rarespecies to colonize open sites, thus enhancing species diversity.There was a significant increase in Simpson diversity of seeds intraps after the ice storm. As Simpson diversity is a measure ofspecies evenness and richness (Magurran, 2004), the increase inSimpson index means that the disturbance caused by the ice stormimproved seed richness and/or evenness. One possible explanationfor this result is that one or more abundant species that weredominating the seed rain prior to the storm had reduced seedproduction after the storm (Table 1; Fig. 4; e.g. C. eyrei, S. superba, F.insularis, Daphniphyllum oldhamii), while less abundant speciesbegan to fruit and have more seed production after the storm(Table 1; Fig. 4; Acer cordatum; Lindera glauca; Symplocos sumuntia;Manglietia yuyuanensis). The other possibility is that some speciesthat didn’t fruit before the ice storm may have put all theirresources into producing seeds when conditions became stressful.The opening of gaps in the forest could promote seedling establishment,and the increased seed diversity after the ice storm couldcause potential increased seedling diversity. There was moderatesimilarity between species composition before and after the storm.This means that there was a large variation in seed production.Some species stopped fruiting after the ice storm damage, whilesome species began to produce fruits.4.2. Seed limitationOur study showed that there were no significant differences inseed limitation twenty months before and after the ice storm. Therewas no significant difference in dispersal limitation before and afterthe storm, but there was a significant difference in source limitation.Seed limitation can be caused by source limitation or dispersallimitation; in this study, source limitation was the primary cause ofseed limitation after the ice storm. Dispersal limitation of specieswas not affected by the ice storm. This result may be also be partlycaused by the recovery of seed production to previous levels in thesecond year after the ice storm. On the other hand, some treespecies began to fruit immediately after the disturbance by the icestorm, which acts to buffer seed limitation somewhat.In general, seed limitation and dispersal limitation were notcorrelated with dispersal mode. Source limitation of rodentdispersed species increased after the storm. The seed production oftree species that depend on scatter-hoarding animals for dispersalmay be variable and more likely to be influenced by weather.195

Y. Du et al. / Acta Oecologica 39 (2012) 87e93 935. ConclusionAt the community level, seed production decreased dramaticallyimmediately after the ice storm but recovered to pre-storm levels inthe second year following the ice storm. The disturbance by the icestorm increased seed diversity. There was moderate similarity inspecies composition before and after the storm. The ice stormgreatly increased source limitation. Seed limitation and dispersallimitation were not correlated with dispersal modes; only sourcelimitation of rodent dispersed species increased after the ice storm.Though our data are consistent with the interpretation of the stormwhich caused decreased seed production, we are not able to test forthem really. Thus, we cannot exclude the possibility that we wereobserving the natural fluctuations that were independent of the icestorm. As seedling data were not available, it is not known whichspecies’ seedlings best survived the storm and which grew,survived, and recruited best after the storm. In the future, it isnecessary to investigate the importance of fluctuations in seedarrival in determining seedling dynamics, and whether seedbanking could buffer recruitment against fluctuations in seedproduction caused by a heavy storm. Unusual years may lead tounusual patterns that are important to the long-term dynamics ofthe forest. Continued studies of seed rain and seedling recruitmentare crucial for a more complete understanding of forest dynamics.AcknowledgmentWe would like to give our special thanks to Dr. Joe Wright andDr. Helene Muller-Landau for frequent advice. We thank Zhenxi Lai,and Zhiyong Jiang for help with data collection, Teng Fang and BinChen for assistance with dispersal modes identifications, and JinlongZhang for statistical advice. We would also like to thank AnneBjorkman at the University of British Columbia for her assistancewith English language and grammatical editing of the manuscript.We gratefully acknowledge the support from the AdministrationBureau of the Gutianshan National Nature Reserve. This study wasfinanced by Key Innovation Project of CAS (KZCX2-YW-430).ReferencesBrokaw, N., Fraver, S., Grear, J.S., Thompson, J., Zimmerman, J.K., Waide, R.B.,Everham, E.M., Hubbell, S.P., Foster, R.B., 2004. Disturbance and canopy structurein two tropical forests. In: Losos, E., Condit, R., La Frankie, J. (Eds.), TropicalForest Diversity and Dynamism. Center for Tropical Forest Science, SmithsonianInstitution, Washington, DC, pp. 177e194.Chen, J., Feng, Z., 2002. Study on geographical compositions of seed plant flora inGutianshan Mountain of Zhejiang Province. J. East. China. Norm. Univ (Nat. Sci.)48, 104e111.Clark, J.S., Macklin, E., Wood, L., 1998. Stages and spatial scales of recruitmentlimitation in southern Appalachian forests. Ecol. Monogr. 68, 213e235.Condit, R., 1998. Tropical Forest Census Plots. Springer-Verlag.Corlett, R.T., 2009. The Ecology of Tropical East Asia. Oxford University Press, USA.Downs, A.A., 1938. Glaze damage in the birch-beech-hemlock type of Pennsylvaniaand New York. J. Forest. 36, 63e70.Darwin, A.T., Ladd, D., Galdins, R., Contreras, T.A., Fahrig, L., 2004. Response of forestunderstory vegetation to a major ice storm. J. Torrey. Bot. Soc. 131, 45e52.Du, Y.J., Mi, X.C., Liu, X.J., Chen, L., Ma, K.P., 2009. Seed dispersal phenology anddispersal syndromes in a subtropical broad-leaved forest of China. Forest. Ecol.Manag. 258, 1147e1152.Eriksson, O., Ehrlen, J., 1992. Seed and microsite limitation of recruitment in plantpopulations. Oecologia 91, 360e364.Hu, Z., Yu, M.J., Ding, B.Y., Fang, T., Qian, H., Chen, Q., 2003. Types of evergreenbroadleaved forests and the species diversity in Gutianu Mountain NatureReserve. Chin. J. Appl. Environ. Biol. 9, 341e345.Legendre, P., Mi, X.C., Ren, H.B., Ma, K.P., Yu, M.J., Sun, I.F., He, F.L., 2009. Partitioningbeta diversity in a subtropical broad-leaved forest of China. Ecology 90,663e674.Lomascolo, T., Aide, T.M., 2001. Seed and seedling bank dynamics in secondaryforests following hurricane Georges in Puerto Rico. Caribb. J. Sci. 37, 259e270.Magurran, A.E., 2004. Measuring Biological Diversity. Blackwell Publishing, Oxford,pp. 26e116.Man, X.X., Mi, X.C., Ma, K.P., 2011. Effects of an ice storm on community structure ofan evergreen broad-leaved forest in Gutianshan National Nature Reserves,Zhejiang Province. Biodivers. Sci. 19, 197e205.Mou, P., Warrillow, M.P., 2000. Ice storm damage to a mixed hardwood forest andits impacts on forest regeneration in the ridge and valley region of southwesternVirginia. J. Torrey. Bot. Soc. 127, 66e82.Muller-Landau, H.C., Wright, S.J., Calderon, O., Hubbell, S.P., Foster, R.B., 2002.Assessing recruitment limitation: concepts, methods and case-studies froma tropical forest. In: Levey, D., Silva, W., Galetti, M. (Eds.), Seed Dispersal andFrugivory: Ecology, Evolution and Conservation. CAB International, pp. 35e53.Muller-Landau, H.C., Wright, S.J., Calderon, O., Condit, R., Hubbell, S.P., 2008.Interspecific variation in primary seed dispersal in a tropical forest. J. Ecol. 96,653e667.Nathan, R., Muller-Landau, H.C., 2000. Spatial patterns of seed dispersal, theirdeterminants and consequences for recruitment. Tree 15, 278e285.Nicholas, N.S., Zedaker, S.M., 1989. Ice damage in spruce-fir forests of the BlackMountain, North Carolina. Can. J. Forest Res. 19, 1483e1491.R Development Core Team, 2009. R: A Language and Environment for StatisticalComputing. Available at:. R Foundation for Statistical Computing, Vienna,Austria, ISBN 3-900051-07-0 http://www.R-project.org :(accessed on 25 August2009).Rebertus, A.J., Shifley, A.R., Richards, R.H., Roovers, L.M., 1997. Ice storm damage toan old- growth oak-hickory forest in Missouri. Am. Midl. Nat. 137, 48e61.Rhoades, R.W., 1995. Succession in a mature oak forest in southwest Virginia.Castanea 60, 98e106.Ribbens, E., Silander Jr., J.A., Pacala, S.W., 1994. Seedling recruitment in forest: calibratingmodels to predict patterns of tree seedling dispersion. Ecology 75,1794e1806.Seischab, F.K., Bernard, J.M., Eberle, M.D., 1993. Glaze storm damage to western NewYork forest communities. Bull. Torrey Bot. Club 120, 64e72.Siccama, T.G., Weier, G., Wallace, K., 1976. Glaze damage in a mixed hardwood forestin Connecticut in relation to Vitis infestation. Bull. Torrey Bot. Club 103,180e183.Song, H.G., 2008. Survey of destruction of snow calamity on ecological environmentof South China Nature Reserves. Chin. J. Wildl. 29, 147e149.Stone, R., 2008. Natural disasters-ecologists report huge storm losses in China’sforests. Science 319, 1318e1319.Terborgh, J., Alvarez-Loayza, P., Dexter, K., Cornejo, F., Carrasco, C., 2011. Decomposingdispersal limitation: limits on fecundity or seed distribution? J. Ecol. 99,935e944.Turnbull, L.A., Crawley, M.J., Rees, M., 2000. Are plant populations seed-limited? Areview of seed sowing experiments. Oikos 88, 225e238.Uriarte, M., Canham, C.D., Thompson, J., Zimmerman, J.K., Brokaw, N., 2005. Seedlingrecruitment in a hurricane-driven tropical forest: light limitation, densitydependenceand the spatial distribution of parent trees. J. Ecol. 93, 291e304.Whigham, D.F., Olmstead, I., Cabrera Cano, E., Harmon, M., 1991. The impact ofhurricane Gilbert on trees, litterfall, and woody debris in a dry tropical forest inthe northeastern Yucatan Peninsula. Biotropica 23, 434e441.Wright, S.J., Calderón, O., 1995. Phylogenetic constraints on tropical floweringphenologies. J. Ecol. 83, 937e948.Wright, S.J., Carrasco, C., Calderón, O., Paton, S., 1999. The El Niño southern oscillation,variable fruit production and famine in a tropical forest. Ecology 80,1632e1647.Wunderle Jr., J.M., 1999. Pre- and post-hurricane fruit availability: implications forPuerto Rican Parrots in the Luquillo Mountains. Caribb. J. Sci. 35, 249e264.Yu, M.J., Hu, Z.H., Yu, J.P., Ding, B.Y., Fang, T., 2001. Forest vegetation types inGutianshan Natural Reserve in Zhejiang. J. Zhejiang. Univ (Agric.& Life. Sci.) 27,375e380.Zhang, L.W., Mi, X.C., Shao, H.B., Ma, K.P., 2011. Strong plant-soil associations ina heterogeneous subtropical broad-leaved forest. Plant Soil 347, 211e220.196

Acta Oecologica xxx (2011) 1e9Contents lists available at ScienceDirectActa Oecologicajournal homepage: www.elsevier.com/locate/actoecOriginal articleComparison of seed rain and seed limitation between communityunderstory and gaps in a subtropical evergreen forestYanjun Du a,b , Xiangcheng Mi a , Keping Ma a, *a State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, No.20 Nanxincun, Xiangshan, Beijing 100093, Chinab Graduate University of Chinese Academy of Sciences, Beijing 100049, ChinaarticleinfoabstractArticle history:Received 5 March 2011Accepted 1 June 2011Available online xxxKeywords:Dispersal limitationDispersal modesSeed dispersalSource limitationSpecies coexistenceTreefall gaps have been identified as important sites for plant recruitment. In this study, we compared seedrain between forest gaps and forest interior using 150 seed traps in the understory and 19 traps in gaps ina 24 ha permanent plot of subtropical evergreen broad-leaved forest in East China. We asked how totalseed abundance and seed species richness, the relative representation of different dispersal modes, andseed limitation and its components differed between canopy gaps and the understory. Results showed thatmost of the species found in both the understory and in gaps were animal-dispersal, but most of the seedrain was comprised of wind-dispersed species in both habitats. No significant differences in either densityor diversity of seeds between gaps and non-gap sites were found. Contrary to expectations, winddispersedseeds did not occur significantly more in treefall gaps than in the forest understory. Therewere also no significant differences in seed limitation and its components (source limitation, dispersallimitation) between the understory and gaps. Seed limitation was strong for all but a few of the bestdispersedspecies in both gap and understory seed traps. Source and dispersal limitation showed largeinter-specific variation in both the understory and in gaps. Overall, our results indicate that: (i) gaps mayplay a neutral role in maintaining seed diversity in this subtropical forest; (ii) under strong seed limitationboth in gaps and in the understory, population and community dynamics slows and ecological drift inspecies composition may become a more important determinant of community structure.Ó 2011 Elsevier Masson SAS. All rights reserved.1. IntroductionTreefall gaps have been identified as important sites for plantrecruitment as well as foraging sites for several frugivorous birds(Hartshorn, 1983; Denslow et al., 1986; Levey, 1988; Clark and Clark,1992; Dalling et al., 2002). Gaps are thought to maintain speciesdiversity primarily in two ways (Schnitzer and Carson, 2001). First,they create high light habitats, providing a regeneration niche forpioneer species and species that are intermediately tolerant toshade (Grubb, 1977). Second, species may specialize on and partitionresources along resource gradients that vary greatly from thegap center to the forest interior, thus permitting species coexistence(Schnitzer and Carson, 2001). Seed output was expected to beenhanced in gaps since seeds from understory species coulddisperse into gaps (Schnitzer et al., 2008). As the ‘more individualshypothesis’ (MIH) predicts that richness increases monotonicallywith density (Evans et al., 2005), species diversity tends to be higherin gaps than in the understory. In addition, because frugivorous* Corresponding author. Tel.: þ86 10 62836223; fax: þ86 10 82599518.E-mail address: kpma@ibcas.ac.cn (K. Ma).birds often concentrate their activities in treefall gaps, gaps may actas recruitment foci and receive a disproportionate number of seeds(Hoppes, 1988; Levey, 1988). Coexistence occurs because the superiorcompetitor lacks the ability to colonize all available habitats,leaving space for more less competitive but better-colonizingspecies (Levine and Murrell, 2003; Tilman, 1994). As colonizationis a function of fecundity and dispersal ability, numerous speciescould be expected to coexist through a competition-colonizationtrade-off. Therefore, shade-tolerant tree diversity does not appearto be maintained by gaps, possibly due to a combination of seed andestablishment limitation (Schnitzer and Carson, 2001). However,empirical studies of how seed limitation and its components differbetween gaps and the understory remain scant.Of the few studies that have compared seed rain in the understoryand in gaps, some found that the seeds of wind-dispersedtrees, lianas, and shrubs were more abundant in seed trapsplaced in treefall gaps than those placed in the understory(Augspurger and Franson, 1988; Denslow and Diaz, 1990). However,Jones et al. (2005) studied seed dispersal patterns in a naturalpopulation of the neotropical tree species Jacaranda copaia, andfound no evidence for increased seed number in tree fall gaps1146-609X/$ e see front matter Ó 2011 Elsevier Masson SAS. All rights reserved.doi:10.1016/j.actao.2011.06.001Please cite this article in press as: Du, Y., et al., Comparison of seed rain and seed limitation between community understory and gaps ina subtropical evergreen forest, Acta Oecologica (2011), doi:10.1016/j.actao.2011.06.001197

2Y. Du et al. / Acta Oecologica xxx (2011) 1e9during the two-year study. Loiselle et al. (1996) studied thepatterns of seed rain in recent treefall gaps and paired understorysites in a tropical wet forest in Costa Rica, and found that understorysites received greater seed rain than did gap sites in three offour samples over the course of one year. More empirical studiesare needed to investigate whether gaps help contribute to seeddiversity and abundance at the community level.The degree of discrepancy in seedfall density and diversitybetween gaps and the forest understory will vary with dispersalagent (Schupp and Frost, 1989). Mammals such as monkeys areunlikely to deposit large seeds in gaps. Instead, they drop seedsunder the canopy trees in which they feed (Schupp et al., 1989).Numerous birds and bats dispersing small-seeded plants are morelikely to carry seeds to gaps. This may be because frugivorous birdsare especially active in and around gaps (Malmborg and Willson,1988). Wind may disperse seeds into gaps more efficiently thananimals. Several studies of tropical tree species have shown thatwind-dispersed seeds may disperse disproportionately more intonew and regenerating treefall gaps than to understory sites(Augspurger and Franson, 1988; Denslow and Diaz, 1990; Loiselleet al., 1996). Turbulence created by air flow across the brokencanopy and convectional currents in gaps should increase theprobability of seed deposition in gaps by altering the wind speedand aerodynamic behavior of seeds and by trapping seeds in eddies(Burrows, 1975; Schupp and Frost, 1989; Bohrer et al., 2008).In this study, we compared the seed rain between forest gapsand the forest interior in a subtropical evergreen broad-leavedforest in East China in terms of seed abundance, seed richness,and seed limitation. We asked:(i) How does seed abundance and seed species richness varybetween gaps and the understory?(ii) How does the relative representation of different dispersalmodes vary between gaps and the understory?(iii) How do seed limitation and its components differ betweengaps and the understory?2. Materials and methods2.1. Study siteThe study was conducted in a 24 ha permanent forest plot(29 15.101 0 -29 15.344 0 N, 118 07.010 0 -118 07.400 0 E) in GutianshanNational Nature Reserve (GNNR), Kaihua County, Zhejiang Provincein eastern China. GNNR covers a total area of approximately 8107 ha.The topography is characterized by mountains with steep slopes.The mean annual temperature is 15.3 C. The hottest month is Julywith a mean temperature of 27.9 C, and the coldest is January witha mean temperature of 4.3 C. The mean annual precipitation is1787 mm, most of which falls between March and July. The meanannual number of frost-free days is 250. The dominant vegetationtype in GNNR is subtropical evergreen broad-leaved forest dominatedby Castanopsis spp., Cyclobalanopsis spp. and Schima superba.A total of 1991 vascular plant species, belonging to 244 families and897 genera, were recorded within the entire GNNR (Chen and Feng,2002; Hu et al., 2003).2.2. Methods1998). The elevation range between the highest and lowest pointin the plot was 269 m (from 446 to 715 m). The first tree census wasconducted in 2005. All woody stems 1 cm in DBH were mapped,measured, identified, and tagged (Legendre et al., 2009).2.2.2. Seed collectionWe placed 130 seed traps in the understory of the plot in June2006, and placed another 20 understory traps and 19 gap traps inSeptember 2006 (Fig. 1A, in electronic appendices). A gap wasdefined as a vertical opening in the forest extending through allfoliage levels to within 2 m of the ground (Brokaw,1982). The sizes ofthe 19 gaps ranged from 15 m 2 to 360 m 2 . Gaps were identified in thefirst permanent plot census in 2005. All 19 seed traps were placed inthe center of the gap. Seed rain has been censused weekly since June2006. Each seed trap consists of a square 0.5 m 2 PVC frame supportinga shallow, open-topped, 1 mm nylon mesh bag suspended0.8 m above the ground on four PVC posts following the methods ofWright and Calderon (1995) and Wright et al. (1999). All seeds, fruits,seed-bearing fruit fragments, flowers, capsules, and other reproductiveparts of plants that fell into the traps were identified tospecies and recorded. Fruits were categorized as aborted, immature,damaged, fragments and mature. Because the seed traps werelocated above the ground, they captured fruits and seeds fallingdirectly from trees, as well as those spat or defecated by birds, batsand arboreal mammals; they did not, however, record secondarydispersal by rodents and other terrestrial animals (Muller-Landauet al., 2008). All data presented refer to seed number, based uponeither a count of actual seeds per fruit, or calculated based upon themean number of seeds per fruit. We used seed rain data from January2007 to December 2007.A dispersal mode was assigned to each species based on fruitmorphology and unpublished observations of fruit consumption.Each species assigned to one main dispersal mode: wind, explosive,animal, and weight.2.3. Data analysis2.3.1. Calculation of seed limitationWe calculated measures of seed limitation, source limitationand dispersal limitation (Clark and Macklin, 1998; Nathan andMuller-Landau, 2000) using data on seed arrival into the 150understory seed traps and 19 traps in gaps. Fundamental seedlimitation ¼ 1- a/n, where a is the number of seed traps reached byseeds and n is the total number of seed traps. Source limitation(stochastic) ¼ exp (- s/n), where s is the total number of seedsa=narriving at all traps. Dispersal limitation is 1-1 sourcel limitation .Total species number45362718940202.2.1. Tree plotIn 2005, a permanent plot covering 24 ha (400 600 m) wasestablished within the evergreen broad-leaved forest in GNNR. Theplot was established and data were collected following the protocolof the CTFS (Center for Tropical Forest Science) network (Condit,0understorygapSeed trap siteFig. 1. Total species numbers in the understory and in gaps.Please cite this article in press as: Du, Y., et al., Comparison of seed rain and seed limitation between community understory and gaps ina subtropical evergreen forest, Acta Oecologica (2011), doi:10.1016/j.actao.2011.06.001198

Y. Du et al. / Acta Oecologica xxx (2011) 1e9 32.3.2. Statistical testsWe used R 2.11 for all data analysis (R core development team).As the sample size of gap traps is clearly small (only 19 traps),which affects the power of the analysis, bootstrapping is by far themost rigorous way to calculate significance. Therefore, we usedbootstrapping in R to test for significant differences in total seednumber, species richness, and seed limitation between gaps andthe understory, and to test for significant differences in the numberof wind-dispersed seeds dispersing into the understory and intogaps. We use 10000 bootstrap replicates. A one-way analysis ofvariance (ANOVA) was used to test whether seed density wassignificantly affected by dispersal mode. A paired-t test was used toanalyze whether there was a difference between the number ofwind-dispersed seeds and the number of animal-dispersal seedsboth in the understory and in gaps.3. Results3.1. Seed richnessThe 150 understory seed traps collected 28285 mature seeds,representing 40 species (belonging to 34 genera, and 20 families,Fig. 1). Seed rain in the understory totaled 356 seeds/m 2 in 2007.We collected 3052 mature seeds in nineteen gap seed traps (20species, belonging to 18 genera, and 12 families, Fig. 1), and seeddensity was 315 seeds/m 2 . The number of species captured in eachseed trap averaged 5.8 1.7 and 5.6 1.7 species (mean 1 SD) inunderstory and gap traps, respectively. Species richness did notdiffer significantly between understory and gap traps (P ¼ 0.85)(Fig. 2A). There were also no significant differences in seed numberbetween the understory and tree gaps (P ¼ 0.78) (Fig. 2B). Thespecies with the most abundant seeds both in gaps and in theunderstory were S. superba, Castanopsis eyrei, and Fraxinus insularis.Seeds from these three species accounted for 78 percent of all seedsin understory traps and 88 percent of seeds in gap traps.3.2. Dispersal modesSeed density differed significantly according to dispersal mode(ANOVA, df ¼ 3, P ¼ 0.03). The seed rain was dominated by specieswith zoochorous dispersal in both the understory and in gaps.Animal-dispersed species constituted 78% of species arriving in theunderstory and 74% of species arriving in gaps. The numbers ofanimal-dispersed seeds in the understory was not significantlydifferent from that in gaps (P ¼ 0.82). However, 69% (19636 individualseeds) of the total number of seeds collected from the understorytraps were wind-dispersed. There were 2366 wind-dispersedseeds collected from gap traps, accounting for 77.5% of the totalnumber of seeds in gaps. The number of animal-dispersed seeds andwind-dispersed seeds in the understory was 50 59 seeds and 13989 seeds per trap (mean 1 SD), respectively (Fig. 3). The number ofanimal-dispersed seeds and wind-dispersed seeds in gaps was 3643 seeds and 124 78 seeds (mean 1 SD) per seed trap, respectively(Fig. 3). The number of wind-dispersed seeds was significantly largerthan the number of animal-dispersed seeds both in the understory(P < 0.001) and in gaps (P ¼ 0.001). There was no significant differencein the number of wind-dispersed seeds arriving in understoryand gaps (P ¼ 0.74, Fig. 3). Five species dispersed to more than 69% ofunderstory seed traps. Of these five species, three were winddispersed,one bird-dispersed, and one rodent-dispersed.Fig. 2. (A) Number of species per trap in gaps and in the understory, (B) number ofseeds per trap in gaps and in the understory.3.3. Seed limitationThere were no significant differences in seed limitation betweenthe understory and gaps (P ¼ 0.095, Fig. 4A). Seed limitation wasFig. 3. Number of seeds per trap for anemochorous-dispersed and zoochorousdispersedspecies in the understory and in gaps.Please cite this article in press as: Du, Y., et al., Comparison of seed rain and seed limitation between community understory and gaps ina subtropical evergreen forest, Acta Oecologica (2011), doi:10.1016/j.actao.2011.06.001199

4Y. Du et al. / Acta Oecologica xxx (2011) 1e9total species, source limitation was less than dispersal limitationboth in the understory and in gaps. Source limitation showed largeinter-specific variation, ranging from

Y. Du et al. / Acta Oecologica xxx (2011) 1e9 5across the broken canopy and convectional currents in heated gapsshould increase the probability of seed deposition in gaps byaltering the wind speed and aerodynamic behavior of seeds and bytrapping seeds in eddies (Burrows, 1975; Schupp and Frost, 1989).However, our results showed no significant difference in thenumber of wind-dispersed seeds in the understory and gaps. Onestudy had similar results as ours. Jones et al. (2005) found noevidence for increased seed number in treefall gaps for winddispersedspecies J. copaia. The mechanisms proposed to accountfor more wind-dispersed seeds in gaps include downdrafts or othermicroenvironmental conditions (eddies, turbulence, temperaturedifferences) above tree fall gaps that capture wind-borne seeds andforce them to drop out of the air column and into the gap (Schuppet al., 1989).4.3. Seed limitationThis study represents the first community-level comparisonof seed limitation between gaps and the understory for anyforest. We found no significant differences in seed limitationbetween the understory and gaps. Most tree populations in ourstudy area faced substantial seed limitation in both habitats.Seed limitation was strong for all but a few of the best-dispersedspecies in both gap and understory traps in this study. Only a fewof the tree species showed little seed limitation. S. superba andP. massoniana are quite common and wind-dispersed species,and their seeds reached a large proportion of the plot, indicatinglittle to no seed limitation. C. eyrei is the most abundant speciesin the plot, and its seed production was also very high. As itsseeds are dispersal by small mammals, like rodents, it hadrelatively low seed limitation; what seed limitation did occurwas due entirely to dispersal limitation. F. insularis andT. succedaneum are not very common species, but they producelarge numbers of well-dispersed seeds, and therefore had verylow seed limitation. Under seed limitation, many species fail tohave any viable juveniles at an available site. Thus, some speciesmight ‘win’ recruitment sites by forfeitdthat is, not because theyare the superior competitors under the given environmentalconditions, but because better competitors never reached thatsite (Hurtt and Pacala, 1995; Muller-Landau, 2002; Schupp andMilleron, 2002). Although seed limitation is an equalizingmechanism that can slow competitive exclusion, it is worthpointing out that higher seed limitation (lower fecundity) couldalso reduce diversity in the system as a whole as it could reducethe possibility of specialization on narrow habitat niches (Hurttand Pacala, 1995).We found no significant differences in source limitationbetween understory and gaps sites. Source limitation showed largeinter-specific variation in the understory and in gaps. This reflectsthe great variation in population level seed availability, arising fromdifferences in adult abundance and/or in seed production per adult.The greatest dispersal limitation was present in animal-dispersedtaxa that have abundant but poorly-dispersed seeds: Q. serrata,V. carlesii, A. millettii, V. bracteatum, C. gracilis, I. micrococca, andE. rubiginosa. The wind-dispersed species R. ovatum and A. fortunei,however, showed little source limitation but great dispersal limitation.This is because each fruit of these two species containsnumerous seeds (342 and 28, respectively), but their fruit numberswere only 2 and 12, respectively.The stage at which limitation occurs varies considerablyamong species. Our results implied that recruitment of Q. serrata,Vaccinium carlesii, E. muricata, A. fortunei, V. bracteatum, C. gracilis,and I. micrococca was limited by dispersal rather than by source,while Camellia fraternal was limited by source. This means thatthe relatively high density of seeds does not necessarily meanthat the seed reaches the soil surface. Species with relatively highimportance value, like Loropetalum chinense and Camellia chekiangoleosa,were limited by both seed output and dispersalability. Although the importance value of T. succedaneum is notvery high, its seeds were not limited by either source or dispersal.These differences mean that the limitations imposed by therecruitment process are likely to be an important factor affectingdiversity in forests, patterns of species composition withinstands, and distributions of species across environmental gradients(Clark and Macklin, 1998). Our results showed that therewere no significant differences in dispersal limitation betweenthe understory and treefall gaps. Considering that some largeseededanimal-dispersed taxa, like Quercus, have significantsecondary dispersal, seed traps are likely to greatly underestimateseed dispersal and overestimate dispersal limitation andseed limitation. It should be stated that the differences weobserved between this group and other animal-dispersed speciescould reflect the limitations of the seed trapping methods weused as much or more than the limitations in dispersal ability ofthis group.ConclusionOur study showed that no more wind-dispersed seeds arrivedinto treefall gaps than into non-gap sites. Therefore, gaps may playa neutral role in maintaining seed diversity. As far as we are aware,this is the first community-wide comparison of seed limitationbetween gaps and the understory in a subtropical forest. Dispersallimitation affected most species in both gap and understory habitats;both source limitation and dispersal limitation were speciesspecific.The winning-by-default theory, which posits that notbecause species are the superior competitors under the givenenvironmental conditions, but because better competitors neverestablished in that site, is supported by our study. An enormousnumber of species can be maintained by strong seed limitation inboth gap and understory sites. Moreover, as seed limitationincreases due to increasing species richness or decreasing meanfecundity, more sites might be won by forefeit rather than byabsolute dominance. As a result, population and communitydynamics might slow and ecological drift in species compositionmay become a more important determinant of communitystructure.AcknowledgementWe would like to thank Dr. Helene Muller-Landau for manydiscussions about the calculation of recruitment limitation, andP. Legendre from Université Montréal, Canada, who taught us muchabout statistics and the R language. We will also thank Mr. FangTeng and Mr. Chen Bin for identifying dispersal modes for somespecies, Xiaojuan Liu for collecting seed size data, and two anonymousreviewers who gave valuable comments on the manuscript.We gratefully acknowledge the support from the AdministrationBureau of the Gutianshan National Nature Reserve. We would alsolike to thank Anne Bjorkman at the University of British Columbiafor her assistance with English language and grammatical editing ofthe manuscript. This study was financed by Key Innovation Projectof CAS (KZCX2-YW-430).AppendixFig. A.1 Seed traps in GTS 24 ha permanent plot. The triangleswith black background represent seed traps in gaps. The size ofeach grid is 20 20 m.Please cite this article in press as: Du, Y., et al., Comparison of seed rain and seed limitation between community understory and gaps ina subtropical evergreen forest, Acta Oecologica (2011), doi:10.1016/j.actao.2011.06.001201

6Y. Du et al. / Acta Oecologica xxx (2011) 1e9Table A. 1Species, dispersal modes, seed number, and seed limitation. Dispersers give the major classes of dispersal agents: wind (W), explosive dispersal (B), animals (A), and weight (S).Traps are the number of those 150 traps that captured seeds and/or fruits in understory and of 19 traps in gaps during that period. The numbers in brackets were all values fromgaps. It was first sorted by dispersal mode and then ordered by declining seed number in the understory within each dispersal mode.Tree species Dispersers Traps Seed number Seed limitation Source limitation Dispersal limitationCastanopsis eyrei A 105(13) 3716(416) 0.30(0.32) 0.00(0.00) 0.30(0.32)Eurya muricata A 12(0) 880(0) 0.92(1.00) 0.00(1.00) 0.92(1.00)Quercus serrata A 32(2) 777(37) 0.79(0.89) 0.01(0.14) 0.79(0.88)Toxicodendron succedaneum A 127(12) 727(35) 0.15(0.37) 0.01(0.16) 0.15(0.25)Ilex micrococca A 2(0) 708(0) 0.99(1.00) 0.01(1.00) 0.99(1.00)Cyclobalanopsis gracilis A 12(2) 483(32) 0.92(0.89) 0.04(0.19) 0.97(0.87)Daphniphyllum oldhamii A 72(3) 246(7) 0.52(0.84) 0.19(0.69) 0.40(0.49)Vaccinium bracteatum A 4(0) 199(0) 0.97(1.00) 0.27(1.00) 0.96(1.00)Ternstroemia gymnanthera A 39(5) 145(12) 0.74(0.74) 0.38(0.53) 0.58(0.44)Adinandra millettii A 1(0) 110(0) 0.99(1.00) 0.48(1.00) 0.99(1.00)Euonymus oblongifolius A 3(0) 110(0) 0.98(1.00) 0.48(1.00) 0.96(1.00)Vaccinium carlesii A 2(0) 55(0) 0.99(1.00) 0.69(1.00) 0.96(1.00)Cyclobalanopsis glauca A 11(2) 50(4) 0.93(0.89) 0.72(0.81) 0.74(0.45)Myrica rubra A 8(0) 50(0) 0.95(1.00) 0.72(1.00) 0.81(1.00)Eurya rubiginosa A 1(0) 39(0) 0.99(1.00) 0.77(1.00) 0.97(1.00)Lithocarpus glaber A 5(1) 35(15) 0.97(0.95) 0.79(0.45) 0.84(0.90)Camellia chekiangoleosa A 5(1) 31(10) 0.97(0.95) 0.81(0.59) 0.82(0.87)Vaccinium mandarinorum A 1(0) 26(0) 0.99(1.00) 0.84(1.00) 0.95(1.00)Idesia polycarpa A 1(0) 17(0) 0.99(1.00) 0.89(1.00) 0.94(1.00)Castanopsis fargesii A 1(0) 7(0) 0.99(1.00) 0.95(1.00) 0.85(1.00)Camellia fraterna A 6(1) 6(1) 0.96(0.95) 0.96(0.95) 0.02(-0.03)Sorbus folgneri A 0(1) 0(6) 1.00(0.95) 1.00(0.73) 1.00(0.81)Elaeocarpus decipiens A 2(1) 4(2) 0.99(0.95) 0.97(0.90) 0.49(0.47)Michelia skinneriana A 1(0) 4(0) 0.99(1.00) 0.97(1.00) 0.75(1.00)Styrax dasyanthus A 2(0) 3(0) 0.99(1.00) 0.98(1.00) 0.33(1.00)Malus leiocalyca A 1(0) 3(0) 0.99(1.00) 0.98(1.00) 0.66(1.00)Dalbergia hupeana A 2(1) 3(1) 0.99(0.95) 0.98(0.95) 0.33(-0.03)Euscaphis japonica A 1(0) 3(0) 0.99(1.00) 0.98(1.00) 0.66(1.00)Diospyros glaucifolia A 1(0) 3(0) 0.99(1.00) 0.98(1.00) 0.66(1.00)Albizia kalkora A 2(0) 2(0) 0.99(1.00) 0.98(1.00) 0.01(1.00)Sloanea sinensis A 1(0) 1(0) 0.99(1.00) 0.99(1.00) 0.00(1.00)Rhaphiolepis indica A 1(0) 1(0) 0.99(1.00) 0.99(1.00) 0.00(1.00)Styrax odoratissimus A 0(1) 0(1) 1.00(0.95) 1.00(0.95) 1.00(-0.03)Rhus hypoleuca A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Sapium japonicum A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Symplocos paniculata A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Litsea coreana A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Wikstroemia monnula A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Symplocos anomala A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Viburnum setigerum A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Sassafras tzumu A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Rhamnus crenata A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Syzygium buxifolirm A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Please cite this article in press as: Du, Y., et al., Comparison of seed rain and seed limitation between community understory and gaps ina subtropical evergreen forest, Acta Oecologica (2011), doi:10.1016/j.actao.2011.06.001202

Y. Du et al. / Acta Oecologica xxx (2011) 1e9 7Table A. 1 (continued )Tree species Dispersers Traps Seed number Seed limitation Source limitation Dispersal limitationEvodia faugeaii A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Meliosma flexuosa A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Prunus spinulosa A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Aralia chinensis A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Euonymus myrianthus A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Clerodendrum cyrtophyllum A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Ilex latifolia A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Ilex chinensis A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Premna microphylla A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Castanopsis tibetana A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Diplospora dubia A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Hovenia trichocarpa A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Photinia glabra A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Pittosporum illicioides A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Clerodendrum trichotmum A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Cleyera japonica A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Machilus thunbergii A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Styrax suberrifolia A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Meliosma oldhamii A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Callicarpa rubella A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Ehretia acuminata A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Ilex elmerriliana A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Machilus leptophylla A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Elaeocarpus chinensis A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Litsea elongata A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Magnolia cylindrica A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Machilus grijsii A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Camellia cuspidata A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Tilia endochrysea A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Viburnum sempervirens A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Picrasma quassioides A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Mahonia bealei A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Nyssa sinensis A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Symplocos stellaris A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Callicarpa giraldii A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Carpinus viminea A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Zanthoxylum austrosinense A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Chimonanthus salicifolius A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Diospyros morrisina A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Alangium kurzii A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Ilex pubescens A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Lespedeza Formosa A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Castanopsis carlesii A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Tarenna mollissima A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Ilex litseaefolia A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Vernicia montana A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Osmanthus cooperi A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Machilus pauhoi A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Aidia cochinchinensis A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Phyllanthus glaucus A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Lasianthus japonicus A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Ilex ficoidea A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Euonymus carnosus A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Manglietia yuyuanensis A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Symplocos sumuntia A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Lindera glauca A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Litsea cubeba A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Lindera reflexa A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Photinia serrulata A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Elaeocarpus japonicus A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Dendropanax dentiger A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Symplocos setchuensis A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Dendrobenthamia japonica A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Antidesma japonicum A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Glochidion Puberum A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Ficus erecta A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Ilex rotunda A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Maclura cochinchinensis A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Ilex wilsonii A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Quercus phillyraeoides A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Lindera aggregata A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Eurya loquaiana A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Ilex suaveolens A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Cinnamomum subavenium A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)(continued on next page)Please cite this article in press as: Du, Y., et al., Comparison of seed rain and seed limitation between community understory and gaps ina subtropical evergreen forest, Acta Oecologica (2011), doi:10.1016/j.actao.2011.06.001203

8Y. Du et al. / Acta Oecologica xxx (2011) 1e9Table A. 1 (continued )Tree species Dispersers Traps Seed number Seed limitation Source limitation Dispersal limitationTutcheria microcarpa A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)C.yclobalanopsis gracilis A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Photinia parvifelia A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Viburnum erosum A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Rubus chingii A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Cinnamomum chekiangense A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Prunus schneideriana A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Gardenia jasminoides A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Photinia beauverdiana A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Ardisia crenata A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Celtis biondii A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Callicarpa bodinieri A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Xylosma racemosum A 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Loropetalum chinense B 21(9) 69(67) 0.86(0.53) 0.63(0.03) 0.62(0.51)Corylopsis glandulifera B 5(0) 11(0) 0.97(1.00) 0.93(1.00) 0.53(1.00)Illicium lanceolatum B 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Distylium myricoides B 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Neolitsea aurata S 18(2) 117(40) 0.88(0.89) 0.46(0.12) 0.78(0.88)Choerospondias axillaris S 1(0) 3(0) 0.99(1.00) 0.98(1.00) 0.66(1.00)Schima superba W 150(19) 16789(1919) 0.00(0.00) 0.00(0.00) 0.00(0.00)Fraxinus insularis W 103(17) 1430(302) 0.31(0.63) 0.00(0.00) 0.31(0.11)Rhododendron ovatum W 2(0) 684(0) 0.99(1.00) 0.01(1.00) 0.99(1.00)Pinus massoniana W 110(12) 403(33) 0.27(0.37) 0.07(0.18) 0.21(0.23)Alniphyllum fortunei W 3(1) 335(112) 0.98(0.95) 0.11(0.00) 0.98(0.95)Itea oblonga W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Liquidambar formosana W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Acer olivaceum W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Pertusadina hainanensis W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Platycarya strobilacea W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Sinadina racemosa W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Rhododendron latoucheae W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Pieris japonica W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Rhododendron mariesii W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Lyonia ovalifolia W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Pieris formosa W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Reevesia pycnantha W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Acer wilsonii W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Cunninghamia lanceolata W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Weigela japonica W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Rhododendron simsii W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Hydrangea chinensis W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)Acer cordatum W 0(0) 0(0) 1.00(1.00) 1.00(1.00) 1.00(1.00)ReferencesAugspurger, C.K., Franson, S.E., 1988. Input of wind-dispersed seeds into light-gapsand forest sites in a Neotropical forest. Journal of Tropical Ecology 4, 239e252.Bohrer, G., Katul, G.G., Nathan, R., Walko, R.L., Avissar, R., 2008. Effects of canopycanopy heterogeneity, seed abscission and inertia on wind-driven dispersalkernels of tree seeds. Journal of Ecology 96, 569e580.Brokaw, N.V.L., 1982. The definition of treefall gap and its effect on measures offorest dynamics. Biotropica 11, 158e160.Burrows, F.M., 1975. Wind-Borne seed and fruit movement. New Phytologists 75,405e418.Chen, J., Feng, Z., 2002. Study on geographical compositions of seed plant flora inGutianshan Mountain of Zhejiang Province. Journal of East China NormalUniversity (Natural Science) 48, 104e111.Clark, D.A., Clark, D.B., 1992. Life history diversity of canopy and emergent trees ina Neotropical rain forest. Ecological Monographs 62, 315e344.Clark, J.S., Macklin, E., Wood, L., 1998. Stages and spatial scales of recruitmentlimitation in southern Appalachian forests. Ecological Monographs 68,213e235.Condit, R., 1998. Tropical Forest Census Plots. Springer-Verlag, Berlin, Germany.Dalling, J.W., John, R., 2008. Seed limitation and the coexistence of pioneer treespecies. In: Carson, W.P., Schnitzer, S.A. (Eds.), Tropical Forest CommunityEcology. Wiley-Blackwell, pp. 242e253.Dalling, J.W., Muller-Landau, H.C., Wright, S.J., Hubbell, S.P., 2002. Role of dispersalin the recruitment limitation of Neotropical pioneer species. Journal of Ecology90, 714e727.Denslow, J.S., Diaz, A.E.G., 1990. Seed rain to tree-fall gaps in a Neotropical rainforest.Canadian Journal of Forest Research 20, 642e648.Denslow, J.S., Moermond, T.C., Levey, D.J., 1986. Spatial components of fruit displayin understory trees and shrubs. In: Estrada, A., Fleming, T.H. (Eds.), Frugivoresand Seed Dispersal. Dr. W. Junk Publishers, Dordrecht, The Netherlands, pp.37e44.Evans, K.L., Greenwood, J.J.D., Gaston, K.J., 2005. Dissecting the species-energyrelationship. Proceedings of the Royal Society B 272, 2155e2163.Grubb, P.J., 1977. The maintenance of species-richness in plant communities: theimportance of the regeneration niche. Biological Reviews 52, 107e145.Harms, K.E., Habitat-specialization and seed dispersal-limitation in a Neotropicalforest. (1997). Ph. D Dissertation, Princeton University.Hartshorn, G.S., 1983. Plants: introduction. In: Janzen, D.H. (Ed.), Costa Rican NaturalHistory. University of Chicago Press, Chicago, Illinois, pp. 118e157.Hoppes, W.G., 1988. Seedfall pattern of several species of bird-dispersed plants in anIllinois woodland. Ecology 69, 320e329.Hu, Z., Yu, M.J., Ding, B.Y., Fang, T., Qian, H., Chen, Q., 2003. Types of evergreenbroadleaved forests and the species diversity in Gutian MountainNature Reserve. Chinese Journal of Application Environmental Biology 9,341e345.Hurtt, G.C., Pacala, S.W., 1995. The consequences of recruitment limitation: reconcilingchance, history, and competitive differences between plants. Journal ofTheoretical Biology 176, 1e12.Jones, F.A., Chen, J., Weng, G.J., Hubbell, S.P., 2005. A genetic evaluation of seeddispersal in the Neotropical tree Jacaranda copaia (Bignoniaceae). The AmericanNaturalists 166, 543e555.Legendre, P., Mi, X.C., Ren, H.B., Ma, K.P., Yu, M.J., Sun, I.F., He, F.L., 2009. Partitioningbeta diversity in a subtropical broad-leaved forest of China. Ecology 90,663e674.Levey, D.J., 1988. Spatial and temporal variation in Costa Rican fruit and fruit-eatingbird abundance. Ecological Monographs 58, 251e269.Levine, J.M., Murrell, D.J., 2003. The community-level consequences of seeddispersal patterns. Annual Review of Ecology, Evolution, and Systematics 34,549e574.Loiselle, B.A., Ribbens, E., Vargas, O., 1996. Spatial and temporal variation of seedrain in a tropical lowland wet forest. Biotropica 28, 82e95.Malmborg, P.K., Willson, M.F., 1988. Foraging ecology of avian frugivores and someconsequences for seed dispersal in an Illinois woodlot. Condor 90, 173e186.Please cite this article in press as: Du, Y., et al., Comparison of seed rain and seed limitation between community understory and gaps ina subtropical evergreen forest, Acta Oecologica (2011), doi:10.1016/j.actao.2011.06.001204

Y. Du et al. / Acta Oecologica xxx (2011) 1e9 9Muller-Landau, H.C., Wright, S.J., Calderon, O., Condit, R., Hubbell, S.P., 2008.Interspecific variation in primary seed dispersal in a tropical forest. Journal ofEcology 96, 653e667.Muller-Landau, H.C., Wright, S.J., Calderon, O., Hubbell, S.P., Foster, R.B., 2002. In:Levey, D.J., Silva, W.R., Galetti, M. (Eds.), Seed Dispersal and Frugivory: Ecology,Evolution, and Conservation. CAB International, Oxon, UK, pp. 35e53.Nathan, R., Muller-Landau, H.C., 2000. Spatial patterns of seed dispersal, theirdeterminants and consequences for recruitment. Trends in Ecology andEvolution 15, 278e285.R Development Core Team., 2009. R: A Language and Environment for StatisticalComputing. R Foundation for Statistical Computing, Vienna, Austria. http://www.R-project.orgi.Schnitzer, S.A., Carson, W.P., 2001. Treefall gaps and the maintenance of speciesdiversity in a tropical forest. Ecology 82, 913e919.Schnitzer, S.A., Mascaro, J., Carson, W.P., 2008. Treefall gaps and the maintenance ofplant species diversity in tropical forests. In: Carson, W.P., Schnitzer, S.A. (Eds.),Tropical Forest Community Ecology. Blackwell Publishing Ltd, Oxford, UK, pp.196e209.Schupp, E.W., Frost, E.J., 1989. Differential predation of Welfia Georgii seeds intreefall gaps and the forest understory. Biotropica 21, 200e203.Schupp, E.W., Howe, H.F., Augspurger, C.K., Levey, D.J., 1989. Arrival and survival intropical treefall gaps. Ecology 70, 562e564.Schupp, E.W., Milleron, T., 2002. Dissemination limitation and the origin andmaintenance of species-rich tropical forests. In: Levey, D.J., Silva, W.R.,Galetti, M. (Eds.), Seed Dispersal and Frugivory: Ecology, Evolution, andConservation. CAB International, Oxon, UK, pp. 19e33.Selwyn, M.A., Parthasarathy, N., 2006. Reproductive traits and phenology of plantsin tropical dry evergreen forest on the Coromandel coast of India. Biodiversityand Conservation 15, 3207e3234.Tilman, D., 1994. Competition and biodiversity in spatially structured habitats.Ecology 75, 2e16.Wright, S.J., Calderon, O., 1995. Phylogenetic patterns among tropical floweringphenologies. Journal of Ecology 83, 937e948.Wright, S.J., Carrasco, C., Calderon, O., Paton, S., 1999. The El Nino Southern Oscillationvariable fruit production, and famine in a tropical forest. Ecology 80,1632e1647.Please cite this article in press as: Du, Y., et al., Comparison of seed rain and seed limitation between community understory and gaps ina subtropical evergreen forest, Acta Oecologica (2011), doi:10.1016/j.actao.2011.06.001205

Journal of Theoretical Biology 276 (2011) 99–105Contents lists available at ScienceDirectJournal of Theoretical Biologyjournal homepage: www.elsevier.com/locate/yjtbiA genome evolution-based framework for measures of originality for cladesJianxiong Huang a,b , Xiangcheng Mi a,n , Keping Ma aa State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, The Chinese Academy of Sciences, 20 Nanxingcun, Xiangshan, Beijing 100093, Chinab Graduate School of Chinese Academy of Sciences, Yuquanlu, Beijing 100049, Chinaarticle infoArticle history:Received 3 June 2010Received in revised form25 January 2011Accepted 25 January 2011Available online 1 February 2011Keywords:Genome evolutionPhylogeneticsCharacter distributionOriginalityConservation priorityabstractTo evaluate the originality of a species for determining its conservation priority, most indices use thebranching pattern and the branch length of a phylogenetic tree to represent the diversification patternand the number of characters. One limitation of these indices is their lack of consideration of thedynamic process, such as character changes and distribution along lineages during evolution. In thisstudy, we propose a robust framework incorporating the underlying dynamic processes under aframework of genome evolution to model character changes and distribution along different lineages ina given phylogenetic tree. Our framework provides a more transparent modeling, instead of the simplesurrogates of branching pattern and branch length previously employed. Nonrandom extinction hasbeen found to be clustered within old and species-poor clades, thus it is desirable to combine theevaluation of originality of clades, which will provide a more complete picture and a useful tool forsetting global conservation priorities. Using a phylogenetic tree consisting of 70 species of New Worldterrestrial Carnivora, we demonstrate that the index derived from our framework can discern thedifference in originality of clades. Moreover, we demonstrate that the originality of clades and speciesin a tree changes with different scenarios of dynamic processes, which were neglected by previousindices. We find that the originality of clades should be one of the criteria for setting globalconservation priorities.& 2011 Elsevier Ltd. All rights reserved.1. IntroductionOver the past few decades, we have faced an acceleratedextinction rate equivalent to the mass extinctions of the paleontologicalpast. In response to this rapid change, a range of indicesfor quantifying biodiversity has been advocated to prioritize andmaximize conservation efforts. One way biodiversity can bedefined as character diversity of a sample, where characterincludes both the observed morphological and behavioral traitsand the unobserved physiological traits of species (Faith, 1992;Faith and Walker, 1996). The expected contribution of a species toa set of species is the complementarity of all its characters to theoverall set of characters. The complementarity of a species can bemeasured by the average rarity of its characters, e.g., the originalityof a species (see Pavoine et al., 2005, for a current review). Theconservation of highly original species with a high probability ofextinction will fundamentally maximize this aspect of biodiversity,which has the potential to offer greater ecosystem functioning(Humphries et al., 1995; Crozier, 1997). As species charactersn Corresponding author. Tel.: +86 10 62836507; fax: +86 10 82596146.E-mail addresses: setoutsoft@qq.com (J. Huang),mixiangcheng@ibcas.ac.cn (X. Mi), kpma@ibcas.ac.cn (K. Ma).cannot be counted directly, phylogenetic trees are used as surrogates(Crozier, 1997). Early approaches focused on measurementsof branching pattern of a phylogenetic tree but were unable toincorporate branch length data (May, 1990; Vane-Wright et al.,1991; Nixon and Wheeler, 1992). When complete dated specieslevelphylogenies for large taxonomic groups became available,measurements of species-based originality combined both branchingpattern of a tree and branch length (Pavoine et al., 2005;Crozier, 1992; Redding and Mooers, 2006; Isaac et al., 2007). Theseindices, such as the quadratic entropy (QE) based index (Pavoineet al., 2005) and evolutionary distinctness (Redding and Mooers,2006; Isaac et al., 2007), simply used branch length to representgenetic or character richness. Branch length in ultrametric treesrepresents the time span of the corresponding evolutionary process.These indices are inevitably relatively coarse due to their lackof consideration of the dynamic processes, such as characterchanges and character distribution along different lineages duringevolution (Crozier, 1997; Petrov, 2002; Oliver et al., 2007). A bettermeasure of phylogenetic relationship among species requires amore transparent and explicit process-based model to expresscharacter evolution and the pattern of character distribution alongdifferent lineages in a phylogenetic tree.Nonrandom extinction has been demonstrated in clades(i.e., genera and families) of birds and mammals (Bennett and0022-5193/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.doi:10.1016/j.jtbi.2011.01.037206

100J. Huang et al. / Journal of Theoretical Biology 276 (2011) 99–105Owens, 1997; Russell et al., 1998; Purvis and Hector, 2000;Johnson et al., 2002; Mooers et al., 2009) and angiosperms(Vamosi and Wilson, 2008), and tends to be clustered withinclades that have few species. For example, 22 families of angiospermsand 19 families of mammals are considered to be at highrisk of extinction (Vamosi and Wilson, 2008; IUCN, 2006).Phylogenetically original species of rockfish are morphologicallydistinctive and more likely to be fished (Magnuson-Ford et al.,2009). All members of some species-poor clades may also sharethe same phylogenetically distributed vulnerabilities (Purvis andHector, 2000). These selective extinctions in species-poor cladesgreatly increase the loss of phylogenetic diversity (PD) and morehigher taxonomic diversity relative to the loss expected by chance(Purvis and Hector, 2000; Heard and Mooers, 2000). Phylogeneticallybased measures of conservation value have taken threedifferent approaches. First, they estimate species uniqueness, i.e.,the number of characters possessed by only a particular speciesand not shared with others (Faith, 1992, 2004; Pavoine et al.,2005; Altschul and Lipman, 1990), and calculate it using thelength of the terminal branch of a tip species to where it subtendsto a phylogenetic tree (Faith, 1992, 2004) or using the probabilitythat the species has a unique character (Crozier, 1992). Second,they evaluate the originality of individual species by calculatingthe average rarity of a character and pick out the relativelyisolated species (Redding et al., 2008), such as the QE-based index(Pavoine et al., 2005). Third, they capture information about theamount of unique evolutionary history a species has and theredundancy of its shared evolutionary history, such as the equalsplit(ES) (Redding and Mooers, 2006) and fair proportion (FP)approaches (Isaac et al., 2007; Steel et al., 2007). Although a rangeof indices has been advocated by different researchers, we stillneed to expand these indices to provide information on theoriginality of clades (i.e., the average rarity of the characters ofspecies in clades) in the tree of life. Therefore, it is desirable tocombine the evaluation of the originality of clades in conservationdecisions, which will provide a more complete picture for settingglobal priorities (Vamosi and Wilson, 2008).In this paper, we provide a novel process-based modeling ofcharacter change and distribution based on the framework ofgenome evolution to quantify the originality of a species. In ourframework, we treat a target species as a set of characters andmodel the process of character distribution along differentlineages in a given phylogenetic tree analogous to the behaviorof genome evolution. The applications of the framework areillustrated by proposing a new index—character rarity (CHR)—tofill the aforementioned gap in evaluating the originality of cladesin the tree of life. We evaluate CHR by comparing it with otherexisting indices of originality for a set of 70 species of the NewWorld terrestrial Carnivora. We also demonstrate that the originalityof a species in a tree changes with different scenarios ofdynamic processes, which were neglected by previous indices.2. Model2.1. Genome evolution modelCurrent genome studies have demonstrated that genomecomplexity, i.e., genomic size (Petrov, 2002; Oliver et al., 2007)and number of genes (Crozier, 1997; Lynch and Conery, 2003),increases during evolution over long time scales. Genome sizeevolution is governed by the rates of DNA insertion and deletion;the genome size usually remains relatively constant due to theapproximate equilibrium of sequence insertion and deletion(Petrov, 2002). Once a large insertion occurs and is fixed, theold equilibrium of genome size is broken and a new equilibriumFig. 1. Genome evolution model. An evolutionary process of a species is made upof two sub-processes and three stages. Stage I represents the composition ofgenome before the evolutionary process. After the sub-process of insertion ofnongenic sequence, the process reaches stage II, in which the size of insertednongenic sequence should be enough to break the balance of the old genome size.The species evolves to stage III through a whole evolutionary process, in whichgenic and nongenic sequences have a chance to mutate to a new functionalsequence.around an increased genome size will be established (Fig. 1). Thelarge insertions of nongenic (non-functional) DNA caused byreplication errors depend strongly on the initial genome sizeassociated with the number of initial transposon copies andpotential target insertion sites (Zhu et al., 2003; Kazazian,2004). Oliver et al. (2007) analyzed the genome sizes of 168species belonging to 20 eukaryotic taxonomic groups and foundstrong evidence that the rate of genome size evolution is positivelyrelated to the initial genome size.In a genome, some sequences are functional (genic DNA),whereas some are non-functional (nongenic DNA). When a newbalanceofgenomesizeisattained,weuseparametera to representthe ratio of nongenic DNA to genic DNA. Here, 1/a can be explainedas the maximal capability of genic DNA for maintaining nongenicDNA because the insertion of nongenic DNA adds energetic costsand has longer replication time, as well as other deleteriousphenotypic effects (Petrov, 2002). Therefore, it is reasonable toassume that a will remain constant during the entire evolutionaryprocess for all species in a given phylogenetic tree.During the subsequent evolution, we assume that both genesequences and nongenic DNA mutate following the rule of neutralmutation. We also presume that nongenic DNA has a mutationrate of m per unit evolutionary interval per unit sequence lengthat the initial genome size, while genic DNA has a relativemutation rate of b to m. This presumption is based on the factthat the mutation of genic DNA is strongly restricted by naturalselection and other associated genes during evolution becausemost of the mutation of genic DNA is deleterious. Therefore,b should theoretically be larger than 0 and less than 1. Thenumber of new genes created by gene duplication and lateralgene transfer, among others (Long et al., 2003), is positivelyrelated to the number of initial transposon copies and potentialtarget insertion sites in nongenic DNA (Long et al., 2003). Thesenew genes are also correlated with the length of nongenic DNA,which is similar to the new genes mutated from nongenic DNA.Thus, for simplicity of simulation, we treat all new genes from amutation process of nongenic DNA.2.2. Modeling of character distributionFollowing the above-described behavior of genome evolution,we define the number of characters as the number of genes incoding regions found within the DNA of a species. We integratecharacter changes, including character inheritance, mutation, andincrements, and character distribution along lineages into adynamic process modeling framework of genome evolution.207

J. Huang et al. / Journal of Theoretical Biology 276 (2011) 99–105 101For the convenience of describing evolutionary processes, in thispaper, evolutionary stage refers to the internodal distance on adirect path from root to tip. For example, species A in Fig. 2undergoes three evolutionary stages in the hypothetical tree.One character may be changed in state during an evolutionarystage, the probability of which can be expressed as the mutationrate. As Drake (1991) found, the average mutation rate per basepair per unit time is inversely proportional to the genome size,and the mutation rate of the whole genome is constant. Accordingto this rule, the product of m (mutation rate) and the genomelength (the length of gene sequence and nongenic DNA) areconstant during evolution. For example, species A in Fig. 2 undergoesthree hypothetical evolutionary stages and has time spansof T 1 , T 2 , and T 3 , respectively. Let TC i denote the number ofcharacters at the beginning of the ith evolutionary stage andTC i (1+a) denote genome length at the ith stage (a is the ratioof nongenic sequence to genic sequence as defined in the previoussection). A constant genome mutation rate means thatm 1 TC 1 (1+a)¼m 2 TC 2 (1+a)¼m 3 TC 3 (1+a). As shown inTable 1, if we assume the total number of characters at stage1 is 1 unit and the mutation rate of nongenic DNA is m, thenm i ¼m/TC i . Following the molecular clock theory that the numberof expected changes of a gene is linearly correlated with time,then the character mutation rate at the ith evolutionary stage isV i ¼b m i T i , whereas the inherited rate of the characters at theith evolutionary stage is I i ¼1 V i , where T i is the time of the ithevolutionary stage. We can divide the total characters of a targetspecies into three groups:(1) Derived characters (MC i ), which are new character statesmutated from the characters of an immediate ancestor duringthe ith evolutionary stage, can be calculated as TC i V i .(2) Inherited characters, which are directly inherited from aspecies’ immediate ancestor and have no mutation duringthe ith evolutionary stage, can be expressed as TC i I i .(3) New characters (IC i ), which are new characters acquired fromthe mutation of nongenic DNA. The number of new charactersduring the ith evolutionary stage can be expressed asTC i a m i T i , where TC i a is the length of nongenic DNA.Derived and new characters of a species are novel and are notyet shared by its sister species. For example, the novel characterspossessed by Z in Fig. 2 are not shared by its sister species C. Wefocus on novel characters (CN i ), which are the sum of MC i and IC iat each evolutionary stage, and their inheritance pattern on aphylogenetic tree. Table 1 illustrates how novel characters ofdifferent ancestors are inherited by target species A in Fig. 2during evolution. We assume that the first ancestor of species Anear the root possesses one unit of characters (CN 0 ¼CT 0 ¼1) andthat all of them are novel characters. These characters possessedby species A can be calculated as CN 0 I 1 I 2 I 3 because CN 0 undergoesthree hypothetical evolutionary stages. Similarly, the novel charactersof Y possessed by species A are CN 1 I 2 I 3 , the novel charactersof Z possessed by species A are CN 2 I 3 , and the novel characters ofspecies A are CN 3 . Finally, the characters of species A consist ofthese parts: CN 0 I 1 I 2 I 3 +CN 1 I 2 I 3 +CN 2 I 3 +CN 3 (Fig. 2). Hence, weknow how the characters of all ancestors are distributed alongdifferent lineages, which provide the basis for estimation of theaverage rarity of the target species.2.3. Application of modelFig. 2. A hypothetical phylogenetic tree illustrating how ancestors’ charactersaccumulated in a target species. A–C are different tip species in the tree, X and Yare hypothetical ultimate common ancestors of all species, and Z is a hypotheticalancestor of A and B. T 1 –T 3 are the evolution intervals of each divergence beforespecies A. CN i is the number of novel characters at the ith evolutionary stage, I i isthe inheritance rate of the novel characters at the ith evolutionary stage to thenext descents, and i refers to the ith evolutionary stage, i.e., the evolutionaryinterval between the ith and the (i-1)th internal nodes on a direct path from rootto tip.Under the framework of explicit process-based modeling ofgenome evolution, we suggest a new index, CHR, to measure theoriginality of clades.Pavoine et al. (2005) defined the originality of a species as theaverage rarity of characters, and we defer to this definition andquantify CHR under our framework. We set the rarity of acharacter of a target species to 1 when the character is not sharedby any other sister species or to 0 when the character is shared byall other sister species in a phylogenetic tree. According to thedefinition, supposing a target species has S 1 sister species in aphylogenetic tree and a character of the target species hasprobability P¼(P 1 , P 2 , P 3 , y,P S 1 ) to be shared by its sisterspecies, where P 1 , P 2 , P 3 , y,P S 1 is the probability that acharacter is shared by its sister species, thus the character willTable 1Illustration of character decomposition and species A in Fig. 1 as target species in an ultrametric phylogenetic tree.Evolutionary stage E 0 E 1 E 2 E 3Evolutionary interval (T i ) T 1 T 2 T 3Mutation rate of nongenic DNA (m i ) m (m/TC 2 ) (m/TC 3 )Mutation rate of TC i (V i ) V 1 ¼bmT 1 V 2 ¼bm 2 T 2 V 3 ¼bm 3 T 3Inheritance rate of CN i to the next descents (I i ) I 1 ¼1 V 1 I 2 ¼1 V 2 I 3 ¼1 V 3Inheritance rate of CN i to species A (P i ) P 0 ¼I 1 I 2 I 3 P 1 ¼I 2 I 3 P 2 ¼I 3 P 3 ¼1The number of total characters (TC i ) TC 0 ¼1 TC 1 ¼TC 0 TC 2 ¼ TC 1 þTC 1 am 2 T 2 TC 3 ¼ TC 2 þTC 2 am 3 T 3The number of novel characters (CN i ) CN 0 ¼1 CN 1 ¼TC 1 V 1 +TC 1 amT 1 CN 2 ¼ TC 2 V 2 þTC 2 am 2 T 2 CN 3 ¼ TC 3 V 3 þTC 3 am 3 T 3The number of characters of CN i possessed by species A CN 0 P 0 CN 1 P 1 CN 2 P 2 CN 3i: refers to the ith evolutionary stage, i.e., the evolutionary interval between the ith and the (i-1)th internal nodes on a direct path from root to tip; m i : the mutation rate ofnongenic DNA at the ith evolutionary stage as defined in Section 2.2; V i : mutation rate of total characters at the ith evolutionary stage; I i : Inheritance rate of novelcharacters at the ith evolutionary stage to the next descents; P i : Inheritance rate of novel characters at the ith evolutionary stage to species A; TC i : the number of totalcharacter at the ith evolutionary stage; CN i : the number of novel characters at the ith evolutionary stage.208

102J. Huang et al. / Journal of Theoretical Biology 276 (2011) 99–105have an expected rarity of 1 S S 1i ¼ 1 P i=ðS 1Þ. IfP 1 , P 2 , P 3 , y, P S 1are all 0, the expected rarity of the character is at the maximum of1. Otherwise, if P 1 ,P 2 ,P 3 , y,P S 1 are all 1, then the expectedrarity of the character will be at the minimum of 0.Therefore, in our framework, the character rarity of a particularspecies can be expressed as follows:CHR ¼ XnCN i P i ½1 XS 1P ij =ðS 1ÞŠ ð1Þi ¼ 0j ¼ 1where S is the number of species in a phylogenetic tree, n is thenumber of evolutionary stages that a target species undergoes,i¼0 indicates the initial stage, P i is defined as in Table 1, P ij is theinheritance rate of sister species j at the ith evolutionary stage,that is, p ij ¼ P n k ¼ i I k for species j, P ij is the probability of sisterspecies sharing the new characters at the ith evolutionary stagewith the target species, and I k is the inheritance rate of novelcharacters at kth stage to the next descendents (defined asin Table 1). P ij is 0 when species j does not share a commonancestor with the target species at or after evolutionary stage i.When b¼0, then P i ¼1 and S S 1j ¼ 1 P ij is simply equal to thenumber of sister species sharing a common ancestor at the ithstage with the target species (which we call SP i ). CHR can then besimplified as follows:CHR ¼ XnCN i ½1 SP i =ðS 1ÞŠ ¼ XnCN i ðS 1 SP i Þ=ðS 1Þ ð2Þi ¼ 0i ¼ 0When the phylogenetic tree is ultrametric or when the leaves ofthe tree are equidistant from the root and b¼0, CHR is equal to themean phylogenetic distance between the target species and the otherspecies, i.e., the taxonomic distinctiveness (TD) proposed by Ricotta(2004). For example, the CHR value of species A in the phylogenetictree in Fig. 2 is T 1 [(3 1 2)/(3 1)]+T 2 [(3 1 1)/(3 1)]+T 3 [(3 1 0)/(3 1)]¼T 2 /2+T 3 , that is, the mean phylogenetic distancebetween species A and B and between species A and C is[T 3 +(T 3 +T 2 )]/2¼CHR. TD has not been applied to conservationprioritization in previous studies. Since our framework is flexible,we can also quantify species uniqueness under this framework(see Appendix B).2.4. Indices and dataThe QE-based index (Pavoine et al., 2005) is different fromearlier indices measuring branching pattern because it incorporatesbranch length into measures by using frequency distributionsto maximize the expected number of characters shared bytwo randomly chosen species in a phylogenetic tree. Pavoine et al.(2005) found that their QE-based index and four other previousones, such as May’s (1990) index, are highly correlated. Therefore,this study only compares CHR with one branching pattern-basedindex, Nixon and Wheeler’s (1992) weighted index (NWW). Tomeasure the relative contributions of each species to phylogeneticdiversity (Faith, 1992; Crozier, 1997), Redding and Mooers’ (2006)ES approach equally apportions branch length among daughterclades, while the FP approach of Isaac et al. (2007) equally dividesthe branch length among descendent species. The comparisonbetween CHR and these indices of known properties will facilitatethe understanding of characteristics of CHR.We consider three properties in assessing the robustness ofCHR: (1) the capacity to discern the difference in the originality ofdifferent clades, (2) the sensitivity to species richness in clades,and (3) the capacity to discern the difference in the originality ofindividual species. We evaluate CHR by comparing its performancewith existing indices of originality: QE-based, NWW, FP,ES, TD, and species uniqueness (SU). We conduct this comparisonusing a ultrametric phylogenetic tree comprising a total of 70species of extant terrestrial New World Carnivora developedby Diniz and Torres (2002). The tree comprising 70 extant speciesof Carnivora was also constructed from the phylogeny of 271extant species of Carnivora by Bininda-Emonds et al. (1999). Weemploy the tree of 70 species as a methodological illustration andexclude improvements made by Johnson et al. (2006) and others.The tree includes 12 Felidae, 14 Canidae, 3 Ursidae, 13 Procyonidae,and 28 Mustelidae. All statistical analyses were performedusing R 2.9.0 (R Development Core Team, 2009) with the ‘‘ade4’’package (Chessel et al., 2004). The phylogenetic tree data used inthis study are also available in the ade4 package.3. ResultsWhen comparing CHR with other previous indices, we use therelative originality (l i ¼ l i =ðS s i ¼ 1 l iÞ, where s is the number ofspecies in the tree) to facilitate the comparison among indices. CHRhas three parameters: a, b,andm. To compare the behavior of CHR,we set a as 25, which means the length of nongenic DNA is 25times the length of gene sequences. We assume that nongenic DNAhas a mutation rate of 0.5 during the entire evolution process of thetree and obtain the value of m. We also set b to 0.5, whichrepresents a moderate mutation rate of gene sequences.As shown in Table 2, CHR and TD display the same rank oforiginality. Therefore, we only compare the performance of one ofthese similar indices in this study (Fig. 3). We can divide theindices measuring originality in Table 2 into three groups.(1) Group A includes only a QE-based index. As shown in Table 2and Fig. 3, the QE-based index is highly correlated with otherindices measuring originality, including FP, ES, and CHR, but isrelatively less correlated with SU. This suggests that theQE-based index is more sensitive to the originality of individualspecies, but less sensitive to species uniqueness.(2) Group B includes ES and FP indices. ES and ED are all closelycorrelated with SU and the QE-based index, as shownin Table 2 and Fig. 3. This suggests that ES and FP are highlysensitive to both originality and species uniqueness. As seen inEq. (2), CHR shares similar statistical properties with FP and ES;they all consider redundancy by weighting the branch lengthat different positions. The weight of CHR assigned to branchlength approximates (S 1 SP i )/(S 1), which is much higherthan the weight of FP and ES assigned to branch length, 1/SP i(S and SP i are the same as in Eq. (1)). Therefore, CHR stronglydepends on the branches of earlier divergence, whereas FP andES are highly correlated with terminal branches.(3) Group C comprises indices, including CHR and NWW, whichare highly correlated with the QE-based index, but negativelycorrelated with SU (Table 2, Fig. 3). This suggests that CHRand NWW are more sensitive to the originality of clades andare less sensitive to species uniqueness. CHR is roughlyconcordant with NWW (Table 2, Fig. 3).CHR is highly correlated with NWW, which measures theoriginality of clades, and the QE-based index, which measuresthe originality of individual species. This suggests that CHRintegrates the properties of the QE-based and NWW indices. Webelieve that CHR is more appropriate for discerning the variationof originality of clades based on the following aspects:(1) CHR can discern the difference in originality between differentclades and weighs the originality near the root differentlythan that near the tip, which is clearly illustrated in Fig. 3.This property is due to the relatively higher weighting given209

J. Huang et al. / Journal of Theoretical Biology 276 (2011) 99–105 103Table 2Correlation coefficients between CHR and seven other indices—NWW (Nixon and Wheeler, 1992), QE-based index (Pavione et al., 2005), TD (Ricotta, 2004), CHR, FP(Isaac et al., 2007), ES (Redding and Mooers, 2006), and SU (Faith, 1992, 2004)—and b for the ratio of the mutation rate of gene sequence to nongenic DNA sequence.NWW QE-based TD CR FP ES SUQE-based 0.6600TD 0.8519 0.6396CR 0.8428 0.6606 0.9989FP 0.0292 0.5915 0.0876 0.121ES 0.1390 0.7020 0.1086 0.141 0.9419SU 0.1698 0.2932 0.1246 0.0999 0.8745 0.8032b¼0.0 0.8519 0.6396 1.0000 0.9989 0.0876 0.1086 0.1246b¼0.2 0.8484 0.6485 0.9998 0.9996 0.1014 0.1220 0.1144b¼0.4 0.8443 0.6578 0.9992 1.0000 0.1165 0.1366 0.1033b¼0.6 0.8397 0.6675 0.9981 0.9999 0.1328 0.1523 0.0913b¼0.8 0.8345 0.6777 0.9964 0.9992 0.1506 0.1693 0.0781b¼1.0 0.8285 0.6883 0.9939 0.9980 0.1699 0.1877 0.0637Fig. 3. An ultrametric phylogenetic tree illustrating the measures of species originality for 70 species of Carnivora according to the following indices: NWW (Nixon andWheeler, 1992), QE-based index (Pavione et al., 2005), CHR, FP (Isaac et al., 2007), ES (Redding and Mooers, 2006), and SU (Faith, 1992, 2004). The indices are indicatedwith Cleveland’s dot plots (Cleveland, 1994) and presented as centralized percentages.to branches near the root by (S 1 SP i )/(S 1) in CHR. Forexample, in Felidae (12 species), the branch length near theroot is 37.6 million years (MYs) and CHR apportions approximately37.6 (70 1 12)/69¼31.06 to all 12 species,whereas FP and ES assign close to 37.6/12¼3.13 to all species.Therefore, CHR allocates the highest rank of originality tospecies of Felidae. Similar to the QE-based index, CHR canidentify the ranks of originality of clades. For example, itcorrectly gives the highest weights to species of Felidae,Ursidae, and Canidae, which are the three most divergentclades on the tree (Fig. 3). CHR also assigns the greatestoriginality to Puma concolor and Herpailurus yaguaroundibecause these two species are the most divergent in Felidaeand also the most isolated clades in the tree.210

104J. Huang et al. / Journal of Theoretical Biology 276 (2011) 99–105(2) CHR is sensitive to species richness in clades, particularlywhen clades are very small. For example, the weight assignedto Felidae, Canidae, Procyonidae, and Mustelidae rangesfrom (70 1 28)/69 to (70 1 12)/69 (i.e., 0.586–0.786),although the clade size varies from 12 species (Felidae) to28 (Mustelidae), with the weight CHR assigned to Ursidaebeing (70 1 3)/69¼0.957. Therefore, CHR appoints a higherrank of originality to Ursidae than Canidae, although the latteris more isolated than the former (Fig. 3).(3) CHR can also precisely interpret the originality of a tipspecies. Although the behavior of CHR is similar to NWW,the latter measures originality depending on a clade’s size,whereas the former assigns higher weights to branches ofearlier divergence. In Fig. 3, NWW can hardly distinguish theoriginality between species from the same clade. CHR can alsoclearly distinguish the originality of species from the sameclade, rather than a large clade. As an example, in the subtreewith three groups—group 1 includes Leopardus wiedii andLeopardus pardalis, group 2 contains Oreailurus jacobita andOncifelis colocolo, and group 3 contains Oncifelis guigna,Oncifelis geoffroyi, and Leopardus tigrinus—CHR, ES, and theQE-based index weigh these species differently in terms oforiginality than NWW and FP. Disregarding the time ofdivergence, the indices of NWW and FP rank the originalityof three groups as group 14group 24group 3 due to theearlier divergence of group 1 than that of group 2 and group 3,and group 3 having more species than group 2. Instead,considering the time of divergence of the three groups as0.3 MYs for group 1, 1.9 MYs for group 2, and 3.2 MYs forgroup 3, the QE-based, ES, and CHR indices rank the originalityof the three groups as group 34group 24group 1.However, the three groups of species have much smallerdifferences in originality as measured by CHR than by the QEbasedindex. CHR is clearly an integration of the behaviors ofNWW and the QE-based index when assigning the weights oforiginality to species.3.1. Sensitivity analysisAlthough our model is based on the mechanistic processes ofgenome evolution, it is difficult for us to assign precise values tothe three parameters a, b, and m according to the current theoryof genome evolution. To increase the confidence of the mainresults presented above, we also assess the sensitivity of CHR tothese three parameters when varying one of the parameters andfixing the other two parameters at a moderate value. On the onehand, when we fix b at 0.5 and vary a and m, the value ofcharacter rarity for 70 species of Carnivora is highly correlatedwith the value of character rarity when a is fixed at 25(i.e., assuming that 96.15% DNA of the ancestor of Carnivora isnongenic) and m is fixed at the value of the mutation rate ofnongenic DNA at 0.5 (see Table 1 in Appendix A). This indicatesthat CHR is insensitive to the variation of a and m, but sensitive tothe effect of varying b (Table 2). On the other hand, CHR values ofdifferent species vary with b, as shown in Table 2. CHR of eachspecies varies in a larger range, but its relative value has a lowvariation range with increasing b and is still highly correlatedwith the other indices (Table 2). This result indicates that theoriginality of clades and species changes with different evolutionarydynamic processes.4. DiscussionTo our knowledge, our study is the first of this type to assessoriginality by using explicit modeling of character distributionbased on genome evolution. Our framework provides a mechanisticmodeling of character distribution among the differentclades in a phylogenetic tree. When b¼0, CHR is equal to thetaxonomic distinctness proposed by Ricotta (2004), indicatingthat our framework offers a biologically meaningful interpretationof a previous index (see also Appendix B). Under our framework,we assess the originality of clades and provide a new toolfor setting global conservation priorities.CHR provides a useful tool for measuring the originality ofclades. The QE-based index tends to rank the originality ofindividual species (Pavoine et al., 2005), whereas CHR apportionshigher weight to branches near the root and behaves similarly toNWW (r¼0.8302–0.8519) (Nixon and Wheeler, 1992; Krajewski,1994). ES and FP rely on the originality of a species when a targetspecies is relatively isolated from other species in the tree, butthey also strongly depend on species uniqueness when a targetspecies is not very isolated from other species in the tree, as foundby Isaac et al. (2007). Thus, ES and FP can obtain the highestamount of phylogenetic diversity in a conservation prioritysetting (Redding and Mooers, 2006; Isaac et al., 2007), but theycapture the uniqueness more than the originality of a species(Isaac et al., 2007). Species uniqueness focuses on endemism ofcharacters and contains only part of the information a targetspecies contributes to diversity (Barker, 2002). The behavior ofCHR is an integration of the QE-based index and NWW, and theintegration provides a more balanced ranking for priority decisions.Unlike the QE-based index, the behavior of CHR is stronglydependent on earlier divergent branches and is sensitive to thespecies richness of clades. This is more accordant with theprinciple of priority setting, i.e., assigning higher priority toisolated, species-poor clades that have distinctive traits, sharesome phylogenetically distributed vulnerability, and are extinction-prone.Our sensitivity analysis indicates that the originality of cladesand species in a tree changes under different scenarios of dynamicprocesses, which were not taken into account by previous indices.The consistency of results with varying a and m increases theconfidence that the behavior of CHR is relatively invariant againsta and m. Therefore, although we set moderate values for a, b, andm, the behavior of CHR indicates that our evaluation of thebehavior of CHR and its comparison with other indices measuringoriginality in Fig. 3 are robust. On the other hand, the originalityof species and clades changes with varying b (Table 2). This resultindicates that dynamic processes should be taken into account forthe measure of the originality of species and clades. One potentialcaveat of our study is that there are three free parameters to bedetermined before its application and that it is difficult todetermine these parameters according to the present knowledgeof genome study. Fortunately, these parameters are biologicallymeaningful; therefore, we can give them reasonable estimatedvalues even if we do not know their actual value. For example, weassume that 96.15% of the DNA of the ancestor of Carnivora isnongenic and thus fix a at 25 in this study. On the other hand, oursensitivity analysis suggests that CHR is insensitive to the effectsof varying a and m. We can also easily predict the behavior of CHRwhen varying b because the value of CHR monotonically changeswith increasing b, as shown in Table 2. Another potential caveat isthe assumption that the rate of character change is even acrossthe entire phylogenetic tree. However, rates of phenotypic changeare known to vary throughout the tree of life (Ricklefs, 2007).Further work should integrate this interesting parameter intothe model.In summary, our study represents an initial step towarddeveloping more biologically-derived criteria for preferentialconservation based on modeling the dynamic process of characterchanges and distribution along different lineages. We also highlight211

J. Huang et al. / Journal of Theoretical Biology 276 (2011) 99–105 105that the originality of clades and species in a tree may change withdifferent scenarios of dynamic processes. Faced with the ‘‘the agonyof choice’’ in conservation priority settings, indices for the evaluationof species-based originality, uniqueness, and contribution of aspecies to the total PD have been proposed for different conservationpurposes (Faith, 1992; Pavoine et al., 2005; Vane-Wright et al.,1991; Redding and Mooers, 2006; Isaac et al., 2007). In this study,CHR was demonstrated to be closely allied to ED, FP, and QE-basedindices, and these indices were demonstrated to capture PD quitewell (Pavoine et al., 2005; Redding et al., 2008).WepredictthatCHRcaptures the ‘‘feature diversity’’ of clades and species well based onthe biologically meaningful optimization in our modeling framework,but this needs to be tested in further work. CHR should be oneof the criteria used to select evolutionarily distinct species forpreferential conservation because it is based on sensitivity-tested,biologically-derived parameters, thus providing a solid scientificfoundation to this form of conservation triage.AcknowledgementsThe authors are grateful to Fangliang He and Xinsheng Hufor their constructive comments on our manuscript. We thankA. Mooers, D.W. Redding, R.H. Crozier, and anonymous reviewersfor their insightful comments. We also thank Anne Bjorkman andThomas New for their assistance with English language andgrammatical editing of the manuscript. This work was financiallysupported by Key Innovation Project of CAS (KZCX2-YW-430).Appendix A. Supplementary materialSupplementary data associated with this article can be foundin the online version at doi:10.1016/j.jtbi.2011.01.037.ReferencesAltschul, S.F., Lipman, D.J., 1990. Equal animals. Nature 348, 493–494.Bennett, P.M., Owens, I.P.F, 1997. Variation in extinction risk among birds: chanceor evolutionary predisposition? Proc. R. Soc. Lond. B–Biol. Sci. 264, 401–408.Bininda-Emonds, O.R.P., Gittleman, J.L., Purvis, A., 1999. Building large trees bycombining phylogenetic information: a complete phylogeny of the extantCarnivora (Mammalia). Biol. Rev. 74, 143–175.Barker, G.M., 2002. Phylogenetic diversity: a quantitative framework for measurementof priority and achievement in biodiversity conservation. Biol. J. Linn.Soc. 76, 165–194.Cleveland, W.S., 1994. The Elements of Graphing Data. AT & T Bell Laboratories,Murray Hill, NJ.Crozier, R.H., 1997. Preserving the information content of species: genetic diversity,phylogeny, and conservation worth. Annu. Rev. Ecol. Syst. 28, 243–268.Crozier, R.H., 1992. Genetic diversity and the agony of choice. Biol. Conserv. 61,11–15.Chessel, D., Dufour, A.B., Thioulouse, J., 2004. The ade4 package-I—one-tablemethods. R News 4, 5–10.Drake, J.W., 1991. A constant rate of spontaneous mutation in DNA-basedmicrobes. Proc. Natl. Acad. Sci. USA 88, 7160–7164.Diniz, J.A.F., Torres, N.M., 2002. Phylogenetic comparative methods and thegeographic range size—body size relationship in new world terrestrialcarnivora. Evol. Ecol. 16, 351–367.Faith, D.P., 1992. Conservation evaluation and phylogenetic diversity. Biol.Conserv. 61, 1–10.Faith, D.P., Walker, P.A., 1996. Environmental diversity: on the best-possibleuse of surrogate data for assessing the relative biodiversity of sets of areas.Biodiversity Conserv. 5, 399–415.Faith, D.P., 2004. From species to supertrees: popperian corroboration and somecurrent controversies in systematics. Aust. Syst. Bot. 17, 1–16.Humphries, C.J., Williams, P.H., Vanewright, R.I., 1995. Measuring biodiversityvalue for conservation. Annu. Rev. Ecol. Syst. 26, 93–111.Heard, S.B., Mooers, A.O., 2000. Phylogenetically patterned speciation rates andextinction risks change the loss of evolutionary history during extinctions.Proc. R. Soc. Lond. B–Biol. Sci. 267, 613–620.Isaac, N.J.B., Turvey, S.T., Collen, B., Waterman, C., Baillie, J.E.M., 2007. Mammals onthe EDGE: conservation priorities based on threat and phylogeny. Plos One 2,e296.IUCN, 2006. IUCN Red List of Threatened Species. /http://www.iucnredlist.orgS.Johnson, C.N., Delean, S., Balmford, A., 2002. Phylogeny and the selectivity ofextinction in Australian marsupials. Anim. Conserv. 5, 135–142.Johnson, W.E., Eizirik, E., Pecon-Slattery, J., Murphy, W.J., Antunes, A., Teeling, E.,O’Brien, S.J., 2006. The late Miocene radiation of modern Felidae: a geneticassessment. Science 311, 73–77.Kazazian, H.H., 2004. Mobile elements: drivers of genome evolution. Science 303,1626–1632.Krajewski, C., 1994. Phylogenetic measures of biodiversity: a comparison andcritique. Biol. Conserv. 69, 33–39.Lynch, M., Conery, J.S., 2003. The origins of genome complexity. Science 302,1401–1404.Long, M., Betran, E., Thornton, K., Wang, W., 2003. The origin of new genes:glimpses from the young and old. Nat. Rev. Genet. 4, 865–875.May, R.M., 1990. Taxonomy as destiny. Nature 347, 129–130.Mooers, A.O., Goring, S.J., Turvey, S., Kuhn, T., 2009. Holocene extinctions and theloss of feature diversity. In: Turvey, S.T. (Ed.), Holocene Extinctions. OxfordUniversity Press, Oxford, UK, pp. 263–277.Magnuson-Ford, K., Ingram, T., Redding, D.W., Mooers, A.O., 2009. Rockfish(Sebastes) that are evolutionarily isolated are also large, morphologicallydistinctive and vulnerable to overfishing. Biol. Conserv. 142, 1787–1796.Nixon, K.C., Wheeler, Q.D., 1992. Measures of phylogenetic diversity. In: Novacek,M.J., Wheeler, Q.D. (Eds.), Extinction and Phylogeny. Columbia UniversityPress, New York, pp. 216–234.Oliver, M.J., Petrov, D., Ackerly, D., Falkowski, P., Schofield, O.M., 2007. The modeand tempo of genome size evolution in eukaryotes. Genome Res. 17, 594–601.Pavoine, S., Ollier, S., Dufour, A.B., 2005. Is the originality of a species measurable?Ecol. Lett. 8, 579–586.Petrov, D.A., 2002. Mutational equilibrium model of genome size evolution. Theor.Popul. Biol. 61, 531–544.Purvis, A., Hector, A., 2000. Getting the measure of biodiversity. Nature 405,212–219.R Development Core Team, 2009. R: a Language and Environment for StatisticalComputing. R Foundation for Statistical Computing, Vienna, Austria.Redding, D.W., Mooers, A.O., 2006. Incorporating evolutionary measures intoconservation prioritization. Conserv. Biol. 20, 1670–1678.Russell, G.J., Brooks, T.M., McKinney, M.M., Anderson, C.G., 1998. Present andfuture taxonomic selectivity in bird and mammal extinctions. Conserv. Biol.12, 1365–1376.Redding, D.W., Hartmann, K., Mirnoto, A., Bokal, D., Devos, M., Mooers, A.O., 2008.Evolutionarily distinctive species often capture more phylogenetic diversitythan expected. J. Theor. Biol. 251, 606–615.Ricotta, C., 2004. A parametric diversity measure combining the relative abundancesand taxonomic distinctiveness of species. Divers. Distrib. 10, 143–146.Ricklefs, R.E., 2007. Estimating diversification rates from phylogenetic information.Trends Ecol. Evol. 22, 601–610.Steel, M., Mimoto, A., Mooers, A.O., 2007. Hedging our bets: the expectedcontribution of species to future phylogenetic diversity. Evol. Bioinform. 3,237–244.Vane-Wright, R.I., Humphries, C.J., Williams, P.H., 1991. What to protect?Systematics and the agony of choice. Biol. Conserv. 55, 235–254.Vamosi, J.C., Wilson, J.R.U., 2008. Nonrandom extinction leads to elevated loss ofangiosperm evolutionary history. Ecol. Lett. 11, 1047–1053.Zhu, Y.X., Dai, J.B., Fuerst, P.G., Voytas, D.F., 2003. Controlling integrationspecificity of a yeast retrotransposon. Proc. Natl. Acad. Sci. USA 100,5891–5895.212

Plant Soil (2011) 347:211–220DOI 10.1007/s11104-011-0839-2REGULAR ARTICLEStrong plant-soil associations in a heterogeneous subtropicalbroad-leaved forestLiwen Zhang & Xiangcheng Mi & Hongbo Shao &Keping MaReceived: 27 January 2011 /Accepted: 19 May 2011 /Published online: 2 June 2011# Springer Science+Business Media B.V. 2011Abstract The relative importance of deterministic andneutral processes on community assembly is currently atopic of much debate among ecologists. Analyzingspecies-environment associations is an effective way toassess the importance of deterministic process such asniche differentiation, but both habitat association anddispersal limitation can produce similar patterns ofspatial aggregation in species. Therefore, it is crucial toResponsible Editor: Hans Lambers.Electronic supplementary material The online version of thisarticle (doi:10.1007/s11104-011-0839-2) containssupplementary material, which is available to authorized users.L. Zhang : H. ShaoThe CAS/Shandong Provincial Key Laboratory of CoastalZone Environmental Processes, Yantai Institute of CoastalZone Research,Chinese Academy of Sciences,Yantai 264003, ChinaL. Zhange-mail: lwzhang@yic.ac.cnH. Shaoe-mail: hbshao@yic.ac.cnL. Zhang : X. Mi : K. Ma (*)State Key Laboratory of Vegetation and EnvironmentalChange, Institute of Botany, Chinese Academy of Sciences,20 Nanxincun, Xiangshan,Beijing 100093, Chinae-mail: kpma@ibcas.ac.cnX. Mie-mail: mixiangcheng@ibcas.ac.cncontrol for the impact of dispersal limitation on speciesdistributions when analyzing species-environment associations.We sampled soil with high resolutions in a24 ha stem-mapped subtropical forest and tested plantsoilassociations. We controlled for the influence ofdispersal limitation by employing the homogeneousThomas process to simulate the effect of dispersallimitation on the aggregation of tree species. Aftercontrolling for the effect of dispersal limitation, wefound that the spatial heterogeneity of soil propertieswas associated with distributions of 88.2% (90 of 102species) of tree species in this subtropical forest.Furthermore, not only did soil properties influence thedistribution of tree species, but also tree species tendedto affect properties of the soil around them. The soilfactors most strongly influencing species distributionswere TC, TN, TP, K, Mg, Si, soil moisture, and bulkdensity. We found the spatial heterogeneity of soilproperties to be strongly associated with tree speciesdistributions. Niche partitioning of soil gradients contributedsubstantially to species coexistence in thissubtropical forest.Keywords Dispersal limitation . Plant-soilassociations . Soil nutrients . Spatial patterns .Species coexistence . Subtropical forestAbbreviationsGTS 24 ha Gutianshan ForestDynamic PlotNmin Nitrogen mineralization rate213

212 Plant Soil (2011) 347:211–220TCTNTPBDPCACCAPCsIntroductionTotal CTotal NTotal PBulk densityPrincipal CorrespondenceAnalysisCanonical CorrespondenceAnalysisPrincipal ComponentsIt is widely recognized that species are oftenaggregately distributed in natural forests (He et al.1997; Condit et al. 2000; Li et al. 2009). Ecologistsare interested in the mechanisms that determinespecies distributions in forests because understandingthese mechanisms will improve our understanding ofspecies coexistence overall (He et al. 1997; Law et al.2009).Two contrasting mechanisms have been proposedto explain the formation of species aggregationin forests: habitat association and dispersal limitation(Hubbell 2001; Chave et al. 2002; Clark et al. 2007;Chave 2008; Clark 2009). As a part of niche assemblyrules, habitat association predicts that species distributionis determined by environmental characteristicsin a local community (Chase and Leibold 2003). Incontrast, the dispersal limitation paradigm anticipatesthat the dispersal abilities of species, species compositionin the ambient metacommunity, immigrationrates, and the distance from maternal parent treesdetermine species distribution regardless of habitatconditions (Hubbell 2001). Ecologists have not yetreached a consensus on the relative importance ofthese competing hypotheses in generating speciesaggregation. Some studies have found that aggregateddistributions of species are attributed to differentialhabitat preference among species (e.g. Clark et al.1998, 1999; John et al. 2007), but other studies haveshown that dispersal limitations could yield similaraggregated distributions if a species were not able todisperse to all suitable sites and instead establishednear the parent trees (Hubbell 1997; Dalling et al.2002). Some researchers argue that species distributionsin plant communities are determined by bothprocesses (Coomes et al. 1999; Plotkin et al. 2002;Lancaster 2006; Wiegand et al. 2007). Analyzingspecies-environment associations is one effective wayto assess the influence of habitat association onspecies distributions. However, as described above,the fact that dispersal limitation may also lead tospecies aggregation challenges researchers to understandthe underlying process of species distributions.Many statistical methods were proposed to resolvethis problem (Harms et al. 2001; Plotkin et al. 2002;Diggle 2003). In particular, spatial point process modelsare a particularly promising method recently developedto address this issue (Moller and Waagepetersen 2003;Waagepetersen 2007). Spatial point process modelscan mimic the observed spatial structure produced byecological processes such as dispersal limitation, andcan contribute to disentangling the compound effect ofhabitat association and dispersal limitation in the studyof community assembly rules (e.g. Shen et al. 2009;Lin et al. 2011).Studies of species-environment associations inforest ecosystems are often based on habitat types interm of topographic factors previously (Harms et al.2001; Lai et al. 2009; Legendre et al. 2009).However, some studies have shown that topographicvariables may not be a good surrogate for soilproperties meant to reveal the influence of environmentalfactors on species distribution. Hall et al.(2004) found that four species in the genus Entandrophragmadid not show any habitat specialization fortopographic factors, but did associate with differentsoil nutrients (Ca, Mg, P) in a 100 ha forest plot inAfrica. Therefore, it is important to consider moredirectly influential environmental factors such as soilnutrients in studies of plant-environment association.Many studies have focused on species coexistencein the rainforest because of its high species diversity,and several of these found that habitat associationplayed a key role in maintaining species diversity (e.g. Clark et al. 1998, 1999; John et al. 2007). Althoughthe species diversity of the subtropical evergreenbroad-leaved forest is next to that of the rainforest,relatively few studies have examined the mechanismsof species coexistence in this ecosystem. Most ofthese subtropical forests are located in easternEurasia, and a large proportion of these forests arein southern China (Ni and Song 1997), whereas muchof the land at the same latitude elsewhere in the worldis covered by desert. Thus, subtropical forests arevital for sustainable development of this region (e.g.mediating global change). To fill this research gap, we214

Plant Soil (2011) 347:211–220 213established a 24-ha Gutianshan forest dynamics plot(GTS) with high species richness (159 species) in theevergreen broadleaved subtropical forest in the Zhejiangprovince of eastern China (Legendre et al. 2009). Here,we use this long-term forest plot to investigate theassembly rules that allow so many species to coexist insuch a small area. We ask the following questions, (1)Are soil properties strongly associated with speciesdistributions in this subtropical forest? (2) Which soilproperties closely associated with species compositionin this subtropical forest? To address these questions,we sampled soils from the GTS plot at high resolutionand analyzed as many soil properties as possible. Wecontrolled for the effect of dispersal limitations using asimulation technique in the statistical tests.Materials and methodsStudy siteThe 24 ha (400 m×600 m) GTS plot is located inthe Gutianshan National Nature Reserve, ZhejiangProvince, East China (29.25°N, 118.12°E). The GTSplot was established in 2004–2005 following the CTFS(Center for Tropical Forest Science) standard censusprotocols (Condit 1998), and 140,700 individuals(DBH≥1 cm) belonging to 159 species were recordedin the plot. The dominant species were Castanopsiseyrei(Fagaceae), Schimasuperba (Theaceae) andPinusmassoniana (Pinaceae). The topography of theGTS plot is rugged terrain, and the elevation rangesfrom 446.3 m to 714.9 m (Online Resources: FigureS1).The soil is a subtropical red soil (equivalent toUltisols in US soil taxonomy). We sampled 893 pointsusing both regular and random sampling techniques tocover the entire GTS plot from July to August, 2007,following the sampling protocol of the 50 ha BCI(Barro Colorado Island) plot soil survey (Robertson etal. 1997; Johnetal.2007). The main part of the 24 haplot was divided using a 30×30 m grid (390×600 m),and the intersection points were taken as base points.At each base point two extra points were selectedalong a random compass direction (N, E, S, W, NE,NW, SE or SW), while keeping all sample pointswithin the boundaries of the plot (Online Resources:Figure S2). The distances from a base point to its twoextra points were sampled randomly without replacementfrom 2 m, 5 m and 15 m. In order to cover thewhole plot, we divided the remaining area (i.e., 10 m×600 m) into 10 m×30 m grids and sampled thecrossing points (Online Resources: Figure S2).We obtained 22 soil physical and chemical factors:available elements: Al , B, Ca, Cu, Fe, K, Mg, Mn, N,Na, Si, Zn, P; nitrogen mineralization rate (Nmin), N-NO 3 − , N-NH 4 + , pH, soil moisture, bulk density (BD),total C (TC), total N (TN), and total P (TP) by soilanalysis. We measured total C following the Walkley-Black method, total N with the Kjeldahl NitrogenDetermination method, and total P by UV-Spectrometerin the State Key Laboratory of Vegetation and EnvironmentalChange, Institute of Botany, Chinese Academyof Science. Available cations were extracted withMehlich-III solution and determined by AtomicEmission-Inductively Coupled Plasma (AE-ICP). N-NH 4 + and N-NO 3 − were analyzed with a ContinuousFlow Analyzer in the Key Laboratory of Plant-SoilInteractions, China Agriculture University. For soilvariables, we first performed a Box-Cox transformationon the original data, then performed trend surfaceregression, and used the residuals to computeanisotropic empirical semi-variograms and fittedthe empirical semi-variograms with different semivariogrammodels (Legendre and Legendre 1998).The range parameters of semi-variogram models for allsoil properties were larger than the sampling distances,and which suggests the sampling intensity can capturesignificant variation in soil properties. Next, we selectedthe models with the lowest residual sums of squaresandperformed Ordinary Kriging using GS+ version 9.0(GeoStatistics for the Environmental Sciences, byGamma design software) to obtain soil variables forevery 5×5 m quadrat. The average values for 50×50 mquadrats were also calculated. Finally we added thetrend and transformed the data back to its original scale(John et al. 2007).Data analysisPCA (Principal Correspondence Analysis) wasconducted on 22 soil factors to extract 4PCs(Principal Components) that could represent 76%of the spatial variation in these variables. In orderto simulate the spatial autocorrelation structure ofspecies caused by dispersal limitation, a homogeneousThomas model was employed to simulate theaggregation of species based on observed data(Waagepetersen 2007; Shenetal.2009; Lin et al.215

214 Plant Soil (2011) 347:211–2202011). These mimicked distributions were used as anull model in the analysis (John et al. 2007).Moreover, to guarantee samples large enough forthe statistical models, we excluded rare species(those species for which there were fewer than 24individuals in the analysis), and thus 102 specieswere included in the following analysis.The formulas of the homogeneous Thomasmodels as follows (Waagepetersen 2007; Shen etal. 2009):r c ðuÞ ¼akðuc; dÞWhere, u is the expected number of daughter treesper mother tree; α dictates tree density per unit area;k(u–c;δ) is the function that can simulate thedistributions of daughter trees surrounding theirmother tree (c), k is the density of mother trees perunit area, and δ is the standard deviation of thelocation distribution of the offspring for a givenmother tree.The K-function is the average number of neighboringtrees within the entire circle of a given radiusaround a typical point of the pattern divided by theintensity of the pattern. The L-function is a squareroot transformation of the K-function. The formulasof the K-function and the L-function are as follows(Wiegand and Moloney 2004):KðrÞ ¼ A n 2 X ni¼1rffiffiffiffiffiffiffiffiffiffiKðrÞLðrÞ ¼pX nj¼1rI r ðd ij ÞW ijði 6¼ jÞWhere r is the radius of the sampled circle; i and jsubscripts refer to the observed points, the ith pointbeing the centre of the circle; n is number of points inarea A of the study site; d ij is the distance from jth andthe centre point (ith point) of the circle; I rðd ij ÞW ijis an edgeeffect correction weight, if dij≤r, I r (d ij )=1, if d ij >r,I r (d ij )=0, W ij is the proportion of the area of a circlecentered at the ith point with radius d ij that lies withinarea A.We performed the species-soil association analysisusing R 2.9.0 (R Development Core Team, 2009;Package “spatstat”; Package “vegan”). Firstly, weestimated the optimally fitted dispersal parameters (k,δ, u) using the K-function with the Method ofMinimum Contrast (Diggle and Gratton 1984; Mollerand Waagepetersen 2003) for 102 species. If the bestfitteddispersal parameters could be obtained from theobserved spatial distribution of the species, weapplied the best-fit parameters in a homogeneousThomas model to simulate the spatial point distributionof the species subject to dispersal limitation.Otherwise, we simulated point patterns of thesespecies with a homogeneous Poisson model, whichis a complete random process. Including the 499simulated species distribution maps and the observeddistribution map, every species had 500 distributionmaps. We compared the relationship of observedspecies distributions and PCs with the simulatedspecies distributions and PCs in a 50×50 m quadrat.If the Pearson coefficient of the observed speciesdistribution and specific PCs were above the 95%confidence envelope of the Pearson coefficients of thesimulated species distributions and PCs, the associationof the observed species distribution and its PCswas significantly positive, otherwise, significantlynegative.We also employed a similar method to examine therelationship between individual soil variables andspecies composition (50×50 m quadrat) to find outwhich soil factors were significantly associated withcommunity composition. We pooled the simulateddistribution maps of all species together to gain 499artificial communities, and then added the observedcommunity to get 500 communities in total. In thenext step, we used CCA (Canonical CorrespondenceAnalysis; Legendre and Legendre 1998), which is acanonical ordination method to analyze the relationshipbetween soil properties and community composition,and obtained correlation coefficients. If theobserved correlation coefficient was >97.5% or

Plant Soil (2011) 347:211–220 215ResultsThe first four PCs together represented 76% of thespatial variation in 22 soil factors; PC1 through PC4represented 39%, 17%, 13%, and 7% of the variation,respectively (spatial distributions of PC1-PC4 areshown in Online Resources: Figure S3). Hence, PC1was the most important principal component. TC, TN,TP, N, Zn, K, Ca, Mg, and N-NH + 4 were stronglynegatively correlated with PC1 (Table 1), indicatingthat PC1 represents the gradient of macro-nutrients.Mn, Cu, B, Si, BD and soil moisture were stronglynegatively correlated with PC2; Fe, Na, B werenegatively correlated and pH was positively correlatedwith PC3, and N-NO −3 was strongly positivelycorrelated with PC4.Thus, PC2 and PC3 representTable 1 Loadings of the top 4 principal components of 22 soilfactors at GTSPC1 PC2 PC3 PC4TC −0.27 0.12 −0.29 −0.04TN −0.31 0.09 −0.19 0.02TP −0.31 0.05 −0.17 0.15N −0.27 −0.13 0.10 −0.19P −0.18 0.22 −0.10 0.24Fe 0.05 −0.18 −0.39 −0.29Mn −0.16 −0.31 0.28 0.17Zn −0.29 −0.12 −0.01 −0.23Cu −0.19 −0.30 0.20 −0.06K −0.27 −0.16 −0.05 0.25Ca −0.27 −0.20 0.21 −0.05Mg −0.29 −0.18 0.15 0.02Na −0.02 −0.25 −0.30 0.19B 0.00 −0.32 −0.34 −0.27Si 0.20 −0.31 −0.03 −0.10Al −0.09 0.01 −0.25 −0.21−N-NO 3 −0.11 −0.18 −0.12 0.56+N-NH 4 −0.26 −0.10 0.12 −0.27Nmin −0.18 0.23 0 0.14pH −0.09 0.19 0.41 −0.18BD 0.18 −0.31 0.17 0.08Soil moisture −0.23 0.30 0.01 −0.21PC1~PC4 are the top four principal components of 22 soilfactorsAl, B, Ca, Cu, Fe, K, Mg, Mn, N, Na, Si, Zn, P denoteavailable elements, Nmin is N mineralization rate, BD is Bulkdensity, TC, TN, and TP indicate Total C, Total N, and Total Pspatial variation in micro-nutrients and other variables(pH, BD, soil moisture) whereas PC4 represent thespatial variation in N-NO 3 − .The result of the speciessoilassociation analysis revealed that 88.2% (90species) of 102 species were significantly correlatedwith at least one principal component (Online Resources:Table S1). 53 species were significantly correlatedwith PC1, 45 species were significantly correlated withPC2, 35 species were significantly correlated withPC3, and 43 species were significantly correlated withPC4 (Table 2). 39 species (38.2%) associated with onlyone principal component (Table 2).According to their height at maturity, 159 species atGTS wereclassified into three growth groups:canopytrees (≥15 m, 41 species), understory trees (≥5 mand ≤15 m, 49 species), and shrubs (

216 Plant Soil (2011) 347:211–220Table 2 Species-PCsassociations at GTSPC1~PC4 is the top fourprincipal components of 22soil factors. (8, 15) indicates8 species positively associatedwith PC1, and 15 speciesnegatively associated withPC1Canopy trees Understorey trees Shrubs41 species 49 species 12 speciesSpecies associated with PC1 23 (8,15) 28 (17,11) 2 (2,0)Species associated with PC2 23 (15,8) 19 (10,9) 3 (0,3)Species associated with PC3 17 (6,11) 15 (2,13) 3 (1,2)Species associated with PC4 14 (7,7) 24 (13,11) 5 (1,4)Non association 4 5 3Species only associated with 1 PC 15 19 5Species only associated with 2 PCs 10 15 4Species only associated with 3 PCs 6 3 0Species associated with 4 PCs 6 7 0showed significant correlations with PC4 and 5 of 9species (55.6%, p-value=0.196) in shrubs.We found a positive correlation between the speciesaggregation index and the number of significantspecies-soil correlations (r=25 m, correlation coefficientis 0.29, p-value=0.008, A two-tailed permutationtest, 999 permutations, see Online Resource: Table S1)at the 50×50 m quadrat. Thus, at this scale, the morehighly aggregated species were, the more likely theywere to show significant relationships with soilproperties.We also found that TC, TN, TP, K, Mn, Nmin, P, Si,and soil moisture were significantly correlated withspecies composition in the 50×50 m quadrat (Table 4).Thus, these soil factors were more important than theother soil factors measured (Al, B, Ca, Cu, Fe, Mg, N,Na, Zn, bulk density, N-NH 4 + ,N-NO 3 − )instructuringcommunity at GTS.DiscussionSpecies-soil associationsThere are many methods to divide a community intodifferent habitat types, such as clustering analysis andmultivariate regression tree analysis with indicativespecies (Legendre et al. 2009). In our analysis, we didnot attempt to divide the plot into different habitattypes in term of soil factors but instead used PCA toextract orthogonal axes to represent multiple correlatedsoil factors because these quantitative axes can beeasily incorporated into a statistical analysis. Afterconducting a PCA of soil factors, we extracted thefirst four PCs to represent the spatial variation indifferent soil characteristics. PC1 represents thespatial variation in soil elemental nutrients (N, P, K,Ca, Mg), PC2 and PC3 indicate spatial variation ofTable 3 Associationsbetween dominant speciesin different growth groupsand PCsIf the observed correlationcoefficient was below97.5% of simulationenvelope, the relationship ofspecies and PCs wassignificantly negative (“−”)and if it was above 97.5%of simulation envelope, therelationship was significantlypositive. “ns” indicates therelationship is not significantGrowth group Species Abundance PC1 PC2 PC3 PC4Canopy trees Castanopsis eyrei 12406 + ns ns nsSchima superba 8514 ns ns ns nsPinus massoniana 2061 + ns ns nsCyclobalanopsis glauca 1620 − ns ns nsDaphniphyllum oldhamii 2718 ns ns + nsUnderstorey trees Quercus serrata 3508 + ns ns nsRhododendron ovatum 10793 ns + ns nsLoropetalum chinense 4461 ns ns ns nsTernstroemiagymnanthera 3177 ns − + nsDistylium myricoides 3468 − ns ns nsShrubs Chimonanthus salicifolius 7835 ns ns − nsEurya rubiginosa 2770 ns ns ns nsRaphiolepis indica 1995 ns ns + −218

Plant Soil (2011) 347:211–220 217Table 4 Pivotal soil factors at GTSSoil factors Obs. p-valueTC 0.598 0.002*TN 0.643 0.002*TP 0.608 0.002*N 0.191 0.078P 0.399 0.004*Fe 0.069 0.314Mn 0.281 0.016*Zn 0.070 0.276Cu 0.098 0.332K 0.341 0.002*Ca 0.009 0.106Mg 0.090 0.212Na 0.005 0.088B 0.039 0.368Si 0.329 0.006*Al 0.065 0.286N-NO3 0.135 0.072N-NH4 0.164 0.128Nmin 0.266 0.006*pH 0.003 0.032BD 0.318 0.002*Soil moisture 0.361 0.002*Al, B, Ca, Cu, Fe, K, Mg, Mn, N, Na, Si, Zn denote availablecations, Nmin is N mineralization rate, BD is Bulk density, TC,TN, and TP indicate Total C, Total N, and Total P“Obs” are the observed correlation coefficients of soil andcommunity composition and “*” indicates p-value

218 Plant Soil (2011) 347:211–220nutrients, leading to niche partitions underground.Species in the same growth group grow in a similarlight environment, but they may exploit different soilenvironment sunder ground and results in speciescoexistence. At GTS, the most dominant canopy treesC. eyrei and D. oldhamii, showed different patterns;one was significantly positively associated with PC1,while the other was significantly positively correlatedwith PC3. Of the understory species, the distributionof Q. serrata was significantly positively correlatedwith PC1, while R. ovatum was significantly positivelycorrelated with PC2. However, some specieswere associated with the same PCs, thus the coexistenceof these species may be due to other ecologicalprocesses, such as negative density dependence (Zhuet al. 2010; Chen et al. 2010).Previous studies have revealed that tree speciescould cause spatial variation of soil properties throughlitterfall inputs (e.g. Finzi et al. 1998). Our resultindirectly supports this conclusion. We found moreaggregation of the species which were more likely tobe correlated with soil properties at the at spatial scale50×50 m quadrat, indicating that the tree specieschanged soil properties around themselves. Thus, thespecies-soil association at GTS is bidirectional.Pivotal soil factors at GTSThe pivotal soil factors associated with speciescomposition at the GTS were nitrogen, phosphorus,total C, K, Mg, Si, soil moisture, and bulk density.Both available P and total P were strongly correlatedwith species composition in this typical montanesubtropical forest, which supports the view thatphosphorus is an important limiting factor in montaneforest soils (Sollins 1998; Chave 2008). The utilizationof Fe-bound and Al-bound phosphorus or thepartitioning of organic phosphorus may provide otherways for species to coexist by reducing competitions(Oluleye et al. 2008; Turner 2008). Meanwhile,nitrogen was also strongly correlated with speciescomposition at GTS, although N has always beenregarded as a limiting factor in lowland forests. AtGTS, the amount of nitrogen and phosphorus inleaves is even less than at other montane forests ineastern China (Jin 2010). Therefore, we confirmedthat both nitrogen and phosphorus are limitingresources for species growth or reproduction in thissubtropical forest. Besides the parent rock, humuscoming from the decomposition of litterfall is themost crucial source of soil nutrients, and couldinfluence species distributions as an integrated factor(Baillie et al. 2006). The content of TC reflects humuscondition, thus TC is strongly related to speciescomposition and suggests that microorganisms mayplay a significant role in the dynamics of soil resourceavailability at GTS. K and Mg had stronger associationswith community composition than other cations, whichis likely due to the very low pH at GTS, as low pH maycause K and Mg to leach from soil. Si was also stronglycorrelated with species composition. One possiblereason is that Si may protect plants from toxic cations,such as Mn and Al, and promote P absorption (Epstein1994). The relatively small associations of Al and Cuwith community composition may also be due to theabundance of Si in the plot, but the effect of Si maynot be enough to balance the negative influence of Mnon species fitness. The distribution of soil moisturewas also strongly correlated with species composition.Soil moisture was measured in July, in one of the twodrought seasons (July–August and October–February)at GTS; hence, the measured soil moisture availabilityis particularly important as compared to other sites.In summary, by employing recently developedspatial statistical techniques, we found strongspecies-soil associations at GTS. Spatial variation insoil factors was correlated with the distributions of88.2% of 102 species. Thus, niche partitioning of thesoil environment may contribute to species coexistencein this subtropical forest. Furthermore, theinfluence of plants on soil properties may intensifythe contribution of this process. Niche partitioning ofthe soil environment not only occurs among specieswithin the same growth group, but also tends to showamong species in different growth groups, which maypromote species coexistence.Acknowledgements The authors would like to thank FangliangHe and Peter Kuehn for providing useful suggestions on soilsample design and analysis. We are very grateful to Fusuo Zhangand Jianlan Song at the China Agricultural University forproviding help in analyzing soil nutrients. We are grateful forhelp in the field from graduate students XingXing Man and BoYang. We thank two anonymous reviewers for their comments andsuggestions, which have been very helpful for improving themanuscript. We would also like to thank Anne Bjorkman at theUniversity of British Columbia for her assistance with Englishlanguage and grammatical editing of the manuscript. This researchis funded by Key Innovation Project of Chinese Academic Science(KZCX2-YW-430).220

Plant Soil (2011) 347:211–220 219ReferencesBaillie IC, Ashton PS, Chin SP, Davies SJ, Palmiotto PA,Russo SE, Tan S (2006) Spatial associations of humus,nutrients and soils in mixed dipterocarp forest at Lambir,Sarawak, Malaysian Borneo. J Trop Ecol 22:543–553Chase JM, Leibold MA (2003) Ecological niches: linkingclassical and contemporary approaches. The University ofChicago Press, Chicago and LondonChave J (2008) Spatial variation in tree species compositionacross tropical forests: pattern and process. In: Carson WP,Schnitze SA (eds) Tropical forest community ecology.Blackwell Scientific Publications, Oxford, pp 11–30Chave J, Muller-Landau HC, Levin SA (2002) Comparingclassical community models: theoretical consequences forpatterns of diversity. Am Nat 159:1–23Chen L, Mi XC, Comita L, Zhang LW, Ren HB, Ma KP (2010)Community-level consequences of density dependenceand habitat association in a subtropical broad-leavedforest. Ecol Lett 13:695–704Clark JS (2009) Beyond neutral science. Trends Ecol Evol24:8–15Clark DB, Clark DA, Read JM (1998) Edaphic variation andthe mesoscale distribution of tree species in a Neotropicalrain forest. J Ecol 86:101–112Clark DB, Palmer MW, Clark DA (1999) Edaphic variables andthe landscape-scale distributions of tropical rain foresttrees. Ecology 80:2662–2675Clark JS, Dietze M, Chakraborty S, Agarwal PL, Ibanez I,LaDeau S, Wolosin M (2007) Resolving the biodiversityparadox. Ecol Lett 10:647–659Condit R (1998) Tropical forest census plots: methods andresults from Barro Colorado Island, Panama and acomparison with other plots. Springer-Verlag, HeidelbergCondit R, Ashton PS, Baker P, Bunyavejchewin S, GunatillekeS, Gunatilleke N, Hubbell SP, Foster RB, Itoh A,LaFrankie JV, Lee HS, Losos E, Manokaran N, SukumarR, Yamakura T (2000) Spatial patterns inthe distribution oftropical tree species. Science 288:1414–1418Coomes DA, Rees M, Turnbull L (1999) Identifying aggregationand association in fully mapped spatial data. Ecology80:554–565Dalling JW, Muller-Landau HC, Wright SJ, Hubbell SP (2002)Role of dispersal in the recruitment limitation of Neotropicalpioneer species. J Ecol 90:714–727Diggle PJ (2003) Statistical analysis of spatial point patterns.Arnold, LondonDiggle PJ, Gratton RJ (1984) Monte Carlo methods ofinference of implicit statistical models. J R Stat SocB46:193–212Epstein E (1994) The anomaly of silicon in plant Biology. ProcNatl Acad Sci USA 91:11–17Finzi AC, Canham CD, Breemen NV (1998) Canopy tree-soilinteractions within temperate forests: species effects on pHand cations. Ecol Appl 8:447–454Hall JS, Mckenna JJ, Ashton PMS, Gregoire TG (2004) Habitatcharacterizations underestimate the role of edaphic factorscontrolling the distribution of Entandrophragma. Ecology85:2171–2183Harms KE, Condit R, Hubbell SP, Foster RB (2001) Habitatassociations of trees and shrubs in a 50-ha Neotropicalforest plot. J Ecol 89:947–959He FL, Legendre P, LaFrankie JV (1997) Distribution patternsof tree species in a Malaysian tropical rain forest. J Vet Sci8:105–114Hubbell SP (1997) A unified theory of biogeography andrelative species abundance and its application to tropicalrain forests and coral reefs. Coral Reefs 16:S9–S21Hubbell SP (2001) The unified neutral theory of biodiversityand biogeography. Princeton University Press, PrincetonJin DM (2010) Leaf traits of woody species in east Chinaforests and their relationships with climate. PHD dissertation,Institute of Botany, Chinese Academy of ScienceJohn R, Dalling JW, Harms KE, Yavitt JB, Stallard RF,Mirabello M, Hubbell SP, Valencia R, Navarrete H,Vallejo M, Foster RB (2007) Soil nutrients influencespatial distributions of tropical tree species. Proc NatlAcad Sci USA 104:864–869Lai JS (2008) Species habitat associations and species coexistenceon Evergreen Broadleaved forest in Gutianshan, Zhejiang.PHD dissertation, Institute of Botany, Chinese Academy ofScienceLai JS, Mi XC, Ren HB, Ma KP (2009) Species-habitatassociations change in a subtropical forest of China. J VetSci 20:415–423Lancaster J (2006) Using neutral landscapes to identify patterns ofaggregation across resource points. Ecography 29:385–395Law R, Illian J, Burslem D, Gratzer G, Gunatilleke CVS,Gunatilleke I (2009) Ecological information from spatialpatterns of plants: insights from point process theory. JEcol 97:616–628Legendre P, Legendre L (1998) Numerical ecology, 2ndEnglish edition. Elsevier Science BV, AmsterdamLegendre P, Mi XC, Ren HB, Ma KP, Sun IF, Yu MJ, He FL(2009) Partitioning beta diversity in a subtropical broadleavedforest of China. Ecology 90:663–674Li L, Huang ZL, Ye WH, Cao HL, Wei SG, Wang ZG, Lian JY,Sun IF, Ma KP, He FL (2009) Spatial distributions of treespecies in a subtropical forest of China. Oikos 118:495–502Lin YC, Chang LW, Yang KC, Wang HH, Sun IF (2011)Point patterns of tree distribution determined by habitatheterogeneity and dispersal limitation. Oecologia165:175–184Moller J, Waagepetersen RP (2003) Statistical inference andsimulation for spatial point processes. Chapman and Hall/CRC, Boca RatonNakashizuka T (2001) Species coexistence in temperate, mixeddeciduous forests. Trends Ecol and Evol 16:205–210Ni J, Song YC (1997) Relationship between climatic anddistribution of main species of subtropical evergreenbroad-leaved forest in China. Acta Phytoecologia Sinica(Chinese Version) 21:115–129Oluleye AK, Akinrinde EA, Obigbesan GO (2008) Phosphorusdynamics in forest and savannah agro-ecological Alfisolsincubated with different phosphorus fertilizers. J FoodAgric Environ 6:534–537Plotkin JB, Chave J, Ashton PS (2002) Cluster analysis ofspatial patterns in Malaysian tree species. Am Nat160:629–644221

220 Plant Soil (2011) 347:211–220Robertson GP, Klingensmith KM, Klug MJ, Paul EA, Crum JR,Ellis BG (1997) Soil resources, microbial activity, andprimary production across an agricultural ecosystem. EcolAppl 7:158–170Shen G, Yu M, Hu X, Mi XC, Ren HB, Sun IF, Ma KP (2009)Species-area relationships explained by the joint effects ofdispersal limitation and habitat heterogeneity. Ecology90:3033–3041Sollins P (1998) Factors influencing species composition in tropicallowland rain forest: Does soil matter? Ecology 79:23–30Turner BL (2008) Resource partitioning for soil phosphorus: ahypothesis. J Ecol 96:698–702Waagepetersen R (2007) An estimating function approach toinference for inhomogeneous Neyman-Scott processes.Biometrics 63:252–258Wiegand T, Moloney KA (2004) Rings, circles, and nullmodelsfor point pattern analysis inecology. Oikos104:209–229Wiegand T, Gunatilleke CVS, Gunatilleke IAUN (2007)Species associations in a heterogeneous Sri Lankandipterocarp forest. Am Nat 170:77–95Zhu Y, Mi XC, Ren HB, Ma KP (2010) Density dependence inprevalent in a heterogeneous subtropical forest. Oikos119:109–119222

Oikos 121: 236–244, 2012doi: 10.1111/j.1600-0706.2011.19428.x© 2011 The Authors. Oikos © 2012 Nordic Society OikosSubject Editor: Stefano Allesina. Accepted 5 May 2011Separating the effect of mechanisms shaping species-abundancedistributions at multiple scales in a subtropical forestJiajia Cheng , Xiangcheng Mi , Karin Nadrowski , Haibao Ren , Jintun Zhang and Keping MaJ. J. Cheng, X. C. Mi (mixiangcheng@ibcas.ac.cn), H. B. Ren and K. P. Ma, State Key Laboratory of Vegetation and Environmental Change,Inst. of Botany, Chinese Academy of Sciences, 20 Nanxincun, Xiangshan, Beijing 100093, PR China. – J. J. Cheng and J. T. Zhang, College ofLife Science, Beijing Normal Univ., Beijing 100875, PR China. – K. Nadrowski, Inst. Biologie I - Special Botany and Functional Biodiversity,Univ. of Leipzig, Johannisallee 21-23, DE-04103 Leipzig, Germany.Species abundance distributions (SADs) play an important role in the current dispute over mechanisms shaping communityassembly. Niche theory assumes differential occurrence of species in different habitats while neutral theory emphasizesstochastic events and dispersal. The previous tests of niche and neutral models shaping SADs lead to the claim that SADsare not informative for inferring underlying processes. Using spatial statistical models in a fully mapped 24-ha subtropicalforest in China, we first demonstrate that one can not distinguish between the effect of habitat heterogeneity and dispersallimitation on SADs by inspecting whether the observed SADs fall within 95% confidence intervals of the simulated SADs.Subsequently, we demonstrate that SADs can be used to detect mechanisms shaping SADS by comparing alternativeprocess-based models using model selection techniques. We found that dispersal limitation explain SADs at smaller spatialscales, while the combination of niche and dispersal limitation explain SADs at larger scales. These processes are linkedwith the degree of conspecific aggregation, informing further attempts to refine and parameterize the statistical theory ofsampling SADs.Species abundance distributions (SAD) describe richnessas well as dominance structure for a community in spaceor time (McGill et al. 2007). Recently, SAD patterns haveplayed an important role in the search for the underlyingmechanisms of community assembly (Hubbell 2001,Magurran and Henderson 2003, Sugihara et al. 2003, Volkovet al. 2005, 2007, Š izling et al. 2009). Theoretical as well asempirical studies have explained the form of SAD in termsof niche differentiation or neutral processes (Volkov et al.2005, McGill et al. 2007, Ulrich et al. 2010). Niche theoryassumes that species differ in their habitat use and thus canavoid competition by habitat partitioning (Nee et al. 1991,Sugihara et al. 2003). In contrast, neutral theory emphasizesthe importance of stochastic events and dispersal (Hubbell2001, Volkov et al. 2003). It was shown that SADs varyingfrom straight lines to entirely concave curves can be generatedassuming only different dispersal distances (Hubbell2001, Chave et al. 2002). However, most of current methodsinferring underlying processes shaping SADs could notseparate the effect of neutral and niche processes (McGillet al. 2006, Ulrich et al. 2010, but see Chave et al. 2006). Thisleads to the general consensus that analyzing SADs may notbe particularly informative for underlying processes (Chaveet al. 2002, Purves and Pacala 2006, Ulrich et al. 2010).Current testing of the importance of neutral or nicheprocesses in shaping SADs is usually based on curve fittingor model community simulation using different combinationsof mechanisms (McGill et al. 2006). Curve-fittingapproaches, i.e. estimating parameters from observed SADs,use free parameters of questionable ecological relevance tomaximize fit (Alonso et al. 2008, McGill 2003, McGillet al. 2006). These approaches usually fail to reject competingmechanistic explanations based on either neutral orniche theory (McGill 2003, McGill et al. 2006, 2007). Analternative is to simulate model communities based on differentcombinations of ecological processes (Chave et al. 2002,Zillio and Condit 2007). However, the parameters chosento simulate these communities are only rarely extractedfrom real data and thus lead to unrealistic scenarios (Chaveet al. 2002, McGill et al. 2006). This restricted the results ofthese simulations to qualitative assessments, concluding thatthe same patterns of SAD can be qualitatively yielded byniche or neutral processes (Chave et al. 2002, Purves andPacala 2006, Ulrich et al. 2010). Recently, Alonso et al.(2008) theoretically demonstrated that the SADs predictedby different underlying processes will be essentially differentalthough they are qualitatively very similar. They arguedthat it is possible to detect the underlying processes shapingempirical SADs by comparing alternative process-basedmodels using model selection techniques such as Chaveet al. (2006). We here attempt to separate the effect of dispersallimitation and habitat heterogeneity on SADs by comparingprocess-based models using model selection techniques.SAD of a community is determined by the abundancedistribution of the species, i.e. it results from the aggregationof all species-level (population-level) abundance distribution236223

(Plotkin and Muller-Landau 2002, Harte et al. 2005, Greenand Plotkin 2007, Alonso et al. 2008). Green and Plotkin(2007) and Alonso et al. (2008) derived exact expressionsof SADs as functions of sampling scale and degree of conspecificclustering respectively. Both habitat filtering asassumed by niche based approaches and dispersal limitationas assumed by neutral approaches are mechanisms inducingconspecific spatial aggregation. Habitat parameters, such aselevation, slope and soil nutrients that usually vary at largescales along environmental gradients would be expectedto cause lower degree of conspecific aggregation (Wiegandet al. 2007, Getzin et al. 2008), while short distance dispersalis expected to generate higher level of local conspecific clustering(Plotkin et al. 2000, Seidler and Plotkin 2006, Shenet al. 2009). Therefore, to empirically link between mechanisticforces to degree of conspecific aggregation would shedlight on statistical theory of sampling SAD.We employ recently developed spatial statistics methods(Waagepetersen 2007, Waagepetersen and Guan 2009) toseparate effects of habitat heterogeneity as assumed by nichebased approaches, dispersal limitation as assumed by neutralapproaches, and their combination on SADs. We used fourspatial statistical processes, 1) homogeneous Poisson processesrepresenting pure random effects; 2) heterogeneousPoisson processes for the effect of habitat heterogeneity(Illian et al. 2008); 3) homogeneous Thomas processes, alsocalled Poisson cluster models, implementing the effect ofdispersal limitation without habitat heterogeneity (Plotkinet al. 2000, Potts et al. 2004, Seidler and Plotkin 2006, Johnet al. 2007); (4) and finally heterogeneous Thomas processesfor the joint effect of habitat heterogeneity and dispersallimitation (Shen et al. 2009, Waagepetersen 2007, Waagepetersenand Guan 2009). Unlike traditional spatial statisticalmodels, these models do not require unrealistic assumptionssuch as stationarity in geostatistics (Stoyan and Stoyan 1994,Shen et al. 2009) and easily integrate habitat heterogeneityand dispersal limitation.In this study we first attempt to distinguish the effect ofniche (habitat heterogeneity) or neutral (dispersal) based ona comparison to empirical data by using spatial statisticalmodels as null models from a 24-ha stem-mapping plot ina subtropical forest in China. Second, we use informationcriteria to separate niche or neutral effects by comparingcompeting models. This analysis allows us to compare thecontribution of all mechanisms discussed above to the formof observed SADs in subtropical forests at multiple spatialscales. Finally, we also explore the relationship between thesemechanisms and the population-level abundance distribution,and suggest further refinement and parameterization inthe statistical theory of sampling SADs.Material and methodsThe study site and plot descriptionGutianshan National Nature Reserve (29 ° 10’19.4” –29 ° 17’41.4”N, 118 ° 03’49.7” – 118 ° 11’12.2”E) is located inthe west of Kaihua County, Zhejiang Province. The evergreenbroad-leaved subtropical forest in the Reserve, representativevegetation of subtropical China, is dominated by Castanopsiseyrei (Fagaceae), Schima superba (Theaceae) and Pinus massoniana(Pinaceae). Detailed descriptions of climate, topographyand flora can be found in Legendre et al. (2009) andZhu et al. (2010).The 24-ha Gutianshan plot (29 °15.101 ’ –29 °15.344’N,118 °07.010 ’ –118 °07.400’E), a rectangle of 600 400 m, isestablished according to CTFS plot census protocol ( www.ctfs.si.edu/ ). All trees with DBH 1 cm in the plot weretagged, identified, measured, and georeferenced. We recorded140 700 individuals belonging to 48 families and 159 species.Data and spatial statistical modelsAbundance of 102 tree species with DBH 1cm is usedin this study. Species having less than 24 individuals wereexcluded from analyses due to requirement of minimal populationsizes for the accuracy of spatial pattern modelling(Baddeley et al. 2005). In this study, we employ topographicand edaphic variables to represent habitat heterogeneity.During plot establishment in 2004 – 2005, elevations for eachintersection of the 20 m grid (651 intersections) and part ofintersections of the 10 m and 5 m grids (212 intersections)throughout the plot were measured in a topographic survey.We obtained elevation of four corners for every 5 5 mquadrat using ordinary kriging following John et al. (2007).The topographic variables of every 5 5 m quadrat includingaltitude, convexity, slope and aspect were calculated (seedetails in Legendre et al. 2009). We also sampled soil in GTSplot following the protocols of BCI soil sample ( http://ctfs.si.edu/datasets/bci/soilmaps/BCIsoil.html ). Initially,we divided the whole 24-ha plot into 30 30 m cells exceptan extra area of 6000 m 2 , which was portioned into 60 cellsof 30 10 m. We then sampled soil at the cell intersectionpoints together with two additional points at 2, 5 or 15 min a randomly chosen direction to obtain the distributioninformation of soil nutrients at finer scales, but the cells inthe extra area do not have additional points. Overall, wetook 892 soil samples in the whole plot (see detail in Zhanget al. 2011). Finally, we measured soil moisture, bulk density,nitrogen mineralization rate (Nmin), pH, together with16 soil nutrients including total carbon (TC), total nitrogen(TN), total phosphor (TP), available Fe, Mn, Zn, Cu, K, P,Ca, Mg, Na, B, Si, N (including NH 4 and NO - 3 ), and Alfollowing the lab protocol of BCI soil sampling protocols(John et al. 2007). We obtained spatial predictions of elevationsand edaphic variables for the 5 5 m blocks usingordinary kriging following John et al. (2007). Subsequently,we chose the first four principal components (PCs) of 25topographic and edaphic variables explaining 99.94% of thevariation of 25 variables (Supplementary material AppendixA1) to reduce redundant information among covaryinghabitat variables and minimize the possibility of overfitting(John et al. 2007, Shen et al. 2009).Model descriptionWe employ four spatial statistical processes, i.e. the homogeneousPoisson process, the heterogeneous Poisson process,the homogenous Thomas process and the heterogeneousThomas process, to investigate the contribution of differentmechanisms in structuring SADs.224237

1) The homogeneous Poisson process randomly generatespoints with average density α of each species in a givenarea (Plotkin et al. 2000, Illian et al. 2008). This process,assuming that each individual is distributed randomly andindependently, represents pure random arrangement andprovides a null model of spatial pattern in ecology.2) Compared with the homogeneous Poisson process, theheterogeneous Poisson process incorporates habitat variablesinfluencing the distribution of points into model. Itproduces clustered distributions with respect to environmentvariables and generally improves the prediction ofpopulation distribution. The process has four more parameters( β j , j 1 – 4) than the homogeneous Poisson processbecause it models the correlation between tree density andhabitat variables represented by first four PCs of environmentalvariables (Illian et al. 2008, Shen et al. 2009).3) The homogeneous Thomas process is a cluster processformed by randomly located cluster centers from a Poissonprocess with intensity κ . The offspring trees are alsoassumed to be a Poisson process with intensity μ , andthe locations of offspring trees are isotropically andnormally distributed around each cluster center with amean being zero and a standard deviation δ . For eachspecies, the distribution is defined by three parametersin the process, κ , μ and δ (Potts et al. 2004, Seidler andPlotkin 2006). This process represents the effect of dispersallimitation without habitat heterogeneity.4) The heterogeneous Thomas process is similar to thehomogeneous Thomas process except that the treedensity in each quadrat is associated with the habitatvariables (thus assuming niche processes in communityassembly) in the quadrat. Additional parameters,β j ( j 1 – 4) were also used to model the correlationsbetween the tree density and habitat variables representedby the first four PCs of the environmental variables(Waagepetersen 2007, Waagepetersen and Guan2009). Thus, this process includes the joint effects ofhabitat heterogeneity and dispersal limitation.Model parameterizationWe estimate the model parameters from data of Gutian plot.Parameter α is simply estimated by density of each speciesin the homogeneous Poisson process, and parameters of species-habitatassociation ( β j , j 1 – 4 ) in the heterogeneousPoisson process are estimated using maximum likelihood(M ø ller and Waagepetersen 2004). Dispersal related parameters( κ , μ , δ ) are estimated by comparing the empirical Kˆ(r)with the theoretical K function using minimum contrastmethods (Stoyan and Stoyan 1994, Diggle 2003); parametersin the heterogeneous Thomas process were estimatedusing Waagepetersen and Guan (2009) ’ s two-step approach(Supplementary material Appendix A4). When the specieshabitatassociation can be quantified by habitat variables,inhomogeneous K functions are used instead of homogeneousK functions (Baddeley et al. 2000).Simulations of communitiesUnder assumption of species independence (Volkov et al.2005, 2007, Alonso et al. 2008), we use the four parameterizedmodels to simulate the spatial distribution of each speciesin the entire plot, and then overlaid simulated trees of allspecies to get the simulated community composition. Inthis study, we obtained 100 simulated communities foreach processes. Then we randomly place a quadrat with size10 10 m onto the simulated community, and constructedthe SADs of Preston-like and rank-abundance type. Weobtain 10 SADs from each simulated community. Finallythe predicted SADs of each model at the scale of 10 10 mare obtained with the mean value of 1000 SADs of Prestonlikeand rank-abundance type from 100 simulated communities,and a 95% confidence interval is constructed for eachpredicted SAD. We also construct SADs of rank-abundancetype and Preston-like at scales 20 20, 40 40, 80 80,100 100 and 200 200 m using the same method.Statistical analysesWe first assess models by inspecting whether the observedempirical SADs fall within 95% confidence intervals ofthe simulated SADs. Subsequently, the performance of thefour models is compared with an approximation of Akaike ’ sinformation criterion (AIC) and Bayesian information criterion(BIC). The AICs and BICs of the four models can beapproximated as following (Webster and McBratney 1989,Shen et al. 2009):AIC ∼ n ln R 2k (1)BIC ∼ n ln R k lnn (2)where n is the number of ranks or octaves classes, k is thenumber of parameters in a model and R is the sum of residualsquares of difference in species abundance at each species rankor octave class (Supplementary material Appendix A5).Since forms of SADs are strongly influenced by the degreeof conspecfic spatial aggregation (Plotkin et al. 2000, Plotkinand Muller-Landau 2002, Harte et al. 2005, Green andPlotkin 2007), we compare the degree of conspecific spatialaggregation between simulated and observed communities.For this, we first estimate the degree of conspecific aggregationusing nearest neighbor distance function G ( r ) for allobserved and simulated communities except those generatedby the homogeneous Poisson process, which is not expectedto cause spatial aggregation. Next we compare the degree ofconspecific aggregation between observed distribution andsimulated distribution using major axis regression (Legendreand Legendre 1998). If the regression line is below identityline, the model underestimates the degree of spatial aggregation,whereas if the regression line is above identity line, themodel overestimates the degree of spatial aggregation.The PCs of environmental variables was extracted usingthe package ‘ vegan ’ (Oksanen et al. 2010), major axis regressionwas implemented using the package ‘ lmodel2 ’ (Legendre2009), and simulations of spatial distribution were carriedout using the Package ‘ spatstat ’ (Baddeley and Turner 2005).ResultsIn this study, we present results of Preston-like and rankabundancetype distribution at scales from 10 10 to200 200 m 2 . The results in Fig. 1 and Supplementary238225

Figure 1. 95% confidence intervals of the four models at scales of 40 40 m (upper) and 100 100 m (lower). Solid line representsobserved mean value of species abundance distribution (SAD) of 100 samples at corresponding scales. Dotted lines are the average valuesof SADs from 1000 (10 of each simulated communities) samples at corresponding scale, and dashed lines are 95% confidence intervalsconstructed using the 25th-lowest and 25th-highest value of 1000 simulated SADs. Different symbols represents different SADs simulatedby different processes: Δ : homogeneous Poisson process, : heterogeneous Poisson process, ° :homogeneous Thomas process, ∇ :heterogeneousThomas process. 95% confidence intervals of the four processes at scales of 10 10 m, 20 20 m, 80 80 m, 200 200 m wereshown in Appendix 3.material Appendix A2 – A3 show that the observed SADs ofrank-abundance type and Preston-like fall within the 95%confidence intervals estimated by heterogeneous Poisson processes,homogeneous Thomas processes and heterogeneousThomas processes, while not including the SAD generatedby homogeneous Poisson processes at all scales. However,using approximated AIC and BIC values, models can be discriminated(Table 1). Additionally, on all but one scale scale,both Preston-like and rank-abundance type SADs supportedthe same model. None of the models using purely random oronly niche-based processes was chosen across all scales, however,on small scales, homogenous Thomas processes assumingdispersal limitation only were preferred, while on largerscales heterogeneous Thomas processes assuming dispersallimitation as well as niche separation performed best.At a scale of 10 10m, homogenous and heterogeneousPoisson processes often underestimate abundance of abundantspecies, and overestimate abundances of rare species (Supplementarymaterial Appendix A5a). At larger scales, estimationof abundant species improves, but rare species continue to be226239

overestimated (Supplementary material Appendix A5b – f).In contrast, homogenous and heterogeneous Thomas processesslightly overestimate abundance of abundant speciesand underestimate abundance of rare species at small scales(Supplementary material Appendix A5a – c). With increasingscales, estimation of abundant species improves, but underestimationof less abundant species at the tail of SADs remainsfor homogeneous Thomas processes (Supplementary materialAppendix A5d – f). There is greater discrepancy betweenobserved SADs and simulated SADs by homogeneous andheterogeneous Poisson processes, although addition of environmentalfactors greatly improves the estimation of SADsfor heterogeneous Poisson process (Supplementary materialAppendix A5a – e). Yet at scales of 200 200 m, the heterogeneousPoisson process, as well as the heterogeneousThomas process, seems to be a good estimator of abundanceof common species.Th e results of Preston-like SADs exhibit similar patternas those of rank-abundance type at scales of 10 10 –200 200 m (except at scale of 20 20 m, see the resultsfrom AIC, BIC and MR a in Table 1, Supplementary materialAppendix A5, A6). Patterns of observed Preston-like SADsvary with sampling scales (Fig. 2). As sample scale increases,SADs change from typical log-series SAD with a large numberof singletons (Fig. 2a – b) to log-normal type (Fig. 2c – f).The homogeneous Poisson process often predicts more rareand common species, but less abundant species. The heterogeneousPoisson process also predicts more rare and commonspecies than observed distribution, but did a bit betterthan homogeneous Poisson process (Table 1). The PrestonlikeSADs predicted by Thomas processes, including bothhomogeneous and heterogeneous Thomas processes, seemflatter when shifting from rare species to abundant ones thanPoisson processes, and are closer to observed distribution.But as the scale increases, the homogeneous Thomas processestimates less rare species compared to observed distributions,while heterogeneous Thomas process predict moreoctaves than observed distribution at most scales (Fig. 2).As shown in Fig. 3, the heterogeneous Poisson and theheterogeneous Thomas process usually underestimate thedegree of conspecific aggregation, whereas the homogeneousThomas process usually overestimates the degree ofconspecific aggregation. At all examined scales, the heterogeneousPoisson process underestimates the degree of conspecificaggregation (except at scale of 200 200 m wherethe heterogeneous Poisson process overestimates the degreeof aggregation of species with lower clustering, Fig. 3). Incontrast, homogeneous Thomas process usually overestimatesthe degree of conspecific aggregation at larger scales(40 40 – 200 200 m). But the conspecific aggregationsimulated by homogeneous Thomas process is close to theobserved aggregation at small scales (10 10 – 20 20 m)(Fig. 3). The conspecific aggregation simulated by the heterogeneousThomas process is between those simulated byheterogeneous Poisson process and homogeneous Thomasprocess (Fig. 3) on most spatial scales.DiscussionExploration of mechanisms structuring SAD of communityis a central question of ecology. The previous tests ofniche and neutral models lead to the claim that SADs arenot informative for inferring underlying processes (Chaveet al. 2002, McGill et al. 2006, Purves and Pacala 2006).However, Alonso et al. (2008) argued that this claim is toocrude because it is possible to detect the underlying mechanismsshaping SADs using quantitative alternative processbasedmodel comparison. Here we show that mechanismsshaping SAD can be detected by comparing quantitativelypredicted SADs from alternative process-based models withempirical SADs. Integrating niche, neutral and combinedprocesses into spatial point process models at multiple spatialscales in a subtropical forest, we first demonstrate thatSADs cannot be separated based on 95% confidence intervals.Subsequently, we quantitatively separate mechanismsshaping the same SADs and show that different processesshape SADs at small as opposed to large spatial scales.We could not separate the effect of habitat heterogeneity,dispersal limitation and their joint effect by inspecting whetherthe observed SADs fall with 95% confidence intervals ofsimulated SADs at different scales as previous tests found(Fig. 1, S1 – 2) (Chave et al. 2002, McGill et al. 2006, Purvesand Pacala 2006). This parallels Chisholm and Pacala ’ s (2010)Table 1. Approximated Akaike ’ s information criterion (AIC) and Bayesian information criterion (BIC) (in brackets) values of four spatial pointpattern models for species abundance distributions (SADs) of Preston-like and rank – abundance type.SAD type Scale ( m 2 )Homogeneous Poisson(purely random)HeterogeneousPoisson (niche)Homogenous Thomas(dispersal)Heterogeneous Thomas(niche dispersal)Preston-like 10 10 26.4396 (26.3855) 27.7114 (27.4410) 10.5345 (10.3723) 24.4457 (24.0130)20 20 40.8054 (41.1079) 40.5983 (42.1112) 26.1243 (27.0320) 28.7906 (32.6698)40 40 44.7240 (45.1219) 41.4363 (43.8608) 36.4261 (37.6198) 22.1818 (26.0611)80 80 51.8551 (52.3400) 39.0015 (41.4261) 40.6793 (42.1340) 26.2544 (30.7740)100 100 50.0912 (50.5761) 39.0930 (41.5176) 41.4759 (43.1708) 23.6288 (28.1484)200 200 56.8818 (57.4467) 46.4603 (49.2850) 49.7586 (47.6757) 42.4023 (47.5147)Rankabundancetype10 10 64.0878 (65.7253) 39.3313 (47.4681) –16.5805 (–12.27849) –1.76156 (8.8960)20 20 156.6813 (158.7588) 119.0804 (129.3905) 33.8626 (39.53806) 17.21254 (32.3471)40 40 243.9256 (246.3200) 157.5563 (169.4664) 139.7302 (146.5602) 43.3616 (61.5749)80 80 351.4052 (354.0104) 192.8835 (205.4378) 119.3711 (126.7342) –97.7453 (–76.1105)100 100 359.8841 (362.4992) 195.5229 (208.3965) 97.3080 (104.8074) –122.5364 (–102.3621)200 200 339.6987 (342.3236) 107.9668 (121.0917) 105.7122 (113.5575) –152.6998 (–131.7000)The lowest AIC value in the four spatial process models was bolded. The abundance of rank-abundance type SADs are log-transformed usingln( A i 0.5), where A i is the abundance of the i th species.240227

Figure 2. Plots of the observed and simulated species-abundance distribution (SAD) at different scales in Gutian plot at scales of 10 10 –200 200 m. Solid line represents observed mean value of SADs from 100 samples. Blue dotted-lines represents homogeneous Poissonprocess, pink dotted-lines represents heterogeneous Poisson process, red dotted-lines represents homogeneous Thomas process, green dottedlinesrepresents heterogeneous Thomas process. The SADs are plotted using Preston ’ s binning method. The numbers on the x axis representPreston ’ s octave classes, and octaves represent the number of species with abundance of 1, 2, 3 – 4, 5 – 8, 9 – 16, and so on (Hubbell 2001).findings that a strong niche model using many parametersproduces the same form of SAD as a much simpler neutralmodel. The observed patterns of SADs at multiple scales fallwithin the 95% of CI of four models except the homogeneousPoisson process (Fig. 1, S1 – 2). This non-parametrictest indicates that species aggregation is important in shapingSADs, and random placement inadequately describesspatial patterns of species composition as well as SADsregardless the spatial scale (Plotkin et al. 2000, Green et al.2003, Green and Plotkin 2007). Random placement slightlyunderestimated abundances of abundant species and clearlyoverestimated abundances of rare species (Supplementarymaterial Appendix A5), consistent with findings of Harteet al. (2005).However, quantitative model comparison enables us tochoose between the competing models (Table 1, Supplementary228241

Figure 3. Comparison of observed aggregation of 100 species (having 24 individuals/24ha) with aggregation simulated by three spatialstatistical models at scales of (a) 10 10 m; (b) 20 20 m; (c) 40 40 m; (d) 80 80 m; (e) 100 100 m; (f) 200 200 m. Filledcircles represent the comparison between observed aggregation and aggregation simulated by heterogeneous Poisson process, while opencircles stands for the homogeneous Thomas process, and open triangles represent the heterogeneous Thomas process. The solid line standsfor the identity relationship with equal aggregation of observed and simulated point patterns, the dashed line for major axis regression linebetween observed aggregation and simulated aggregation by heterogeneous Poisson process, and the dotted line for the homogeneousThomas process, and the dot-dashed line for the heterogeneous Thomas process.material Appendix A5). We found that dispersal limitationis sufficient to explain pattern of SADs at smaller scales(10 10 – 20 20 m), whereas the joint effect of dispersallimitation and habitat heterogeneity performs best at largerscales (40 40 – 200 200 m; Table 1, Fig. 2, Supplementarymaterial Appendix A5). This result complementthe finding of Alonso et al. (2008) that qualitatively similarSADs generated by different underlying processes will beessentially different. The patterns of SADs have been shownto vary with the degree of conspecific spatial aggregationand sampling scale (Plotkin et al. 2000, Plotkin and Muller-Landau 2002, Harte et al. 2005, Green and Plotkin 2007,Š izling et al. 2009). Habitat heterogeneity often causesspatially aggregated patterns at larger scales. Gong et al.(2007) found that 79% of the examined species showedsignificant correlation with at least one habitat type in the242229

Gutian plot. But large-scale habitat heterogeneity alonemodeled by heterogeneous Poisson process often underestimatesconspecific aggregation (Fig. 3) (Shen et al. 2009),resulting in the underestimation of abundant species at thehead of SADs and overestimation of less abundant speciesat tail of SADs (Supplementary material Appendix A5).Particularly at smaller scales (10 10 – 20 20 m),habitat heterogeneity cannot explain structure in SADs(Table 1). On the other hand, dispersal limitation is anotherimportant force inducing aggregated conspecific patternat smaller scales and shaping SADs demonstrated byVolkov et al. (2003), Etienne and Olff (2005) and numerousother studies. Legendre et al. (2009) found that dispersallimitation determines community composition of theforest in Gutian plot at smaller scale (10 10 m). However,the homogeneous Thomas process representing dispersallimitation exaggerates the degree of conspecific aggrega tion(Fig. 3), leading to the overestimation of abundant speciesat the head of SADs and underestimation of less abundantspecies at the tail of SADs at multiple scales (Supplementarymaterial Appendix A5).Our results indicate that the joint effect of habitat heterogeneityand dispersal limitation can best explain SADs atlarge scales, and confirms the continuum hypothesis (Gravelet al. 2006). The heterogeneous Thomas process incorporatesthe joint effect of habitat filtering at larger scales and dispersallimitation at smaller scales, and improves the power ofpredicting SADs at larger scales (Table 1). But dispersal limitationwithout incorporating habitat heterogeneity can stillfit SADs quite well at smaller scales (10 10 and 20 20m), indicating that dispersal limitation is the only leadingprocess at smaller scales (Table 1). Our results parallel recentfindings in Gutian plot: Legendre et al. (2009) found thatdispersal limitation and habitat heterogeneity equally governthe community composition at larger scales (20 20 –50 50 m). Shen et al. (2009) demonstrated that the jointeffect of habitat heterogeneity and dispersal limitation canbest explain the species-area relationship in forest communitiesof Gutian plot, partly inconsistent with our results thatdispersal limitation is the leading process at smaller scales.However, this disparity may result from the effort to find asingle best model across all scales.When developing a statistical theory of sampling fromspecies abundance distributions, Green and Plotkin (2007)and Alonso et al. (2008) showed the form of the SADs differdepending on spatial scale and population-level abundancedistribution, i.e. conspecific aggregation respectively. Assuminghomogeneous Poisson processes, no conspecific aggregationis generated and the form of the SAD remains thesame in a smaller sample from a larger region. However, theobserved SAD shows not as heavy tails as the SAD generatedby assuming negative binomial distribution. This means thatwe observe relative less abundant as well as rare species thanwe would expect form a negative binomial distribution. Herewe show two patterns that may be of interest in this respect.First, our different models cause different degree of spatialaggregation in the simulated communities (Fig. 3). Whilethe heterogeneous Poisson process simulating habitatfiltering consistently produces the lowest aggregation, thehomogeneous Thomas process simulating dispersal limitationonly generally produces the highest agglomeration. Theheterogeneous Thomas process combining these two aspectsresults in intermediate agglomeration. Although Green andPlotkin (2007) consider both processes to result in agglomeration,they do not differentiate between agglomeration causedby habitat filtering and agglomeration caused by dispersallimitation. As a second point we show that including theseprocesses, the resulting SADs do not result in more abundantas well as more rare species relative to those simulatedby the homogeneous Poisson process. While there are indeedmore abundant species in the simulated communities includingaggregation (Supplementary material Appendix A5), thereare less instead of more rare species (Supplementary materialAppendix A5, Fig. 2). Green and Plotkin (2007) consider thecase where the intensity of agglomeration may changes withthe overall abundance of species in. However, our results maysuggest that the intensity of agglomeration may also changewith the underlying processes of community assemblage. Ourresults support Š izling et al. ’ s (2009) finding that a higher degreeof conspecific aggregation leads to right-skewed SAD whenaccounting for autocorrelation between samples. Our resultsdemonstrate that the degree of local conspecific aggregationis determined by habitat heterogeneity and dispersal limitation,suggesting that SADs should not be evoked by only statisticallyconvergent processes as found by Š izling et al. (2009).One of the potential caveats is that the models in the workare static and retroactive and ignore community dynamicprocesses such as birth, death, extinction and speciation.For example, overestimating abundant species at multiplescales by heterogeneous Thomas processes suggests that negativeconspecific density dependence at local scales shouldbe taken into account. Another potential caveat is that 57(of 159) species with less than 24 individuals are excludedfrom our analysis due to requirement of minimal populationsizes for the accuracy of spatial pattern modeling, which mayweaken the general application of our conclusions.In summary, in this paper we demonstrate that indeedunderlying mechanisms shaping SADs can be separated.This is accomplished by comparing process-based modelsthat share the same philosophy and can then be subjected tomodel selection techniques. We find that neutral processesshape SADs at smaller scales, while the combined effects ofniche and neutral processes act at larger scales in a subtropicalforest. We show that these processes are linked with thedegree of conspecific aggregation, which may inform furtherattempts to refine and parameterize the statistical theory ofsampling species abundance distributions.Acknowledgements – We thank Arno š t L. Š izling, Liza Comita,Yongtao Guan, Fangliang He and Guochun Shen for their insightfulcomments and suggestions on this manuscript. We thank FangTeng, Chen Shengwen and many other field workers who help usto carry out censuses during 2005. This study was funded by NSFCgrant no. (31061160188), Key Innovation Project of CAS (KSCX2-EW-Z-5) and German Science Foundation (DFG FOR 891/1).ReferencesAlonso, D. et al. 2008. The implicit assumption of symmetry andthe species abundance distribution. – Ecol. Lett. 11: 93 – 105.Baddeley, A. and Turner, R. 2005. Spatstat: an R package for analyzingspatial point patterns. – J. Stat. Software 12: 1 – 42.230243

Baddeley, A. J. et al. 2000. Non- and semi-parametric estimationof interaction in inhomogeneous point patterns. – Stat. Neerl.54: 329 – 350.Baddeley, A. et al. 2005. Residual analysis for spatial point processes(with discussion). – J. R. Stat. Soc. Ser. B. Stat. Methodol.67: 617 – 666.Burnham, K. P. and Anderson, D. R. 2002. Model selection andmultimodel inference: an information theoretic approach. –Springer.Chave, J. et al. 2002. Comparing classical comunity models: theoreticalconsequences for patterns of diversity. – Am. Nat. 159: 1 – 23.Chave, J. et al. 2006. Theoretical biology: comparing models ofspecies abundance. – Nature 441: E1 – E2.Chisholm, R. A. and Pacala, S. W. 2010. Niche and neutral modelspredict asymptotically equivalent species abundance distributionsin high-diversity ecological communities. – Proc. NatlAcad. Sci. USA 107: 15821 – 15825.Diggle, P. J. 2003. Statistical analysis of spatial point patterns, 2ndedi. – Academic Press.Etienne, R. S. and Olff, H. 2005. Confronting different models ofcommunity structure to species-abundance data: a Bayesianmodel comparison. – Ecol. Lett. 8: 493 – 504.Getzin, S. et al. 2008. Heterogeneity influences spatial patterns anddemographics in forest stands. – J. Ecol. 96: 807 – 820.Gong, G. Q. et al. 2007. Habitat associations of wood species inthe Gutianshan subtropical broad-leaved evergreen forest. –Sci. Soil Water Conserv. 5: 79 – 83.Gravel, D. et al. 2006. Reconciling niche and neutrality: the continuumhypothesis. – Ecol. Lett. 9: 399 – 409.Green, J. L. and Plotkin, J. B. 2007. A statistical theory for samplingspecies abundances. – Ecol. Lett. 10: 1037 – 1045.Green, J. L. et al. 2003. Species richness, endemism, and abundancepatterns: tests of two fractal models in a serpentinegrassland. – Ecol. Lett. 6: 919 – 928.Harte, J. et al. 2005. A theory of spatial structure in ecologicalcommunities at multiple spatial scales. – Ecol. Monogr. 75:179 – 197.Hubbell, S. P. 2001. The unified neutral theory of biodiversity andbiogeography. – Princeton Univ. Press.Illian, J. et al. 2008. Statistical analysis and modelling of spatialpoint patterns. – Wiley.John, R. et al. 2007. Soil nutrients influence spatial distributionsof tropical tree species. – Proc. Natl Acad. Sci. USA 104:864 – 869.Legendre, P. 2009. Model II Regression. – http://cran.r-project.org/ .Legendre, P. and Legendre, L. 1998. Numerical ecology. –Elsevier.Legendre, P. et al. 2009. Partitioning beta diversity in a subtropicalbroadleaved forest of China. – Ecology 90: 663 – 674.Magurran, A. and Henderson, P. A. 2003. Explaining the excessof rare species on natural species abundance distributions.– Nature 422: 714–716.McGill, B. 2003. Strong and weak tests of macroecological theory.– Oikos 102: 679 – 685.McGill, B. J. et al. 2006. Empirical evaluation of neutral theory.– Ecology 87: 1411 – 1423.McGill, B. J. et al. 2007. Species abundance distributions: movingbeyond single prediction theories to integration within an ecologicalframework. – Ecol. Lett. 10: 995 – 1015.M ø ller, J. and Waagepetersen, R. P. 2004. Statistical inference andsimulation for spatial point processes. – Chapman and Hall/CRC Press.Nee, S. et al. 1991. Lifting the veil on abundance patterns. – Proc.R. Soc. B 243: 161 – 163.Oksanen, J. et al. 2010. vegan: community ecology package. Rpackage ver. 1.17-4.Plotkin, H. B. and Muller-Landau, H. C. 2002. Samplingthe species composition of a landscape. – Ecology 83:3344 – 3356.Plotkin, J. B. et al. 2000. Species – area curves, spatial aggregation,and habitat specialization in tropical forests. – J. Theor. Biol.207: 81 – 99.Potts, M. D. et al. 2004. Habitat heterogeneity and niche structureof trees in two tropical rain forests. – Oecologia 139: 446 – 453.Purves, D. W. and Pacala, S. W. 2006. Ecological drift in nichestructuredcommunities: neutral pattern does not imply neutralprocess. – In: Burslem, D. et al. (eds), Biotic interactionsin the tropics. Cambridge Univ. Press, pp. 107 – 138.Seidler, T. G. and Plotkin, J. B. 2006. Seed dispersal and spatialpattern in tropical trees. – PLoS Biol. 4: 2132 – 2137.Shen, G. C. et al. 2009. Species-area relationships explained by thejoint effects of dispersal limitation and habitat heterogeneity.– Ecology 90: 3033 – 3041.Š izling, A. L. et al. 2009. Species abundance distribution resultsfrom a spatial analogy of central limit theorem. – Proc. NatlAcad. Sci. USA 106: 6691 – 6695.Stoyan, D. and Stoyan, H. 1994. Fractals, random shapes, andpoint fields : methods of geometrical statistics. – Wiley.Sugihara, G. et al. 2003. Predicted correspondence between speciesabundances and dendrograms of niche similarities. – Proc.Natl Acad. Sci. USA 100: 5246 – 5251.Ulrich, W. et al. 2010. A meta-analysis of species – abundance distributions.– Oikos: 1149 – 1155.Volkov, I. et al. 2003. Neutral theory and relative species abundancein ecology. – Nature 424: 1035 – 1037.Volkov, I. et al. 2005. Density dependence explains tree speciesabundance and diversity in tropical forests. – Nature 438:658 – 661.Volkov, I. et al. 2007. Patterns of relative species abundance inrainforests and coral reefs. – Nature 450: 45 – 49.Waagepetersen, R. 2007. An estimation function approach to inferencefor inhomogeneous Neyman-Scott processes. – Biometrics63: 252 – 258.Waagepetersen, R. and Guan, Y. 2009. Two-step estimation forinhomogeneous spatial point processes and a simulation study.– J. R. Stat. Soc. Ser. B 71: 685 – 702.Webster, R. and McBratney, A. B. 1989. On the Akaike informationcriterion for choosing models for variograms of soil properties.– Eur. J. Soil Sci. 40: 493 – 496.Wiegand, T. et al. 2007. Species associations in a heterogeneous SriLankan dipterocarp forest. – Am. Nat. 170: E77 – E95.Zhang, L. W. et al. 2011. Strong plant–soil associations in aheterogenous subtropical broad-leaved forest. – Plant Soil,doi: 10.1007/s11104-011-0839-2.Zhu, Y. et al. 2010. Density dependence is prevalent in a heterogeneoussubtropical forest. – Oikos 119: 109 – 119.Zillio, T. and Condit, R. 2007. The impact of neutrality, nichedifferentiation and species input on diversity and abundancedistributions. – Oikos 116: 931 – 940.Supplementary material /available online as AppendixO19428 at www.oikosoffice.lu.se ). Appendix A1 – A5244231

Oikos 119: 109119, 2010doi: 10.1111/j.1600-0706.2009.17758.x,# 2009 The Authors. Journal compilation # 2009 OikosSubject Editor: Thorsten Wiegand. Accepted 5 June 2009Density dependence is prevalent in a heterogeneous subtropicalforestYan Zhu, Xiangcheng Mi, Haibao Ren and Keping MaY. Zhu, X. C. Mi, H. B. Ren and K. P. Ma (kpma@ibcas.ac.cn), State Key Laboratory of Vegetation and Environmental Change, Inst. ofBotany, Chinese Academy of Sciences, 20 Nanxincun, Xiangshan, CNBeijing 100093, China.Although negative conspecific density dependence among neighbours is widely studied, the general prevalence of theeffects is still poorly understood due to a lack of studies from zonal forests other than the tropics. In addition, thedetection of density dependence may be confounded by the influence of habitat heterogeneity. Here we examinedthe spatial distributions of 47 common tree species (diameter at breast height]1 cm) using the pair-correlation functiong(r) in a fully mapped 24-ha subtropical forest in China. We first investigated whether habitat heterogeneity influencedtree distributions, and then examined the conspecific tree patterns and density dependence after removing the effects ofhabitat heterogeneity. We found that the forest plot exhibited strong large-scale heterogeneity in the distribution of bothlarge adult trees of different growth forms and individual species. After the habitat heterogeneity was accounted for, 39 ofthe 47 species (83.0%) were found to exhibit density dependence predominantly at close distances among neighbors. Ourfindings highlight density dependence as a prevalent mechanism for regulating the population spatial structure of mosttree species in the species-rich subtropical forest studied here. Furthermore, the occurrence of density dependence isclosely associated with species abundance and the strength of conspecific aggregation at local scales. Abundant specieswith high strength of conspecific aggregation tend to show density dependence.Since Janzen (1970) and Connell (1971) reported theimpairing performance of conspecific neighbours in tropicalforest communities 40 years ago, negative density dependencehas long been considered a potentially importantmechanism regulating population dynamics and facilitatingspecies coexistence (Wright 2002). Numerous studies ofindividual species have found evidence for negative densitydependence (Hubbell et al. 1990, Zimmerman et al. 2008).Some studies even directly verified that pests and pathogenscould cause density-dependent mortality (Bell et al.2006). Community-level evaluations of conspecific densitydependence have been provided by recent articles fromfully-censused tropical forest plots. For example, Harmset al. (2000) have shown prevailing evidence of densitydependent recruitment at seed-to-seedling transition andseedling stages, further verifying that density dependencepromotes coexistence of diverse species in the 50-ha BarroColorado Island (BCI) plot in Panama. Strong densitydependence was also found to be responsible for themortality of over 80% of tree species in the BCI plot andPasoh plot of Malaysia (Peters 2003) among trees largerthan 1 cm dbh. Nonetheless, there has been only weakevidence of density dependence in other zonal forests ascontrasted with the tropics (Hille Ris Lambers et al. 2002,Hyatt et al. 2003). Carson et al. (2008) found that 81% ofstudies of density dependence were from tropical forestplots, mainly conducted in BCI and Pasoh, and few studiesin subtropical forests have found evidence for densitydependence at the community level (Connell et al. 1984).Therefore, an important question is raised: are thesefindings from the tropical forests geographically biased?No comparable results from other large-scale subtropicalforest plots have been reported so far.Although the above-mentioned studies have pointedto the importance of density-dependent regulation on thedynamics of tree populations in tropical forests, densitydependence may be confounded by other factors, notablyhabitat heterogeneity (Plotkin et al. 2000, Wiegand et al.2007, Murrell 2009). For example, spatial habitat heterogeneity(e.g. less suitable habitats versus favorable habitats)might result in higher mortality in less suitable habitatsand increased survival in favorable habitats (Wright 2002,Getzin et al. 2008). He and Duncan (2000) found thatresults of detecting density-dependent effects could bealtered by controlling for the confounding influence ofelevation on tree survival, and they emphasized theimportance of considering habitat factors in exploringdensity dependence effects. These studies suggested theneed to control for or remove habitat heterogeneity ininferring the importance of density dependence in regulatingtree communities.There are several possible density dependent effects,e.g. distance dependence of increased mortality of offspringnear parent trees (JanzenConnell hypothesis) (Hubbell232109

et al. 1990, Condit et al. 1992, Hyatt et al. 2003), densitydependent thinning (random mortality hypothesis) (Kenkel1988), and a community compensatory trend (CCT)describing an inverse interspecific relationship betweenabundant species and rare species (Connell et al. 1984).The common method of detecting these effects is tocompare growth or mortality of individuals with the densityof a focal species, or the distance from parent trees (Wright2002). Another widely used method is to infer densitydependence from the spatial distribution of trees (Ford1975, Kenkel 1988, Getzin et al. 2006, Wiegand et al.2007). Density-dependence is a mechanism occurringamong neighbours due to resource competition, pestfacilitation, and allelopathy (Janzen 1970, Wright 2002).If there is strong density dependence, the distribution oflive trees that survived the effects is expected to be moreregular and conspecific clustering is expected to declinewith increasing size classes (Sterner et al. 1986, Barot et al.1999). Thus, comparison of the spatial patterns of differentsize classes may be used to indirectly detect densitydependence if no data on tree mortality is available. Thismethod requires a temporal invariance of the spatialpatterns of the different size classes.Although spatial point pattern analysis is an effectiveapproach for detecting density dependence, the methodsuffers from the problem that habitat heterogeneity can alsosubstantially influence the spatial distribution of trees bothtoward aggregation (more common, Getzin et al. 2008)and toward regularity (less common, Chapin et al. 1989).Therefore, it is critical to take into account habitatheterogeneity when evaluating density dependence.Following this line of reasoning, Getzin et al. (2008)analyzed density-dependent thinning in western hemlockpopulations using adults of western hemlock as a ‘control’to account for the biasing effects of habitat heterogeneity.Another work that explicitly considered habitat heterogeneityis the one of Wiegand et al. (2007) who analyzed largesize trees (10 cm in dbh) in the Sinharaja forest of SriLanka and found 24 of 46 common tree species showedregular or random patterns after conditioning on the largescale patterns which were likely to be driven by habitatheterogeneity. These studies have greatly improved ourunderstanding of the magnitude of density dependence inforest communities and have also provided sophisticatedanalytical tools to make unbiased inferences of densitydependence.In this study, we explore the prevalence of densitydependence in a species-rich subtropical forest consideringthe potentially confounding effect of habitat heterogeneity.We also study the effect of species abundance andconspecific aggregation on the detection of density dependence.We analyze data from a newly established 24 hastem-mapping plot, the first of its kind in a subtropicalforest in China, named Gutianshan Forest Dynamic Plot(hereafter Gutianshan FDP; Legendre et al. 2009). Morecomprehensive than Wiegand et al. (2007), our analysisincludes trees with dbh]1 cm. By including small trees theeffects of density dependence will be more evident, sincesmall trees are likely more susceptible to the effects.Our study consisted of three analytical steps. We firstassessed whether the environmental conditions of thestudy plot were heterogeneous. Second, we factored outlarge-scale effects in the spatial patterns (beyond the scaleof direct treetree interactions) to selectively study smallscaleeffects (i.e. conspecific aggregation and regularity)which are likely to be caused by treetree interactionsbut could be caused by small-scale heterogeneity as well.This analysis allowed us to examine whether the majorityof examined species are still spatially aggregated afterremoving confounding large-scale effects and explain theoccurrence of density dependence combined with conspecificaggregation. Finally, we used a case-control designto examined ‘pure’ density dependence after factoring outhabitat heterogeneity, and explored the relationship ofdensity dependence with species abundance and conspecificaggregation.Material and methodsStudy site and data collectionGutianshan FDP is located in Gutianshan NationalNature Reserve (29810?19ƒ29817?41ƒN, 118803?49ƒ118811?12ƒE) of eastern China (Fig. 1). Mean temperatureat Gutianshan Reserve ranges from 4.38C in Januaryto 27.98C in July with an annual mean of 15.38C. Themean annual frost-free period lasts 250 days. The meanannual precipitation is 1963.7 mm. There are two wetseasons, from March to June and September, and two dryseasons, from July to August and October to February.The plot encompasses a complicated and steep terrain withelevation ranging from 446.3 to 714.9 m a.s.l., and themean slope is about 37.528 and ranges from 12.798 to 628(Fig. 1).The 600400 m plot was established from November2004 to September 2005 as a part of the Chinese ForestBiodiversity Monitoring Networks with the aim to monitorlong-term changes in a subtropical evergreen broadleavedforest. All woody stems ]1 cm dbh in the plot weremapped, measured, identified to species and tagged. Wedocumented 140 700 individuals ]1 cm dbh belongingto 49 families, 103 genera and 159 species. Fagaceae,Lauraceae, Theaceae and Magnoliaceae are the dominantfamilies in the old-growth forest (Zhu et al. 2008). Specieswere grouped by maximum attainable height into threegrowth forms: shrubs (B5 m), understory trees (]5 andB15 m), and canopy trees (]15 m).To obtain a sufficiently large sample size, we analyzedthe spatial pattern of 47 common species with ]40individuals at each life history stage (Supplementarymaterial Appendix 1). These species comprised 90.1% ofthe stems in the plot.Point pattern analysisWe mainly used the pair-correlation functions g(r) toanalyze the spatial pattern of tree distributions at differentscales. The pair-correlation function g(r) is a counterpart tothe widely used K-function (Ripley 1976, Wiegand andMoloney 2004, Illian et al. 2008), which is calculated as:110233

Figure 1. The location and contour map of the 24-ha Gutianshan FDP, China. The number in the contour map is relative height (m),and the unit of (x, y) axes is meters.g(r) 1 dk(r)2pp d(r)K(r) is a cumulative distribution function where K(r) is theexpected number of points within the entire circle of agiven radius r around a typical point of the pattern dividedby the intensity l of the pattern. The values of K(r) atlarger scales include the values of K(r) at smaller scales,so that K(r) cannot detect significant pointpoint interactionsat specific spatial scales (Wiegand and Moloney2004, Loosmore and Ford 2006). In contrast, the g(r)is a non-accumulative distribution function in which g(r)is the expected density of points in a ring of a givendistance r around a focal point divided by the intensityl of the pattern (Stoyan and Stoyan 1994, Dale et al.2002). The g-function may effectively identify what pointpoint interactions occur at specific scales and estimatethe strength of aggregation/regularity. If g(r)1 there aremore points at distance r than expected under a randompattern, which indicates aggregation at scale r.Biological questions and null modelsAnalysis 1. The effects of habitat heterogeneity onspatial patternsHabitat heterogeneity and the effects of plant interactionsmay cause locally elevated point densities and are thereforedifficult to separate. In general, only certain aspects ofheterogeneity can be detected (Illian et al. 2008). Getzinet al. (2008) used a relatively simple method to test forlarge-scale effects of heterogeneity on local tree density.They argued that mature trees can be expected to exploit allavailable sites (except canopy gaps) which have undergoneexcessive thinning due to habitat quality, so that thecombined spatial pattern of large adult trees of all speciesshould capture strong habitat factors common to all species.Thus, aggregation of all adults at larger scales (e.g. r10 min boreal forests) should indicate heterogeneity (Stoyan andPenttinen 2000, Wiegand et al. 2007, Getzin et al. 2008).Note that there may be effects of plantplant interactionsat small scales.(1)The pair correlation g-function and the L-function(the transformed K-function L(r)[k(r)/p] 0.5 r) show atypical behavior if the intensity of the pattern varies at largerscales (Wiegand and Moloney 2004). In this case theg-function does not approach the expected value of g(r)1for scales rrp where rp is the scale at which directtree-tree interactions should become unimportant. TheL-function shows a strong increase and does not drop forlarger scales to the expected value of L(r)0. Cumulativeproperties of L(r) at larger scales (Besag 1977) facilitatevisual interpretation of large-scale habitat heterogeneity, sowe also used L(r) to test it.We analyzed the joined spatial pattern of all old adulttrees of three different growth forms separately: shrubs withdbh]3 cm, under-story trees with dbh]10 cm, andcanopy trees with dbh]20 cm. The joined spatial patternof all old adult trees of a given growth form should reflect theeffects of heterogeneity common to all species. However, thejoined spatial pattern of all old adult trees may overrideeffects of individual niches and of species where the adults donot show aggregation at larger scales. Therefore we furtheranalyzed the spatial pattern of adults of each speciesindividually (Supplementary material Appendix 2).Analysis 2. Population spatial patterns after removinglarge-scale effectsTo investigate the second-order characteristics (i.e. treetreeinteractions) of the spatial patterns of tree species, we usedthe heterogeneous Poisson process (HP; Wiegand et al.2007) as the null model. Our approach is based on theassumption of separation of scales (Wiegand et al. 2007);in other words, habitat heterogeneity influences spatialdistribution of trees only at larger scales, typically alonggradients related to topography, whereas direct treetreeinteractions take place at smaller scales. In this case, theheterogeneous Poisson process can reveal the second-ordercharacteristics of the spatial patterns of tree species byconditioning on the large scale pattern.The heterogeneous Poisson process displaces the originallocation of all trees randomly within a circular moving234111

window of radius R. This small displacement destroys smallscale patterns of the locations of the trees while keeping thelarge scale pattern unchanged. This allows us to selectivelystudy small-scale effects. However, this method is not ableto disentangle potential effects of small-scale heterogeneities(e.g. microsites see Fajardo and McIntire 2007) and treetree interactions.In order to implement the heterogeneous Poissonprocess, we first estimated the intensity function l(x, y)of the pattern by using a moving window of bandwidth R(Wiegand and Moloney 2004, Wiegand et al. 2007). Theintensity l(x, y) is conducted by summing all stems ofthe circle window and then weighting them with a nonparametrickernel (Wiegand and Moloney 2004, Wiegandet al. 2007). The nonparametric kernel is defined as:8< 31 d2R 5d5Re R (d) 4R R 2 (2):0; otherwisewhere R is the bandwidth, d is the distance from a focalpoint. In the second step we generated random pointswithin the study plot but retained a point with a probabilitygiven by the intensity function l(x, y). This producespatterns with intensity function l(x, y). We selected aconservative value of R30 m, and expected that thesignificant aggregation of treetree interactions shouldoccur at small scales and disappear well below 30 m. Ifthis is not the case the assumption of separation of scalesmay not hold (Wiegand et al. 2007).We used a goodness-of-fit (GOF) test to assesssignificant departures from the null model. The p-valueof the observed pattern is calculated as followsˆp1 rank[u 0 ] 1(3)swhere u 0 is a summary statistic that measures thediscrepancy between the empirical and the expected paircorrelation function over a distance interval of interest. Thefunction rank[u 0 ] returns the rank of u 0 within the valuesof the corresponding summary statistics u i (i1, . . . s)for each of the s simulations of the null model. We furtheranalyzed only those data sets with an observed p-valueB0.005 and a rank995 (Loosmore and Ford 2006,Wiegand et al. 2007).We used Spearman’s rho statistic (Hollander and Wolfe1973) to estimate the relationship between species abundanceand conspecific aggregation measured by the maximalvalue of the pair correlation function, species abundanceand density dependence, and conspecific aggregation anddensity dependence.Analysis 3. Conspecific density dependence using a casecontrol designAccounting for heterogeneity when testing for densitydependence is not easy because numerous environmentalcovariates are difficult to quantify (He and Duncan 2000,Wiegand et al. 2007). Getzin et al. (2008) approximatelysolved the problem through a population-based case-controlstudy using random labeling (RL) as the null model. Theapproach rests on three assumptions: 1) cases (smaller sizeclass trees) and controls (adults) deriving from the samepopulation are exposed to the same stochastic processesunder habitat heterogeneity (Wiegand and Moloney 2004),2) small-sized trees as well as adult trees should beapproximately in an equilibrium stage and not e.g. invadingthe study plot or declining, 3) the large-scale pattern ofadult trees reflects the underlying habitat heterogeneity,which is created by excessive thinning through life-historystages (Getzin et al. 2008). Under the null hypothesis ofrandom labeling, cases are a random sub-sample of thejoined patterns of cases and controls (Wiegand andMoloney 2004, Diggle et al. 2007). By comparing thecase patterns with control patterns, the case-controlapproach may identify specific factors (i.e. density dependence)other than habitat heterogeneity that may influencethe patterns of trees of smaller size classes (cases) up throughincreasing size classes if the null hypothesis is rejected.We used adults as controls (pattern 1), to account forunderlying heterogeneity and smaller size class trees as cases(pattern 2) with g 21 (r)g 22 (r) as test statistics in practicalapplications. Under the random labeling null model,g 21 (r)g 22 (r), but if the cases show additional aggregationwhich is independent from the control pattern we findg 21 (r)g 22 (r)B0 (Getzin et al. 2008). We expected theextent of additional aggregation to decline with size classes,thus revealing the effects of density dependence on theestablishment of cases.A formula can express this: d(r)d j (r)d s (r). Thefunction d j (r) returns the value taken by g 21 (r)g 22 (r)over the scales r when the ‘cases’ are juveniles and thefunction d s (r) is the value when the ‘cases’ are saplings.The function d(r) reflects density-dependent thinningif d(r)0, d max is the maximum strength of densitydependentthinning when d(r) takes the maximal value.The r thin reflects the scales at which density-dependentthinning takes place, and r max is the scale at which thestrength of density-dependent thinning peaks. We providedan example to illustrate our method on how to analyzedensity dependence by using the spatial distribution ofDistylium myricoides (Fig. 2ac).We classified population of different growth forms intothree size classes to define three life history stages: sapling,juvenile, and adult (see examples in Fig. 3). For shrubs, weclassified individuals of 11.5 cm dbh as saplings, 1.52 cmdbh as juveniles, and ]2 cm dbh as adults; similarly, forunder-story trees, we considered 12.5 cm dbh as saplings,2.55 cm dbh as juveniles, and ]5 cm dbh as adults; andfor canopy trees, we categorized 15 cm dbh as saplings,510 cm dbh as juveniles, and ]10 cm dbh as adults.In all analyses, we used a grid size of 1 m 2 and a ring widthof 3 m for analysis of plantplant interactions at spatial scalesof 0 m up to 30 m. This is a fine resolution compared to the600400 m size of the study plot and is sufficient to capturedetailed variation in the pair-correlation function over therange of scales where we expected significant effects up to30 m. Significant departure from null models was evaluatedusing the 5th-lowest and 5th-highest value of 999 MonteCarlo simulations of the null model to generate approximately99% simulation envelopes.All point pattern analyses in this paper were done usingsoftware R 2.6.0 (R Development Core Team) and the gridbasedsoftware Programita (Wiegand and Moloney 2004).112235

Figure 2. Example of the analysis of conspecific density dependence (a, b, c) and the relationship of the maximal strength of densitydependence with species abundance (d). Distylium myricoides as an example illustrates the analysis with a case-control where the largerscalepattern of adult trees as controls accounts for the effects of habitat heterogeneity. The function d s (r) returns the value taken byg 21 (r)g 22 (r) over the scales r when the ‘cases’ are saplings, the function d j (r) returns the value when the ‘cases’ are juveniles. Thefunction d(r) reflects the strength of density dependence if d(r)d j (r)d s (r) 0, d max is the maximum strength of density dependencefor per species when d(r) takes the maximal value. (a) If d s (r)B0, saplings exist in additional aggregated pattern independent from adults.(b) If d j (r)B0, and d(r)0 (c), the strength of additional aggregation declines from saplings to juveniles, which indicates densitydependence works without regard to habitat heterogeneity (d max 7.4 at scale of 0 m). (d) The horizontal axis is species abundance inlogarithm base 2 and the vertical axis is the d max of each species. Approximately 99% simulation envelopes (dashed lines) were constructedusing the 5th-lowest and 5th-highest value of 999 Monte Carlo simulations of the random-labeling null model.ResultsThe effects of habitat heterogeneity on spatial patternWe analyzed the joint patterns of all adults for canopy trees,under-story trees and shrubs in the plot with the CSR nullmodel to investigate whether the plot showed large-scaleheterogeneity or not (Fig. 4ac). Habitat heterogeneitywas indicated by the departure of the g-function andthe L-function from CSR. For each growth form, theL-function showed a clear large-scale aggregation even upto scales of r30 m. In addition, we analyzed the patternsFigure 3. Examples with maps of species (Distylium myricoides), all individuals (a), 12.5 cm dbh as saplings (b), 2.55 cm dbh asjuveniles (c), and ]5 cm dbh as adults (d) in the Gutianshan FDP. The unit of (x, y) axes is meters.236113

of adults of individual species (Supplementary materialAppendix 2). Adults of all 47 tested species showedaggregation at larger scales even up to 30 m by theL-function; only two species showed aggregation atrelatively small scales of 10 m and 12 m by the g-function.These findings provided evidence that the GutianshanFDP exhibited large-scale heterogeneity, suggesting that weshould use specific methods to account for habitatheterogeneity.Population spatial patterns after removing large-scaleeffectsAll 47 species examined and contrasted to the HP nullmodel showed significant aggregation (i.e. the rank of theGOF test was995), 19 (40.4%) of these species exhibitedsignificant regularity at detailed scales (Supplementarymaterial Appendix 2, Fig. 5b).The maximal strength of conspecific aggregation(g max , the maximal value g(r) at a given scale) was negativelycorrelated with species abundance (Spearman’s rho0.55, S26762, pB0.0001). The g max of all examinedspecies occurred at scales of 02 m, so g max valuesapproximately represented the strength of populationaggregation at local scales.We assessed the scale-dependent effects of the secondordercharacteristics of population patterns by calculatingthe percentage of species exhibiting different spatial patternsat each detailed scale r (i.e. being outside the simulationenvelopes). The percentage of aggregation (95.7%) peakedat the scale of 1 m, then decreased sharply with increasingr, and reached 0 at a scale of approximately 26 m butwell below 30 m. In contrast, the percentage of regularityincreased with the increasing scales of 1330 m (Fig. 5b).However, this regularity is an effect of the null model incombination with strong small-scale clustering. The heterogeneousPoisson process distributes the points of small-scaleclusters randomly over a 30 m neighborhood, therebyproducing distances greater than 20 m, higher than observedneighborhood densities (e.g. Fig. 5a).Conspecific density dependence using a case controldesignFor saplings, 40 of 47 tested species exhibited additionalaggregated patterns relative to adults, six species showedrandom patterns, and three species showed more regularpatterns. Of juveniles, however, only 29 of 47 tested speciesexhibited additional clustering relative to adults, 17 speciesshowed random patterns, and three species showed regularpatterns. The percentage of species displaying additionalclustering relative to adults decreased from saplings tojuveniles over examined scales (Fig. 6ab).We found that 39 (83.0%) of the 47 tested speciesshowed a decline of strength of additional clustering fromsaplings to juveniles (Table 1, Supplementary materialAppendix 2), indicating that the majority of species showeddensity dependence across the subtropical forest plot. Eight(17.0%) species did not display a decline of strength of theadditional clustering within size classes, including fourabundant species (n1000) (e.g. Schima superba, Lithocarpusglaber) and four less abundant species (nB1000)(e.g. Elaeocarpus japonicus, Ilex elmerrilliana).The maximum decline d max in clustering was negativelycorrelated with species abundance (Spearman’s rho0.53, S15072, p0.0007) (Fig. 2d), and positivelyFigure 4. Analysis of the pattern of all old adult trees. (a) adults of canopy trees with a dbh]20 cm; (b) adults of under-story trees with adbh]10 cm; (c) adults of shrubs with a dbh]3 cm. The analysis used the homogeneous pair-correlation g-function and L-function. Theconfidence limits (dashed lines) were constructed from the 5th-lowest and 5th-highest g(r) values of 999 simulations of a null model ofcomplete spatial randomness (CSR). The ring width was 3 m.114237

As a stabilizing force in the maintenance of species diversity,negative density dependence is one of the essentialmechanisms for species coexistence in a forest community(Chesson 2000, Zimmerman et al. 2008). However, fewstudies have supported the generality of density dependencefrom other zonal forests as contrasted with the tropics, andprevious studies testing density dependence were confoundedby other factors, particularly habitat heterogeneity(He and Duncan 2000, Wright 2002, Carson et al. 2008).Our study, which approximately considered the effect ofhabitat heterogeneity, demonstrated that density dependenceis prevalent among most species at the communitylevel in the Gutianshan FDP subtropical forest. Our studyalso provided evidence that habitat heterogeneity canintroduce bias in the detection of density dependence.The disentanglement of the confounding effects of thisheterogeneity enables a clear, unbiased evaluation of therelative importance of density dependence in regulatingpopulation structure.Habitat heterogeneityFigure 5. Analysis of population spatial pattern. (a) as an example,Distylium myricoides was used to illustrate the approach comparingthe actual population pattern to the heterogeneous Poison nullmodel (HP) by means of function l (x, y) varying with location(x, y) and accounting for habitat heterogeneity. The simulationenvelopes (dashed lines) were constructed using the 999 simulationsof the null model. The solid circles denote the paircorrelationfunctions of the observed data over scale r. The ringwidth of the pair-correlation function was 3 m. (b) Proportion ofspecies showing significant aggregation (solid circles), regularity(open circles) and no interaction (open squares) over differentscales in the Gutianshan FDP.with the maximal strength of conspecific aggregation(Spearman’s rho0.70, S2936, pB0.0001).The percentage of cases showing density dependencedecreased with increasing scales r (Fig. 6c). Meanwhile, themaximal strength of density dependence predominantlyoccurred at scales smaller than 14 m (Fig. 6d). Thirtysix(92.3%) species reached a maximum of density dependenceat scales of 0 to 5 m, with 21 species having d max peak at ascale of 0 m (in a 11 m grid cell). This suggests thatdensity dependence occurs at very close distances amongneighbors. This result implies that density dependenceoccurs predominantly at local scales and d max valuesapproximately indicated the strength of density dependenceat these scales.DiscussionHabitat heterogeneity may introduce a bias in thedetection of density dependence. The same method wasused in both cases with and without factoring out habitatheterogeneity, but results were different (Fig. 2, 6,Supplementary material Appendix 3). With factoring outhabitat heterogeneity, the percentage of cases showingdensity dependence was higher at small scales than that atlarger scales (Fig. 6c) when the adult pattern was used tocontrol for possible large-scale heterogeneity. Withoutfactoring out habitat heterogeneity, the percentage of casesshowing density dependence was not different at smallerscales from that at larger scales when the randomized adultpattern was used as pattern 1 (Supplementary materialAppendix 3c). Possible heterogeneity cannot be removed,thus the decline in aggregation with increasing size classesat larger scales may be driven by unfavorable habitatinstead of density dependence; in addition, signatures ofdensity dependence at smaller scales may be overriddenbecause the increase of establishment driven by favorablehabitat offsets the reduction of conspecific trees caused bydensity dependence (Wright 2002). The number of speciesshowed density dependence was also different with andwithout factoring out habitat heterogeneity. Five of 47examined species did not show density dependencewithout factoring out habitat heterogeneity. Of thesefive species, only Camellia fraterna and Schima superbadid not display density dependence with factoring outhabitat heterogeneity. Additionally, the maximum declined max in clustering also was negatively associated withspecies abundance (Spearman’s rho0.39, S17200,p0.0103) (Supplementary material Appendix 3e) withoutfactoring out habitat heterogeneity, but the resultcannot express the relationship of the extent of densitydependence with species abundance. As shown in Fig. 4and Supplementary material Appendix 2, GutianshanFDP exhibited strong large-scale heterogeneity in the distributionof both large adult trees of different growthforms and individual species. Recent studies from GutianshanFDP (Gong et al. 2007, Legendre et al. 2009) alsodemonstrated that strong habitat heterogeneity accountedfor 25.834.7% of the variation in species compositionof the forest community in Gutianshan FDP. Habitatheterogeneity can confound the detection of densitydependence. Therefore, it is necessary to disentangle theconfounding effect in order to detect the ‘true’ densitydependent effects.238115

Figure 6. The analysis of scale-dependence of density dependence. (a, b) Proportion of species showing the test statistic g 21 (r)g 22 (r)B0(solid circles), g 21 (r)g 22 (r)0 (open circles) and g 21 (r)g 22 (r)0 (the data are inside the simulation envelopes, open squares) overscales at the Gutianshan FDP. (a) saplings as ‘cases’, (b) juveniles as ‘cases’, (c) proportion of examined species showing densitydependence at detailed scales. (d) The number of tested species for which the strength of density dependence peaked at the range of scalesof 0 to 30 m. An arrow denotes that no species shows the maximal strength of density dependence beyond 14 m.Conspecific density dependenceThe prevalence of density dependenceThe majority of evidence for the prevalence of densitydependence comes from tropical forests, with only limitedevidence from other zonal forests (Carson et al. 2008). Forexample, Connell et al. (1984) found little evidence fordensity dependent growth or mortality at the communitylevel through 18 years of study in an Australian subtropicalforest. In the Gutianshan FDP, 83.0% of examined speciesdisplayed density dependent effects among established trees(]1 cm dbh) (Table 1). This finding clearly providedsupport for the generality of density dependence afteraccounting for heterogeneity in a subtropical forest.Additionally, it is worth an emphasis that the findingbenefits from the analysis of fine scales. Not like otherstudies that used larger scales of ]5 m for this analysis(Wills et al. 1997, Peters 2003), our study applied a 1 mfine-scale of pair correlation function g(r). As densitydependence was strongest among neighbours at a 1 m scale,the analysis at fine scales may increase the probability ofdetecting density-dependent signatures, and facilitates usdetecting the prevalence of density dependence.The proportion of species in our study affected bydensity dependence at the community level was similar tothat found in other tropical forests (Wills et al. 1997,Harms et al. 2000, Peters 2003, Uriate et al. 2004). Thissuggests that density dependence is not more prevalent intropical forests, although the number of species exhibitingdensity dependence in studies by Peters (2003)was considered to be overestimated (Leigh et al. 2004,Zimmerman et al. 2008). In temperate deciduous forests,Hille Ris Lambers et al. (2002) also found the proportionof species exhibiting density dependence was equivalent toTable 1. The number of species examined (]100 trees in the forest plot) in each growth form, and the number of species exhibiting densitydependence or not (The percentage of species showing such effects follows in parenthesis).Growth forms Examined No. species showing density dependent effects or notshowing the effectsnot showing the effectsCanopy trees 18 15 (83.3) 3 (16.7)Under-story trees 24 21 (87.5) 3 (12.5)Shrubs 5 3 (60.0) 2 (40.0)In total 47 39 (83.0) 8 (17.0)116239

that in tropical forests. The prevalent extent of densitydependence is similar in different zonal forests acrosslatitudes, which suggested that density dependence cannotexplain why species is more diverse in tropical forests thanin other zonal forests unless the strength of densitydependence could be stronger in tropical forests (HilleRis Lambers et al. 2002, Carson et al. 2008).Ecologists expected that density dependent effects shouldbe stronger in the younger life-history stages of trees dueto higher susceptibility to pathogens, pests and abioticstressors, and there had been few studies of densitydependence for established trees at the community level(Condit et al. 1992, Wills et al. 1997, Peters 2003). Thus,Carson et al. (2008) suggested that ecologists should paygreater attention to density dependent effects occurringin later life-history stages. In this study, we demonstratedthat density dependence is important from the sapling tojuvenile stages of tree species in the entire forest ofGutianshan FDP. Stoll and Newberry (2005) also foundstrong density dependent effects for tree species withdbh]10 cm in a 10-year study of a Malaysian tropicalforest. These results, together with those of Peters (2003),suggested that density dependent effects also worked onestablished trees of entire communities, but little wasknown about the impact of density dependence on speciesdiversity (Harms et al. 2000, Wills et al. 2006). Suchstudies are critical to understanding the relative importanceof density dependence in maintaining species diversity.Density dependence and species abundanceThe relationship between species abundance and densitydependence is still a hot issue in the field of communityecology (Carson et al. 2008). We found that the maximumdecline in clustering from saplings to juveniles (d max ) of lessabundant species was higher than that of abundant species(Fig. 2d) predominantly at scales of rB5 m. This isconsistent with the result that less abundant species had ahigher maximum strength of aggregation g max at scales of02 m, and tended to aggregate at narrow sites.Although in species showing density dependence themean value (6.5) of g max for abundant species was muchlower than that (13.8) of less abundant species, theproportion of abundant species showing density dependence(85.7%) was slightly higher than that of lessabundant species (78.9%). This indicates that densitydependence tends to take place in abundant species.Recently, Murrell (2009) also proposed that densitydependentthinning and subsequent pattern formationmay strongly depend on the mutual effects of scaledependentheterogeneity and species abundance. If heterogeneityoccurs at intermediate scales, then it does affecttree-tree interactions and species succession directly (Getzinet al. 2008). However, if abiotic heterogeneity occurs overlarge spatial scales relative to individual interactions andseed dispersal, then density-dependent effects and theresulting patterns may be governed by species abundanceas well as by individual species identity (Murrell 2009). Ourfindings also lead us to argue that conspecific densitydependence is likely less prevalent in tropical forests withhigh species diversity and fewer highly abundant speciesthan in temperate forests (Wright 2002).Density dependence and conspecific aggregationDensity dependence has a close association with conspecificaggregation. After factoring out large-scale effects, allexamined species exhibited locally aggregated patterns(Fig. 5b), which was consistent with prevalent densitydependence at local scales (Fig. 6c, 6d, Supplementarymaterial Appendix 2). Moreover, the maximum declined max in clustering from sapling to juveniles was positivelycorrelated with the maximum strength of aggregation g max .Lithocarpus glabra (n1309) and Camellia fraterna(n4108) had much lower g max ’s (3.6, 4.6, respectively)than the mean g max value (9.6) of all examined species,which could partially explain why these two abundantspecies did not show density dependence. In contrast,Cleyera japonica, a less abundant species (n479), hadthe highest value of g max (56.7) and the highest value ofd max (63.4) of all examined species. Hardy and Sonké(2004) suggested that dispersal limitation is the major factoraffecting the degree of conspecific aggregation, althoughhabitat heterogeneity also affects the distribution of manyspecies. Legendre et al. (2009) confirmed that spatialprocesses, such as dispersal limitation, could account for53.262.6% of the variation of community composition inthe forest community in Gutianshan FDP at scales (unitsize) from 2020 to 5050 m. Seeds of short-distancedispersal can only arrive at certain parts of suitable habitats(Howe and Smallwood 1982, Nathan and Muller-Landau2000, Wright 2002), which leads to higher local aggregationintensity. Dispersal limitation leads to the aggregationof plant individuals, and interspecific exclusion is rare in theinterior of a cluster. In addition, conspecific interactionswill easily take place in the proximity of conspecificneighbours (Chesson and Neuhauser 2002). ThoughHubbell et al. (2001) suggested that local density dependencecould not maintain species diversity at the communitylevel, density dependence, niche differentiation andinterspecific facilitation can be responsible for increasedspecies diversity with time (Wills et al. 2006). However,aside from species abundance and conspecific aggregation,other species traits which we did not examine, such as shadetolerance and water use efficiency for different species,might also contribute to density dependence (Comita andHubbell 2009).The potential caveat of this study is that we cannotremove possible temporal and spatial disturbances overridingsignatures of density dependence, e.g. ice-storms andnatural fire. On the other hand, studies of densitydependence in tropical forests usually focus on analyzingmortality patterns of focal species after a few years ofobservation, which is too short to find the lagged effectof density dependence for established trees that have alifespan of several hundred years (Ratikainen et al. 2008).Therefore, detection of density dependence using static dataof multiple life-history stages can take advantage of theconsideration of the lagged effects of density dependence.Interspecific interactions also might influence the detectionof density dependence but the strong conspecific clusteringfound at the Gutianshan FDP may cause intraspecificsegregation which will reduce the probabilities ofheterospecific encounters and therefore the importance ofinterspecific interactions (Wiegand et al. 2007). However,240117

additional studies are required to quantify the effect ofintraspecific interactions.In summary, our study focused on the spatial patternfound in Gutianshan FDP and clearly highlighted densitydependence as a prevalent mechanism regulating thepopulation structure of woody species in a subtropicalforest. Our findings revealed that it is crucial to remove theconfounding effect of habitat heterogeneity in studies ofdensity dependence. Our findings also reflected that theoccurrence of density dependence has a close associationwith high species abundance and the strength of conspecificaggregation at local scales.Acknowledgements We are grateful to Dr Fangliang He forinsightful suggestions on the revision of the ms, and Fang Teng,Chen Shengwen and Prof. Ding Bingyang for species identification.Dr. Thorsten Wiegand helped with guidance of data analysisand insightful comments. We also thank many field workers fortheir contributions to the establishment and census of the 24 hapermanent forest plot. We gratefully acknowledge the supportfrom the Administration Bureau of the Gutianshan NationalNature Reserve. This study was financially supported by KeyInnovation Project of CAS (KZCX2-YW-430).ReferencesBarot, S. et al. 1999. Demography of a savanna palm tree:predictions from comprehensive spatial pattern analyses. Ecology 80: 19872005.Bell, T. et al. 2006. Plant pathogens drive density-dependentseedling mortality in a tropical tree. Ecol. Lett. 9: 569574.Besag, J. 1977. Contribution to the discussion of Dr Ripley’spaper. J. R. Stat. Soc. B 39: 193195.Carson, W. P. et al. 2008. Challenges associated with testing andfalsifying the JanzenConnell hypothesis: a review andcritique. In: Carson, W. P. and Stefan, A. S. (eds), Tropicalforest community ecology. Blackwell, pp. 210241.Chapin, F. S. et al. 1989. Competition causes regular spacing ofalder in Alaskan shrub tundra. Oecologia 79: 412416.Chesson, P. 2000. Mechanisms of maintenance of speciesdiversity. Annu. Rev. Ecol. Syst. 31: 343366.Chesson, P. and Neuhauser, C. 2002. Intraspecific aggregationand species coexistence. Trends Ecol. Evol. 17: 210211.Comita, L. S. and Hubbell, S. P. 2009. Local neighborhood andspecies’ shade tolerance influence survival in a diverse seedlingbank. Ecology 90: 328334.Condit, R. et al. 1992. Recruitment near conspecific adults andthe maintenance of tree and shrub diversity in a neotropicalforest. Am. Nat. 140: 261286.Connell, J. H. 1971. On the role of natural enemies in preventingcompetitive exclusion in some marine animals and in rainforest trees. In: Den Boer, P. J. and Gradwell, G. (eds),Dynamics of populations. PUDOC, pp. 298312.Connell, J. H. et al. 1984. Compensatory recruitment, growth,and mortality as factors maintaining rain forest tree diversity. Ecol. Monogr. 54: 141164.Dale, M. R. T. et al. 2002. Conceptual and mathematicalrelationships among methods for spatial analysis. Ecography25: 558577.Diggle, P. J. et al. 2007. Second-order analysis of inhomogeneousspatial point processes using case-control data. Biometrics63: 550557.Fajardo, A. and McIntire, E. J. B. 2007. Distinguishing micrositeand competition processes in tree growth dynamics: an a priorispatial modeling approach. Am. Nat. 169: 647661.Ford, E. D. 1975. Competition and stand structure in some evenagedplant monocultures. J. Ecol. 63: 311333.Getzin, S. et al. 2006. Spatial patterns and competition of treespecies in a douglas-fir chronosequence on vancouver island. Ecography 29: 671682.Getzin, S. et al. 2008. Heterogeneity influences spatial patternsand demographics in forest stands. J. Ecol. 96: 807820.Gong G. Q. et al. 2007. Habitat associations of wood species inthe Gutianshan subtropical broad-leaved evergreen forest, Sci. Soil Water Conserv. 5: 7983, in Chinese.Hardy, O. J. and Sonké, B. 2004. Spatial pattern analysis of treespecies distribution in a tropical rain forest of cameroon:assessing the role of limited dispersal and niche differentiation. For. Ecol. Manage. 179: 191202.Harms, K. E. et al. 2000. Pervasive density-dependent recruitmentenhances seedling diversity in a tropical forest. Nature 404:493495.He, F. L. and Duncan, R. P. 2000. Density-dependent effects ontree survival in an old-growth douglas fir forest. J. Ecol. 88:676688.Hille Ris Lambers, J. et al. 2002. Density-dependent mortality andthe latitudinal gradient in species diversity. Nature 417:732735.Hollander, M. and Wolfe, D. A. 1973. Nonparametric statisticalinference. Wiley.Howe, H. F. and Smallwood, J. 1982. Ecology of seed dispersal. Annu. Rev. Ecol. Syst. 13: 201228.Hubbell, S. P. et al. 1990. Presence and absence of densitydependence in a neotropical tree community. Philos. Trans.R. Soc. Lond. B 330: 269281.Hubbell, S. P. et al. 2001. Local neighborhood effects on longtermsurvival of individual trees in a neotropical forest. Ecol.Res. 16: 859875.Hyatt, L. A. et al. 2003. The distance dependence prediction ofthe JanzenConnell hypothesis: a meta-analysis. Oikos 103:590602.Illian, J. et al. 2008. Statistical analysis and modelling of spatialpoint patterns. Wiley.Janzen, D. H. 1970. Herbivores and the number of tree species intropical forests. Am. Nat. 104: 501528.Kenkel, N. C. 1988. Pattern of self-thinning in jack pine: testingthe random mortality hypothesis. Ecology 69: 10171024.Legendre, P. et al. 2009. Partitioning beta diversity in asubtropical broad-leaved forest of China. Ecology 90:663674.Leigh, E. G. Jr. et al. 2004. Why do some tropical forests havesome many species of trees? Biotropica 36:447473.Loosmore, N. B. and Ford, E. D. 2006. Statistical inferenceusing the G or K point pattern spatial statistics. Ecology 87:19251931.Murrell, D. J. 2009. On the emergent spatial structure of sizestructuredpopulations: when does self-thinning lead to areduction in clustering? J. Ecol. 97: 256266.Nathan, R. and Muller-Landau, H. C. 2000. Spatial patterns ofseed dispersal, their determinants and consequences forrecruitment. Trends Ecol. Evol. 15: 278285.Peters, H. A. 2003. Neighbour-regulated mortality: the influenceof positive and negative density dependence on tree populationsin species-rich tropical forests. Ecol. Lett. 6: 757765.Plotkin, J. B. et al. 2000. Species-area curves, spatial aggregation,and habitat specialization in tropical forests. J. Theor. Biol.207: 8199.118241

Ratikainen, I. I. et al. 2008. When density dependence is notinstantaneous: theoretical developments and managementimplications. Ecol. Lett. 11: 184198.Ripley, B. D. 1976. The second-order analysis of stationary pointprocesses. J. Appl. Probabil. 13: 255266.Sterner, R. W. et al. 1986. Testing for life historical changesin spatial patterns of four tropical tree species. J. Ecol. 74:621633.Stoll, P. and Newbery, D. M. 2005. Evidence of species-specificneighborhood effects in the Dipterocarpaceae of a Borneanrain forest. Ecology 86: 30483062.Stoyan, D. and Stoyan, H. 1994. Fractals, random shapes andpoint fields. Methods of geometrical statistics. Wiley.Stoyan, D. and Penttinen, A. 2000. Recent applications of pointprocess methods in forestry statistics. Stat. Sci. 15: 6178.Uriate, M. et al. 2004. A maximum-likelihood, neighborhoodanalysis of tree growth and survival in a tropical forest. Ecol.Monogr. 74: 591614.Wiegand, T. and Moloney, K. A. 2004. Rings, circles and nullmodelsfor point pattern analysis in ecology. Oikos 104:209229.Wiegand, T. et al. 2007. Species associations in a heterogeneousSri Lankan dipterocarp forest. Am. Nat. 170: E77E95.Wills, C. et al. 1997. Strong density- and diversity-related effectshelp to maintain tree species diversity in a neotropical forest. Proc. Natl Acad. Sci. USA 94: 12521257.Wills, C. et al. 2006. Nonrandom processes maintain diversity intropical forests. Science 311: 527531.Wright, S. J. 2002. Plant diversity in tropical forests: a review ofmechanisms of species coexistence. Oecologia 130: 114.Zimmerman, J. K. et al. 2008. Large tropical forest dynamicsplots: testing explanations for the maintenance of speciesdiversity. In: Carson, W. P. and Stefan, A. S. (eds), Tropicalforest community ecology. Blackwell, pp. 98117.Zhu, Y. et al. 2008. Community composition and structure ofGutianshan forest dynamic plot in a mid-subtropical evergreenbroad-leaved forest, east China. J. Plant Ecol. 32: 262273,in Chinese.Supplementary material (available as Appendix O17758 atwww.oikos.ekol.lu.seappendix). Appendix 13.242119

ARTICLE IN PRESSFlora 205 (2010) 399–403Contents lists available at ScienceDirectFlorajournal homepage: www.elsevier.de/floraClonal integration increases performance of ramets of the fernDiplopterygium glaucum in an evergreen forest in southeastern ChinaJuan Du b,1 , Ning Wang c,1 , Peter Alpert d , Ming-Jian Yu e , Fei-Hai Yu a,b,n , Ming Dong ba College of Nature Conservation, Beijing Forestry University, Beijing 100083, Chinab State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, Beijing 100093, Chinac College of Life Sciences, Jinggangshan University, Ji’an 343009, Chinad Biology Department, University of Massachusetts, Amherst, MA 01003-9297, USAe College of Life Sciences, Zhejiang University, Hangzhou, Zhejiang 310058, Chinaarticle infoArticle history:Received 24 February 2009Accepted 8 May 2009Keywords:Clonal integrationPteridophyteResource sharingRhizome severingSubtropical forestSurvivorshipabstractPhysiological integration is a major ecological advantage of clonal growth in angiosperms. Clonalgrowth is also common in pteridophytes, but almost no study has tested whether clonal integrationincreases performance in ramets of pteridophytes in natural populations. To test this hypothesis andalso whether the positive effect of integration is greater on smaller ramets, we severed the connectingrhizomes of individual ramets of the common, understory fern Diplopterygium glaucum in an evergreen,broadleaf forest in southeastern China. In another experiment, we severed rhizomes around the edgesof small plots each containing several ramets. After 19.5 weeks, survival was 100% in intact individualramets but only 27% in severed ones. Among surviving ramets, final dry mass and lamina mass werealso less in severed than in intact ramets, though stalk, rhizome, and root mass and maximum quantumyield of PSII (F v /F m ) were not reduced. Individual ramets with fewer stalk nodes had lower dry mass butwere not more affected by severing than ramets with more stalk nodes. Severance around the edge ofplots did not significantly affect the combined final mass of the ramets within a plot. We conclude thatclonal integration can have significant positive effects on both survival and growth of individual rametsof ferns in natural populations.& 2009 Elsevier GmbH. All rights reserved.IntroductionMany angiosperms have the capacity for clonal growth, inwhich a plant and its vegetative offspring remain connected bystems or roots at least until the offspring establish (de Kroon andvan Groenendael, 1997). A major ecological advantage of clonalgrowth is physiological integration between connected ramets,including translocation of resources such as carbohydrates, water,and nutrients (Alpert et al., 2003; Gómez et al., 2007; Jónsdóttirand Watson, 1997; Nilsson and D’Hertefeldt, 2008; Zhang and He,2009). Evidence that clonal integration can increase the performanceof clonal angiosperms comes from numerous studies inwhich the prevention of integration by severing connectionsbetween ramets decreases survival, growth, or further vegetativereproduction (Roiloa et al., 2007; Yu et al., 2008). A few studieshave also shown effects of connection on physiological parameterssuch as capacity for resource uptake or photosynthesisn Corresponding author at: College of Nature Conservation, Beijing ForestryUniversity, Beijing 100083, China.E-mail address: feihaiyu@bjfu.edu.cn (F.-H. Yu).1 These authors contributed equally to this work.(de Kroon et al., 1996; Nielsen and Pedersen, 2000; Roiloa et al.,2007; Wang et al., 2008).Although many pteridophytes also show clonal growth, oftenvia rhizomes (Carlquist and Schneider, 2001; Klimeš et al., 1997;Lu, 2007), much less is known about the effects of clonalintegration on their performance, especially in pteridophytesother than lycopods. A number of studies have demonstratedextensive sharing of resources between ramets in lycopods(Callaghan, 1980; Carlsson et al., 1990; Headley et al., 1985,1988a, b). Two papers have reported that severing connections innatural populations of lycopods in deciduous forests in theeastern US increased the mortality of ramets (Lau and Young,1988; Railing and McCarthy, 2000).Another open question in clonal plant biology is the degree towhich clonal integration is mutualistic in connected ramets innatural populations. Greenhouse and garden experimentsshow that connection between ramets given different levelsof a resource generally increases the performance of the rametsgiven the lower level but not of the ramets given the higherlevel (Alpert, 1999; Chen et al., 2006; D’Hertefeldt andFalkengren-Grerup, 2002). This suggests that integration maybenefit only certain ramets in a clone in habitats where a singleresource is patchy. Similarly, ramets whose size or rooting depth0367-2530/$ - see front matter & 2009 Elsevier GmbH. All rights reserved.doi:10.1016/j.flora.2009.12.018243

ARTICLE IN PRESS400J. Du et al. / Flora 205 (2010) 399–403gives them greater access to resources may subsidize connected,smaller or more shallowly rooted ramets in natural populations(Alpert, 1990, 1996; Marshall and Anderson-Taylor, 1992; Zhangand He, 2009).We therefore hypothesized that (1) clonal integration increasesperformance of ramets in natural populations of pteridophytes,as measured both by growth and physiology, and that(2) this positive effect of integration is greater in smaller ramets.To test these hypotheses, we conducted two experiments on acommon fern, one at the level of individual ramets and one at thelevel of small groups of neighboring ramets. We predicted that (1)severing the connecting rhizomes to a ramet would decrease itssurvivorship, accumulation of dry mass, and photosyntheticcapacity, and that (2) these effects of severing would be greaterin ramets with fewer stalk nodes.Material and methodsThe speciesDiplopterygium glaucum (Thunb. ex Houtt.) Nakai (Gleicheniaceae),formerly called Gleichenia glauca Hook or Hicriopteris glauca(Thunb.) Ching, is a perennial, terrestrial fern that grows clonallyvia rhizomes that bear vertical, perennial fronds (i.e., ramets) withadventitious roots (Lu, 2007). The rhizomes are about 3 mm indiameter (Qian and Chen, 1959), and the inter-ramet distance is onaverage 8.2 cm (n=507, S.E.=0.199; G.-L. Yu, Y.-B. Song, and F.-H.Yu unpublished data). The stalk of the frond is pseudodichotomousat the first node, producing a pair of opposite rachises and a budbetween them. After the rachises grow into pinnae, the budproduces another internode and node with a new pair of rachisesand a bud, and the process repeats. D. glaucum mainly inhabits theunderstories of evergreen forests in mountain ravines at elevationsbelow 1500 m in much of southeast China (Qian and Chen, 1959).ramets and the others. For the intact treatment, rhizomes wereexposed and reburied without being cut.After 137 days on 1–2 October, near the end of the growingseason, we measured the maximum quantum yield (F v /F m ) ofphotosystem II (PSII) on a fully developed, healthy pinna at theupper node of each of the ramets, using a portable chlorophyllfluorometer (Fluorescence Monitoring System 2, Hansatech, UK;saturation pulse method). Measurements were made at 0800–1000 h, after a preliminary, dark adaptation period of at least30 min (Maxwell and Johnson, 2000). We then harvested eachramet, including the portions of the rhizome within 4 cm of thebase of the frond of each intact ramet. Laminas, stalks, rhizomes,and roots were dried separately at 70 1C for 48 h, and weighed.Final sample sizes were 57 intact and 56 severed ramets, because10 intact and 6 severed ramets could not be relocated at the endof the treatments and 5 additional severed ramets werediscovered to have an uncut connection.Effects of severance (intact or severed) and size (1 or 41 node)on survivorship were tested using logistic regression (GENMODprocedure with a logit link function; SAS Institute Inc., 1999).Differences in F v /F m and final dry mass were tested in separateANOVAs with severance and size as fixed effects.Plot experimentFor the experiment in plots, we randomly located 7 pairs of50 cm 50 cm plots with 5–7 ramets of D. glaucum each; plotswithin pairs were 2–4 m apart. On 16 May 2008, the soil was cutdown to a depth of 30 cm with a sharp blade around theperimeter of one plot in each pair, chosen at random, a depthsufficient to sever all rhizomes of D. glaucum crossing into or outof the plot. On 1–2 Oct 2008, we harvested all the ramets in eachplot as described above. We used paired t-tests to compare thecombined dry mass of the ramets in each plot between thesevered and intact treatments.Study areaThe study site was located in an evergreen, broadleaf forest(29114 0 29.5 00 N–29114 0 55.8 00 N, 118106 0 31.4 00 E–118108 0 06.8 00 E, 351–633 m a.s.l.) in the Gutian Mountain National Nature Reserve inZhejiang Province, China. The study site has a subtropical, moist,monsoon climate, with mean annual precipitation of 1964 mm,temperature of 15.3 1C, 1334 h of sunlight, and frost-free period of250 days (Ding et al., 2001). D. glaucum is one of the dominantunderstory herbs in the forest.ResultsIn the experiment on individual ramets, severing rhizomesdecreased the survivorship of ramets by 73% (Fig. 1; w 2 =82.91,100IntactSeveredIndividual ramet experimentFor the experiment on individual ramets, we haphazardlyselected 70 ramets of D. glaucum with 1 stalk node and 64 with41 stalk node shortly after the start of the growing season, on17–18 May 2008. Ramets were at least 1 m apart, and it wasassumed that each ramet came from a different rhizome. Half ofthe ramets in each size class were randomly assigned to each of thetwo treatments, severed or intact. For the severed treatment, therhizome at the base of each ramet was carefully excavated and cut4 cm away from the base in both the distal and proximal directions,after which the soil was replaced. Consider the average distance(8.2 cm) between adjacent ramets and the thin rhizome diameter(about 3 mm), it is unlikely that a cut 4 cm away from the rametbase would bring much reserve loss to the target ramets. Thismeans that the severance treatment mainly functioned to preventresource sharing (i.e., physiological integration) between the targetSurvivalship (%)80604020032 2530 261 >1Number of stalk nodesFig. 1. Survivorship of the intact and severed D. glaucum ramets of different sizes(1 or 41 stalk node). Numbers within the bars are the final number of replicatesin the analysis.244

ARTICLE IN PRESSJ. Du et al. / Flora 205 (2010) 399–403 40120Lamina0.90IntactSevered15StalkRoot0.85Biomass (g)10RhizomeF v /F m0.800.7550.701 >1Number of stalk nodes0IntactSevered Intact SeveredFig. 3. F v /F m (means and SE) of the intact and severed D. glaucum ramets ofdifferent sizes (1 or 41 stalk node).1 node>1 nodeFig. 2. Biomass (means and SE [up for total, down for components]) of the intactand severed D. glaucum ramets of different sizes (1 or 41 stalk node).Table 1Effects of severance and size (number of stalk nodes) on biomass and themaximum quantum yield of PSII (F v /F m ) of the D. glaucum ramets.Severance Size InteractionF 1, 68 P F 1, 68 P F 1, 68 PTotal biomass 1 6.78 0.011 25.17 o0.001 0.90 0.346Lamina biomass 2 7.77 0.007 30.77 o0.001 0.50 0.482Stalk biomass 0.07 0.799 11.97 o0.001 1.73 0.193Rhizome biomass 1.74 0.192 0.05 0.816 0.13 0.724Root biomass 0.12 0.731 0.95 0.334 0.21 0.645F v /F m 1.32 0.255 4.05 0.048 1.33 0.2531 ln transformation.2 square-root transformation.df=1, Po0.001). All ramets with intact connections survived, butonly 27% of the ramets whose rhizomes had been severed. Size oframets did not affect survival (Fig. 1; w 2 =0.001, df=1, P40.9).Final total dry mass of surviving ramets was also less insevered than in intact ramets (Fig. 2, Table 1; F 1,68 =6.78, P=0.01).Severance decreased lamina (F 1,68 =7.77, P=0.007), but not stalk(F 1,68 =0.07, P=0.8), rhizome (F 1,68 =1.74, P=0.2), or root mass(F 1,68 =0.12, P=0.7). Ramets with 41 stalk nodes had more finaltotal mass than ramets with just one node (Fig. 2, Table 1;F 1,68 =25.17, Po0.001); size affected lamina (F 1,68 =30.77,Po0.001), and stalk mass (F 1,68 =11.97, Po0.001), but notrhizome mass (F 1,68 =0.05, P=0.8), or roots (F 1,68 =0.95, P=0.3).Effects of severing did not differ between size classes (eachF 1,68 o1.73, and each P40.19 for the interaction effects). F v /F mwas nominally lower in severed ramets with only one stalk nodethan in other ramets (Fig. 3, Table 1) but effects of severance(F 1,68 =1.32, P=0.3) and severance x size (F 1,68 =1.33, P=0.3) werenot significant and effect of size was only marginally so(F 1,68 =4.05, P=0.05).In the experiment on plots (Fig. 4A), severing rhizomes at theedge of a plot nominally decreased total, lamina, and stalk mass ofthe combined ramets within the plot, but effects of severancewere not significant (each t 6 o0.13 and each P40.1 for total,lamina, stalk, rhizome, and root mass). Number of ramets in eachplot was not different between the intact and severed treatments(Fig. 4B, t 6 o0.01, P40.9).DiscussionResults largely fulfilled the prediction that severing rhizomeswould decrease the performance of ramets. Severing connectionsto individual ramets greatly reduced their survival and also thefinal dry lamina and total mass of the surviving ramets, thoughnot their stalk, rhizome, or root mass, nor their photosyntheticcapacity as measured by maximum quantum yield of photosystemII. Together, these results support the hypothesis thatclonal integration can increase the performance of ramets of fernsin natural populations.Results are broadly similar to the few available previousresults for natural populations of clonal lycopods. Lau and Young(1988) and Railing and McCarthy (2000) each found that severingramets of different lycopod species in forest understoriesincreased the mortality of ramets by 50%. Severance alsodecreased growth by 50% in the latter study, and did not affectwater potential in the former.In contrast, severing connections to plots of D. glaucum rametsdid not significantly affect the combined mass of ramets withinplots, although a proportional reduction in lamina mass wassimilar to that seen in individual ramets. Lack of significant effectsof severing around the edges of small plots has been observed inseveral previous studies (Adachi et al., 1996; Wang et al., 2004; Yuet al., 2008) and could be explained by the likelihood that at leastsome ramets within severed plots remain connected to eachother. If so, this suggests that the ecologically important effects ofintegration may occur mainly within relatively small subgroups ofclosely adjacent ramets, at least in some species.Results failed to fulfil the prediction that severance would havegreater effects on ramets with fewer stalk nodes, even thoughthese ramets had less lamina, stalk, and total dry mass. Resultsthus do not support the hypothesis that the positive effects ofclonal integration are greater in smaller ramets within clones. Wefound no directly comparable previous studies on clonal plants. Anumber of papers note more positive effects of integration onoffspring than on parents (Cullen et al., 2005; Nielsen andPedersen, 2000; Pauliukonis and Gough, 2004; Zhang et al.,2006) or on younger than on older ramets along rhizomes orstolons (Alpert, 1996; Alpert et al., 2002; Marshall and Anderson-Taylor, 1992). Moreover, self-thinning, which typically involves245

ARTICLE IN PRESS402J. Du et al. / Flora 205 (2010) 399–403120100LaminaStalkRootRhizome76Biomass (g)806040Number of ramets54322010Intact0Severed Intact SeveredFig. 4. Effect of severance of plots of ramets of D. glaucum from surrounding ramets on (A) biomass (mean and SE [up for total, down for components]) and (B) number oframets (mean and SE). None of the traits shows significant difference (paired t-tests, all P40.1).greater mortality of smaller individuals, is frequently absent inclonal plants (Scrosati, 2006). However, it may be that thesedifferences are not primarily due to differences between ramets insize.Together with previous evidence that clonal integration cangreatly promote survival in natural populations of at least twolycopods studied so far (Lau and Young, 1988; Railing andMcCarthy, 2000), these results, even though for a single speciesof fern at one site, suggest that integration may be an importantfactor in the performance of forest pteridophytes. Future workmay eventually show that resource sharing and other aspects ofphysiological integration in clonal plants are as widespread inrhizomatous pteridophytes as they appear to be in angiospermsthat grow clonally via stolons, rhizomes, or roots.AcknowledgementsWe thank Mr. Yao-Bin Song and Wei Guo for help withharvesting and the Administration of the Gutian MountainNational Nature Reserve for permission to carry out the study.Research was supported by the Important Directional Item of CASKnowledge Innovative Project (KZCX2-YW-430) and NSFC(30770357, 30860054).ReferencesAdachi, N., Terashima, I., Takahashi, M., 1996. Nitrogen translocation via rhizomesystems in monoclonal stands of Reynoutria japonica in an oligotrophic deserton Mt Fuji: field experiments. Ecol. Res. 11, 175–186.Alpert, P., 1990. Water sharing among ramets in a desert population of Distichlisspicata (Poaceae). Am. J. Bot. 77, 1648–1651.Alpert, P., 1996. Nutrient sharing in natural clonal fragments of Fragaria chiloensis.J. Ecol. 84, 395–406.Alpert, P., 1999. Clonal integration in Fragaria chiloensis differs betweenpopulations: ramets from grassland are selfish. Oecologia 120, 69–76.Alpert, P., Holzapfel, C., Benson, J.M., 2002. Hormonal modification of resourcesharing in the clonal plant Fragaria chiloensis. Funct. Ecol. 16, 191–197.Alpert, P., Holzapfel, C., Slominski, C., 2003. Differences in performance betweengenotypes of Fragaria chiloensis with different degrees of resource sharing.J. Ecol. 91, 27–35.Callaghan, T.V., 1980. Age-related patterns of nutrient allocation in Lycopodiumannotinum from Swedish Lapland. Oikos 35, 373–386.Carlquist, S., Schneider, E.L., 2001. Vessels in ferns: structural, ecological, andevolutionary significance. Am. J. Bot. 88, 1–13.Carlsson, B.A., Jónsdóttir, I.S., Svensson, B.M., Callaghan, T.V., 1990. Aspects of clonalityin the arctic: a comparison between Lycopodium annotinum and Carex bigelowii.In: van Groenendael, J., de Kroon, H. (Eds.), Clonal Growth in Plants: Regulationand Function (Eds.) SPB Academic Publishing, The Hague, pp. 131–151.Chen, J.-S., Lei, N.-F., Yu, D., Dong, M., 2006. Differential effects of clonal integrationon performance in the stoloniferous herb Duchesnea indica, as growing at twosites with different altitude. Plant Ecol. 183, 147–156.Cullen, B.R., Chapman, D.F., Quigley, P.E., 2005. Carbon resource sharing andrhizome expansion of Phalaris aquatica plants in grazed pastures. Funct. PlantBiol. 32, 79–85.De Kroon, H., Fransen, B., van Rheenen, J.W.A., van Dijk, A., Kreulen, R., 1996. Highlevels of inter-ramet water translocation in two rhizomatous Carex species, asquantified by deuterium labelling. Oecologia 106, 73–84.De Kroon, H., van Groenendael, J., 1997. In: The Ecology and Evolution of ClonalPlants. Backhuys Publishers, Leiden.D’Hertefeldt, T., Falkengren-Grerup, U., 2002. Extensive physiological integrationin Carex arenaria and Carex disticha in relation to potassium and wateravailability. New Phytol. 156, 469–477.Ding, B.-Y., Zeng, H.-Y., Fang, T., Chen, S.-W., Yu, J.-P., 2001. A study on the fernsflora in Gutianshan Nature Reservation in Zhejiang Province. J. Zhejiang Univ.(AgricLife Sci.) 27, 370–374.Gómez, S., Latzel, V., Verhulst, Y.M., Stuefer, J.F., 2007. Costs and benefits ofinduced resistance in a clonal plant network. Oecologia 153, 921–930.Headley, A.D., Callaghan, T.V., Lee, J.A., 1985. The phosphorus economy of theevergreen tundra plant Lycopodium annotinum. Oikos 45, 235–245.Headley, A.D., Callaghan, T.V., Lee, J.A., 1988a. Phosphate and nitrate movement inthe clonal plants Lycopodium annotinum L. and Diphasiastrum complanatum (L.)Holub. New Phytol. 110, 487–495.Headley, A.D., Callaghan, T.V., Lee, J.A., 1988b. Water uptake and movement in theclonal plants, Lycopodium annotinum L. and Diphasiastrum complanatum (L.)Holub. New Phytol. 110, 497–502.Jónsdóttir, I.S., Watson, M.A., 1997. Extensive physiological integration: anadaptive trait in resource-poor environments?. In: de Kroon, H., vanGroenendael, J. (Eds.), The Ecology and Evolution of Clonal Plants (Eds.)Backbuys Publishers, Leiden, pp. 109–136.Klimeš, L., Klimešová, J., Hendriks, R., van Groenendael, J., 1997. Clonal plantarchitecture: a comparative analysis of form and function. In: de Kroon, H., vanGroenendael, J. (Eds.), The Ecology and Evolution of Clonal Plants (Eds.)Backbuys Publishers, Leiden, pp. 1–29.Lau, R.R., Young, D.R., 1988. Influence of physiological integration on survivorshipand water relations in a clonal herb. Ecology 69, 215–219.Lu, S.-G., 2007. In: Pteridology. High Education Press, Beijing.Marshall, C., Anderson-Taylor, G., 1992. Mineral nutritional inter-relations amongststolons and tiller ramets in Agrostis stolonifera L. New Phytol. 122, 339–347.Maxwell, K., Johnson, G.N., 2000. Chlorophyll fluorescence: a practical guide. J. Exp.Bot. 51, 659–668.Nielsen, S.L., Pedersen, M.F., 2000. Growth, photosynthesis and nutrient content ofseedlings and mature plants of Cymodocea nodosa: the importance of clonalintegration. Aquat. Bot. 68, 265–271.Nilsson, J., D’Hertefeldt, T., 2008. Origin matters for level of resource sharing in theclonal herb Aegopodium podagraria. Evol. Ecol. 22, 437–448.Pauliukonis, N., Gough, L., 2004. Effects of the loss of clonal integration on foursedges that differ in ramet aggregation. Plant Ecol. 173, 1–15.Qian, C.-S., Chen, H.-Y., 1959. In: Flora Republicae Popularis Sinicae. Science Press,Beijing, p. 128.Railing, C.A., McCarthy, B.C., 2000. The effects of rhizome severing and nutrientaddition on growth and biomass allocation in Diphasiastrum digitatum. Am.Fern J. 90, 77–86.Roiloa, S.R., Alpert, P., Tharayil, N., Hancock, G., Bhowmik, P.C., 2007. Greatercapacity for division of labour in clones of Fragaria chiloensis from patchierhabitats. J. Ecol. 95, 397–405.SAS Institute Inc., 1999. In: SAS/STAT User’s GuideVersion 8. SAS Institute, Inc.,Cary, NC, USA.246

ARTICLE IN PRESSJ. Du et al. / Flora 205 (2010) 399–403 403Scrosati, R., 2006. Crowding in clonal seaweeds: does self-thinning occur inMastocarpus papillatus shortly before stand biomass peaks? Aquat. Bot. 84,233–238.Wang, N., Yu, F.-H., Li, P.-X., He, W.-M., Liu, F.-H., Liu, J., Dong, M., 2008. Clonalintegration affects growth, photosynthetic efficiency and biomass allocation,but not the competitive ability, of the alien invasive Alternanthera philoxeroidesunder severe stress. Ann. Bot. 101, 671–678.Wang, Z.-W., Li, L.-H., Han, X.-G., Dong, M., 2004. Do rhizome severing and shootdefoliation affect clonal growth of Leymus chinensis at ramet population level?Acta Oecol. 26, 255–260.Yu, F.-H., Wang, N., He, W.-M., Chu, Y., Dong, M., 2008. Adaptation of rhizomeconnections in drylands: increasing tolerance of clones to wind erosion. Ann.Bot. 102, 571–577.Zhang, L.-L., He, W.-M., 2009. Consequences of ramets helping ramets: no damageand increased nutrient use efficiency in nurse ramets of Glechoma longituba.Flora 204, 182–188.Zhang, X.-Q., Liu, J., Welham, C.V.J., Liu, C.-C., Li, D.-N., Chen, L., Wang, R.-Q., 2006.The effects of clonal integration on morphological plasticity and placementof daughter ramets in black locust (Robinia pseudoacacia). Flora 201,547–554.247

Journal of Ecology 2009, 97, 1383–1389doi: 10.1111/j.1365-2745.2009.01560.xFactors affecting detection probability in plantdistribution studiesGuoke Chen 1,2 , Marc Ke´ry 3 , Jinlong Zhang 1 and Keping Ma 1 *1 State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences,Beijing 100093, China; 2 Graduate School of Chinese Academy of Sciences, Yuquanlu, Beijing 100049, China; and3 Swiss Ornithological Institute, 6204 Sempach, SwitzerlandSummary1. Plant ecologists have been rather slow to appreciate the existence and the effects of imperfectdetection probability in plants. Sources of heterogeneous detectability include differences in morphologyor life-form, patch size, observers and survey effort. Understanding the relationshipbetween such factors and detectability is crucial for the efficient design of new plant distributionstudies and for the interpretation of existing ones.2. We have studied the factors affecting detectability in a large permanent plot (24 ha) in EastChina where the true distribution of six shrub and tree species was known from a detailed earlierinventory. Two observers independently resurveyed and recorded detection and non-detection ofeach species in each 20 · 20 m sampling quadrat. A total of 288 quadrats were resurveyed (218 byobserver A, 211 by observer B and 141 by both). We used generalized linear mixed modelling tostudy the relationships between detection and species, observer, survey effort and patch size.3. Detectability of an occupied quadrat was remarkably low and ranged from 0.09 to 0.34 on averagefor the six shrub and tree species. Differences of detection among species were mainly as a resultof distinctive morphology rather than life-form. There was no significant difference of overall detectionprobability between the two observers. Detectability increased to 0.95 as the survey pathapproached 20% area of the sampling quadrat and as a plant patch covered c. 19% of the area ofthe sampling quadrat.4. Synthesis. Our results suggest that imperfect detection is much more widespread than currentlyacknowledged by most plant ecologists. We identify several sources of heterogeneity in detectability(species, survey effort and patch size) that ought to be considered when studying and modelling thedistribution of plant species. Detectability should be accounted for in plant distribution studies toavoid spurious inferences.Key-words: biodiversity, China, detection probability, forest dynamic plot, generalized linearmixed model, monitoring, plant distributionIntroductionThe study of the geographical distribution of species lies atthe heart of ecology (Krebs 2000). Distribution is frequentlyinvestigated by the occupancy metric (MacKenzieet al. 2002, 2006), that is, by the proportion of a studyarea that is occupied by a target species. However, someauthors have long acknowledged that a species may goundetected in a survey even when it is actually presentwithin a sampling unit (Kéry 2002, 2004; MacKenzie et al.2002, 2006; Kéry et al. 2006). Thus, the observation of a‘zero’ (an apparent absence) may represent either the trueabsenceofaspeciesoralternatively, non-detection in spite*Correspondence author. E-mail: kpma@ibcas.ac.cnof presence of that species. That is, true occurrenceand detection are confounded in our observations of occurrence.Unfortunately, this situation does not appear to be sufficientlyacknowledged at present in plant distribution studies.This can be seen, for example, in the improper use of the term‘presence–absence data’ for distribution observations that arein fact ‘detection–non-detection data’. Observed ‘absences’may in reality be ‘false absences’ (Royle & Nichols 2003; Tyreet al. 2003), i.e. non-detected presences.Not accounting for the possibility of ‘false absences’ mayhave serious consequences in many respects. Geographicalrange size and extent of occurrence will be underestimated withimperfect detection (Ke´ry 2002; Anderson 2003). Inferencesabout habitat selection will be misleading, especially if detect-Ó 2009 The Authors. Journal compilation Ó 2009 British Ecological Society248

1384 G. Chen et al.ability, not only true occurrence, depends on the habitat (Gu &Swihart 2004). Finally, local extinction rates will be overestimated(Williams, Nichols & Conroy 2002; Ke´ry 2004; Ke´ryet al. 2006), as will turnover rates.The solution to the problem of confounded occurrence anddetection lies in conducting replicate observations of a closedsystem (Burnham & Overton 1978; MacKenzie et al. 2002,2006; Royle & Nichols 2003; Tyre et al. 2003). The pattern ofdetection–non-detection of a species at an occupied site yieldsthe information about detection probability, which enables usto correct the observed distribution for imperfect detection aswell as to recover unbiased functional relationships betweenhabitat covariates and true occurrence.There is a wealth of studies that show how to correct forimperfect detection in animal ecology (Burnham & Overton1978; Williams, Nichols & Conroy 2002; MacKenzie et al.2006), but only very few plant ecological studies have acknowledgedthe existence of imperfect detection and dealt with thechallenge in an appropriate way (a few examples include Alexander,Slade & Kettle 1997; Shefferson et al. 2001; Ke´ry &Gregg 2004). Kéry & Gregg (2003) illustrated the bias incurredin estimates of demographic parameters from studies in permanentplots when life state-specific detection probability is unaccountedfor. Furthermore, local extinction rates of plants willbe overestimated and their relationship with covariates such aspopulation size or habitat may be completely distorted whendetection probability is < 1 and depends on the same factors(Kéry 2004; Ke´ry et al. 2006).Detection probability may vary in time because of surveyspecificconditions and in space owing to site-specific characteristics(Bailey, Simons & Pollock 2004). In addition, factorssuch as the size of a plant patch, plant architecture and growthform, and differences among observers have been hypothesizedto affect plant detection probability (Ke´ry et al. 2006).However, much remains to be learnt about the effects of suchfactors on detection probability.In this article, we highlight the effects on detection probabilitybased on a resurvey of six shrub and tree species in a large(24 ha) permanent plot in East China in 2007. The trueoccupancy state of these species in each 20 · 20 m quadratis known from an intensive, original inventory conducted in2005. In 2007, two observers independently resurveyed theplot once each and we analysed the resulting detection–nondetectiondata to gain insight into the factors affectingdetection probability of these plant species.In our study, we conditioned the analysis on the quadratsthat were known to be occupied in the 2005 inventory. A zeroobservation thereby was assumed to represent the overlookingof a species rather than its absence, and hence, we directlymodelled detection events and did not need to account for possiblenon-occurrence by use of a site-occupancy model (Mac-Kenzie et al. 2002). For each species, the probability ofdetecting at least one individual in each quadrat was investigatedin relation to observer, survey effort and patch size.In addition, species-specific detection probability was exploredin relation to differences in morphology and life-form. Thus,we were able to study not only survey-specific factors but alsosite- and species-specific factors that influence heterogeneousdetection probability.Materials and methodsSTUDY SITE AND SPECIESWe conducted our study at Gutianshan (GTS) permanent plot(29°15¢ N, 118°07¢ E), located within the Gutianshan National NatureReserve, in Kaihua County, East China. According to recordsfrom 1958 to 1986, annual mean temperature in this region is 15.3 °Cand annual mean precipitation is 1964 mm (Yu et al. 2001). Thereserve was designed to protect the old-growth evergreen broadleavedforest in this region and 1426 seed plant species have beenrecorded in the reserve (Legendre et al. 2009). Our study plot covereda rectangle of 24 ha montane uplands (446–715 m a.s.l.), divided regularlyinto 600 20 · 20 m quadrats. The vegetation in the plot is evergreenbroad-leaved forest (Wu 1980). Dominant species areCastanopsis eyrei (Fagaceae), Schima superba (Theaceae) and Pinusmassoniana (Pinaceae). According to the 2005 inventory, 159 seedplant species (belonging to 49 families) occur within the plot (seeLegendre et al. (2009) for more information about GTS plot). These159 species are divided into 63 canopy tree species, 70 understoreytree species and 26 shrub species (Lai et al. 2009). The vegetation isdense and thick with a c. 12-m high canopy layer, a rather closed, c.5-m high understorey and a dense, c. 1.8-m high shrub layer. In total140 676 individuals with diameter at breast height (d.b.h.) ‡ 1cmwere recorded in the 2005 inventory. The shrub species contain 13%of the total individuals, the understorey species of 57% of the individualsand the canopy species of 30% of the individuals.We selected six focal species to conduct a resurvey in the plot(Table 1). The choice was based on different conspicuousness in termsof their mean height, different leaf size and leaf colour. To determinethe age structure of the six species in the plot, we took the d.b.h. datafrom the 2005 inventory and divided them, for each species, into 10classes using the kmeans function in R 2.8.1 (R Development CoreTeam 2008). The distributions of d.b.h. for all six species were skewedtoward young stages (Fig. 1), indicating an uneven age structure.FIELD WORKIn 2005, a detailed inventory was conducted within the 24-ha plot.Over 9 months, a team of 20 people measured the d.b.h. and identified,mapped and tagged the individual trees. Coordinates of all stemswith d.b.h. ‡ 1 cm were mapped and numbered tags were attached tothe stems in the field.In 2007, two observers (GC and JZ) conducted the resurvey of sixspecies in the plot. Neither of them participated in the original inventoryin 2005, hence, they had no information about the distribution ofthe six species in the plot. As one of them was more experienced thanthe other, they conducted a 5-h training session on 4 December 2007.They first studied specimens of the six focal species and then walkedin the community next to our study plot to familiarize themselves withthe search image of these six species, i.e. with how the species look intheir natural environment.On 5, 6 and 8 December 2007, between 9 am and 3 pm, the twoobservers independently surveyed the plot along the main path withinthe plot, which was formed during the 2005 inventory. Some segmentsof the path were not obvious, but most were clearly recognizable.The observers walked through the quadrats along the path. Asthey stepped into a quadrat they took a photograph and recorded thetag of the first individual of any of the six focal species they found. AtÓ 2009 The Authors. Journal compilation Ó 2009 British Ecological Society, Journal of Ecology, 97, 1383–1389249

Plant detection probability 1385Table 1. Basic information of the six species in our study. Life-form follows Lai et al. (2009) and mean height is based on field experience in the2005 inventory. The descriptions of leaf size and colour follow Kuang et al. (1979), Li et al. (1982), Wu & Huang (1987), Zhang & Ren (1998) andour field experienceSpeciesLife-formMeanheight (m)Leaf size(cm 2 )Leaf abaxialcolourLeaf adaxialcolourCamellia fraterna (Theaceae) Shrub 1.5 4–8 · 2–3.5 Pale green Dark green, shinyMyrica rubra (Myricaceae) Understorey tree 4.5 5–14 · 1–4 Pale green Dark greenSymplocos stellaris (Symplocaceae) Understorey tree 4 6–23 · 1.8–5 Pale green Dark greenCamellia chekiangoleosa (Theaceae) Understorey tree 3.5 7–13 · 3–6 Pale green to Green, shinyyellowish greenTernstroemia gymnanthera (Theaceae) Understorey tree 3.5 4–12 · 1.5–5.5 Pale green Dark green, shinyNeolitsea aurata var. chekiangensis (Lauraceae) Understorey tree 2 8–14 · 2.5–4 Pale green Green0 1000 4000Camche0 500 1800Camfra0 1000 4000NeoaurFig. 1. Distributions of diameter at breastheight (d.b.h.) for the six species: Camfra(Camellia fraterna), Myrrub (Myrica rubra),Symste (Symplocos stellaris), Camche(Camellia chekiangoleosa), Tergym (Ternstroemiagymnanthera) and Neoaur (Neolitseaaurata var. chekiangensis). The d.b.h. isclustered into 10 classes using the kmeansfunction in R.Frequency0 500 17001–1.61.7–2.32.4–33.1–3.73.8–4.64.7–5.65.7–6.66.8–7.98–9.516.1–16.1Tergym1–2.32.4–4.34.4–77.1–9.910–12.212.3–14.214.3–16.316.4–18.919–22.823.4–30.10 50 2401–1.51.6–2.22.3–2.82.9–3.53.6–4.14.2–4.64.7–5.45.5–6.87–1021–21Symste1–1.92–2.93–44.1–5.15.2–66.1–6.86.9–7.98–9.29.4–1111.7–13.4d.b.h. (cm)0 50 100 2001–1.71.8–2.72.8–3.83.9–55.1–6.46.5–7.98–9.79.9–12.713.3–18.940–40Myrrub1–2.12.2–3.94–6.46.5–9.49.5–12.913–1717.1–21.922–29.129.3–4450.4–66the same time, they recorded other individuals of the six focal speciesif they noticed any of them in the same quadrat. Each species wasrecorded at least one time in each quadrat. Then, they walked downthe path to the next quadrat. To record a detailed survey path, theobservers took a photograph of a numbered tag about every 2–3 mon their survey path.Numbered tags were recorded along the survey path and coordinatesof these tags can be obtained from data base of the GTS plot.A detailed survey path can thereby be reconstructed for each observer.Each quadrat (20 · 20 m) was divided regularly into 400 grid cells(1 · 1 m); hence, the number of grid cells covered by the survey pathin each quadrat can be computed. We define the number of grid cellscovered by the survey path within each quadrat as a measure of surveyeffort in that quadrat. It indicates roughly how much area of the quadrathas been surveyed. Besides, coordinates of all stems of the sixspecies in each quadrat had been recorded in the 2005 inventory. Thetotal number of 1 · 1 m grid cells occupied by each species in eachquadrat was defined as a measure of patch (‘target’) size. It indicatesroughly how much area of the quadrat was occupied by the species.Thus, for each quadrat surveyed, we had information on the observerwho conducted the survey, survey effort, species patch size andspecies detection–non-detection. Assuming a stable occurrence stateof the six species in each quadrat from 2005 until 2007, we tested therelationship between detection and non-detection and the followingexplanatory variables: observer, survey effort and patch size.STATISTICAL ANALYSISThe original inventory in 2005 is close in time to our survey in 2007and hence we conditioned on those quadrats where each species wasknown to be present 2 years before and an observed ‘zero’ wasassumed to mean that a species had been overlooked.Conditioning on species presence in a quadrat, we estimated detectionprobability and tested its relationships with explanatory variablesusing a generalized linear mixed model (GLMM; Breslow & Clayton1993; Kéry 2002). We fitted a random quadrat effect to account forthe non-independence of detections in the same quadrat owing to theeffects of unmeasured factors. Fixed effects in this model were species,observer, survey effort and patch size as well as all pairwise interactionsbetween the main explanatory factors. Under this model, meandetection probability P ij for species i in quadrat j can be written as:logitðP ij Þ¼a 0 þ a 1 x 1 þ a 2 x 2 þ a 3 x 3 þ a 4 x 4 þ a 5 x 1 x 2 þ a 6 x 1 x 3þ a 7 x 2 x 3 þ a 8 x 1 x 4 þ a 9 x 2 x 4 þ a 10 x 3 x 4 þ d j þ e ijIn this model, a 0 is the logit-linear mean for species 1 and observer1, a 1 is a vector of coefficients of species 2–6 and x 1 denotes the indicatorvariables for species 2–6, a 2 is the effect of observer 2 and x 2denotes the corresponding indicator variable, a 3 is the effect of surveyeffort for species 1 with x 3 the corresponding covariate value and a 4 isthe effect of patch size of species 1 with x 4 the corresponding covariatevalue. Next are the interaction effects: a 5 is a vector of coefficientsÓ 2009 The Authors. Journal compilation Ó 2009 British Ecological Society, Journal of Ecology, 97, 1383–1389250

1386 G. Chen et al.for the species by observer effects, a 6 is a vector of coefficients for thespecies by effort effects, a 7 is the coefficient for the observer by surveyeffort effect, a 8 is a vector of coefficients for the species by patch sizeeffects, and a 9 and a 10 are the coefficients of the observer by patch sizeand the survey effort by patch size interactions respectively. Finally,d j is the random quadrat effect assumed to come from a zero-meannormal distribution with variance r 1 2 ,ande ij is an overdispersionterm assumed to come from another zero-mean normal distributionwith variance r 2 2 .For inference about the fixed effects, we used the Wald statistic(McCulloch & Searle 2001). To explore the functional form of therelationship between these factors and detection probability, we usedour model to form predictions of detection probability for each statisticallysignificant factor. We fitted the GLMM using the statisticalpackage GenStat (Payne et al. 2006).ResultsA total of 288 quadrats were surveyed: 218 quadrats surveyedby JZ, 211 by GC and 141 by both of them. Averaging over allquadrats for two observers separately, the mean survey time ineach quadrat ranged from 4 to 7 min. Quadrats detected by atleast one observer ranged from 24 to 60. Camellia fraterna hadthe highest proportion (29%) of quadrats detected (Table 2).The two observers did not differ significantly in their detectionprobability (Table 3) and were similar at detecting all sixspecies (i.e. there is no significant species · observer interaction,Table 3).In contrast, detection probability was significantly differentamong six species (Table 3; Fig. 2). Myrica rubra had the highestvalue of 0.34, while Neolitsea aurata var. chekiangensis wasthe lowest at 0.09 (Fig. 2). Thus, detection probabilities for thesix species were remarkably low in our study.As expected, detection probability increased with surveyeffort (Table 3; Fig. 3). Among all visited quadrats, surveyeffort ranged from 5 to 44 m 2 and the associated estimates ofdetection probability ranged from 0.11 to 0.58. The increasedpattern did not differ between two observers, as there was noeffort · observer interaction. Likewise, the increase did notdiffer among species, as there was no species · effort interactioneither (Table 3).Table 2. Detection information of the six species in our study. Foreach of the six species, the number of quadrats known to be occupiedis based on the 2005 inventory. Number of quadrats detected and theassociated proportions detected are based on the 2007 resurvey in theplotSpeciesNumberof quadratsknownto beoccupiedNumberofquadratsdetectedin 2007Camellia fraterna 207 60 29Myrica rubra 196 37 18.9Symplocos stellaris 166 24 14.5Camellia chekiangoleosa 247 59 23.9Ternstroemia gymnanthera 264 36 13.6Neolitsea auratavar. chekiangensis271 39 14.4Proportionofquadratsdetected (%)Table 3. Relationships between plant detection probability andseveral explanatory variables under a generalized linear mixed modelfor all six species combined in the Gutianshan permanent plot. For allexplanatory variables, both their main effects as well as their pairwiseinteractions are included. Estimated variance component for theeffects of quadrat = 0.6013 (SE = 0.1619)Source of variation d.f. Wald statistic PSpecies 5 27.80 < 0.001Observer 1 2.29 0.13Survey effort 1 22.80 < 0.001Patch size 1 33.40 < 0.001Species · observer 5 2.24 0.815Species · survey effort 5 0.56 0.990Observer · survey effort 1 3.71 0.054Species · patch size 5 7.47 0.188Observer · patch size 1 1.82 0.177Survey effort · patch size 1 7.03 0.008Detection probability increased with plant patch size(Table 3; Fig. 4). The estimated detection probability rangedfrom 0.13 with patch size of 5 m 2 to 0.98 with patch size of90 m 2 . Detection probability increased to 0.95 (i.e. ‘almost certaindetection’) with a patch size of 75 m 2 . Still, the relationshipbetween detection probability and patch size did notdepend on the observer (no patch size · observer interaction,Table 3), or the species (no species · patch size interaction,Table 3). Finally, a joint effect of patch size and survey efforton detection probability existed in our case (Table 3; Fig. 5).DiscussionDIFFERENCES IN DETECTION PROBABILITY AMONGSPECIESWe can think of two possible explanations for differences ofdetection probability among six species: life-form and distinctivemorphology. In our study, life-form was related to plantheight. According to the 2005 inventory, most individualsof C. fraterna and large number individuals of N. aurata var.chekiangensis were in the same shrub layer. While C. fraternawas detected most reliably, N. aurata var. chekiangensis hadthe lowest detection. Likewise, detection rates of the other fourspecies differed significantly, even though the species were in asimilar understorey layer. That is to say, similar height did notmean similar detection probability. Life-form, therefore, didnot affect detection to the extent we expected.On the other hand, distinctive morphology had a significanteffect on detection probability. Camellia fraterna had the highestproportion of quadrats in which the species was detectedbecause a small number individuals of C. fraterna had whiteflowers at the time of the survey. White flowers constitute a distinctivesearch image within the surrounding vegetation, andthus are easily detectable. We have to note that Camellia chekiangoleosahas conspicuous flowers and M. rubra has distinctivefruits. Obviously therefore, to get an unbiased estimate ofdistributions of these species, field work should be conductedduring the species’ flowering (or fruiting) season. However,Ó 2009 The Authors. Journal compilation Ó 2009 British Ecological Society, Journal of Ecology, 97, 1383–1389251

Plant detection probability 13870.61.0Detection probability0.50.40.30.2Detection probability0.80.60.40.10.2Neoaur Tergym Camche Symste Camfra MyrrubSpeciesFig. 2. Detection probability for six species (predicted mean ± 1 SE)under the generalized linear mixed model analysis in Table 3. Interpretationof the species name is as in Fig. 1. Standard errors (SE) areasymmetric around the predictions of detection probability as bothpredictions and standard errors are back-transformed from the logitlinearpredictor.0 50 100 150 200Patch size (m 2 )Fig. 4. Effect of patch size on detection probability under the generalizedlinear mixed model analysis in Table 3. Patch size is the area perquadrat occupied by species. Effect of patch size has been computedfor the mean of survey effort and averaged over species and observer.The solid line indicates predicted means. Dashed lines indicate theupper and lower band of standard errors. The horizontal line indicatesa detection probability of 0.95. Interpretation of the asymmetricstandard errors is as in Fig. 2.1.0Detection probability0.80.60.41.00.80.6Detection probability0.20 50 100 150 200Survey effort (m 2 )Fig. 3. Effect of survey effort on detection probability under the generalizedlinear mixed model analysis in Table 3. Survey effort is thearea per quadrat occupied by survey path. Effect of survey effort hasbeen computed for the mean of patch size and averaged over the levelsof species and observer. The solid line indicates predicted means.Dashed lines indicate the upper and lower bands of standard errors.The horizontal line indicates a detection probability of 0.95. Interpretationof the asymmetric standard errors is as in Fig. 2.Ternstroemia gymnanthera, Symplocos stellaris and N. auratavar. chekiangensis do not have distinctive flowers or fruits thatcan make their search image conspicuous. Therefore, to getunbiased estimates of their distribution, much higher surveyeffort and more experienced observers are needed for suchspecies.DIFFERENCES IN DETECTION PROBABILITY AMONGOBSERVERSIn our study, detection probabilities of the two observers werenot significantly different. This result is consistent with ourexpectation, as they had been trained prior to the survey. Our0.40.220015010050Survey effort (m 2 )results therefore suggest that training may be an efficientway to reduce observer-specific heterogeneity in detectionprobability. Similarly, no difference exists between two observerswhen the inexperienced one has been trained before thesurvey (Kéry & Gregg 2003). Thus, we argue that training ishelpful to ensure consistency of detection probability, andtherefore of observed distributions, among observers. Webelieve that sufficient training of field personnel is importantfor large-scale monitoring programmes of plant diversity,where plenty of amateur observers are available, but fewexperts.0020015010050Patch size (m 2 )Fig. 5. Joint effects of survey effort and patch size on detection probabilityunder the generalized linear mixed model analysis in Table 3.Interpretations of survey effort and patch size are as in Figs 3 and 4respectively. These effects have been computed by averaging over thelevels of species and observer.Ó 2009 The Authors. Journal compilation Ó 2009 British Ecological Society, Journal of Ecology, 97, 1383–1389252

1388 G. Chen et al.DIFFERENCES IN DETECTION PROBABILITYASSOCIATED WITH SURVEY EFFORT AND PATCH SIZEWe predicted ‘almost certain’ detection (95%) at a surveyeffort of 80 m 2 per quadrat, representing 20% of its area(Fig. 3). This is reasonable because the sampling unit (quadrat)was relatively small (20 · 20 m). Walking through one-fifth ofits area, the observer was able to view most of the quadrat.However, survey effort was relatively low for both observers inall quadrats, with a maximum of 44 m 2 (about 10% of its area)in our study. Low survey effort therefore can explain in partthe low detection probabilities of the six species.It can be assumed, however, that more individuals would beencountered and identified if the observers were to make athorough survey of each quadrat. A higher detection probabilitywould therefore be possible, which in fact is exactly whatthe predictions of detection probability mean (Fig. 3).Similarly, averaging over species, detection probabilityincreases to 0.95 with a target (i.e. patch) size of c. 75 m 2 perquadrat (Fig. 4), representing c. 19% of quadrat area. Royle &Nichols (2003) have pointed out the direct linkage betweensite-specific detection probability P and local abundance N:P =1) (1)r) N . This emphasizes that detection probability isbound to increase with local abundance of a species. Our caseshows the similar trend between detection probability andpatch size of a plant species (Fig. 4). Indeed, patch size in ourcase can be regarded as a rough measure of local abundance ofa species in each quadrat. Therefore, it is even possible to estimatethe distribution of local abundance over all samplingunits within a region using ‘detection ⁄ non-detection’ data only(Royle & Nichols 2003; Dorazio 2007).Interestingly, there is an interaction between survey effortand patch size, that is, a joint effect on detection probability(Fig. 5). The two sides are, to some degree, exchangeable intheir effect on detection probability. Thus, for a small target,i.e. a small patch, more grid cells need to be surveyed to ensure95% certainty of detection, while for larger targets, a muchsmaller effort is sufficient for that level of detection certainty(McArdle 1990).POTENTIAL EFFECTS OF COMMUNITY DYNAMICS ONDETECTION PROBABILITYOur approach assumes that the system remained stablebetween the original inventory in 2005 and the resurveys in2007, i.e. that no occupied quadrat was abandoned in themeantime. On the one hand, however, any mortality mightlead to a decrease in patch size. That is to say, our results mightoverestimate patch size effect. Moreover, it is possible thatsome extinction took place and this would bias our estimatesof detection probability. Arguably, extinction may have takenplace more frequently in quadrats occupied by small than bylarge patches and this would exaggerate our estimates of patchsize effects on detection probability. On the other hand, it islikely that some recruitment took place, thus an increase ofpatch size. That is to say, our results may underestimate patchsize effect on detection probability. Furthermore, these effectscandifferbyspeciesandthatmayexplainpartofthespeciesdifferences in detection probability.Nevertheless, there is a 5-ha forest plot adjacent to andslightly overlapping our study plot from which some informationabout community dynamics is available. In 2002, a totalof 19 183 individuals with d.b.h. ‡ 1 cm were recorded in theoriginal inventory of this 5-ha plot. Five years later, in 2007, atotal of 19 902 individuals with d.b.h. ‡ 1 cm were recorded inthe second inventory (K.P. Ma, X.C. Mi, X.J. Du & M.J. Yu,unpublished data). That is, the annual net change in populationsize is about 0.75% in this area and this result is within therange of turnover rate – averaging of mortality and recruitment– reviewed by Stephenson & van Mantgem (2005). The0.75%changeinpopulationsizeislikelytocauseamuchlowerchange rate of proportion of quadrats occupied, which is verysmall compared to the detection probability of 9–34%. Therefore,we believe that our estimates of detection probabilitywere hardly affected at all by community turnover.IMPERFECT DETECTION IN PLANT DISTRIBUTIONSTUDIESThere is a famous saying by Harper (1977) that plants areeasier to study than animals as they do not run away. However,it is now clear that they can be difficult to detect, andthis has, in our opinion, not been acknowledged sufficientlywidely in the plant ecology community. In our current study,even for moderate-sized shrubs and little trees, we founddetection probabilities of occupied 20 · 20 m quadrats thatwere far below 1. The consequences of such imperfect detectionfor plant distribution studies are clear: it will result in a– possibly serious – underestimation of the true distributionof a species. Furthermore, as ‘target’ (i.e. patch) size in oursurvey was also related to detection, it is clear that distributionstudies that do not correct for imperfect detectabilitymay also result in a size-biased sample of occurrences (i.e.larger patches are more likely tobedetectedthansmallerpatches). Finally, observed distributions of different specieswill be biased to different degrees and studies employing differentfield efforts will not be comparable.In general, imperfect detection will lead one to underestimatethe distribution of a plant. Furthermore, any relationshipsthat may exist between detection and other factors, e.g.habitat covariates, will erroneously be ascribed to true occurrenceunless a proper field design and suitable analytical methodsare used (Ke´ry & Schmidt 2008). If comparable detectionprobability cannot be ensured across the desired dimensions ofcomparison (e.g. time, space and habitat), methods must beemployed that can explicitly account for imperfect detection.Site-occupancy models (MacKenzie et al. 2002), a sort of hierarchicallogistic regression, can be applied to plant distributiondata (i.e. detection–non-detection, also naively called ‘presence–absence’data) provided that at least some sites are surveyedmore than once, allowing detection to be formallyestimated. From the pattern of detection and non-detection ofthe species at occupied sites, one can estimate true distributionfree from any distorting effects of detection probability. TheseÓ 2009 The Authors. Journal compilation Ó 2009 British Ecological Society, Journal of Ecology, 97, 1383–1389253

Plant detection probability 1389useful methods deserve to be known and used much morewidely in plant distribution studies.AcknowledgementsTwo anonymous referees provided valuable comments that significantlycontributed to improvement of our paper. We thank D.I. MacKenzie forcomments on occupancy. Drs Mi Xiangcheng and Ren Haibao established the24-ha permanent forest plot in Gutianshan, Zhejiang Province, China. Thiswork was financially supported by a Key Innovation Project of ChineseAcademy of Sciences (KZCX2-YW-430).ReferencesAlexander, H.M., Slade, N.A. & Kettle, W.D. (1997) Application of markrecapturemodels to estimation of the population size of plants. Ecology, 78,1230–1237.Anderson, R.P. (2003) Real vs. artefactual absences in species distributions:tests for Oryzomys albigularis (Rodentia : Muridae) in Venezuela. Journal ofBiogeography, 30, 591–605.Bailey, L.L., Simons, T.R. & Pollock, K.H. (2004) Estimating site occupancyand species detection probability parameters for terrestrial salamanders.Ecological Applications, 14, 692–702.Breslow, N.E. & Clayton, D.G. (1993) Approximate inference in generalizedlinear mixed models. Journal of the American Statistical Association, 88,9–25.Burnham, K.P. & Overton, W.S. (1978) Estimation of the size of a closed populationwhen capture probabilities vary among animals. Biometrika, 65,625–633.Dorazio, R.M. (2007) On the choice of statistical models for estimating occurrenceand extinction from animal surveys. Ecology, 88, 2773–2782.Gu, W. & Swihart, R.K. (2004) Absent or undetected? Effects of non-detectionof species occurrence on wildlife-habitat models. Biological Conservation,116, 195–203.Harper, J.L. (1977) Population Biology of Plants. Academic Press, London.Kéry, M. (2002) Inferring the absence of a species – a case study of snakes. Journalof Wildlife Management, 66, 330–338.Kéry, M. (2004) Extinction rate estimates for plant populations in revisitationstudies: importance of detectability. Conservation Biology, 18, 570–574.Kéry, M. & Gregg, K.B. (2003) Effects of life-state on detectability in a demographicstudy of the terrestrial orchid Cleistes bifaria. Journal of Ecology, 91,265–273.Kéry, M. & Gregg, K.B. (2004) Demographic analysis of dormancy and survivalin the terrestrial orchid Cypripedium reginae. Journal of Ecology, 92,686–695.Kéry, M. & Schmidt, B.R. (2008) Imperfect detection and its consequences formonitoring for conservation. Community Ecology, 9, 207–216.Kéry, M., Spillmann, J.H., Truong, C. & Holderegger, R. (2006) How biasedare estimates of extinction probability in revisitation studies? Journal ofEcology, 94, 980–986.Krebs, C.J. (2000) Ecology: The Experimental Analysis of Distribution andAbundance, 5th edn. Addison-Wesley Longman, San Francisco.Kuang, K., Zheng, S., Li, P. & Lu, A. (1979) Flora of China. Vol. 21 (Myricaceaethrough Betulaceae). Science Press, Beijing, China [In Chinese].Lai, J., Mi, X., Ren, H. & Ma, K. (2009) Species-habitat association change ina subtropical forest of China. Journal of Vegetation Science, 20, 415–423.Legendre, P., Mi, X.C., Ren, H.B., Ma, K.P., Yu, M.J., Sun, I.F. & He, F.L.(2009) Partitioning beta diversity in a subtropical broad-leaved forest ofChina. Ecology, 90, 663–674.Li, X., Bai, P., Li, Y., Li, S., Wei, F., Wei, Y., Yang, X., Huang, P., Cui, H.,Xia, Z. & Li, J. (1982) Flora of China. Vol. 31 (Lauraceae and Hernandiaceae).Science Press, Beijing, China [In Chinese].MacKenzie, D.I., Nichols, J.D., Lachman, G.B., Droege, S., Royle, J.A. &Langtimm, C.A. (2002) Estimating site occupancy rates when detectionprobabilities are less than one. Ecology, 83, 2248–2255.MacKenzie, D.I., Nichols, J.D., Royle, J.A., Pollock, K.H., Bailey, L.L. &Hines, J.E. (2006) Occupancy Estimation and Modeling: Inferring Patternsand Dynamics of Species Occurrence. Elsevier⁄ Academic Press, Burlington,MA, USA.McArdle, B.H. (1990) When are rare species not there? Oikos, 57, 276–277.McCulloch, C.E. & Searle, S.R. (2001) Generalized, Linear, and Mixed Models.John Wiley & Sons, New York.Payne, R.W., Murray, D.A., Harding, S.A., Baird, D.B. & Soutar, D.M.(2006) GenStat for Windows, 9th edn. VSN International, Hemel Hempstead.R Development Core Team (2008) R: A Language and Environment for StatisticalComputing. R Foundation for Statistical Computing, Vienna, AustriaISBN 3-900051-07-0, URL http://www.R-project.org.Royle, J.A. & Nichols, J.D. (2003) Estimating abundance from repeated presence-absencedata or point counts. Ecology, 84, 777–790.Shefferson, R.P., Sandercock, B.K., Proper, J. & Beissinger, S.R. (2001) Estimatingdormancy and survival of a rare herbaceous perennial using markrecapturemodels. Ecology, 82, 145–156.Stephenson, N.L. & van Mantgem, P.J. (2005) Forest turnover rates followglobal and regional patterns of productivity. Ecology Letters, 8, 524–531.Tyre, A.J., Tenhumberg, B., Field, S.A., Niejalke, D., Parris, K. & Possingham,H.P. (2003) Improving precision and reducing bias in biological surveys:estimating false-negative error rates. Ecological Applications, 13, 1790–1801.Williams, B.K., Nichols, J.D. & Conroy, M.J. (2002) Analysis and Managementof Animal Populations. Academic Press, San Diego, CA, USA.Wu, R. & Huang, S. (1987) Flora of China. Vol. 60 (Symplocaceae and Styracaceae).Science Press, Beijing, China [In Chinese].Wu, Z. (1980) Vegetation of China. Science Press, Beijing, China [In Chinese].Yu, M., Hu, Z., Yu, J., Ding, B. & Fang, T. (2001) Forest vegetation types inGutianshan Natural Reserve in Zhejiang. Journal of Zhejiang University(Agriculture and Life Science), 27, 375–380 [In Chinese].Zhang, H. & Ren, S. (1998) Flora of China. Vol. 49 (Theaceae). Science Press,Beijing, China [In Chinese].Received 21 March 2009; accepted 28 July 2009Handling Editor: Pieter ZuidemaÓ 2009 The Authors. Journal compilation Ó 2009 British Ecological Society, Journal of Ecology, 97, 1383–1389254

Ecology, 90(11), 2009, pp. 3033–3041Ó 2009 by the Ecological Society of AmericaSpecies–area relationships explained by the joint effectsof dispersal limitation and habitat heterogeneityGUOCHUN SHEN, 1,2 MINGJIAN YU, 1,2,6 XIN-SHENG HU, 3 XIANGCHENG MI, 4 HAIBAO REN, 4 I-FANG SUN, 5AND KEPING MA 41 College of Life Sciences, Zhejiang University, Hangzhou2 Key Laboratory of Conservation Biology for Endangered Wildlife, Ministry of Education, Hangzhou3 Department of Renewable Resources, 751 General Service Building, University of Alberta, Edmonton, Alberta T6G 2H1 Canada4 Institute of Botany, Chinese Academy of Sciences, Beijing5 Center for Tropical Ecology and Biodiversity, Tunghai University, TaichungAbstract. Species–area relationships (SARs) characterize the spatial distribution ofspecies diversity in community ecology, but the biological mechanisms underlying the SARshave not been fully explored. Here, we examined the roles of dispersal limitation and habitatheterogeneity in shaping SARs in two large-scale forest plots. One is a 24-ha subtropical forestin Gutianshan National Nature Reserve, China. The other is a 50-ha tropical rain forest inBarro Colorado Island, Panama. Spatial point pattern models were applied to investigate thecontributions of dispersal and habitat heterogeneity and their interactions to the formation ofthe SARs in the two sites. The results showed that, although dispersal and habitatheterogeneity each could significantly contribute to the SARs, each alone was insufficient toexplain the SARs. Their joint effects sufficiently explained the real SARs, suggesting thatheterogeneous habitat and dispersal limitation are two predominant mechanisms formaintaining the spatial distributions of the species in these two forests. These results add toour understanding of the ecological processes underlying the spatial variation of SARs innatural forest communities.Key words: Barro Colorado Island, Panama; dispersal limitation; Gutianshan National Nature Reserve,China; heterogeneous habitat; point pattern modeling; Poisson processes; species–area relationship (SAR);subtropical forest; Thomas processes; tropical forest.INTRODUCTIONA species–area relationship (SAR) describes how thenumber of species changes with the size of the samplingarea (Gleason 1922, Connor and McCoy 1979). Thisrelationship has been studied for more than one hundredyears, and its importance has long been appreciated inbiogeography, community ecology, and conservationbiology (de Candolle 1855, Sugihara 1980, Higgs 1981,He and Legendre 1996, 2002, Desmet and Cowling 2004,Thomas et al. 2004). However, current understanding ofSARs mainly comes from empirical data fitting byvarious statistical and ad hoc models (e.g., exponentialand power curves; Tjørve 2003), and the processes thatproduce the SARs are still not fully understood(McGuinness 1984, Storch et al. 2007).Conventionally, three principle hypotheses have beenproposed to account for SARs. The random placementhypothesis proposes that nothing other than a randomManuscript received 3 September 2008; revised 19 January2009; accepted 17 February 2009. Corresponding Editor: N. J.Gotelli.6 Corresponding author: College of Life Sciences, ZhejiangUniversity, Hangzhou, China 310058.E-mail: fishmj202@hotmail.com3033placement of species and individuals in an area isresponsible for the shape of SARs, thus leaving no roomfor habitat differences and other ecological processes forexplaining species richness (Arrhenius 1921, Coleman1981). On the contrary, the habitat diversity hypothesisattributes the increase of species to the addition of newhabitats when the size of sampling area increases(Williams 1964). Meanwhile, the equilibrium theoryassumes that the number of species in an island is aresult of dynamic equilibrium between the effect ofimmigration and extinction (Preston 1960, 1962, Mac-Arthur and Wilson 1963, 1967).The recently developed neutral theory of macroecologyintroduces speciation and dispersal limitationinto the mechanisms for interpreting species diversityand SARs (Hubbell 2001). It has been shown thatdispersal limitation is a key process generating species–area curves (Hubbell 1999). Different dispersal kernels(e.g., narrow vs. fat tails) can generate SARs varyingfrom log-log linear to triphasic shape (Hubbell 2001,Chave et al. 2002). Rosindell and Cornell (2007) showeda triphasic SARs with a log-log linear central phase in aninfinite landscape.It has been well recognized that each of the abovehypotheses can be applied to explain the observed255

3034 GUOCHUN SHEN ET AL.Ecology, Vol. 90, No. 11TABLE 1.A comparison of the two forest plots (Gutianshan, China, and Barro Colorado Island, Panama [BCI]).Characteristic Gutian plot Barro Colorado IslandPlot location 29810 0 19 00 –29817 0 41 00 N,118803 0 50 00 –118811 0 12 00 E989 0 4.5 00 –989 0 20.7 00 N,79851 0 18.6 00 –79851 0 19.1 00 WCommunity typesubtropical forest,old-growth evergreen,broad-leavedtropical forest,old-growth, rain,semi-deciduous plantsPlot setting-up year 2005 1980Plot size (ha) 24 50Mean annual temperature 15.38C 278CMean annual precipitation (mm) 1964 2600Species richness (number of species with dbh 1 cm) 159 301Number of individuals 140 676 229 049species–area curve in some kinds of communities(Connor and McCoy 1979). These hypotheses emphasizeroles of different ecological and evolutionaryprocesses (Gotelli and Graves 1996), but the methodfor quantitatively assessing the contributions of individualprocesses remains to be explored. This has been amajor challenge to the study of the SARs and a sourceof controversies.In this study, we address this challenge using therecently developed spatial statistics methods (Waagepetersen2007, Waagepetersen and Guan 2009). The spatialstatistical models can be used to analyze spatialdistribution of individuals through either single processor multiple processes (Stoyan 2000, Cottenie 2005), thusallowing for assessing the additive effects of differentgenerating mechanisms. Because the new methods donot require the unrealistic assumptions of stationary andisotropic spatial distribution (Baddeley and Turner2005), more accurate estimates of the joint effects ofdispersal and habitat heterogeneity can be obtained(John et al. 2007).Our interest here is to investigate the individual effectsof random placement, dispersal limitation, habitatheterogeneity, and their joint effects on the formationof the SARs. We start by testing the effect of the randomplacement model and then examine more complexspatial models by including the processes of dispersallimitation and habitat heterogeneity. Specifically, fourmodels were assessed: (1) the homogeneous Poissonprocess for examining the effect of a pure randomprocess; (2) the heterogeneous Poisson process forexamining the effect of habitat heterogeneity; (3) thePoisson cluster process for examining dispersal limitation.Hereafter, this process is called homogenousThomas model based on previous studies (Plotkin etal. 2000, Seidler and Plotkin 2006, John et al. 2007); and(4) the heterogeneous Thomas model for examining thejoint effects of dispersal and habitat heterogeneity.These different spatial models were applied to generateSARs at different spatial scales and the results obtainedfrom each of the above four models were comparedacross scales. Inferences on the effect of each of thesemodels on SARs were then drawn from comparing thegoodness of fit of the models.MATERIALS AND METHODSSubtropical and tropical forest community data setsTwo different types of tree communities were chosento examine the mechanisms generating the SARs. Thefirst data is a 24-ha stem-mapped subtropical forest plot,located in Gutianshan National Nature Reserve, westernZhejiang Province, China (Table 1, hereafter calledGutian plot). Detailed descriptions of the climate,geology, flora, and fauna in the Gutian plot can befound in Zhu et al. (2008) and Legendre et al. (2009).The Gutian plot was stem mapped in 2005. There are intotal 140 676 stems (dbh 1 cm) belonging to 159species. The second data is the widely known 50-haForest Dynamics Plot of Barro Colorado Island (BCI),Panama (Table 1). Detailed descriptions of the climate,geology, flora, and fauna of BCI can be found in Croat(1978), Leigh et al. (1982), and Gentry (1990). The BCIdata we used is the sixth census data collected in 2005.The climate and community composition of the twoplots are summarized in Table 1. Because the accuracyof spatial pattern modeling relies on reasonable minimumpopulation size (Baddeley and Turner 2005), rarespecies with fewer than 50 individuals were not includedin all analyses.In order to quantify the effect of habitat heterogeneityon the SARs of Gutian and BCI plots, we included fourtopographical variables, the tree density per quadrat,and 12 soil nutrient elements and pH value in the soil inour analysis. Specifically, the topographic variables aremean elevation, mean convexity, mean aspect, and meanslope in each 4 3 4 m quadrat (Harms et al. 2001,Valencia et al. 2004). Similarly, based on the originalBCI soil data, we generated the maps of 4 3 4m 2 scalefor the concentrations of nutrient elements including Zn,Al, B, Ca, Fe, K, Cu, Mg, Mn, N, P, and N(mineralization) and pH value in the soil using geostatisticalmethods. The total tree density in each 4 3 4m 2 quadrat was used as a comprehensive bioenvironmentalindex for analysis.Testing the effects of random placement, dispersallimitation, and habitat heterogeneity on SARsFour distinct processes with a progressive increase incomplexity were used to explain the SARs in the two256

November 2009 TWO MAIN DRIVING FORCES OF SARS3035plots. The first is a homogeneous Poisson process wherethe spatial location of a given point (tree) is independentof any other trees. This process only has one parameter,a, the average tree density per unit area for each species(Table 2, Appendix).The second model is a heterogeneous Poisson processwhere the density of each tree species in each quadrat isassociated with the environmental factors in thequadrat. Compared with the homogeneous Poissonprocess, heterogeneous Poisson process has additionalparameters, b j ( j ¼ 1, 2, ...), for describing thecorrelations between the tree density and habitatconditions (Table 2, Appendix). This process can beused to examine the effects of the interaction betweentree density and habitat factors (e.g., topography andsoil nutrient properties) on the SARs.The third model is a homogeneous Thomas processwhere the aggregative distribution of offspring due todispersal limitation is considered, distinct from theprevious two processes. Homogeneous Thomas processis a cluster process, which is formed by the distributionof parent trees generated by a Poisson process (j),together with the distribution of a random number ofoffspring around each parent tree. Here, the number ofoffspring for a parent is also assumed to be a Poissondistribution (l), and the locations of the offspring ofeach parent are assumed to be independent andisotropically normally distributed around the parenttree, with mean being zero and standard deviation d(Table 2, Appendix).The fourth model is a heterogeneous Thomas processwhere the relation between the density of each treespecies and the environmental factors in each quadrat isconsidered in addition to the three parameters in thehomogeneous Thomas process (Table 2, Appendix).Thus, the joint effects of dispersal limitation and habitatheterogeneity are included.In our analysis, parameters in each of the above fourprocesses were estimated using Waagepetersen andGuan’s two-step approach (Waagepetersen and Guan2009; Appendix). The maximum likelihood method,Eq. A.4 in the Appendix, was used to estimate theenvironmental (habitat heterogeneous) parameters. Theminimum contrast method, Eq. A.5 in the Appendix,was used to estimate the dispersal related parameters. Inthe heterogeneous models with soil nutrient variables,we calculated principal components (PCs) from 13 soilvariables and used only the first three components(condensed variables, explained 80.2% of total variancein soil nutrient variables) together with four othertopographic parameters and the tree density per quadratfor analysis. This approach was also used by John et al.(2007) and can help to minimize the possibility of overfittingthe models. To compare different models,Akaike’s information criterion (AIC) was calculated toassess the gain in explanatory power due to the additionof more parameters. Since the parameters were estimatedthrough two-step approach, our AIC calculationsFour different processes and their parameters usedfor testing species–area relationships (SARs) pattern atdifferent spatial scales in Gutian and BCI plots.TABLE 2.Processeswere based on the sum of residuals and the number ofparameters used in different processes (Webster andMcbratney 1989; Appendix).For each of the four processes, we used theparameterized processes to simulate spatial distributionof each species, and then overlaid each species distributiongenerated by the simulations to recover a communitythat was estimated from the actual community. Aspecies–area curve was then constructed by randomlythrowing quadrats onto one simulated community,similar to the method used in previous studies (Dunganet al. 2002, Manly 2006). Finally, the predicted SARs ofeach model were calculated by averaging the SARs on100 simulated communities and a 95% confidenceinterval (CI) was constructed for each predicted SARs.The observed (true) SARs from the original data ofGutian and BCI plots were compared against thepredicted SARs. The model is considered adequate ifthe observed SARs fall within the 95% CI of eachpredicted SARs, otherwise, the model is rejected. Forillustration, we presented the observed and foursimulated spatial distribution maps of Cupania seemannii(Triana & Planch.) in the BCI plot. The nearestneighbor distance function G(r) was also calculated forthose different distributions to evaluate the goodness offit of each model (Ripley 1988, Møller and Waagepetersen2004). All calculations were conducted using theprogram R package ‘‘spatstat’’ and the main codes foranalyses were included in the Supplement.RESULTSParametersHomogeneous PoissonaHeterogeneous Poisson a, b j ( j ¼ 1, 2, . . .)Homogeneous Thomasj, l, dHeterogeneous Thomas j, l, d, a, b j ( j ¼ 1, 2, . . .)Note: Parameters are: a, the tree density per unit area; j, thedensity of parent trees per unit area; l, the expected number ofoffspring trees per parent tree; d, the standard deviation for thelocation distribution of the offspring for a given parent, whichis assumed to be independently and isotropically normallydistributed for the spatial distances between a parent and itsoffspring; and b j , the log linear regression coefficients of the treedensity on the jth environmental factor in the focal quadrat.Homogeneous Poisson processOur results showed that the SARs produced by thehomogeneous Poisson process significantly overestimatedspecies diversity at most scales in both Gutian andBCI plots (Fig. 1, cyan lines). In each plot, the observedSAR was distributed outside the range of 95% CIgenerated by the homogeneous Poisson process. Fig.1B, D, F clearly show the discrepancies between thepredicted and observed SARs at different scales. TheAICs of homogeneous Poisson process models were thehighest in the two plots. These significant differences257

3036 GUOCHUN SHEN ET AL.Ecology, Vol. 90, No. 11indicated that the random placement model is notadequate for explaining the SARs for Gutian and BCIplots. This is because species in both plots are notrandomly distributed. For example, results in Fig. 2A, Eshow that the homogeneous Poisson process fails todescribe spatial distribution of C. seemannii in the BCIplot. This result strongly suggests that nonrandomprocesses should be invoked to explain the SARs forthe two forests.Note that the seemingly ‘‘good’’ estimates of the SARsnear the two ends of the SARs curve, i.e., near 0 and 24ha in the Gutian plot (or near 0 and 50 ha in the BCIplot), were an artifact. If the results were magnified atvery small scales, the SARs driven by the homogeneousPoisson process significantly overestimated speciesrichness at most small scales (Fig. 1). The artifact arisesfrom the fact that the total richness is fixed at the largespatial scale regardless what models are used. Thus, withthe sampling area approaching the total plot, thepredicted species richness is forced to converge to thetotal species richness. This was also true in the followinganalyses with different processes.Heterogeneous Poisson processCompared with the results of the homogeneousPoisson process, a better predicted SAR was obtainedusing the heterogeneous Poisson process (green lines inFig. 1). The AICs of the heterogeneous Poisson processmodel were smaller than those of the homogeneousPoisson process model in the two plots (Table 3).However, the SAR predicted using the heterogeneousPoisson process still significantly overestimated speciesrichness at most scales, especially when some importantheterogeneity habitats (e.g., topography and soil nutrients)were not included (green lines in the middle of Fig.1). The green lines in middle of Fig. 1, generated by theheterogeneous Poisson process, included four topographicparameters (elevation, slope, aspect, and convex)and the total tree density indices. These green lineswere significantly different from those generated by thehomogeneous Poisson process at most large spatialscales. The difference in species richness at each spatialscale and the difference in the AICs between thehomogeneous and heterogeneous processes indicatedthat topography and the total tree density wereimportant for explaining the SARs.Soil nutrients and pH value are important factorswhich can change the SAR pattern. The inclusion of fiveaboveground habitat factors (four topography variablesplus the number of trees in each quadrat) and three mainPCA components from 13 soil parameters with the BCIplot data, substantially improved the prediction of theSARs (green lines at the bottom of Fig. 1). Fig. 2Bshows the spatial distribution of C. seemannii in the BCIplot predicted from the heterogeneous Poisson process.The inclusion of the effects of local soil nutrientproperties produced the clustered distribution that wascloser to the real situation (Fig. 2E). However, there isstill noticeable overestimation in species richness at mostspatial scales even when information of 17 habitatvariables was considered in the BCI plot. This overestimationis largely due to the underestimation ofaggregation of species. Fig. 3 exactly shows thatheterogeneous Poisson model underpredicts the aggregationof the spatial distribution of C. seemannii in theBCI plot. These results indicated that random placementand habitat heterogeneity together were still insufficientto explain the SARs.Homogeneous Thomas processOur results showed that the homogeneous Thomasprocess fitted the SARs better than the homogeneousPoisson process, with smaller AICs (Table 3). Thissuggests that dispersal limitation could be an importantfactor in affecting the SARs. The homogeneous Thomasprocess predicted aggregated distribution of species. Forexample, Fig. 2C shows the distribution of C. seemanniiin the BCI plot predicted from the homogeneousThomas process.In contrast to the overestimation with the homogeneousPoisson process, the homogeneous Thomas modelunderestimates species richness at intermediate spatialscales (blue lines in Fig. 1). The homogeneous Thomasprocess in general overpredicts the degree of aggregationthat causes the underestimation of the SARs (Fig. 3).The underestimations at the scale of 0.015–6.000 ha inthe Gutian plot or at 0.04–6.00 ha in the BCI plotindicated that the sole dispersal process was notsufficient to explain species distribution pattern. Thissuggested that other aggregation processes, such as theheterogeneous habitats, could affect the SARs as well.The presence of other aggregation processes could bringabove biased estimation of the clustering intensity ofeach species.Heterogeneous Thomas processThe last model was the heterogeneous Thomasprocess which significantly improved the explanationof the SARs in both plots. No significant differenceswere observed between the predicted and observedSARs at most sampling scales, except a very slightunderestimation at the scale of 0.04–2.2 ha for the BCIplot (red lines in Fig. 1C–F), and at the scale of 0.02–0.18 ha for the Gutian plot (red lines in Fig. 1A, B). Theheterogeneous Thomas process models had the lowestAIC (Table 3). Apparently, the heterogeneous Thomasprocess is the best-fitted process among the fourprocesses studied here.Fig. 2D shows the spatial distribution of C. seemanniipredicted using the heterogeneous Thomas process inthe BCI plot. Differences in spatial patterns of C.seemannii (Fig. 2) and nearest neighbor distance curves(Fig. 3) again show heterogeneous Thomas process is thebest process of the four studied here. The samequalitative results were also observed in most of otherspecies (not shown here). These results explicitly258

November 2009 TWO MAIN DRIVING FORCES OF SARS3037FIG. 1. The observed (black dots) and predicted species–area curves for the two data sets (A, C, E) and their differences (B, D,F). (A, B) Gutian plot (China) without soil data, (C, D) Barro Colorado Island (BCI; Panama) plot without soil data, (E, F) BCIplot with soil data. In each figure, the cyan line is the species–area curve predicted from the homogeneous Poisson process, thegreen line is the prediction of the heterogeneous Poisson process, the blue line is the prediction of the homogeneous Thomasprocess, and the red line is that of the heterogeneous Thomas process. The differences between results with soil data and resultswithout soil data by the same analysis method reflect the potential impact of soil nutrients and soil pH values on species–arearelationships. The vertical bars indicate the 95% confidential intervals.demonstrated that the joint effects of habitat heterogeneityand dispersal limitation determine the SARs inboth Gutian and BCI plots.DISCUSSIONUsing different types of spatial point pattern models,we demonstrated that the joint effects of dispersallimitation and heterogeneous habitats could explain thepattern of SARs in both Gutian and BCI forest plots.Compared with the significant effects of dispersallimitation or habitat heterogeneity, their joint effectsincrease the predictive power of the SARs in forestcommunities (Table 3). These results are consistent withthe previous studies observed from seed trapping259

3038 GUOCHUN SHEN ET AL.Ecology, Vol. 90, No. 11FIG. 2. The natural distribution of Cupania seemannii (Triana & Planch.) in the BCI plot and its distribution predicted fromfour different processes. (A) Distribution predicted by the homogeneous Poisson process, with parameter â ¼ 0.0026. (B)Distribution predicted by the heterogeneous Poisson process, with parameters â ¼ 17.2541, ˆb 1 ¼ 4.1274 for elevation, ˆb 2 ¼ 0.1113for slope, ˆb 3 ¼ 0.0261 for aspect, ˆb 4 ¼ 0.4145 for convex, ˆb 5 ¼ 1.6390 for tree density, ˆb 6 ¼ 0.2057 for the first component of soilnutrients, ˆb 7 ¼ 0.2057 for the second component of soil nutrients and pH value, and ˆb 8 ¼ 1.6736 for the third component of soilnutrients and pH value. (C) Distribution predicted by the homogeneous Thomas process, with parameters j ¼ 0.0005, l ¼ 4.8786,and d ¼ 5.8189. (D) Distribution predicted by heterogeneous Thomas process, with parameters j ¼ 0.0003, l ¼ 40.7715, d ¼ 17.0692,and the same ˆb j ( j ¼ 1, ..., 8) as those in panel (C). (E) Natural distribution of Cupania seemannii.experiments and environmental association tests (Levineand Murrel 2003, John et al. 2007). Although theimportance of dispersal limitation and habitat heterogeneityin explaining species diversity have beenseparately stressed in forest or other communities(Boecklen 1986, Hart and Horwitz 1991, Plotkin et al.2000, Condit et al. 2002, Clark et al. 2004, Wiegand etal. 2007), our results highlight the significance of theirjoint effects in determining the SARs. This result has notbeen emphasized in lowland rain forest in Panama (BCI)although effects of dispersal limitation have beenreported (Condit et al. 2002, Seidler and Plotkin 2006).Although the random placement hypothesis explainsthe SARs quite well in some communities, it oftenoverestimates species richness in other cases (Ryti 1984,TABLE 3. A comparison of Akaike’s information criterion(AIC) among the four spatial process models.PlotPoissonThomasHomogeneousHeterogeneousHomogeneousHeterogeneousGutian 235.5 217.6 178.8 91.7BCI 251.4 246.2 , 239.5à 187.2 164.5 , 150.1àNote: AICs were calculated according to the sum of residualsand the number of parameters used in different processes (seeAppendix). The differences between results with soil data andresults without soil data by the same analysis method reflect thepotential impact of soil nutrients and soil pH values on species–area relationships.Without soil data.à With soil data.260

November 2009 TWO MAIN DRIVING FORCES OF SARS3039FIG. 3. The relative nearest-neighbor distance (r) function G(r) (the observed G function minus that for a completely randompoint process) for Cupania seemannii in the BCI plot and the theoretical G(r)’s for the fitted homogeneous Poisson, heterogeneousPoisson, homogeneous Thomas, and heterogeneous Thomas processes.Gotelli and Graves 1996, Poltkin et al. 2000, He et al.2002), similar to the results of our homogeneous Poissonprocess models. Ryti (1984) suggested that overestimationof richness by this process might be caused byhabitat heterogeneity and dispersal limitation. Habitatheterogeneity and dispersal limitation could cause theassumption of random spatial distribution of individualsand species invalid. Habitat diversity hypothesis assumesthat species diversity is controlled by theavailability of different habitat types (Williams 1964).However, our results from the heterogeneous Poissonmodels and other habitat-associated tests suggest thathabitat heterogeneity is an important but not sufficientprocess in affecting spatial species diversity. Dispersallimitation is likely to be another important factor thatchanges the SARs (Levine and Murrel 2003). Theoreticaland empirical studies both showed that speciesdiversity depended on the strength of dispersal limitation(Hubbell et al. 1999, Chave et al. 2000, Levine andMurrel 2003, Rosindell and Cornell 2007). Our resultsfrom the Thomas process model confirmed it andfurther indicated that dispersal limitation was not thesole main force in changing the SARs. In summary, thedispersal limitation hypothesis and habitat diversityhypothesis only emphasize different individual processbut not the joint processes. Heterogeneous Thomasprocess used in this study fills the gap in this area.The results of our study suggest that the mechanismsfor maintaining species diversity in forest communitycould be distinct from those of other types ofcommunities, such as forest bird (Boecklen 1986) andherbivore communities (Rigby and Lawton 1981) wherehabitat heterogeneity is considered as the dominantprocess. An important feature of forest community isthat seed dispersal first sets the template for treedistribution. This template is subject to the effects oflocal environments through a variety of forms ofenvironmental filtering. Although the real dynamics offorest communities may be more complicated, dispersaland environmental filtering are perhaps the two mostfundamental steps to determine spatial distribution ofspecies and our study supports this hypothesis. Furtherinquiries on the effects of different types of processes,such as density-dependent selection and interspecificcompetition, could be interesting in predicting SARs,and more sophisticated analyses are needed. Our studyshows the effectiveness of the spatial point process fordescribing species distribution. This is consistent to thefinding of He and Legendre (2002) who have shown thatspatial distribution of species is one of the two primaryfactors that directly determine the shape of SARs, andthe other factor is abundance. The contribution of anyother factors to SARs is through their indirect effects onthe spatial distribution and the abundance of species.The advantage of the point pattern modeling methodused in this study is that it can incorporate bothdispersal and heterogeneous habitats into the pointpattern models (Stoyan 2000, Cottenie 2005, John et al.2007). This method relaxes two unrealistic assumptions(stationarity and isotropy) that are required in traditionalstatistical analysis (Baddeley and Turner 2005),and hence can more accurately estimate the effects ofdispersal and habitat heterogeneity. However, we alsonoticed some limitations in the present modeling261

3040 GUOCHUN SHEN ET AL.Ecology, Vol. 90, No. 11framework. For instance, the spatial statistics cannotevaluate the processes in a manner analogous to themethods of variance partitioning (Dungan et al. 2002,Baddeley et al. 2005), owing to the fact that there is nonatural generalization of the conditional intensity of atemporal process given the ‘‘past’’ or ‘‘history’’ up totime t (Ripley 1988, Møller and Waagepetersen 2004).This begs for the development of more advanced spatialmodels for modeling species distributions.ACKNOWLEDGMENTSForemost, we are grateful to Fangliang He for supportingthe establishment of the Gutian plot. This work was inspiredfrom the many discussions G. Shen had with him. We thankJoshua Plotkin and James Rosindell for very useful commentson the manuscript. We also thank Jianhua Chen, Teng Fang,Shengwen Chen, Binyang Ding, Chaozong Zheng, and manyfield workers who provided support on field investigation. Thiswork was supported by Key Innovation Project of CAS(KZCX2-YW-430) and NSFC (30200034) and was conductedwhile G. Shen visited Fangliang He’s lab in 2008. The BCIforest dynamics research project was made possible by NationalScience Foundation, the John D. and Catherine T. MacArthurFoundation, the Mellon Foundation, the Celera Foundation,and numerous private individuals, and through the hard workof over 100 people from 10 countries over the past two decades.The plot project is part of the Center for Tropical ForestScience, a global network of large-scale demographic tree plots.LITERATURE CITEDArrhenius, O. 1921. Species and area. Journal of Ecology 9:95–99.Baddeley, A., and R. Turner. 2005. Spatstat: An R package foranalyzing spatial point patterns. Journal of StatisticalSoftware 12:1–42.Baddeley, A., R. Turner, J. Møller, and M. Hazelton. 2005.Residual analysis for spatial point processes (with discussion).Journal of the Royal Statistical Society: Series B(Statistical Methodology) 67:617–666.Boecklen, W. J. 1986. Effects of habitat heterogeneity on thespecies–area relationships of forest birds. Journal of Biogeography13:59–68.Chave, J., H.C. Muller-Landau, and S. A. Levin. 2002.Comparing classical community models: theoretical consequencesfor patterns of diversity. American Naturalist 159:1–23.Clark, C. J., J. R. Poulsen, E. F. Connor, and V. T. Parker.2004. Fruiting trees as dispersal foci in a semi-deciduoustropical forest. Oecologia 139:66–75.Coleman, B. D. 1981. Random placement and species–arearelations. Mathematical Biosciences 54:191–215.Condit, R., N. Pitman, E. G. Leigh, J. Chave, J. Terborgh,R. B. Foster, P. Nunez, S. Aguilar, R. Valencia, G. Villa,H. C. Muller-Landau, E. Losos, and S. P. Hubbell. 2002.Beta-diversity in tropical forest trees. Science 295:666–669.Connor, E. F., and E. D. McCoy. 1979. The statistics andbiology of the species–area relationship. American Naturalist113:791–833.Cottenie, K. 2005. Integrating environmental and spatialprocesses in ecological community dynamics. Ecology Letters8:1175–1182.Croat, T. B. 1978. Flora of Barro Colorado Island. StanfordUniversity Press, Stanford, California, USA.de Candolle, A. 1855. Géographie botanique raisonnée: oul’exposition des faits principaux et des lois concernant ladistribution géographique des plates de l’epoque Actuelle.Maisson, Paris, France.Desmet, P., and R. Cowling. 2004. Using the species–arearelationship to set baseline targets for conservation. Ecologyand Society 9:11. hhttp://www.ecologyandsociety.org/vol9/iss2/art11/iDungan, J. L., J. N. Perry, M. R. T. Dale, P. Legendre, S.Citron-Pousty, M. J. Fortin, A. Jakomulska, M. Miriti, andM. S. Rosenberg. 2002. A balanced view of scale in spatialstatistical analysis. Ecography 25:626–640.Gentry, A. H. 1990. Four neotropical rainforests. YaleUniversity Press, New Haven, Connecticut, USA.Gleason, H. A. 1922. On the relation between species and area.Ecology 3:158–162.Gotelli, N. J., and G. R. Graves. 1996. Null models in ecology.Smithsonian Institution Press, Washington, D.C., USA.Harms, K. E., R. Condit, S. P. Hubbell, and R. B. Foster. 2001.Habitat associations of trees and shrubs in a 50-haneotropical forest plot. Journal of Ecology 89:947–959.Hart, D. D., and R. J. Horwitz. 1991. Habitat diversity and thespecies–area relationship: alternative models and tests. Pages47–68 in S. S. Bell, E. D. McCoy, and H. R. Mushinsky,editors. Habitat structure: the physical arrangement ofobjects in space. Chapman and Hall, London, UK.He, F., J. V. LaFrankie, and B. Song. 2002. Scale dependenceof tree abundance and richness in a tropical rain forest,Malaysia. Landscape Ecology 17:559–568.He, F., and P. Legendre. 1996. On species–area relations.American Naturalist 148:719–737.He, F., and P. Legendre. 2002. Species diversity patternsderived from species–area models. Ecology 83:1185–1198.Higgs, A. J. 1981. Island biogeography and nature reservedesign. Journal of Biogeography 8:117–124.Hubbell, S. P. 2001. The unified neutral theory of biodiversityand biogeography. Princeton University Press, Princeton,New Jersey, USA.Hubbell, S. P., R. B. Foster, S. T. O’Brien, K. E. Harms, R.Condit, B. Wechsler, S. J. Wright, and S. Loo de Lao. 1999.Light-gap disturbances, recruitment limitation, and treediversity. Science 238:554–557.John, R., J. W. Dalling, K. E. Harms, J. B. Yavitt, R. F.Stallard, M. Mirabello, S. P. Hubbell, R. Valencia, H.Navarrete, M. Vallejo, and R. B. Foster. 2007. Soil nutrientsinfluence spatial distributions of tropical tree species.Proceedings of the National Academy of Sciences (USA)104:864–869.Legendre, P., X. Mi, H. Ren, K. Ma, M. Yu, Y. Sun, and F.He. 2009. Partitioning beta diversity in a subtropical broadleavedforest of China. Ecology 90:663–674.Leigh, E. G., Jr., A. S. Rand, and D. M. Windsor. 1982. Theecology of a tropical forest: seasonal rhythms and long-termchanges. Smithsonian Institution Press, Washington, D.C.,USA.Levine, J. M., and D. J. Murrel. 2003. The community-levelconsequences of seed dispersal patterns. Annual Review ofEcology, Evolution, and Systematics 34:549–574.MacArthur, R. H., and E. O. Wilson. 1963. An equilibriumtheory of insular biogeography. Evolution 17:373–387.MacArthur, R. H., and E. O. Wilson. 1967. The theory ofisland biogeography. Princeton University Press, Princeton,New Jersey, USA.Manly, B. F. J. 2006. Randomization, bootstrap and MonteCarlo methods in biology. Chapman and Hall/CRC Press,Virginia.McGuinness, K. A. 1984. Species–area curves. BiologicalReviews 59:423–440.Møller, J., and R. P. Waagepetersen. 2004. Statistical inferenceand simulation for spatial point processes. Chapman & HallCRC Press, New York, New York, USA.Plotkin, J. B., M. D. Potts, N. Leslie, N. Manokaran, J.Lafrankie, and P. S. Ashton. 2000. Species–area curves,spatial aggregation, and habitat specialization in tropicalforests. Journal of Theoretical Biology 207:81–99.262

November 2009 TWO MAIN DRIVING FORCES OF SARS3041Preston, F. W. 1960. Time and space and the variation ofspecies. Ecology 41:785–790.Preston, F. W. 1962. The canonical distribution of commonnessand rarity. Ecology 43:185–215.Rigby, C., and J. H. Lawton. 1981. Species–area relationshipsof arthropods on host plants: herbivores on Bracken. Journalof Biogeography 8:125–133.Ripley, B. D. 1988. Statistical inference for spatial processes.Cambridge University Press, Cambridge, UK.Rosindell, J., and S. J. Cornell. 2007. Species–area relationshipsfrom a spatially explicit neutral model in an infinitelandscape. Ecology Letters 10:586–595.Ryti, R. T. 1984. Perennials on rock islands: testing for patternsof colonization and competition. Oecologia 64:184–190.Seidler, T. G., and J. B. Plotkin. 2006. Seed dispersal andspatial pattern in tropical trees. PLoS Biology 4:2132–2137.Storch, D., P. Marquet, and J. Brown. 2007. Scalingbiodiversity. Cambridge University Press, Cambridge, UK.Stoyan, D., and A. Penttinen. 2000. Recent applications ofpoint process methods in forestry statistics. Statistical Science15:67–78.Sugihara, G. 1980. Minimal community structure: an explanationof species abundance patterns. American Naturalist 116:770–787.Thomas, C. D., A. Cameron, R. E. Green, M. Bakkenes, L. J.Beaumont, Y. C. Collingham, B. F. N. Erasmus, M. F. D.Siqueira, A. Grainger, and L. Hannah. 2004. Extinction riskfrom climate change. Nature 427:145–148.Tjørve, E. 2003. Shapes and functions of species–area curves: areview of possible models. Journal of Biogeography 30:827–835.Valencia, R., R. B. Foster, G. Villa, R. Condit, J. C. Svenning,C. Hernandez, K. Romoleroux, E. Losos, E. Magard, and H.Balslev. 2004. Tree species distributions and local habitatvariation in the Amazon: large forest plot in eastern Ecuador.Journal of Ecology 92:214–229.Waagepetersen, R. P. 2007. An estimating function approach toinference for inhomogeneous Neyman-Scott processes. Biometrics63:252–258.Waagepetersen, R., and Y. Guan. 2009. Two-step estimationfor inhomogeneous spatial point processes and a simulationstudy. Journal of the Royal Statistical Society, Series B 71, inpress.Webster, R., and A. B. Mcbratney. 1989. On the Akaikeinformation criterion for choosing models for variograms ofsoil properties. Journal of Soil Science 40:493–496.Wiegand, T., C. V. S. Gunatilleke, I. A. U. N. Gunatilleke, andA. Huth. 2007. How individual species structure diversity intropical forests. Proceedings of the National Academy ofSciences (USA) 104:19029–19033.Williams, C. B. 1964. Patterns in the balance of nature.Academic Press, New York, New York, USA.Zhu, Y., G.-F. Zhao, L. Zhang, G. Shen, X. Mi, H. Ren, M.Yu, J. Chen, S. Chen, T. Fang, and K. Ma. 2008. Communitycomposition and structure of Gutianshan forest dynamic plotin a mid-subtropical evergreen broad-leaved forest, EastChina. [In Chinese.] Journal of Plant Ecology 32:262–273.APPENDIXMain algorithms for the four spatial point models and the steps for estimating parameters used in addressing the joint effects ofhabitat heterogeneity and dispersal limitation on the species–area relationships (SARs), and calculations of AIC (EcologicalArchives E090-217-A1).SUPPLEMENTR program code for estimating parameters of spatial point models and generating the expected SARs for each model (EcologicalArchives E090-217-S1).263

Ecological Archives E090-217-A1Guochun Shen, Mingjian Yu, Xin-Sheng Hu, Xiangcheng Mi, Haibao Ren, I-Fang Sun,and Keping Ma . 2009 . Species–area relationships explained by the joint effects of dispersallimitation and habitat heterogeneity. Ecology 90:3033–3041.Appendix A. Main algorithms for the four spatial point models and the steps for estimatingparameters used in addressing the joint effects of habitat heterogeneity and dispersallimitation on the SARs. Calculations of AIC are also detailed.The algorithm of each model used in the present paper is described in detail in the book ofMøller and Waagepetersen (2003). The steps for parameter estimation in each model are alsoexplicitly illustrated in Waagepetersen and Guan (2007). For convenience, we summarized themain framework of the four models used in our study and the main steps of parameterestimation. For more details, please refer to the two literatures mentioned above.POISSON PROCESSA Poisson process X, defined in a two dimensional region2S ⊂ R , with intensity measureμ and intensity function ρ , satisfies for any bounded subregionB ⊆ S with μ ( B)> 0 .Meanwhile,N (B) is a Poisson variable distributed with mean μ (B). Conditional on N(B) ,the points in X B are i.i.d. with density proportional to ρ (u) , which has the form ofTρ u)= α exp( z ( u)β )(A.1)(1: k 1:kwhereu ∈ B and α > 0 , ( ) denotes thez1 : ku1 × k vector of nonconstant environmentalvariables;β1: kis a corresponding regression parameter.If ρ (u)is an constant λ for all u ∈ S , the Poisson process is homogeneous or stationary.This is a model for ‘no interaction’ and ‘complete spatial randomness process. If ρ (u)is nota constant, but a function of environmental variables ( ) on location u ∈ S , we say thatz1 : ku264

this is an inhomogeneous Poisson process. It implies that there is no interaction betweenpoints, but the intensity could vary according to environmental factors.THOMAS PROCESSSince the independence properties of Poisson process are usually not realistic for real data, wechoose two kinds of Cox processes to model the aggregation patterns. The Thomas pointprocess X is a superposition of clusters X c of offspring associated with mother points c in astationary Poisson point process of intensity κ . Given c, the clusters X c are independentPoisson processes with intensity functionsTρ ( u)= exp( z:( u)β1:c) αk(u − c;1 k kδ)where α >0, k( u − c;δ )is a probability density depending on a parameter δ >0determining the spread of offspring points around c.Texp( z1:k( u)β1: k)still representscovariance between event density and environment at point u. X is a homogeneous Thomasprocess whenTexp( z1 : k( u)β:)1 k=1. Otherwise, X is a heterogeneous Thomas processAssume thatTexp( z1:k( u)β1: k)is bounded by some constant M, A cluster X c may then beregarded as an independent thinning of a cluster Y c with intensity functionMk( ⋅ − c;δ )where the spatially varying thinning probability isTexp( z1 : k( u)β1: k) / M. Using this thinningperspective, the intensity function of Thomas process X isTρ u)= ακ exp( z ( u)β )(A.2)(1: k 1:kPARAMETER ESTIMATIONFor the above four processes, the intensity functions (A.1 and A.2) could also be written as:Tρ ( u)= exp( z(u)β ) , (A.3)wherez u)= (1, z 1( u))and β = β , β ) , β = log( ) for Poisson process and(: k(0 1:k0αβ = log(κ ) for Thomas process. Therefore, following Waagepetersen’s two-step approach0α(Waagepetersen and Guan 2007), we could maximize the following log-likelihood function265

ased on the above intensity function A3 to obtain βˆ :∑ − ∫TTl( β ) = z(u)β exp( z( u)β ) duu∈X∩SFor Poisson process models, we can get all the parameters using maximum likelihoodS(A.4)methods based on A.4. Other parameters κˆ andby minimum contrast methods:δîn Thomas process could be estimatedrcc 2m( κ , δ ) = ( Kˆ( t)− K( t; κ,δ ) ) dt(A.5)∫rlwhere rl, r, and c are user-specified constants, and K is the inhomogeneous K-function of Xwhich is defined asKˆ=∑u,η∈X∩S1[0 < u −η < t]eTexp(( z(u)− z(η))ˆ β )u,ηwheree u,ηis an edge-effect correction. Here, considering the bias of K increases with r, wechoose rl = 0 and r = 100 meters. Following Diggle’s (2003) recommendation, we choose c as1/4. Border edge correction method (Ripley 1988) is used for faster compute.μˆ , theexpected number of offspring of each mother point can be got by following ˆ μ = M / ˆ κ . Finallyˆ α = exp(βˆ0) / ˆ κfor Thomas process.Akaike's Information criterionAkaike's information criterion (AIC) was used to measure the goodness of fit of an estimatedstatistical model. In the general case, AIC isAIC = 2k− 2ln(L)where k is the number of parameters in the statistical model, and L is the maximized value ofthe likelihood function for the estimated model. A problem in the application of this criterionin our study is that estimation of our model parameters is not totally based on maximum266

likelihood method. Although likelihood based parameter estimation methods have beendeveloped in recent years, problems of unstable and extremely time consuming restricted theirapplication (Moller and Waagepetersen. 2004, Guan 2006). Fortunately, we can use thefollowing estimation Â in our current modeling framework (Webster and Mcbratney 1989):A ˆ 2π= { nln[] + n + 2} + nlnR 2kn+where n is the number of observations, k is the number of parameters estimated and R is thesum of residual squares. The quantity in the curly brackets is constant for a given set of dataand so models can be compared by computing: AIC = nln( R)+ 2k.LITERATURE CITEDDiggle, P. J. 2003. Statistical analysis of spatial point patterns. Second edition. AcademicPress, London, UK.Guan, Y. 2006. A composite likelihood approach in fitting spatial point process models.Journal of the American Statistical Association 101:1502–1512.Møller, J., and R. P. Waagepetersen. 2004. Statistical inference and simulation for spatial pointprocesses. Chapman and Hall CRC Press, Virginia, USA.Ripley, B. D. 1988. Statistical inference for spatial processes. Cambridge University Press,Cambridge, UK.Waagepetersen, R., and Y. Guan. 2007. Two-step estimation for inhomogeneous spatial pointprocesses. Department of Mathematical Sciences, Aalborg University, Aalborg, Denmark.Webster R., and A. B. Mcbratney. 1989. On the Akaike Information Criterion for choosingmodels for variograms of soil properties. Journal of Soil Science 40:493–496.267

Forest Ecology and Management 258 (2009) 1147–1152Contents lists available at ScienceDirectForest Ecology and Managementjournal homepage: www.elsevier.com/locate/forecoSeed dispersal phenology and dispersal syndromes in a subtropical broad-leavedforest of ChinaYanjun Du a,b , Xiangcheng Mi a , Xiaojuan Liu a , Lei Chen a , Keping Ma a, *a State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, No. 20 Nanxincun, Xiangshan, Beijing 100093, Chinab Graduate University of Chinese Academy of Sciences, Beijing 100049, ChinaARTICLEINFOABSTRACTArticle history:Received 3 March 2009Received in revised form 31 May 2009Accepted 2 June 2009Keywords:Seed rainDispersal modeFruiting phenologyFruit colorSeasonalityEvergreen forestThis study describes the dispersal phenology and syndromes in Gutianshan 24 ha plot in a subtropicalbroad-leaved forest of China. The 130 0.5 m 2 seed traps collected 69,115 mature seeds, representing 27species (belonging to 24 genera, and 15 families) in 12 months. One marked peak in the number of seedsand species during the year was found in dry season (November). Zoochory was the most commondispersal syndrome (70.4%), followed by anemochory (18.5%), ballistic dispersal (11.1%). Among fruittypes, berry (33%), capsule (22%), nut (18%), and drupe (11%) were common in the subtropical evergreenforest. In fruit color, brown was the commonest (40%), followed by dark brown (30%), black (15%), red(11%), and yellow (4%). Overall, the community level seed rain study revealed that one marked peak inseed number occurred in the middle of dry season; zoochory was the principal dispersal mode of woodyplants in subtropical forest, and dry seasons favor seed dispersal by animal and wind.ß 2009 Elsevier B.V. All rights reserved.1. IntroductionSeed dispersal is a critical event in plant life history for thesurvival of populations, and the survival must link to various bioticand abiotic factors (Vanschaik et al., 1993). The diasporas of manyplant species have characteristic morphological structures andtraits that enhance their probability of being dispersed away fromthe mother plant (Tiffney, 1984; Hughes et al., 1994; Griz andMachado, 2001). Among these traits are fruit color, seed size andshape, and time of fruit ripening (Willson and Whelan, 1990). Themost commonly used classification system of dispersal syndromesis based on the agent or vector of dispersal, typically inferred fromseed morphology (Levin et al., 2003). The principal agents ofdispersal are either abiotic or biotic, and the dispersal syndromesare termed, respectively, anemochory, ballistic, and zoochory.The proportion of dispersal modes in a particular vegetationtype is defined as the dispersal spectrum, which is influenced bycommunity attributes, environmental circumstances and floristiccomposition (Van der Pijl, 1982; Hughes et al., 1994). Althoughdispersal syndromes have many exceptions and are moderatelypredictive about dispersal mechanisms (Fleming et al., 1993),knowledge on dispersal spectra of plant communities is helpful forinterpreting local ecology and for understanding factors that* Corresponding author. Tel.: +86 10 62836223; fax: +86 10 62590835.E-mail address: kpma@ibcas.ac.cn (K. Ma).control composition and structure of communities (Howe andWestley, 1988; Arbelaez and Parrado-Rosselli, 2005).Seed dispersal by animals predominates tropical forest plantspecies (Willson et al., 1989), and involves a tremendous diversityof animal species and behaviors. Animals may consume fruit anddrop, spit or defecate the seeds, carry seeds in their coats or scatterhoardseeds for later consumption. Abiotic strategies such as wind,water and ballistic dispersal form the main mode of seedmovement for the remaining 10–30% of tropical tree species(Willson et al., 1989). However, little research studies the seeddispersal syndromes in subtropical tree species.Seasonality exposes plants to periodic changes in the qualityand abundance of resources (Fretwell, 1972). And almost allsubtropical environments vary seasonally in temperature, rainfall,wind speed and daylength. All of these factors would play a role intriggering phenological changes in subtropical plants. A seasonalclimate also brings about fluctuations in pollinators, seed dispersalagents, predators, and competitors (Griz and Machado, 2001). Theactivity of pollinators and seed dispersers, and breeding periods ofanimals depend on the seasonal production of flowers and fruits inthe community. Phenological patterns are of great importance indetermining the temporal changes, which constrain the physiologicaland morphological adaptations in plant community forutilization of resources (Herrera, 1986; Vanschaik et al., 1993;Selwyn and Parthasarathy, 2006) by fauna. However, phenologicalpatterns of seed dispersal in subtropical broad-leaved forests arelittle known.0378-1127/$ – see front matter ß 2009 Elsevier B.V. All rights reserved.doi:10.1016/j.foreco.2009.06.004268

1148Y. Du et al. / Forest Ecology and Management 258 (2009) 1147–1152In this paper, we address the following questions:(1) What is the pattern of seed dispersal phenology in thesubtropical forest understory?(2) What is the dispersal mode for major woody plant species insubtropical forest?(3) Do climate or seasons favor one dispersal mode over another?2. Materials and methods2.1. Study siteThe study was conducted in a 24-ha permanent forest plot(29815.101 0 –29815.344 0 N, 118807.010 0 –118807.400 0 E) in GutianshanNational Nature Reserve (GNNR), Kaihua County, ZhejiangProvince in eastern China. GNNR covers a total area of approximately8107 ha. The topography is characterized by mountainswith steep slopes. The substrate consists mainly of granite. Thedominant soils can be classified into four types: red soil, red-yellowsoil, yellow-red soil and marsh soil. The mean annual temperatureis 15.3 8C. The hottest month is July with a mean temperature of27.6 8C, and the coldest is January with a mean temperature of4.7 8C. The mean annual precipitation is 1787 mm with seasonaldistribution over the year (Yu et al., 2001). Wet season occurs fromMarch to July, whereas dry season runs from August to Februarynext year. The mean annual number of frost-free days is 250. Atotal of 1991 vascular plant species, belonging to 244 families and897 genera, were recorded within the entire GNNR. The dominantvegetation type in GNNR is subtropical evergreen broad-leavedforest dominated by Castanopsis spp., Cyclobalanopsis spp. andSchima superba (Chen and Feng, 2002; Hu et al., 2003).2.2. Methods2.2.1. Plot setIn 2005, a permanent plot covering 24-ha (400 m 600 m,horizontal distance) was established within the evergreen broadleavedforest in GNNR. The plot was established and data werecollected following the plot standards of the CTFS (Center forTropical Forest Science) network (Condit, 1998). The elevationrange between the highest and lowest point in the plot was 269 m(from 446 to 715 m). The first tree census was conducted in 2005.All woody stems 1 cm in DBH were mapped, measured,identified, and tagged (Legendre et al., 2009).2.2.2. Seed collectionSeed rain has been censused weekly since June 2006, using 130seed traps set along 2.3 km of trails within the plot (Wright andCalderón, 1995; Wright et al., 1999). Each seed trap consists of asquare, 0.5 m 2 PVC frame supporting a shallow, open-topped, 1-mm nylon mesh bag, and suspended 0.8 m above the ground onfour PVC posts. All seeds, fruits, seed-bearing fruit fragments,flowers, capsules, and other reproductive parts of plants that fallinto the traps were identified to species and recorded. Fruits werecategorized as aborted, immature, damaged, fragments andmature. Because the seed traps were located above the ground,they captured fruits and seeds falling directly from trees, as well asthose spat or defecated by birds, bats and arboreal mammals; theydid not, however, record secondary dispersal by rodents and otherterrestrial animals (Muller-Landau et al., 2008). All data presentedrefer to seed number, based upon either a count of actual seeds perfruit, or calculated based upon the mean number of seeds per fruit.Dispersal syndrome was assigned to each species based on fruitmorphology and unpublished observations of fruit consumption.Each plant species was assigned to one main dispersal mode. FruitFig. 1. The mean monthly temperature and rainfall during the period of study and inthe 50-year mean.morphology was the basis for classifying species as dispersed bywind, animals, or explosion. We also assigned the dispersal modeto each species by ‘‘Exclusion hypotheses’’ (Hughes et al., 1994).3. ResultsSeed rain. The 130 understory seed traps collected 69,115mature seeds, representing 27 species (belonging to 24 genera, and15 families) in 52 weekly censuses between June 2006 and May2007. Seed rain understory totaled 1064 seeds/m 2 during the 12-month period (total trap area = 65 m 2 ). The Pearson correlationcoefficient value between adult tree basal area and seedproduction is 0.456 (p = 0.000 < 0.01, n = 147).Dispersal Phenology. There were negative correlations betweenrainfall and seed number (R = 0.494, p = 0.103), between rainfalland seed species number (R = 0.812, p =0.001< 0.05) (Figs. 1 and2). We found one marked peak in the number of seeds and speciesduring the year (Fig. 2). The highest peak occurred in the middle ofdry season (November), which accounts for 47.6% of seeds all theyear (Figs. 1 and 2). The lowest number of fruiting species wasfound in June (55 seeds/month), July (188 seeds/month) andAugust (403 seeds/month) also showed few fruiting species. Thenumber of seeds in wet season is 6582, which accounts for 10% ofthe total seeds of the year, and the number in dry season is 62,533accounting for 90%. Seeds of ten species were collected during thedry season, while 23 species were collected during the wetseason.Dispersal modes. Dispersal modes for all species are listed inTable 1. Three dispersal syndromes were considered: zoochory(animal dispersal), anemochory (wind dispersal), and ballisticdispersal. Zoochory was the most common dispersal mode,represented by 70.4% of all studied species, followed byanemochory (18.5%), and ballistic dispersal (11.1%) (Fig. 3). Themost frequent families were Fagaceae (19%), Ericaceae (15%) andFig. 2. The number of seeds and number of species falling into the traps per monthduring the period of study.269

Y. Du et al. / Forest Ecology and Management 258 (2009) 1147–1152 1149Table 1Fruit type, dispersal units, diaspore color and dispersal syndromes of woody species from subtropical forest in East China (T, tree; S, shrub; Z, zoochory; A, anemochory; B,ballistic; F, fruit; S, seed).Species Family Life-forms Fruit type Diaspore color Dispersal syndrome UnitStyrax dasyanthus Styracaceae S or T Achene Brown Z SQuercus serrata Murray Fagaceae T Nut Green Z SVaccinium carlesii Ericaceae S Berry Dark brown Z FEurya muricata Theaceae S Berry Dark brown Z FMachilus thunbergii Lauraceae T Berry Dark brown Z FTernstroemia gymnanthera Theaceae S or T Berry Red Z SDaphniphyllum oldhami Daphniphyllaceae S or T Drupe Dark brown Z FCorylopsis glandulifera Hamamelidaceae S Capsule Black B SLoropetalum chinense Hamamelidaceae S or T Capsule Black B SVaccinium mandarinorum Ericaceae S or T Berry Black Z FItea chinensis Saxifragaceae S Capsule Brown A SFraxinus insularis Oleaceae T Samara Black A SPinus massoniana Pinaceae T Cone Brown A SRhododendron ovatum Ericaceae S Capsule Brown A SSchima superba Theaceae T Capsule Brown A SNyssa sinensis Nyssaceae T Drupe Dark brown Z FCyclobalanopsis glauca Fagaceae T Nut Brown Z SCyclobalanopsis myrsinaefolia Fagaceae T Nut Brown Z SAlbizia kalkora Leguminosae T Legume Brown Z SIdesia polycarpa Flacourtiaceae T Berry Red Z FLithocarpus glabra Fagaceae T Nut Brown Z SCastanopsis eyrei Fagaceae T Nut Dark brown Z SVaccinium bracteatum Ericaceae S Berry Dark brown Z FIlex micrococca Aquifoliaceae T Berry Red Z FDistylium myricoides Hamamelidaceae S or T Capsule Brown B SToxicodendron succedaneum Anacardiaceae S or T Drupe Yellow Z FEurya rubiginosa Theaceae S or T Berry Dark brown Z FTheaceae (15%) (Fig. 4). During one year’s collection, 11 familieswere only represented by one species each, 1 family by threespecies, 2 families by four species each, and 1 family by five species(Fig. 5).Among fruit types, berry (33%), capsule (22%), nut (18%), anddrupe (11%) were common in the subtropical evergreen forest(Fig. 6). In fruit color, brown was the commonest (40%), followedby dark brown (30%), black (15%), red (11%), and yellow (4%)(Fig. 7). All berry (9 species) and nut fruits (5 species) wereanimal dispersal. Dark brown diaspore could be berry, nut,drupe, but all of them were dispersed by animals. All red fruitswere berry, and only dispersed by animals. Nearly all zoochoryspecies have deep fruit/seed color, like dark brown and red(Table 1).All five Fagaceae species were animal dispersed. Three-fourthEricaceae species were also animal dispersed. All three Hamamelidaceaespecies were ballistic dispersed. Three of four Theaceaespecies were animal dispersed. All anemochory species havebrown or black fruits (Table 1).4. Discussion4.1. Seed rainThe seed rain of the Gutianshan was highly variable, butcontinuous throughout the year.During a one-year period, we collected seeds belonging to 27species, out of a total of 159 species in the whole plot. Theregression between adult basal area and seed rain suggests thatFig. 4. Percentage of family during the study period.Fig. 3. Dispersal modes in sub-tropical forest in East China. Percentage of species perdispersal mode relative to the total number of plant species collected.Fig. 5. The number of species in each family.270

1150Y. Du et al. / Forest Ecology and Management 258 (2009) 1147–1152Fig. 6. Percentage of fruit types during the study period.November must wait until March or April before reliable rainspermit germination. If most seeds dispersed in wet season insubtropical forest, the seed would germinate immediately, andseedlings will suffer a long winter season which is unfavorable forseedling establishment. Here is a study on the reproductivephenology of Hong Kong shrubland in South China which exhibitedsimilar patterns (Corlett, 1993). Seed rain has a December peak anda May–June low. The simplest adaptive explanations are either aphenological match with the migratory movements of millions ofpartly frugivorous birds (thrushes etc.) through subtropical Asia inOctober–November and/or diet switching by resident insectivorefrugivoresas insect availability declines in the cool-dry season(Corlett, 1993). Liu (2000) studied the soil seed bank dynamics in asubtropical evergreen broad-leaved forest in China and found thatmost seeds germinated in April, which shows that most seedspostpone germination until more favorable conditions take placein the spring.4.2. Dispersal syndromesFig. 7. Percentage of diaspore color during the study period.interspecific variations in primary seed dispersal could beexplained by average basal area of adults. The fundamentallimitation on recruitment could be absence of parent trees. Unlessseed production and dispersal are high, recruitment limitation islikely for many taxa simply on the basis of parent tree abundance.The seed rain averaged 1064 seeds m 2 yr 1 , but the seed densitywe got maybe underestimated because some seeds might havebeen removed from seed traps by seed predator prior to oursampling. The time of reproduction is one of the traits that aredirectly affected by the availability of limiting resource, such aswater (Volis, 2007). We found one marked peak both in the numberof seeds and in the number of species in November. Even fleshyfruits matured during the dry season (e.g., Vaccinium carlesii, Euryamuricata, Vaccinium mandarinorum, Idesia polycarpa, Vacciniumbracteatum). This is not in agreement with previous reports atother sites (Frankie et al., 1974; Lieberman, 1982; Murali andSukumar, 1994; Griz and Machado, 2001), where fruiting peakedduring the wet season. This does not facilitate the germination ofseeds, which depend largely on sufficient water availability.Although researchers held the view that fruiting peak may berelated to environmental conditions both for dispersal (e.g.,stronger winds during the dry season) and for germination, wethought that fruiting peak in dry season in our study is not relatedto germination, but for seedling establishment. This suggests thatthere are other advantages in winter fruiting. Seed dispersed inWe found that zoochory was the principal dispersal mode ofplants in subtropical forest understory. Our results are similar tostudies conducted in tropical forests around the world (Table 2, seealso Selwyn and Parthasarathy, 2006). Study found that dispersalsyndrome could explain significant variation in clumping of seeddeposition (Muller-Landau et al., 2008). Dispersal by animals(particularly birds) could help seeds escape high mortality conditionsnear their parents, where predation, abundance of pathogensand intraspecific competition are at their highest (Janzen, 1970).Birds vary in their movement and fruit handling patterns, andtherefore exert differing influences on seed viability and the spatialpatterns of seed deposition (Howe, 1989, 1993; Clark et al., 2005).Anemochorous species only represent 18.5% of all species at this site,which is lower than other studies (Table 2). We think this may bebecause these forests have more open canopy vegetation, allowinggreater wind circulation at all forest levels when compared to denseevergreen forests, particular during the dry season. In our study,anemochorous fruits were produced exclusively in the dry season.Dispersal by wind is more efficient during the dry season, becausedry conditions favor the liberation of the seeds and allow their wingsand plumes to fully expand (Sharpe and Fields, 1982).Ballistic dispersal represented 11.1% of all species at this site. Allballistic dispersal occurred in dry season. This is different from theprevious studies which showed positive correlation with theoccurrence of rain, suggesting the influence of rain in fruitdehiscence (Murray, 1986; Griz and Machado, 2001). Thisdifference may be explained by the fact that precipitation insubtropical region is greater than that of arid environments, whereit favors dispersal of seeds when conditions for germination,seedling establishment and growth are optimal (Gutterman, 1994).Many temperate-zone plants produce fruits that are adapted forseed dispersal by birds, and many of these fruits are conspicuouslyTable 2Comparison of frequency distribution and seasonal dominance of dispersal syndromes with other forests (blanks represent data not observed or not comparable fromreference; A, anemochory; Z, zoochory).Sites Zoochroy Anemochory Duration of rainySeason (month)Annual rainfall(mm)Seasonal dominance% (n) % (n) Dry season Rainy seasonReferenceCosta rica 50 (53) 30 (32) 4 1533 A Z Frankie et al. (1974)Cerrado, Brazil 52 (142) 30 (81) 6 1300 A Z Gottsberger and Silberbauer-Gottsberger (1983)Ghana 75 (59) 25 (20) 5 1100 A Z Lieberman (1982)Caatinga, Brazil 36 (15) 32 (13) 7 549 A Z Griz and Machado (2001)Colombia 47 (137) 23 (68) No marked 3060 – – Arbelaez and Parrado-Rosselli, 2005rainy seasonGutianshan, China 70 (19) (18) (5) 5 1787 Z – This study271

Y. Du et al. / Forest Ecology and Management 258 (2009) 1147–1152 1151colored (Willson and Thompson, 1982). In our study, brown wasthe commonest, followed by dark brown, black, red, and yellow.This result is similar to other studies around the world (Turcek,1963; Willson and Thompson, 1982; Van der Pijl, 1982). Turcek(1963) showed that black and red were common colors forEuropean fruits whose seeds were dispersed by birds (see also Vander Pijl, 1982). Corlett (1996) studied 255 native plant species inHong Kong and also showed that the commonest color of fruit isblack, followed by red. It is similar in our sub-tropical forest wherewe differentiate deep brown and black, but actually they are close.Colorful displays of ripe fruit likely evolved in order to attract aviandispersal agents. Frugivorous birds thus serve as selective agents ofplants by favoring those species whose seeds could disperse topotential ‘safe sites’. The subsequent seed dispersal pattern notonly determines the potential area of plant recruitment, but alsoserves as a template for subsequent processes, such as predation,competition and mating (Nathan and Muller-Landau, 2000). Othercolors such as yellow, green, and orange, frequently the colors ofunripe fruits, were less popular (Willson and Thompson, 1982).The fruits of red, brown, deep brown color with pulp as a rewardexhibit animal dispersal mode. The anemochory and ballisticspecies were almost brown fruits without any reward.4.3. SeasonalityNatural selection might favor plants with a pattern ofreproductive phenophases dispersed throughout the growingseason (Vanschaik et al., 1993). Our study showed that mostseeds dispersed during the dry season, and it was also possible tofind 10 species dispersed seeds during the wet season. This may bebecause the efficiency of dispersal depends on the number of seedseaten, and if at one time of the year there are a greater number ofseeds available than can be eaten, then the energy that went intothe production of the surplus will be wasted. Conversely, if thereare not enough seeds at one time of the year, then the potentialdispersal agents will either starve or will eat something else(Smythe, 1970). And any species that fruits during this time willincrease its chances of dispersal, which will ultimately promoterecruitment.Our study showed that zoochory dominated the dry season, andthis is different from previous studies. Griz and Machado (2001)found that zoochorous (i.e. fleshy) diasporas were most commonduring the rainy season at Caatinga, Brazil seasonally, which is inaccordance with a previous case (Barbosa et al., 1989). The fact thatzoochorous species showed a fruiting peak during the wet seasonfollows a general pattern among tropical dry forests (Table 2, seealso Bullock, 1995), where periods of greater precipitation seem tofavors animal dispersal (Jackson, 1981; Gottsberger and Silberbauer-Gottsberger,1983; Machado et al., 1997). There is evidencethat the period of fleshy fruit production is associated with changesin behavioral patterns of dispersers (Smythe, 1970; Stiles, 1980;Wheelwright, 1985). However, studies on animal behaviors inGutianshan subtropical forests are still scarce, and further work onpossible relationships between the foraging behaviors of animalsand fruiting patterns of zoochorous species in this communitywould be of great value.Although only one year of data was considered, the dispersalphenology pattern could be representative of subtropical evergreenbroad-leaved forests based on findings by other studies atsimilar latitudes. One such study on the reproductive phenology ofHong Kong shrublands in South China (22817 0 N, 114809 0 E) over athree years period showed similar patterns to what we found(Corlett, 1993). Community patterns of reproductive phenologyare highly seasonal and vary little between years. Most speciesflowered and fruited every year but Machilus cf. thunbergii did notflower in 1988 and Rapanea neriifolia flowered but did not fruit in1989. A similar pattern, with a November–December fruiting peak,was seen in mostly woody species at Dinghushan BiosphereReserve in South China (23810 0 N, 112831 0 E) (Li and Wang, 1984).Au et al. (2006) studied seed rain in woody plant communities inHong Kong (22825 0 N, 114807 0 E), China, and also found similarpatterns to those seen in our study in Gutianshan. They found thatmost seed rain occurred between September and January, withpeaks in December and January.5. ConclusionOverall, the community level seed rain study revealed that onemarked peak in seed number occurred in the middle of dry season(November). The lowest seed number of fruiting species was foundin wet season (June). Zoochory was the principal dispersal mode ofplants in subtropical forest understory. Dry seasons favor seeddispersal by animal and wind. Our study will provide the neededknowledge which would contribute to study on seedling establishmentand recruitment in subtropical forests. Further works shouldfocus on possible relationships between the foraging behaviors ofanimals and fruiting patterns of zoochorous species in thiscommunity. We should also conduct a more detailed evaluationof the spatial patterns of adults and seed rain. Since only one yearof data was considered in the study, we should continue to studythe year to year variability in seed production and the phenology ofdispersal.AcknowledgementsWe would like to thank Dr. Joe Wright and Dr. Helene Muller-Landau for many discussion sessions which contributed to theformulation of the ideas presented here. We will also thank Mr.Fang Teng and Mr. Chen Bin for identifying dispersal modes forsome species and two anonymous reviewers who help us improvethis paper. We gratefully acknowledge the support from theAdministration Bureau of the Gutianshan National Nature Reserve.This study was financed by Key Innovation Project of CAS (KZCX2-YW-430).ReferencesArbelaez, M.V., Parrado-Rosselli, A., 2005. Seed dispersal modes of the sandstoneplateau vegetation of the middle Caqueta river region, Colombian Amazonia.Biotropica 37, 64–72.Au, A.Y.Y., Corlett, R.T., Hau, B.C.H., 2006. Seed rain into upland plant communitiesin Hong Kong, China. Plant. Ecol. 186, 13–22.Barbosa, D.C., Alves, J.L., Prazeres, S.M., Paiva, A.M.A., 1989. Dados fenológicos de 10espécies arbóreas de uma área de caatingga (Alagoinha PE). Acta Bot. Bras. 3,109–117.Bullock, S.H., 1995. Plant reproduction in neotropical dry forests. In: Bullock, S.H.,Mooney, H.A., Medina, E. (Eds.), Seasonally Dry Tropical Forests. CambridgeUniversity Press, Cambridge, UK, pp. 277–297.Chen, J., Feng, Z., 2002. Study on geographical compositions of seed plant flora inGutianshan Mountain of Zhejiang Province. J. East China Norm. Univ. (Nat. Sci.)48, 104–111.Clark, C.J., Poulsen, J.R., Bolker, B.M., Connor, E.F., Parker, V.T., 2005. Comparativeseed shadows of bird-, monkey-, and wind-dispersed trees. Ecology 86, 2684–2694.Condit, R., 1998. Tropical Forest Census Plots. Springer-Verlag, Berlin, Germany.Corlett, R.T., 1993. The reproductive phenology of Hong Kong shrubland. J. Trop.Ecol. 9, 501–510.Corlett, R.T., 1996. Characteristics of vertebrate-dispersed fruits in Hong Kong. J.Trop. Ecol. 12, 819–833.Fleming, T.H., Venable, D.L., Herrera-M, L.G., 1993. Opportunism vs. specialization:the evolution of dispersal strategies in fleshy-fruited plants. Vegetatio 107–108,107–120.Frankie, G.W., Baker, H.G., Opler, P.A., 1974. Comparative phenological studies oftrees in tropical wet and dry forests in lowlands of Costa-Rica. J. Ecol. 62, 881–919.Fretwell, S.D., 1972. Populations in a seasonal environment. Monogr. Popul. Biol. 5,1–217.Gottsberger, G., Silberbauer-Gottsberger, I., 1983. Dispersal and distribution in theCerado vegetation of Brazil. Sonderbaende des Naturwissenschaftlichen Vereinsin Hamburg 7, 315–352.272

1152Y. Du et al. / Forest Ecology and Management 258 (2009) 1147–1152Griz, L.M.S., Machado, I.C.S., 2001. Fruiting phenology and seed dispersal syndromesin caatinga, a tropical dry forest in the northeast of Brazil. J. Trop. Ecol. 17, 303–321.Gutterman, Y., 1994. Strategies of seed dispersal and germination in plants inhabitingdeserts. Bot. Rev. 60, 373–425.Herrera, J., 1986. Flower and fruiting phenology in the coastal shrublands ofDonana, South Spain. Vegetatio 68, 91–98.Howe, H.F., Westley, L.C., 1988. Mechanics and ecology of mutualism. In: Howe,H.F., Westley, L.C. (Eds.), Ecological Relationships of Plants and Animals. OxfordUniversity Press, Oxford, pp. 107–160.Howe, H.F., 1989. Scatter and clump-dispersal and seedling demography: hypothesisand implications. Oecologia 79, 417–426.Howe, H.F., 1993. Aspects of variation in a neotropical seed dispersal system.Vegetatio 107/108, 149–162.Hu, Z., Yu, M., Ding, B., Fang, T., Qian, H., Chen, Q., 2003. Types of evergreenbroadleaved forests and the species diversity in Gutian Mountain NatureReserve. Chin. J. Appl. Environ. Biol. 9, 341–345.Hughes, L., Dunlop, M., French, K., Leishman, M.R., Rice, B., Rodgerson, L., Westoby,M., 1994. Predicting dispersal spectra—a minimal set of hypotheses based onplant attributes. J. Ecol. 82, 933–950.Jackson, J.F., 1981. Seed size as a correlate of temporal and spatial patterns of seedfall in a neotropical forest. Biotropica 13, 121–130.Janzen, D.H., 1970. Herbivores and number of tree species in tropical forests. Am.Nat. 104, 501–528.Legendre, P., Mi, X.C., Ren, H.B., Ma, K.P., Yu, M.J., Sun, I.F., He, F.L., 2009. PartitioningBeta diversity in a subtropical broad-leaved forest of China. Ecology 90, 663–674.Levin, S.A., Muller-Landau, H.C., Nathan, R., Chave, J., 2003. The ecology andevolution of seed dispersal: a theoretical perspective. Annu. Rev. Ecol. Syst.34, 575–604.Li, M.J., Wang, Z.H., 1984. The phenology of common plants in Ding Hu Shan (inChinese). Trop. Subtrop. Forest Ecosys. 2, 1–11.Lieberman, D., 1982. Seasonality and phenology in a dry tropical forest in Ghana. J.Ecol. 70, 791–806.Liu, J.M., 2000. A preliminary study on the soil seed bank dynamics of the DistyliumChinensis community in the Maolan Karst forst. Acta Phytoecol. Sin. 24, 366–374.Machado, I.C.S., Barros, L.M., Sampaio, E., 1997. Phenology of caatinga species atSerra Talhada, PE, northeastern Brazil. Biotropica 29, 57–68.Muller-Landau, H.C., Wright, S.J., Calderon, O., Condit, R., Hubbell, S.P., 2008.Interspecific variation in primary seed dispersal in a tropical forest. J. Ecol.96, 653–667.Murali, K.S., Sukumar, R., 1994. Reproductive phenology of a tropical dry forest inMudumalai, southern India. J. Ecol. 82, 759–767.Murray, D., 1986. Seed Dispersal. Academic Press, California.Nathan, R., Muller-Landau, H.C., 2000. Spatial patterns of seed dispersal, theirdeterminants and consequences for recruitment. TREE 15, 278–285.Selwyn, M.A., Parthasarathy, N., 2006. Reproductive traits and phenology of plantsin tropical dry evergreen forest on the coromandel coast of India. Biodivers.Conserv. 15, 3207–3234.Sharpe, D.M., Fields, D.E., 1982. Integrating the effects of climate and seed fallvelocities on seed dispersal by wind: a model and application. Ecol. Model. 17,297–310.Smythe, N., 1970. Relationships between fruiting seasons and seed dispersalmethods in a neotropical forest. Am. Nat. 104, 25–35.Stiles, E.W., 1980. Patterns of fruit presentation and seed dispersal in bird-disseminatedwoody-plants in the eastern deciduous forest. Am. Nat. 116, 670–688.Tiffney, B.H., 1984. Seed size, dispersal syndromes, and the rise of the angiosperms—evidence and hypothesis. Ann. Miss. Bot. Gard. 71, 551–576.Turcek, F.J., 1963. Color preference in fruit- and seed-eating birds. Proc. Int. Ornithol.Congra. 13, 285–292.Van der Pijl, L., 1982. Principles of Dispersal in Higher Plants. Springer-Verlag,Berlin.Vanschaik, C.P., Terborgh, J.W., Wright, S.J., 1993. The phenology of tropical forests—adaptive significance and consequences for primary consumers. Annu. Rev.Ecol. Syst. 24, 353–377.Volis, S., 2007. Correlated patterns of variation in phenology and seed production inpopulations of two annual grasses along an aridity gradient. Evol. Ecol. 21, 381–393.Wheelwright, N.T., 1985. Fruit size, gape width, and the diets of fruit-eating birds.Ecology 66, 808–818.Willson, M.F., Irvine, A.K., Walsh, N.G., 1989. Vertebrate dispersal syndromes insome Australian and New-zealand plant-communities, with geographic comparisons.Biotropica 21, 133–147.Willson, M.F., Thompson, J.N., 1982. Phenology and ecology of color in bird-dispersedfruits, or why some fruits are red when they are green. Can. J. Bot. 60,701–713.Willson, M.F., Whelan, C.J., 1990. The evolution of fruit color in fleshy-fruited plants.Am. Nat. 136, 790–809.Wright, S.J., Calderón, O., 1995. Phylogenetic constraints on tropical floweringphenologies. J. Ecol. 83, 937–948.Wright, S.J., Carrasco, C., Calderón, O., Paton, S., 1999. The El Niño southernoscillation, variable fruit production and famine in a tropical forest. Ecology80, 1632–1647.Yu, M.J., Hu, Z.H., Yu, J.P., Ding, B.Y., Fang, T., 2001. Forest vegetation types inGutianshan natural reserve in Zhejiang. J. Zhejiang Univ. (Agric. Life Sci.) 27,375–380.273

Journal ofPlant EcologyVolume PAGES 1–10 5, Number 3,Pages 346–355doi: 10.1093/jpe/rtr048July 2012doi:10.1093/jpe/rtr048Advanced Access publishedon 11 January 2012available online atwww.jpe.oxfordjournals.orgSpatial associations of tree speciesin a subtropical evergreenbroad-leaved forestZheng R. Luo 1 , Ming J. Yu 1 , De L. Chen 2 , You G. Wu 2 andBing Y. Ding 3, *AbstractAimsThe spatial segregation hypothesis and the low-frequency hypothesisare two important proposed mechanisms that delay or prevent competitiveexclusion in ecosystems. Because tree species interact withtheir neighbors, the importance of these potential processes can beinvestigated by analyzing the spatial structures of tree species.MethodsThe distribution of the adults of 27 common tree species in a fullymapped 5-ha subtropical forest plot in Baishanzu, eastern China,was analyzed to investigate the community-level intra- and interspecificspatial association patterns. We first tested for the overall spatialpattern in the 5- to 40-m neighborhoods and classified first-order bivariateassociations with a diametric scheme based on Ripley’s K andnearest-neighbor statistic (G-function). Then heterogeneous Poissonnull models were used to distinguish second-order interactions fromoverall spatial associations (including first-order effects). Finally, weanalyzed correlations between the existence of species interactionsand some attributes of the species involved.1 Key Laboratory of Conservation Biology for Endangered Wildlife of the Ministry of Education, and College of Life Science,Zhejiang University, Hangzhou 310058, China2 Management of Baishanzu, Fengyangshan-Baishanzu National Nature Reserve, Qingyuan 323800, China3 Department of Biology Science, The School of Life and Environmental Science, Wenzhou University, Wenzhou 325027, China*Correspondence address. The School of Life and Environmental Science, Wenzhou University, Wenzhou 325027,China. Tel/Fax: 086-577-86689259; E-mail: dby@wzu.edu.cnImportant FindingsPartial overlap and segregation increased with scale, whereas mixingdecreased. Nearly 70% of the species pairs occurred less thanexpected at random, and only 3.4% of the species pairs were wellmixed; 11.0% of all species pairs showed significant small-scaleinteractions, which was a greater frequency than expected by chanceif species are abundant or prefer the same habitat, but less frequentthan expected if species are highly aggregated. This suggests thatboth spatial segregation and low frequency of species facilitate speciescoexistence by reducing the opportunity that trees of two speciesencounter each other. The study also revealed that positive interactionswere more prevalent than negative interactions in the forest,which indicates that positive interactions may have important effectson forest species assemblies.Keywords: Baishanzu d point pattern analysis d spatialsegregation d low frequency d interaction opportunityReceived: 1 May 2011 Revised: 30 October 2011 Accepted: 3December 2011Downloaded from http://jpe.oxfordjournals.org/ at Zhejiang University on September 7, 2012INTRODUCTIONThe relative importance of inter- and intraspecific interactionsis one of the central debates in ecology. Although positiveinteractions between species have recently been reported, especiallyin stressful habitats (Callaway et al. 2002; Martinezet al. 2010), competition has been accepted as a prevalent processin nature since the 1930s (Gause 1934). Specifically, competitionhas important effects on composition and structure ofplant communities (e.g. Chesson 2000; Grime 1977; Tilman1994). Due to interspecific competition, species in a communitywill exclude each other until only one species remains (Wright2002). However, many mechanisms have been proposed thatdelay or prevent competitive exclusion. Given the fact thatplants are sessile organisms, the frequency at which two speciesdirectly interact will be influenced by the spatial patterns they(separately and jointly) exhibit and their relative abundances.The Spatial Segregation Hypothesis contends that highly patchy© Ó The Author 2012. Published by Oxford University Presson on behalf of of the the Instituteof of Botany, Chinese Academy of of Sciences and and the the Botanical Society of China. of China.All rights reserved. For permissions, please email: journals.permissions@oup.com274

Luo 2 et al. | Analyzing plant species association347Journal of Plant Ecologydistributions lead to interspecific segregation and, thus, on averageindividuals will interact with conspecifcs rather than heterospecifics(Pacala 1997). Most tropical and subtropical forestsare species rich, which usually means low densities of individualspecies. This generally weakens direct interspecific interactionsamong species. If two species occur at low frequencies, they dorarely encounter each other as proximal neighbors (Perry et al.2009). The Low-Frequency Hypothesis states that, due to thelow frequencies of most species in forests, pairwise species interactionsare rare occurrences. The common ground betweenthese hypotheses is that reducing the probability of encountersbetween pairs of species avoids competitive exclusion.Whatever coexistence mechanisms are operating in the forest,they should leave a spatial signature (Hubbell et al. 2001).Space can be used as a surrogate for uncovering ecological processthrough the study and analysis of spatial patterns (McIntireand Fajardo 2009). If spatial mechanisms, such as the spatialsegregation hypothesis and the low-frequency hypothesis, areindeed important for species coexistence, one would expectemergence of distinct spatial patterns, not only intraspecificbut also interspecific. However, while there have been plentyof studies that investigate intraspecific spatial structure, fewattempts have been made to analyze species spatial associationsas a subset of species biological interactions (e.g. Kubota et al.2007; Martinez et al. 2010; Perry et al. 2009; Wang et al.2010; Wiegand et al. 2007). Such studies are necessary to revealpotential biological processes that control the assembly, dynamicsand functioning of forest ecosystems.Interspecific spatial repulsion is demonstrated by fewer heterospecificneighbors on average than expected in a randomdistribution and can be explained by negative interactions(e.g. competition). Interspecific spatial attraction, on the otherhand, can be a result of positive interactions (e.g. facilitation)and manifests itself as higher than expected neighborhooddensities of heterospecifics. However, other substantially differentprocesses performed at different scales can create similarspatial patterns. For example, spatial repulsion (or attraction)between species may be also explained by different (or similar)microhabitat preferences (i.e. shading, soil moisture or nutrientlevels; Wright 2002). Large-scale interspecific distributionpatterns that are usually determined by species’ habitat preferenceswill confound small-scale patterns if the analyses arenot conducted appropriately (Wang et al. 2010). This makesthe true underlying ecological processes behind patterns inspecies distributions difficult to elucidate (i.e. biological limitation).Thus, it is essential to integrate all relevant ecologicalinformation when making a priori inferences from ecologicaltheory, which will help break up the biological limitation(McIntire and Fajardo 2009). For instance, if the Spatial SegregationHypothesis is true in the forest, (i) most species shouldbe aggregated and most species pairs should be spatially segregatedespecially at large spatial scales, (ii) the probabilities ofsignificant interspecific associations should negatively correlatewith the degrees of aggregations of species involved and(iii) species pairs with the same habitat preference should associatemore frequently at small spatial scales. Alternatively,given that the Low-Frequency Hypothesis applies to an ecosystem,(i) the probabilities of significant interspecific associationsshould positively correlate with densities (or abundances) ofinvolved species and (ii) a higher mean relative species densityof a forest (i.e. abundance of one species relative to the totalabundance of all species or the inverse of species richness) willlead to a lower frequency of interspecific associations in theforest.Precise application of analytical tools is important to effectivelytest the above inferences we have made (McIntire andFajardo 2009). A considerable problem in this study is to separatethe effect of the abiotic environment from the effect ofdirect plant interactions. One effective approach to studyingspecies interactions is to use heterogeneous Poisson processesas null models to account for larger spatial scale patterns whensmall-scale patterns are of primary interest (Wiegand et al.2007). This approach is in the spirit of the empirically basedknowledge that plant–plant interactions (second-order effects)usually work at small scales (e.g. 0–15 m) and species environmentinteractions (first-order effects) work at large scales (e.g.>15m). Another problem is to precisely describe the overallheterogeneous patterns of interspecific associations. Informationof relative neighborhood density or nearest-neighbor distancealone does not unambiguously characterize heterogeneouspatterns. Wiegand et al. (2007) developed a simple scheme integratinginformation of relative neighborhood density andnearest-neighbor distance to classify overall associations.In this article, we use the above-mentioned analytical toolsto assess the type and frequency of species distributions andassociations in a subtropical forest in eastern China. In orderto test the (i) Spatial Segregation Hypothesis and (ii) Low-Frequency Hypothesis for species coexistence, we tested thepriori inferences drawn from the hypotheses. Specifically,we analyzed (i) how many species pairs were spatially segregated,(ii) how many species pairs associated at small scales,(iii) how the probability of interspecific association was relatedto the species’ properties, such as species density, degree of intraspecificaggregation, and species habitat preference, and(iv) the relationship between community-wide relative densityand frequency of association. In order to answer the lasttwo questions, we conducted further analyses: we correlatedthe P values of interspecific associations with the intraspecificg(r) values (see Methods for calculations of these twostatistics) and abundances of species involved; we checkedthe relationship between the frequency of significant interspecificassociations and habitat preference of speciesinvolved with permutation tests, and finally, we comparedthe percentages of significant interspecific associations amongdifferent forests.Study siteThe study site (Baishanzu Forest Dynamic Plot, hereafterBaishanzu FDP), a 250 3 200 m permanent forest plot, is locatedin the Baishanzu mountains in southeast China (119°3#53$E,Downloaded from http://jpe.oxfordjournals.org/ at Zhejiang University on September 7, 2012275

Luo 348et al. | Analyzing plant species association Journal of Plant Ecology 327°40#54$N). In summer, the region is influenced by the southeastmonsoon, which carries a large amount of water from thePacific Ocean. The Baishanzu FDP is representative of midaltitudesubtropical mountain evergreen broadleaf forests, whichis the typical vegetation in the subtropical monsoon climate.Many rare species exist in the old-growth species-rich communityof the Baishanzu FDP, while the abundance of dominantspecies is much higher than other species (Xu et al. 2007).The site is on a north-facing hillside with a slope ranging from7.27° to 44.12°.The Baishanzu FDP (1 400–1 600 m elevation) was establishedin July 2003, when it was divided into 120 20 3 20 mand 10 20 3 10 m grids by using a total station. All trees with>1 cm diameter at 1.3 m height (DBH) and seedlings (individualswith DBH < 1 cm and/or height < 1.3 m) of non-shrub specieswere tagged, measured, stem mapped and identified to thespecies level. We monitored 156 tree species in the census. Theyshow varying degrees of association to topography (Wang et al.2011).Study speciesIn this study, all species with >30 adult trees in the census wereincluded. In order to include more shrub and understory species,we developed criteria to define adult trees. Trees wereconsidered adult when DBH > 8 cm for canopy species;DBH > 4 cm for understory species and DBH > 2.5 cm forshrub species. Adult specimens of 27 species (7 canopy species,13 understory species and 7 shrub species) were analyzed.These 27 species accounted for ;93.8% of the total numberof adults in the plot. There were 6 species with >1 000 adults,including the most abundant species, Rhododendron latoucheae,with >6 000 adults. All these species showed aggregation atsome scale (Luo et al. 2009). There were 14 species that showedsignificant habitat preference (Wang et al. 2011). The ecologicalattributes of these species are shown in Table S1 in theonline supplementary material.Spatial pattern analysisSummary second-order statistics.Sophisticated techniques of spatial point pattern analysiswere used to quantify species associations. Specifically, wechose the summary statistics of Ripley’s (1976) K-function,pair-correlation function (Stoyan and Stoyan 1994) and thedistribution function of near neighbor distances G(y) (Diggle2003) to analyze the data. The bivariate K-function K 12 (r)was defined as the expected number of Type 2 points withinradius r of an arbitrary Type 1 point, divided by intensity k 2 ofType 2 (i.e. the expected number of points of Type 2 per unitarea). The bivariate pair-correlation function g 12 (r) based onpoint-to-point distance is related to the derivative of theK 12 -function, i.e. g 12 (r) = K’ 12 (r)/2pr (Stoyan and Stoyan1994). K 12 (r) is a cumulative distribution function wherethe values of K 12 (r) at larger scales include the values ofK 12 (r) at smaller scales. In contrast, the pair-correlation functiong 12 (r) is a non-cumulative distribution function in whichg 12 (r) is the expected density of Type 2 points in a ring (itswidth Dr is very small) of a given distance r around a Type1 focal point divided by intensity k 2 of Type 2 (Wiegandand Moloney 2004). By factoring out the interference amongdifferent distances, the pair-correlation function allows for aprecise assessment of scales where significant point-to-pointinteractions occur (Wiegand et al. 2007). The univariate Kand pair-correlation functions are analogous to the bivariatefunctions, but the focal point is not counted (Wiegand andMoloney 2004). The statistic G 12 (y) evaluates the fractionof points of the focal Type 1 that have their nearest Type2 neighbor within distance y (Diggle 2003; Illian et al.2008). This statistic provides information on the bivariateemptiness probability not provided by K 12 (r) (Wiegandet al. 2007).Statistical hypotheses and null modelsSpatial analysis 1: Overall non-random association.The null statistical hypothesis of this analysis (i.e. homogeneousprocess) was that Species 2 is randomly distributed inthe plot, irrespective of the distribution of Species 1. In orderto distinguish the difference between observed associationsand random distribution hypothesis, we implemented completespatial randomness (CSR) as the null model of this test.In this null model, the locations of the focal species were fixed,but Species 2 were distributed randomly and independentlyof the locations of Species 1. Additionally, with the schemedeveloped by Wiegand et al. (2007), we categorized all associationsof the heterogeneous bivariate patterns. The commonlyused statistics bivariate K 12 (r) and the G 12 (r) were implementedto construct the two axes of the scheme. The two axesP and M are defined asPðrÞ = Ĝ 12 ðrÞ G exp12 ðrÞ: ð1ÞMðrÞ = lnð ˆK 12 ðrÞÞ lnðK exp12 ðrÞÞ: ð2ÞThe theoretical value of the two summary statistics underCSR is known (G exp12 r =1 e k2pr2 and K exp12 ðrÞ = pr2 ), but weused the mean of the Monte Carlo simulations because theedge-corrected estimates of K 12 (r) are not unbiased (Perryet al. 2006). The M axis and P axis evaluate two fundamentalaspects of bivariate point patterns like the bivariate K 12 - andG 12 -functions. The M axis indicates the extent to which thereare more (positive value) or less (negative value) points ofSpecies 2 than expected (K-function), while the P axis indicatesthe extent to which the probability of having a nearestneighbor of Species 2 within distance r is higher (positivevalue) or lower (negative value) than expected (emptinessprobability). P(r) can distinguish the state in which manySpecies 1 points have no Species 2 neighbor but few Species1 points have many Species 2 neighbors (with low values ofP(r)) from the state where all Species 1 points have a similarnumber of Species 2 neighbors (with high values of P(r)); thisDownloaded from http://jpe.oxfordjournals.org/ at Zhejiang University on September 7, 2012276

4Luo et al. | Analyzing plant species association349Journal of Plant Ecologycan not be distinguished by M(r). P(r) will be negative if theproportion of nearest neighbors within distance r is smallerthan expected and positive if the proportion is larger thanexpected. Similarly, M(r) will be negative if the average numberof neighbors within distance r is smaller than expected and positiveif the number is larger than expected. Further discussionabout this scheme can be found in Wiegand et al. (2007) andWang et al. (2010). Following this scheme, species spatial associationscan be roughly classified into five types:1. ‘segregation’ (P(r) < 0 and M(r) < 0) where two species aresegregated in space;2. ‘partial overlap’ (P(r) < 0 and M(r) > 0) where many plants ofSpecies 1 have no Species 2 neighbors within distance r, butother plants of Species 1 have plenty of Species 2 neighbors;3. ‘mixing’ (P(r) > 0 and M(r) > 0) where two species are mixedwell in space;4. ‘strong interaction’ (P(r) >0 and M(r) < 0) where mostindividuals of the two species are segregated, butoccasionally plants of Species 2 are the common nearestneighbors of all the focal plants in the strongly aggregatedcluster; and5. ‘null association’ (or Type 0, P(r) = 0 and M(r) = 0) proposedby Martinez et al. (2010). A null association arises ifneither K 12 (r) nor G 12 (r) shows significant departures fromthe CSR.Spatial analysis 2: small-scale plant–plant interactionsA univariate g-function was used to investigate the spatialpatterns of conspecific adults. In order to reveal significantsecond-order effects in the univariate patterns (i.e. regularityor aggregation), we examined the discrepancies betweenthe observed g-function and the 95% confidence envelope ofthe simulated null model. Here we used heterogeneous Poissonpoint processes (HP) as the null model because it isa shortcut for separating the first- (i.e. habitat preference)and second (i.e. direct plant-plant interactions)-order effects(Wiegand et al. 2007). Diggle (2003) suggested that small-scaleeffects are attributed to second-order plant–plant interactionsand large-scale effects are attributed to environmental heterogeneity.In HP, the occurrence of any point is independent ofthat of any other, but the points are distributed in accordancewith an intensity function k(x, y) that varies with location (x, y)(Stoyan and Stoyan 1994; Wiegand and Moloney 2004). TheEpanechnikov kernel function, a nonparametric method, wasused to estimate the intensity function k(x, y) of a given pointpattern (Wiegand et al. 2007).In order to reveal significant second-order effects in thebivariate patterns (i.e. repulsion or attraction), we kept the locationof the individuals of the first species fixed and randomizedthe locations of the individuals of the second species underthe HP model. Estimation of intensity of Species 2 was conductedas in the univariate patterns analysis. This allowedus to assess, given the intensity of Species 2, whether pointsof Species 2 were found more or less frequently than expectedaround points of Species 1, which would suggest interspecificattraction or repulsion, respectively. We tested all possible pairsof the 27 species. Each species was analyzed as Species 1 andSpecies 2 because there was no reason to assume that the interactionswould be symmetric.Significant tests comparing patterns with the nullmodelSignificant departures from the underlying null model weretested by overlap with 95% confidence envelope, whichwas constructed by using the 5th lowest and the 5th highestvalue of 199 Monte Carlo iterations, or, if the number of treesof Species 2 was u jð3Þs + 1where I(u 0 > u j ) is an indicator function that equals 1 if u 0 > u j andequals 0 otherwise, and s represents the number of simulationiterations. u i is a summary statistic that measures the discrepancybetween the empirical measurement and the theoretical pair correlationfunction over a distance interval of interest, withr maxu i = + ½ĝ i ðr k Þg i ðr k Þ 2 dr k ;r k = r minwhere r k is distance, r min and r max are lower and upper limits ofthe summation in terms of distance. ĝ i ðr k Þ is the empirical measurementfor pattern i for g(r), g i ðr k Þ is the mean result computedfor all simulated patterns except for i, and dr k =(t k+1 t k )is the width of the distance interval (Diggle 2003; Loosmoreand Ford 2006).Bandwidth selection for intensity estimation of eachspeciesThe Epanechnikov kernel function that was applied to estimatethe intensities of species depends on the parameter bandwidth R.It is important to select a suitable bandwidth parameter for separatingbiological effects, as an unsuitable R may underestimateor overestimate intensity k(x,y). For example, Zhu et al. (2010)reported that the HP null model led many randomly distributedspecies show fake large-scale regularity in terms of statistical test.This may be caused by overestimation of intensity k. In general,plant–plant interactions are restricted to within a limited spatialseparation. An individual-based analysis of plant survival inthis forest revealed that direct interactions usually occuramong plants within 15 m of each other. This observationwas used to select the bandwidth for each species.ð4ÞDownloaded from http://jpe.oxfordjournals.org/ at Zhejiang University on September 7, 2012277

350 Luo et al. | Analyzing plant species association Journal of Plant Ecology5We used a procedure to select the best bandwidths for eachspecies (listed in Table S1 in the online supplementarymaterial), as follows:1. Given R = R*, we estimated the intensity function of aspecies using the Epanecnikov kernel function.2. Then, we simulated 199 patterns under the HP null modelwith the intensity function estimated from the species.3. We calculated g(r) values of these modeled patterns and theobserved pattern.4. After that, we estimated the P value of the GOF test usingEquations (3) and (4) and restricting r k to 15–35 m (i.e.t min = 15, t max = 35).5. We repeated Step 1 through 4 using R* values ranging from10 to 25 m, with step intervals of 1 m and assessed P values ateach step.6. We chose the bandwidth corresponding to the largestP value as the suitable bandwidth for the species.Additional testsIn order to assess how the probability of significant interspecificinteractions varies with species aggregation and abundance,we correlated the P value of the GOF with thenumber of stems of species pairs and the univariate g(r) of focalor neighbor species at scales r = 0, 2, 6, 10, 15 and 30 m usingSpearman’s q. Significance of these correlations was determinedwith a t-test. Testing for similar habitat preferencesof species pairs between which interactions have been identifiedis much more difficult. Here we used a permutationtest: (i) based on the matrix of interspecific interactions (i.e.Table S2 in the online supplementary material), the observednumber of species pairs (note that Species 1 vs. Species 2 andSpecies 2 vs. Species 1 are considered as different species pairs)which had the same habitat preference and showed a significantinteraction (i.e. ‘p’ or ‘n’ in Table S2 (see online supplementarymaterial) indicating positive and negative interactionrespectively) were counted; (ii) ps and ns were randomlyreplaced over the matrix (but the diagonal of the matrix shouldalways keep empty); (iii) the number of species pairs with thesame habitat preference and a significant interaction werecounted again based on the new distribution of ps and ns inthe matrix; and (iv) we repeated steps 2 and 3 999 times resultingin 1 000 predictions of number of interacting species pairswith similar habitat preferences if the relationship were random.If the observed number belong to the largest (or smallest)2.5% of the total numbers, interspecific interaction was concludedto occur more (or less) frequently between species withthe same habitat preference than between species with differenthabitat preferences.All these analyses were conducted in R statistical software(R Development Core Team 2009). We used the package spatstat(Baddeley and Turner 2005) in R to perform spatial pointpattern analyses.RESULTSAnalysis 1: Overall non-random associationsAs shown in Fig. 1, the relative frequencies of overall interspecificassociations depended on scale. Most changes in the relativefrequencies of the different bivariate association typesoccurred at scales smaller than 15 m (Fig. 1). The frequencyof species pairs where neither summary statistic (i.e. K 12 norG 12 ) showed significant departure from the CSR model was relativelyhigh in small neighborhoods. Partial overlap and segregationincreased and mixing decreased with increasingspatial scale. The analyzed 702 species pairs were not equallydistributed within the 2D classification space (Fig. 1, 2). Takingthe proportion of cases (i.e. species pairs) at the 20-m spatialscale as an example, segregation and partial overlap were thetwo most frequent associations, making up 34.2 and 35.6%of all cases, respectively. As expected, type IV associationsFigure 1: assessment of scale dependence on overall associations. Thefigure shows the proportion of the 702 species pairs studied categorizedinto five association types.Figure 2: allocation of the overall association of the 702 species pairsinvolving 27 species in the Baishanzu FDP at 20 m based on the classificationscheme described in Analysis 1.Downloaded from http://jpe.oxfordjournals.org/ at Zhejiang University on September 7, 2012278

6Luo et al. | Analyzing plant species association351Journal of Plant Ecology(i.e. strong interaction) occurred in only 1.2% of all cases andonly 3.4% species pairs were well mixed. This distribution patternof classifications suggested that for most species pairs,trees encounter a neighbor from the counterpart species lessoften than expected by chance [P(r) < 0], though there are still25.6% of species pairs whose association can be predicted bytheir abundance alone (i.e. not significant under CRS model).Analysis 2: small-scale plant–plant interactionsUnivariate case.In 27 adult patterns examined, 20 of them showed significantdeparture from HP null models (i.e. P value from GOFtest < 0.05). Nineteen species were fine-scale aggregated.The univariate pair-correlation function showed that noneof the species was significantly regular at scale < 5 m and 3 speciesexhibited regularity at mid-scale (5–15 m). Among thesethree species, only Acer olivaceum did not show aggregation atsmall scale while the other two species did. The remainingseven species followed the HP null models.In order to roughly estimate the effect of scale on speciesspatial patterns, we counted the number of species (withP value < 0.05 by GOF test) for each detail scale r wherethe observed pattern showed significant aggregation or repulsion(i.e. g(r) value was above or below the envelopes constructedby Monte Carlo simulations). Results showed thatthe number of species exhibiting aggregation peaked at 3–4 mand decreased with the increasing scale, while the numberof species exhibiting regularity fluctuated with the increasingscale from 5 to 15 m (Fig. 3a).Bivariate case.A total of 702 bivariate point pattern analyses for all adult pairsof the 27 species were executed. For 77 of these species pairs,the GOF test revealed a significant association (11.0%); in 58of these cases, the small-scale association was positive (attraction),and in 19, it was negative (repulsion). In order to estimatethe magnitude of scale dependence, we also counted foreach scale the number of species exhibiting significant attractionor repulsion (using only species pairs where the P value ofthe GOF test was < 0.05). We found that attraction occurredmore frequently than repulsion, especially at small scales; thecount of attraction showed a decreasing trend with increasingscale and stabilized for scales >15 m at a count of ;0–2 species,while the count of repulsion peaked at 2 m with 13species pairs and decreased gradually with increasing scale(Fig. 3b).Among 77 significantly interacting species pairs, 42 of themwere symmetric, while the other 35 species pairs were asymmetric(see Table S2 in the online supplementary material).Both positive and negative interactions showed some symmetricalcases (19 and 2 cases, respectively). Symmetric interactionsusually involved an abundant species. From the 27species analyzed, only one species (Illicium angustisepalum)did not show any significant small-scale (0–15 m) associationto another species. This species showed a relatively high degreeof univariate clustering and relatively low abundance. On theother hand, two species (R. latoucheae and Cleyera pachyphylla)showed significant interaction with >10 other species (14 and17 species, respectively). Rhododendron latoucheae was the mostabundant species in the community and was scatteredthroughout the plot. Most interactions involving this specieswere positive and symmetric; only one species (Sycopsis sinensis)negatively interacted with R. latoucheae. Cleyera pachyphyllashowed negative association with 9 species. It is interesting tonote that this species showed strong clonal reproduction.The P value of the GOF test was moderately and negativelycorrelated with the number of stems of neighbor species (Species2) (q = 0.25; P < 0.01), weakly and negatively correlatedwith the number of stems of focal species (Species 1) (q = -0.17;p< 0.01) and negatively correlated with the product and sum ofthe number of stems of two species (q = 0.31 and q = 0.30,respectively, both P < 0.01, see Table 1). This suggests that theabundant species interacted more frequently with other species.We also found a positive but weak correlation of P valuewith the clumping of two species. The two correlations bothpeaked at scale 2 m (q = 0.19 for Species 1 and q = 0.12 forSpecies 2, see Table 1). This suggested that small-scale clumpingspecies were less likely to interact with other species.Downloaded from http://jpe.oxfordjournals.org/ at Zhejiang University on September 7, 2012Figure 3: Scale-dependent biotic interactions. The figures exhibit number of species (univariate) and species pairs (bivariate) showing significantpositive and negative adult interactions over different scales in the Baishanzu FDP.279

352 Luo et al. | Analyzing plant species association Journal of Plant Ecology7Table 1: correlation of the P value of the GOF test of the bivariateanalysis 2 with several variables describing species abundanceand aggregationVariableAdditionally, the permutation test showed that significantinteractions, especially for positive interactions (P < 0.01), occurredsomewhat more frequently for those species with thesame habitat preference (P < 0.05). These results indicate thatspecies with similar habitat preference tended to interact.DISCUSSIONCorrelation qn s1 3 n s2 0.31***n s1 + n s2 0.30***n s1 0.17***n s2 0.25***g 11 (r = 0.1 m) 0.05g 11 (r = 2 m) 0.19***g 11 (r = 6 m) 0.18***g 11 (r = 10 m) 0.12***g 11 (r = 15 m) 0.14***g 11 (r = 30 m) 0.14***g 22 (r = 0.1 m) 0.03g 22 (r = 2 m) 0.12***g 22 (r = 6 m) 0.06g 22 (r = 10 m) 0.06*g 22 (r = 15 m) 0.05g 22 (r = 30 m) 0.06The correlations are not corrected for multiple testing. n s1 , the numberof individuals of Species 1; n s2 , the number of individuals of Species 2;g 11 and g 22 are the values of the univariate pair correlation functions ofSpecies 1 and 2, respectively, at the specified spatial scale r.*P < 0.1; **P < 0.05; ***P < 0.01.The comprehensive spatial analyses of species distributionsand associations among adult trees of 27 common species ina subtropical evergreen broad-leaved forest in eastern Chinarevealed a variety of strong scale-dependent spatial structures.We found that most species were aggregated at small scales,and segregation and overlap were dominant in overall speciesassociations. More than a quarter of species pairs co-occurredat scales >10 m, by chance alone. Selective analysis of smallscaleeffects revealed that 89% of species pairs did not exhibitsignificant interactions between adult plants (note that we didnot perform multiple testing here, if we had, the percentagewould have been higher). The findings highlight that neutralspecies–species associations are prevalent in this subtropicalforest, which would facilitate species coexistence. Manytheories may explain the lack of positive or negative species–species interactions (e.g. Chesson 2000; Hubbell 2001; Wright2002). Here we mainly discuss the importance of the segregationhypothesis and the low-density hypothesis in interpretingthe patterns we observed.The segregation hypothesis and the low-frequencyhypothesisThese two hypotheses are known as mechanisms that violatethe condition of the competitive exclusion principle (Chesson2000; Wright 2002). The common underlying mechanism ofthese hypotheses is that spatial segregation or low densitiesamong species decreases the probability of interspecificencounters with the effect of weakening interspecific interaction(Palmer 1994; Vazquez et al. 2007). All 27 species weresignificantly aggregated, which separate different species inspace (Luo et al. 2009). Our study showed that spatial segregationand partial overlap were the most dominant (near70%) overall species association types, which suggests thattrees encounter a neighbor from the counterpart species lessoften than expected by chance [P(r) < 0]. More than a thirdof species pairs were segregated completely; these pairs of speciesrarely occupy the same areas, allowing plants of differentspecies to ‘avoid’ each other. This observation is compatiblewith the segregation hypothesis and explains why 86.6% ofall species pairs did not exhibit significant small-scale associations.The finding that small-scale clumping species are lesslikely to interact with other species (i.e. the P value of theGOF test positively correlated with the clumping of two species)also strongly supports the segregation hypothesis. Thishypothesis is also supported by the finding that species withsimilar habitat preferences tended to interact (especiallyattract) significantly. Habitat heterogeneity attributed to differencesin topography had profound effects on species associations.On one hand, species with the same associationusually inhabit the same area, which gives them more opportunityto interact; on the other hand, adults usually modifytheir local environment, benefiting heterospecifics with similarpreferences and allowing them to take up the same area.Low frequency of most species is also a reason that results ina low percentage of significant species interactions. If most specieswere present at low abundances relative to the number ofspecies, chance alone would make it unlikely that they encountereach other as neighbors (Lieberman and Lieberman2007; Perry et al. 2009). Our results show that significantsmall-scale interspecific interactions are more likely if oneor both species are more abundant. A good example is the mostabundant species, R. latoucheae, which shows significant interactionswith 14 other species, while the rare canopy species,I. angustisepalum, is independent of all other species. An importantdifference between species-rich and species-poor communitiesis that the mean relative species density of species-richcommunities is often low, while that of species-poor communitiesis often high. Under the low-frequency hypothesis, thehigh relative species density in species-poor communitieswould lead to increased frequencies of species–species interactions.A comparison of results from Changbaishan (Wang et al.2010), Sinharaja (Wiegand et al. 2007) and Baishanzu (this paper)supported our expectation that lower mean relative speciesdensities (i.e. higher species richness) in a forest lead toDownloaded from http://jpe.oxfordjournals.org/ at Zhejiang University on September 7, 2012280

8Luo et al. | Analyzing plant species association353Journal of Plant Ecologya lower frequency of interspecific association (see Table S3 inthe online supplementary material). These results suggest thatthe low-frequency hypothesis elucidates important mechanismsinfluencing the frequency of small-scale interspeciesinteractions within forests. However, further comparisonswith statistical test across sites are still necessary to confirmthe effect of low species density across communities becausethe frequency of significant interactions and mean relativespecies density are reciprocally dependent.Furthermore, the low probability of interspecific encountershas profound influence on species trait evolution that is relatedto a species’ ability to compete. Since the set of neighbors encounteredby individuals of a given species within the immediateneighborhood is quite variable and unpredictable for theindividual (i.e. high biotic uncertainty), Hubbell and Foster(1986) argued that natural selection may be diffuse. Underthese conditions, species are unlikely to develop specific interactionswith other species. In this scenario, interspecific interactionwould be weak even if individuals of the two speciessometimes encountered each other.Our study supports the idea that reducing the probability ofinterspecific encounters may strikingly weaken species interactions.Naturally, if species live in different habitats and haveno direct or indirect interactions with each other, they shouldhave no difficulty coexisting in a region (Chesson 2000). However,species do not have to be strictly segregated in space forregional coexistence. There are still a third of species pairs thatoverlapped and some species pairs mixed well. Spatial patternsare usually controlled by multiple processes, which work atdifferent scales (He et al. 1997; Luo et al. 2009). In a broadsense, species differing in the resources they exploit (i.e. nichecomplementarity) would weaken interspecific interactions(Chesson 2000; Wills et al. 2006).Interpretations of intraspecific aggregationSimilar to results from other tropical, subtropical and temperateforests (e.g. Condit et al. 2000; Li et al. 2009; Wang et al.2010), we found that intraspecific aggregation is prevalentin this subtropical forest. The high degree of individual speciesaggregation has important effects on species spatial separationand their interactions (DeBoeck et al. 2006; Seidler and Plotkin2006). Species habitat preference is an important process leadingto the aggregation pattern here (Luo et al. 2009). More thanhalf of these species exhibited habitat preference (see Table S1in the online supplementary material). This suggests that thebenefit of growing in favorable habitats may overwhelm thenegative effects of sharing that habitat with conspecifics(Getzin et al. 2008). Our study showed that the observed patternsof seven species were undistinguishable from predictionsof a heterogeneous Poisson model, suggesting that patterns ofthese species can be explained by habitat heterogeneity alone.Preceding studies also indicate that dispersal limitation alsocontributes to aggregation patterns (Grubb 1977; He et al.1997). Indeed, we found that even after accounting for theeffects of habitat preference, there were 19 species that wereaggregated at a small scale. This suggests that the effects of dispersallimitation are very strong. Fruit type and seed size ofthese species show that most species in the community arenot dispersed far from their parent trees, which disperse theirseed by gravity or wind. Habitat heterogeneity and dispersallimitation are the two most important effects that determinedspecies distribution patterns in forests (Shen et al. 2009). Moreover,the lack of evidence of spatial regularity of individualspecies at small scales suggests Janzen–Connell effects and intraspecificcompetition is absent or weak at least. The Janzen–Connell hypothesis posits that mother trees impair survivalof offspring where natural enemies are aggregated (Connell1971; Janzen 1970). The effect will lead to a regular spatial distributionof adults. However, further studies are needed toconfirm the absence of Janzen–Connell effects and intraspecificcompetition because strong effects of dispersal limitationand habitat preference may mask them.The importance of positive interactions in forestsPositive associations have been documented in stressed environmentslike alpine, arid and Mediterranean-type plant communities(Armas and Pungnaire 2005; Callaway et al. 2002;Riginos et al. 2005), but are rare in forest communities (butsee Kubota et al. 2007 and Martinez et al. 2010). Our selectiveanalysis of small-scale effects revealed that positive associationsare more prevalent than negative associations in BaishanzuFDP, especially between species pairs involving non-canopyspecies with similar habitat preferences. The finding is not surprisingwhen species have similar requirements for establishment;canopy trees modify the local environment in theirvicinity, which facilitates small conspecifics and heterospecificswith similar preferences (Dovciak et al. 2001; Kubota et al.2007); and/or species’ niches differ (Chesson 2000).Such small-scale attractions, which enable species to exploita greater portion of available resources, profoundly influencecommunity structure. We find that the most abundant species,R. latoucheae, associates with 14 other species and only one ofthese associations is negative. Most species can recruit underadults of R. latoucheae. This has interesting implications for communityassembly. By providing habitats for numerous species,the net effect of foundation species on species diversity can actuallybe positive. This phenomenon may need to be integratedinto forest ecological theory (Bruno et al. 2003). For example,Shen et al. (2009) used the joint effects of dispersal limitationand habitat heterogeneity to explain species-area curves inGutianshan plot (25 ha, in subtropics) and BCI (Barro ColoradoIsland) plot (50 ha, in tropics) and showed that the joint effects ofthese two processes did not fit the observed species-area curveswell at intermediate spatial scales (they usually predicted lowerthan the observed species area). The influence of positive interactionsbetween species may explain this departure. Positive facilitativeinteractions between species occurring at some lifestages can compensate for negative competitive interactions atother stages (Callaway et al. 2002; Illian et al. 2009; Martinezet al. 2010). In this case, the patterns that finally emerged wouldDownloaded from http://jpe.oxfordjournals.org/ at Zhejiang University on September 7, 2012281

354 Luo et al. | Analyzing plant species association Journal of Plant Ecology9be predominantly neutral as those found here (Wiegand et al.2007). It would be interesting to study species interactions at differentlife stages and habitats.SUPPLEMENTARY MATERIALSupplementary Tables S1–S3 are available at Journal of PlantEcology online.FUNDINGAppropriative Foundation of Ecology and EnvironmentProtection of Zhejiang Province (ZCJ200317); China NationalProgram for R & D Infrastructure and Facility Development(2008BAC39B02).ACKNOWLEDGEMENTSWe are grateful to Chaozong Zheng Xiaofeng Jin, Jianguo Ai, and XuYang for their technical support for species identification. The authorsthank the many students from Zhejiang University and many colleaguesfrom Management of Baishanzu for their contributions tothe establishment of this 5 ha permanent forest plot. We also thankGuochun Shen, Helge Bruelheide and three anonymous reviewersfor critical comments of this manuscript. Last but not least, we appreciateChristine Verhille at the University of British Columbia andLauren Barry for their assistance with English language and grammaticalediting of the manuscript.Conflict of interest statement. None declared.REFERENCESArmas C, Pungnaire FI (2005) Plant interactions govern populationdynamics in a semi-arid plant community. J Ecol 93:978–89.Baddeley A, Turner R (2005) Spatstat: an R package for analyzing spatialpoint patterns. J Stat Softw 12:1–42.Bruno JF, Stachowicz JJ, Bertness MD (2003) Inclusion of facilitationinto ecological theory. Trends Ecol Evol 18:119–25.Callaway RM, Brooker RW, Choler P, et al. (2002) Positive interactionsamong alpine plants increase with stress. Nature 417:844–8.Chesson P (2000) Mechanisms of maintenance of species diversity.Annu Rev Ecol Syst 31:343–66.Condit R, Ashton PS, Baker P, et al. (2000) Spatial patterns in the distributionof tropical tree species. Science 288:1414–8.Connell JH (1971) On the role of natural enemies in preventing competitiveexclusion in some marine animals and in rain forest trees. IndenBoer PJ, Gradwell GR (eds). Dynamics of Populations. Wageningen,The Netherlands: Centre for agricultural publication and documentation,298–312.DeBoeck HJ, Nijs I, Lemmens CMHM, et al. (2006) Underlying effectsof spatial aggregation (clumping) in relationships between plant diversityand resource uptake. Oikos 113:269–78.Diggle P (2003) Statistical Analysis of Spatial Point Patterns. London,UK: Arnold.Dovciak M, Frelich LE, Reich PB (2001) Discordance in spatial patternsof white pine (Pinus strobus) size-classes in a patchy near-borealforest. J Ecol 89:280–91.Gause GF (1934) The struggle for Existence. Baltimore, MD: Williams& Wilkins.Getzin S, Wiegand T, Wiegand K, et al. (2008) Heterogeneity influencesspatial patterns and demographics in forest stands. J Ecol96:807–20.Grime JP (1977) Evidence for the existence of three primary strategiesin plants and its relevance to ecological and evolutionary theory.Am Nat 111:1169–94.Grubb PJ (1977) The maintenance of species-richness in plant communities:the importance of the regeneration niche. Biol Rev53:107–45.He FL, Legendre P, LaFrankie JV (1997) Distribution patternsof tree species in a Malaysian tropical rain forest. J Veg Sci8:105–14.Hubbell SP (2001) The Unified Neutral Theory of Biodiversity and Biogeography.Princeton, NJ: Princeton University Press.Hubbell SP, Ahumada JA, Condit R, et al. (2001) Local neighborhoodeffects on long-term survival of individual trees in a neotropical forest.Ecol Res 16:859–75.Hubbell SP, Foster RB (1986) Biology, chance and history and thestructure of tropical rain forest tree communities. In Diamond JM, CaseTJ (eds). Community Ecology. New York: Harper and Row, 314–29.Illian JB, Penttinen A, Stoyan H, et al. (2008) Statistical Analysis and Modelingof Spatial Point Patterns. Chichester, UK: John Wiley & Sons.Illian JB, Moller J, Waagepetersen RP (2009) Hierarchical spatial pointprocess analysis for a plant community with high biodiversity.Environ Ecol Stat 16:389–405.Janzen DH (1970) herbivores and the number of tree species in tropicalforests. Am Nat 104:501–28.Kubota Y, Kubo H, Shimatani K (2007) Spatial pattern dynamics over10 years in a conifer/broadleaved forest, northern Japan. Plant Ecol190:143–57.Li L, Huang ZL, Ye WH, et al. (2009) Spatial distributions of tree speciesin a subtropical forest of China. Oikos 118:495–502.Lieberman M, Lieberman D (2007) Nearest-neighbor tree species combinationsin tropical forest: the role of chance, and some consequencesof high diversity. Oikos 116:377–86.Loosmore NB, Ford ED (2006) Statistical inference using the G or Kpoint pattern. Ecology 87:1925–31.Luo ZR, Ding BY, Mi XC, et al. (2009) Distribution patterns of tree speciesin an evergreen broadleaved forest in eastern China. Front BiolChina 4:531–8.Martinez I, Wiegand T, Gonzalez-Taboada F, et al. (2010) Spatial associationsamong tree species in a temperate forest community innorth-western Spain. For Ecol Manag 260:456–65.McIntire EJB, Fajardo A (2009) Beyond description: the active andeffective way to infer processes from spatial patterns. Ecology 90:46–56.Pacala SW (1997) Dynamics of plant communities. In Crawley MJ(ed). Plant Ecology. Oxford: Blackwell Science, 532–55.Palmer MW (1994) Variation in species richness: towards a unificationof hypotheses. Folia Geobot 29:511–30.Perry GLW, Enright NJ, Miller BP (2009) Nearest-neighbour interactionsin species-rich shrublands: the roles of abundance, spatial patternsand resources. Oikos 118:161–74.Perry GLW, Miller BP, Enright NJ (2006) A comparison of methods forthe statistical analysis of spatial point patterns in plant ecology. PlantEcol 187:59–82.Downloaded from http://jpe.oxfordjournals.org/ at Zhejiang University on September 7, 2012282

10 Luo et al. | Analyzing plant species association355Journal of Plant EcologyR Development Core Team (2009) R: A Language and Environment forStatistical Computing. Vienna, Austria: R Foundation for StatisticalComputing.Riginos C, Milton SJ, Wiegand T (2005) Context-dependent interactionsbetween adult shrubs and seedlings in a semi-arid shrubland.J Veg Sci 16:331–40.Ripley BD (1976) The second-order analysis of stationary point processes.J Appl Probab 13:255–66.Seidler TG, Plotkin JB (2006) Seed dispersal and spatial pattern in tropicaltrees. PLoS Biol 4:2132–7.Shen GC, Yu MJ, Hu XS, et al. (2009) Species-area relationshipsexplained by the joint effects of dispersal limitation and habitat heterogeneity.Ecology 90:3033–41.Stoyan D, Stoyan H (1994) Fractals, Random Shapes and Point Fields:Methods of Geometrical Statistics. Chichester, UK: John Wiley & Sons.Tilman D (1994) Competition and biodiversity in spatially structuredhabitat. Ecology 75:2–16.Vazquez DP, Melian CJ, Williams NM, et al. (2007) Species abundanceand asymmetric interaction strength in ecological networks. Oikos116:1120–7.Wang XG, Wiegand T, Hao ZQ, et al. (2010) Species associations in anold-growth temperate forest in north-eastern China. J Ecol 98:674–86.Wang W, Luo ZR, Zhou RF, et al. (2011) Habitat associations of woodyplant species in Baishanzu subtropical broad-leaved evergreen forest.Biodiv Sci 19:134–42.Wiegand T, Gunatilleke S, Gunatilleke N (2007) Species associationsin a heterogeneous Sri Lankan dipterocarp forest. Am Nat170:77–95.Wiegand T, Moloney KA (2004) Rings, circles, and null-models forpoint pattern analysis in ecology. Ecography 104:109–29.Wills C, Harms KE, Condit R, et al. (2006) Nonrandom processes maintaindiversity in tropical forest. Science 311:527–31.Wright J (2002) Plant diversity in tropical forests: a review of mechanismsof species coexistence. Oecologia 130:1–14.Xu M, Luo ZR, Yu MJ, et al. (2007) Floristic composition and communitystructure of mid-montane evergreen broad-leaved forest innorth slope of Baishanzu mountain. J Zhejiang Uni (Agric & LifeSci) 33:450–7.Zhu Y, Mi XC, Ren HB, et al. (2010) Density dependence is prevalent ina heterogeneous subtropical forest. Oikos 119:109–19.Downloaded from http://jpe.oxfordjournals.org/ at Zhejiang University on September 7, 2012283

Oikos 121: 1239–1250, 2012doi: 10.1111/j.1600-0706.2011.20079.x© 2012 The Authors. Oikos © 2012 Nordic Society OikosSubject Editor: Martin F. Quigley. Accepted 23 September 2011Density dependence is not very prevalent in a heterogeneoussubtropical forestZhengrong Luo, Xiangcheng Mi, Xiaorong Chen, Zhenlin Ye and Bingyang DingZ. R. Luo, College of Life Science, Zhejiang Univ., CN-310058 Hangzhou, PR China. – X. C. Mi, Inst. of Botany, Chinese Academy ofSciences, CN-100093 Beijing, PR China. – X. R. Chen and Z. L. Ye, Management of Baishanzu, Fengyangshan-Baishanzu National NatureReserve, CN-323800 Qingyuan, PR China. – B. Y. Ding (dby@wzu.edu.cn), The School of Life and Environmental Science, Wenzhou Univ.,CN-325035 Wenzhou, PR China.There is a growing body of evidence demonstrating that tree survival is influenced by negative density-dependence, butit is still controversial how the effect may vary with life-stage, and to what extent it plays a role in regulating tree survivalin heterogeneous subtropical forests. In this study, we investigated density-dependent effects on tree survival of six treespecies in a 5-ha subtropical forest in eastern China. The roughly 45 000 individuals in the forest were fully censused in2003 and 2008. For each of these species, we used an inhomogeneous pair-correlation function to quantify the changein spatial distribution for different size classes, and a case-control design to study seedling–adult associations in 2003.Autologistic regression was used to determine the influence of neighborhood factors on individual survival from 2003to 2008. We found that seedlings of five species were repulsed by distance to nearest conspecific adults in terms of theirsurvival, consistent with predictions of the Janzen–Connell mechanism. By contrast, only the least shade-tolerant Schimasuperba had a negative relationship with individual survival and conspecific distance-weighted basal area. This suggests thatthe Janzen–Connell effect is only prevalent at the early seedling stage or seed-to-seedling phase. The strength of clusteringsignificantly declined at sapling–pole and pole–adult transitions for Sycopsis sinensis and at seedling–sapling transition forCleyera pachyphylla. Correlations between individual survival and conspecific abundance for these species were consistentwith trends in the strength of clustering. These results suggest that density dependence plays a limited role in individualsurvival and species spatial structure beyond the early seedling stage (i.e. after true leaves growing) in this forest. In addition,this study indicates that including individuals from early life-stages and factoring out potential confounding factorssuch as habitat preference are important in studies that seek evidence for density dependence in forest trees.Population regulation is often thought to be a function ofnegative density-dependence at one or more life-historystages, because higher conspecific density can impair performancethrough greater competition for resources, easierdetection by predators and more susceptibility to pathogens(Janzen 1970, Connell 1971, Moeur 1997, Nishimura et al.2002, Wright 2002). More and more evidence has demonstratedthat species’ demographic rates, such as recruitment(Condit et al. 1992, Wills et al. 1997), growth (Uriarte et al.2004, Stoll and Newbery 2005) and survival (Hubbell et al.2001, Peters 2003, Comita and Hubbell 2009) are negativelyrelated to conspecific density in tropical forests. At thesame time, these trends have also been found to be prevalentin temperate (Kenkel 1988, He and Duncan 2000,HilleRisLambers et al. 2002) and subtropical forests (Chenet al. 2010, Zhu et al. 2010). However, inconsistent resultshave been observed in some tree species (Peterson andSquiers 1995) or in more heterogeneous forest stands(Dovciak et al. 2001, McDonald et al. 2003, Bin et al. 2011).The controversial evidence for negative density dependencemay be due to the confounding effects of life-historystage and species habitat preference. Many previous empiricalstudies have tested density dependence by presenting a staticmapping of saplings ( 1 cm dbh) and/or juvenile trees.These types of studies do not provide a direct test of densitydependent effects in earlier stages, because they ignorethe possibility of changes in spacing patterns (Cintra 1997).Density dependence is thought to occur most strongly duringearly life-stages (e.g. seedlings) when cohorts are densestand most susceptible to mortality (Howe and Smallwood1982, Norghauer et al. 2006). An important and unresolvedissue is whether density dependence acting later in the lifecycle is also important, or even overrides the density dependenceacting on seedlings (Freckleton and Lewis 2006). Toanswer this question, it is better to integrate informationfrom a variety of life-stages ranging from seedling to adult.Furthermore, negative density dependence is usuallyconfounded by the effect of habitat preference showing upas positive density dependence. In empirical observationalstudies, ecologists usually infer the importance of densitydependence from the spatial pattern of species, under thepremise that whatever coexistence mechanisms are operatingin the forest, they should leave a spatial signature that canbe detected (Hubbell et al. 2001). As a result, the results of2841239

density dependence do not always hold even though densitydependence is present, because of species habitat preference(Dovciak et al. 2001, McDonald et al. 2003). For example,if density dependence works strongly and constantly duringthe whole life-cycle of trees, density dependence should leavethe following signatures on space at the community level:1) trees survive better with fewer conspecific neighbors, 2)the clustering declines with time (self-thinning effect), and3) offspring are fewer within the vicinity of conspecific adults(repulsion effect) (Hubbell 1979, Moeur 1997). However,this prediction may be obscured by the effect of habitatpreference. A patchy distribution of limiting resources ofteninfluences demographic processes and the emerging speciespatterns (Getzin et al. 2008). In some favorable sites such astreefall gaps, both the abundance and survival of a habitatspecialist species can be high (Augspurger 1984). As a result,adults might be collected in sites most favorable for the species,whereas saplings and seedlings are widely dispersed(Condit et al. 2000). He and Duncan (2000) found that thedetection of density dependence could be altered by removingthe confounding effect of elevation on tree survival.Thus, it is necessary to account for species-specific habitatpreferences when interpreting the importance of negativedensity dependence in tree communities.At the same time, many previous applications of patternanalysis in the literature are largely based on static data (i.e.one census data) which does not always provide direct evidenceof density dependence. Indeed, it is the processes ofpopulation dynamics that are truly influenced by densitydependence(Condit et al. 1992). However, few studies lookat the change of point patterns over time. One reason forthe lack of long-term dynamics analyses may be that relativelong-term observational data is difficult to collect. Comparingpatterns among different life-stages (which reflect longtermoutcomes), together with neighborhood analysis (i.e.an individual-based approach to determine the influence ofseveral neighborhood factors on short-term plant survival),could remedy this deficit and provide clearer insights intocommunity dynamics. Surprisingly, few attempts have beenmade to combine static data and short-term dynamic data(e.g. individual survival over a short time interval) to detectdensity dependence in large stem-mapped plots.In this study, we sought to assess the role of density dependencein regulating the dominant tree species in subtropicalforest beyond the early seedling stage. To do this, we investigated1) how spatial distribution patterns change from seedlingsto adults, 2) what spatial relationships exist betweenadults and seedlings, and 3) how the density of conspecificneighbors influences survival of individual trees consideringthe potentially confounding effect of habitat preference. Weanalyzed two full sets of census data (with a 5-year interval)from a 5-ha stem-mapped plot in a subtropical evergreenbroad-leaved forest in eastern China. More advanced thanother similar analyses done in large plots where seedlingswere not included (Hubbell et al. 2001, Zhang et al. 2009),our study integrated seedlings from the whole plot into theanalyses.Corresponding to our question, we divided our study intothree complimentary parts. To begin with, we corrected forlarge-scale heterogeneity and compared the strength of clusteringamong size classes (part 1). This comparison exploredwhether and at which life-stage density dependence occurs(Getzin et al. 2008). Then, we used a case-control design toassess the spatial relationship between adults and seedlings(part 2). This allowed us to explore whether seedlings arerepulsed by distance to nearest conspecific adults in terms oftheir survival. Finally, we applied autologistic regression todetect the independent effects of conspecific neighborhooddensity or basal area on tree survival (part 3). Through theseanalyses, we integrated spatial and temporal information intothe study.Material and methodsStudy site, species and data collectionThe study site is an old-growth forest that forms part ofa larger evergreen broad-leaved forest in the Baishanzumountainous region (119°3′53″E, 27°40′54″N), which is partof the Fengyangshan-Baishanzu National Nature Reserve.In summer and early autumn (July to October), typhoonscarrying a large quantity of water can often influence thisregion. The regional climate is described as subtropical monsoonclimate, and is characterized by high relative humidity,substantial rainfall (2342 mm per year), hot summers (highof 32°C) and mild winters (low of 13°C). The study plot islocated at 1400–1600 m elevation on a north facing hillsidewith slopes ranging from 7.27° to 44.12°, and is influencedby habitat heterogeneity such as edaphic gaps (e.g. gravelsor runnels).A rectangular plot measuring 250 200 m (5 ha) wasestablished in an old-growth forest in July 2003. We measuredthe elevation at each intersection of a 20 20 m gridoverlaying the plot (last grid column is 10 20 m). All treeswith dbh (diameter at 1.3 m height) 1 cm and seedlings(with true leaves and dbh 1 cm) of tree species were tagged,stem-mapped and identified to species. The dbh was recordedfor all measurements larger than 1 cm, whereas for seedlingheights and basal diameters at the ground were measured.The roughly 45 000 individuals observed in the July 2003census were completely re-censused in August of 2008.We chose six dominant tree species (with 5% importantvalue) whose seedlings were fully-mapped (Table 1).These evergreen species prevail in the mid-subtropical evergreenbroad-leaved forest in east China (Wu 1980). Althoughthese species only accounted for 27.3% of the total individualswith dbh 1 cm, they took up 54.2% of the total basalarea in the plot. The diameter-distributions of Lithocarpusbrevicaudatus, Cyclobalanopsis multinervis, Schima superbaand Sycopsis sinensis were bimodal with abundant seedlingsand adult trees, while those of Cleyera pachyphylla andCyclobalanopsis stewardiana were reverse J-shape with a highnumber of small individuals. Mortalities of these specieswere highest among the smallest plants and declined withincreasing size class (Appendix 1, Fig. A1).Spatial point pattern analysisEmpirical studies have repeatedly shown that spatial pointpattern analysis is an effective approach for detecting negativedensity dependence when sophisticated analytical tools1240285

Table 1. Basic information for six dominant species in the Baishanzu evergreen-broad leaved forest. The important value was the sum of relative abundance (i.e. abundance of a species divided by totalabundance) and relative basal area (basal area of a species divided by total basal area).Importantvalue (%) Aridity-tolerance Shade-toleranceDispersal mode (dispersalunit: mm) Growth formNo. ofindividuals(and adults)Species code FamilyLithocarpus brevicaudatus Lt Fagaceae 4930 (1093) gravity or rodent (nut: 13.7) canopy 23.2 aridity-tolerant shade-tolerantCyclobalanopsis multinervis Mo Fagaceae 3500 (891) gravity or rodent (nut: 14.7) canopy 16.6 mid-tolerant shade-tolerantSchima superba Gt Theaceae 1988 (759) gravity or bird (seed: 4.5) canopy 14.5 aridity-tolerant mid-tolerantCleyera pachyphylla Tc Theaceae 2986 (428) gravity or bird (Berry: 5.0) midstory 11.5 aridity-tolerant shade-tolerantSycopsis sinensis Cf Hamamelidaceae 2935 (127) gravity or explosion (seed: 2) midstory 8.2 water-demanding shade-tolerantCyclobalanopsis stewardiana Bo Fagaceae 1098 (312) gravity or rodent (nut: 14.4) canopy 7.8 mid-tolerant shade-tolerantare used (He and Duncan 2000, Getzin et al. 2008, Zhuet al. 2010). We used the pair-correlation function g(r) toanalyze spatial patterns based on point-to-point distances(Stoyan and Stoyan 1994, Wiegand and Moloney 2004).The univariate pair-correlation function g(r) is relatedto the widely used Ripley’s K-function (Ripley 1976),where the g(r) function is the probability density functionof K-function. Due to its non-cumulative property, g(r) isrecommended for exploratory data analysis to identify specificscales of deviation from the null model (Wiegand andMoloney 2004). Similar to the K-function, g(r) can also beextended to describe point patterns with two types of points:the bivariate pair-correlation function g 12 (r) is the expecteddensity of points of patterns 2 at distance r from an arbitrarypoint of pattern 1, divided by intensity l 2 of pattern 2. Incase of inhomogeneous g(r) function, the intensity l is notconstant but varies with the location (x, y). In this study,we applied the Monte Carlo simulation to test significanceof departure from an underlying null model. To control forboth the effects of habitat preference and random mortality,we selected heterogeneous Poisson process and random-labelingas null model in the point pattern analyses (discussedin the following parts). Approximate 95% confidence intervalswere estimated from 5th highest and 5th lowest g(r)values obtained from 199 simulations. We used the packagespatstat in R statistical software (R Development Core Team2009) to perform spatial point pattern analysis.In this analysis, we divided individuals of a species alivein 2003 into four basic size classes: 1) ‘seedlings’ with dbh1 cm, 2) ‘saplings’ with dbh ranging from 1 to 5 cm, 3)‘poles’ with dbh ranging from 5 to 10 cm, and 4) ‘adults’with dbh 10 cm.Biological hypotheses and null modelsPoint pattern analysis 1: strength of clustering of differentsize classesIt is usually expected that clustering of trees declines withincreasing size class due to density-dependent self-thinning(Hubbell 1979, Moeur 1997). Therefore, we comparedthe spatial patterns for four different size classes mentionedabove using g(r) functions. To factor out the confoundingeffect of environmental heterogeneity, we used inhomogeneousg(r) functions based on the intensity function l(x, y)of the locations of individuals belonging to each size-class.We constructed a statistic g dif (r) g l (r) g s (r) to describe thedifference between pattern l and s, where g l (r) and g s (r) areg(r) functions for pattern l and s respectively. g dif (r) 0 indicatesthat pattern 1 is more aggregated than pattern s, whileg dif (r) 0 indicates the opposite. We implemented a nullmodel of heterogeneous Poisson process for each size class tocreate simulation g dif (r) values and the 95% envelope interval.It is the simplest null model that accounts for first-order effectssuch as habitat preference. In this null model, the intensity l isnot approximately constant but varies with the location (x, y).We used a nonparametric method to estimate intensity l(x, y)of the spatial distribution of a species which combines a movingwindow estimator with an Epanechnikov kernel (Stoyanand Stoyan 1994, Zhu et al. 2010). The fixed bandwidth of themoving window was 20 m which is little larger than scales atwhich local plant–plant interactions are likely to be present.2861241

Point pattern analysis 2: seedling–adult associationTo investigate the effects of density-dependent repulsionon the spatial association between offspring and conspecificadults, we implemented the random-labeling null modelwithin case-control design (Getzin et al. 2008). In the casecontroldesign, the control pattern is available to act as a surrogatefor the varying environmental factor. Assuming thatthe large-scale pattern of adult trees reflects the underlyingheterogeneity and intensity of seed rain, here we used adultclasses as control. The patterns of seedlings as cases werethen compared to the control patterns of adult trees. If thecases and the controls follow the same pattern, the randomlabelingnull model applies, where the cases represent arandom sub-sample of the joined pattern of controls andcases (Wiegand and Moloney 2004). From their definition,g(r) functions are invariant under random thinningand therefore we would expect g 12 (r) g 11 (r). We assessedthe significance of the difference g 12 (r) g 11 (r) in order toinvestigate departures from random labeling. With pattern1 adults control and pattern 2 cases seedlings, thepair-wise difference evaluates different biological effects. Avalue of g 12 (r) g 11 (r) » 0, for example, means that seedlingcases surround adult trees at scale r in the same way thatadults surround adults. Hence the distribution of seedlingsfollows the pattern of adults (Getzin et al. 2008). On theother hand, g 12 (r) g 11 (r) 0 indicates an overall establishmentlimitation due to abiotic or biotic constraints.Neighborhood analysisMany previous tests of the neighborhood effect on treedynamics are questionable, because these tests did notaccount for potential spatial autocorrelation in mortality(Leigh et al. 2004). In this study, we used autologistic regressionto address this issue, which models the spatial autocorrelationin survival to test the effects of conspecific densityand basal area (weighted by distance) on individual tree andseedling survival. The term measuring spatial autocorrelationin survival in the autologistic model describes the potentialresult of any number of biotic and abiotic factors that maycause survival to be patchy, such as light, nutrients, and moistureheterogeneity (Hubbell et al. 2001, Queenborough et al.2007). For every individual (including seedlings) of the sixspecies alive in 2003, we defined neighborhoods with radiiof 5, 10, 15 and 20 m from the focal tree. Plants with neighborhoodsthat exceeded the confines of the 5-ha plot wereonly as neighbors of other plants, not as focal plants. Withineach circle, we calculated the following independent neighborhoodvariables: 1) the number of neighbors that were ofthe same species as the focal plant (conspecific abundance),and 2) the effective basal area of these neighbors (conspecificbasal area). Effective basal area was quantified as the sum ofbasal area of each conspecific tree 1 cm dbh (within a givenradius) divided by the distance of that tree from the focalplant (Comita and Hubbell 2009). Effective basal area reflectsthe competition intensity from an individual’s neighbors. Tocontrol for the effects of topographic variation on plant survival,we used the elevation, convexity and slope of the 5 5 mgrid where the focal individual was located as predictors in anautologistic regression. The definition of elevation, convexityand slope followed Valencia et al. (2004).To account for the correlation between a plant’s competitiveability and its size, we incorporated the size of each focalindividual (i.e. basal area of focal trees or height of focal seedling)into the autologistic regression. The response variablewas the survival of the focal plant in August 2008, which wasassigned a value of one (alive) or zero (dead).For each focal species, autologistic regression was appliedto two life-history stages: seedling (with true leaves and dbh 1 cm) and tree (dbh 1 cm). The model was fitted witha Monte Carlo Markov chain simulation. For detailed discussionabout this simulation, see Hubbell et al. (2001).We used the R2WinBUGS package in the R software andWinBUGS software to execute the simulations.ResultsPoint pattern analysis 1: strength of clustering ofdifferent size classesFor the six species, there was no systematic decline in thestrength of aggregation from seedlings to adults. The changesof spatial pattern along with size classes were disordered. Thestrengths of clustering without the confounding effect ofenvironmental heterogeneity significantly declined at thesapling–pole and pole–adult transitions for Sycopsis sinensis,but declined only at the seedling–sapling transition forCleyera pachyphylla (Table 2). However, there was no significantincrease in the strength of clustering at any stagetransition of any species. Most of the shifts in strength ofclustering were consistent with the null model of heterogeneousPoisson processes.Point pattern analysis 2: seedling–adult associationThis analysis showed that the distributions of seedlingsdeparted from adults for five of the six species we studied(except S. sinensis) in our heterogeneous plot. The test statisticg 12 (r) – g 11 (r) showed significant departures from randomlabeling between seedlings and adults, especially atsmall distances (Fig. 1). These results indicate that patchesof seedlings were significantly segregated from adults, i.e.many adults had few seedlings in their immediate vicinity.For Lithocarpus brevicaudatus and C. pachyphylla, this segregationpersisted even to the scale of 20 m. For S. sinensis,Table 2. Shifts in pattern of different size-classes of the six species.The first census data (2003) was used for these analyses. standsfor more clumped, stands for more regular, and r for non significantshift.Species Seedling–sapling Sapling–pole Pole–adultLithocarpusr r rbrevicaudatusCyclobalanopsisr r rmultinervisSchima superba r r rCleyera pachyphylla r rSycopsis sinensis r Cyclobalanopsisstewardianar r r1242287

1 (a) Lt0.8 (b) Mo0.50.50–0.50.2–0.1–0.4–1–0.71.5 (c) Gt6.5 (d) Tcg 12 (r)–g 11 (r)10.5041.5–0.5–1–110 (e) Cf–3.52 (f) Bo1.5510.500–0.5–5–1–100 5 10 15 20–1.5–20 5 10 15 20Distance r (m)Figure 1. Seedling–adult association analyses of the six species with random labeling case-control design. The pattern of adult trees of eachspecies serves as the control pattern, and seedlings as case. If the seedlings follow the adult pattern, the test statistic g 12 (r) g 11 (r) » 0.Grey solid lines show approximate 95% simulation envelopes and the solid lines with points show observed values of test statistics. Thesix species are L. brevicaudatus (Lt), C. multinervis (Mo), S. superba (Gt), S. sinensis (Cf), C. pachyphylla (Tc), and C. stewardiana (Bo).however, seedlings surround conspecific adults in the sameway that adults surround adults.Neighborhood effectsThe survival of seedlings was strongly spatially autocorrelatedfor each species, but its significance declined rapidly withincreasing distance. At distances 15 m from the focal seedling,the spatial autocorrelation did not differ significantlyfrom zero or even become negative. For most species, survivalwas not spatially structured beyond this distance. In contrast,for trees, only the survival of L. brevicaudatus, and S. sinensiswere significantly spatially autocorrelated, while that of theother four species had no spatial autocorrelation (Table 3).As expected, individual survival was strongly positively correlatedto its height or basal area for all the species (Table 3),where focal plants with larger size survived better than smallerones. This pattern did not change significantly with distance.2881243

We detected some relationships between individual survivaland micro-topography. There was a strong general effectof elevation on seedling survival for L. brevicaudatus, butthe effect of elevation on tree survival was only significantat 10 m. At the same time, a positive relationship betweenelevation and C. pachyphylla seedling survival was detected.Our results also showed that tree survival of S. sinensis wassignificantly correlated with convexity at the 15 m scale,while seedling survival of the species was positively correlatedwith slope.Significant relationships between survival of these speciesand conspecific abundance/basal area were few. Conspecificabundance had significant negative effects on the survival ofS. sinensis trees and C. pachyphylla seedlings, and the effectwas still significant at greater distances for C. pachyphylla(Table 3). Our results also showed that individual survivalof S. superba was negatively related to conspecific basalarea within 10 m distance. In contrast, for L. brevicaudatus,Table 3. Results of autologistic regression modeling the survival oftarget groups with 5 predictor variables at four neighborhood size(r 5, 10, 15, 20 m). The predictor variables are basal area (BA, fortrees) or height (FH, for seedlings) of focal plant, conspecific abundance(CA), conspecific basal area (CB), elevation (ELEV), convexity(CONV), slope (SLOPE), and spatial autocorrelation of survival (SP).A predictor is not shown if it is not significant at any radii. Thesigns , and indicate significant positive correlations atp-value 0.05, 0.01 and 0.001 respectively. Similarly, indicatesnegative correlations. The zero sign (0) indicates non-significantrelationships between survival and these variables. The target speciesare L. brevicaudatus (Lt), C. multinervis (Mo), S. superba (Gt),C. pachyphylla (Tc), S. sinensis (Cf), and C. stewardiana (Bo)Target group Predictor 5 m 10 m 15 m 20 mLt seedlings ELEV FH CB 0 0SP Lt tress ELEV 0 0 0BA CA 0SP 0 0 0Mo seedlings FH CB 0 0SP 0Mo trees BA Gt seedlings FH CB 0 0 0SP 0Gt trees BA CB 0 0Tc seedlings ELEV 0 FH CA SP 0 0 0Tc trees BA Cf seedlings SLOPE 0 FH SP 0 Cf trees CONV 0 0 0BA CA 0 0 0SP 0Bo seedlings FH SP 0 0 Bo trees BA individual survival increased with increasing conspecificabundance or basal area. For other groups, individual survivalwas not significantly affected by conspecific abundanceor by conspecific basal area.DiscussionThe effect of negative density dependence on plant survivaland regulation of species spatial structure in temperate andtropical forest has been debated for decades (Janzen 1970,Connell 1971, Kenkel 1988, Hubbell 1979, Silva Matoset al. 1999, He and Duncan 2000, Peters 2003, Zhanget al. 2009). Recently, it was found to affect subtropicalplant populations, at least in abundant species (Chen et al.2010, Zhu et al. 2010). In our study, the inclusion of seedlingsand the disentanglement of the confounding effect ofhabitat preference gave us a clear and unbiased insight intothe importance of density dependence in regulating speciesspatial structures. By integrating short-term dynamic data(i.e. individual survival) and static data (i.e. trends in spatialpatterns over size-classes of the six populations alive in 2003)into spatial and demographic pattern analyses, our studydemonstrated that density dependence in this forest maynot be prevalent. Density-dependence is only important forearly life-history stages (e.g. seedling establishment or earlyseedling). Later life stages rarely show density-dependence,and even if present, it is not strong enough to eliminate thestrong clustering pattern formed at the seedling establishmentphase.Self-thinning effect of density dependence(part 1 and 3)The self-thinning effect of density dependence has beenshown to be prevalent even among trees larger than 10 cmdbh in tropical and temperate forests (Kenkel et al. 1997,Peters 2003). However, in our study, little evidence for thiseffect was detected. Prior to considering the effects of spatialautocorrelation, our results from neighborhood analysesshowed that conspecific abundance had significant negativeeffects on the survival of S. sinensis trees and C. pachyphyllaseedlings only. The nonrandom patterns of mortality, with ahigher mortality risk within neighborhoods of higher conspecifics,are compatible with the outcome expected fromthe self-thinning effect of density dependence. Interestingly,these two species were both highly aggregated (Luoet al. 2009). This finding is consistent with results fromthe Gutianshan subtropical forest, where Zhu et al. (2010)found that the occurrence of negative density dependence ispositively associated with the strength of conspecific aggregationat a local scale. This phenomenon suggests that highaggregation usually strengthens intraspecific competition(Zhang et al. 2009), because sessile terrestrial plants usuallyinteract mainly with their nearest neighbors, and not withthe population as a whole. Direct competition for limitingphysical resources, such as light and soil nutrients, may beimportant in influencing individual mortality of these twospecies (Peters 2003). However, natural enemies such aspathogens and insects, in theory, could also contribute todensity dependence.1244289

Neighborhood analysis alone does not provide solidevidence for lack of density dependence, because the effectmay be underestimated due to using short-term time seriesdata only. Sometimes density dependence may not bestrong enough to cause substantial mortality, but insteadcause limited growth over short time-scales (Getzin et al.2006). However, negative effects on growth and survivalgenerally reinforce one another over longer time intervals(Wright 2002).The cumulative effects of density dependence may have astrong influence on species distribution patterns. Comparisonof the strength of clustering among different size-classesgives us much information about the effect of density dependenceon individual survival, because the individuals fromlarge size-classes may suffer density dependence for a longtime. If survival is lower for plants that are more crowded,then as the cohort ages the strength of clustering shoulddecrease, as individuals within high density patches areselectively removed. After the habitat preference effect wasaccounted for, the strengths of clustering were found to declineat one or two transitions for S. sinensis and C. pachyphyllaonly. This is consistent with the finding from neighborhoodanalysis: there is lack of signature of negative density dependencefor the other four species and other transition stagesof these two species in this forest region. Similar results havebeen reported in many other studies (Zhang et al. 2009, Binet al. 2011). Furthermore, even the cumulative effects presentin S. sinensis and C. pachyphylla are not strong enough toentirely eliminate the strong spatial clustering patterns whichinitially formed in seed dispersal and seedling establishmentstages (Luo et al. 2009). Many researchers have suggestedthat density dependent plant survival has the greatest impacton seedlings, given their susceptibility to natural enemies andabiotic stresses (Connell 1971, Hyatt et al. 2003). However,our study showed that there is no difference in the prevalenceof density dependence between seedling and later life stages,notwithstanding the absence of a statistical test. Therefore,we conclude that the self-thinning effect of density dependenceplays a limited role in plant survival and species spatialstructure beyond the early seedling stage (i.e. after true leavesgrowing). However, our result differs from the evidence fromthe Gutianshan subtropical forest, where, by comparing thestrength of clustering of different size-classes without statisticaltests (i.e. if the observed value of the statistic d(r) they used 0, density-dependent thinning was considered taking placein the species), density dependence has been demonstrated asa prevalent mechanism for regulating the population spatialstructure of the majority of species (Zhu et al. 2010).Repulsion effects of density dependence(part 2 and 3)The repulsion effect of density dependence, where offspringare inhibited in the vicinity of their conspecific adults, ismost notable during early life-stages such as the seedlingestablishment phase. Our results show that densities of seedlingsare lower within the vicinity of their conspecific adultsfor five species; a large number of seedlings established infavorable sites distant from conspecific adults. This scenariois consistent with the Janzen–Connell hypothesis, whichdescribes that host-specific pests and/or pathogens reducerecruitment near conspecific adults (Janzen 1970, Connell1971). This suggests that a large proportion of seeds or smallseedlings dispersed beneath parents are killed because of highconspecific sibling density or proximity to species pathogens.On the other hand, seeds escaping to a distance furtherfrom the parent tree have more chance of arriving in favorablesites (Hood et al. 2004). Howe and Smallwood (1982)hypothesized that directed dispersal to habitats favoringseedling growth also increased the chances of recruitment.The propagules of these five species can be dispersed by animals(birds and/or rodents), which are assumed to be betterdispersed than purely wind- or explosively-dispersed species(Condit et al. 2000). This may have given them a temporaladvantage in the space occupation of newly formed tree-fallgaps. Early stage seedlings dispersed to light gaps are lesslikely to suffer from damping-off disease than those presentin the understory (Augspurger 1984). Our finding is consistentwith a meta-analysis result that seedlings benefit fromincreased distance from conspecific adults at the communitylevel (Hyatt et al. 2003). However, it is difficult to determinewhether the repulsion effect was taking place at pre- orpost-seedling establishment with present data. We can notexclude the possibility that the repulsion effect was presentat seed stage.Despite these results, these effects do not persist into laterlife-stages. The relationship between individual survival andconspecific effective basal area describes the performance ofindividuals when they are near large trees, because the variableof effective basal area integrates the effects of size anddistance. In the neighborhood analysis, we found that onlyfor plants of S. superba, individual survival was negativelyrelated to conspecific effective basal area within a small distance.This result is consistent with the findings from BCIthat shade-intolerant species are more vulnerable to competitionfrom neighbors, and are more sensitive to conspecificsin their neighborhood (Hubbell et al. 2001, Comita andHubbell 2009). As a mid-shade-tolerant species, seedlingsor saplings of S. superba usually grow worse when under acanopy tree. For the other five species, individual survival isunrelated to whether or not it lives near conspecific adults.This suggests that the repulsion effect of density dependenceis absent at later life stages. This would leave a profoundinfluence on the distribution pattern of adults. The repulsioneffect of density dependence at early stages repulsesmost seedlings from establishing near conspecific adults, butit is usually not strong enough to completely remove seedsor seedlings dispersed beneath parent trees (Hubbell 1980).Consequently, the absence of the repulsion effect at laterstages gives seedlings an opportunity to grow under theirparent tree, and regular distribution of adults would not belikely to emerge. This interpretation is supported by the factthat patterns of adults of all these species in the plot werestrongly aggregated (Luo et al. 2009).Although the repulsion effect was only present at veryearly stages, it likely plays an important role in populationregulation and species coexistence. Silva Matos et al. (1999)demonstrated that negative density dependence acting onseedling recruitment overrode potential density dependenteffects later in the life cycle. They argued that if there existsa difference between maximum survival near adult trees andsurvival at further distances, then density dependence may2901245

e stabilizing at level of the population. Harms et al. (2000)have documented that even a partial reduction of recruitmentwhere seed density is high can increase alpha diversity.Density dependence and species’habitat preferenceSpecies’ habitat preferences have an important influenceon species’ spatial patterns and thus on their coexistence.Habitat preferences are one of the primary mechanisms thatpromote species coexistence in subtropical and tropical forests(Shen et al. 2009). Our results showed that most of thespecies studied did not exhibit significant shifts in clusteringof different size-classes under the null model of heterogeneousPoisson processes. At the same time, the survival ofseedlings was found to be strongly spatially autocorrelatedfor each species, although only half the groups of individualsshowed significant relationships between survival and microtopography.Together, this evidence suggested that specieshabitat preference plays an important role in this forest.The effect of strong habitat preference can sometimes overwhelmthe effect of density dependence on spatial patternsand introduce a bias in the detection of density dependence(Chen et al. 2010, Zhu et al. 2010). Generally, an increasein aggregation with heterogeneity may elevate risk, wheresmall individuals are more likely to be detected by naturalenemies or suppressed by other conspecifics. However, thiseffect may be outweighed by the initial benefit of favorablehabitat association (Wright 2002, North and Ovaskainen2007), especially when individuals grow into stages that arenot as sensitive to natural enemies or intraspecific competition.In this study, we considered the potential confoundingfactor of species’ habitat preference when seeking evidencefor density dependence. Little evidence of density dependencein this forest does not likely result from the stronginterference of species’ habitat preference.The importance of including early life-stages(part 1 and 2)The inclusion of seedlings when using spatial point patternanalysis to detect signatures of negative density dependenceis important. Most previous studies using the Forest DynamicsPlot data, which include only individuals greater than 1cm dbh, provide no direct evidence for processes at smallersize classes (Muller-Landau et al. 2004, Freckleton and Lewis2006). It is possible that ecologically important densitydependenteffects have occurred by the time the plants recruitinto these size classes. Sometimes, the microsite conditionsthat are suitable for sustained sapling and juvenile growthmay differ from optimal conditions for initial establishment(Lai et al. 2009). In this case, using different life-stages toinfer density dependence may yield different results. In thisstudy, we found that five species exhibited spatial signaturesof density dependence at an early stage by seedling–adult association analysis. However, if we use sapling–adultassociation or pole–adult instead, many fewer species willshow evidence of density dependence (Appendix 1 Fig.A2, A3). This suggests that the signature of density dependenceon the seedling stage may not necessarily be maintainedat later size classes. Censuses that do not considerseedlings may provide an incomplete picture of the extentand importance of density dependence (Freckleton andLewis 2006). Therefore, without specific knowledge of spatialdistribution at smaller size classes, inferences from the indirectevidence from patterns of larger size classes alone (e.g.large saplings or juveniles) should be drawn with caution.Acknowledgements – We thank Drs. Fangliang He and GuochunShen for their critical comments which greatly improved our study.We also appreciate many people from Zhejiang Univ., WenzhouUniv. and the Management of Baishanzu for their contributions tothe establishment and re-census of this 5-ha permanent forest plot.We are grateful to Lauren Barry for English editing. The study wassupported by the Appropriative Foundation of Ecology and EnvironmentProtect of Zhejiang Province and the China NationalProgram for R & D Infrastructure and Facility Development.ReferencesAugspurger, C. K. 1984. Seedling survival of tropical tree species:interactions of dispersal distance, light-gaps and pathogens.– Ecology 65: 1705–1712.Bin, Y. et al. 2011. Tree mortality and recruitment in a subtropicalbroadleaved monsoon forest in south China. – J. Trop. For.Sci. 23: 57–66.Chen, L. et al. 2010. Community-level consequences of densitydependence and habitat association in a subtropical broadleavedforest. – Ecol. Lett. 13: 695–704.Cintra, R. 1997. A test of the Janzen–Connell model with twocommon tree species in Amazonian forest. – J. Trop. Ecol. 13:641–658.Comita, L. S. and Hubbell, S. P. 2009. Local neighborhood andspecies’ shade tolerance influence survival in a diverse seedlingbank. – Ecology 90: 328–334.Condit, R. et al. 1992. Recruitment near conspecific adults andthe maintenance of tree and shrub diversity in a neotropicalforest. – Am. Nat. 140: 261–286.Condit, R. et al. 2000. Spatial patterns in the distribution oftropical tree species. – Science 288: 1414–1418.Connell, J. H. 1971. On the role of natural enemies in preventingcompetitive exclusion in some marine animals and in rainforest trees. – In: den Boer, P. J. and Gradwell, G. R. (eds),Dynamics of populations. Centre for Agricultural Publicationand Documentation, pp. 298–312.Dovciak, M. et al. 2001. Discordance in spatial patterns of whitepine (Pinus strobus) size-classes in a patchy near-Boreal forest.– J. Ecol. 89: 280–291.Freckleton, R. P. and Lewis, O. T. 2006. Pathogens, densitydependence and the coexistence of tropical trees. – Proc. R.Soc. B 273: 2909–2916.Getzin, S. et al. 2006. Spatial patterns and competition of treespecies in a Douglas-fir chronosequence on Vancouver Island.– Ecography 29: 671–682.Getzin, S. et al. 2008. Heterogeneity influences spatial patternsand demographics in forest stands. – J. Ecol. 96: 807–820.Harms, K. E. et al. 2000. Pervasive density-dependent recruitmentenhances seedling diversity in a tropical forest. – Nature 404:493–495.HilleRisLambers, J. et al. 2002. Density-dependent mortality and thelatitudinal gradient in species diversity. – Nature 417: 732–735.He, F. and Duncan, R. P. 2000. Density-dependent effects on treesurvival in an old-growth Douglas fir forest. – J. Ecol. 88:676–688.Hood, L. A. et al. 2004. The influence of spatial patterns ofdamping-off disease and arbuscular mycorrhizal colonization1246291

on tree seedling establishment in Ghanaian tropical forestsoil. – J. Ecol. 92: 816–823.Howe, H. F. and Smallwood, J. 1982. Ecology of seed dispersal.– Annu. Rev. Ecol. Syst. 13: 201–228.Hubbell, S. P. 1979. Tree dispersion, abundance, and diversity ina tropical dry forest. – Science 203: 1299–1309.Hubbell, S. P. 1980. Seed predation and the coexistence of treespecies in tropical forests. – Oikos 25: 214–229.Hubbell, S. P. et al. 2001. Local neighborhood effects on long-termsurvival of individual trees in a neotropical forest. – Ecol. Res.16: 859–875.Hyatt, L. A. et al. 2003. The distance dependence prediction ofthe Janzen–Connell hypothesis: a meta-analysis. – Oikos 103:590–602.Janzen, D. H. 1970. Herbivores and the number of tree species intropical forests. – Am. Nat. 104: 501–528.Kenkel, N. C. 1988. Pattern of self-thinning in jack pine:testing the random mortality hypothesis. – Ecology 69: 1017–1024.Kenkel, N. C. et al. 1997. Along-term study of Pinus banksianapopulation dynamics. – J. Veg. Sci. 8: 241–254.Lai, J. S. et al. 2009. Species–habitat associations change in a subtropicalforest of China. – J. Veg. Sci. 20: 415–423.Leigh, E. G. et al. 2004. Why do some tropical forests have somany species of trees? – Biotropica 36: 447–473.Luo, Z. R. et al. 2009. Distribution patterns of tree species in anevergreen broadleaved forest in eastern China. – Front. Biol.China 4: 531–538.McDonald, R. I. et al. 2003. Spatial pattern of Quercus regenerationlimitation and Acer rubrum invasion in a Piedmont forest.– J. Veg. Sci. 14: 441–450.Moeur, M. 1997. Spatial models of competition and gap dynamicsin old-growth Tsuga heterophylla – Thuja plicata forest. – For.Ecol. Manage. 94: 175–186.Muller-Landau, H. C. et al. 2004. Seed dispersal and densitydependentseed and seedling survival in Trichilia tuberculataand Miconia argentea. – In: Losos, E. C. et al. (eds), Tropicalforest diversity and dynamism: findings from a large-scale plotnetwork. Univ. of Chicago Press, pp. 340–362.Nishimura, N. et al. 2002. Tree competition and species coexistencein a warm-temperate old-growth evergreen broad-leavedforest in Japan. – Plant Ecol. 164: 235–248.Norghauer, J. et al. 2006. An experimental test of densityanddistant-dependent recruitment of mahogany (Swieteniamacrophylla) in southeastern Amazonia. – Oecologia 148:437–446.North, A. and Ovaskainen, O. 2007. Interactions between dispersal,competition and landscape heterogeneity. – Oikos 116:1106–1119.Peters, H. A. 2003. Neighbour-regulated mortality: the influenceof positive and negative density dependence on treepopulations in species-rich tropical forests. – Ecol. Lett. 6:757–765.Peterson, C. J. and Squiers, E. R. 1995. An unexpected change inspatial across 10 years in an aspen–white-pine forest. – For.Sci. 50: 299–311.Queenborough, S. A. et al. 2007. Neighborhood and communityinteractions determine the spatial pattern of tropical tree seedlingsurvival. – Ecology 88: 2248–2258.Ripley, B. D. 1976. The second-order analysis of stationary pointprocesses. – J. Appl. Probab. 13: 255–266.Shen, G. C. et al. 2009. Species–area relationships explained by thejoint effects of dispersal limitation and habitat heterogeneity.– Ecology 90: 3033–3041.Silva Matos, D. M. et al. 1999. The role of density dependence inthe population dynamics of a tropical palm. – Ecology 80:2635–2650.Stoll, P. and Newbery, D. M. 2005. Evidence of species-specificneighborhood effects in the Dipterocarpaceae of a Borneanrain forest. – Ecology 86: 3048–3062.Stoyan, D. and Stoyan, H. 1994. Fractals, random shapes andpoint fields: methods of geometrical statistics. – Wiley.Uriarte, M. et al. 2004. A neighborhood analysis of tree growthand survival in a harricane-driven tropical forest. – Ecology74: 591–614.Valencia, R. et al. 2004. Tree species distributions and local habitatvariation in the Amazon: large forest plot in eastern Ecuador.– J. Ecol. 92: 214–229.Wiegand, T. and Moloney, K. A. 2004. Rings, circles, and nullmodelsfor point pattern analysis in ecology. – Ecography 104:109–229.Wills, C. et al. 1997. Strong density- and diversity-related effectshelp to maintain tree species diversity in a neotropical forest.– Proc. Natl Acad. Sci. USA 94: 1252–1257.Wright, S. J. 2002. Plant diversity in tropical forests: a review ofmechanisms of species coexistence. – Oecologia 130: 1–14.Wu, Z. Y. 1980. Vegetation of China. – Science Press.Zhang, J. et al. 2009. Density dependence on tree survival in anold-growth temperate forest in northeastern China. – Ann.For. Sci. 66: 51–59.Zhu, Y. et al. 2010. Density dependence is prevalent in a heterogeneoussubtropical forest. – Oikos 119: 109–119.2921247

Appendix 11500(a) Lt800(b) Mo1000Dead treesLive trees600Dead treesLive trees4005002000(0.3) (0.9) 1 8 15 22 29 60.50(0.5) (1) 1 5 10 15 20 25 46.7600(c) Gt800(d) TcNumber of individuals400200Dead treesLive trees600400200Dead treesLive trees0(0.5) (1) 1 6 11 16 21 26 47.90(0.5) 1 4 7 10 13 16 45.1900(e) Cf400(f) Bo600Dead treesLive trees300Dead treesLive trees2003001000(1) 1 2.5 4 5.5 7 8.5 310Size-class (cm)(0.5) 1 8 15 22 29 47.2Figure A1. Frequency distributions of size classes of Lithocarpus brevicaudatus (Lt), Cyclobalanopsis multinervis (Mo), Schima superba (Gt),Sycopsis sinensis (Cf), Cleyera pachyphylla (Tc) and Cyclobalanopsis stewardiana (Bo). Values with and without brackets on the horizontal axisindicate diameter on the ground and dbh, respectively.1248293

0.8 (a) Lt0.6 (b) Mo0.20.1–0.4–0.4–1.01.5 (c) Gt–0.95 (d) Tc0.73g12(r)–g11(r)–0.1–0.91–1–1.7–31.5 (f) Bo1.010.50.0–4–0.5–1.00 5 10 15 20–1.5–90 5 10 15 20–2.06 (e) CfDistance r (m)Figure A2. Sapling–adult association analyses of the six species with random labeling case-control design. The pattern of adult trees for eachspecies serves as control pattern and seedlings as case. If saplings follow the adult pattern, the test statistic g 12 (r) – g 11 (r) ≈ 0. The shade areasshow approximate 95% simulation envelopes and the solid lines show observed values of test statistics. Species names correspond to thecodes in Fig. A1.2941249

0.7 (a) Lt0.6 (b) Mo0.30.1–0.1–0.5–0.4–0.90.9 (c) Gt–0.92.5 (d) Tc0.40.5g12(r)–g11(r)–0.1–0.6–1.1–1.5–3.52 (f) Bo1.510.50–0.5–1.5–1–22.5 (e) Cf0 5 10 15 20–2.50 5 10 15 20Distance r (m)Figure A3. Pole–adult association analyses of the six species with random labeling case-control design: conventions as in Fig. A2.1250295

Oikos 000: 001–008, 2012doi: 10.1111/j.1600-0706.2012.20483.x© 2012 The Authors. Oikos © 2012 Nordic Society OikosSubject Editor: Matthijs Vos. Accepted 27 April 2012Genetic groups in the common plant species Castanopsis chinensisand their associations with topographic habitatsZheng-Feng Wang, Ju-Yu Lian, Guo-Min Huang, Wan-Hui Ye, Hong-Lin Cao and Zhang-Ming WangZ.-F. Wang, J.-Y. Lian, G.-M. Huang, W.-H. Ye (why@scbg.ac.cn), H.-L. Cao and Z.-M. Wang, Key Laboratory of Plant ResourcesConservation and Sustainable Utilization, South China Botanical Garden, Chinese Academy of Sciences, Guangzhou 510650, PR China,and Guangdong Key Laboratory of Digital Botanical Garden, South China Botanical Garden, Chinese Academy of Sciences, Guangzhou510650, PR China.In general, even within a local area, many common plant species are found in different types of environment. We proposethat if the association of a common plant species with different types of environment is investigated, by analysing allindividuals in a given population as a single entity, the results might be misleading or incomplete owing to intraspecificvariation. To test this hypothesis, we used molecular markers to classify mature Castanopsis chinensis individuals witha diameter at breast height 40 cm into different genetic groups and analysed the associations of these groups withtopographic features and habitats within a 20-ha Dinghushan forest plot, South China. Our results indicated that thedifferent groups had different topographical associations, and that the spatial distributions and genetic structures ofindividuals varied among the groups. Therefore, if significant genetic structure exists in the population of a commonspecies within a community, to understand the relationship between the spatial distributions of individuals in thepopulation and the environment, it is necessary to classify the individuals into genetic groups and analyse the data forthese groups, rather than for a combined group of all individuals.Niche and neutral processes are two major theories thatexplain the spatial distributions of species in plant communities(Leibold and McPeek 2006, Kraft et al. 2008,Legendre et al. 2009). Niche processes can be identifiedby analysing the spatial association of species with certainenvironmental descriptors (Plotkin et al. 2000). However,within a local area, the individuals of a common plant speciescan occupy areas with various different environmentalconditions (Gram and Sork 2001, Volis et al. 2004, Noguchiet al. 2007, Queenborough et al. 2007, Nazre et al. 2009),and some conditions might even be considered unfavourable(Miyamoto et al. 2003, Comita et al. 2007). Consequently,if the differences among individuals within the same speciesin the same general area are not examined properly, associationsbetween the spatial distributions of a species and typesof environment might not be identified accurately. This lackof identification might arise because, for example, differentindividuals of a species might occupy different types of environmentsuch that there is an even distribution of the speciesoverall and thus we might conclude that neutral processes,rather than niche processes, govern the spatial distribution ofthe species. Alternatively, we might conclude that a species isassociated significantly with one major environmental typebut not with others, because only a small number of individualsoccupy the latter. However, this distribution mightbe due to the association of different groups of individualswithin the species with different environmental types. Toinvestigate this possibility, we investigated whether there aredifferent genetic groups in Castanopsis chinensis (Fagaceae)that are associated with different topographic features in atopographically complex 20-ha plot in a lower subtropicalevergreen forest in Dinghushan (DHS plot), South China(Wang et al. 2009).Castanopsis chinensis is one of the most important foundationtree species in lower subtropical China. It is a pioneerspecies and its establishment provides suitable microenvironmentsfor late successional species in the community(Ren et al. 2008). In the DHS plot, mature C. chinensisindividuals with a diameter at breast height (DBH) 40 cm(Li et al. 2008) occupy many topographically differenthabitats, from low to high elevations and from ridges tolowland areas (Fig. 1A). In the present study, we intentionallysampled these mature individuals because we consideredthat, given their survival to maturity, their present growingenvironments must be suitable. In other words, sufficienttime might have passed since the species became establishedin the area for the particular microenvironments to selectthe fittest individuals. Consequently, it might be possibleto identify associations between genetic groups withinC. chinensis and their preferred types of environment. In2961

Figure 1. (A) The spatial distribution of the 212 C. chinensis individuals with DBH 40 cm in the 20-ha (400 500 m) DHS plot. Thecolours of blue, red and green indicate genetic groups 1 and 2 and the individuals with an intermediate status between the groups, respectively.The sizes of the tree symbols are related to different values of DBH. Panels (B)–(D) show the spatial distributions of G1, G2, and theindividuals with an intermediate status between the groups in relation to five topographic habitats. Contour lines represent changes inelevation of 20 m.the study reported herein, we used topography to characterizeenvironmental types. This variable was chosen becauseit is the most important environmental factor in determiningthe spatial variability of microclimate and soil propertiesand can serve as a proxy for environmental properties(Hubbell and Foster 1986, Svenning 1999, Comita et al.2007, Noguchi et al. 2007, Nazre et al. 2009).Material and methodsStudy speciesCastanopsis chinensis is an evergreen tree that belongs tothe Fagaceae family. It is distributed from South China toVietnam, below 1500 m a.s.l. (Wu 2009). As a pioneercanopy tree, it can grow up to 20 m in height and usuallyplays a keystone role in the ecosystems in which it isfound (Ma 1999). It is monoecious with unisexual staminateand pistillate flowers on the same plant. Its flowers arewind pollinated. Its seeds are oval and are dispersed bymammals, such as rodents, pigs, cats, and birds (Peng andXu 2005). Castanopsis chinensis is shade intolerant, and itrequires high levels of light for seed germination (Du andHuang 2008).Study siteOur study was conducted in the 20-ha DHS plot, whichlies just to the south of the Tropic of Cancer in the subtropicalpart of South China. The plot is in a 1155-ha nationalnature reserve (112°30¢39²–112°33¢41²E, 23°09¢21²–23°11¢30²N) that was established in 1956 to preserve themonsoon evergreen broadleaved forest there. The region ischaracterized by a south subtropical monsoon climate, witha mean annual temperature of 20.9°C and precipitation of1929 mm.The DHS plot has a complex topography. It includesat least three ridges and four valleys, with an elevation thatranges from 240 to 470 m (Fig. 1A). The plot was surveyedin 2005 using the standard methods of the Center for TropicalForest Science (Condit 1998). All free-standing woodyplants with DBH 1 cm were mapped, tagged and identifiedto the species level, and their DBH was recorded (Yeet al. 2008, Wang et al. 2009).According to the survey data from 2005, the DHS plotcontained a total of 2311 C. chinensis individuals (DBH 1cm), which accounted for 11% of the total ‘importancevalue’ (sum of relative frequency, relative density, and relativebasal area, Ye et al. 2008). The C. chinensis individuals had aunimodal distribution of DBH: the frequency increased toa peak at DBH 20–30 cm and then decreased gradually inlarger size classes (Fig. 1 in Li et al. 2008). The individualsthat we studied (DBH 40 cm) were clearly distributed indistinct locations from most of the other individuals (Supplementarymaterial Appendix A1 Fig. A1). According to localhistory (Z.-L. Huang pers. comm.), the entire C. chinensispopulation was assumed to be established approximately 60years ago from a single source population that is located justto the southeast of the plot. The source population itself hasa history of more than 400 years according to the records ofa nearby Buddhist monastery.2297

Sample collection and microsatellite analysisLeaf samples from 212 C. chinensis individuals in theplot with DBH 40 cm were collected between July andOctober 2008. The samples were placed in sealed plasticbags that contained silica gel until DNA extraction. Eightmicrosatellite markers, including a newly isolated locus,Cch40, were analysed using the procedures described byHuang et al. (2009) and Dong et al. (2010), except formodifications in the annealing temperatures used in theamplification of some loci (Table 1).Data analysisAll conversions of the format of the genetic data were performedusing the software CONVERT 1.3 (Glaubitz 2004)and GENALEX 6.2 (Peakall and Smouse 2006).Genetic variationThe genetic diversity parameters of allelic richness (A),observed heterozygosity (H O ), unbiased expected heterozygosity(H E ) (Nei 1978), and fixation index (f) were estimatedusing the software GENETIX 4.05 (Belkhir et al.1996–2004). Deviations from Hardy–Weinberg equilibrium(HWE) for each locus and genotypic linkage disequilibrium(LD) between all pairs of loci were tested using the softwareGENEPOP 4.0.7 (Raymond and Rousset 1995, Rousset2008), and the significance level was adjusted with theBonferroni correction.Genetic cluster analysisBayesian cluster analysis was performed to assign individualsinto clusters on the basis of their multilocus genotypesusing the software STRUCTURE 2.3.1 (Falush et al.2003, Pritchard et al. 2000), because STRUCTURE canreveal within-population genetic substructure effectively(Anderson and Dunham 2008, Gao et al. 2007). Using amodel that assumed no admixture and no correlated allelefrequencies, 20 independent runs were performed for eachvalue of K (putative cluster numbers, from 1 to 10) with10 6 iterations after a burn-in period of 10 6 on a total of212 multilocus genotypes. Given that the highest meanlog-likelihood [ln Pr(X|K)] value, an estimate of the posteriorprobability of the data for a given K, does not identifythe number of clusters reliably, the preferred K was alsoexamined using ∆K (Evanno et al. 2005). The latter comparesthe rate of change in ln Pr(X|K) between successiveK values, and the value of K with the highest ∆K is usuallychosen. For the inferred K, CLUMPP 1.1.2 (Jakobsson andRosenberg 2007) was used to calculate the average membershipcoefficient for each individual by aligning the results ofthe above-mentioned 20 runs, because clustering algorithmsin STRUCTURE result in slightly different outcomes ineach independent stochastic simulation.The distributions of the individuals that were clusteredby STRUCTURE were visualized further using principalcomponent analysis (PCA) implemented in the softwareGENALEX 6.2 (Peakall and Smouse 2006). To performPCA, pairwise relatedness coefficients (Ritland 1996)between individuals were calculated using the same software,and then they were multiplied by 1 because thePCA requires a dissimilarity matrix in GENALEX.Analyses of the association between topography andgenetic groupsTwo methods were used for the analyses of the associationbetween topography and genetic groups: generalized leastsquares (GLS) and a modified c 2 statistic. The former wasused to determine which topographic features (variables)contributed significantly to the variation in the distributionof individuals, and the latter, to estimate the associationof the spatial distribution of the genetic groups withhabitats.1) GLSThe GLS model was chosen to accommodate spatial autocorrelationamong quadrates (Schabenberger and Gotway2005). To identify important predictors, we used the bestfitmodel developed by McEwan et al. (2011). To constructthis model, factors are eliminated sequentially from the fullmodel until all remaining factors are significant. If there areseveral possible models, the model with the smallest valueof Akaike’s information criterion (AIC) is chosen. In thepresent study, four topographic variables, namely elevation,convexity, slope, and aspect, were measured in accordancewith Wang et al. (2009) for each 20 20 m quadrate withinthe DHS plot in which at least one of the 212 C. chinensisindividuals was distributed. Five of the 212 individuals thatwere located on a border between quadrates were excludedfrom the analysis. The full model contained the linear andTable 1. Locus name, GenBank accession number, annealing temperature (Ta), allelic richness (A), observed (H O ) and expected heterozygosity(H E ), and fixation index (f) for eight microsatellite loci used to characterize 212 C. chinensis individuals.Locus GenBank accession number Ta (°C) A H O H E f ReferenceCch11 EU846108 60 10 0.755 0.780 0.033 Huang et al. 2009Cch12 EU846109 60 8 0.745 0.760 0.019 Huang et al. 2009Cch13 EU846110 58 26 0.948 0.920 0.030 Huang et al. 2009Ccu62F15 AB092346 64 4 0.590 0.610 0.033 Huang et al. 2009Ms04 GU097387 58 6 0.608 0.613 0.008 Dong et al. 2010Ms06 GU097389 56 17 0.910 0.883 0.031 Dong et al. 2010Ms09 GU097392 60 4 0.288 0.285 0.009 Dong et al. 2010Cch40* HM123729 58.5 6 0.682 0.713 0.043Total 0.691 0.696 0.007*forward primer sequence: 5¢-GGAGAAGACCGTACGTGGAA-3¢reverse primer sequence: 5¢-ACACATACACCCACACACACA-3¢2983

quadratic terms of these four independent variables. Themodel was performed using R package ‘nlme’ (Pinheiro et al.2009), and the parameters were estimated using maximumlikelihood methods.2) Modified c 2 statistic for habitat associationWith multivariate regression tree analysis (MRT, De’ath2002), five habitat types: low valley (H1), low hillside (H2),high valley (H3), high hillside (H4), and ridge (H5), weredefined for the plot (Fig. 1B–D). H1 and H2 are separatedfrom H3 and H4 by the threshold of 339.2 m above sealevel, which is the mean elevation of the DHS plot. H1is separated from H2 using the mid-value of the slope,and H5 is also separated from H3 and H4 by the midvalueof the slope in the same manner. H3 contains concavecells (convexity 0), whereas H4 contains convex cells(convexity 0).The analysis consisted of the following steps. First, usinga c 2 statistic, we obtained the expected numbers of individualsin the three genetic groups (refer to Genetic clusterin Results) in the five habitat types. Second, the overallvalue of c 2 was calculated from the expected and observednumbers of individuals in each group for each habitat type.Third, torus translation tests were carried out. Torus translationtests compare the observed overall c 2 of focal groups ina habitat with the expected overall c 2 under a null modelthat assumes that the genetic groups do not differentiate interms of habitat preferences. We divided the DHS plot into500 [25 (N S) 20 (E W)] 20 20 m quadrates, and thusthere were 500 unique torus translations. Three simulatedmaps were generated from each translation: 180° rotation,mirror image, and 180° rotation of the mirror image. Theseprocedures generated a total of 1999 translated maps, eachof which provided an expected overall c 2 . If the observed valueswere higher than 95% of the values obtained under thenull hypothesis for one genetic group in one habitat type,we considered that the genetic group was associated with thehabitat type.Spatial genetic structure (SGS)To analyse SGS within genetic groups, pairwise relatednesscoefficients (Ritland 1996) were regressed on geographicaldistance using SPAGEDI 1.2 g (Hardy and Vekemans2002). We set our first distance class at 20 m, and then setthe other classes at 20-m intervals up to 400 m. The referenceallele frequencies that were used to compute pairwiserelatedness coefficients (Ritland 1996) were based on theentire sample. The confidence interval (95%) for the averagerelatedness coefficient in a particular distance class wasobtained from 10 000 permutations of individual spatiallocations. Relatedness coefficients above or below the 95%confidence envelopes indicate a degree of genetic structuringthat is significantly larger or smaller than would be expectedat random, respectively.The strength of SGS was quantified further using theSp statistic (Vekemans and Hardy 2004). This is calculatedusing Sp b(1 F 1 ), where b is the slope of regression ofthe relatedness coefficient on the logarithm of spatial distance,and F 1 is the average relatedness coefficient in the firstdistance class.Spatial patternThe spatial pattern of individuals was analysed using thedensity-corrected univariate neighbourhood density function(NDF) (Condit et al. 2000). We calculated the NDFfor the same distance intervals as in the SGS analysis, upto half the length of the shortest side of the plot, which is200 m in our study. Theoretically, when the corrected NDFvalue is close to 1.0, less than 1.0, and greater than 1.0, thespatial pattern is random, uniform, and clumped at the givendistance class, respectively. To assess statistical significance,NDF values were compared with 95% confidence envelopesgenerated from 999 Monte Carlo simulations of the nullhypotheses. All calculations and simulations were performedusing SPPACK 1.39 (Perry 2004).ResultsGenetic variationThe number of alleles per locus ranged between 4 and26 (Table 1). H E for the loci ranged from 0.285 to 0.920.No locus showed significant deviation from HWE beforeBonferroni correction (average f 0.007). Nine of the28 locus pairs showed significant deviation from LD at the5% level after Bonferroni correction. However, all locuspairs were in LD when only individuals with DBH 70 cmwere analysed.Genetic clusterThe mean log-likelihood, ln Pr(X|K), increased from K 1to 6 and then decreased, and the maximum ∆K was at K 2.Thus, we consider that the optimal number of clusters is 2.The CLUMPP results indicated strong agreement among 20independent runs at K 2 (H¢ 0.998, average pairwisesimilarity of the 20 runs). On the basis of these findings,individuals were assigned to genetic groups. If their coefficientsof membership of cluster one or two were larger than0.9, they were assigned to group 1 (G1) and 2 (G2), respectively;otherwise, they were considered to be intermediatebetween the groups (Fig. 2B). In the PCA, the individuals inG1 and G2 were separated by Axis 1, with the intermediateindividuals in the middle (Fig. 2A).There were 63, 86 and 63 individuals in G1, G2 andthe intermediate group, respectively, and the frequencies ofindividuals within the same DBH classes in these groupswere similar (Fig. 2C). Therefore, differences in sample sizesamong the groups were not considered for intergroup comparisonin the following analyses.Association between topography and genetic groupsIn the GLS analysis, G1 and G2 had significant positiveand negative correlations with elevation in the linear andquadratic forms, respectively. No topographic features weresignificant for all C. chinensis individuals when combinedtogether as a single group, or for the individuals with anintermediate status between the groups.In the torus translation tests, all C. chinensis individualswhen combined together as a single group did not have4299

Axis 20.080.060.040.020.00–0.02–0.04–0.06–0.08–0.10(A)–0.10 –0.05 0.00 0.05 0.1Axis 1Frequency(C)0.50.40.30.20.101 2 3 4 5 6 7DBH size classes(B)Group 2Intermediate individualsGroup 1Membershipcoefficient10.50Figure 2. (A) Principal component analysis (PCA) using the Ritland (1996) relatedness coefficient for the 212 C. chinensis individuals withDBH 40 cm in the DHS plot. Individuals were assigned into three groups on the basis of their cluster membership coefficients. The firstand second axes explained 22.13% and 16.97% of the total variation, respectively. (B) Membership coefficients of the individuals in differentclusters from the STRUCTURE analysis. (C) Distribution of individuals among the different DBH size classes for the three geneticgroups: 1: 40 DBH 50 cm; 2: 50 DBH 60 cm; 3: 60 DBH 70 cm; 4: 70 DBH 80 cm; 5: 80 DBH 90 cm;6: 90 DBH 100 cm; and 7: DBH 100 cm.significant relationships with any of the five habitats. For thec 2 statistic, the individuals in G1 were significantly associatedpositively with high hillsides and ridges, and significantlyassociated negatively with low valleys, whereas thosein G2 were significantly associated negatively with high hillsides.However, the individuals with an intermediate statusbetween the groups did not have significant associationswith any of the five habitats.Spatial genetic structure (SGS)There was significant SGS with positive relatedness coefficientsup to 100 m for all the individuals when combinedtogether as a single group, and most of the coefficients werenegative and significant at distances greater than 100 m (Fig.3). The relatedness coefficients in all distance classes werepositive for G1 and G2, and those in the first five (20–100 m)and three (20–60 m) distance classes were significant at the5% level for G1 and G2, respectively. In all distance classes,the relatedness coefficients for G1 and G2 were also higherthan those for the combined group of all individuals andfor the individuals with an intermediate status between thegroups, except that the latter had the second highest coefficientin the first distance class (Fig. 3). Therefore, in general,G1 and G2 had higher SGS than the combined group ofall individuals and the intermediate group. The Sp statisticvalues were 0.025, 0.013, 0.011, and 0.012 for G1, G2, theintermediate group, and the combined group of all individuals,respectively.Spatial patternWhen all the C. chinensis individuals were combined togetheras a single group, their distribution was clumped significantlyup to the distance class of 60 m by NDF analysis (Fig. 4).This distribution was also found for G1 up to 140 m, exceptat 80 m, and for G2 up to 80 m. The individuals with anintermediate status between the groups had a random distributionat all distance classes. In addition, G1 generally had amore clumped distribution than the other groups (Fig. 4).DiscussionOur study addresses the question of whether a species canbe considered as a single combined entity across a landscape,Relatedness coefficient0.10.080.060.040.020–0.022060100140Distance class180220260300340Figure 3. Relatedness coefficients in the different geographical distanceclasses for the 212 C. chinensis individuals with DBH 40 cm.Filled symbols indicate significant coefficients at the 0.05 level. Thecolours of blue, red, green, and black indicate genetic group 1 and2, the individuals with an intermediate status between the groups,and all individuals when combined together as a single group,respectively.380m3005

NDF32.521.510.5020 40 60 80 100 120 140 160 180 200Distance classFigure 4. Results of spatial pattern analyses by neighbourhooddensity function (NDF) for the 212 C. chinensis individuals withDBH 40 cm for the different genetic groups. Filled symbols indicatesignificant departure from spatial randomness at the 0.05 level.The colours of blue, red, green, and black indicate genetic group1 and 2, the individuals with an intermediate status between thegroups, and all individuals when combined together as a singlegroup, respectively.or whether it should be considered in terms of genetic subgroupsthat may each be specialized to a particular type ofhabitat within a general area of the landscape.Our data show that the distributions of mature individualsof the tree species Castanopsis chinensis when allindividuals were analysed as a single combined group, orof individuals that were intermediate between the twogenetic groups G1 and G2, were not associated withtypes of topography; a pattern that might be attributableto a neutral dispersal process. However, individuals ingenetic group G1 were associated with elevation and threehabitat types, preferring high hillsides and ridges whileavoiding low valleys, whereas those in G2 showed theopposite associations with elevation and avoidance of highhillsides. The distribution of the two genetic groups maybe attributable to a deterministic niche process. Thesefindings imply that the distributions of the individualsin G1 and G2 vary with topographic features and habitats,and that different genetic groups within C. chinensisare in fact functionally different and specialized for topographicallydifferent environments. However, if we had notclassified the individuals into genetic groups, we would nothave detected this association.In contrast to our method of grouping individuals genetically,the classification of individuals into different life-stagegroups on the basis of DBH and height is a general approachthat can be used if no other information is available (Manabeet al. 2000, Wang et al. 2003, Comita et al. 2007, Li et al.2008, Nishimura et al. 2008). Some previous studies haveindicated that the habitats in which species establish themselvesvary with life stage (Manabe et al. 2000, Comita et al.2007, Nishimura et al. 2008), and have also emphasized theimportance of classifying individuals before studying speciesmdistributions at local scales (Comita et al. 2007). However,in these previous studies, it was unknown whether the individualsin different life stages were genetically different.On the basis of DBH, we classified the 212 matureC. chinensis trees into two life-stage groups (stage I: DBH 50 cm, 111 individuals; stage II: 40 DBH 50 cm,101 individuals). GLS and c 2 statistic analyses indicatedthat the distribution of individuals in life stage I was significantlynegatively associated with high valleys, whereas thatfor life stage II was significantly positively associated withhigh valleys (Supplementary material Appendix A1 Fig. A2).This indicates that the individuals in different life stages haddifferent associations with topographic environments, whichsupports the assertion that habitat associations vary with lifestage (Manabe et al. 2000, Comita et al. 2007, Nishimuraet al. 2008). However, when we analysed the genetic differentiationbetween these two life stages, the F ST value was0.0035, which was much smaller than the values of 0.0968(G1 versus G2), 0.0298 (G1 versus intermediate individuals),and 0.0237 (G2 versus intermediate individuals) thatwere obtained for other comparisons. This indicates that thedifferences in the distribution of individuals in the two lifestages are confounded by genetic differences between thelife stages. In addition, DBH cannot be used definitivelyto distinguish between individuals of different life stages(Harper 1977), and thus the classification of life stage on thisbasis could be inaccurate. Therefore, environment-associateddifferences in the distribution of groups of individualsclassified in this way might be misleading. We think thatbasing analyses on genetic groups can provide a betterunderstanding of the spatial distributions of a species andassociations with the environment in a community than theuse of DBH or height.Theoretically, thinning processes, wind pollination, longdistanceseed dispersal by animals (Peng and Xu 2005),and the mixing of seeds from different mother trees owingto seeds rolling downhill will break down SGS at shortdistances, for example, up to 30–40 m (Hamrick et al. 1993,Epperson 2000, Chung et al. 2003, Luna et al. 2005, Ohsawaet al. 2007, Jump and Peñuelas 2007). In addition, thinningprocesses could also decrease the intensity of clumping andlead to random or regular distributions at later life stages(Sterner et al. 1986, Moeur 1993, Wang et al. 2003). Giventhat the individuals included in the study were mature, theymust have experienced thinning processes. On the basis ofthese assumptions and the biology of C. chinensis, we wouldnot expect to find significant SGS and clumping among the212 mature individuals of C. chinensis in the plot as we did.According to the population history of the DHS plot, theC. chinensis population was established recently from a singlesource population, located near the plot (see the descriptionof the study site in Material and Methods). To verify this, weanalysed variation in the chloroplast DNA (cpDNA) genomeusing five intergenic spacers (atpB-rbcL, trnL UAA -trnF GAA ,psbF-psbB, petN-psbM, and rpoB-trnC GCA ; Supplementarymaterial Appendix A1 Table A1) for nine individuals fromG1 (with membership coefficients of 1.000 for cluster one)and 14 individuals from G2 (with membership coefficientsfrom 0.999 to 1.000 for cluster two). The results indicatedthat all these individuals shared the same cpDNA haplotype(total length of 4558 bp). Given that cpDNA is generally6301

inherited maternally in angiosperms, its analysis is ideal fordetecting seed migration (Schaal et al. 1998). Therefore, theshared cpDNA haplotype in these individuals confirms it islikely that the two groups (G1 and G2) of C. chinensis thatwe defined in the present study are originated from the samegene pool or source population. Therefore, the two distinctgenetic groups that we observed might not have been separatebefore the recent establishment of the population. Assuch, we think that the observed findings can be attributedto environmental selection pressures because it is possiblethat, during development of the population, genetically similarindividuals were selected in patches with similar environmentalconditions, which resulted in significant SGS andclumped distribution in and among these patches.Gram and Sork (2001) demonstrated the presence of significantgenetic variation in association with environmentalheterogeneity for three tree species that differed in life-historytraits, such as pollination, seed dispersal, and successionaland canopy status. Although their studies were conductedamong populations, it is possible that, if the divergent environmentalselection pressures are strong enough and consistentwithin a population, genetic subgroups could developeasily regardless of the life-history traits of the species, even ifthere is high intra-population gene flow (Nevo et al. 2005).To our knowledge, the present study is the first to haveclassified individuals of a tree species within a populationinto genetic subgroups and detected the associations of thesesubgroups with different types of environment. It is likelythat our observation is not specific to C. chinensis because, asdiscussed above, different environmental selection pressureswithin a population that result in the association of differentgenetic subgroups with particular environments couldalso occur in other plant species with different life-historytraits. However, studies on different plant species are neededto confirm this.In conclusion, we detected distinct genetic groups withina C. chinensis population, and these groups showed environmentalspecialization within the study plot. However,given that different groups of individuals occupied differenthabitats, C. chinensis is a generalist at the species level, atleast in terms of the variety of environments that we studied.Our results indicate that, at the species level, considerationof only neutral or niche-based spatial patterns in commonspecies might result in an incomplete understanding withrespect to the environments that the species occupy, andgenetic subgroups within a species that show environmentalspecialization could be a key attribute below the species level.Consideration of such genetic subgroups and their associationwith types of environment will contribute to a betterunderstanding of species persistence and coexistence incommunities (Clark 2010).Acknowledgements – The first two authors, Zheng-Feng Wang andJu-Yu Lian, contributed equally to this paper. We thank Lin-FangWu, Dan-Dan Gao, Hong-Yu Niu, Xin Zhang, Xiao-Yi Li, LeiDong, Peng Zhu, Jian He and Gui-Lian Tao for help in data collection;the Subject Editor, Olivier Hardy for helpful comments onthe manuscript; and Guillaume Blanchet, Guochun Shen, RichardCondit and Ryan Chisholm for data analysis. This work was supportedby the Knowledge Innovation Program of the ChineseAcademy of Sciences (KSCX2-EW-Z, KSCX2-EW-J-28), ForeignExchange Program National Founder (31061160188), the NationalKey Technology R&D Program (2008BAC39B02), the NationalNatural Science Foundation of China (31170352, 31100312), andthe Chinese Forest Biodiversity Monitoring Network.ReferencesAnderson, E. C. and Dunham, K. K. 2008. The influence offamily groups on inferences made with the program Structure.– Mol. Ecol. Resour. 8: 1219–1229.Belkhir, K. et al. 1996– 2004. GENETIX 4.05, logiciel sousWindows TM pour la génétique des populations. – LaboratoireGénome, Populations, Interactions, CNRS UMR 5000,Université de Montpellier II, Montpellier (France).Chung, M. Y. et al. 2003. Genetic structure of age classes inCamellia japonica (Theaceae). – Evolution 57: 62–73.Clark, J. S. 2010. Individuals and the variation needed for highspecies diversity in forest trees. – Science 327: 1129–1132.Comita, L. S. et al. 2007. Developmental changes in habitatassociations of tropical trees. – J. Ecol. 95: 482–492.Condit, R. 1998. Tropical forest census plots. – Springer.Condit, R. et al. 2000. Spatial patterns in the distribution oftropical tree species. – Science 288: 1414–1418.De’ath G. 2002. Multivariate regression trees: a new technique formodeling species – environment relationships. – Ecology 83:1105–1117.Dong, L. et al. 2010. Isolation and characterization of microsatelliteloci in Castanopsis fissa in lower subtropical China. – SilvaeGenet. 59: 299–300.Du, Y. and Huang, Z. 2008. Effects of seed mass and emergencetime on seedling performance in Castanopsis chinensis. – ForestEcol. Manage. 255: 2495–2501.Epperson, B. K. 2000. Spatial genetic structure and nonequilibriumdemographics within plant populations. – PlantSpec. Biol. 15: 269–279.Evanno, G. et al. 2005. Detecting the number of clusters ofindividuals using the software STRUCTURE: a simulationstudy. – Mol. Ecol. 14: 2611–2620.Falush, D. et al. 2003. Inference of population structure usingmultilocus genotype data: linked loci and correlated allelefrequencies. – Genetics 164: 1567–1587.Gao, H. et al. 2007. A Markov chain Monte Carlo approach forjoint inference of population structure and inbreeding ratesfrom multilocus genotype data. – Genetics 176: 1635–1651.Glaubitz, J. C. 2004. CONVERT: A user-friendly program toreformat diploid genotypic data for commonly used populationgenetic software packages. – Mol. Ecol. Notes 4:309–310.Gram, W. K. and Sork, V. L. 2001. Association between environmentaland genetic heterogeneity in forest tree populations.– Ecology 82: 2012–2021.Hamrick, J. L. et al. 1993. The influence of seed dispersal mechanismson the genetic-structure of tropical tree populations.– Vegetatio 108: 281–297.Hardy, O. J. and Vekemans, X. 2002. SPAGeDi: a versatilecomputer program to analyse spatial genetic structure at theindividual or population levels. – Mol. Ecol. Notes 2:618–620.Harper, J. L. 1977. Population biology of plants. – AcademicPress.Huang, G. et al. 2009. Isolation and characterization of polymorphicmicrosatellite loci in Castanopsis chinensis Hance(Fagaceae). – Conserv. Genet. 10: 1069–1071.Hubbell, S. P. and Foster, R. B. 1986. Biology, chance, and historyand the structure of tropical rain forest tree communities. – In:Diamond, J. and Case, T. J. (eds), Community ecology. Harperand Row, pp. 314–329.3027

Jakobsson, M. and Rosenberg, N. A. 2007. CLUMPP: a clustermatching and permutation program for dealing with labelswitching and multimodality in analysis of populationstructure. – Bioinformatics 23: 1801–1806.Jump, A. S. and Peñuelas, J. 2007. Extensive spatial geneticstructure revealed by AFLP but not SSR molecular markers inthe wind-pollinated tree, Fagus sylvatica. – Mol. Ecol. 16:925–936.Kraft, N. J. B. et al. 2008. Functional traits and niche-based treecommunity assembly in an amazonian forest. – Science 322:580–582.Legendre, P. et al. 2009. Partitioning beta diversity in a subtropicalbroad-leaved forest of China. – Ecology 90: 663–674.Leibold, M. A. and McPeek, M. A. 2006. Coexistence of the nicheand neutral perspectives in community ecology. – Ecology87: 1399–1410.Li, L. et al. 2008. Spatial patterns and interspecific associations ofthree canopy species at different life stages in a subtropicalforest, China. – J. Integr. Plant Biol. 50: 1140–1150.Luna, R. et al. 2005. Spatial genetic structure of two sympatricneotropical palms with contrasting life histories. – Heredity95: 298–305.Ma, K. P. 1999. Ecosystem diversity in key area of China.– Zhejiang Sci. and Tec Press, Hangzhou, China.Manabe, T. et al. 2000. Population structure and spatial patternsfor trees in a temperate old-growth evergreen broad-leavedforest in Japan. – Plant Ecol. 151: 181–197.McEwan, R. W. et al. 2011. Topographic and biotic regulation ofaboveground carbon storage in subtropical broad-leavedforests of Taiwan. – For. Ecol. Manage. 262: 1817–1825.Miyamoto, K. et al. 2003. Habitat differentiation among tree specieswith small-scale variation of humus depth and topographyin a tropical heath forest of Central Kalimantan Indonesia.– J. Trop. Ecol. 19: 43–54.Moeur, M. 1993. Characterizing spatial patterns of trees usingstem-mapped data. – For. Sci. 39: 756–775.Nazre, M. et al. 2009. Effect of topography and soil on the distributionof under canopy trees of Garcinia (Guttiferae) inlowerland forest of Peninsular Malaysia. – Int. J. Bot. 5:287–294.Nei, M. 1978. Estimation of average heterozygosity and geneticdistance from a small number of individuals. – Genetics 89:583–590.Nevo, E. et al. 2005. Genomic microsatellite adaptive divergenceof wild barley by microclimatic stress in ‘Evolution Canyon’,Israel. – Biol. J. Linn. Soc. 84: 205–224.Nishimura, S. et al. 2008. Spatial patterns and habitat associationof Fagaceae in a hill dipterocarp forest in Ulu Gadut, WestSumatra. – J. Trop. Ecol. 24: 535–550.Noguchi, H. et al. 2007. Habitat divergence in sympatric Fagaceaetree species of a tropical montane forest in northern Thailand.– J. Trop. Ecol. 23: 549–558.Ohsawa, T. et al. 2007. Steep slopes promote downhill dispersalof Quercus crispula seeds and weaken the fine-scale geneticstructure of seedling populations. – Ann. For. Sci. 64:405–412.Peakall, R. and Smouse, P. E. 2006. GENALEX 6: genetic analysisin Excel. Population genetic software for teaching and research.– Mol. Ecol. Notes 6: 288–295.Peng, S.-J. and Xu, G.-L. 2005. Seed traits of Castanopsis chinensisand its effects on seed predation patterns in DinghushanBiosphere Reserve. – Ecol. Environ. 14: 493–497.Perry, G. L. W. 2004. SpPack: spatial point pattern analysis in Excelusing Visual Basic for Applications (VBA). – Environ. Modell.Softw. 19: 559–569.Pinheiro, J. et al. 2009. nlme: Linear and nonlinear mixed effectsmodels. – R package ver. 3.1-96.Plotkin, J. B. et al. 2000. Species-area curves, spatial aggregation,and habitat specialization in tropical forests. – J. Theor. Biol.207: 81–99.Pritchard, J. K. et al. 2000. Inference of population structure usingmultilocus genotype data. – Genetics 155: 945–959.Queenborough, S. A. et al. 2007. Habitat niche partitioning by 16species of Myristicaceae in Amazonian Ecuador. – Plant Ecol.192: 193–207.Raymond, M. and Rousset, F. 1995. GENEPOP (ver. 1.2): populationgenetics software for exact tests and ecumenicism.– J. Hered. 86: 248–249.Ren, H. et al. 2008. Nurse plant theory and its application inecological restoration in lower subtropics of China. – Prog.Nat. Sci. 18: 137–142.Ritland, K. 1996. Estimators for pairwise relatedness and individualinbreeding coefficients. – Genet. Res. 67: 175–185.Rousset, F. 2008. Genepop’007: a complete reimplementation ofthe Genepop software for Windows and Linux. – Mol. Ecol.Resour. 8: 103–106.Schaal, B. A. et al. 1998. Phylogeographic studies in plants: problemsand prospects. – Mol. Ecol. 7: 465–474.Schabenberger, O. and Gotway, C. A. 2005. Statistical methodsfor spatial data analysis. – Chapman and Hall.Sterner, R. W. et al. 1986. Testing for life historical changes inspatial patterns of four tropical tree species. – J. Ecol. 74:621–633.Svenning, J. C. 1999. Microhabitat specialization in a speciesrich palm community in Amazonian Ecuador. – J. Ecol. 87:55–65.Vekemans, X. and Hardy, O. J. 2004. New insights from fine-scalespatial genetic structure analyses in plant populations. – Mol.Ecol. 13: 921–935.Volis, S. et al. 2004. The influence of space in genetic-environmentalrelationships when environmental heterogeneity and seeddispersal occur at similar scale. – Am. Nat. 163: 312–327.Wang, Z. et al. 2009. Species-topography association in a subtropicalforest of China. – Basic Appl. Ecol. 10: 648–655.Wang, Z.-F. et al. 2003. Spatial pattern of Cryptocarya chinensis lifestages in lower subtropical forest, China. – Bot. Bull. Acad.Sin. 44: 159–166.Wu, T. 2009. Flora of Guangdong (IX). – Guangdong Sci. andTech. Press, Guangzhou, China.Ye, W.-H. et al. 2008. Community structure of a 20 hm 2 lowersubtropical evergreen broadleaved forest plot in Dinghushan,China. – J. Plant Ecol. 32: 274–286.Supplementary material (available online as AppendixO20483 at www.oikosoffice.lu.se/appendix). AppendixA1.8303

American Journal of Botany: e123–e126. 2012.AJB PRIMER NOTES & PROTOCOLS IN THE PLANT SCIENCESI SOLATION AND CHARACTERIZATION OF 36 POLYMORPHICMICROSATELLITE MARKERS IN S CHIMA SUPERBA (THEACEAE) 1H ONG-YU N IU 2,3,4 , X IAO-YI L I 2,3,4 , W AN-HUI Y E 2,3,5 , Z HENG-FENG W ANG 2,3,5 ,H ONG-LIN C AO 2,3 , AND Z HANG-MING W ANG 2,32Key Laboratory of Plant Resources Conservation and Sustainable Utilization, South China Botanical Garden, Chinese Academyof Sciences, Guangzhou 510650, People ’ s Republic of China; 3 Guangdong Key Laboratory of Digital Botanical Garden, SouthChina Botanical Garden, Chinese Academy of Sciences, Guangzhou 510650, People ’ s Republic of China; and 4 GraduateUniversity of the Chinese Academy of Sciences, Beijing 100049, People ’ s Republic of China• Premise of the study: Our objective was to develop microsatellite markers to investigate the level of genetic diversity withinand among populations in a dominant evergreen broad-leaved tree, Schima superba , in southern China.• Methods and Results: Thirty-six microsatellite markers were developed and showed polymorphism in three populations. Thenumber of alleles per locus ranged from six to 34, with an average of 19. The observed and expected heterozygosities rangedfrom 0.242 to 1.000 and from 0.504 to 0.945, respectively.• Conclusions: The developed microsatellites will be useful for studying genetic diversity and population structure in S.superba .Key words: evergreen broad-leaved tree; genetic diversity; genetic markers; Schima superba ; Theaceae.Schima superba Gardner & Champ. (Theaceae) is a dominantbroad-leaved evergreen tree with a wide distribution insouthern China. It grows in various soil types and withstandslow temperatures. It grows quickly and can grow up to 30 m tall;its wood is hard and is useful in construction; and its leavesare thick and have a high water content. These features makethe tree important for reforestation, site restoration programs,and fire prevention in plantations ( Yang et al., 2008 ). It hasbeen reported that morphological variation in S. superba , incharacters such as seed mass, leaf width and thickness, barkthickness, and wood color, is high and influenced by environmentalconditions including temperature, altitude, and precipitation( Zhang et al., 2004 ; Wang et al., 2011 ). However, it isunknown whether these variations are the result of phenotypicplasticity and/or ecotypic variations, and markers for the geneticanalysis of the species are not available. Therefore, wereport here the development of microsatellites that we will useto study its genetic diversity, population structure, and the correlationbetween genetic diversity and both phenotypic variationand geographic distance to understand the effects of geneflow and environments on the genetic structure of the species.1Manuscript received 14 September 2011; revision accepted 30 October2011.This work was supported by the Knowledge Innovation Project of theChinese Academy of Sciences (KSCX2-EW-Z), the Foreign ExchangeProgram National Fund (31061160188), and the National Key TechnologyR & D Program (2008BAC39B02).5Author for correspondence: why@scib.ac.cn; wzf@scib.ac.cndoi:10.3732/ajb.1100454METHODS AND RESULTSTotal genomic DNA was extracted using a modified cetyltrimethylammoniumbromide (CTAB ) method ( Doyle, 1991 ) from one dry leaf tissue ofS. superba from POP-Zhaoqin ( Table 1 ). Approximately 250 ng genomic DNAwas digested into 300 – 1000 bp fragments using the restriction enzyme Mse I(New England Biolabs, Beijing, China). The resulting fragments were thenligated to Mse I adapters ( Mse I F: 5 ′ -TACTCAGGACTCAT-3 ′ and Mse I R:5 ′-GACGATGAGTCCTGAG-3 ′ ) using T4 DNA ligase (New England Biolabs)overnight at 16 ° C. The digestion – ligation mixture was subsequently diluted10 × , and 2 μ L of dilution was used for PCR amplification using Mse I-adapterspecific primers (5 ′ -GATGAGTCCTGAGTAAN-3 ′ , i.e., Mse I-N). Amplifiedproducts were hybridized with 5 ′ -biotinylated (AG) 15 and (AAG) 8 probes. Subsequentprobe-bound DNA fragments were enriched for AG or AAG repeatsusing streptavidin-coated magnetic beads (New England Biolabs). Enrichedfragments were recovered with PCR reaction using Mse I-N as primer. The selectedfragments were then ligated into the pGEM-T plasmid vector (PromegaCorporation, Shanghai, China) and transformed into Escherichia coli DH5 αcompetent cells (TaKaRa Biotechnology Co., Dalian, China). A PCR-basedmethod ( Lunt et al., 1999 ) was used to screen the recombinant clones. A totalof 339 positive clones were identified and sequenced by Majorbio Biotech Co.,Ltd. (Shanghai, China), with M13R or M13F as primers. We detected microsatelliteshaving at least five AG or AAG repeats in 296 sequences. We then usedPrimer Premier 5.0 (PREMIER Biosoft International, Palo Alto, California,USA) to design primers for these sequences, and 134 of the sequences werediscarded due to primer design failure (i.e., repeats close to one end of a sequenceor low quality of primers indicated by Primer Premier 5.0 software).A total of 93 samples of S. superba collected from three populations inGuangdong Province ( Table 2 , Appendix 1), China, were used to assess thepolymorphism of the microsatellites. These populations cover a relatively smallpart of the whole distribution of the species in China. PCR amplifications wereperformed in a 20 μ L volume containing 20 mM Tris-HCl (pH 8.4), 100 mM(NH 4 ) 2 SO 4 , 3 mM MgCl 2 , 0.4 mM dNTPs, 0.4 μ M each primer (5 ′ labeled withFAM, HEX, or TET), 50 ng of genomic DNA, and 1 U Taq polymerase (Ta-KaRa Biotechnology Co.). The amplification program was 95 ° C for 5 min;35 cycles of 94 ° C for 30 s, optimized annealing temperature ( Table 1 ) for 30 s, and72 ° C for 45 s; with a final extension at 72 ° C for 10 min. Electrophoresis of theproducts was performed on an ABI 3730 sequencer (Applied Biosystems,American Journal of Botany: e123–e126, 2012; http://www.amjbot.org/ © 2012 Botanical Society of Americae123304

e124 AMERICAN JOURNAL OF BOTANY [Vol. 0T ABLE 1. Characteristics of 36 microsatellite primers developed in Schima superba.Locus Repeat motif Primer sequences (5 ′ – 3 ′ ) Size range (bp) T a ( ° C) N a 5 ′ -fluorescence labelGenBankaccession no.SS01 (AG) 23 F: CTCATCACTCCTGCACAGATTCCAC 318 – 364 52 29 FAM JN642179R: AATATCGGTAATGTACTTCGCTCCTSS02 ATCT(AT) 4 GG(AG) 19 ACAGF: TTCGTTCCTCATCTTTCTGCTTGTT 261 – 305 55 20 HEX JN642180R: CATCAAAGGCAGTGTCAGCTCTATSS03 (AG) 15 TAG(TA) 2 GGAG F: AAAAGAGGGTGAGGGTGCT 221 – 333 52 13 HEX JN642181R: CCTTTTTCTTTGACTTCGSS04 (AG) 14 F: CTAAGCACATTCATCCCTACAT 152 – 176 52 12 FAM JN642182R: TGTGAAGAATTGGAGATCCTSS05 (AG) 2 TG(AG) 17 (ATAGAG) 2 F: CCAGAATCGACCAGCAAG 243 – 289 55 15 FAM JN642183R: CAGACAAAGGATGGGACAACSS06 (A) 5 TGTA(TG) 2 (AG) 14 F: CTAAGCAACTTCAATAGGACAT 209 – 289 52 28 HEX JN642184R: TACTTGAGATGCCCTAATSS07 (AT) 2 (CT) 9 F: ATGAGACCGCCTAACCTG 225 – 235 55 6 TET JN642185R: GCTTCGGAACCCTACACCSS08 (AG) 13 F: TCTTGAATCAGTGCGAGTT 100 – 124 49 11 FAM JN642186R: CTTCATCGTCCGTGTTCTSS09 (AC) 3 (AG) 18 CGAGCGCT F: TTCCCTTGATTTTGTGCTAC 200 – 250 52 20 HEX JN642187R: CCAAACAAGAAGAACCAAGTCSS10 (AG) 12 G(AG) 2 F: CCAACAAACGGCTTACAT 168 – 194 55 10 HEX JN642188R: ACACCGCAACAGAAATCGSS11 (AG) 2 (A) 3 T(AG) 10 F: CACAAGTACAGCAGGTTGAATCC 146 – 184 55 20 FAM JN642189R: TGTGCCGCTGCTGCTCTACTTCSS12 (AG) 19 F: AGTGTGTTTGGAATCTCCTCAT 199 – 251 52 15 FAM JN642190R: CCTCCTTTACCTGTTGTATTTGSS13 (AAGG) 2 GG(GA) 12 F: TTGGAACCGTCCCCACTCTAT 118 – 143 52 22 HEX JN642191R: TTGGGGCAAAGCAGAGGTATSS14 (AG) 13 TCC(A) 4 F: TCTGGGATGCTAAATGGT 195 – 257 52 16 FAM JN642192R: GTGGTTATTTCAAAAGGTCCSS15 (AGC) 2 AAC(AAG) 9 (AG) 2 F: TGATGTGGGTTTACTGTTTG 244 – 267 55 15 HEX JN642193R: ATCAATAGTCCTCAACAAGCSS16 AGTC(AG) 14 F: GAAAACTAAATGGTCCCTAC 276 – 322 55 18 HEX JN642194R: AGTTAGACTTAGCACTACGGTTSS17 Complex a F: ACTGCCAAAGCCAATCTG 228 – 264 55 19 FAM JN642195R: ATCTTGCCGACAATGACCSS18 (CT) 19 (CG) 2 F: ACCACCAGTAGCAGCCATC 93 – 220 52 24 HEX JN642196R: CAAGCCAACTCCGACAATSS19 (AG) 2 AA(AG) 9 CG(AG) 4 F: GATTGATGTTCAAAGGATGG 240 – 276 52 18 HEX JN642197R: GTTATTACTGGTTTGGTCGTSS20 (CT) 5 CC(CT) 10 F: CTCCGCCGTTTATCTTC 205 – 255 52 21 FAM JN642198R: GTTTTGCTGGACCGTTAGSS21 AGTT(AG) 21 F: GCTTTCAAAACAGATAAGGTCG 226 – 279 52 21 HEX JN642199R: TCGTCCCTGTCAGTCTCATCSS22 AGCA(CG) 2 (AG) 21 F: TCAAGCAGGAGTGAAAGC 285 – 363 52 19 FAM JN642200R: AAAGGTTGGGGTGGATAGSS23 (AG) 13 F: TGCTACTTGCTTTATTGGTG 258 – 368 55 16 HEX JN642201R: CATTCAATAGTGGGATAACGSS24 (AG) 13 AATGAT(AG) 8 F: ATAGCCTCTGGCAAATCC 194 – 204 55 6 HEX JN642202R: ACGAGGACGGTGTTGATGSS25 (AG) 30 F: ACAATCTCACAGCCACCG 277 – 335 55 21 FAM JN642203R: TGTAGAAGCCTTATCCCAATSS26 (AG) 22 AA(CA) 2 F: CCACTTCACCTTTCATCAT 120 – 148 49 14 HEX JN642204R: CACACTCATCTTCCAGACAATSS27 AGAA(AC) 2 (AG) 13 F: AAATCCACAGCAAAATGAGC 103 – 151 52 24 FAM JN642205R: TGTAAGGGTGAGGACTAAGGTSS28 (CT) 9 (CACT) 3 (CT) 5 F: TTGTAACTCCCTCTTCACCT 321 – 390 52 34 HEX JN642206R: GCCCTTCTCATTCCGTCTSS29 (AG) 7 AA(AG) 15 F: AAAGAAGGGCATAATCCAAC 127 – 176 52 20 FAM JN642207R: GGTCAAGCCAGGCAAATCSS30 (CA) 2 TA(GA) 16 F: GCTTGCTTCAGGGTTTTAT 110 – 182 52 30 HEX JN642208R: CTTGGGCTGTATTTATTGGSS31 (AG) 19 ATGA(AG) 2 AT(AG) 2 F: TTGACCTACCTCCAAAATC 200 – 242 52 21 FAM JN642209R: TATGCTCCGTTGCTGCTTSS32 (AG) 20 AAAG F: TCCCAAAACAACCCTCAT 320 – 370 52 22 HEX JN642210R: GGACTGTTGTCGGTGTTGSS33 (AC) 15 (AT) 5 (AG) 8 F: TTGGAGGTGCCTGCTTTC 283 – 345 52 21 FAM JN642211R: TTTCAGTGTAAAGGAGCCAGSS34 (AG) 8 G(AG) 3 (T) 4 G(T) 15 F: TGAGACTGGGTGGAGGAA 290 – 394 52 21 HEX JN642212R: AAGGGTTATTACTCAACAGGTC305

March 2012] AJB PRIMER NOTES & PROTOCOLS — SCHIMA SUPERBA MICROSATELLITESe125TABLE 1. Continued.Locus Repeat motif Primer sequences (5′– 3 ′) Size range (bp) T a ( ° C) N a 5 ′-fluorescence labelGenBankaccession no.SS35 (AG) 8 GAGTG(AG) 2 G(AG) 8 F: CATCTCCGTTTCATCCTG 154 – 218 52 11 HEX JN642213R: ATCCCGATGAAATAGTGGSS36 (AAG) 6 AATGGATGAATG(AG) 14 F: ATCCACCGAATACCAAAG 205 – 269 52 33 FAM JN642214R: GTCGCTTCCCTCTGGTTTNote : N a = number of alleles; T a = annealing temperature.a(AG) 10 TG(AG) 2 (G) 5 TTGGATGGTCAATGCAAGG(A) 3 GATGAGGAGTGG(T) 3 GAGCAAGGA(AG) 2 ACCA(AG) 8Guangzhou, China), and the fragment lengths were analyzed using ABI GeneMappersoftware version 3.7 (Applied Biosystems).Among the 162 microsatellites retained, 36 were successfully amplifiedwith clear and stable polymorphism ( Table 1 ). Number of alleles ( N a ), observedheterozygosity ( H o ), unbiased expected heterozygosity ( H e ), and fixation index( F ) were obtained using GenAlEx version 6.2 ( Peakall and Smouse,2006 ). Deviation from Hardy – Weinberg equilibrium (HWE) and genotypiclinkage disequilibrium among all pairs of loci in each population were analyzedwith GENEPOP version 3.4 ( Raymond and Rousset, 1995 ). Significancelevels were adjusted using Holm ’ s sequential Bonferroni correction( Holm, 1979 ) implemented in R software version 2.13.0 ( R DevelopmentCore Team, 2011 ).Alleles per locus varied from six to 34 ( Table 1 ) with an average of 19 inthe 93 individuals we sampled. Within populations, H o ranged from 0.242 to1.000 and H e ranged from 0.504 to 0.945 ( Table 2 ). There were 12 loci thatshowed significant deviation from expected proportions under HWE due toheterozygote deficiency in one or two of the three populations. No consistentlysignificant linkage disequilibrium was detected among all pairs ofloci within each population, which indicates the independence of the 36microsatellite markers.T ABLE 2. Genetic diversity of 36 loci in three populations a of Schima superba .LocusPOP-Zhaoqin ( N = 33) POP-Dongguan ( N = 32) POP-Shaoguan ( N = 28)N a H o H e F N a H o H e F N a H o H e FSS01 16 0.909 0.915 − 0.009 23 0.938 0.936 − 0.018 21 0.964 0.945 − 0.039SS02 11 1.000 0.874 − 0.162 15 0.938 0.911 − 0.045 14 0.857 0.900 0.030SS03 9 0.758 0.823 0.066 10 0.531 0.840 0.357* 9 0.571 0.840 0.308SS04 8 0.848 0.786 − 0.095 11 0.781 0.821 0.033 8 0.821 0.830 − 0.008SS05 7 0.758 0.799 0.037 12 0.813 0.862 0.042 9 0.750 0.856 0.108SS06 13 0.818 0.859 0.033 22 0.844 0.945 0.093 20 0.929 0.943 − 0.003SS07 3 0.606 0.642 0.041 5 0.344 0.504 0.308 4 0.429 0.506 0.137SS08 7 0.710 0.770 0.063 11 0.813 0.838 0.015 8 0.786 0.827 0.033SS09 14 0.970 0.894 − 0.101 14 0.938 0.920 − 0.035 12 0.929 0.849 − 0.113SS10 7 0.818 0.805 − 0.032 8 0.813 0.840 0.017 8 0.786 0.826 0.031SS11 13 0.788 0.818 0.022 19 0.844 0.929 0.077 14 0.857 0.905 0.035SS12 12 0.879 0.872 − 0.024 12 0.875 0.891 0.003 13 0.889 0.882 − 0.027SS13 12 0.818 0.847 0.019 16 0.750 0.917 0.169 15 0.857 0.903 0.034SS14 7 0.242 0.709 0.653* 14 0.469 0.896 0.469* 7 0.679 0.760 0.091SS15 5 0.576 0.594 0.016 13 0.875 0.878 − 0.012 10 0.821 0.818 − 0.023SS16 7 0.667 0.776 0.127 10 0.625 0.78 0.186 13 0.893 0.854 − 0.065SS17 12 0.909 0.874 − 0.056 17 0.935 0.937 − 0.015 14 0.893 0.890 − 0.021SS18 15 0.909 0.896 − 0.031 13 0.875 0.894 0.006 15 0.964 0.923 − 0.064SS19 12 0.688 0.818 0.146 17 0.813 0.916 0.099* 13 0.857 0.878 0.006SS20 9 0.758 0.814 0.056 15 0.813 0.896 0.079 19 1.000 0.909 − 0.120SS21 16 0.818 0.869 0.043 10 0.906 0.813 − 0.132 11 0.893 0.816 − 0.114SS22 13 0.879 0.860 − 0.038 18 1.000 0.934 − 0.088 12 0.929 0.901 − 0.049SS23 10 0.909 0.865 − 0.067 15 0.906 0.897 − 0.026 8 0.786 0.827 0.033SS24 4 0.576 0.587 0.004 6 0.500 0.623 0.184 3 0.536 0.608 0.104SS25 11 0.818 0.830 − 0.001 19 0.781 0.929 0.146* 15 0.857 0.932 0.063SS26 9 0.667 0.834 0.188 13 0.625 0.914 0.305* 13 0.714 0.895 0.187*SS27 10 0.879 0.857 − 0.041 16 0.719 0.923 0.209* 16 0.857 0.901 0.032SS28 19 0.970 0.937 − 0.051 15 0.688 0.857 0.185* 22 0.857 0.935 0.067SS29 11 0.909 0.883 − 0.046 14 0.875 0.900 0.013 15 0.929 0.904 − 0.046SS30 17 0.788 0.915 0.126 21 0.719 0.900 0.189* 19 0.643 0.931 0.297*SS31 16 0.818 0.917 0.094 14 0.906 0.931 0.011 17 0.857 0.940 0.072SS32 14 1.000 0.928 − 0.094 18 0.844 0.888 0.035 14 0.893 0.887 − 0.025SS33 12 0.727 0.813 0.091 15 0.656 0.859 0.224* 12 0.679 0.854 0.191*SS34 9 0.719 0.778 0.061 19 0.719 0.929 0.214* 13 0.593 0.883 0.316*SS35 9 0.303 0.754 0.592* 9 0.531 0.762 0.292 8 0.464 0.818 0.422SS36 18 0.909 0.932 0.010 17 0.906 0.931 0.011 20 0.821 0.918 0.089mean 11 0.781 0.826 0.011 14 0.775 0.871 0.033 13 0.801 0.861 0.027Note : F = fixation index; H e = unbiased expected heterozygosity; H o = observed heterozygosity; N = sample sizes; N a = number of alleles.aGeographic coordinates of populations: POP-Zhaoqin: 23 ° 10 ′ 10 ″ N, 112 ° 32 ′ 21 ″ E; POP-Dongguan: 22 ° 54 ′ 39 ″ N, 114 ° 13 ′ 21 ″ E; and POP-Shaoguan:24 ° 46 ′ 53 ″ N, 113 ° 36 ′ 16 ″ E.* Indicates a significant deviation ( P < 0.05) from Hardy – Weinberg equilibrium after Holm ’ s sequential Bonferroni correction.306

e126 AMERICAN JOURNAL OF BOTANY [Vol. 0CONCLUSIONSThirty-six microsatellites of S. superba were isolated andtested. Our data indicate that they are highly polymorphic.Therefore, they are useful to investigate genetic diversity, geneflow, population structure, and phylogeography of the species.We also plan to use them to study the fine-scale spatial geneticstructure, mating systems, and pollen and seed flow in a 20 haplot on Dinghushan ( Ye et al., 2008 ), Guangdong Province,China , in which more than 2000 S. superba individuals withdbh ≥ 1 cm have been mapped. The results from these and fromassociated studies of phenotypic diversity and population dynamicswill provide useful information for the sustainable managementof this species.LITERATURE CITEDD OYLE , J. J. 1991 . DNA protocols for plants — CTAB total DNA isolation.In G. M. Hewitt and A. Johnston [eds.], Molecular techniques in taxonomy,283 – 293. Springer-Verlag, Berlin, Germany.H OLM , S. 1979 . A simple sequentially rejective multiple test procedure.Scandinavian Journal of Statistics 6 : 65 – 70 .L UNT , D. H. , W. F. HUTCHINSON , AND G. R. C ARVALHO . 1999 . An efficientmethod for PCR-based isolation of microsatellite arrays (PIMA).Molecular Ecology 8 : 891 – 894 .P EAKALL , R. , AND P. E. S MOUSE . 2006 . GenAlEx 6: Genetic analysisin Excel. Population genetic software for teaching and research.Molecular Ecology Notes 6 : 288 – 295 .R D EVELOPMENT CORE TEAM . 2011 . R: A language and environmentfor statistical computing. R Foundation for Statistical Computing,Vienna, Austria.R AYMOND , M. , AND F. R OUSSET . 1995 . GENEPOP (version 1.2): Populationgenetics software for exact tests and ecumenicism. Journal of Heredity86 : 248 – 249 .W ANG , X. H. , X. H. MA , G . Q . J IN , L. Y. CHEN , AND Z . C . Z HOU . 2011 .Variation pattern of individual types and wood characters in naturalstands of Schima superba . Scientia Silvae Sinicae 47 : 133 – 139 .Y ANG , S. X. , H. DENG , AND M. S. LI . 2008 . Manganese uptake and accumulationin a woody hyperaccumulator, Schima superba . Plant, Soiland Environment 10 : 441 – 446 .Y E , W. H. , H. L. CAO , Z. L. HUANG , J. Y. L IAN , Z. G. WANG , L. LI , S. G.W EI , AND Z. M. WANG . 2008 . Community structure of a 20 hm 2lower subtropical evergreen broadleaved forest plot in Dinghushan,China. Chinese Journal of Plant Ecology 32 : 274 – 286 .Z HANG , P. , G. Q. J IN , Z. C. ZHOU , L. Y U , AND H. H. FAN . 2004 . Provenance differenceand geographic variation pattern for seedling trait of Schimasuperba . Forest Research 17 : 192 – 198 .A PPENDIX 1. Voucher specimens deposited in the herbarium of the South ChinaBotanical Garden, Guangzhou, China .Schima superba Gardner & Champ.: Guo-Liang Shi 2859 , Zhaoqin, GuangdongProvince, China; Rui-fen Lin 549 , Dongguan, Guangdong Province,China; Hua-gu Ye 21427 , Shaoguan, Guangdong Province, China.307

Plant Species Biology (2012) ••, ••–••doi: 10.1111/j.1442-1984.2011.00365.xTopographic effects on fine-scale spatial genetic structurein Castanopsis chinensis Hance (Fagaceae)psbi_365 1..7JIAN HE,*‡ XIAOYI LI,*‡ DANDAN GAO,*‡ PENG ZHU,*‡ ZHENGFENG WANG,*† ZHANGMING WANG,*†WANHUI YE*† and HONGLIN CAO*†*Key Laboratory of Plant Resources Conservation and Sustainable Utilization, †Guangdong Key Laboratory of Digital BotanicalGarden, South China Botanical Garden, Chinese Academy of Sciences, Guangzhou 510650, and ‡Graduate University of ChineseAcademy of Sciences, Beijing 100049, ChinaAbstractEcological study plots are usually treated as if they are flat. This does not hold for manysituations such as mountains where topography is complex. In areas with complex topographyindividual relationships are not only determined by projection distance, but alsoby surface distance. To demonstrate this we compared projection and surface distances byanalyzing spatial genetic autocorrelation for Castanopsis chinensis in two subplots (A andB) in the Dinghushan (DHS) national nature reserve in subtropical South China. Weobserved that the two types of distances generally result in similar fine-scale spatialgenetic structure (SGS) patterning for the spatially less structured subplot B, but not forthe highly structured subplot A. The present study shows clearly that accounting for plotarchitecture in plant species on topographically complex areas enables a more accuratepicture of the underlying spatial genetic structure to emerge.Keywords: Dinghushan, lower subtropical China, projection distance, spatial genetic structure,surface distance.Received 7 August 2011; revision received 20 August 2011; accepted 22 August 2011IntroductionCorrespondence: Zhengfeng WangEmail: wzf@scbg.ac.cnFine-scale spatial genetic structure (SGS) is an importantconsideration when analyzing population dynamics andpersistence, and provides fundamental information forplant species management (Epperson 1992; Chung et al.1998; Troupin et al. 2006). To quantify SGS, spatial autocorrelationanalysis is generally used, which tests the relationshipsin particular attributes of individuals separatedby geographic distances (Sokal & Oden 1978; Smouse &Peakall 1999; Vekemans & Hardy 2004). In flat study areasgeographic distances between individuals can be easilyand accurately computed from 2-D coordinates for subsequentlyuse in analysis. However, if a study area is topographicallycomplex the distances computed from 2-Dcoordinates are unlikely to accurately reflect the true distancesbetween individuals. In general, researchers haveignored topography and used only projection distancesinstead of true surface distances (Hardesty et al. 2005;Kitamoto et al. 2005; Ng et al. 2006). The use of such projectiondistances may result in misleading results. Forexample, if there are three individuals, A, B and C, on aslope (Fig. 1), using projection distance at a distance scaleof 5 m, we will include all individual pairs A and B, A andC and B and C in a particular attribute to carry out thespatial autocorrelation analysis. However, if we use thesurface distance along the slope, only pair A and B can beincluded in the distance scale of 5 m. Therefore, these twotypes of distances may yield different spatial autocorrelationresults because different numbers of pairs will beincluded in the same distance classes.Our objective in the present study is to investigate topographicinfluences on SGS by comparing results obtainedusing projection and surface distances at a topographicallycomplex site. To our knowledge, this is the firstempirical study to take surface distance on a fine scale intoaccount in an SGS study.Materials and methodsStudy speciesCastanopsis chinensis Hance is an evergreen tree in thefamily Fagaceae. It is distributed from South China to© 2012 The AuthorsJournal compilation © 2012 The Society for the Study of Species Biology308

2 J. HE ET AL.1.8 mAB5.2 m1.2 m 3.7 mCDVietnam below 1500 m a.s.l. (Wu 2009). As a pioneercanopy tree, it can grow up to 20 m and usually plays akeystone role in the ecosystems where it is found (Ma1999). It is a monoecious wind-pollinated species, withunisexual staminate and pistillate flowers on the sameplant. Seeds are oval and dispersed by animals, such asrodents, pigs, cats and birds (Peng & Xu 2005).Study site and field methodsSurface distanceProjection distanceFig. 1 Diagram showing the surface and projection distancesbetween individuals at a non-flat site.This study was conducted in a 20 ha Dinghushan (DHS)plot that lies on the south verge of the tropic of Cancer inthe subtropical area of South China. The plot is in the1155 ha DHS national nature reserve (112°30′39″–112°33′41″E, 23°09′21″–23°11′30″N), which was establishedin 1956. This region is characterized by southsubtropical monsoon climate, with a mean annual temperatureof 20.9°C and mean annual precipitation of1929 mm. Its typical vegetation is monsoon evergreenbroadleaved forest.The landform of the DHS plot is complex. It includes atleast three ridges and four valleys, and its elevation rangesfrom 240 to 470 m a.s.l. Plot assessments were made in2005 using standard methods of the Center for TropicalForest Science (Condit 1998). All free-standing woodyplants with d.b.h. 1 cm were mapped, tagged, identifiedto species, and the d.b.h. was recorded (Ye et al. 2008;Wang et al. 2009).The present study makes use of data from a study siteon the south part of the plot (Fig. 2). It includes one clearsteep downward stretched ridge (subplot A) and a hillside(subplot B; Fig. 2). These subplots were selectedbecause they represent the two main topographic complexitiesfound in the main plot. Subplot B was square(140 m ¥ 140 m) to facilitate sampling.Sample collection and microsatellite analysisLeaf or cambium tissues of 390 mapped living C. chinensisindividuals with d.b.h. 1 cm (based on 2005 censuseddata) in the subplots were collected between Septemberand November 2009. Tissue samples were placed insealed plastic bags containing silica gel until DNA extraction.Seven microsatellites (Cch11, Cch12, Ccu62f15,Cch40, Ms04, Ms06 and Ms09) were analyzed followingHuang et al. (2009) and Dong et al. (2010), with the exceptionthat newly optimized annealing temperatures in thepolymerase chain reaction (PCR) in loci Cch40, Ccu62f15,Ms04, Ms06 and Ms09 were used (Table 1). To minimizegenotyping errors, at least one-quarter of the individualsfor each microsatellite locus were re-genotyped.Data analysisThe number of alleles (N A) and the observed and unbiasedexpected heterozygosity (H O and H E) were calculatedfor each locus using GENALEX 6.3 (Peakall & Smouse2006), and the inbreeding coefficient (F IS, an estimation ofdeviation from random mating) was estimated usingGENEPOP 4.0.7 (Raymond & Rousset 1995; Rousset2008). Deviations from Hardy–Weinberg equilibrium(HWE) for each locus and genotypic linkage disequilibrium(LD) between all pairs of loci were tested usingGENEPOP 4.0.7, with significance levels adjusted using aBonferroni correction.To analyze SGS, autocorrelation coefficients (r) wereobtained by regressing the multilocus genetic distances tothe projection and surface distances between individualsusing GENALEX 6.3. Pairwise projection distancesbetween individuals were directly calculated using theindividual’s X and Y coordinates recorded in the field.Pairwise surface distances were estimated using the algorithm‘Surface Length’ implemented in ArcToolboox inArcGIS 9.2 (ESRI 2008) and the topographic map of Wanget al. (2009). We also used two distance intervals, 5 m and10 m, to a total distance of 100 m. Because the numberof pairs at the distance class of 5 m was less than 10for subplot A, at the 5 m distance interval we set ourfirst distance class at 10 m for subplot A. The confidenceinterval (95%) for the average autocorrelation coefficientin a particular distance class was obtained from 999permutations.ResultsThe number of alleles (N A) ranged between 4 and 18, andexpected heterozygosity (H E) ranged from 0.424 to 0.889(Table 1). No locus deviated significantly from HWE afterBonferroni correction (Table 1). Ten of the 21 locus pairsshowed significant deviation from LD at the 1% level after© 2012 The Authors Plant Species Biology ••, ••–••Journal compilation © 2012 The Society for the Study of Species Biology309

TOPOGRAPHIC EFFECTS ON GENETIC STRUCTURE 3Fig. 2 Spatial distribution of the 390 Castanopsis chinensis individuals in the 20 ha DHS plot. The grey circles represent C. chinensisindividuals not included in the present study. Circle sizes relate to d.b.h. The bar charts show the frequency distributions at different d.b.h.for individuals in subplots A and B.Table 1 Locus name, GenBank accessionnumber, annealing temperature (Ta), thenumber of alleles (N A), observed (H O) andexpected heterozygosity (H E) and inbreedingcoefficient (F IS) for the seven microsatelliteloci used in our analysis forCastanopsis chinensisLocusGenBank accessionnumber Ta (°C) N A H O H E F ISCch11 EU846108 60 10 0.782 0.806 0.030Cch12 EU846109 60 8 0.723 0.753 0.040Ccu62F15 AB092346 64 5 0.608 0.597 –0.017Ms04 GU097387 58 6 0.659 0.623 –0.057Ms06 GU097389 56 18 0.867 0.889 0.025Ms09 GU097392 60 4 0.390 0.424 0.080Cch40† HM123729 58.5 6 0.690 0.703 0.020Mean 0.674 0.685 0.037† Forward primer sequence, 5′-GGAGAAGACCGTACGTGGAA-3′; reverse primersequence, 5′-ACACATACACCCACACACACA-3′.Bonferroni correction, but all loci pairs were in linkageequilibrium if only individuals with d.b.h. 70 cm (54individuals) in the 20 ha DHS plot were tested (data notshown).The extent of SGS in subplot A was higher than thatin subplot B using both projection and surface distancesat both 5 and 10 m intervals in most distance classes(Fig. 3). At the 5 m interval (Fig. 3a) in subplot A, SGS wasPlant Species Biology ••, ••–••© 2012 The AuthorsJournal compilation © 2012 The Society for the Study of Species Biology310

4 J. HE ET AL.Surface distanceProjection distance0.160a)Subplot ASubplot B0.1400.1200.1000.080r0.0600.0400.0200.000-0.020-0.0405 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 1000.1600.1400.1200.100b)Distance classPairs in projection distancePairs in surface distanceSubplot ASubplot B0.080r0.0600.0400.0200.000-0.020-0.04010 20 30 40 50 60 70 80 90 100Distance classes (m)Fig. 3 Spatial autocorrelation analysis for subplot A and B with (a) 5 m and (b) 10 m distance intervals. Filled symbols indicate significantcoefficients at P < 0.05. The number of individual pairs for each distance class is shown on the right corner for each panel and thedescriptions are given in panel (b).© 2012 The Authors Plant Species Biology ••, ••–••Journal compilation © 2012 The Society for the Study of Species Biology311

TOPOGRAPHIC EFFECTS ON GENETIC STRUCTURE 5significant in classes from 10 to 35 m, and in the 45 m and60 m distance classes using projection distance, whereasusing surface distance there was significant SGS in thedistance classes from 10 to 45 m, and in the 85 m and 95 mclasses. In subplot B, SGS was significant at 5 m, 10 m and20 m using projection distance and at 5 m, 20 m and 60 musing surface distance.At the 10 m interval (Fig. 3b), for subplot A, SGS wassignificant at all distance classes, except for classes from70 m to 80 m using surface distance and classes in 50 m,70 m, 90 m and 100 m using projection distance. Forsubplot B, SGS was significant only at the first two distanceclasses in both projection and surface distances.DiscussionTopography influences geographic distance measurementson both large and small spatial scales (Sprague2000; Finn et al. 2006). Owing to topographic effects, thetravel distances (surface distance) for Japanese macaque(Macaca fuscata yakui) are on average 9.5% longer than theprojection distances (Sprague 2000). Finn et al. (2006) conductedan isolation by distance analysis using differenttypes of distance in one stream insect Prosimulium neomacropyga.They found that even under heterogeneousenvironments, straight-line distance (corresponding toour projection distance) could still be an acceptable surrogatefor other distances including surface distance atsmall scales, but may not be acceptable at broad spatialscales.Unlike these movable species, plants are sessile. Heterogeneousenvironments could genetically separate individualsfrom each other causing ‘biological meaningfulisolation’. No studies have considered topographicsurface distance, particularly at a local scale.In the present study we found a very different extent ofinfluence when using surface distance on SGS in our twosubplots. For subplot B extensive non-significant autocorrelationcoefficients based on projection distance areessentially unchanged when using surface distance,whereas for subplot A autocorrelation coefficientschanged markedly from significant to non-significant orvice versa. Therefore, even at a topographically complexsite, if the population is less structured a projection distancewill still work, particularly for larger distance intervals(such as 10 m in our study). In spite of this, projectiondistance actually curtails the surface distance betweenindividuals the use of projection distance may not fullycapture the true SGS status among individuals.In subplot B at 5 m intervals (Fig. 3a), we found that theautocorrelation coefficient estimated using surface distanceis higher than that using projection distance at thefirst distance, and from the first distance class to thesecond distance class the autocorrelation coefficientdecreases to non-significant by using surface distance, butremains significant when using projection distance. Theseeds of C. chinensis are dispersed first by gravity neartheir mother trees and second mainly by small rodentsover a short distance scale (Du & Huang 2008). Therefore,seed dispersion of C. chinensis depends heavily on topography.Under biological dispersal processes, spatial autocorrelationfor the first distance class is very sensitive tothe dispersal level (Epperson 2003). Therefore, accordingto our results, differences between the two types of distancein the first two distance classes reflect whethermicro-topography has an influence on the dispersion of C.chinensis seeds over a short distance. We clearly show inthe present study that accounting for surface distancewithin plots leads to a different conclusion as to the influenceof micro-topography and that follow-up studies arewarranted.Our results show that individuals in subplot A havehigher SGS than those in subplot B. Theoretically, thinningprocesses, small sampling size, overlapping generationsamong individuals and seeds rolling downhill (i.e.causing seeds from different mother trees to mix) willcontribute to the lower SGS in subplot A than subplot B(Hamrick et al. 1993; Jensen et al. 2003; Luna et al. 2005;Ohsawa et al. 2007). Two possible reasons may explain thisdisparity, local environmental selection and historicallyirregular recruitment patterns. In the former case, throughpopulation development, genetically similar individualscould be preferred in certain environmental conditions,increase in density in distinct patches and then causeincreased SGS. In the latter case, C. chinensis is a mastingspecies with large temporal variation in seed reproduction.Thus, higher genetic structure in subplot A couldrepresent past events that a few ancestors producedmassive genetically similar descendents.Our data show that in subplot A, consistent significantSGSs occur up to the 45 m distance class at 5 m distanceintervals and up to the 60 m class at 10 m distance intervalsbased on surface distance, and these compare withthe smaller distance classes up to 35 and 40 m, respectively,based on projection distance. These results indicatethat the scale of individuals’ similarities under surfacedistances is far more extensive than projection distancesreflect.The SGS measures genetic relationships among individualsin relation to distance. However, the effects ofdistance on SGS are not only affected by pollen and seedflows, although these flows sometimes are determinative(Epperson 2000), because distance can be related to environmentalheterogeneity and it determines spatial relationshipsamong individuals. For the latter in terms ofecological competition, surface and projection distancesmay yield completely different spatial relationshipsamong the same individuals. For example, individuals CPlant Species Biology ••, ••–••© 2012 The AuthorsJournal compilation © 2012 The Society for the Study of Species Biology312

6 J. HE ET AL.and D (Fig. 1) may be considered close neighbors in projectiondistance, but not so in surface distance. Becausemost biological effects, such as competitive interactions,happen among spatially adjacent neighbors (Boyden et al.2008), the first few shorter distance classes generally coverthese neighbors and they are more meaningful thanlonger ones in fine-scale SGS analysis (Marquardt &Epperson 2004; Vekemans & Hardy 2004; Epperson 2005;Andrew et al. 2007). Thus, including or excluding suchpairs in some distance classes because of the surface orprojection distances used may yield different SGS results.Therefore, it might be necessary to take the effects of differenttypes of distances on environmental variations andspatial relationships among individuals into considerationin future fine-scale SGS studies.AcknowledgmentsWe are grateful to Professor Zhongliang Huang andLinfang Wu for assistance in the field and to YongmingYang and Chun Rui for data analysis. We also thank DrJosquin Tibbits for English language improvement. Thiswork was supported by the Knowledge InnovationProgram of the Chinese Academy of Sciences(KSCX2-EW-J-28), the Knowledge Innovation Project ofthe National Key Technology R & D Program(2008BAC39B02), the Chinese Academy of Sciences(KZCX2-YW-430) and the ‘11th Five-Year’ plan onNational Scientific and Technological Support Projects(2008BADB0B05).ReferencesAndrew R. L., Peakall R., Wallis I. R. & Foley W. J. (2007) Spatialdistribution of defense chemicals and markers and the maintenanceof chemical variation. Ecology 88: 716–728.Boyden S., Binkley D. & Stape L. (2008) Competition amongEucalyptus trees depends on genetic variation and resourcesupply. Ecology 89: 2850–2859.Chung M. Y., Chung G. M., Chung M. G. & Epperson B. K. (1998)Spatial genetic structure in populations of Cymbidium goeringii(Orchidaceae). Genes & Genetic Systems 73: 281–285.Condit R. (1998) Tropical Forest Census Plots. Springer, Berlin.Dong L., Wang Z. F., Zhu P. & Ye W. H. (2010) Isolation andcharacterization of microsatellite loci in Castanopsis fissa inlower subtropical China. Silvae Genetica 59: 299–300.Du Y. J. & Huang Z. L. (2008) Effects of seed mass and emergencetime on seedling performance in Castanopsis chinensis. ForestEcology and Management 255: 2495–2501.Epperson B. K. (1992) Spatial structure of genetic-variation withinpopulations of forest trees. New Forests 42: 257–278.Epperson B. K. (2000) Spatial genetic structure and nonequilibriumdemographics within plant populations. PlantSpecies Biology 15: 269–279.Epperson B. K. (2003) Geographical Genetics. Princeton UniversityPress, Princeton.Epperson B. K. (2005) Estimating dispersal from short distancespatial autocorrelation. Heredity 95: 7–15.ESRI (2008) ArcGIS—A Complete Integrated System. EnvironmentalSystems Research Institute, Redlands. [Cited 25 July 2010.]Available from URL: http://www.esri.com/software/arcgisFinn D. S., Theobald D. M., Black W. C. & Poff L. (2006) Spatialpopulation genetic structure and limited dispersal in a RockyMountain alpine stream insect. Molecular Ecology 15: 3553–3566.Hamrick J. L., Murawski D. A. & Nason J. D. (1993) The influenceof seed dispersal mechanisms on the genetic structure oftropical tree populations. Vegetatio 107/108: 281–297.Hardesty B. D., Dick C. W., Kremer A., Hubbell S. & BerminghamE. (2005) Spatial genetic structure of Simarouba amara Aubl.(Simaroubaceae), a dioecious, animal-dispersed Neotropicaltree, on Barro Colorado Island, Panama. Heredity 95: 290–297.Huang G. M., Hong L., Ye W. H., Shen H., Cao H. L. & Xiao W.(2009) Isolation and characterization of polymorphic microsatelliteloci in Castanopsis chinensis Hance (Fagaceae). ConservationGenetics 10: 1069–1071.Jensen J. S., Olrik D. C., Siegismund H. R. & Lowe A. J. (2003)Population genetics and spatial autocorrelation in an unmanagedstand of Quercus petraea in Denmark. ScandinavianJournal of Forest Research 18: 295–304.Kitamoto N., Honjo M., Ueno S., Takenaka A., TsumuraY., Washitani I. & Ohsawa R. (2005) Spatial geneticstructure among and within populations of Primula sieboldiigrowing beside separate streams. Molecular Ecology 14: 149–157.Luna R., Epperson B. K. & Oyama K. (2005) Spatial genetic structureof two sympatric neotropical palms with contrasting lifehistories. Heredity 95: 298–305.Ma K. P. (1999) Ecosystem Diversity in Key Area of China. ZhejiangScience and Technology Press, Hangzhou.Marquardt P. E. & Epperson B. K. (2004) Spatial and populationgenetic structure of microsatellites in white pine. MolecularEcology 13: 3305–3315.Ng K. K. S., Lee S. L., Saw L. G., Plotkin J. B. & Koh C. L. (2006)Spatial structure and genetic diversity of three tropical treespecies with different habitat preferences within a naturalforest. Tree Genetics & Genomes 2: 121–131.Ohsawa T., Tsuda Y., Saito Y., Sawada H. & Ide Y. (2007) Steepslopes promote downhill dispersal of Quercus crispula seedsand weaken the fine-scale genetic structure of seedling populations.Annals of Forest Science 64: 405–412.Peakall R. & Smouse P. E. (2006) GENALEX 6: genetic analysis inExcel. Population genetic software for teaching and research.Molecular Ecology Notes 6: 288–295.Peng S. J. & Xu G. L. (2005) Seed traits of Castanopsis chinensis andits effects on seed predation patterns in Dinghushan BiosphereReserve. Ecology and Environment 14: 493–497.Raymond M. & Rousset F. (1995) GENEPOP (version 1.2): populationgenetics software for exact tests and ecumenicism.Journal of Heredity 86: 248–249.Rousset F. (2008) GENEPOP ‘007: a complete re-implementationof the GENEPOP software for Windows and Linux. MolecularEcology Resources 8: 103–106.Smouse P. E. & Peakall R. (1999) Spatial autocorrelation analysisof individual multiallele and multilocus genetic structure.Heredity 82: 561–573.© 2012 The Authors Plant Species Biology ••, ••–••Journal compilation © 2012 The Society for the Study of Species Biology313

TOPOGRAPHIC EFFECTS ON GENETIC STRUCTURE 7Sokal R. R. & Oden N. L. (1978) Spatial autocorrelation in biology1. Methodology. Biological Journal of the Linnean Society 10:199–228.Sprague D. S. (2000) Topographic effects on spatial data at aJapanese macaque study site. American Journal of Primatology52: 143–147.Troupin D., Nathan R. & Vendramin G. G. (2006) Analysis ofspatial genetic structure in an expanding Pinus halepensispopulation reveals development of fine-scale genetic clusteringover time. Molecular Ecology 15: 3617–3630.Vekemans X. & Hardy O. J. (2004) New insights from fine-scalespatial genetic structure analyses in plant populations.Molecular Ecology 13: 921–935.Wang Z. G., Ye W. H., Cao H. L., Huang Z. L., Lian J. Y., Li L., WeiS. G. & Sun I. F. (2009) Species-topography association in aspecies-rich subtropical forest of China. Basic and AppliedEcology 10: 648–655.Wu T. (2009) Flora of Guangdong (IX). Guangdong Science andTechnology Press, Guangzhou.Ye W. H., Cao H. L., Huang Z. L., Lian J. Y., Wang Z. G., Li L., WeiS. G. & Wang Z. M. (2008) Community structure of a 20 hm 2lower subtropical evergreen broadleaved forest plot in Dinghushan,China. Journal of Plant Ecology 32: 274–286.Plant Species Biology ••, ••–••© 2012 The AuthorsJournal compilation © 2012 The Society for the Study of Species Biology314

G ModelECOCOM-323; No. of Pages 7Ecological Complexity xxx (2011) xxx–xxxContents lists available at SciVerse ScienceDirectEcologicalComplexityjo ur n al ho mep ag e: www .elsevier .c om /lo cate/ec o co mMultifractal analysis of diversity scaling laws in a subtropical forestShi-Guang Wei a , Lin Li a , Zhong-Liang Huang b, *, Wan-Hui Ye b , Gui-Quan Gong b ,Xiao-Yong Zhou c , Ju-Yu Lian ba Guilin University of Electronic Technology, Guilin 541004, PR Chinab South China Botanical Garden, The Chinese Academy of Sciences, Guangzhou, Guangdong 510650, PR Chinac Sun Yat-sen University, Guangzhou 510275, PR ChinaA R T I C L E I N F OArticle history:Received 4 February 2010Received in revised form 29 July 2011Accepted 31 October 2011Available online xxxKeywords:Diversity indicesPlant transectScaleScale effectsSpecies abundance distributionA B S T R A C TUnderstanding pattern of species diversity is a central goal of the science of ecology, and scaling laws areuseful for revealing biodiversity patterns across scales. A transect along an altitudinal gradient inDinghushan Reserve was used to test for the fractal effect in subtropical forest, and the multifractalmethod was used to validate the common scaling law of diversity. The results showed that: (1) richnessabundancepattern has self-similar relations (fractal effect) in the community despite a significantaltitudinal gradient and habitat heterogeneity; (2) the power-law scaling relationship holds for all stratallevels of the forest (trees, shrubs and herbs), and hence scaling laws were significant; and, (3) theShannon index was the optimal descriptor of tree species diversity information, but not shrub or herbdiversity information in this subtropical forest. We also found that diversity indices that correspondingto q > 1 are descriptive of communities dominated by common species. In contrast, diversity indices thatcorresponding to q < 1 are suitable for communities with large numbers of rare species and high speciesevenness. The range of values of q for which scaling laws existed increased with the increasing latitude.ß 2011 Elsevier B.V. All rights reserved.1. IntroductionThe effect of scale on patterns of species diversity has long beenof interest in ecology (Arrhenius, 1921; Bormann, 1953; Greig-Smith, 1952; Imagawa et al., 1966; Tillyard, 1914), and has beenthe focus of many studies over the past two decades (Dungan et al.,2002; Jelinski and Wu, 1996; Legendre and Fortin, 1989; Petersonand Parker, 1998; Turner et al., 1989; Wiens, 1989). This recentincrease in attention has resulted from the increasing focus ofbiodiversity researchers on the mechanisms maintaining diversity.Species are heterogeneously distributed across landscapes andare often organized within distinct communities (Gaston, 1994b,2000) over several scales (Auerbach and Shmida, 1987; Lyons andWillig, 1999). Species diversity is usually measured for a given area(and, hence, scale), and thus depends on the sampling area. Thereare no simple and reliable rules to compare studies conducted atdifferent scales (He et al., 2002). Therefore, developing a universalscaling law of biodiversity is one of the greatest challenges toecologists (Hubbell, 2001). Such a law could be useful forunderstanding patterns of species richness, species extinction* Corresponding author at: Forest Ecosystem Research Station, South ChinaBotanical Garden, The Chinese Academy of Sciences, DingHuShan Nature Reserve,ZhaoQin, Guangdong 526070, PR China. Tel.: +86 0758 262 1169.E-mail address: Huangzl@scbg.ac.cn (Z.-L. Huang).probabilities, species coexistence, resource-partitioning processes,and for reserve design (Gaston, 1994a).As the area sampled increases in size, both species richness andabundance increase monotonously (Ye et al., 1998), and thenumber of species encountered is proportional to a power of thearea sampled (SAR) (Macarthur and Wilson, 1967):SðAÞ ¼ cA Dwhere S(A) is species richness in a given area A; c is a constant; andD is the scaling exponent or fractal dimension (Li, 2000a). Thispower-law revealed to researchers the fractal nature of ecologicalcommunities, and is considered one of the few robust laws ofecology (Borda-de-Agua et al., 2002; Harte et al., 1999; Ostlinget al., 2003, 2004). The power-law has been used to characterizecommunity structure, estimate species richness, measure disturbanceeffects, determine the appropriate size of natural reserves inconservation biology (He and Legendre, 1996) and estimate speciesextinction rates (May et al., 1995; Pimm et al., 1995).Species richness is one of the diversity metrics that can berelated to each other using Rényi’s generalized entropy function(Hill, 1973; Pielou, 1975; Rényi, 1970):H q ¼ log P si¼1 pq i1 qwhere q is an integer and p i is the relative abundance. Thelogarithm of most classical diversity metrics are special cases of H q .1476-945X/$ – see front matter ß 2011 Elsevier B.V. All rights reserved.doi:10.1016/j.ecocom.2011.10.004315Please cite this article in press as: Wei, S.-G., et al., Multifractal analysis of diversity scaling laws in a subtropical forest. Ecol. Complex.(2011), doi:10.1016/j.ecocom.2011.10.004

G ModelECOCOM-323; No. of Pages 72S.-G. Wei et al. / Ecological Complexity xxx (2011) xxx–xxxFor example, the logarithm of species richness (S(A)), the Shannonand Inverse Simpson indices are H q at q = 0, q = 1 and q = 2,respectively. Thus, the SAR can be generalized as follows:expðH q ðAÞÞ ¼ C q A Dqwhere exp(H q (A)) is the species richness (or other diversity index),C q is a constant, and D q is the scaling exponent or fractal dimension.A is the sample area. This function represents an ecological scalinglaw. If scaling relationships are examined over the range of q, thenthe analysis of the SAR becomes multifractal. When a fractal isobserved, typically a more general class of self-similar relationscalled ‘‘multifractals’’ also exists (Mandelbrot and Evertsz, 1995).In contrast to a fractal object, or signal, where a single number or its‘‘dimension’’ is sufficient for its complete characterization,characterization of a multifractal requires an infinite number ofindices, frequently called ‘‘a spectrum’’. Multifractals are mainlyprobabilistic (Borda-de-Agua et al., 2002). Though fractal theoryhas existed for some time, multifractal analysis has seldom beenapplied to analyze cross-scale biodiversity patterns in subtropicalforests.In this study, a plant transect over an elevational gradient in theDinghushan Mountains was used to test for the fractal effect in asubtropical forest, and a multifractal method was used to validatethe common scaling law of diversity. The study objectives were to:(1) test whether or not the power law exists in the biodiversitypatterns of subtropical forest; (2) test whether or not fractal effectscan be found across several spatial scales; and (3) determine acriterion for choosing the optimal diversity index for differentforest communities. Various factors that may influence thesignificance of the scaling law were also examined, in order tofind the relationship between latitude and the values of q for whichthe scaling law is significant. This analysis contributes to ourunderstanding of how species are distributed, and also how speciesdiversity in subtropical forests.2. Methods2.1. Study areaThe study site was located in the Dinghushan Mountains(112830 0 39 00 –112833 0 41 00 E, 23809 0 21 00 –23811 0 30 00 N) in GuangdongProvince. Dinghushan was the first Nature Reserve established inChina (in 1956), and it has played a significant role in theconservation of forest ecosystems over the past 50 years. Thereserve comprises low mountains and hilly landscapes. Its totalarea is 1155 ha, with an elevational range of 14.1–1000.3 m (abovesea level), and it is composed of tropical-subtropical forest.Dinghushan has a southern subtropical monsoon climate with amean annual temperature of 20.9 8C, and mean monthly temperaturesof 12.6 8C in January and 28.0 8C in July. Average annualprecipitation is 1929 mm, with most precipitation occurringbetween April and September. Annual evaporation is 1115 mmand relative humidity 82 percent.2.2. Sampling methodsA 10 1160 m transect was established from the foot to the topof Sanbaofeng hill in Dinghushan in 2003. The transect covered analtitudinal range from 50.2 m to 476.5 m, and included monsoonevergreen broad-leaf forest, coniferous and broad-leaf mixedforest. The transect was subdivided into 232.5 m 10 m quadratsfor convenience in taking field surveys. The survey consisted ofenumerating all free standing trees and shrubs at least 1 cm indiameter at breast height (DBH), locating each tree usinggeographic coordinates on a reference map, and identifying it tospecies. Within each quadrat, a 2 m 5 m sub-quadrats wasselected to survey shrubs and herbs. For the shrubs, individualswith a DBH of less than 1.0 cm were counted, their heightsmeasured, and they were identified to species. The percent cover ofeach species was also estimated. For the herbs, the percent cover,individual heights, and number of individuals were recorded foreach herbaceous species. Treat tree layer, shrub layer, and herblayer separately in order to be able to compare our results to thoseof other studies, some of which only worked with trees species.2.3. Multifractal analysis2.3.1. Theoretical backgroundIn multifractal analysis, Rényi dimensions (Hentschel andProcaccia, 1983; Rényi, 1970) are defined as:D q ¼ lim d ! 0 log½ P nðdÞi¼1 u iðdÞ q Šðq 1ÞlogðdÞwhere q is any real integer other than 1. When q = 1, D q becomes(Loehle and Li, 1996):D 1 ¼lim d ! 0X nðdÞi¼1½u i ðdÞlogðu i ðdÞÞŠlogðdÞD 1 is called the entropy dimension or information dimension of thedistribution (Hentschel and Procaccia, 1983; Rényi, 1970), and itquantifies the rate of growth of entropy with respect to d. Thisexpression can be considered as a u-weighted average of thesingularity strength values. It is also related to the size (dimensions)of the minimal set, where the whole measure isconcentrated (Evertsz and Mandelbrot, 1992). The dimension D 0is called the capacity dimension and D 2 is the correlation dimension(Peitgen et al., 2004). The greater the variation of D q with respect toq, the higher the degree of heterogeneity of the measure. When anadequate scaling behavior takes place for an experimentalmeasure, the spectrum of Rényi dimensions, D q , provides avaluable characterization of the singular behavior of the measureand the respective interpretation within each context (Harte,2001).Taking the negative of the numerator of Eq. (2) gives (Jumarie,2000; Li, 2000a):HðdÞ ¼X nðdÞi¼1(1)(2)½u i ðdÞlogðu i ðdÞÞŠ (3)This is the Shannon entropy (Loehle and Li, 1996; Shannon andWeaver, 1948), a measure of the heterogeneity or unevenness of aforest community. u i (d) is the probability of observing a species inthe ith cell using samples of d units in size (Chen et al., 2005)2.3.2. Rényi dimensions calculationFor the sampling transect data, the original quadrat data werepooled to create tree, shrub and herb data sets for samples ofdifferent areas (called ‘‘cell size’’), d: 10 10, 20 10, 40 10,145 10, 290 10, 580 10 and 1160 10 m. Hence, there werea total of 116, 58, 29, 8, 4, 2 and 1 replicate(s) or cell(s) (n(d)), foreach sample area listed above, respectively. Thus, species j in cell iof cell size d has a relative abundance value of u ij (d) (j = 1, 2, 3, . . . N i ,where N i was the number of species in cell i).The relativeabundance u ij (d) is defined as(Alatalo, 1981):m i ðdÞ ¼ h iðdÞþ c i jðdÞþ a i jðdÞ(4)3th i 3tc i 3ta iwhere h ij (d), c ij (d) and a ij (d) are the sum of the height, percent coverand abundance of species j in cell I, with cell size d(scale); th i , tc i ,Please cite this article in press as: Wei, S.-G., et al., Multifractal analysis of diversity scaling laws in a subtropical forest. Ecol. Complex.316(2011), doi:10.1016/j.ecocom.2011.10.004

G ModelECOCOM-323; No. of Pages 7S.-G. Wei et al. / Ecological Complexity xxx (2011) xxx–xxx 3and ta i are the total height, percent cover and abundance of allspecies in cell i, respectively.Rényi dimensions spectra D q were computed for values10 q 10 with a step of 1, using the formula (Montero, 2005):D q log½P nðdÞi¼1 u iðdÞ q Šðq 1ÞlogðdÞEq. (5) is an approximation of Eq. (1). Regression analysis was usedto examine the relationship between the numerator and thedenominator of Eq. (5) for each q. Next, the slope of the regressionwas obtained, that was the value of D q . If the log–log plot describedin the regression analysis is a straight line over a certain range ofvalues of q (Li, 2000b), then the scaling laws were significantly for agiven ecological community. The goodness-of-fit of Eq. (5) wasquantified using the adjusted coefficient of determination, R 2 a (Heand Legendre, 1996).In order to calculate D q more easily, Equation (Zhang et al.,2006):I q ðdÞ ¼ 1q2P nðdÞ1 log 4i¼1nðdÞP Nij¼1 uq i(5)3j ðdÞ 5 (6)was derived from Eqs. (4) and (5) in order to calculate the averagevalue of the Rényi information for all cells with size d (Borda-de-Agua et al., 2002). In this equation, N i is the species richness of cell i,and n(d) is the number of cells with cell size d.According to Montero (2005), when q = 1, Eq. (5) changes to:Species richnessSpecies richnessTree layerabundance classShrub layerD 1 X nðdÞi¼1½ u i ðdÞlogðu i ðdÞÞŠlogðdÞEq. (7) is an approximation of Eq. (2).Similarly, at q = 1, Eq. (6) becomes:(7)abundance classHerb layerI 1 ðdÞ ¼3. ResultsP nðdÞ P Nii¼1 j¼1 u i jðdÞlog½u i jðdÞŠnðdÞFig. 1 presents the species-abundance class curves for trees,herbs and shrubs for four representative sampling cell sizes (d):d = 20 10 m, 145 10 m, 580 10 m, and 1160 10 m. Thespecies-abundance curves have similar shapes with increasingarea (cell size) in both the tree and the shrub data. In the herb data,no obvious similarity among scales was evident. In all data sets,there were a large number of rare species, with common speciesfewer in number at all scales.While the relationship of species richness with scale wassimilar in all 3 data sets (Fig. 2), the fractal effects occurred ondifferent scales. But only at larger sampling scales could see thedistinct trend along with the elevation. Species richness of threelayers reached a maximum from the position about 600–800 m.Fig. 3 presents classical species–area curves (q = 0) for differentdata sets. In the tree and shrub curves, species richness increasedmore rapidly with area, and attained a greater maximum valuethan in the herb curve. However, herb species richness did increaseslowly with area. Both trees and shrubs had a greater maximumdiversity than herbs. There were 93 species of trees, 91 species ofshrubs, and 31 species of herbs.The results showed that most linear regressions representedgood fits to the data, with high R 2 a and low p-values (most R 2 a > 0:85and ***p < 0.001) for all three data sets (Fig. 4). As the regressionswere generally significant, scaling laws are common in oursubtropical community. All nine regressions showed a good linear(8)Species richnessabundance classFig. 1. Examples of abundance class curves for tree, shrub, and herb species insamples of different areas d (that is the cell size in this article). The abundance classone includes 0–28 individual, while class two includes 2 1 to 2 2 1 individuals, andso on, until class n: [2 n 1 , 2 n 1]. * – d = 20 10 m; & – d = 145 10 m; ^ –d = 580 10 m; ~– d = 1160 10 m.fit (R 2 a > 0:9, ***p < 0.001), with the possible exception of the q = 1regression in the shrub layer (p = 0.0019, R 2 a ¼ 0:85). The dimensionvalues were D 1 /D 0 = 0.4022 (trees), D 1 /D 0 = 0.6148 (shrubs),and D 1 /D 0 = 0.5502 (herbs), indicating that the shrub communitieswere more homogeneous than the herb or tree communities.The multifractal Rényi spectra D q q curve for trees, shrubs,and herbs are shown in Fig. 5. The Rényi spectra showed the samesigmoidal shape for the different groups. The range of D q was0.0943 (q = 3) to 0.8313 (q = 10) for the trees. The maximumvalue of D q was 0.8111 (q = 10) for the shrubs, and the minimumvalue was 0.1794 (q = 4). For the herbs, the maximum value of D qwas 0.8428 (q = 10), and the minimum value was 0.1494 (q = 3).Please cite this article in press as: Wei, S.-G., et al., Multifractal analysis of diversity scaling laws in a subtropical forest. Ecol. Complex.317(2011), doi:10.1016/j.ecocom.2011.10.004

G ModelECOCOM-323; No. of Pages 74S.-G. Wei et al. / Ecological Complexity xxx (2011) xxx–xxxTree layerSpecies richnessSpecies richnessPosition (m)Shrub layerArea (100m 2 )Fig. 3. Species–area curves of tree, shrub and herb species. The SR value on thefigure is the mean value of SR on the corresponding scale. ^ – tree layer; * – ~ shrublayer; ~ – herb layer.Species richnessSpecies richness4. DiscussionPosition (m)Herb layerPosition (m)Fig. 2. Examples of variation in species richness along the transect for tree, shrub,and herb species. The position represents the distance between the mid-point of thecell plot and the start point (origin). * – d = 20 10 m; ~ – d = 40 10 m; & –d = 145 10 m; ^ – d = 290 10 m.Robust scaling relationships existed in the species abundancedistributions (Fig. 4 straight lines) in the forest. Scaling laws werecommonplace in the subtropical forest analyzed in this study. Thisstudy demonstrated the existence of scaling laws not only onShannon diversity indices (Fig. 4 q = 1), but also on other higherorderRényi entropy dimensions (other diversity indices).For the tree species, as the p-value (significance probability)was smallest when q = 1, the Shannon diversity index bestdescribes the diversity information in the subtropical forest. Bothrare and common species contribute relatively equally to totalspecies diversity, as q = 1 is in the middle of the range of values of q( 10 to 10).For the shrub species, scaling laws were most significant whenq = 3. When q is small, rare species contribute relatively more tospecies diversity than common species; the situation is reversedfor large values of q. Hence, for shrub species diversity information,rare species contribute slightly more than common species.However, some analyses (Patil and Taillie, 1982) have suggestedthat negative values of q in generalized entropies, such as the Rényientropy, are not meaningful in ecology (though they may bemeaningful in physics or some other fields). Therefore, q shouldhave a minimum value of nonnegative number, and we shouldselect a species richness index corresponding to q = 0 as the bestdescriptor of shrub diversity in a subtropical forest.For herb species, the R 2 a of the regression was maximized andthe p-value minimized when q = 10, indicating that rare speciescontributed much more than common species to diversityinformation of herb. Also, as q < 0 are not meaningful in ecology,we should select a species richness index corresponding to q = 0 asthe best descriptor of herb diversity in the subtropical forest. Theseresults suggest that ecological communities with differentstructures may require different diversity indices in order to bestdescribe their structure; species richness, or the Shannon andSimpson diversity indexes were not always effective as diversityindices. Therefore, Rényi fractal analyses could provide anobjective criterion to determine the most suitable diversity indexfor a given community.In Eq. (6), the parameter q acts as a scanning tool for scrutinizingsmaller and larger values of the measure u i (d based on supportfrom the data (Kravchenko et al., 1999). For q 1, regions with ahigh degree of concentration are amplified, while regions with asmall degree of concentration are magnified for q 1 (Montero,2005). In this study, as q increased common species contributedmore to the diversity of the community, and the high-order Rényientropy dimensions were weighted towards common species,while rare species were ignored. As there were many rare speciesin the study transect, higher-order Rényi entropy dimensions(q > 1) were not optimal for describing patterns of diversity.In the tree data set, D q began to increase when q > 2 (Fig. 5), incontrast to the theoretical prediction that D q should be amonotonically decreasing function of q (Grassberger, 1983). Thisincrease of D q with q occurred for q > 2 in the trees and herbs, andfor q > 3 in the shrubs. One possible explanation is that whilePlease cite this article in press as: Wei, S.-G., et al., Multifractal analysis of diversity scaling laws in a subtropical forest. Ecol. Complex.318(2011), doi:10.1016/j.ecocom.2011.10.004

G ModelECOCOM-323; No. of Pages 7S.-G. Wei et al. / Ecological Complexity xxx (2011) xxx–xxx 5Fig. 4. The lines with points are plots of Rényi information I q vs. log(cell size) for q = 0, q = 1, q = 2 of 3 forest layers and the continuous lines without points plots of I q againstlog(cell size) for q = 10 to 10.common species did not contribute greatly to species diversity inthe community, rare species did.In previous studies, D q has generally been found to be amonotonically decreasing function of q (Peitgen et al., 2004).However, in a study conducted on Barro Colorado Island (BCI), D qwas not strictly a decreasing function of q (Borda-de-Agua et al.,2002). The BCI study is intriguing as z q increased when q > 1, incontrast to theoretical predictions. This increase may have been anartefact of the small size of the data set; if further data, from largerareas, was added the increase observed in D q (for positive values ofq) could have been reversed. Our results, from a subtropical forest,also showed an increase in D q with q > 2. Our interpretation of theincrease in D q was that communities might vary in their structuresand species evenness across climatic zones, and so scaling lawscould exist for different ranges of q-values in different communities.The increase in D q (e.g. at q > 2) indicates that the scaling lawswere not significant for this range of parameter space. Scaling lawsare not significant when q > 1 in tropical forest, when q > 2 orq > 3 in subtropical forest, and when q > 4 in temperate forest(Borda-de-Agua et al., 2002; Zhang et al., 2006).From the above arguments, we concluded that the range ofvalues of q for which scaling laws are significant increases withincreasing latitude, the reason being that evenness decreases withincreasing latitude. Common species contributed relatively moreto biodiversity information at higher versus lower latitudes. Incommunities with large numbers of common species, diversityindices that corresponding to q > 1 are therefore needed toadequately describe community structure. In contrast, communitieswith large numbers of rare species and high evenness are bestdescribed using diversity indices that corresponding to q < 1.Whether or not this rule is meaningful across study systemrequires further research in additional forest communities.Please cite this article in press as: Wei, S.-G., et al., Multifractal analysis of diversity scaling laws in a subtropical forest. Ecol. Complex.319(2011), doi:10.1016/j.ecocom.2011.10.004

G ModelECOCOM-323; No. of Pages 76S.-G. Wei et al. / Ecological Complexity xxx (2011) xxx–xxxseed dispersal and soil conditions on the transect that affect seedgermination, tree growth and survival.AcknowledgementsThe authors acknowledge the National Natural ScienceFoundation of China (General Program: 30970544), National KeyTechnology R&D Program (2008BAC39B02), Research Project ofGuangXi education department (201010LX148), and the ChineseForest Biodiversity Monitoring Network for technical support ofthis research work. The Biodiversity Committee and the Bureau ofScience and Technology for Resources and Environment of theChinese Academy of Sciences provided financial assistance. Wethank Dr. Zhang Ming Wang for a careful review and valuablesuggestions on this article. We would also like to thank Ms. Emilyat the University of British Columbia for her assistance withEnglish language and grammatical editing of the manuscript andthanks YanJun Du’s help. We thank Dr. Ke Ming Ma for the projectidea. We also thank Hong Lin Cao for field help, and speciesidentifications.ReferencesFig. 5. Multifractal Rényi spectra D q –q curves of tree, shrub, and herb species. Eachpoint on these graphs is determined from a linear fitting as illustrated in Fig. 4.In conclusion, we have shown that species-abundance distributionsare self similar (fractal effects) across our studycommunity, which spanned a significant altitudinal gradientand heterogeneous habitat types. Power-laws existed for tree,shrub and herb species at all spatial scales, and scaling laws weresignificant in our study communities. The Shannon index was theoptimal descriptor of diversity information in tree but not shrub orherb species in our subtropical forest. Diversity indices correspondingto larger q (e.g. q > 1) were descriptive of communitiesdominated by common species. However, diversity indicescorresponding to smaller values of q (e.g. q < 1) were suitablefor communities with large numbers of rare species and a highdegree of species evenness. The range of values of q for whichscaling laws were significant increased with latitude. In order tofully understand the mechanisms generating the patterns observedin the scaling laws, we are collecting data on factors such asAlatalo, R.V., 1981. Problems in the measurement of evenness in ecology. Oikos 37,199–204.Arrhenius, O., 1921. Species and area. J. Ecol. 9, 95–99.Auerbach, M., Shmida, A., 1987. Spatial scale and the determinants of plant speciesrichness. Trends Ecol. Evol. 2, 238–242.Borda-de-Agua, L., Hubbell, S.P., McAllister, M., 2002. Species–area curves, diversityindices, and species abundance distributions: a multifractal analysis. Am. Nat.159, 138–155.Bormann, F.H., 1953. The statistical efficiency of sample plot size and shape in forestecology. Ecology 34, 474–487.Chen, X., Li, B.L., Collins, S.L., 2005. Multiscale monitoring of a multispecies casestudy: two grass species at Sevilleta. Plant Ecol. 179, 149–154.Dungan, J.L., Perry, J.N., Dale, M.R.T., Legendre, P., Citron-Pousty, S., Fortin, M.J.,Jakomulska, A., Miriti, M., Rosenberg, M.S., 2002. A balanced view of scale inspatial statistical analysis. Ecography 25, 626–640.Evertsz, C.J.G., Mandelbrot, B.B., 1992. Multifractal measures. In: Peitgen, H.O.,Jürgens, H., Saupe, D. (Eds.), Chaos and Fractals: New Frontiers in Science.Springer-Verlag, Berlin, pp. 849–881.Gaston, K.J., 1994a. Rarity. Chapman & Hall, London.Gaston, K.J., 1994b. Scaling the heights of species diversity. Biodivers. Lett. 2,125–126.Gaston, K.J., 2000. Global patterns in biodiversity. Nature 405, 220–227.Grassberger, P., 1983. Generalized dimensions of strange attractors. Phys. Lett. 97,227–230.Greig-Smith, P., 1952. The use of random and contiguous quadrats in the study ofthe structure of plant communities. Ann. Bot. 16, 293–316.Harte, D., 2001. Multifractals: Theory and Applications. Chapman & Hall, London.Harte, J., Kinzig, A., Green, J., 1999. Self-similarity in the distribution and abundanceof species. Science 284, 334–336.He, F.L., LaFrankie, J.V., Song, B., 2002. Scale dependence of tree abundanceand richness in a tropical rain forest, Malaysia. Landscape Ecol. 17,559–568.He, F.L., Legendre, P., 1996. On species–area relations. Am. Nat. 148, 719–737.Hentschel, H.G.E., Procaccia, I., 1983. The infinite number of generalized dimensionsof fractals and strange attractors. Physica D 8, 435–444.Hill, M.O., 1973. Diversity and evenness: a unifying notation and its consequences.Ecology 54, 427–432.Hubbell, S.P., 2001. The Unified Neutral Theory of Biodiversity and Biogeography.Princeton University Press, Princeton.Imagawa, Y., Nakamura, J., Matsue, I., Sueda, T., Hara, K., 1966. Biological investigationon treatment effect of scaling and brushing upon periodontal diseases.Parodontol 20, 1–9.Jelinski, D.E., Wu, J., 1996. The modifiable areal unit problem and implications forlandscape ecology. Landscape Ecol. 11, 129–140.Jumarie, G., 2000. Maximum Entropy, Information Without Probability and ComplexFractals: Classical and Quantum Approach. Springer-Verlag, Berlin.Kravchenko, A.N., Boast, C.W., Bullock, D.G., 1999. Multifractal analysis of soilspatial variability. Agron. J. 91, 1033–1041.Legendre, P., Fortin, M.J., 1989. Spatial pattern and ecological analysis. Plant Ecol. 80,107–138.Li, B.L., 2000a. Fractal geometry applications in description and analysis of patchpatterns and patch dynamics. Ecol. Model. 132, 33–50.Li, B.L., 2000b. Why is the holistic approach becoming so important in landscapeecology? Landscape Urban Plan. 50, 27–41.Loehle, C., Li, B.L., 1996. Statistical properties of ecological and geologic fractals.Ecol. Model. 85, 271–284.Please cite this article in press as: Wei, S.-G., et al., Multifractal analysis of diversity scaling laws in a subtropical forest. Ecol. Complex.320(2011), doi:10.1016/j.ecocom.2011.10.004

G ModelECOCOM-323; No. of Pages 7S.-G. Wei et al. / Ecological Complexity xxx (2011) xxx–xxx 7Lyons, S.K., Willig, M.R., 1999. A hemispheric assessment of scale dependence inlatitudinal gradients of species richness. Ecology 80, 2483–2491.Macarthur, R., Wilson, E.O., 1967. Island Biogeography. Princeton University Press,Princeton, NJ.Mandelbrot, B.B., Evertsz, C.J.G., 1995. Exactly self-similar left-sided multifractals.In: Bunde, A., Havlin, S. (Eds.), Fractals and Disordered Systems Table ofContents. Springer-Verlag, Berlin, pp. 367–399.May, R.M., Lawton, J.H., Stork, N.E., 1995. Assessing extinction rates. In: Lawton,J.H., May, R.M. (Eds.), Extinction Rates. Oxford University Press, Oxford, pp.1–24.Montero, E., 2005. Renyi dimensions analysis of soil particle-size distributions. Ecol.Model. 182, 305–315.Ostling, A., Harte, J., Green, J.L., Kinzig, A.P., 2003. A community-level fractalproperty produces power-law species–area relationships. Oikos 103, 218–224.Ostling, A., Harte, J., Green, J.L., Kinzig, A.P., 2004. Self-similarity, the power law formof the species–area relationship, and a probability rule: a reply to maddux. Am.Nat. 163, 627–633.Patil, G.P., Taillie, C., 1982. Diversity as a concept and its measurement. J. Am. Stat.Assoc. 77, 548–561.Peitgen, H.O., Jurgens, H., Saupe, D., 2004. Chaos and Fractals. Springer-Verlag,Berlin.Peterson, D., Parker, V.T., 1998. Scale Issues in Ecology. Columbia University Press,New York.Pielou, E.C., 1975. Ecological Diversity. Wiley-Interscience Publication, New York.Pimm, S.L., Russell, G.J., Gittleman, J.L., Brooks, T.M., 1995. The Future of Biodiversity.Science 269, 347.Rényi, A., 1970. Probability Theory. North-Holland Pub. Co., Amsterdam.Shannon, C.E., Weaver, W., 1948. A mathematical theory of communications. BellSyst. Tech. J. 27, 632–656.Tillyard, R.J., 1914. On the study of zoogeographical regions by means of specificcontours, with application to the Odanata of Australia. Proc. Linn. Soc. N. S. W.39, 21–43.Turner, M.G., O’Neill, R.V., Gardner, R.H., Milne, B.T., 1989. Effects of changingspatial scale on the analysis of landscape pattern. Landscape Ecol. 3,153–162.Wiens, J.A., 1989. Spatial scaling in ecology. Funct. Ecol. 3, 385–397.Ye, W.H., Ma, K.P., Ma, K.M., Sang, W.G., Gao, X.M., 1998. Studies on plant communitydiversity in Donlingshan mountain, Beijing, China, IX. The influence of scaleon adiversity. Acta Ecol. Sin. 18, 10–14 (in Chinese).Zhang, Y.X., Ma, K.M., Anand, M., Fu, B.J., 2006. Do generalized scaling laws exist forspecies abundance distribution in mountains? Oikos 115, 81–88.Please cite this article in press as: Wei, S.-G., et al., Multifractal analysis of diversity scaling laws in a subtropical forest. Ecol. Complex.321(2011), doi:10.1016/j.ecocom.2011.10.004

Eur J Forest Res (2012) 131:453–461DOI 10.1007/s10342-011-0519-zORIGINAL PAPERSeedling recruitment patterns in a 20 ha subtropical forest plot:hints for niche-based processes and negative density dependenceYue Bin • Guojun Lin • Buhang Li •Linfang Wu • Yong Shen • Wanhui YeReceived: 17 August 2010 / Revised: 8 April 2011 / Accepted: 18 April 2011 / Published online: 11 May 2011Ó Springer-Verlag 2011Abstract Seedling recruitment can be influenced by seeddispersal, conspecific density dependence, and environmentalfactors. These forces are variant in space. In thisstudy, seedling recruitment was investigated by the inversemodeling method. The inverse modeling framework herewas made up of two components: a conspecific effect and adeclining function. Power functions (P) and constant conspecific(C) effects were tried. Two types of decliningfunctions were tried: isotropic (I) and anisotropic (A).Thus, the combination of conspecific effect and decliningfunction generated four candidate models: PI, PA, CI, CA.These four models were used to study the seedlingrecruitment of 13 species in a 20 ha forest plot in subtropicalChina. It was found that PI, PA, CI, CA are thebest models for two, three, five, and three species,respectively. Negative exponents in P were found in threespecies, which may indicate negative density-dependentmortality. Among those species that supported an anisotropiccomponent, all moderately shade-tolerant and shadetolerantspecies except Calophyllum membranaceum hadCommunicated by C. Ammer.Y. Bin G. Lin L. Wu Y. Shen W. Ye (&)Key Laboratory of Plant Resources Conservationand Sustainable Utilization, South China Botanical Garden,Chinese Academy of Sciences, 510650 Guangzhou, Chinae-mail: why@scbg.ac.cnY. Bin G. Lin B. Li Y. ShenGraduate University of Chinese Academy of Sciences,10049 Beijing, ChinaB. LiDepartment of Ecology, School of Life Science/State KeyLaboratory of Biocontrol, Sun Yat-sen University,510275 Guangzhou, Chinahigher possibilities of successful recruitment if their altitudeswere relatively low, consistent with their ecologicalniches. The shade intolerant species, Castanopsis fissaproduces seeds weighing 6–250 times more than otherspecies. Yet, its seedling recruitment was more successfulat higher altitudes, which again was consistent with itsecological niche. Our research indicated that it is necessaryto take anisotropic forces into account when investigatingseed dispersal and seedling recruitment in regions withcomplex topography, and that the niche-based processesand density-dependent mortality at least play some part inconstructing the seedling distribution pattern.Keywords Conspecific effect Inverse modeling Density dependence Forest dynamic plot NicheIntroductionSeedling recruitment is a bottleneck in tree establishment(Queenborough et al. 2007). The spatial pattern of seedlingrecruitment influences the long-term distribution patternsof species (Queenborough et al. 2007), and can have significanteffects on the composition and abundance of plantcommunities (Leak and Graber 1976). Furthermore,recruitment limitation has been theoretically demonstratedto be a mechanism promoting species coexistence andcommunity diversity maintenance (Abrams 1984; Pacalaet al. 1996). Therefore, factors that influence seedlingrecruitment are of great importance to forest ecologists andresearchers.The spatial pattern of seedling recruitment is primarilyconstrained by seed dispersal in the forest. There have beenongoing interests in long distance seed dispersal (e.g.,Bullock and Clarke 2000; Bohrer et al. 2005), which is123322

454 Eur J Forest Res (2012) 131:453–461possibly related to metapopulation survival and the maintenanceof genetic variability (Bohrer et al. 2005). However,given that only a small fraction of seeds are dispersedrelatively far away from the mother plant (Nathan andMuller–Landau 2000), the major interest of plant ecologistsremains in short-distance dispersal. Limited seed dispersalhas been cited as a mechanism for species coexistence inspecies-rich communities (Hubbell 2001; Tuomisto et al.2003; Harms et al. 2001; Valencia et al. 2004). Scientistshave found some indirect support for limited seed dispersal:for example, Turnbull et al. (2000) reported thatseedling recruitment increased in response to seed additionfor about half of the species tested. Several field studiesshowed that the composition of seedlings in tropical forestcanopy gaps closely reflected that of the adult communitiesaround them (Dalling et al. 1998; Hubbell et al. 1999).Since Ribbens et al. (1994) proposed the inverse modelingmethod, seed and seedling patterns have been studied asthe inverse modeling approach has developed (LePageet al. 2000; Uriarte et al. 2005; Muller–Landau et al. 2008;Schurr et al. 2008). For example, LePage et al. (2000)incorporated habitat information into the inverse modelingframework and studied the seedling survivorship in space.Uriarte et al. (2005) considered the effects of densitydependentmortality on the seedling survival pattern.Various dispersal functions have been used, such as twoparameterWeibull function (e.g., Ribbens et al. 1994;LePage et al. 2000), lognormal (e.g., Uriarte et al. 2005)and 2Dt (e.g. Schurr et al. 2008). Some ecologists have alsocompared the performance of various dispersal functions(Greene et al. 2004; Uriarte et al. 2005; Schurr et al. 2008),and it turns out that no dispersal function is consistentlysuperior to the others. Though the dispersal kernelsinvolved vary in these researches, they all agree with ageneral pattern that the probability that a seed arrives atspecific position declines with increasing distance from thematernal tree.Most research concerning inverse modeling assumedthat dispersal possibility decreases as distance increases, inthe same manner in all directions. This isotropic assumptionis often violated in many forests (Wagner et al. 2004;Wälder et al. 2009). In regions with slopes and valleys,seed dispersal can be extensively influenced by gravity andmountain–valley breezes. These forces, however, are notisotropic in space. For example, gravity takes seeds andfruits to the earth. If the velocity of seeds and fruits is notzero after hitting the ground, gravity is more likely to takethem downwards to the valley than up to the ridge. This is asource of anisotropic dispersal. Wind-dispersed species canexhibit anisotropic dispersal patterns due to prevailingwind directions; in mountain regions, mountain–valleybreezes, the up-slope and down-slope winds, or a combinationof both may occur because of solar radiation (Stull1989). Different strength of mountain–valley breezes mayalso give rise to anisotropic distribution of seeds and fruits,especially for light seeds or seeds with wings. Consideringsuch anisotropic behavior can lead to more realistic models(Näther and Wälder 2003; Wagner et al. 2004).Seedling recruitment is then influenced by the survivorshipin space. Conspecifics frequently are considered tonegatively influence seedling survival. Such density dependenceeffects are generally supposed to result from theenhanced propagation of pests and species-specific herbivoresand pathogens as conspecific biomass nearby becomeslarge (Janzen 1970; Connell 1971). Seedling recruitment ismeanwhile influenced by a variety of microclimatic andedaphic factors (Augspurger 1984; Scholl and Taylor 2006).Such factors are often not isotropic in space.To understand how the above factors influence seedlingrecruitment, we use data from seedling stations scattered ina 20 ha plot to analyze anisotropic forces and the effectsconspecifics have on seedling recruitment. Specifically,this study was conducted with the inverse modelingapproach to answer three main questions: (1) How isseedling recruitment influenced by anisotropic forces? (2)How does the species’ ecological status influence seedlingrecruitment? (3) What is the role conspecifics play inseedling recruitment?Materials and methodsStudy siteDinghushan (112°30 0 39 00 –112°33 0 41 00 E, 23°09 0 21 00 –23°11 030 00 N) is located in Guangdong province, China. TheReserve covers an area of 1,155 ha, with low mountainsand hills. This region is characterized by a south subtropicalmonsoon climate, with a mean annual temperature of20.9°C and a mean annual precipitation of 1,929 mm.A permanent 20 ha plot was established in DinghushanNature Reserve (the DHS plot) in November, 2004. All thetrees with diameter at breast height (DBH) larger than1 cm were tagged, mapped, measured for DBH and identifiedto species. The DHS plot is characterized by roughterrain with deep valleys, and the elevation ranges from240 to 470 m (Wang et al. 2009). The plot was divided into500 20 9 20 m subplots for mapping the trees. The altitudesat the four corners of every contiguous 20 9 20 msubplot were measured with a theodolite and cement pileswith marks were erected on the corners. In the plot, thereare 210 species and 71,617 individuals mapped, falling into56 families and 119 genera (Ye et al. 2008). The altitudesof all the recorded individuals were obtained by theinterpolation function in R 2.9.0 (R development core team2008).123323

Eur J Forest Res (2012) 131:453–461 455DataIn November 2007, 149 seed-seedling stations, each with one0.7 9 0.7 m seed trap and three 1 9 1 m seedling quadrates,were set up along the trails but at least 7 m away in theplot to obtain data for long-term seed production and seedlingrecruitment (Fig. 1). We believe that trail dynamics doesnot have much influence on the demographic processes of themonitored seedlings because the trails are so narrow that itjust allows a single person to walk through, and the branchesof the trees along the trails can touch one’s shoulders whenwalking along. The stations were censused in March, 2008and recensused every 3 months. Seedling is defined as thewoody plant individuals with less than 1 cm DBH, the sameas in Chen et al.’s research (2010).Because these 149 stations were highly aggregated insome part and they can hardly represent the seedlingrecruitment of the whole 20 ha plot, another 99 additionalstations were constructed and censused in March, 2009(Fig. 1). To locate these stations, we chose the positions ofthe cement piles erected in 2004 for potential additionalstations because their coordinates in the plot were preciselymeasured when the plot was setup. The 99 additional stationswere randomly selected from the cement piles whosedistances to both the nearest erected seed-seedling stationand the borders of the plot were larger than or equal to20 m. On the chosen points, three 1 9 1 m seedling quadrateswere erected on three sides of the cement piles. Thesites for seedling quadrates were set to be 2 m away on theleft hand side, on the right hand side, and in the front whenstanding behind the cement pile and facing the ridge. Wepredetermined the positions of the sites for seedling quadratesso as to avoid any arbitrary influences. Tree seedlingswere measured and identified to the species level.This study was based on the live tree seedlings in theoriginal 149 stations in March 2009 and census data for the99 additional stations, which means we had a total of 248stations (Fig. 1). The combined data were a snap shot oftree seedlings in the 248 stations in March, 2009. We onlyanalyzed the species with more than 50 individuals andpresent in over 25 stations. Yet, they represent a wide rangeof life histories of plants (Table 1).The measurements of seed and fruit traits of all theseexcept Calophyllum membranaceum, Castanopsis fissa,Lindera metcalfiana, Neolitsea umbrosa, and Syzygiumrehderianumb were obtained from the fruits and seedscollected from the 149 seed traps since they were setup.The diameters of fruits and seeds in three orthogonaldirections were measured. The average diameter of threedirections was taken as size (Table 1). The CV (standarddeviation divided by mean) of three diameters was taken asshape, thus spherical fruits and seeds have a CV that is veryclose to zero. Not a single seed or fruit was collected forCalophyllum membranaceum, Castanopsis fissa, Linderametcalfiana, Neolitsea umbrosa, and Syzygium rehderianumb.For those species, more limited information aboutthe fruits and seeds was obtained from Song and Yi (1985).Fig. 1The distribution of stations in the plot324MethodWe predicted that the number of seedlings (S i ) at stationi was a combined result of the innate conspecific fecundity(CF) in the plot and the change of conspecific effect due todistance (P).The first component was D bjCF j ¼ a ; ð1ÞD minwhere D j was the diameter of the jth tree at breast height,D min was the minimal diameter of reproducing trees, anda and b were parameters to be calibrated.We also used another conspecific effect, assuming thatthe conspecific effect was a constant after the individualshad achieved a minimum DBH. It took the formCF j ¼ a:ð2ÞWe supposed that the conspecific effect of the jth tree toreach the ith station followedpIP ij ¼ pþ1; ð3Þpl 1 þ r2 iju123

456 Eur J Forest Res (2012) 131:453–461Table 1The information for the species relevant to this studySpecies Adaptation to light GrowthformD min(cm) aFruit mass(10 -2 g)Fruit size(mm)FruitshapeSeed mass(10 -2 g)Seed size(mm)SeedshapeAidia canthioides Shade-tolerant Subarbor 2 6.10 5.46 0.04 0.78 2.63 0.35Aporosa yunnanensis Shade-tolerant Subarbor 3 5.94 6.62 0.13 3.16 4.83 0.22Ardisia quinquegona Intermediate shade-tolerant Shrub 1.2 2.91 4.44 0.27 2.90 4.44 0.27Blastus cochinchinensis Shade-tolerant Shrub 1 0.81 2.41 0.08 0.53 2.53 0.06Calophyllum membranaceum Shade-tolerant Shrub 1 – 88.33 0.46 – – –Castanopsis fissa Shade intolerant Arbor 6 – 66.67 1.41 133.40 – –Cryptocarya concinna Intermediate shade-tolerant Arbor 8 12.24 6.73 0.40 12.24 6.73 0.40Lindera metcalfiana Intermediate shade-tolerant Subarbor 5 – 50.33 0.01 – – –Machilus chinensis Intermediate shade-tolerant Arbor 8 31.00 9.48 0.14 15.00 7.31 0.24Memecylon ligustrifolium Shade-tolerant Subarbor 1.5 20.26 8.06 0.03 16.22 7.34 0.01Neolitsea umbrosa Intermediate shade-tolerant Arbor 6 – 51.67 0.22 – – –Ormosia glaberrima Intermediate shade-tolerant Arbor 6 58.09 16.68 0.77 18.69 6.86 0.22Syzygium rehderianum Intermediate shade-tolerant Arbor 6 – – – – – –a The minimum diameter for reproductionwhere p, p and l were parameters, r ij was the distancebetween the ith station and the jth tree (only horizontal andvertical axes were used). The Eq. 3 is actually a 2Dt dispersalkernel. Conspecific effect here mainly refers to seeddispersal and conspecific density dependence. 2Dt was awidely used seed dispersal kernel (Clack et al. 1999; Schurret al. 2008), for simplification we supposed that otherconspecific effects besides seed dispersal also took thatform.When the anisotropic forces were taken into account, theconspecific effect of the jth tree to reach the ith station wasmodified slightly based on Eq. 3:AP ij ¼ IP ij þ c At jAs ir ij; ð4Þwhere At j denoted the altitude of the jth tree, As i the altitudeof the ith station, r ij was the distance between the ithstation and the jth tree (only horizontal and vertical axesare used), and c was a parameter to be calibrated.The number of seedlings predicted to be growing at theith station wasConstant conspecific effect and anisotropic decliningfunction (CA): Eqs. 2 and 4We predicted that the number of seeds and seedlingsobserved as following a Poisson distribution, where themean of the Poisson distribution was predicted as S i(Ribbens et al. 1994). The likelihood of observing O i seedsor seedlings when a mean of S i seeds or seedlings wereexpected under a Poisson distribution wase S iS O iiO i !; ð6Þand the likelihood (L) for a set of N station was (Ribbenset al. 1994)L ¼ YNi¼1e S iS O iiO i !ð7ÞThe set of values of parameters that maximized L werethe values we are looking for. The best model was selectedusing Akaike’s information criterion (AIC). Both dataanalysis and figure drawing were done with R 2.9.0(R development core team 2008).S i ¼ Xnj¼1CF j P ij :ð5ÞResultsBy combining these functions, four models wereobtained:Power conspecific effect and isotropic declining function(PI): Eqs. 1 and 3Power conspecific effect and anisotropic decliningfunction (PA): Eqs. 1 and 4Constant conspecific effect and isotropic decliningfunction (CI): Eqs. 2 and 3Generally, seedlings are distributed where there are conspecificadult trees (Fig. 2). Yet, it is not necessary thatwhere there are adult trees, there are seedlings of the correspondingspecies (Fig. 2b, c, d, f, g, i, j, k, m). For somespecies, the abundance of seedling does not seem to berelated to the abundance of adult trees (Fig. 2c, f, j, k, l, m).The best model for every species was selected accordingto AIC (Table 2). For 6 of the 13 species, the best model123325

Eur J Forest Res (2012) 131:453–461 457contained an anisotropic declining function (Table 3).These six species included two shrubs, three speciesreaching the canopy and one subcanopy member. Theremaining seven species had models with an isotropicdeclining function, of which five were moderately shadetolerant.Whether the species favor an isotropic or anisotropicdeclining function was not related to fruit and seedtraits (Fig. 3). Constant conspecific effect was supported byeight species, of which six were subcanopies or shrubs. Allexcept one species that supported power conspecific effectwere canopy members.The method produced poor fits (adjusted r square: R a 2 =0.07–0.15) to four species, and good fits (R a 2 = 0.16–0.34)to six species. The method fitted Blastus cochinchinensisthe best (R a 2 = 0.34; Fig. 4), but failed to produce a statisticallyvalid model to explain the seedling distribution ofCalophyllum membranaceum (R a 2 \ 0) and Memecylonligustrifolium (R a 2 \ 0). The performance of the approachseemed to be related with the shade tolerance of the species:among the five shade-tolerant species in this study,four had relatively small (R a 2 B 0.15) or even negative R a 2 ,and Blastus cochinchinensis was the only exception.Among the seven moderately shade-tolerant species, fivehad relatively large R a 2 (R a 2 [ 0.15), except for Ardisiaquinquegona and Syzygium rehderianum. Growth form wasanother possible factor related to the R a 2 of the result. Twoof the 3 shrubs and 3 of the 4 subcanopy species studiedhad relatively small or negative R a 2 (R a 2 B 0.15), and 4 ofthe 6 canopy species had relatively large R a 2 (R a 2 [ 0.15).Among different species, a, u, and p varied over severalorders of magnitude (Table 3). Among those species thatsupported AI or AC, the parameter c was positive for allmoderately shade-tolerant species and shade-tolerant speciesexcept Calophyllum membranaceum, and negative forCalophyllum membranaceum and the only light demandingspecies in this study. One of the species with negative c has6 to over 250 times larger seed mass than the other speciesthat support AI or AC but had positive c (Tables 1, 3).DiscussionThe inverse modeling method has been used widely in thefield of predicting seed and seedling distribution (Ribbenset al. 1994; LePage et al. 2000; Uriarte et al. 2005; Muller-Landau et al. 2008). Yet, previous studies usually did notconsider an anisotropic effect. Wagner et al. (2004) andWälder et al. (2009) were two of the very few exceptions.The simplest reason why directionality should be taken intoconsideration in seed dispersal models is that it can beobserved in nature (e.g., Wagner et al. 2004). Anisotropymay be closer to reality for seed dispersal in some case dueto wind direction (Stull 1989) and gravity, and for seedlingdistribution possibly due to anisotropic distribution ofseeds and subsequent ecological processes influencing thesurvival of seedling. In our study, we modified the classicinverse modeling framework slightly so as to integrate ananisotropic effect due to altitude. The result showed thatthe anisotropic models can fit the data of six species betterthan the isotropic models, indicating that seedling recruitmentfor some species can be anisotropic around conspecificindividuals. In contrast to our study, Wagner et al.(2004) found that seed dispersal consistently showedanisotropic dispersion. In their research, the studied objectsare seeds and, as the authors implied in the context, theseeds are dispersed mainly by wind (Wagner et al. 2004).The two points are the main causes of anisotropy. On thecontrary, the objects we studied are seedlings with a heightof at least 20–30 cm with complex vascular tissue in stemsand leaves, indicating that they have survived in the understoreyfor a long time and probably undergone severecompetition and many environmentally filtering processes.Unsurprisingly, the resulting spatial pattern is more vagueand complex, making it difficult to detect any underlyingtrends.Yet, some species still show anisotropic seedlingrecruitment around conspecifics. In contrast to wind-dispersedseeds for which wind direction is responsible foranisotropy (Wagner et al. 2004), the ecological preferenceof these species explain much of the observed pattern.Among those species that show anisotropic seedlingrecruitment, the parameter c was positive for all moderatelyshade-tolerant species and shade-tolerant speciesexcept Calophyllum membranaceum, suggesting that thesespecies gain an extra benefit for successful recruitmentwhen the altitude of their location decreases. This is consistentwith their ecological trait because the sites withlower altitude are very easy to be shaded by the ridges inthis topographically complex forest plot. Furthermore, it isevident that shade tolerance is significantly negativelycorrelated with drought tolerance, according to Niinemetsand Valladares’s (2006) research involving 806 shrubs andtrees. In mountain regions, the distribution of soil moistureis closely related to the relative altitude and soil moisturecontent on the lower slope is larger than on the upper slope(Xu et al. 2003). In an old-growth Douglas fir forest,researchers also found a strong moisture gradient related toelevation at soil depths of 30 and 50 cm; deeper soils weredrier at higher elevations (He and Duncan 2000). Theshade-tolerant and moderately shade-tolerant seedlingsmay have undergone some niche-based processes duringwhich the seeds that germinate at higher altitude meetdifficulty in survival because of a lack of soil moisture. Thecase of Castanopsis fissa provides evidence for this fromthe opposite aspect. Seeds of Castanopsis fissa weigh 6 toover 250 times more than the species with positive c. From123326

458 Eur J Forest Res (2012) 131:453–461Fig. 2 The distribution of seedlings and adult trees of the 13 speciesin the plot. Gray dashed lines are the contour lines depicting thelandform of the plot. A gray triangle denotes an adult tree of thecorresponding species. Black circle denotes the existence of seedlingsat specific station. The radius of a circle is proportional to thelogarithm of the abundance of seedlings at the stationthis biological trait, we can infer that these individuals mayleave most of their seeds at low altitudes, resulting in largeamount of seedlings at sites with lower altitudes. Yet, thecurrent situation is that Castanopsis fissa has a negative c,indicating that the higher the altitude, the more seedlingscan be found. This is in agreement with the lightdemanding feature of Castanopsis fissa rather than thebiological feature of its seeds. Thus, niche-based processesand species ecological preference for specific habitats mayat least play some part in forming the seedling distributionpattern in this subtropical forest.In previous researches involving inverse modeling, theexpected number of seeds or seedlings often increases withconspecific DBH (Uriarte et al. 2005; LePage et al. 2000).In those researches, the seedlings are in the newly emergedstage before competition or density dependence processes123327

Eur J Forest Res (2012) 131:453–461 459Table 2 The AIC (Akaiker’s information criterion) for the fourmodels fitted to the species seedling recruitment dataSpecies PI a PA b CI c CA dAidia canthioides 1,977.45 1,964.63 e 1,991.36 1,984.38Aporosa yunnanensis 335.95 336.54 334.57 334.89Ardisia quinquegona 466.15 464.38 460.62 462.36Blastus cochinchinensis 593.83 536.85 596.85 531.66Calophyllum527.69 526.31 526.78 525.97membranaceumCastanopsis fissa 377.84 367.84 376.01 365.51Cryptocarya concinna 1,681.05 1,678.06 1,730.46 1,832.70Lindera metcalfiana 224.19 224.68 222.69 223.10Machilus chinensis 361.50 363.20 372.38 377.02Memecylon ligustrifolium 329.20 329.13 327.22 327.37Neolitsea umbrosa 368.26 369.63 367.98 369.87Ormosia glaberrima 680.29 677.57 715.50 734.62Syzygium rehderianum 760.06 760.54 776.48 800.80a Power conspecific effect and isotropic declining functionb Power conspecific effect and anisotropic declining functionc Constant conspecific effect and isotropic declining functiond Constant conspecific effect and anisotropic declining functione The lowest AIC for each species is in boldtake place, thus the DBH involved mainly stands for thefecundity of conspecifics. In our study, the seedlings arenot newly emerged seedlings but have survived in theunderstory for a long time. This can be identified by theirheight (at least 20 cm) and the developed vascular tissue instems and leaves. In this situation, we need to considerpostdispersal processes (Nathan and Muller–Landau 2000).The initial positive relationship between adult DBH and theabundance of newly emerged seedling can be masked ifseedling success is negatively related to adult DBH. Inother words, if there is strong, positively density-dependentmortality as a result of the conspecific adult nearby, the netconspecific effect can be negative. Conspecific densitydependence has been proposed as a mechanism maintainingspecies diversity in forest communities (Janzen 1970;Connell 1971; Peters 2003; He and Duncan 2000). Densitydependentmortality possibly results from the enhancedpropagation of pests and species-specific herbivores andpathogens (Janzen 1970; Connell 1971). Some researchershave pointed out that high conspecific density and conspecificbasal area result in low survival for conspecifictrees, especially for seedlings and saplings (Peters 2003;Webb and Peart 2000; He and Duncan 2000; Pigot andLeather 2008). Uriarte et al. (2005) also found that themajority of species studied supported a model with adensity-dependent effect. Density-dependent effect isprobably the reason why some species have negativeexponents for the conspecific effect function (Eq. 1),indicating that the number of recruits expected at a givendistance decline with increasing conspecific DBH, thusproviding indirect evidence for the density dependencephenomenon. In our study, those species are Memecylonligustrifolium, Syzygium rehderianum, and Cryptocaryaconcinna. In the Dinghu Mountain Nature Reserve,Cryptocarya concinna is the main food for Thelassodesquadraia, an insect (Huang et al. 1998). The speciesspecificpest possibly explains why Cryptocarya concinnaseedling recruitment show a negative relationship withconspecific diameter. The other two species, Memecylonligustrifolium and Syzygium rehderianum, are not as wellstudied as Cryptocarya concinna is.Table 3Adjusted R square and parameter estimates for the best modelSpecies R square Best model a u p a c bAidia canthioides 0.103 PA 6.10 9 10 4 1.07 9 10 1 1.37 9 10 2 1.23 9 10 -6 1.17Aporosa yunnanensis 0.150 CI 2.43 9 10 2 1.16 9 10 -3 1.30 9 10 4 – –Ardisia quinquegona 0.096 CI 8.63 9 10 0 2.84 9 10 -1 1.13 9 10 1 – –Blastus cochinchinensis 0.343 CA 6.12 9 10 1 1.14 9 10 0 1.36 9 10 1 8.59 9 10 -6 –Calophyllum membranaceum -0.005 CA 4.12 9 10 4 1.10 9 10 -2 1.48 9 10 5 -3.41 9 10 -8 –Castanopsis fissa 0.106 CA 2.03 9 10 2 9.84 9 10 -4 2.51 9 10 5 -2.69 9 10 -8 –Cryptocarya concinna 0.223 PA 1.05 9 10 3 1.92 9 10 -2 3.54 9 10 5 8.95 9 10 -8 -15.24Lindera metcalfiana 0.181 CI 1.95 9 10 3 1.27 9 10 0 1.45 9 10 1 – –Machilus chinensis 0.277 PI 8.60 9 10 0 2.74 9 10 -4 6.57 9 10 3 – 2.74Memecylon ligustrifolium -0.014 CI 1.79 9 10 0 1.85 9 10 -4 1.87 9 10 4 – -0.08Neolitsea umbrosa 0.219 CI 5.24 9 10 2 6.63 9 10 -4 4.99 9 10 4 – –Ormosia glaberrima 0.290 PA 7.00 9 10 0 6.42 9 10 -2 6.52 9 10 2 1.18 9 10 -6 2.66Syzygium rehderianum 0.072 PI 8.67 9 10 2 4.53 9 10 -2 4.13 9 10 4 – -78.2a Best model: PI: Power conspecific effect and isotropic declining function; PA: power conspecific effect and anisotropic declining function; CI:constant conspecific effect and isotropic declining function; CA: constant conspecific effect and anisotropic declining function328123

460 Eur J Forest Res (2012) 131:453–461Fig. 4 The predicted values by the best model versus the observedvalues of Blastus Cochinchinensis, the species with the highestadjusted R square in this studyFig. 3 Boxplots of seed/fruit traits of species grouped by thedeclining function they supported. In this figure, the studied speciesare classified by the declining function they support. Thus, weobtained two groups of species, the anisotropic group and theisotropic group (the horizontal axis). The distributions of seed/fruittraits (e.g., fruit mass in panel a, the vertical axis) were plotted as abox for the two groups of species. The notch in each ‘‘box’’ denotesthe 95% confidence interval. If the notches of the two boxes in a panel(e.g., panel a) do not overlap, the traits (e.g., fruit mass in panel a) ofthe groups of species do not have significant difference. From thisfigure, we know that all the difference in traits were not significantThe relation between DBH and number of recruits atgiven distance is very diverse in form. Not all the speciestake the form of Memecylon ligustrifolium, Syzygiumrehderianum, and Cryptocarya concinna. Aidia canthioides,Ormosia glaberrima, and Machilus chinensis have morerecruits as conspecific DBH becomes larger. The otherspecies have constant conspecific effects, including all ofthe shrubs, 2 of 3 subcanopy species, and one canopyspecies. Although assuming that seedling abundanceincrease with tree diameter, Uriarte et al. (2005) found thatthe relationship between tree diameter and the number ofseedlings produced is fairly flat for the majority of species.Our research is not alone in appointing a constant to therelationship between the number of offspring and conspecificsize. Wagner et al. (2004) also assumed that tree ofdifferent sizes gave birth to identical numbers of fruits. Inour study, the species supporting constant conspecificeffect are mainly shrubs and subarbors which do not havewide ranges of DBH. The resulting variation in conspecificeffect due to DBH may be masked by the variation fromother factors that are not considered in this research, e.g.,the habitat conditions of adult trees and the seedlings(Schurr et al. 2008). Another possibility is that this effectwas not calibrated well, given that Ribbens et al. (1994)found strong positive interactions between conspecificDBH and the number of recruits.Although widely used, the inverse modeling is notwithout methodological challenges. First, we cannotinclude all conspecifics that have an effect on the recruitment,although most of our seedling stations are over 20 mfrom the border of the plot. This is a problem that manyother researchers also cannot avoid (LePage et al. 2000;Uriarte et al. 2005). Second, the best fit obtained is only thebest among those simple models that were tried, and thusmay not well describe the complex and individually variableseedling recruitment that are influenced by seed dispersalprocess, competition, and density dependenceprocess. Third, animal’s effect on seed dispersal was nottaken into account in this research. Despite their shortcomings,our fitted models explain an average of 15.7% ofthe total variations in seedling distribution among stations,slightly smaller than other similar studies (Muller–Landauet al. 2008). These models capture a substantial first outlineof seedling recruitment of the studied species and providesan initial basis for predicting seedling recruitment in thisforest plot.Acknowledgments The authors thank Yong Shen, Wenping Liuand many other individuals for their help with the field work. Thisstudy was funded by the Knowledge Innovation Project of theChinese Academy of Sciences (KZCX-YW-430-03), the NationalKey Technology R&D Program (2008BAC39B02).123329

Eur J Forest Res (2012) 131:453–461 461ReferencesAbrams PA (1984) Recruitment, lotteries, and coexistence in coralreef fish. Am Nat 123:44–55Augspurger CK (1984) Seedling survival of tropical tree species:interactions of dispersal distance, light-gaps, and pathogens.Ecology 65:1705–1712Bohrer G, Nathan R, Volis S (2005) Effects of long-distance dispersalfor metapopulation survival and genetic structure at ecologicaltime and spatial scales. J Ecol 93:1029–1040Bullock JM, Clarke RT (2000) Long distance seed dispersal by wind:measuring and modelling the tail of the curve. Oecologia124:506–521Chen L, Mi X, Comita LS, Zhang L, Ren H, Ma K (2010)Community-level consequences of density dependence andhabitat association in a subtropical broad-leaved forest. EcolLett 13:695–704Clack JS, Silman M, Kern R, Macklin E, HilleRisLambers J (1999)Seed dispersal near and far: patterns across temperate andtropical forests. Ecology 80:1475–1494Connell JH (1971) On the role of natural enemies in preventingcompetitive exclusion in some marine animals and in rain foresttrees. In: den Boer PJ, Gradwell GR (eds) Dynamics ofpopulations. Centre for Agricultural Publishing and Documentation,Wageningen, pp 298–312Dalling JW, Hubbell SP, Silvera K (1998) See dispersal, seedlingestablishment and gap partitioning among tropical pioneer trees.J Ecol 86:674–689Greene DF, Canham CD, Coates KD, Lepage PT (2004) Anevaluation of alternative dispersal functions for trees. J Ecol92:758–766Harms KE, Condit R, Hubbell SP, Foster RB (2001) Habitatassociations of trees and shrubs in a 50-ha neotropical foretplot. J Ecol 89:947–959He F, Duncan RP (2000) Density-dependent effect on tree survival inan old-growth Douglas fir forest. J Ecol 88:676–688Huang Z, Kong G, Wei P (1998) Plant species diversity dynamics inDinghu Mountain. Biodiv Sci 6:116–121 (in Chinese withEnglish abstract)Hubbell SP (2001) The unified neutral theory of biodiversity andbiogeography. Princeton University Press, PrincetonHubbell SP, Foster RB, O’Brien ST, Harms KE, Condit R, WechslerB, Wright SJ, Loo de Lao S (1999) Light gap disturbances,recruitment limitation, and tree diversity in a neotropical forest.Science 283:554–557Janzen DH (1970) Herbivores and the number of tree species intropical forests. Am Nat 104:501–527Leak WB, Graber RE (1976) Seedling input, death and growth inuneven-aged northern hardwoods. Can J For Res 6:368–374LePage PT, Canham CD, Coates KD, Bartemucci P (2000) Seedabundance versus substrate limitation of seedling recruitment innorthern temperate forests of British Columbia. Can J For Res30:415–427Muller-Landau HC, Wright SJ, Calderón O, Condit R, Hubbell SP(2008) Interspecific variation in primary seed dispersal in atropical forest. J Ecol 96:653–667Nathan R, Muller-Landau HC (2000) Spatial patterns of seeddispersal, their determinants and consequences for recruitment.Trends Ecol Evol 15:278–285Näther W, Wälder K (2003) Experimental design and statisticalinference for cluster point processes—with applications to thefruit dispersion of Anemochorous forest trees. Biometrical J45:1006–1022Niinemets U, Valladares F (2006) Tolerance to shade, drought andwaterlogging of temperate, Northern hemisphere trees andshrubs. Ecol Monogr 76:521–547Pacala SW, Canham CD, Saponara J, Silander JA, Kobe RK, RibbensE (1996) Forest models defined by field measurements: estimation,error analysis and dynamics. Ecol Monogr 66:1–43Peters HA (2003) Neighbour-regulated mortality, the influence ofpositive and negative density dependence on tree populations inspecies-rich tropical forests. Ecol Lett 6:757–765Pigot AL, Leather SR (2008) Invertebrate predators drive distancedependentpatterns of seedling mortality in a temperate tree Acerpseudoplatanus. Oikos 117:521–530Queenborough SA, Burslem DFRP, Garwood NC, Valencia R (2007)Neighborhood and community interactions determine the spatialpattern of tropical tree seedling survival. Ecology 88:2248–2258R Development Core Team (2008) R: a language and environment forstatistical computing. R Foundation for Statistical Computing,Vienna, Austria. http://www.R-project.orgRibbens E, Silander JA Jr, Pacala SW (1994) Seedling recruitment inforests: calibrating models to predict patterns of tree seedlingdispersion. Ecology 75:1794–1806Scholl AE, Taylor AH (2006) Regeneration pattern in old-growth redfir-western white pine forests in the northern Sierra Nevada,Lake Tahoe, USA. For Ecol Manage 235:143–154Schurr FM, Steinitz O, Nathan R (2008) Plant fecundity and seeddispersal in spatially heterogeneous environments: models,mechanisms and estimation. J Ecol 96:628–641Song S, Yi J (1985) Seed and seedling morphology of Dinghushantree species. Hainan People’s Press, HainanStull RB (1989) An introduction to boundary layer meteorology.Kluwer Academic Publishers, DordrechtTuomisto H, Ruokolainen K, Yli-Halla M (2003) Dispersal, environment,and floristic variation of western Amazonian forests.Science 299:241–244Turnbull LA, Crawley MJ, Rees M (2000) Are plant populations seedlimited?A review of seed sowing experiments. Oikos 88:225–238Uriarte M, Canham CD, Thompson J, Zimmerman JK, Brokaw N(2005) Seedling recruitment in a hurricane-driven tropical forest:light limitation, density-dependence and the spatial distributionof parent trees. J Ecol 93:291–304Valencia R, Foster RB, Villa G, Condit R, Svenning JC, Hernandez C,Romoleroux K, Losos E, Magard E, Balslev H (2004) Treespecies distributions and local habitat variation in the Amazon:large forest plot in eastern Ecuador. J Ecol 92:214–229Wagner S, Wälder K, Ribbens E, Zeibig A (2004) Directionality infruit dispersal models for anemochorous forest trees. Ecol Model179:487–498Wälder K, Näther W, Wagner S (2009) Improving inverse modelfitting in trees—anisotropy, multiplicative effects, and Bayesestimation. Ecol Model 220:1044–1053Wang Z, Ye W, Cao H, Huang Z, Lian J, Wei S, Sun I-F (2009)Species-topography association in a species-rich subtropicalforest of China. Basic Appl Ecol 10:648–665Webb CO, Peart DR (2000) Habitat associations of trees andseedlings in a Bornean rain forest. J Ecol 88:464–478Xu X, Liu W, Gao P, Mu X (2003) The discussion on soil moisturedistributional diversity in hilly Loess Plateau region. EcolEnviron 12:52–55 (in Chinese with English abstract)Ye WH, Cao HL, Huang ZL, Lian JY, Wang ZG, Li L, Wei SG,Wang ZM (2008) Community structure of a 20 hm 2 lowersubtropical evergreen broadleaved forest plot in Dinghushan,China. Acta Ecol Sin (Chinese version) 32:274–286 (in Chinesewith English abstract)123330

Journal of Tropical Forest Science 23(1): 57–66 (2011)Bin Y et al.TREE MORTALITY AND RECRUITMENT IN A SUBTROPICALBROADLEAVED MONSOON FOREST IN SOUTH CHINAY Bin 1, 2 , J Lian 1 , Z Wang 1 , W Ye 1, * & H Cao 11Key Laboratory of Plant Resources Conservation and Sustainable Utilization, South China Botanical Garden, ChineseAcademy of Sciences, Guangzhou 510650, China2Graduate University of Chinese Academy of Sciences, Beijing 10049, ChinaReceived February 2010BIN Y, LIAN J, WANG Z, YE W & CAO H. 2011. Tree mortality and recruitment in a subtropical broadleavedmonsoon forest in south China. Mortality and recruitment are key factors influencing forest succession.Using spatial pattern and neighbourhood analyses, effects of conspecifics and heterospecifics upon mortalityand recruitment of some locally dominant tree species were investigated in a 1-ha forest plot. Specifically,we were interested in how species in different layers of the forest responded to such effects in a subtropicalforest in China. During a seven-year period (1992–1999), mortality rates of the studied species ranged from2 to 7% per year while recruitment rates ranged from 0 to 3% per year. At this small spatial scale, mortalityof all but one species was random in space. Unlike mortality, however, recruitment into the ≥1-cm size classconsistently occurred where local conspecific density was high. This suggests that this process may be limitedby seed dispersal. Heterospecific individuals did not influence recruitment significantly for any species. Bothcanopy species had difficulty recruiting into the ≥1-cm size class during the study period. In conclusion, treemortality in this patch of forest was random and recruitment for six non-canopy species and two canopyspecies was possibly limited by seed availability and ecological needs respectively.Keywords: Density dependence, spatial pattern, neighbourhood analysisBIN Y, LIAN J, WANG Z, YE W & CAO H. 2011. Kematian serta perekrutan pokok di hutan monsun daunlebar subtropika di selatan China. Kematian serta perekrutan merupakan faktor utama yang mempengaruhipenggantian hutan. Kesan konspesies dan heterospesies terhadap kematian dan perekrutan beberapaspesies pokok dominan dikaji dalam plot 1 ha dengan menggunakan corak analisis ruang serta analisisjiran. Tujuannya adalah untuk melihat bagaimana spesies dalam lapisan hutan yang berbeza bergerak balasterhadap kesan tersebut di hutan subtropika di China. Sepanjang tempoh tujuh tahun (1992–1999), kadarkematian spesies yang dikaji berjulat antara 2% hingga 7% sementara kadar perekrutan berjulat antara0% hingga 3% setiap tahun. Pada skala ruang yang kecil ini, kematian semua spesies, kecuali satu, adalahsecara rawak. Tidak seperti kematian, perekrutan dalam kelas saiz ≥ 1 cm wujud secara konsisten apabilakepadatan konspesies tempatan adalah tinggi. Ini mencadangkan bahawa proses ini mungkin terhad olehpenyebaran biji benih. Individu heterospesies tidak mempengaruhi perekrutan secara signifikan untukmana-mana spesies. Bagaimanapun, dalam tempoh kajian, kedua-dua spesies kanopi menghadapi kesukaranperekrutan dalam kelas saiz ≥ 1 cm. Sebagai kesimpulan, didapati bahawa kematian pokok di tapak hutanini adalah rawak dan perekrutan mungkin dihadkan oleh penyebaran biji benih. Juga keperluan ekologiuntuk enam species bukan kanopi serta dua spesies kanopi masing-masing dihadkan oleh kewujudan bijibenih dan keperluan ekologi.INTRODUCTIONMortality and recruitment are key factorsinfluencing the dynamics and structure offorest tree populations (Lewis et al. 2004)and the succession and composition of forestcommunities. Consequently, an understandingof these processes is important for ecologists andforest managers (Silk et al. 2003).Conspecific density-dependent mortalityis one possible mechanism responsible forthe maintenance of species diversity in forestcommunities (He & Duncan 2000, Peters 2003).Negative conspecific density-dependent mortality,possibly resulting from the enhanced propagationof pests and species-specific herbivores andpathogens, is defined as an inverse relationshipbetween plant survival and conspecific density.Under the regulation of this negative feedback,rare species can achieve a higher rate of population*Author for correspondence. E-mail: why@scbg.ac.cn© Forest Research Institute Malaysia57331

Journal of Tropical Forest Science 23(1): 57–66 (2011)Bin Y et al.growth than common species and, thus, maintainspecies diversity (Janzen 1970, Connell 1971).Some researchers have pointed out that highconspecific density and/or basal area resultsin low survival for conspecific trees, especiallyfor seedlings and saplings (Peters 2003, Pigot& Leather 2008). Others have suggested thatdistance from conspecific trees has little impacton seed survival but may increase seedling survival(Hyatt et al. 2003), or that performances of plantsare negatively related to the overall densityirrespective of the species of their neighbours(Sletvold 2005). Therefore, the extent of densitydependentmortality most likely varies within andacross forest ecosystems.Dispersal limitation is another mechanismfor maintenance of species diversity. Dispersallimitation refers to the phenomenon thatseed density declines rapidly with distanceaway from the maternal tree (Muller-Landauet al. 2008). Under this hypothesis, seeds ofsuperior competitors may fail to arrive at suitablemicrosites and less competitive species will havemore chances to take their places, thus, slowingcompetitive exclusion and promoting speciescoexistence (Hurtt & Pacala 1995). Scientists havefound some indirect support for this, for example,in a meta-analysis, seedling recruitment increasedin response to seed addition for about half ofthe species tested (Turnbull et al. 2000), and intropical forest communities, several field studiesfrom Panama have shown that the composition ofseedlings in canopy gaps closely reflected that ofthe adult communities around them (Dalling etal. 1998, Hubbell et al. 1999). An analysis of therelations between recruit density and conspecificdensity in the neighbourhood of recruits andhow spatial pattern is influenced by recruits maythus provide further indirect information on therelationship between dispersal limitation andrecruitment patterns in forests.Mortality and recruitment have been studiedextensively in tropical forests, especially in searchof density dependence and limited dispersal(Condit et al. 1994, Peters 2003, Uriarte et al.2005, Wright et al. 2005, Queenborough etal. 2007). The conspecific negative densitydependenteffect and limited dispersal may alsoplay a role in maintaining the species diversityin a subtropical area. Yet there is a lack ofcomparative information for subtropical forestsystems for which mortality and recruitmentare little explored and thus poorly understood.We speculate that the conspecific negativedensity-dependent mortality also exists in thissubtropical forest but the effects differ amongspecies, and that although heterospecifics canalso affect survival, the effect is relatively weakcompared with conspecifics. We also speculatethat recruits may gather around conspecificsbecause of limited dispersal of seeds and fruits,and with regard to seedling stage before severecompetition takes place, heterospecifics generallydo not influence the seedling recruitmentpattern. In this study, we aimed to test the abovespeculations on recruitment and mortality withdata obtained from a forest plot in a subtropicalarea.MATERIALS AND METHODSStudy siteA 1-ha permanent plot was set up in November1992 in Dinghu Mountain, a nature reservelocated in Zhaoqing (112° 30'–112° 33' E, 23°09'–23° 11' N), Guangdong province, China. Thereserve occupies an area of 1155 km 2 , coveredmostly by hills and valleys at an altitude rangingfrom about 14 to 1000 m above sea level (asl).This area has a typical monsoonal climate with anannual average precipitation of 1927 mm. Apriltill September is the main rainy season. The meanannual temperature is 21 °C. The lowest monthlyaverage temperature is 12.6 °C in January and thehighest is 28 °C in July. The annual mean relativehumidity is 80%.The soil in Dinghu Mountain is composedmainly of lateritic red and mountain yellowbrownsoil. The lateritic red soil occurs in hillyland below an altitude of 300 m, and in hills andlow mountains at an altitude of 300 to 900 m asl,whereas the mountain yellow-brown soil occurspartially on the top of overlying hills.The plot was constructed in November 1992.To map the stems accurately, the 1-ha plot wassubdivided into 25 subplots, each measuring20 × 20 m, that we further divided into sixteen5 × 5 m quadrates. All trees and shrubs withdiameters at breast height (dbh) ≥1 cm weretagged, measured for dbh with callipers andidentified to species. Using a measuring tapewith a precision of 1 dm, coordinates inside eachquadrate and a detailed map of all stem positionsin the 1-ha plot were obtained. New recruits andmortality were recorded in a subsequent census© Forest Research Institute Malaysia58332

Journal of Tropical Forest Science 23(1): 57–66 (2011)Bin Y et al.in November 1999. Recruits were defined here astrees that were less than the 1-cm dbh thresholdin November 1992 but had reached 1 cm or morein November 1999. Sprouts were excluded fromthe analysis.The structure of the forest is complex. Thereare five layers in the forest from the top of thecanopy to the ground floor, namely, three treelayers (top: height above 15 m, middle: 10–15m and low: 3–9 m), one shrub layer (0.5–2 m)and one herb layer (below 0.5 m). Based onthe importance value (IV), which is the sumof relative abundance, relative dominance andrelative frequency (Song 2001), Castanopsischinensis and Schima superba are the two mostdominant species in the top tree layer (i.e. thecanopy). Cryptocarya chinensis and Cryptocaryaconcinna are the two most dominant species inthe middle layer, Acmena acuminatissima, Aporosayunnanensis are the two most dominant speciesin the low layer, and Blastus cochinchinensis andPsychotria rubra dominate the herb/shrub layer.These eight species—as opposed to only canopyspecies—were chosen because they likely haveimportant ecological roles in different layersof forest and this approach provided a morecomprehensive description of forest dynamics.Neighbourhood analysisWe used the logistic regression model to study therelationship between the mortality probabilityof an individual tree and its neighbourhooddensity (He & Duncan 2000, Suzuki et. al2003). Predictive variables were the number ofconspecifics (Cons-N), basal area of conspecifics(Cons-BA), number of heterospecifics (Het-N)and basal area of heterospecifics (Het-BA)within 5, 7.5 and 10 m from the focal tree. Eachpredictive variable was used alone in the logisticregression model.The significance of the logistic regressioncannot be tested in the usual way because oursample violates the assumption of statisticalindependence (He & Duncan 2000, Suzuki et al.2003). Instead, a randomisation procedure wasused to remove the distortion resulting from anyspatial pattern of plants. By holding the numberof dead trees the same as observed, the fate ofa tree was randomly assigned to the predictivevariable. One thousand randomisations wereconducted, generating the null distribution oflog-likelihood ratio. The observed log-likelihoodratio was then compared with the null distributionunder the assumption that tree mortality wasindependent of the predictive variable. If theobserved log-likelihood ratio was outside the95% confidence level of the null distribution,the effect was judged to be significant.To determine whether recruitment was relatedto density, Cons-N, Cons-BA, Het-N and Het-BAwithin 5, 7.5 and 10 m of recruits of the eightdominant species were compared with those of500 randomly generated points. If their 95%confidence intervals did not overlap they wereconsidered to be significantly different. Data wereanalysed using the statistical computing programR 2.6.0 (2008).Spatial pattern analysisSecond order point pattern analyses were usedto detect the relative importance of densitydependentand density-independent processes.K(r) is defined as the expected number of pointswithin distance r from randomly chosen points(He & Duncan 2000). K(r) was calculated at1-m intervals until a maximum of 25 m. Thetransformation L(r) = (K(r)/pi) × 0.5 – r isexpected to be zero when the pattern is spatiallyrandom. An L value greater than zero revealsclustering whereas an L value less than zeroindicates regular distribution.We tested the random mortality hypothesisby determining whether the spatial pattern ofindividuals in 1992 with ≥1-cm dbh that were stillalive in 1999 differed from that expected if themortality of living trees in 1992 had occurredrandomly. Random mortality was simulated byrandomly deleting individuals from the originaldata set. The number of individuals removedwas set to be equal to the number of individualsthat died from 1992 till 1999. This procedurewas repeated 1000 times and the significantdifference from randomness was tested as above.Also tested in the same way was whether recruitsoccurred randomly in space. The only differencewas that trees were not randomly deleted butrather the number of recruited was added intothe original data set. This procedure was alsoconducted on R 2.6.0.RESULTSIn 1992, 3535 trees ≥1-cm dbh were mapped inthe 1-ha forest plot. In 1999 census, 2597 of them© Forest Research Institute Malaysia59333

Journal of Tropical Forest Science 23(1): 57–66 (2011)Bin Y et al.were alive and 602 recruits were observed. Theeight species accounted for 58.7% of the totalnumber of stems, 71.3% of the total basal area,60.2% of the dead trees and 34.1% of the recruits.Annual death rates ranged from 2.04 to 7.55%and recruitment rates, from 0 to 3.57% amongthe eight species (Table 1). Castanopsis chinensisand S. superba accounted for 35.04% of the totalbasal area. Almost all standing C. chinensis andS. superba were above 10 m tall and the majorityof them were about 20 m tall (Zhou, personalcommunication).MortalityCons-N had significantly positive impact on themortality of A. acuminatissima across all threeneighbourhood radii (Table 2). At the 5-mneighbourhood distance, mortalities of C. concinnaand B. cochinchinensis were significantly negativelyrelated to Het-N and Het-BA respectively. Mortalityof A. yunnanensis was significantly positivelyassociated with Het-BA at 7.5 and 10 m (Table2). Regression could not be conducted for threespecies because of their too small sample sizes.The L(r) lines of the live trees of C. chinensis,S. superba , C. chinensis, C. concinna, A. yunnanensisand P. rubra were inside the 95% confidenceenvelopes, indicating that mortality of thesespecies occurred randomly during the timeinterval 1992–1999 (Figures 1a–d, f and h). Thecurve of live trees of A. acuminatissima exceededthe lower envelope at small scales, indicating thatmortality of this species was possibly negativelydensity-dependent (Figure 1e), and part of theL-curve for B. cochinchinensis exceeded the upperenvelope at scales around 2 to 5 m, indicatingthat mortality of this species was possibly relatedto habitat heterogeneity (Figure 1g).RecruitmentExcept for Cons-N of C. concinna at the 7.5-mradius, the recruits of six of the eight speciesstudied had larger observed Cons-N and Cons-BA means than randomly generated points didat all three neighbourhood radii although mostof the differences were not significant (Table3). Tests were not conducted for S. superba as ithad no recruits and for C. chinensis as it had onlyone recruit.The density and basal area of heterospecificsdid not show a clear trend among species acrossall neighbourhood radii (Table 3). The observedHet-N and Het-BA were significantly larger thanrandom points only for P. rubra at radii 7.5 and10 m. Castanopsis chinensis had all observed pointslower than random points at all three radii,except for Het-BA at 5 m. The rest of the speciesdid not show any significant difference betweenobserved and simulated random points at anyradius (Table 3).Table 1Survey summary of the eight species studied in a 1-ha plot at Dinghu Mountain reservein ChinaSpeciesForestlayerNo. No. No.1992 a 1999 b died cNo.recruited dAnnualmortalityrate e (%)Annualrecruitmentrate f (%)Castanopsis chinensis Top 12 10 3 1 3.57 1.19Schima superba Top 35 30 5 0 2.04 0.00Cryptocarya chinensis Middle 22 19 7 4 4.55 2.60Cryptocarya concinna Middle 201 146 84 29 5.97 2.06AcmenaacuminatissimaLow 111 94 25 8 3.22 1.03Aporosa yunnanensis Low 1100 950 184 34 2.39 0.44BlastuscochinchinensisShrub 373 269 197 93 7.55 3.57Psychotria rubra Shrub 220 196 60 36 3.90 2.34a Abundance in 1992 census; b abundance in 1999 census; c number of individuals died from 1992 till 1999;d number of individuals recruited into the ≥ 1cm dbh size class from 1992 till 1999; e proportions of trees diedand recruited per year from 1992 till 1999; f proportions of trees recruited per year from 1992 till 1999© Forest Research Institute Malaysia60334

Journal of Tropical Forest Science 23(1): 57–66 (2011)Bin Y et al.Table 2Results of the analysis of the relationships between neighbourhood densityand tree mortality for five species studied in a 1-ha plot at Dinghu Mountainreserve in ChinaSpecies Parameter Neighbourhood distance5 m 7.5 m 10 mRegression coefficientCryptocarya concinna Cons-N -0.0441 -0.0543 -0.0400Het-N -0.0341* -0.0077 -0.0023Cons-BA -0.0002 -0.0002 -0.0001Het-BA -0.0001 0.0000 0.0001Acmena acuminatissima Cons-N 0.4526** 0.3634** 0.2550**Het-N -0.0362 -0.0157 -0.0161Cons-BA 0.0017 0.0003 0.0001Het-BA 0.0000 -0.0001 -0.0001Aporosa yunnanensis Cons-N 0.0038 0.0003 0.0023Het-N 0.0097 0.0049 0.0049Cons-BA 0.0010 0.0002 0.0002Het-BA 0.0000 0.0001* 0.0001**Blastus cochinchinensis Cons-N -0.0024 0.0076 0.0243Het-N -0.0093 0.0014 0.0065Cons-BA -0.0030 0.0003 0.0051Het-BA -0.0001** 0.0000 0.0000Psychotria rubra Cons-N 0.0078 0.0022 0.0065Het-N 0.0303 0.0145 0.0043Cons-BA 0.0008 0.0001 0.0006Het-BA 0.0001 0.0001 0.0000Cons-N = number of conspecifics; Het-N = number of heterospecifics; Cons-BA = basal area ofconspecifics; Het-BA = basal area of heterospecifics; * = p < 0.05; ** = p < 0.01Recruits increased spatial aggregations ofthe six species analysed but did so at differentspatial scales (Figure 2). Recruitments of A.acuminatissima and C. concinna significantlyincreased the spatial aggregation at scales greaterthan 10 and 20 m respectively (Figures 2c and b).Recruitment contributed to spatial aggregationof C. chinensis at scales up to 10 m (Figure 2a), butbeyond 10 m, recruitments seemed to be random(Figure 2a). The recruits of B. cochinchinensis, A.yunnanensis and P. rubra significantly increasedspatial aggregation at all the scales studied(Figures 2d, e and f).DISCUSSIONWhether a given forest community is regulatedby density-dependent processes has been debatedfor a long time. Some have found evidenceconsistent with density-dependent tree mortality(Wills et al. 1997, Peters 2003). However, in astudy on tree survival in an old-growth temperateforest in north-eastern China, significant negativedensity-dependent mortality was not detectedwhen trees reached 1-cm dbh (Zhang et al. 2009).In our research, mortality of A. acuminatissima waspositively related to the number of conspecifics© Forest Research Institute Malaysia61335

Journal of Tropical Forest Science 23(1): 57–66 (2011)Bin Y et al.L (r)-10 10aL (r)-2 0 2 4b0 5 10 15 20 25r (m)0 5 10 15 20 25r (m)Ripley’s L-functionL (r)L (r)-5 0 5 100 1 2 3c0 5 10 15 20 25r (m)eL (r)-0 5 1 0 2 5L (r)0 0 1 5 3 00 5 10 15 20 25r (m)df0 5 10 15 20 25r (m)0 5 10 15 20 25r (m)ghL (r)0 2 40 5 10 15 20 25r (m)L (r)0 2 4 60 5 10 15 20 25r (m)Figure 1Tests of random mortality by spatial pattern analysis in a 1-ha plot at Dinghu Mountain reserve inChina. The solid line is the observed L(r) after excluding trees that died in 1992 and 1999; thedash lines are the 95% confidence envelops for L(r) if mortality is random. a: Castanopsis chinensis,b: Schima superba, c: Cryptocarya chinensis, d: Cryptocarya concinna, e: Acmena acuminatissima, f: Aporosayunnanensis, g: Blastus cochinchinensis, h: Psychotria rubra. If the solid line is above the dash lines,mortality has enhanced the original spatial aggregation and mortality possibly results from habitatheterogeneity. If the solid line is below the dash lines, mortality alleviated the spatial aggregation ofthe original pattern, and mortality is possibly due to the negative effect of conspecific individuals.If the solid line is within the two dash lines, spatially random mortality is observed.up to 10-m neighbourhood radius analysed.However, we did not find such a consistentdensity-dependent mortality in the other fivespecies. Therefore, generally, we believe thatmortality is not dependent on density of theconspecifics in our study. Our results alsosuggested that mortality of individuals of thestudied species occurred, generally, at random,although it seemed that A. acuminatissima and B.cochinchinensis deviated from this at up to 5- and10-m neighbourhood radii respectively.On the other hand, a more plausibleexplanation for why we did not detect conspecificdensity-dependent mortality in our species,except possibly for A. acuminatissima, might bedue to the small sample sizes of our species andlimited spatial scale of 1 ha. For example, stemslying near plot borders could have been affectedby trees lying outside the plot. In order to obtainmore reliable results, a 20-ha forest plot wasestablished in Dinghu Mountain reserve in 2005(Li et al. 2009). Seed traps and seedling plots have© Forest Research Institute Malaysia62336

Journal of Tropical Forest Science 23(1): 57–66 (2011)Bin Y et al.Table 3Conspecific and heterospecific neighbourhood densities of observed recruits of the six studiedspecies and the randomly selected points at three radii (r) in a 1-ha plot at Dinghu Mountain reservein ChinaCryptocarya chinensisr = 5m r = 7.5m r = 10mObserved Random Observed Random Observed RandomCons-N 1.0 ± 0* 0.2 ± 0.1 1.3 ± 0.7* 0.3 ± 0.7 1.3 ± 0.7 0.7 ± 0.1Het-N 18.0 ± 6.9* 27.0 ± 1.2 31.3 ± 2.7* 57.8 ± 2.5 54.0 ± 6* 102.3 ± 4.4Cons-BA 1819.8 ± 1309.7* 102.6 ± 36.5 1875.6 ± 644.6* 161.4 ± 51.2 1875.6 ± 644.6* 390.2 ± 79.8Het-BA 926.8 ± 1324.7 2170.2 ± 227.9 3398.6 ± 621.7* 5164.6 ± 391.0 4847.0 ± 885.1* 8820.7 ± 484.7Cryptocarya concinnaCons-N 1.7 ± 2.0 1.6 ± 0.2 3.2 ± 0.6 3.3 ± 0.3 6.1 ± 0.8 5.6 ± 0.4Het-N 26.7 ± 17.0 25.4 ± 1.2 60.1 ± 5.8 56.3 ± 2.4 104.4 ± 9.0 95.8 ± 4.0Cons-BA 442.8 ± 680.5 305.9 ± 34.6 687.0 ± 177.5 641.5 ± 54.4 1204.1 ± 241.3 1126.8 ± 80.0Het-BA 2229.8 ± 5805.7 1851.0 ± 215.7 5219.4 ± 1567.3 5124.1 ± 387.7 10057.2 ± 1771.8 8023.5 ± 507.7Acmena acuminatissimaCons-N 1.0 ± 2.0 0.8 ± 0.1 2.7 ± 1.2 2.0 ± 0.2 4.9 ± 1.9 3.8 ± 0.3Het-N 17.1 ± 22.2 26.5 ± 1.2 42.0 ± 17.2 57.8 ± 2.5 73.7 ± 29.4 102.5 ± 4.3Cons-BA 253.0 ± 482.6 98.3 ± 25.5 346.9 ± 139.9 220.1 ± 36.6 459.9 ± 219.7 416.6 ± 50.3Het-BA 2341.0 ± 3102.6 2223.9 ± 240.7 3830.5 ± 1130.3 4954.6 ± 342.1 10666.4 ± 3412.8 9110.0 ± 495.0Aporosa yunnanensisCons-N 12.4 ± 10.6 8.8 ± 0.6 27.1 ± 4.2* 19.7 ± 1.2 46.4 ± 5.4* 35.1 ± 1.8Het-N 16.6 ± 17.0 17.5 ± 0.9 40.3 ± 8.2 41.0 ± 2.0 70.2 ± 14.7 69.7 ± 3.4Cons-BA 264.8 ± 222.4 194.1 ± 12.8 621.3 ± 91.2* 448.5 ± 23.8 1089.8 ± 108.7* 822.3 ± 38.7Het-BA 1698.4 ± 3272.4 2255.1 ± 242.9 4156.4 ± 1212.9 4876.1 ± 370.5 9162.2 ± 1679.3 8792.5 ± 471.5Blastus cochinchinensisCons-N 4.4 ± 8.4 2.6 ± 0.3 8.0 ± 1.6 5.5 ± 0.6 12.0 ± 2.0 9.6 ± 1.0Het-N 24.4 ± 23.4 24.3 ± 1.1 54.6 ± 5.6 55.4 ± 2.4 101.0 ± 9.3 96.3 ± 4.0Cons-BA 12.1 ± 25.3 7.6 ± 1.0 22.8 ± 5.4 15.8 ± 1.8 35.2 ± 6.6 27.2 ± 3.0Het-BA 2069.0 ± 4142.3 2167.4 ± 230.1 4594.8 ± 856.9 5324.4 ± 379.1 9713.2 ± 1207.3 9332.2 ± 487.7Psychotria rubraCons-N 3.6 ± 5.0 1.6 ± 0.2 8.5 ± 1.7* 3.8 ± 0.4 13.4 ± 1.9* 6.9 ± 0.7Het-N 31.5 ± 15.4 25.3 ± 1.1 74.1 ± 5.0* 54.6 ± 2.2 131.4 ± 7.0* 98.1 ± 3.7Cons-BA 19.6 ± 28.4 9.7 ± 1.7 46.8 ± 10.3* 23.0 ± 3.2 78.4 ± 15.0* 40.0 ± 5.0Het-BA 3276.8 ± 6078.7 2253.5 ± 216.5 7187.2 ± 1519.4* 4879.0 ± 327.7 13167.0 ± 1580.5* 10115.0 ± 523.6Mean ± 2SE; Cons-N = number of conspecifics; Het-N = number of heterospecifics; Cons-BA = basal area of conspecifics;Het-BA = basal area of heterospecifics; * = p < 0.05; ** = p < 0.01been constructed and monitored since late 2007.However, the only results obtained from the 20-haplot currently are from two-year-old seedlings andthus are not comparable with results obtained forseedlings from the current study. The 20-ha plotis due for another recensus in 2011. Nonetheless,the present study provided the first preliminaryinformation of tree mortality and recruitmentfor this region. Future studies from the 20-haplot should provide robust information on localforest dynamics and either verify or possiblycontradict our results. In this study, significanteffects of heterospecific density on mortality onlyoccurred positively for basal areas at 7.5- and10-m neighbourhood radius in A. yunnanensisand negatively at 5 m in B. cochinchinensis, and© Forest Research Institute Malaysia63337

Journal of Tropical Forest Science 23(1): 57–66 (2011)Bin Y et al.L (r)-2 2 6 10aL (r)0.0 1.0b0 5 10 15 20 250 5 10 15 20 25r (m)r (m)Ripley’s L-functionL (r)0.0 1.0 2.00 5 10 15 20 25r (m)cL (r)0.0 1.5d0 5 10 15 20 25r (m)L (r)0.5 2.0 3.50 5 10 15 20 25eL (r)1 3 5 70 5 10 15 20 25fr (m)r (m)Figure 2Tests of random recruitment by spatial pattern in a 1-ha plot at Dinghu Mountain reserve in China.The solid line is the observed L(r) for trees recorded in 1992 and recruits found in 1999, the dashlines are the 95% confidence envelops for L(r) if recruitment is random. a: Cryptocarya chinensis,b: Cryptocarya concinna, c: Acmena acuminatissima, d: Aporosa yunnanensis, e: Blastus cochinchinensis,f: Psychotria rubra. If the solid line is above the dash lines, recruitment enhanced the spatialaggregation of the original pattern, possibly due to limited dispersal. If the solid line is below thedash lines, recruitment alleviated the spatial aggregation of the original pattern, possibly due tohabitat heterogeneity. If the solid line is within the two dash lines, spatially random recruitment isobserved.negatively for number of heterospecifics at 5 mfor C. concinna, but not in other combinationsof species and neighbourhood radius. Theseinconsistent relationships between density ofheterospecifics and mortality among species aresimilar to those reported by Zhang et al. (2009)and Queenborough et al. (2007). Thoughplants all consume a set of similar resourcessuch as light, water and soil nutrient, differentspecies may differ in the amount they require,and when they need it. Adjacent heterospecificplants can influence each other by facilitationand competition. When the positive effect offacilitation exceeds that of competition, the netdirection of plant–plant interaction is positive,and vice versa.The two canopy species, C. chinensis and S.superba, had few recruits in the 1-ha plot wesurveyed in 1999. One possible cause for thispoor regeneration is that the relatively matureforest cannot meet the ecological needs of thesespecies. Castanopsis chinensis and S. superba areconsidered to be moderately light-demandingor shade-intolerant species (Huang et al. 1998).The climax of this forest is supposed to be aC. chinensis–C. concinna community; at the time ofcensus the forest was approaching climax basedon results from a 12-year study at the same site(Peng et al. 1998). Thus, perhaps it is not all thatsurprising that C. chinensis and S. superba nowexhibit some signs of in situ population decline,such as severe difficulty in recruitment to the© Forest Research Institute Malaysia64338

Journal of Tropical Forest Science 23(1): 57–66 (2011)Bin Y et al.1-cm dbh size class. Similarly, in an African wetforest, a dominant light-demanding canopy tree,Microberlinia bisulcata, also had scarce recruits> 1-cm dbh (Newbery et al. 2010). Anotherpossible explanation for poor regeneration inC. chinensis is related to the biology of its seeds.Before dispersal, the seeds are predated byCurculio davidi, a weevil, and then after dispersalby rats and birds (Du et al. 2006). Furthermore,pathogens threaten the survival of seeds in bothpre- and post-dispersal periods (Du et al. 2006).Collectively, potentially severe losses of seedsmay partially contribute to the few recruits of C.chinensis in the 1-ha plot. Alternatively, however,recruits of these canopy species may be occurringat some distance away from maternal trees wherecanopy gaps are more prevalent (e.g. Newberyet al. 2010). If these were indeed the case, suchrecruits outside of our 1-ha plot would have beenmissed out in our 1999 survey.Although much work has been donemonitoring tree recruitment and mortality in thisregion of China, more studies over a larger areaand longer time span are needed to elucidatewhether the forest is at equilibrium in the shortterm, and what factors are driving the successionprocess in this subtropical forest and other forestsaround the world.ACKNOWLEDGEMENTSWe thank the many individuals who contributedto the field survey of the plot. This study wasfunded by the Knowledge Innovation Projectof the Chinese Academy of Sciences (KZCX-YW-430-03), the National Key Technology R&DProgram (2008BAC39B02), and the 11th Five-YearPlan on National Scientific and TechnologicalSupport Projects (2008BADB0B05).REFERENCESCo n d i t R, Hubbell SP & Fo s t e r RB. 1994. Density dependencein two understorey tree species in a neotropicalforest. Ecology 75: 671–680.Con n e l l JH. 1971. On the role of natural enemies inpreventing competitive exclusion in some marineanimals and in rain forest trees. Pp. 298–312 inden Boer PJ & Gradwell GR (Eds.) Dynamics ofPopulations. Center for Agricultural Publication andDocumentation, Wageningen.Da l l i n g JW, Hu b b e l l SP & Si lv e r a K. 1998. Seed dispersal,seedling establishment and gap partitioningamong tropical pioneer trees. Journal of Ecology 86:674–689.Du Y, Pe n g S, Hu a n g Z & Xu G. 2006. Analysis of seeddeath of Castanopsis chinensis in the dispersalprocess in Dinghushan biosphere reserve. Ecologyand Environment 15: 1284–1288. (In Chinese withEnglish abstract)He FL & Du n c a n RP. 2000. Density-dependent effect on treesurvival in an old-growth Douglas fir forest. Journalof Ecology 88: 676–688.Hua n g Z, Ko n g G & We i P. 1998. Plant species diversitydynamics in Dinghu Mountain. Biodiversity Science 6:116–121. (In Chinese with English abstract)Hub b e l l SP, Fo s t e r RB, O’Br i e n ST, Ha r m s KE, Co n d i tR, We c h s l e r B, Wr i g h t SJ & Lo o d e La o S. 1999.Light gap disturbances, recruitment limitation, andtree diversity in a neotropical forest. Science 283:554–557.Hur t t GC & Pa c a l a SW. 1995. The consequences ofrecruitment limitation: reconciling chance, history,and competitive differences between plants. Journalof Theoretical Biology 176: 1–12.Hy a t t LA, Ro s e n b e r g MS, Ho wa r d TG, Bo l e G, Fa n g W,An a s ta s i a J, Br o w n K, Gr e l l a R, Hi n m a n K, Ku r d z i e lJP & Gu r e v i t c h J. 2003. The distance dependenceprediction of the Janzen–Connell hypothesis, a metaanalysis.Oikos 103: 590–602.Jan z e n DH. 1970. Herbivores and the number of treespecies in tropical forests. American Naturalist 104:501–527.Le w i s SL, Philips OL, Sh e i l D, Vi n c e t i B, Ba k e r TR, Br o w nS, Gr a h a m AW, Hi g u c h i N, Hi b e r t DW, La u r a n c e WF,Lej o l y J, Ma l h i Y, Mo n t e a g u d o A, Nú ñ e z Va r g a s P,So n k é B, Nu r Su pa r d i MN, Te r b o r g h JW & Vá s q u e zMar t í n e z R. 2004. Tropical forest tree mortality,recruitment and turnover rates, calculation,interpretation and comparison when census intervalsvary. Journal of Ecology 92: 929–944.Li L, We i SG, Hu a n g ZL, Ye WH & Ca o HL. 2009. Spatialdistributions of tree species in a subtropical forest ofChina. Oikos 118: 495–502.Mul l e r-La n d a u HC, Wr i g h t SJ, Ca l d e r ó n O, Co n d i t R &Hu b b e l l SP. 2008. Interspecific variation in primaryseed dispersal in a tropical forest. Journal of Ecology96: 653–667.Ne w b e ry DM, Ch r i s t o p h e JP, Xa n d e r MVDB, Ju l i a n MN &Geo r g e BC. 2010. Recruitment dynamics of thegrove-dominant tree Microberlinia bisulcata in Africanrain forest: extending the light response versusadult longevity trade-off concept. Plant Ecology 206:151–172.Pe n g S, Fa n g W, Re n H, Hu a n g Z, Ko n g G, Yu Q & Zh a n gD. 1998. The dynamics on organization in thesuccessional process of Dinghushan Cryptocaryacommunity. Acta Phytologica Sinica 22: 245–249.Pet e r s HA. 2003. Neighbour-regulated mortality, theinfluence of positive and negative density dependenceon tree populations in species-rich tropical forests.Ecology Letters 6: 757–765.Pi g o t AL & Le at h e r SR. 2008. Invertebrate predators drivedistance-dependent patterns of seedling mortalityin a temperate tree Acer pseudoplatanus. Oikos 117:521–530.Qu e e n b o r o u g h SA, Bu r s l e m DFRP, Ga r w o o d NC & Va l e n c i aR. 2007. Neighborhood and community interactionsdetermine the spatial pattern of tropical treeseedling survival. Ecology 88: 2248–2258.© Forest Research Institute Malaysia65339

Journal of Tropical Forest Science 23(1): 57–66 (2011)Bin Y et al.Sil k JWF, Ke b l e r PJA & Va n We l z e n PC. 2003. Macaragnaand Aallotus species (Euphorbiaceae) as indicatorfor disturbance in the mixed lowland dipterocarpforest of east Kalimantan (Indonesia). EcologicalIndicators 2: 311–324.Sl e t v o l d N. 2005. Density-dependent growth and survivalin a natural population of the facultative biennialDigitalis purpurea. Journal of Ecology 93: 727–736.Su z u k i RO, Ku d o h H & Ka c h i N. 2003. Spatial and temporalvariations in mortality of the biennial plant,Lysimachia rubida, effects of intraspecific competitionand environmental heterogeneity. Journal of Ecology91: 114–125.Son g YC. 2001. Vegetation Ecology. East China NormalUniversity, Shanghai.Tur n b u l l LA, Cr aw l e y MJ & Re e s M. 2000. Are plantpopulations seed-limited? A review of seed sowingexperiments. Oikos 88: 225–238.Ur i a rt e M, Ca n h a m CD, Th o m p s o n J, Zi m m e r m a n JK & Br o k awN. 2005. Seedling recruitment in a hurricane-driventropical forest, light limitation, density-dependenceand the spatial distribution of parent trees. Journalof Ecology 93: 291–304.Wi l l s C, Co n d i t R, Fo s t e r RB & Hu b b e l l SP. 1997. Strongdensity and diversity-related effects help to maintaintree species diversity in a neotropical forest.Proceedings of the National Academy of Sciences 94:1252–1257.Wr i g h t SJ, Mu l l e r-La n d a u HC, Ca l d e r ó n O & He r n a n d é zA. 2005. Annual and spatial variation in seedfall andseedling recruitment in a neotropical forest. Ecology86: 848–860.Zh a n g J, Ha o Z, Su n IF, So n g B, Ye J, Li B & Wa n g X. 2009.Density dependence on tree survival in an old-growthtemperate forest in northeastern China. Annals ofForest Science 66: 204p1–p9.© Forest Research Institute Malaysia66340

Exploring Tree-Habitat Associations in a ChineseSubtropical Forest Plot Using a Molecular PhylogenyGenerated from DNA Barcode LociNancai Pei 1,2. , Ju-Yu Lian 1. , David L. Erickson 3 , Nathan G. Swenson 4 , W. John Kress 3 , Wan-Hui Ye 1 *,Xue-Jun Ge 1 *1 Key Laboratory of Plant Resource Conservation and Sustainable Utilization, South China Botanical Garden, Chinese Academy of Sciences, Guangzhou, People’s Republicof China, 2 The Graduate University of the Chinese Academy of Sciences, Beijing, People’s Republic of China, 3 Department of Botany, National Museum of Natural HistorySmithsonian Institution, Washington, D.C., United States of America, 4 Department of Plant Biology, Michigan State University, East Lansing, Michigan, United States ofAmericaAbstractElucidating the ecological mechanisms underlying community assembly in subtropical forests remains a central challengefor ecologists. The assembly of species into communities can be due to interspecific differences in habitat associations, andthere is increasing evidence that these associations may have an underlying phylogenetic structure in contemporaryterrestrial communities. In other words, by examining the degree to which closely related species prefer similar habitats andthe degree to which they co-occur, ecologists are able to infer the mechanisms underlying community assembly. Here weimplement this approach in a diverse subtropical tree community in China using a long-term forest dynamics plot and amolecular phylogeny generated from three DNA barcode loci. We find that there is phylogenetic signal in plant-habitatassociations (i.e. closely related species tend to prefer similar habitats) and that patterns of co-occurrence within habitats aretypically non-random with respect to phylogeny. In particular, we found phylogenetic clustering in valley and low-slopehabitats in this forest, indicating a filtering of lineages plays a dominant role in structuring communities in these habitatsand we found evidence of phylogenetic overdispersion in high-slope, ridge-top and high-gully habitats, indicating thatdistantly related species tended to co-occur in these high elevation habitats and that lineage filtering is less important instructuring these communities. Thus we infer that non-neutral niche-based processes acting upon evolutionarily conservedhabitat preferences explain the assembly of local scale communities in the forest studied.Citation: Pei N, Lian J-Y, Erickson DL, Swenson NG, Kress WJ, et al. (2011) Exploring Tree-Habitat Associations in a Chinese Subtropical Forest Plot Using aMolecular Phylogeny Generated from DNA Barcode Loci. PLoS ONE 6(6): e21273. doi:10.1371/journal.pone.0021273Editor: Simon Joly, Montreal Botanical Garden, CanadaReceived December 9, 2010; Accepted May 27, 2011; Published June 20, 2011Copyright: ß 2011 Pei et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricteduse, distribution, and reproduction in any medium, provided the original author and source are credited.Funding: This study was financially supported by National Basic Research Program of China (973 Program) (2007CB411600), China National Program for R & DInfrastructure and Facility Development (2008BAC39B02), the Research Fund for the Large-Scale Scientific Facilities of the Chinese Academy of Sciences (2009-LSF-GBOWS-01), Major Innovation Program of CAS (KSCX2-YW-N-0807) and Key Innovation Project of CAS (KZCX2-YW-430). The funders had no role in study design,data collection and analysis, decision to publish, or preparation of the manuscript.Competing Interests: The authors have declared that no competing interests exist.* E-mail: xjge@scbg.ac.cn (X-JG); why@scbg.ac.cn (W-HY). These authors contributed equally to this work.IntroductionDetermining the ecological and evolutionary processes underlyingcommunity assembly remains a central goal in communityecology. Perhaps nowhere has the debate regarding the assemblyof communities been more vigorous than in tropical treecommunity ecology. Proposed assembly mechanisms invoke therelative importance of niche- [1,2,3] and neutral-based [4,5,6]processes. Tests of these niche- and neutral-based mechanismshave occasionally focused on the degree to which species areassociated with the underlying environment where strongassociations are indicative of niche-based mechanisms dominatingthe assembly process [7,8]. Recent work has demonstrated thatboth species and entire clades have strong associations with soilhabitats [9]. In other words there may often be substantialphylogenetic signal in plant-soil habitat associations where closelyrelated species tend to be found on similar soils. This suggests thatthe evolutionary history of species may help explain their presentday distribution and co-occurrence patterns along habitatgradients and that niche-based process can be detected usingphylogenetic information.The use of phylogenetic information in plant communityecology has dramatically increased since the pioneering work ofWebb [10]. A conceptual framework has emerged from thisliterature that is designed to identify the relative influence of nichebasedversus neutral processes during the assembly of communities.Specifically this conceptual framework integrates thephylogenetic signal in species traits or niches (i.e. the degree oftrait or niche similarity between closely related species) withpatterns of community phylogenetic structure (i.e. phylogeneticclustering or overdispersion) in order to infer the relative influenceof habitat filtering, limiting similarity or neutrality duringcommunity assembly (Table 1).In tropical tree community ecology phylogenetic analyses ofcommunities have generally used one of three approaches: (i) theyhave examined only the phylogenetic structure of communitiesPLoS ONE | www.plosone.org 1 June 2011 | Volume 6 | Issue 6 | e21273341

Phylogenetic Community StructureTable 1. A conceptual framework integrating the degree of phylogenetic signal in plant-soil habitat associations and thephylogenetic structure of the assemblage.Phylogenetically ClusteredAssemblagePhylogenetically RandomAssemblagePhylogeneticallyOverdispersed AssemblagePhylogenetic Signal in Plant-Soil Habitat AssociationsPhylogenetic ‘Anti-Signal’ inPlant-Soil Habitat AssociationsHabitat Filtering Neutrality Limiting SimilarityLimiting Similarity Neutrality Habitat FilteringNiche-based processes (i.e. habitat filtering of limiting similarity) are indicated by a non-random phylogenetic structure of the assemblage, but which processes can onlybe inferred by considering the degree of phylogenetic signal in plant-soil habitat associations. A Neutral model is supported when the assemblage is random withrespect to phylogeny regardless of the degree of phylogenetic signal in plant-soil habitat associations. (Adapted from Kraft NJB, Cornwell WK, Webb CO, Ackerly DD(2007) Trait evolution, community assembly, and the phylogenetic structure of ecological communities. Am Nat 170: 271–283).doi:10.1371/journal.pone.0021273.t001[10,11,12,13]; (ii) they have examined only the phylogenetic signalin plant-soil associations [9] or (iii) they have integrated patterns ofphylogenetic signal in functional traits with patterns of phylogeneticcommunity structure [14,15]. While many of these studieshave inferred the relative influence of niche-based or neutralprocesses, none has successfully implemented the conceptualframework presented in Table 1 where phylogenetic signal ofhabitat associations is integrated with patterns of phylogeneticcommunity structure. This is surprising because a popularexplanatory niche-based mechanism for the maintenance oftropical forest tree species diversity and community assembly ishabitat partitioning [16,17,18]. This mechanism is expected tolead to non-random patterns of species-habitat associations. Ifthese associations have strong phylogenetic signal or ‘anti-signal’(i.e. closely related species have non-randomly diverged in theirsoil habitat associations), then niche-based processes should resultin non-random phylogenetic community structure (Table 1).Although there has been no study that has explicitly linkedlevels of phylogenetic signal in tropical tree-habitat associationswith patterns of phylogenetic community structure, there havebeen two studies that have examined the phylogenetic communitystructure of tropical trees in different habitats. Both of thesestudies have been conducted in the 50-ha Barro Colorado Island(BCI) forest dynamics plot in Panama. The first study wasconducted by Kembel and Hubbell [11] who found that speciesin ‘dry plateau’ and ‘young’ forest habitats tended to be morephylogenetically related than expected by chance (i.e. phylogeneticallyclustered) and species in the ‘slope’ and ‘swamp’ habitatstended to be more distantly related than expected by chance (i.e.phylogenetic overdispersion). Kembel and Hubbell [11] inferredthat in the former case environmental filtering acting onevolutionarily conserved traits was the community assemblymechanism and biotic interactions acting on evolutionarilyconserved traits was the community assembly mechanism in thelatter case. This work was important because a previous studyfrom this forest had found little evidence for species-specifichabitat associations [8]. Thus the discrepancies between the twostudies suggested that analyses that include phylogenetic datamay refine our understanding of the role of niche-based processesduring tropical tree community assembly.The study by Kembel and Hubbell [11] utilized a phylogenetichypothesis estimated by an informatics tool called Phylomatic[19]. This phylogeny contained many unresolved relationships (i.e.soft polytomies) particularly within families. It was unclear at thetime of the Kembel and Hubbell [11] study how much this lack ofphylogenetic resolution influenced their results and inferences.Recently Kress et al. [13] revisited the analyses of Kembel andHubbell [11] using a highly resolved molecular phylogenyconstructed from three DNA barcode loci. They concluded thatmany of the results in the original Kembel and Hubbell [11] studywere not supported when using the resolved molecular phylogeny.Further the results from Kress et al. [13] showed that the work byKembel and Hubbell [11], which relied upon the poorly resolvedPhylomatic phylogeny, generally underestimated the degree ofphylogenetic structuring of the tree communities in the seven BCIhabitats. The stronger patterns of structuring found was taken asevidence that terminal phylogenetic resolution provided by amolecular phylogeny generated using DNA barcode loci iscritical for identifying the underlying phylogenetic structure ofcommunities and that a lack of resolution may lead to type IIstatistical errors as previously suggested by Swenson [20]. Thusthe implementation of phylogenetic information that allows forspecies-level resolution should improve the quality of communityphylogenetic analyses.The majority of the phylogenetic analyses of tree communitiesthat have been performed to date are from North and SouthAmerica or in Southeast Asia. Ideally a greater breadth of foreststhat have distinctive biogeographic histories should be studied inorder to elucidate whether or not any general trends or emergentproperties exist. For example, local species richness is known to behighly correlated with regional scale species richness [21,22]. Thusregional scale differences in species richness likely plays apredominant role in determining differences in local scale richnessfrom region to region. That said, it is still of interest how local scaleprocesses ‘scale-up’ to produce differences in regional speciesrichness and/or whether the strength of niche-based or neutralprocesses varies between local communities occurring in differentregions [23]. In other words, are local scale niche-based processessuch as habitat filtering uniformly important in two regions withvastly different levels of biodiversity? In order to answer such aquestion, researchers must continue to sample, analyze andcompare the phylogenetic structure of tree communities from asmany regions as possible.Here we utilize a molecular phylogeny constructed from threeDNA barcode loci (rbcL, matK, and trnH-psbA) representing 183woody plant species in the 20-ha Dinghushan forest dynamics plotin China. The phylogeny, observed spatial distribution patterns forthe 188 species in different habitats and a conceptual frameworkthat integrates the degree of phylogenetic signal in plant-habitatassociations with the phylogenetic structure of communities areused to ask the central question of whether niche-based (i.e. habitatpartitioning or limiting similarity) or neutral processes determinethe assembly of species in this subtropical seasonal forest? Inanswering this central question we also compare and contrastresults of the analyses when utilizing a molecular phylogeny versusa phylogeny derived from Phylomatic.PLoS ONE | www.plosone.org 2 June 2011 | Volume 6 | Issue 6 | e21273342

Phylogenetic Community StructureMaterials and MethodsResearch site and DNA sequencingThe present study took place in the Dinghushan forest dynamicsplot (DHS FDP) located within the Dinghushan National NatureReserve (23u099210–23u119300N, 112u309390–112u339410E) inGuangdong province, south China. The DHS FDP is a key nodein the Chinese Forest Biodiversity Monitoring Network and a partof the Center for Tropical Forest Science (CTFS) global networkof forest dynamics plots, The DHS FDP is a subtropical forest witha mean annual rainfall of 1,985 mm. A total number of 71,336woody stems greater than or equal to 1 cm diameter at breastheight have been mapped and identified to species in the400 m6500 m plot. There are 183 species (188 taxa) of treesand shrubs in the DHS FDP. These 183 species represent 24orders, 51 families, and 110 genera (38 genera containing $ twospecies; 10 genera containing $ four species; and two generacontaining $ eight species).A molecular community phylogeny was generated for the 183species in the DHS FDP by sequencing three DNA barcoding loci(rbcL, trnH-psbA, and matK). DNA sequences were generated for 1–2 tagged individuals located within the plot. Genomic DNA wasextracted from leaf and/or bark tissue using a standard CTABprotocol [24]. Standard barcode primers (rbcL, matK, and trnHpsbA)suggested by the Consortium for the Barcode of Life (http://barcoding.si.edu/) were used in the study. The PCR cyclingconditions utilized in this study were as follows: rbcL and trnH-psbAused 95uC for 3 min, (95uC 30 sec, 53uC 45 sec, 72uC 1 min)634cycles, 72uC 7 min, while matK required lower annealingtemperatures, longer extension time and more cycles (95uC3 min (95uC 30 sec, 51uC 45 sec, 72uC 1.5 min)638 cycles,72uC 7 min) [25], adding a final concentration of 5% for DMSO.Sequences of rbcL (,650 bp for the sequence length) which can besequenced via one reaction, had 1-fold coverage, but the matK(,900 bp) and trnH-psbA (280–870 bp) had 2-fold coverage. AllDNA sequence data were submitted to GenBank and theiraccession numbers are provided in Table S1.Phylogenetic reconstructionWe reconstructed two types of community phylogeniesrepresenting the 183 plant species found in the DHS FDP. Thefirst phylogeny we generated was a molecular phylogeny using thethree sequenced barcode loci described in the previous section. ADNA supermatrix was generated that contained all three markers(rbcL + matK + trnH-psbA) ([13]; see also Text S1 for more detail onalignment and matrix construction). The DNA supermatrix wasthen analyzed using RA6ML [26] via the CIPRES supercomputercluster [27] to infer a maximum likelihood (ML) communityphylogeny. Node support was estimated using bootstrap valueswith nodes with less than 50% support being collapsed into softpolytomies. The familial topology in this molecular phylogeny wasconcordant with that observed in the APG III classification. InFig. 1 we show a comparison of the family-level relationshipswithin the Asterids clade between our ML analysis of the barcodesequence data and that derived by the APG III.A second phylogenetic tree was generated for this work usingthe informatics tool Phylomatic [19]. This tree ‘grafts’ taxonomicrelationship to a stored phylogenetic ‘backbone’ is generallyresolved to the family level. Thus relationships between specieswithin a genus and genera within a family are generally levelunresolved. Community phylogenies derived from Phylomatic arethe typical approach in community phylogenetics investigations,because molecular phylogenies of most tropical taxa are notavailable. As such the Phylomatic tree in this study was generatedFigure 1. A comparison of the family-level relationships withinthe Asterid clade. The topology on the left-hand side represents thephylogenetic relationships of families obtained from the APG IIIconsensus phylogeny, while the topology on the right-hand siderepresents the DHS phylogeny generated with the ML analysis of thebarcode sequence data.doi:10.1371/journal.pone.0021273.g001in order to compare whether any information would be lost if onlya Phylomatic tree, exhibiting lower rates of resolution, was utilized.Habitat types and spatial scales classificationFive habitat types in the DHS FDP were classified using thetopographic variables slope, elevation and convexity [28,29]. Inparticular, habitats were classified using a quantitative methodwhere the observed slope and elevation was compared to the plotmedian value. The specific classification scheme is given inTable 2. The quantitative classification of habitat types allows forthem to be ordered by similarity, as valley (V), low-slope (LS),high-gully (HG), ridge-top (RT) and high-slope (HS). A habitattype was assigned to each given 20620 m quadrat in the DHSFDP. Topographical variations in the DHS FDP are larger thanthat of the BCI forest plot in Panama (Table 2, and Fig. 2). Thus itis difficult to compare the habitat types of the two plots. Themajority of the analyses were conducted by dividing the 20-ha plotinto 500 20 m620 m quadrats. Two additional spatial scales wereused, specifically 40 m640 m and 100 m6100 m. In sum, fivePLoS ONE | www.plosone.org 3 June 2011 | Volume 6 | Issue 6 | e21273343

Phylogenetic Community StructureTable 2. Criterions of habitat classification, areas of each habitat, total numbers of species and stems $1-cm d.b.h. in 2005 census,and total stem densities by habitat for the 20-ha Forest Dynamics Plot of Dinghushan, China.Habitat Valley High-gully Low-slope High-slope Ridge-topArea (ha) 6.92 3.08 4.60 2.92 2.48Slope (degrees) ,33 $33 $33 $33 ,33Elevation (m) ,326.3 $326.3 ,326.3 $326.3 $326.3Convexity (degrees) all ,0 all . .0Mean 6 s.e. (species diversity) 25.5860.46 28.2760.88 27.5760.48 34.7360.84 27.5360.85Total number of species 149 133 135 135 105Total number of stems [density (no.ha 21 )] 19,501 (2828.06) 11,052 (3588.31) 17,215 (3742.39) 14,174 (4854.11) 9,394 (3787.90)Notes: Valley (slope , median (slope), elevation , median (elevation)); Low-slope (slope $ median(slope), elevation , median(elevation)); High-slope (slope $ median(slope), elevation $ median(elevation), convexity .0); High-gully (slope $ median(slope), elevation $ median(elevation), c$onvexity ,0); Ridge-top (slope , median(slope), elevation $ median(elevation), convexity .0).Median slope = 33 degrees; Median elevation = 326.3 m.doi:10.1371/journal.pone.0021273.t002habitat types and three spatial scales were used to quantify thecommunity phylogenetic structure in the DHS FDP.Habitat association – randomization tests andphylogenetic signalIn order to quantify the degree to which individual species in theDHS FDP are associated with specific habitat types we used thehabitat randomization procedure described in Harms et al. [10].Specifically a torus translation was utilized where the habitat mapwas ‘rotated’ or iterated. During each iteration, a ‘null’ specieshabitatassociation was calculated for each of the 99 most commontree species. These 99 species were selected because they werecommon enough (n.20) to provide a robust estimate of theirhabitat association. This generated a null distribution to which wecould compare the observed association. Each simulated mapincluded 173 valley, 77 high-gully, 115 low-slope, 73 high-slope,and 62 ridge-top quadrats. This randomization proceduremaintains the observed spatial autocorrelation of both the habitatdata and the species distributions.We also quantified the phylogenetic signal in plant-habitatassociations in order to implement the conceptual frameworkpresented in Table 1. Phylogenetic signal was measured on themedian habitat in which individuals of each species are foundusing the five habitat categories as ordered variables as describedin the above section. The descriptive statistic K presented inBlomberg et al. [30] was used to measure the phylogenetic signalin habitat associations. The significance of the observed K valuewas determined using a permutation test. Specifically, the namesof taxa were randomized across the tips of the phylogeny 999times. During each iteration, a null K value was calculated andrecorded. This generated a distribution of 999 null K values towhich the observed could be compared.Community phylogenetic structure analysesOne of the 19 equally likely phylogenetic trees of the three-locusML analysis of 183 species was randomly selected to use in thepresent community phylogenetic analysis. Non-parametric ratesmoothing in the R package ‘ape’ [31,32] was used to generate anultrametric phylogeny. This ultrametric barcode phylogeny wasused for all subsequent community phylogenetic analyses.Using both the molecular and Phylomatic phylogenies, wequantified the Net Relatedness Index (NRI) and the NearestTaxon Index (NTI) [33,34] for each 400 m 2 quadrats (n = 500).The NRI and NTI are calculated as follows:NRI ~{1 | (MPD { rndMPD)=sdrndMPDNTI ~{1 | (MMPD { rndMMPD)=sdrndMMPDFigure 2. The spatial distribution of the five habitat types inthe 20-ha Dinghushan plot. Colors represent different habitat typesat the spatial scale of 20 m620 m.doi:10.1371/journal.pone.0021273.g002Where MPD represents the mean pairwise phylogenetic distancebetween all taxa within a local assemblage and MMPD representsthe mean phylogenetic distance for each taxa to its nearest relativewithin a local assemblage. The rndMPD and rndMMPD representthe mean MPD and mean MMPD from 999 randomly generatedassemblages. An independent swap null model was used togenerate these 999 random assemblages [35]. This is the same nullmodel as the ‘constrained’ null model in Kembel and Hubbell[11]. The NRI is generally considered to be a ‘basal’ metric whilethe NTI is generally considered to be a ‘terminal’ metric. NegativePLoS ONE | www.plosone.org 4 June 2011 | Volume 6 | Issue 6 | e21273344

Phylogenetic Community Structurevalues of both metrics indicate phylogenetic overdispersion. Inother words species in local assemblages are more phylogeneticallydiverse than expected by chance. Positive values of both metricsindicate phylogenetic clustering. In other words species in localassemblages are more closely related than expected by chance.Because the NRI and NTI values in the 500 quadrats werespatially autocorrelated, we estimated the mean NRI and NTIvalues within habitats using simultaneous spatial autoregressionanalyses. We used generalized least-squares models with a firstorderspatial neighbor SAR covariance structures in S+Spatial-Stats [36] to perform these analyses. Next, following Kembel andHubbell [11], we defined habitats for each 400 m 2 quadrat andtested whether each habitat type tended to contain quadrats thatwere on average phylogenetically clustered, overdispersed orrandom using t-tests.Then, following the methods of Kress et al. [13], we askedwhether the results for each individual habitat type generated fromthe molecular phylogeny and the Phylomatic phylogeny weresignificantly different. For all of the 500 quadrats combined, wecompared the NRI and NTI values quantified from the molecularphylogeny to those calculated from the Phylomatic phylogenyusing a paired t-test.Lastly, we performed all analyses using the species lists in thefive habitats as individual communities and the forest plot specieslist as the species pool. This analysis was designed to addresswhether or not the entire species assemblage in a habitat wasphylogenetically non-random.ResultsHabitat-species association and community assemblyThe habitat association tests recovered 52 significant positiveand negative plant-habitat associations out of a potential 495species-habitat combinations. Thus 10.5% of the tests werepositive, which is greater than the expected false discovery rateof 5%. There were 29 significant positive or negative associationsin the habitats that were phylogenetically overdispersed (23 for thehigh-slope habitat and 5 for the high-gully habitat) or phylogeneticallyclustered (one for the low-slope habitat, but no significantpositive or negative associations in the valley habitat). Another 23significant positive or negative associations were found in theridge-top habitat which contained phylogenetically randomassemblages. A total of 52 of the 99 most common species had asignificantly positive or negative association with at least onehabitat type (Table 3). The detailed results regarding which specieswere associated with individual habitat types are provided asSupplemental Material (Text S2).The 19 most abundant species accounted for 74.77% of all stemsin the DHS FDP. Of these species 12 had at least one positive ornegative habitat association, while 7 were not significantlyassociated with a habitat (Table 3). We detected significantdifference (F = 26.414, P,0.001) in species richness between theforest communities that were phylogenetically clustered (30.2860.54, mean 6 se) and those that were phylogenetically overdispersed(26.3860.34) using ANOVA. We further found that thehigh-slope habitat had the highest species richness (34.7360.84),followed by habitats of the high-gully (28.2760.88), the low-slope(27.5760.48), the ridge-top (27.5360.85) and the valley(25.5860.46) (Table 2).Phylogenetic signal in habitat associationsWe utilized the descriptive statistic K to quantify the phylogeneticsignal in habitat associations using the median habitat type for allindividuals of a species and treating habitat type as an orderedvariable. The observed K value was 0.80 and this value wascompared to a distribution of 999 null K values generated with apermutation test. The observed K value was significantly higherthan that expected (P = 0.019) using a two-tailed test. A higher thanexpected K value indicates there is phylogenetic signal in specieshabitatassociations. In other words closely related species tended tobe more similar in their habitat associations than expected.Community phylogenetic structures in different habitattypesA total of five habitat types were classified in the DHS FDP andthey are mapped using different colors in Fig. 2. The NetRelatedness Index (NRI) value and the Nearest Taxon Index(NTI) value in each quadrat is marked in the 20-ha plot in Fig. 3.The results from the Phylomatic phylogeny found phylogeneticclustering in the valley habitat using both the NRI and the NTImetrics, and in the low-slope habitat using the NRI metric.Phylogenetic overdispersion was found in the high-gully, the highslope,and the ridge-top habitats using both the NRI and NTImetrics (Table 4 and Fig. 3). When using the molecular phylogeny,we found significant phylogenetic structuring in six of the 10 tests(two metrics per habitat type). Specifically we found phylogeneticclustering in the valley and the low-slope habitats, phylogeneticoverdispersion in the high-slope and the ridge-top habitats, and aphylogenetically random pattern in the high-gully habitat usingboth the NRI and the NTI metrics.When comparing our molecular phylogeny to the less wellresolvedPhylomatic tree, five of ten inferences were similar.Analyses based on the molecular phylogeny identified significantTable 3. Randomized-habitat tests for habitat associations on the 20-ha Forest Dynamics Plot of Dinghushan, China.Habitat association 99 species 19 species Habitat association 99 species 19 speciesValley + 0 0 Valley - 0 0High-gully + 4 1 High-gully - 1 0Low-slope + 1 1 Low-slope - 0 0High-slope + 22 2 High-slope - 1 0Ridge-top + 11 4 Ridge-top - 12 4Total + 38 8 Total - 14 4The first column contains results for the 99 common species for which there were $20 stems in the plot in the 2005 census. The second column contains results for the19 most abundant species, all of which had $1000 stems in the plot in the 2005 census. For each habitat, ‘‘+’’ indicates significant positive association and ‘‘2’’ indicatessignificant negative association (two-tailed test, a = 0.05).doi:10.1371/journal.pone.0021273.t003PLoS ONE | www.plosone.org 5 June 2011 | Volume 6 | Issue 6 | e21273345

Phylogenetic Community StructureFigure 3. The spatial patterns of NRI and NTI values in the forest plot. Values of NRI and NTI for each 400 m 2 quadrat in the 20-ha forestdynamics plot in Dinghushan, south China, are calculated using the molecular phylogeny and the Phylomatic phylogeny. Negative NRI and NTI valuesindicate phylogenetic overdispersion and positive values indicate phylogenetic clustering. The color scales across all NRI and NTI maps are madeequivalent to allow for direct visual comparisons between the four maps. a. Barcode NRI; b. Phylomatic NRI; c. Barcode NTI; d. Phylomatic NTIdoi:10.1371/journal.pone.0021273.g003phylogenetic structuring in the low-slope habitat using the NTImetric for which the Phylomatic phylogeny did not. In theremaining four cases, the Phylomatic phylogeny demonstratedsignificant phylogenetic structuring but the molecular phylogenydid not. It is important to note that although the molecularphylogeny generally provided higher NRI and NTI values, this didTable 4. The estimated mean and standard error of the NRI and the NTI values in the DHS habitat types estimated using first ordersimultaneous spatial autoregression for the molecular phylogeny (columns labeled ‘‘Molecular NRI/NTI’’) or for the Phylomaticphylogeny (columns labeled ‘‘Phylomatic NRI/NTI’’).Habitat type/Spatial scale Molecular NRI Phylomatic NRI NRI difference Molecular NTI Phylomatic NTI NTI differenceValley 0.6160.10 *** 0.5360.09 *** 0.0860.09 0.02±0.09 0.40±0.08 *** 20.38±0.13 **High-gully 20.07?0.13 20.45±0.13 ** 0.38±0.13 ** 20.01±0.12 20.41±0.12 ** 0.40±0.18 *Low-slope 0.3260.09 ** 0.2860.10 ** 0.0460.09 0.19±0.08 * 0.18±0.10 0.0160.14High-slope 20.4560.12 *** 20.4860.12 *** 0.0360.11 20.2860.11 * 20.4760.09 *** 0.1960.14Ridge-top 20.5760.09 *** 20.6560.11 *** 0.0860.13 20.11±0.12 20.41±0.09 *** 0.30±0.17 *20 m620 m 0.14160.053 ** 0.02260.052 0.11960.048 * 20.00860.047 20.00660.046 20.00260.06940 m640 m 0.10460.103 0.01260.097 0.09260.108 0.00960.088 20.02360.090 0.03260.140100 m6100 m 0.07160.247 20.00360.137 0.07460.261 0.15760.025 0.17460.145 20.01760.255Notes: The P values in the ‘‘Molecular NRI/NTI’’ and ‘‘Phylomatic NRI/NTI’’ columns were calculated using a two-tailed t-test to assess whether the mean NRI and NTIvalues in the habitat types and spatial scales were higher or lower than expected. Negative values indicate that the observed average NRI or NTI was phylogeneticallyoverdispersed. Positive values indicate that the observed average NRI or NTI score was phylogenetically clustered. The columns labeled ‘‘NRI or NTI difference’’ providedthe mean of the difference between the molecular phylogeny and Phylomatic NRI and NTI values in each habitat type or spatial scale and were calculated using a twotailedpaired t-test to assess whether the NRI and NTI values in a habitat type or spatial scale calculated from the barcode phylogeny were significantly different thanthose calculated using the Phylomatic phylogeny. We found all differences among habitats in NRI and NTI were statistically significant according to the spatial GLS tests(P,0.01). The asterisk ***, **or * indicate the significance at the level of P,0.001, 0.01 or 0.05 respectively.doi:10.1371/journal.pone.0021273.t004PLoS ONE | www.plosone.org 6 June 2011 | Volume 6 | Issue 6 | e21273346

Phylogenetic Community Structurenot necessarily mean these values were more often non-random.For example, an insignificant mean NTI value of 20.01 in the HGhabitat was recorded using the molecular phylogeny, but a lowerand significant mean NTI value of 20.41 was recorded in thishabitat using the Phylomatic phylogeny (Table 4). When directlycomparing the NRI and NTI values from the 500 individualquadrats within the five habitat types using paired t-tests, four outof the ten comparisons were significantly different whencomparing the results from the molecular and Phylomaticphylogenies (Table 4). We also quantified the NRI and NTI foreach habitat type using all species found in a habitat as theassemblage. We found that NRI values from the molecularphylogeny were positive in seven cases and negative in two otherscases while NRI values from the Phylomatic phylogeny werepositive in two occasions and negative in two cases. The NTIvalues from the molecular phylogeny were negative in nine caseswhile the NTI values from the Phylomatic phylogeny were positivein four cases and negative in six cases (Fig. 4).DiscussionThe niche versus neutral debate is particularly important intropical forest community ecology given the elevated levels ofspecies richness and often low population sizes of many species inthese systems. Previous analyses of tropical tree communities havesuggested that lineages non-randomly sort into different habitattypes [11] thereby indicating the potential importance of nichebasedprocesses during tropical tree community assembly. Beyondsimply finding support for niche-based assembly, this phylogenetically-basedanalysis was important in that it detected non-randomhabitat associations that traditional species-based analyses couldnot detect [8]. This work has been important for our understandingof tropical tree community assembly and for its depiction of theadditional information that may be gleaned when using phylogenetictrees. That said, this work comes from a single forest plot inPanama and similar studies from other tropical regions could helpdetermine the generality of these findings. Further, the nonrandomsorting of lineages into different habitat types in Panamawas not integrated with information regarding the degree to whichclosely related species share similar habitat preferences. In otherwords whether or not there is phylogenetic signal in species-habitatassociations. The present study aimed to quantify whether thephylogenetic structure of a sub-tropical Chinese tree communitywas non-random across habitats as what has been done inPanama. Next, we quantified whether there was phylogeneticsignal in plant-habitat associations – information critical forinferring which ecological process has influenced communityassembly the most (Table 1). Finally, previous work has shown thatmolecular phylogenies generated from three DNA barcode locimay provide substantially different results than those generatedFigure 4. The total distributions of NRI or NTI values in different habitats generated from the molecular and Phylomaticphylogenies. The solid black line across the color box represents the median value. A hollow circle indicates an outlier value of NRI or NTI. HS, Highslope;RT, Ridge-top; HG, High-gully; LS, Low-slope; V, Valley.doi:10.1371/journal.pone.0021273.g004PLoS ONE | www.plosone.org 7 June 2011 | Volume 6 | Issue 6 | e21273347

Phylogenetic Community Structureusing a less-well resolved Phylomatic phylogeny [13]. In this studywe compare and contrast the results from a molecular phylogenyto those to a Phylomatic phylogeny to determine whether theresults of previous work [13] are generally applicable. In thefollowing we discuss the results of our study with respect tocommunity assembly and the use of molecular versus Phylomaticphylogenies.Community Assembly, habitat specialization and speciesdiversityPhylogenetic investigations of plant communities have been usedto determine whether patterns of species co-occurrence havephylogenetic structure [34,37]. It is recognized that non-randomphylogenetic structure (phylogenetic overdispersion or clustering)indeed exists in animal, plant and microbial communities [38,39,40]where approximately sixty percent of previous studies have foundevidence for phylogenetic clustering in contemporary terrestrial andplant communities [41]. Two general types of niche-based processescan produce these patterns of non-random phylogenetic communitystructure – environmental filtering and strong negative or positivebiotic interactions. Environmental filtering during communityassembly dictates that only a small subset of species share similarecological strategies or niches can co-occur in a given environment.If closely related species have similar strategies or niches, thenenvironmental filtering should produce a pattern of phylogeneticclustering. Conversely, if closely related species have very divergentstrategies or niches, then environmental filtering should result inphylogenetic overdispersion. Non-random biotic interactions (i.e.competition, facilitation, etc) dictating community assembly shouldgenerate a pattern of phylogenetic overdispersion if closely relatedspecies are similar or phylogenetic clustering if closely relatedspecies are very dissimilar. Thus while non-random patterns ofphylogenetic community structure are indicative of niche-basedprocesses, it is not possible to identify which process withoutinformation pertaining to the similarity of closely related species (i.e.phylogenetic signal) (Table 1).The present study found that many species (.10%) have asignificant positive or negative association with a habitat type in theDHS FDP. Subsequent analyses of the phylogenetic signal inspecies-habitat associations found that there was significantphylogenetic signal. Thus while only some species were significantlyassociated with a particular habitat, closely related species did onaverage tend to be associated with similar habitats. Thus anyobserved patterns of phylogenetic clustering in this system should beindicative of habitat filtering while patterns of phylogeneticoverdispersion should be indicative of biotic interactions.The phylogenetic structuring analyses showed that mean NRI ofall 0.04-ha quadrats (i.e. local assemblages) was significantlydifferent from the null expectation of zero (P = 0.008). Thisdeviation indicates niche-based community assembly in this forest,but the inferred process varies with the habitat considered. Inparticular, both the valley and the low-slope habitats had assemblagesthat were phylogenetically clustered. We therefore infer thathabitat filtering drives the assembly of the communities and cooccurrenceof species in these two habitats. In the high-slope andridge-top habitats communities were generally phylogeneticallyoverdispersed indicating a large role for biotic interactions drivingthe assembly and co-occurrence of species in these habitats.Interestingly these habitats are apt to suffer drought and they likelyhave low soil nutrients concentrations. Thus it is possible thatfacilitation influences co-occurrence and therefore generates apattern of dissimilar co-occurring species. Lastly, the high-gullyspecies assemblages were no different from those expected bychance, which suggests one of two possibilities. First neutrality maydominate the assembly process in these habitats. Second thestrength of habitat filtering is ‘balanced’ by the strength of bioticinteractions resulting in a random pattern from non-randomprocesses acting in opposing directions [15].The results showed that the phylogenetic structure of the speciesin an entire habitat type often mirrored those found in individualquadrats within that habitat type (Table S2). This finding suggeststhat the filtering of lineages at the ‘habitat-scale’ largely explainsthe local-scale phylogenetic pattern. We do note, however, that insome instances this was not the case where the habitat-scalepattern was not found locally. This suggests that non-randomecological interactions within habitats may play a large role indetermining local phylogenetic structure.Comparative analyses of molecular and PhylomaticphylogeniesPrevious work has suggested that the lack of terminal resolutionin phylogenies generated by the informatics program Phylomaticmay bias investigations of community phylogenetic structure[13,20]. One of these studies was conducted in Panama [13] andthe other was simulation based [20]. Therefore it is unclear howgeneralizable the findings are to other systems. Thus additionalstudies that compare the results generated from the resolvedmolecular phylogenies to those from a Phylomatic phylogeny areneeded. The present study has performed such a comparison.The results from the molecular phylogeny generated from threeDNA barcode loci had higher values of both NRI and NTI metricsthan those generated using the Phylomatic tree in nine out of tencomparisons (Table 4). In other words the results from the resolvedmolecular phylogeny tended to show more phylogenetic clusteringthan that found using the Phylomatic phylogeny. This result issimilar to that found in Panama [12] and in previous simulationwork [20] that suggests that increased resolution provided bymolecular phylogenies should allow for the detection of nonrandomphylogenetic community structure that a Phylomaticphylogeny cannot detect. In other words, the lack of resolution in aPhylomatic phylogeny likely leads to Type II statistical errors.ConclusionsThe present study sought to determine whether niche-based orneutral processes dominate the assembly of tree communities in asub-tropical Chinese forest. The work quantified the phylogeneticstructure of tree communities in five habitat types and thephylogenetic signal in plant-habitat associations. Using a conceptualframework that integrates the level of phylogenetic signal inplant-habitat associations with the phylogenetic dispersion ofspecies in a community (Table 1) we infer that niche-basedprocesses (habitat filtering and facilitation) drive the assembly ofcommunities in this forest. These results are consistent withfindings from a similar study in Panama [13] suggesting that localscaleniche-based processes are important in both of these regionsdespite their very different biogeographic histories and regionalspecies pool compositions. The work also provides furtherevidence that less well-resolved Phylomatic phylogenies tend togenerate Type II statistical errors and that utilizing resolvedmolecular phylogenies is therefore advised when feasible. It issuggested that the feasibility of generating such molecularcommunity phylogenies is enhanced through the utilization ofsequence data from three commonly used DNA barcoding loci(rbcL, trnH-psbA, and matK). We expect that as barcoding becomesmore widespread, community phylogenetics researchers willbenefit from ‘tapping into’ the vast resource that is a DNAbarcode library.PLoS ONE | www.plosone.org 8 June 2011 | Volume 6 | Issue 6 | e21273348

Phylogenetic Community StructureSupporting InformationTable S1 A list of taxa, GenBank accession numbers,and tree tag numbers.(DOC)Table S2 The estimated mean and standard error (S.E.)of the NRI and NTI values in the DHS habitat typesestimated using first order simultaneous spatial autoregressionfor the barcode phylogeny (columns labeled‘‘Barcode NRI/NTI’’) or for the phylomatic phylogeny(columns labeled ‘‘Phylomatic NRI/NTI’’), using habitatsample files.(DOC)Text S1 Sequence editing and alignment of threebarcode loci.(DOC)References1. Diamond JM (1975) Assembly of species communities. In: Cody ML,Diamond JM, eds. Ecology and Evolution of Communities. Cambridge:Belknap Press of Harvard University Press. pp 342–444.2. Weiher E, Keddy P (1999) Ecological assembly rules: perspectives, advances,retreats. Cambridge, UK: Cambridge University Press.3. Tilman D (1982) Resource Competition and Community Structure. PrincetonNJ:Princeton University Press.4. Kelly CK, Bowler MG, Pybus O, Harvey PH (2008) Phylogeny, niches, andrelative abundance in natural communities. Ecology 89: 962–970.5. Jabot F, Chave J (2009) Inferring the parameters of the neutral theory ofbiodiversity using phylogenetic information and implications for tropical forests.Ecol Lett 12: 239–248.6. Hubbell SP (2001) The unified neutral theory of biodiversity and biogeography.Princeton: Princeton University Press.7. John R, Dalling JW, Harms KE, Yavitt JB, Stallard RF, et al. (2007) Soilnutrients influence spatial distributions of tropical tree species. Proc Natl AcadSci USA 104: 864–869.8. Harms KE, Condit R, Hubbell SP, Foster RB (2001) Habitat associations oftrees and shrubs in a 50-ha neotropical forest plot. J Ecol 89: 947–959.9. Schreeg LA, Kress WJ, Erickson DL, Swenson NG (2010) Phylogenetic analysisof local-scale tree soil associations in a lowland moist tropical forest. PLoS ONE5: e13685.10. Webb CO (2000) Exploring the phylogenetic structure of ecological communities:an example for rain forest trees. Am Nat 156: 145–155.11. Kembel SW, Hubbell SP (2006) The phylogenetic structure of a neotropicalforest tree community. Ecology 87: S86–S99.12. Swenson NG, Enquist BJ, Pither J, Thompson J, Zimmerman JK (2006) Theproblem and promise of scale dependency in community phylogenetics. Ecology87: 2418–2424.13. Kress WJ, Erickson DL, Jones FA, Swenson NG, Perez R, et al. (2009) PlantDNA barcodes and a community phylogeny of a tropical forest dynamics plot inPanama. Proc Natl Acad Sci USA 106: 18621–18626.14. Swenson NG, Enquist BJ (2007) Ecological and evolutionary determinants of akey plant functional trait: wood density and its community-wide variation acrosslatitude and elevation. Amer J Bot 94: 451–459.15. Swenson NG, Enquist BJ (2009) Opposing assembly mechanisms in aneotropical dry forest: implications for phylogenetic and functional communityecology. Ecology 90: 2161–2170.16. Tilman D, Pacala SW (1993) The maintenance of species richness in plantcommunities. In: Ricklefs RE, Schluter D, eds. Species diversity in ecologicalcommunities: historical and geographical perspectives. Chicago, USA: Universityof Chicago Press. pp 13–25.17. Hubbell SP, Foster RB (1983) Diversity of canopy trees in a neotropical forestand implications for conservation. In: Sutton SJ, Whitmore TC, Chadwick AC,eds. Tropical rain forest: ecology and management. Oxford, UK: BlackwellScience. pp 25–41.18. Li L, Huang Z, Ye W, Cao H, Wei S, et al. (2009) Spatial distributions of treespecies in a subtropical forest of China. Oikos 118: 495–502.19. Webb CO, Donoghue MJ (2005) Phylomatic: tree assembly for appliedphylogenetics. Mol Ecol Notes 5: 181–183.20. Swenson NG (2009) Phylogenetic resolution and quantifying the phylogeneticdiversity and dispersion of communities. PLoS ONE 4: e4390.21. Ricklefs RE (1987) Community diversity: relative roles of local and regionalprocesses. Science 235: 167–171.Text S2 A description of which individual species wereassociated with particular habitat types.(DOC)AcknowledgmentsWe thank Campbell O. Webb for discussing running the softwarePhylocom, and Jin-Long Zhang for instructing R-package usage. We alsothank Dr. Simon Joly and two anonymous reviewers for their helpfulcomments and criticisms on earlier versions of this manuscript.Author ContributionsConceived and designed the experiments: X-JG W-HY. Performed theexperiments: NP. Analyzed the data: NP J-YL DLE NGS. Contributedreagents/materials/analysis tools: J-YL W-HY. Wrote the paper: NP NGSDLE WJK X-JG W-HY.22. Caley MJ, Schluter D (1997) The relationship between local and regionaldiversity. Ecology 78: 70–80.23. Swenson NG (2011) The role of evolutionary processes in producing biodiversitypatterns, and the interrelationships between taxonomic, functional andphylogenetic biodiversity. Amer J Bot 98: 472–480.24. Doyle JJ, Doyle JL (1987) A rapid DNA isolation procedure for small quantitiesof fresh leaf tissue. Phytochem Bull 19: 11–15.25. Fazekas AJ, Burgess KS, Kesanakurti PR, Graham SW, Newmaster SG, et al.(2008) Multiple multilocus DNA barcodes from the plastid genome discriminateplant species equally well. PLoS ONE 3: e2802.26. Stamatakis A (2006) RAxML-VI-HPC: Maximum likelihood-based phylogeneticanalyses with thousands of taxa and mixed models. Bioinformatics 22:2688–2690.27. Miller M, Holder M, Vos R, Midford P, Liebowitz T, et al. (2009) The CIPRESPortals. Available: http://www.phylo.org/sub_sections/portal. Accessed 2009Aug 4.28. Wang Z, Ye W, Cao H, Huang Z, Lian J, et al. (2009) Species-topographyassociation in a species-rich subtropical forest of China. Basic Appl Ecol 10:648–655.29. Wang ZG (2007) Species diversity and mechanisms for maintenance of monsoonevergreen broadleaved forest in Dinghushan [PhD Thesis]. Guangzhou: SouthChina Botanical Garden, Chinese Academy of Sciences.30. Blomberg SP, Garland T, Ives AR (2003) Testing for phylogenetic signal incomparative data: behavioral traits are more labile. Evolution 57: 717–745.31. Paradis E, Bolker B, Claude J, Cuong HS, Desper R, et al. (2010) Analyses ofPhylogenetics and Evolution. Version: 2.5-3. Available: http://ape.mpl.ird.fr/.Accessed 2011 Mar 10.32. R Development Core Team (2008) R: a language and environment for statisticalcomputing, R foundation for statistical computing, Vienna, Austria. Version2.10. Available: http://www.R-project.org. Accessed 2011 Apr 13.33. Webb CO, Ackerly DD, Kembel SW (2008) Phylocom: software for the analysisof phylogenetic community structure and trait evolution. Bioinformatics 24:2098–2100.34. Webb CO, Ackerly DD, McPeek MA, Donoghue MJ (2002) Phylogenies andcommunity ecology. Annu Rev Ecol Syst 33: 475–505.35. Gotelli NJ, Entsminger GL (2001) Swap and fill algorithms in null modelanalysis: rethinking the knight’s tour. Oecologia 129: 281–291.36. Kaluzny SP, Vega SC, Cardoso TP, Shelly AA (1998) S+SpatialStats: user’smanual for Windows and UNIX. New York, USA: Springer-Verlag, New York.37. Cavender-Bares J, Kozak KH, Fine PVA, Kembel SW (2009) The merging ofcommunity ecology and phylogenetic biology. Ecol Lett 12: 693–715.38. Swenson NG, Enquist BJ, Thompson J, Zimmerman JK (2007) The influence ofspatial and size scale on phylogenetic relatedness in tropical forest communities.Ecology 88: 1770–1780.39. Rabosky DL, Reid J, Cowan MA, Foulkes J (2007) Overdispersion of body sizein Australian desert lizard communities at local scales only: no evidence for theNarcissus effect. Oecologia 154: 561–570.40. Horner-Devine MC, Bohannan BJM (2006) Phylogenetic clustering andoverdispersion in bacterial communities. Ecology 87: S100–S108.41. Vamosi SM, Heard SB, Vamosi JC, Webb CO (2009) Emerging patterns in thecomparative analysis of phylogenetic community structure. Mol Ecol 18:572–592.PLoS ONE | www.plosone.org 9 June 2011 | Volume 6 | Issue 6 | e21273349

Short Note: Isolation and Characterization of Microsatellite Lociin Castanopsis fissa in Lower Subtropical ChinaBy LEI DONG 1),2),3) , ZHENG-FENG WANG 1),2),4) , Peng Zhu 1),2),3) and WAN-HUI YE 1),2)(Received 22 th November 2009)AbstractWe report on the development and characterization often microsatellite markers from repetitive DNAenriched libraries for Castanopsis fissa from lower subtropicalChina. The number of alleles ranged from threeto thirteen. Observed and expected heterozygositiesranged from 0.265 to 0.818, and 0.270 to 0.873, respectively.These microsatellite markers will be used tostudy fine-scale spatial genetic structure of C. fissa in20 ha Dinghushan plot in lower subtropical China.Key words: Castanopsis fissa, microsatellite, genetic marker,population genetics, lower subtropical China, reforestation,spatial genetic structure, marker development, DNA enrichedlibraries, Hardy-Weinberg equilibrium, linkage disequilibrium,Dinghushan.Castanopsis fissa Rehd. & Wils. (Fagaceae) is a fastgrowingbroad-leaved evergreen tree, widely distributedin lower subtropical China. It is shade-tolerant whenyoung and need full illumination when mature (COR-NELISSEN, 1993; TAM and GRIFFITHS, 1994). It can adapta wide variety of soil type, withstand low temperature,and produce a heavy leaf fall (TAM and GRIFFITHS, 1994).These attributes make it ideal for reforestation programms(TAM and GRIFFITHS, 1994). It is also an importanteconomic tree, used for timber, tannin extraction,and paper pulp.Castanopsis fissa is monoecious, with unisexual staminateand pistillate flowers on the same plant. Flowersof C. fissa are wind pollinated. The shape of the seed(nut) ranges from ellipsoid to ovoid. Seeds are animaldispersed. Here, we reported the development ofmicrosatellite markers which will be used to study itsspatial genetic structure in 20 ha Dinghushan plot inlower subtropical China.Genomic DNA was extracted from one dry leaf tissueby using CTAB method (DOYLE, 1991). Approximately250 ng of the total genomic DNA was digested by arestriction enzyme MseI (NEB) and the resulting fragmentsligated with MseI adaptor (5’-TACTCAGGACT-CAT-3’/5’-GACGATGAGTCCTGAG-3’) with T4 ligase1) Key Laboratory of Plant Resources Conservation and SustainableUtilization, South China Botanical Garden, Chinese Academyof Sciences, Guangzhou 510650, P. R. China.2) Guangdong Key Laboratory of Digital Botanical Garden, SouthChina Botanical Garden, Chinese Academy of Sciences,Guangzhou 510650, P. R. China.3) Graduate University of the Chinese Academy of Sciences, Beijing100049, P. R. China.4) Corresponding author: ZHENG-FENG WANG. South China BotanicalGarden, Chinese Academy of Sciences, XingKe Route 723,TianHe Guangzhou 510650, P. R. China. Tel: +86-20-37252996,Fax: +86-20-37252615. E-Mail: wzf@scib.ac.cn(NEB) overnight at 16°C. The digestion-ligation mixturewas subsequently diluted 10 times, and 2 µl was usedfor PCR amplification using adaptor-specific primers (5’-GATGAGTCCTGAGTAAN-3’, i.e. MseI-N). PCR productshybridized to a 5’ biotin-labeled oligonucleotideprobe (GA) 15and (CA) 15. Subsequent probe-bound DNAfragments were enriched for GA or CA repeats usingstreptavidin-coated magnetic beads (NEB). Enrichedfragments were recovered with PCR amplification usingMseI-N as primer. PCR products were then ligated intothe pGEM-T plasmid vector (Promega), and transformedinto the Escherichia coli DH5α competent cells (Takara).The PCR-based method described by LUNT et al. (1999)was used to screen the recombinant clones. Identifiedpositive clones were sequenced by United Gene Holdings,LTD (Shanghai, China) with M13R or M13F asprimer. Primers were designed using OLIGO 6.54 software(MBI) for the sequences contain microsatelliterepeats.Polymorphisms of these micosatellite loci wereassessed by 34 Castanopsis fissa individuals collectedfrom Dinghushan, Guangdong Province, China. PCRamplification were performed in 10 µl reaction mixtures,consisting of approximately 5 ng of template DNA,50 mM KCl, 20 mM Tris-HCl (pH 8.0), 1.5 mM MgCl 2,0.5 µM of each primer, 0.2 mM of each dNTP, and 1U ofTaq DNA polymerase (Takara). The reaction mixturewas subjected to PCR amplification in a PTC-100 (MJ)using a PCR program, 4 min at 95°, followed by 35cycles of 94°C for 30 s, 52–64°C(depending on locus)annealing temperature for 30s, and 72°C for 30s, followedby 10 min at 72°C. PCR products were thenresolved on 6% denaturing polyacrylamide gels andvisualized by silver staining. The sizes of PCR productswere determined with 20-bp DNA ladder (DongshenBiotech Company, China).Observed heterozygosity (H O), the unbiased expectedheterozygosity (H E) and fixation index (F IS) were calculatedusing GDA 1.1 (LEWIS and ZAYKIN, 2001). Deviationsfrom Hardy-Weinberg equilibrium (HWE) for eachlocus and genotypic linkage disequilibrium (LD)between all pairs of loci were tested using GENEPOP4.0.7 (RAYMOND and ROUSSET, 1995; ROUSSET, 2008).The number of allele varied from 3–13 with an averageof 7.2 alleles pre locus. The observed and expectedheterozygosities ranged from 0.265 to 0.818 and from0.270 to 0.873, respectively (Table 1). One loci (Ms07)exhibited significant deviation from HWE after Bonferronicorrection, which could be due to the occurrence ofnull alleles. Only one locus, Ms08, showed significantLD with MS03, Ms09 and Ms10 after Bonferroni correction.350299

Table 1. – Details of microsatellite loci in Castanopsis fissa including locus name, forward and reverse primer sequences, repeatmotif, annealing temperature (T a ), numbers of alleles (A), observed/expected heterozygosities (H O /H E ), fixation index (F IS ) by WEIRand COCKERHAM’s (1984), size range and GenBank accession number.*P < 0.05 after Bonferroni correction.AcknowledgementsThis work was supported by National Basic ResearchProgram of China (973 Program) (No. 2007CB411600),the National Key Technology R&D Program(2008BAC39B02), Knowledge Innovation Project of theChinese Academy of Sciences (KZCX2-YW-430) and the“Eleventh Five” National Science and Technology SupportingProject: Key Technology Research and Demonstrationon Construction of Natural Reserve(2008BAD0B05)ReferencesCORNELISSEN, J. H. C. (1993): Aboveground morphology ofshade-tolerant Castanopsis fargesii saplings in respondto light environment. International Journal of PlantSciences 154: 481–495.DOYLE, J. J. (1991): DNA protocols for plants – CTAB totalDNA isolation, pp. 283–293. In: Molecular Techniquesin Taxonomy, edited by G. M. HEWITT and A. JOHNSTON,Springer-Verlag, Berlin, Germany.LEWIS, P. O. and D. ZAYKIN (2001): Genetic Data Analysis:Computer program for the analysis of allelic data.Version 1.0 (d16c). Free program distributed by theauthors over the internet from http://lewis.eeb.uconn.edu/lewishome/software.html.LUNT, D. H., W. F. HUTCHINSON and G. R. CARVALHO(1999): An efficient method for PCR-based isolation ofmicrosatellite arrays (PIMA). Molecular Ecology 8:891–894.RAYMOND, M. and F. ROUSSET (1995): GENEPOP (version1.2): population genetics software for exact tests andecumenicism. Journal of Heredity 86: 248–249.ROUSSET, F. (2008): Genepop’007: a complete reimplementationof the Genepop software for Windows and Linux.Molecular Ecology Resources 8: 103–106.TAM, P. C. F. and D. A. GRIFFITHS (1994): Mycorrhizal associationsin Hong Kong Fagaceae. VI. Growth and nutrientuptake by Castanopsis fissa seedlings inoculatedwith ectomycorrhizal fungi. Mycorrhiza 4: 169–172.WEIR, B. S. and C. C. COCKERHAM (1984): EstimatingF-statistics for the analysis of population structure.Evolution 38: 1358–1370.Herausgeber: Johann Heinrich von Thünen-Institut. Bundesforschungsinstitut für Ländliche Räume, Wald und Fischerei.Schriftleitung: Institut für Forstgenetik, Sieker Landstraße 2, D-22927 GroßhansdorfVerlag: J. D. Sauerländer’s Verlag, Finkenhofstraße 21, D-60322 Frankfurt a. M.Anzeigenverwaltung: J. D. Sauerländer’s Verlag, Frankfurt am Main.Gesamtherstellung: PPPP Norbert Wege e.K., Gladenbach — Printed in Germany.351© J. D. Sauerländer’s Verlag, Frankfurt a. M., 2011ISSN 0037-5349

Spatial Distribution and Interspecific Associations of TreeSpecies in a Tropical Seasonal Rain Forest of ChinaGuoyu Lan 1,2 , Stephan Getzin 3 , Thorsten Wiegand 3 , Yuehua Hu 1 , Guishui Xie 2 , Hua Zhu 1 , Min Cao 1 *1 Key Laboratory of Tropical Forest Ecology, Xishuangbanna Tropical Botanical Garden, Chinese Academy of Sciences, Kunming, P. R. China, 2 Rubber Research Institute,The Chinese Academy of Tropical Agricultural Sciences, Danzhou City, Hainan Province, P. R. China, 3 Department of Ecological Modelling, UFZ Helmholtz Centre forEnvironmental Research, Leipzig, GermanyAbstractStudying the spatial pattern and interspecific associations of plant species may provide valuable insights into processes andmechanisms that maintain species coexistence. Point pattern analysis was used to analyze the spatial distribution patternsof twenty dominant tree species, their interspecific spatial associations and changes across life stages in a 20-ha permanentplot of seasonal tropical rainforest in Xishuangbanna, China, to test mechanisms maintaining species coexistence. Torustranslationtests were used to quantify positive or negative associations of the species to topographic habitats. The resultsshowed: (1) fourteen of the twenty tree species were negatively (or positively) associated with one or two of thetopographic variables, which evidences that the niche contributes to the spatial pattern of these species. (2) Most saplingsof the study species showed a significantly clumped distribution at small scales (0–10 m) which was lost at larger scales (10–30 m). (3) The degree of spatial clumping deceases from saplings, to poles, to adults indicates that density-dependentmortality of the offspring is ubiquitous in species. (4) It is notable that a high number of positive small-scale interactionswere found among the twenty species. For saplings, 42.6% of all combinations of species pairs showed positive associationsat neighborhood scales up to five meters, but only 38.4% were negative. For poles and adults, positive associations at thesedistances still made up 45.5% and 29.5%, respectively. In conclusion, there is considerable evidence for the presence ofpositive interactions among the tree species, which suggests that species herd protection may occur in our plot. In addition,niche assembly and limited dispersal (likely) contribute to the spatial patterns of tree species in the tropical seasonal rainforest in Xishuangbanna, China.Citation: Lan G, Getzin S, Wiegand T, Hu Y, Xie G, et al. (2012) Spatial Distribution and Interspecific Associations of Tree Species in a Tropical Seasonal Rain Forestof China. PLoS ONE 7(9): e46074. doi:10.1371/journal.pone.0046074Editor: Ethan P. White, Utah State University, United States of AmericaReceived May 16, 2012; Accepted August 27, 2012; Published September 28, 2012Copyright: ß 2012 Lan et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricteduse, distribution, and reproduction in any medium, provided the original author and source are credited.Funding: This project was supported by a grant from the National Science Foundation of China (31061160188-03) and a grant from the National Science &Technology Pillar Program of China (2008BAC39B02). SG was supported by the Environmental Research Center Advanced Grant ‘‘SpatioDiversity’’ (grant number233066) to TW. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.Competing Interests: The authors have declared that no competing interests exist.* E-mail: caom@xtbg.ac.cnIntroductionNumerous mechanisms have been proposed to explain coexistenceof tree species in tropical forests on local scales [1–2]. One ofthe most prominent hypotheses is the Janzen–Connell hypothesisthat states that distance- or density-dependent mortality due topredation or host-specific pests should promote less aggregatedand more mingled spatial distributions of species [3–4]. Condit etal. [5] suggested that the role of density-dependence may only beimportant among those species with the highest populationdensities. Later research showed that density-dependence wasvery common in some tropical forests [6–8]. An extension of theJanzen–Connell hypothesis, the species herd protection hypothesis,suggests that heterospecifc neighbors can promote coexistenceby thwarting the transmission of biotic plant pests [7,9]. Accordingto the species-herd protection hypothesis, heterospecific crowdingmay be of general benefit for the survival of recently establishedseedlings because fewer encounters between a host and its speciesspecificpests and pathogens would occur [9–10].The Janzen–Connell and the species herd protection hypothesesmake specific predictions on the spatial placement of individuals ofdifferent species. First, the spatial pattern of individual speciesshould become with progressing life stage less aggregated becausehigh density clumps of a given species are more prone to predationor host-specific pests. A similar pattern is expected due to selfthinning [11]. However, the mechanism of the herd protectionhypothesis may additionally create or maintain positive spatialassociations among species despite the expected effect of interspecificcompetition. Thus, we expect for progressing life stages thatthe species patterns should become less aggregated and theoccurrence of positive associations among species should increase(or be maintained). One possibility to explore if observed spatialplacement of trees is compatible with this hypothesis is to usetechniques of spatial point pattern analysis [12–14].The dipterocarp tropical seasonal rain forest is one the mostimportant vegetation types in southwest China and harborsbiodiversity that is important both for global and national speciesrichness in China. The 20-ha Xishuangbanna forest dynamics plotwas established in 2007 in this vegetation type to test theories andhypotheses related to biodiversity maintenance. For example, ofspecial interest is to find evidence (direct or indirect) of Janzen-Connell effects (or species herd protection), niche assembly ordispersal limitation [8]. Here we focus on the Janzen-Connell andspecies herd protection hypotheses and conduct a detailed spatialPLOS ONE | www.plosone.org 1 September 2012 | Volume 7 | Issue 9 | e46074352

Spatial Patterns and Associations of the Treespoint pattern analysis of the dominant tree species in this forest.Studying the factors that structure the spatial pattern andrelationship of the dominant species is of interest because theydetermine to a large extend the light climate in the forest and playa major role in structuring the lower layers of the forest. Because oftheir large size, tree species are also expected to exert positive ornegative interactions to other tree species or plants at lower layers.For example, it is believed that the emergent tree species Parashoreachinensis wang hsie at Xishuangbanna could facilitate the survivalof other plants by building up a humid microenvironmentfavorable in the dry season [15]. However, Kohyama [16]demonstrated for a warm-temperate rainforest that one-sidedcompetition for light (i.e., light is intercepted mainly by larger treesand smaller trees have limited access to light) plays a key role inspatial pattern formation of trees. Whether these species promoteor hamper each other to reach the light in the top layer remains anopen question.In this paper we used point pattern analysis to analyze thespatial species distributions, interspecific species associations andtheir change across life stages among the twenty dominant speciesof the Xishuangbanna forest dynamics plot. Our specific objectiveswere to explore if Janzen-Connell effects or species herd protectionare potential mechanisms for structuring the species patterns inour study site. To this end, we formulate two non exclusive guidinghypotheses: (1) the univariate spatial pattern of the twentydominant species should become more regular with increasinglife stage and (2) the bivariate (interspecific) spatial patterns of thetwenty dominant species should show positive association at smallscales and all life stages. Finding evidence for hypothesis (1) wouldsuggest operation of Janzen-Connell effects while evidence for (2)would agree with the species herd protection hypothesis andgenerally facilitation among different species.MethodsStudy siteThe study site was located in the Xishuangbanna NationalNature Reserve in south-western China (101u349E, 21u369N)(Fig. 1). It borders Myanmar in the southwest and Laos in thesoutheast, and has mountainous topography, with mountain ridgesrunning in a north–south direction, decreasing in elevationsouthward. The uplift of the Himalayas leads to the penetrationof warm and moist tropical air mass from the Indian Ocean toXishuangbanna in the summer, and forms a barrier preventingcold air mass from the north reaching the region in the winter,allowing for the existence of tropical seasonal rain forest in itsaltitudinal and latitudinal northern limits. The region is dominatedby a typical monsoon climate with an alternation between a dryseason and a rainy season. As recorded by a weather station 14 kmaway from the study site, the mean annual temperature is 21.0uC,and the mean annual precipitation is 1532 mm, of whichapproximately 80% occurs between May and October. The dryseason is from November to April [17]. Under these climaticconditions, the tropical seasonal rain forest grows in the lowlands,valleys and hills that have a good water supply [15,18–19].Data collectionA 20-ha permanent forest plot was established in theXishuangbanna National Nature Reserve in 2007 following thefield protocol of the Center for Tropical Forest Science [20–23].The plot is rectangular in shape and measures 400 m (north-south)by 500 m (east-west). The elevation of the plot ranges from 709 mto 869 m above sea level; the highest elevation occurs in the northwestcorner of the plot. Three perennial creeks run through theplot and join together in the south-eastern corner of the plot. Theforest occurs mainly on laterite and lateritic red soils with pHvalues ranging from 4.5–5.5.All trees ($1 cm in diameter at breast height, DBH) weremapped and tagged with unique numbers. Tree diameters weremeasured 1.3 m from the ground. All stems were identified tospecies. The top tree layer with very uneven crown canopy reaches30–60 m high and has a coverage of about 30%. As an emergenttree, the single dominant species Parashorea chinensis is the tallesttree, with crown branches near the top and its semi-orbicularcrown soars high. Other top tree species such as Sloanea tomentosa(Benth.) Rehd. et Wils, Pometia tomentosa (Bl.) Teysm. et Binn,Semecarpus reticulata Lecte, Barringtonia pendula (Griff.) Kurz usuallyoccupy a space of 30–45 m high above the continuous crowncanopy of the second tree layer and under the crown of Parashoreachinensis. The second tree layer reaches up to 18–30 m high.Garcinia cowa predominates in this layer and other representativespecies are Ficus langkokensis Drake, Knema furfuracea (Hook.f. &Thomson) Warb., Cinnamomum bejolghota (Buch.-Ham.) Sweet,Castanopsis echidnocarpa Hook.f. & Thoms. ex Miq., The third treelayer is 6–20 m high and can be roughly divided into two sublayers.The upper sub-layer occupies the 10–20 m high space andhas as common species Baccaurea ramiflora Lour., Dichapetalumgenonioides (Roxb.) Engl., Castanopsis hystrix Miq., Castanopsis indica(Roxb.) Miq., The lower sub-layer is 6–10 m high. The speciesPittosporopsis kerrii Craib predominates and other common speciesare Phoebe lanceolata (Nees) Nees, Mezzettiopsis creaghii Ridl., Leeacompactiflora Kurz, Saprosma ternata, ,Nephelium chryseum Bl. [18,24].Relative density (FD), relative dominance (RA, using basal area)and relative frequency (RF) were calculated for each species inorder to estimate the importance value (IV) [24]. We selected thetop twenty tree species with the greatest importance values for ouranalysis. And these species comprised more than 60% of the totalindividuals (95,498) of trees $1.0 cm DBH (see Table 1).Data analysisIn order to investigate the species distribution patterns and howthe species associations change across life stages, individuals ofthese species were classified into three life stages: saplings (1 to5 cm DBH), poles (5 to D95 0.1 cm DBH) and adults (.D95 0.1 cmDBH). Here, D95 0.1 is the 95th percentile of diameter of all trees§0.16Dmax, and Dmax is the diameter of the thickest tree[8,24,25,26]. For treelets (maximum DBH no more than 20 cm),individuals between 1 and 3 cm in diameter were classified assaplings, and poles included stem diameters between 3 and D95 0.1cm. These divisions roughly correspond to trees that are located inthe understory, midstory, and the canopy of the forest [27].Positive and negative associations of species with habitats weredetermined by torus-translation tests. The tests assess the similaritybetween the spatial structure of each focal species population andeach habitat [28]. Habitats of the plot were identified by threephysical parameters ( i.e., elevation, slope and convexity) in each ofthe 20620 m quadrats. We divided the habitat into six types,namely: valley (slope,27.1u; elevation,764.87 m), low-slope(slope.27.1u; elevation,764.87 m), high-slope (slope$27.1u,elevation$764.87 m, convexity.0), high-gully (slope$27.1u, elevation$764.87m, convexity,0), high-plateau (slope,27.1u, elevation$764.87m, convexity.0) and gap. Here we defined a gapas such if it has a total open area greater than 200 m 2 . For the testsof association, each simulated map was overlaid by the truedistribution of trees. Evaluation of all randomly seeded maps(n = 1000) and all torus-translated maps (n = 4999) gave frequencydistributions of relative-density estimates for each species in eachof the six principal habitats, one set of distributions for thePLOS ONE | www.plosone.org 2 September 2012 | Volume 7 | Issue 9 | e46074353

Spatial Patterns and Associations of the TreesFigure 1. Location (marked with ‘‘w’’) of the 20-ha plot in a tropical seasonal rainforest of China.doi:10.1371/journal.pone.0046074.g001randomized habitats tests and one set for the torus-translationtests, respectively.Spatial pattern analysisWe used the pair-correlation function [12,29] as summarystatistic to quantify the spatial structure of the uni- and bivariatepatterns. The pair correlation function g 11 (r) for univariate patternsof a given life stage of a species 1 can be defined based on theneighborhood density O 11 (r)=l 1 g 11 (r) which is the mean density oftrees of species 1 within rings with radius r and width dr centered inthe trees of species 1 [12] where l 1 is the intensity ( = number ofspecies 1 trees in the plot/area of the plot). The pair correlationfunction is therefore the ratio of the observed mean density of treesin the rings to the expected mean density of trees in the rings. Thepair correlation function is especially suitable for exploratoryanalysis because it isolates specific distance classes [12–14,29]. Thepair correlation function for bivariate patterns (i.e., composed ofspecies 1 and species 2 trees) follows intuitively, the quantity g 12 (r)is the ratio of the observed mean density of species 2 trees in therings around species 1 trees to the expected mean density ofspecies 2 trees in these rings [12]. The corresponding neighborhooddensity function yields O 12 (r)=l 2 g 12 (r).The univariate pair correlation function g 11 (r) can be used tofind out if the distribution of a species is random, aggregated, orregular; and at which distances r these patterns occur. Under thenull model of complete spatial randomness (CSR), where thepoints are independently and randomly distributed over the entireplot, the pair correlation function yields g 11 (r) = 1, under aggregationg 11 (r).1, and under regularity g 11 (r),1 [12]. However, thisassessment is more complicated in case of environmentalheterogeneity [30]. If the pattern contains areas with low pointdensity, the local neighborhood density is larger than the expecteddensity under CSR. As a consequence, spurious aggregationappears which may also obscure an existing small-scale regularity(i.e., virtual aggregation; [30]).One approach to account for possible first-order effects resultingfrom larger-scale environmental heterogeneity is to use theheterogeneous Poisson process as null model [12,31]. This nullmodel is able to approximately factor out the effects ofheterogeneity by displacing the points of the pattern only withinlocal neighborhoods of radius h. As a consequence, it maintains theobserved large-scale structure but removes potential non-randomlocal spatial structures at distances r below h [31]. This allows foran assessment of potential (conditional second-order) interactionsamong points if they occur at scales which are smaller than thescales at which the environment varies (i.e., a separation of scales[32]). The occurrence of any point in a heterogeneous Poissonpoint process is independent of that of others, but the points aredistributed in accordance with an intensity function l(x, y) thatvaries with location (x ,y) [12,29,30]. We used non-parametrickernel estimate of the intensity function based on the Epanechnikovkernel [12,30] with a bandwidth of h = 30 m and a spatialresolution of 2 m.We used the bivariate (cross) pair correlation function g 12 (r) tostudy the bivariate species-species associations. Again, first-ordereffects (habitat preference) may confound conditional secondordereffects (direct plant-plant interactions) [30]. To revealsignificant conditional second-order interaction we approximatelyfactored out first-order effects by using the heterogeneous Poissonnull model described above and randomized the locations of thetrees of the second species, but we kept the locations of the trees ofthe first species fixed [30]. The intensity function was thereforeconstructed based on the pattern of the second species. Abandwidth of h = 30 m and a spatial resolution of 2 m were alsoselected for all bivariate analyses. To avoid edge effects, edgecorrection of Donnelly [33], available for rectangular windowsonly, was used in the analysis. Note that we assessed the g 21 -functions also for each species combination because we cannotassume that interactions between species would be symmetric [30].PLOS ONE | www.plosone.org 3 September 2012 | Volume 7 | Issue 9 | e46074354

Spatial Patterns and Associations of the TreesTable 1. Species properties.Rank Species Code Family Life type Fruit type Dispersal modelNo. ofindividualsPercentage(%)No. ofsaplingNo. ofpolesNo. ofadults Habitat association1 Parashorea chinesis PARACH Dipterocarpaceae Emergent Samara Wind, Gravity 7919 8.3 6492 1276 151 High plateau2;Gap22 Sloanea tomentosa SLOATO Elaeocarpaceae Upper canopy Capsule Ballistic 502 0.5 222 196 84 N3 Pometia tomentosa POMETO Sapindaceae Upper canopy Drupe Animal 480 0.5 274 147 59 Low-slope+; Highplateau24 Semecarpus reticulata SEMERE Anacardiaceae Upper canopy Drupe Animal 619 0.6 354 234 31 High-plateau25 Barringtonia pendula BARRPE Lecythidaceae Upper canopy Drupe Gravity 573 0.6 314 234 25 N6 Garcinia cowa GARCCO Guttiferae Lower canopy Drupe Animal, Gravity 4333 4.5 2795 1448 90 High-plateau+; Lowslope27 Knema furfuracea KNEMFU Myristicaceae Lower canopy Drupe Animal 3160 3.3 2543 578 39 Gap+8 Ficus langkokensis FICULA Moraceae Lower canopy Sycarp Animal 1337 1.4 761 537 39 N9 CinnamomumbejolghotaCINNBE Lauraceae Lower canopy Berry Animal 1337 1.4 938 376 23 N10 CastanopsisechidnocarpaCASTEC Fagaceae Lower canopy Nut Animal, Gravity 881 0.9 291 518 72 High-plateau+; Lowslope211 Castanopsis hystrix CASTHY Fagaceae Lower canopy Nut Animal, Gravity 244 0.3 32 146 66 High-plateau+; Lowslope212 Castanopsis indica CASTIN Fagaceae Lower canopy Nut Animal, Gravity 351 0.4 166 132 53 Valley+; High-plateau213 Baccaurea ramiflora BACCRA Euphorbiaceae Understory Berry Animal 3212 3.4 1814 1365 33 N14 Pittosporopsis kerrii PITTKE Icacinaceae Treelet Drupe Animal 20918 21.9 16439 4453 26 low-slope2; high-plateau+15 Mezzettiopsis creaghii MEZZCR Annonaceae Treelet Berry Animal 3300 3.5 1744 1514 42 Valley+; High-plateau216 Nephelium chryseum NEPHCH Sapindaceae Treelet Drupe Animal 1098 1.1 713 343 42 high-slope+16 Phoebe lanceolata PHOELA Lauraceae Treelet Drupe Animal 2409 2.5 895 1496 18 Low-slope2; Highplateau+17 Saprosma ternata SAPRTE Rubiaceae Treelet Drupe Animal 2698 2.8 2332 345 21 Low-slope+; Highplateau219 DichapetalumgelonioidesDICHGE Dichapetalaceae Treelet Drupe Animal 1222 1.3 704 473 45 High plateau220 Leea compactiflora LEEACO Vitaceae Treelet Berry Animal 1051 1.1 974 64 13 NTotal 57644 60.4Topographical types include valley (slope,27.1u; elevation,764.87 m), low-slope (slope.27.1u; elevation,764.87 m), high-slope (slope$27.1u, elevation$764.87 m, convexity.0), high-gully (slope$27.1u, elevation$764.87 m,convexity,0), high-plateau (slope,27.1u, elevation$764.87 m, convexity.0) and gap (with a total open area greater than 200 m 2 ). ‘‘+’’ indicates positive correlation; ‘‘2’’ indicates negative correlation; ‘‘N’’ indicates neutralcorrelation.doi:10.1371/journal.pone.0046074.t001PLOS ONE | www.plosone.org 4 September 2012 | Volume 7 | Issue 9 | e46074355

Spatial Patterns and Associations of the TreesWe used a Monte-Carlo approach to test for significantdepartures from the null models. Each of the 199 simulations ofa point process underlying the null model generates a summarystatistic; and simulation envelopes with a

Spatial Patterns and Associations of the TreesFigure 2. Examples of distribution maps and univariate patterns. Shown are the univariate g 11 pair-correlation functions of the data independence on scale r (solid squares), and the expected g 11 (r) function under the heterogeneous Poisson null model (open squares) and the MonteCarlo simulation envelopes (solid lines) of the null models. Monte Carlo confidence was constructed at approximately 95% confidence level (199simulations). See Table 1 for species codes. Green cross: saplings, blue open circle: poles, red solid circle: adults.doi:10.1371/journal.pone.0046074.g002more heterospecific neighbors. In the following we discuss ourresults in more detail.Sapling clusteringThe twenty dominant species had a large number of juvenileswhich showed small-scale aggregation and were mostly locatedclose to the adults (Fig. S2). Aggregated spatial distributions arecommonly observed in naturally regenerating forests [35–36]. Thepresent study provides evidence of clumping for the dominanttwenty species in a dipterocarp forest in China. However, thedegree of aggregation varied with species, life stages, and spatialscale. Aggregated distributions may result for example fromlimited seed dispersal [37], animal mediated clump dispersal [38],or habitat heterogeneity [39–41]. The topography of our site wasvery diverse, with an elevation ranging from 709 to 869 m abovesea level and three perennial creeks that joined together in thesouth-eastern corner of the plot. Fourteen of the studied speciesshowed a distribution pattern related to topography. This isconfirmed by the species habitat (torus translation) test. However,the positive or negative associations to the topographic habitats aremore likely to cause larger-scale aggregation which was approximatelyfactored out by the heterogeneous Poisson null model.Thus, other factors may be responsible for the strong small-scaleclustering of saplings. For example, we found especially strongsmall-scale clustering for saplings of the two species P. chinensis andS. reticulate (Fig. S2). Although P. chinensis produces winged seeds,the seeds are relatively heavy. Hence, P. chinensis is wind- andgravity-dispersed and as a consequence, nearly 60–70% of theseeds fall within a circle of 1–8 m near the conspecific adults [42]thereby causing the strongly aggregated distribution pattern atsmall scales. The species S. tomentosa produce capsules and whenthe fruits are ripe, seeds are launched upwards and dispersedballistically with limited dispersal distance. The species P. tomentosa,S. reticulata produce drupes and are dispersed by animals. Thespecies of C. echidnocarpa, C. hystrix and C. indica produce nuts andare dispersed by Rattus tanezumi, but the distance dispersed byRattus is no more than 10 m [43]. This phenomenon is interestingbecause animal dispersal is known to be the most efficient dispersalmode in tropical forest [44]. However, both species B. pendula (witha mass about 220 g) and G. cowa (with a mass more than 70 g) havelarge seeds and are dispersed by gravity [45]. For G. cowa, there isthe highest density of seedlings within a circle of 5–7 m near theconspecific adults [46]. Thus, for these tree species limited seeddispersal is likely to be responsible for the observed aggregateddistribution pattern at small scales.The spatial patterns of our study species lost their strongclumping from the transition from juveniles to adults andapproached random patterns as found in previous studies[35,47]. This indicates that density-dependent mortality of theoffspring is ubiquitous in our plot [8]. However, self-thinningespecially in tropical forests does not need to lead to regularPLOS ONE | www.plosone.org 6 September 2012 | Volume 7 | Issue 9 | e46074357

Spatial Patterns and Associations of the TreesFigure 3. Examples of species associations in bivariate patterns. Shown are the bivariate g 12 pair-correlation functions of the data independence on scale r (solid squares), the expected g 12 (r) function under the heterogeneous Poisson null model (open squares) and the simulationenvelopes (solid lines) of the Monte Carlo simulations of the null model. Monte Carlo confidence was constructed at approximately 95% confidencelevel (199 simulations). See Table 1 for species codes. Red open circle is the first species, and black solid circle is the second species.doi:10.1371/journal.pone.0046074.g003Figure 4. Number of significant attraction and repulsion for all species at the three life stages at scales 0–50 m. a) saplings, b) poles, c)adults.doi:10.1371/journal.pone.0046074.g004PLOS ONE | www.plosone.org 7 September 2012 | Volume 7 | Issue 9 | e46074358

Spatial Patterns and Associations of the Treesdistributions of tree species, but just random patterns may emerge[48]. Overall, our results on univariate species patterns of treesagree with our first hypothesis and support the presence of Janzen-Connell effects.Positive species associationsTo investigate our second guiding hypothesis on species herdprotection, we analyzed potential species interactions amongtwenty dominant species across three life stages. In order todisentangle the effect of species interactions (second-order effects)and environment (first-order effects) on the species association, weused heterogeneous Poisson null models, accounting for possibleenvironmental heterogeneity, to reveal significant bivariatesecond-order interactions (repulsion, attraction). Our studyrevealed marked findings. One of the most interesting results isthat second-order effects were strong; overall, 81% of all speciespairs showed significant second-order effects (plant-plant interactions).Another interesting result is that positive associationsbetween species made up a high proportion at small neighbourhoodsof up to five meters. Interestingly, the highest proportion(45.5%) of positive associations occurred in the poles stage but itdeclined only to 29.5% in adults. The positive species interactionin the immediate neighbourhood of the youngest life-stage iscompatible with the presence of species herd protection. With itsemphasis on biotic interactions, this hypothesis is an extension ofthe classical Janzen-Connell hypothesis and states that increasedheterospecific crowding results in fewer encounters between a hostand its species-specific pests and pathogens. Thus, survival shouldincrease with the density of heterospecifics, even if conspecificdensity remains constant [7,49]. This hypothesis has beenproposed to explain, for example, the high tree species diversityon Barro Colorado Island [6]. In addition, species interaction didlikely not depend on phylogeny or taxonomy, for example C.echidnocarpa was positively associated with C. hystrix at scale 0–10 m(Fig. 3j-i), however C. hystrix was negatively associated with C. indicaat scale 0–5m (Table S1), though these three species belong to thesame genus. Only in 72 pairings of saplings showed no significanteffects at small scales (0–5 m).Somewhat unexpectedly we also observed positive associationsbetween adult species. Given the larger size of adults and theobserved (univariate) self thinning one would expect ratherindependence or negative interspecific associations. One explanationfor the observed positive associations is that this pattern is stilla signal from the species herd protection imprinted during earlierlife stages. An alternative, non exclusive explanation is that thepositive association could also be the result of facilitation. Forexample, P. chinensis is a big-statured species of the dipterocarpaceaewith a huge umbrella crown which can grow up to 60 m inheight. In the tropical seasonal rain forest of Xishuangbanna, thelarge trees of P. chinensis usually dominate the emergent layer andbuild up a relatively humid microenvironment favorable to the lifeof plants, especially in the dry season [15]. In other words, P.chinensis can facilitate the survival of other plants. We found thatthe associations among P. chinensis adults and adults of three otherupper canopy species in our plot showed significant attraction(second-order effects) at small scales 0–5 m.Previous research of tropical tree species associations werecarried out in a Sri Lankan Dipterocarp forest, and the resultsshowed that only 5% of common tree species pairs showedsignificant second-order effects (species interactions) [30]. However,in tropical seasonal rain forest in Xishuangbanna, 80% of thespecies pairs showed significant second-order effects (includingboth positive and negative association) at small scale 0–5 m.Although we do not know why there is such a big differencebetween the two dipterocarp forest, at least we can infer thatmechanisms of tree species coexistence in dipterocarp tropicalseasonal rainforest of China are totally different from dipterocarptropical forest in Sri Lankan.In conclusion, our results show that the degree of spatialclumping in the twenty dominant species decreases from saplings,to poles, to adults. The reduction of spatial aggregation with lifestages is indirect evidence of Janzen–Connell spacing effects.Species distribution maps and the torus translation test indicatesome presence of habitat associations and niche assembly. Thepositive spatial associations among the tree species at small spatialscales suggest operation of species herd protection but may also beindicative of facilitative interactions. Overall, these results suggestthat Janzen–Connell spacing effects, habitat association, limiteddispersal and species herd protection may contribute to shape thespatial patterns of the tree species in the tropical rainforest inXishuangbanban, southwest China.Supporting InformationFigure S1 Habitat of the 20-ha permanent plot oftropical seasonal rain forest in China. Valley: slope,27.1u,elevation,764.87 m; Low-slope: slope.27.1u, elevation,764.87m; High-slope: slope$27.1u, elevation$764.87 m,convexity.0; High-gully: slope$27.1u, elevation$764.87 m, convexity,0;High-plateau: slope,27.1u, elevation$764.87 m, convexity.0;Gap: with a total open area greater than 200 m 2 .(TIF)Figure S2 Distribution maps of the twenty dominantspecies across life stages in the 20-ha plot of tropicalseasonal rainforest. See Table 1 for species codes. Green cross:saplings, blue open circle: poles, red solid circle: adults.(TIF)Figure S3 Univariate patterns of the twenty dominantspecies across life stages in the 20-ha plot of tropicalseasonal rainforest. Shown are the univariate g 11 paircorrelationfunctions of the data in dependence on scale r (solidsquares), and the expected g 11 (r) function under the heterogeneousPoisson null model (open squares) and the Monte Carlo simulationenvelopes (solid lines) of the null models. Monte Carlo confidencewas constructed at approximately 95% confidence level (199simulations). See Table 1 for species codes.(TIF)Table S1 Spatial associations of twenty dominantspecies across life stages in 20-ha permanent plot oftropical seasonal rain forest in China. The bivariate statisticof the pair-correlation function was used to analyze the spatialassociations among five canopy species under the heterogeneousPoison null model. ‘‘p’’ stands for positive association, ‘‘r’’ standsfor no spatial association (randomness), and ‘‘n’’for negativeassociation. Monte Carlo simulation envelopes were constructed atthe approximate 95% confidence level. See Table 1 for speciescodes.(DOC)AcknowledgmentsWe are grateful to Prof. Fangliang He for proving learning opportunity inCanada and other helps. We are also grateful to all of the field workers whocontributed to the tree species census of the 20-ha plot.PLOS ONE | www.plosone.org 8 September 2012 | Volume 7 | Issue 9 | e46074359

Spatial Patterns and Associations of the TreesAuthor ContributionsConceived and designed the experiments: G-YL SG TW MC. Performedthe experiments: G-YL Y-HH G-SX HZ. Analyzed the data: G-YL Y-HHReferences1. Wright SJ (2002) Plant diversity in tropical forests: a review of mechanisms ofspecies coexistence. Oecologia 130: 1–14.2. Volkov I, Banavar JR, He FL, Hubbell SP, Maritan A (2005) Densitydependence explains tree species abundance and diversity in tropical forests.Nature 438: 658–6613. Janzen DH (1970) Herbivores and the number of tree species in tropical forests.Am Nat 104: 501–528.4. Connell JH (1971) On the role of natural enemies in preventing competitiveexclusion in some marine animals and in rain forest trees. In: Dynamics ofPopulations (eds Boer PJD & Gradwell GR). PUDOC, Wageningen, pp. 298–312.5. Condit R, Hubbell SP, Foster RB (1994) Density dependence in two understroytree species in a neotropical forest. Ecology 75 (3): 671–680.6. Wills C, Condit R, Foster RB, Hubbell SP (1997) Strong density- and diversityrelatedeffects helps to maintain species diversity in a neotropical forest. P NatlAcad Sci USA 49: 1252–1257.7. Peters HA (2003) Neighbour-regulated mortality: the influence of positive andnegative density dependence on tree populations in species-rich tropical forests.Ecol Lett 6: 757–765.8. Lan GY, Zhu H, Cao M, Hu YH, Wang H, et al. (2009) Spatial dispersionpatterns of trees in a tropical rainforest in Xishuangbanna, southwest China.Ecol Res 24: 1117–1124.9. Wills C, Green DR (1995) A genetic herd-immunity model for the maintenanceof MHC polymorphism. Immunol Rev 143: 263–292.10. Comita LS, Hubbell SP (2009) Local neighbourhood and species’ shadetolerance influence survival in a diverse seedling bank. Ecology 90: 328–334.11. Getzin S, Dean C, He F, Trofymow JA, Wiegand K, et al. (2006) Spatialpatterns and competition of tree species in a Douglas-fir chronosequence onVancouver Island. Ecography 29: 671–682.12. Wiegand T, Moloney KA (2004) Rings, circles, and null-models for pointpattern analysis in ecology. Oikos 104: 209–229.13. Perry GLW, Miller BP, Enright NJ (2006) A comparison of methods for thestatistical analysis of spatial point patterns in plant ecology. Plant Ecol 187: 59–82.14. Law R, Illian J, Burslem DFRP, Gratzer G, Gunatilleke CVS, et al. (2009)Ecological information from spatial patterns of plants: insights from pointprocess theory. J Ecol 97:616–628.15. Cao M, Zhang J (1997) Tree species diversity of tropical forest vegetation inXishuangbanna, SW China. Biodivers Conserv 6: 995–1006.16. Kohyama T (1993) Size-structured tree populations in gap dynamics forest: theforest architecture hypothesis for the stable coexistence of species. J Ecol 81:131–143.17. Zhang JH, Cao M (1995) Tropical forest vegetation of Xishuangbanna, SWChina and its secondary changes, with special reference to some problems inlocal nature conservation. Biol Conserv 73:229–238.18. Zhu H (2006) Forest vegetation of Xishuangbanna, south China. For StudChina 8: 1–27.19. Zhu H, Cao M, Hu H (2006) Geological history, flora, and vegetation ofXishuangbanna, southern Yunnan, China. Biotropica 38: 310–317.20. Condit R, Hubbell SP, Foster RB (1992) Recruitment near conspecific adultsand the maintenance of tree and shrub diversity in a neotropical forest. Am Nat140: 261–286.21. Condit R (1995) Research in large, long –term tropical forest plot. Trends EcolEvol 10(1):18–22.22. Condit R, Hubbell SP, Foster RB (1996) Changes in tree species abundance in aneotropical forest: impact of climate change. J Trop Ecol 12: 231–256.23. Condit R (1998) Tropical forest census plots: Methods and results from BarroColorado Island, Panama and a Comparion with other plot, Springer-Verlagand R.G. Landes Company. pp: 17–19.24. Lan GY, Zhu H, Cao M (2012) Tree species diversity of a 20-ha plot in atropical seasonal rainforest in Xishuangbanna, southwest China. J For Resonline25. King DA, Davies SJ, Noor NSM (2006) Life and mortality are related to adulttree size in a Malaysian mixed dipterocarp forest. For Ecol Manage 206, 152–158.SG. Contributed reagents/materials/analysis tools: TW. Wrote the paper:G-YL SG TW MC.26. Lan GY, Hu YH, Zhu H, Cao M (2011) Topography related spatial distributionof dominant tree species in a tropical seasonal rain forest in China. For EcolManage 262:1507–1503.27. Bunyavejchewina S, LaFrankieb JV, Bakerc PJ, Kanzakid M, Ashtone PS, et al.(2003) Spatial distribution patterns of the dominant canopy dipterocarp speciesin a seasonal dry evergreen forest in western Thailand. For Ecol Manage 175:87–101.28. Harms KE, Condit R, Hubbell SP, Foster RB (2001) Habitat associations oftrees and shrubs in a 50-ha neotropical forest plot. J Eco 89: 947–959.29. Stoyan D, Stoyan H (1994) Fractals, random shapes and point fields. Wiley,Chichester.30. Wiegand T, Gunatilleke S, Gunatilleke N (2007a) Species associations in aheterogeneous Sri Lankan dipterocarp forest. Am Nat 170: E77–E95.31. Wang XG, Wiegand T, Hao Z, Li BH, Ye J, et al. (2010) Species associations inan old-growth temperate forest in north-eastern China. J Ecol 98: 674–686.32. Wang ZG, Ye WH, Cao HL, Huang ZL, Lian JY, et al. (2009) Speciestopographyassociation in a species-rich subtropical forest of China. Basic ApplEcol 10: 648–655.33. Donnelly KP (1978) Simulations to determine the variance and edge effect oftotal nearest neighbourhood distance, In Simulation methods in archeology,Cambridge: Cambridge University Press. pp: 91–9534. Loosmore NB, Ford ED (2006) Statistical inference using the G or K pointpattern spatial statistics. Ecology, 87: 1925–1931.35. He F, Legendre P, LaFrankie JV (1997) Distribution patterns of tree species in aMalaysia tropical forest. J Veg Sci 8: 105–114.36. Condit R, Ashton PS, Baker P, Bunyavejchewin S, Gunatilleke S, et al. (2000)Spatial Patterns in the Distribution of tropical tree Species. Science 288: 1114–1118.37. Grubb PJ (1977) Maintenance of species-richness in plant communities: theimportance of the regeneration niche. Biological Reviews of the Cambridge38. Wiegand T, Huth A, Martínez I (2009) Recruitment in tropical treespecies:Revealing complex spatial patterns. Am Nat 174: E106–E140.39. Harms KE, Wright SJ, Calderon O, Hernandez A, Herre EA (2000) Pervasivedensity-dependent recruitment enhances seedling diversity in a tropical forest.Nature 404: 493–495.40. Getzin S, Wiegand T, Wiegand K, He F (2008) Heterogeneity influences spatialpatterns and demographics in forest stands. J Ecol 96: 807–820.41. Queenborough SA, Burslem D, Garwood NC, Valencia R (2007) Habitat nichepartitioning by 16 species of Myristicaceae in Amazonian Ecuador. Plant Ecol192: 193–207.42. Yin SH, Shuai JG (1990) Study on fruiting behaviour seeding establishment andpopulation age classes of Shorea wantianshuea. Acta Botanica Yunnanic, 12 (1):415–420.43. Liu JX (2008) Predation and dispersal of Castanopsis nuts by rodents and yellowchest rat (Rattus tanezumi) storage strategy in Xishuangbanna, Qupu normaluniversity. Master thesis, 1–40 (In Chinese with English abstract).44. Seidler TG, Plotkin JB (2006) Seed dispersal and spatial pattern in tropical trees.PLoS Biol, 4, 2132–2137.45. Yang XF, Tang Y, Cao M (2010) Diaspore Traits of 145 Tree Species from aTropical Seasonal Rain forest in Xishuangbanna,SW China. Acta BotanicaYunnanic 32: 367–377 (in Chinese with English abstract)46. Liu Y, Chen J, Bai ZL, Deng XB, Zhang L (2002) Seed dispersal, seedpredation, and seedling spatial pattern of Garcinia cowa (Guttiferae). J Plant Ecol,26(4): 427–434(In Chinese with English abstract).47. Wiegand T, Gunatilleke S, Gunatilleke N, Okuda T (2007b) Analyzing thespatial structure of a Sri Lankan tree species with multiple scales of clustering.Ecology 88: 3088–3102.48. Getzin S, Worbes M, Wiegand T, Wiegand K (2011) Size dominance regulatestree spacing more than competition within height classes in tropical Cameroon.J Trop Ecol 27: 93–102.49. Comita LS, Condit R, Hubbell SP (2007) Developmental changes in habitatassociations of tropical trees. J Ecol 95:482–492.PLOS ONE | www.plosone.org 9 September 2012 | Volume 7 | Issue 9 | e46074360

Journal of Plant Ecology Advance Access published September 25, 2012Journal ofPlant EcologyPAGES 1–6doi: 10.1093/jpe/rts031available online atwww.jpe.oxfordjournals.orgButtress trees in a 20-hectaretropical dipterocarp rainforest inXishuangbanna, SW ChinaZhiyuan He 1, 2 , Yong Tang 1, *, Xiaobao Deng 1 , and Min Cao 11Key Laboratory of Tropical Forest Ecology, Xishuangbanna Tropical Botanical Garden, the Chinese Academy of Sciences,Mengla 666303, China2Graduate University of Chinese Academy of Sciences, Beijing 100049, China*Correspondence address. Xishuangbanna Tropical Botanical Garden, the Chinese Academy of Sciences,Mengla, Yunnan 666303, China. Tel: 06918716952; Fax: 06918715070; E-mail: tangy@xtbg.ac.cnAbstractAimsButtresses are prevalent and are important to many ecological processesin tropical rainforests but are overlooked in many rainforeststudies. Based on a buttress survey in a 20-hectare plot, this studyaims to answer the following questions: (I) Is buttress forming a fixedspecies characteristic? (ii) Is there any phylogenetic signal for buttressforming across a broad taxonomic scale? (iii) Is buttress formingan inherent feature or simply induced by environmental factors,and how is this relevant to the size of the tree?MethodsWe surveyed buttresses for all 95 940 trees with diameter at breastheight (DBH) ≥10 mm in a 20-ha tropical dipterocarp rainforestin Xishuangbanna, SW China. The occurrence of buttresseswas compared across different taxa and across different tree-sizeclasses. A phylogenetic analysis was conducted among buttressedand non-buttressed species in order to understand the evolutionarybackground of buttress formation.Important FindingsThis preliminary study showed that buttress trees are very abundant(making up 32% of trees with ≥100 mm DBH) in this 20-ha tropicalrainforest situated at the northern edge of the tropics. Fifty-onepercent of the 468 tree species in the plot had stems that producedbuttresses. Large trees were more likely to develop buttresses thansmaller ones. We found that although buttress formation is not afixed species characteristic, there is a strong phylogenetic signal forbuttress formation in larger species.Keywords: buttress • phylogenetic signal • tropicalrainforest • species size • XishuangbannaReceived: 19 April 2012 Revised: 20 August 2012 Accepted: 28August 2012IntroductionButtresses are prevalent in many tropical forests, in particularin lowland tropical rainforests (Richards 1996CIT0030CIT0030), but trees may also develop significantbuttresses in sub-tropical and wet temperate forests (Francis1924; Nicoll and Ray 1996). The size and frequency of buttressesappear to decrease with increasing latitude, and fromlow to high altitudes (Smith 1972).Buttresses are generally considered mechanical structuresthat support tree boles and balance the trees against unidirectionalstresses received from prevailing winds, asymmetriccanopy, leaning stem and gravity caused by growingon slopes (Navez 1930; Richter 1984; ter Steege et al. 1997;Warren et al. 1988; Young and Perkocha 1994). In swamprainforests of Guyana, buttresses developed mainly on theopposite side of the leaning direction of Caryocar nuciferumL. (ter Steege et al. 1997). In Barro Colorado Island (BCI) andCosta Rica, the largest buttresses occur mainly on the sides oftrees away from the direction of asymmetrical crowns (Youngand Perkocha 1994). However, in another study in BCI, thesize of buttresses was not correlated with crown asymmetricstress but with the prevailing wind load (Richter 1984).Furthermore, Lewis (1988) found that the arrangements ofbuttresses in Pterocarpus officinalis Jacq. showed no associationwith either prevailing wind direction or asymmetrical treecrowns. Subsequently, mechanical models were developed totest supporting hypotheses using engineering and anatomicalstructure analysis (Clair et al. 2003; Crook et al. 1997; Ennos1995; Fisher 1982; Henwood 1973; Young and Perkocha© The Author 2012. Published by Oxford University Press on behalf of the Institute of Botany, Chinese Academy of Sciences and the Botanical Society of China.All rights reserved. For permissions, please email: journals.permissions@oup.com361

Page 2 of 6Journal of Plant Ecology1994). Most of these studies have shown that buttresses aresupporting organs of trees, especially of large trees.Apart from serving as supporting structures for trees,buttresses also have other important ecological functions(Tang et al. 2011). Buttresses may increase the contact areabetween the tree and the ground and become barriers tomatter flow, leading to a high accumulation of litter andhigh soil moisture and nutrients (Pandey et al. 2011; Tanget al. 2011). Buttresses may limit soil erosion and nutrientloss following overland flow around trees by promotinginfiltration of stem flow during heavy rainfall events on hillslopes (Herwitz 1988). Buttresses are also proposed to bean important organ in nutrient acquisition, and trees withbuttresses are more competitive than trees without them,providing an explanation for the dominance of buttressedtrees in rainforests (Newbery et al. 2009). Buttresses alsoprovide important microhabitats for many life forms. Forexample, three species of lizards in Sumatra and a whipspider in Central Amazon are particularly found nearbuttresses (Dias and Machado 2006; Voris 1977) and speciesdiversity of herpetofauna is higher around buttresses thanin other rainforest habitats (Whitfield and Pierce 2005). Theabundance of mycelial mat is also found to be higher nearbuttresses than in conjoint habitats (Guevara and Romero2007). The microhabitats formed by buttresses may alsoaffect seedling germination and establishment and result indifferent seedling assemblages at the upslope and downslopesides of buttress trees, and, consequently, in the long run,contribute to the maintenance of rainforest diversity (Tanget al. 2011).Buttresses occur in many distantly related families andspecies from some families are more likely to develop buttressesthan others (Chalk and Akpalu 1963; Chapman et al.1998; Fisher 1982; Francis 1924; Richards 1996). For example,species in the families Dipterocarpaceae, Leguminosae,Sterculiaceae and Burseraceae are more frequently found tohave large buttresses, while species from Annonaceae andFagaceae rarely have buttressed trees (Porter 1971; Richards1996). The same species may have both buttressed andnon-buttressed individuals and the proportion of buttressedstems appears to increase with tree size (Chapman et al.1998; Kaufman 1988). Emergent trees >30 m high alwaysdevelop large buttresses (Richards 1996) and buttress sizewas correlated with potential height of tree species in atropical montane rainforest on Hainan Island, China (Denget al. 2008). However, there are also exceptions; e.g. somedominant species with large spreading crowns are generallyun-buttressed in rainforests in Southern Queensland,Australia (Francis 1924). A possible reason why those bigtrees do not develop buttresses from their well-developedtap roots may be because it was suggested that buttress treesusually have superficial root systems (Richards 1996 ). Dueto limited information, the prevalence of buttresses acrosstaxa and whether they are associated with tree size is stillnot clear.Although buttresses are very distinctive and may playimportant roles in many ecological processes in rainforests,they have largely been neglected in many rainforest studiesand there is still no clear answer on many aspects of theorigin and functions of buttresses. We compare the occurrenceof buttresses across species and among families in a20-ha tropical dipterocarp rainforest in SW China and testfor the presence of an evolutionary signal in buttress formationusing phylogenetic analyses to determine whether (i)buttress forming is a fixed species characteristic, (ii) there is aphylogenetic signal for buttress formation across broad taxa,(iii) buttress formation is an inherent feature or is simplyinduced by environmental factors and how this is relevant tothe size of a tree.MATERIALS and METHODSStudy siteThis study was conducted in a 20-ha tropical seasonal dipterocarprainforest dynamic plot (101°34′E, 31°36′N) in theMengla National Nature Reserve in Xishuangbanna, SWChina. The Xishuangbanna region is dominated by a typicalmonsoon climate, with alternation between a dry seasonfrom November to April and a rainy season from May toOctober. As recorded by the Mengla weather station 14 kmfrom the study site, the mean annual temperature of the areais 21.0°C, and the mean annual precipitation is 1532 mm, ofwhich ~80% occurs during the wet season (Lan et al. 2011).The 20-ha dynamic plot was established in 2007 following theprotocol for large forest dynamics plot of Center for TropicalForest Science (CTFS; Condit 1998).The plot is 400 by 500m, with elevation ranging from 709.27 to 869.14 m abovesea level. The slopes in the plot range from 7° to 47°. Threeperennial creeks wind through the plot and join togetherat the south-eastern corner of the plot. The forest is developedmainly on laterite and lateritic red soils with pH valuesof ~4.5–5.5 (Cao et al. 2006) and is dominated by Parashoreachinensis (Dipterocarpaceae; Lan et al. 2011). All free-standingstems with DBH ≥10 mm were tagged, mapped and identifiedand their DBH was measured. The initial survey recorded95 940 trees from 468 species.Buttress surveyAll living standing trees (DBH ≥ 10 mm) in the 20-ha plotwere carefully checked for buttresses around their base fromJanuary to March in 2011. The more or less flat triangularwood structure connecting the tree trunk with lateral rootsrunning at or a little below the surface of the soil was regardedas a buttress (Richards 1996). Buttress trees were classifiedinto five categories according to the size of the buttress. Forbuttressed trees in Class 3 and above, the height, length andthickness of each buttress was measured and their orientationswere recorded. Buttress length was measured from itsintersection with the trunk of the tree to the point where theridge of the buttress first entered the ground (Chapman et al.362

Zhiyuan et al. | Buttress trees in a tropical dipterocarp rainforest in China Page 3 of 61998). Buttress height was measured as the vertical distancefrom where the buttress becomes even with the trunk of thetree to the ground. Buttress thickness was measured at a regularpoint in the middle of the buttress. Buttress orientationwas recorded as degrees to the north with a compass (Lewis1988). The identification, location and measurements of eachtree were obtained from the database of the first survey of the20-ha plot conducted in 2007 (Lan et al. 2009).Data analysisSpecies with at least one individual developed buttress wasconsidered to have the potential of producing buttresses andwas defined as a buttress species. The percentage of buttressspecies was calculated for the 20 most dominant families inthe plot. Importance values of the families were calculated bythe sum of the relative diversity, relative density and relativedominance of each family, according to the work by Mori et al.(1983).Among the 468 tree species in the plot, we identified 241buttress species and 86 non-buttress species (SupplementaryTable S1), the latter had no buttress in at least 10 individuals.To compare similarity in buttress formation with phylogeneticsimilarity, a phylogenetic tree including the 327 buttress andnon-buttress species was constructed based on the APGIIIsystem (The Angiosperm Phylogeny Group III 2009) inPhylomatic and then a test of phylogenetic signal of buttressingcharacteristic was conducted using the K statistic (Blomberget al. (2003). All phylogenetic analyses were performed usingthe phylosignal function in the “picante” package of the Rstatistical environment (Kembel et al. 2010).Tree size (DBH) was arbitrarily classified into five classesusing 200-mm intervals, following Slik and Eichhorn (2003):10 to 200 mm (small trees), 200 to 400 mm (lower canopytrees), 400 to 600 mm (middle canopy trees), 600 to 1000 mm(upper canopy trees) and 1000 mm (emergent trees). Thepercentage of buttressed trees that occurred in each tree-sizeclass was calculated. The relationship between tree size (DBH)and the occurrence of buttress trees was examined using abinomial GLM model. We calculated the Pearson’s correlationof the observed buttressing percentage in each DBH sizeclass (n = 138, range: 1–255 cm) and the corresponding fittedvalues. To understand whether the size of a species could berelated to the potential for buttress formation, we defined theDBH size of the largest individual within a species in the plotas the size of this species, represented as DBHmax, and classifiedthe species into four classes according to the DBHmax(Aiba and Kohyama 1996). We used a Pearson’s correlationtest between species size (DBHmax classes) and the percentageof buttress species in each DBHmax class to test whether thereis an association of buttress formation with the size of species.RESULTSOf the 95 940 trees with DBH > 10 mm in the 20-ha plot, weidentified buttresses on 4669 trees (5%), which were from241 out of the 468 species (51%), 132 out of the 213 genera(62%) and 56 out of the 70 families (80%) in the plot. Of thetrees with DBH > 100 mm, 3930 (32%) of 12 344 individualsand 230 (68%) of 339 species were buttressed. Of the 13 mostabundant species (with >1000 individuals), only one species,Leea compactiflora Kurz, was not buttressed. Among the other12 species, the percentage of buttressing varied from 0.07%in Saprosma ternatum Hook. f. to 21.27% in Ficus langkokensisDrake. Buttresses were found in all the dominant emergentand canopy species (e.g. Parashorea chinensis Wang Hsie,Sloanea tomentosa (Benth.) Rehd. et Wils., Pometia tomentosa(Blume) Teijsm. and Binn., Semecarpus reticulata Lecte.) and103 species had large buttresses of category ≥ size Class 3 inthe 20-ha plot.Among 15 of the top 20 most dominant families, >50%of the species and genera produced buttresses (Table 1).Fourteen of the 17 species in Elaeocarpaceae were buttressedand three other species without buttressed individuals had

Page 4 of 6Journal of Plant EcologyDiscussionFigure 1: the percentages of buttressed trees in different DBHsize classes. DBH Classes 1 to 5 represent trees with DBH ranges of10–200, 200–400, 400–600, 600–1000 and ≥1000 mm.Figure 2: number of buttressed species in different DBHmax classes.Independent Contrasts (PIC)’ is significantly higher than therandom values of PIC (K = 10.2456409, P < 0.001, repetitions(reps) = 999), indicating a strong phylogenetic signal for buttressformation among species.The percentage of buttressed trees increased with tree size(t = 9.6969, df = 136, P value < 0.001; Figure 1). For treeswith DBH > 100 mm, 31.87% of individuals from 230 species(67.8%) were buttressed. The percentage of buttressed speciesalso increased with DBHmax of the species(t = 7.8739,df = 2, P value = 0.01575; (Figure 2). The larger the DBHmaxof a species, the more likely that some of the individualswill produce buttresses. Of the 111 species with DBHmax>500 mm, 106 (95%) had buttressed stems. However, only1 (0.78%) of the 129 species with DBHmax

Zhiyuan et al. | Buttress trees in a tropical dipterocarp rainforest in China Page 5 of 6et al. 2011). The advantages buttressed trees receive allow themto gradually become dominant in frequently disturbed and generallynutrient-poor rainforests. However, the increase in thepercentage of buttressed stems with the tree size may also be aresponse to environmental conditions. The longer a tree standsin the forest, the more likely it will form a buttress induced byunidirectional force caused by wind and asymmetrical crownand gravity when growing on slopes (Navez 1930; Richter 1984;ter Steege et al. 1997; Warren et al. 1988; Young and Perkocha1994). In addition, the higher proportion of buttress-formingspecies in taxa with larger maximum DBH suggests there maybe also an evolutionary force on buttress formation towardslarge-sized trees. Answers to these questions may need longtermmonitoring of the dynamics of buttress trees.Buttressed trees have been recorded from many distantlyrelated families and many families have both buttress andnon-buttress species (Chapman et al. 1998; Richards 1996).However, some families such as Dipterocarpaceae, Leguminosae,Sterculiaceae and Burseraceae tend to have more speciesforming large buttresses in many tropical rainforests (Porter1971; Richards 1996). In the dipterocarp rainforest we studied,Elaeocarpaceae and Fagaceae are the two families that had thehighest percentage of buttress-forming species. The clustereddistribution of buttressed and non-buttressed species on thephylogenetic tree suggests that there is a strong evolutionarybackground of buttress formation. Non-buttressed speciesfrom families such as Rubiaceae, Rutaceae and Ardisiaceae,however, are small trees that grow in the understory or subcanopyof the forest. This further emphasizes the effect ofspecies size on buttress formation and provides support for thesupporting hypothesis, i.e. that buttresses function as structuralsupports of large trees. By comparing the heights of 5784 speciesfrom 222 locations ranging from 74°29′N to 54°30′S, Moleset al. (2009) found that plants are significantly taller in the tropicsthan in other regions. Her study, together with the supportinghypothesis, may partly explain why there are more buttresstrees in tropical areas.In conclusion, this preliminary study showed that buttressesare very abundant in the 20-ha tropical rainforestplot studied here, which is situated at the northern edgeof the tropics. As the largest buttress survey so far, we suggestthat buttress formation is a plastic species characteristicand has a strong phylogenetic signal towards large-sizedspecies.SupplemenTARy DATASupplementary material is available at Journal of Plant Ecologyonline.FundingNational Natural Science Foundation of China (31070411);the Biodiversity Committee of the Chinese Academy ofSciences.ACKNOWLEDGEMENTSWe thank the Xishuangbanna Station for Tropical RainforestEcosystem Studies, the Chinese Academy of Sciences for providingresearch facilities, Feng Zili and Ma Lang for their help in the fieldand Dr Sun Zhenhua and Dr Lin Luxiang for help with data analysis.Conflict of interest statement. None declared.ReferencesAiba S-I, Kohyama T. (1996) Tree species stratification in relation toallometry and demography in a warm-temperate rain forest. J Ecol84:207–18.Blomberg SP, Garland T Jr, Ives AR. (2003) Testing for phylogeneticsignal in comparative data: behavioral traits are more labile.Evolution 57:717–45.Cao M, Zou X, Warren M, et al. (2006) Tropical forests ofXishuangbanna, China. Biotropica 38:306–9.Chalk L, Akpalu J. (1963) Possible relation between the anatomy ofthe wood and buttressing. Commonw For Rev 42: 53–8.Chapman C, Kaufman L, Chapman L. (1998) Buttress formation anddirectional stress experienced during critical phases of tree development.J Trop Ecol 14:341–9.Clair B, Fournier M, Prevost MF, et al. (2003) Biomechanics of buttressedtrees: bending strains and stresses. Am J Bot 90:1349–56.Condit R. (1998) Tropical Forest Census Plots: Methods and Results fromBarro Colorado Island, Panama and a Comparison with Other Plots.Berlin, Germany: Springer.Crook MJ, Ennos AR, Banks JR. (1997) The function of buttress roots:a comparative study of the anchorage systems of buttressed (Aglaiaand Nephelium ramboutan species) and non-buttressed (Mallotuswrayi) tropical trees. J Exp Bot 48:1703–16.Deng F, Zang R, Chen B. (2008) Identification of functional groupsin an old-growth tropical montane rain forest on Hainan Island,China. Forest Ecol Manag 255:1820–30.Dias SC, Machado G. (2006) Microhabitat use by the whip spiderHeterophrynus Longicornis (Amblypygi, Phrynidae) in centralAmazon. J Arachnol 34: 540–4.Ennos AR. (1995) Development of buttresses in rainforest trees: theinfluence of mechanical stress. In Coutts MP and Grace J (eds).Wind and Trees. Cambridge, UK: Cambridge University Press,293–300.Fisher JB. (1982) A survey of buttresses and aerial roots of tropicaltrees for presence of reaction wood. Biotropica 14: 56–61.Francis WD. (1924) The development of buttresses in Queenslandtrees. Proc Roy Soc Queensl 36:21–37.Guevara R, Romero I. (2007) Buttressed trees of Brosimum alicatrumSw. affect mycelial mat abundance and indirectly the compositionof soil meso-fauna. Soil Bio Biochem 39:289–94.Henwood K. (1973) A structural model of forces in buttressed tropicalrain forest trees. Biotropica 5:83–93.Herwitz SR. (1988) Buttresses of tropical rainforest trees influencehillslope processes. Earth Surf Proc Land 13:563–7.Kaufman L. (1988) The role of developmental crises in the formationof buttresses: a unified hypothesis. Evol Trend Plant 2:39–51.Kembel SW, Cowan PD, Helmus MR, et al. (2010) Picante: R toolsfor integrating phylogenies and ecology. Bioinformatics 26:1463–4.365

Page 6 of 6Journal of Plant EcologyLan G, Hu Y, Cao M, et al. (2011) Topography related spatial distributionof dominant tree species in a tropical seasonal rain forest inChina. Forest Ecol Manag 262:1507–13.Lan G, Zhu H, Cao M, et al. (2009) Spatial dispersion patterns of treesin a tropical rainforest in Xishuangbanna, southwest China. EcolRes 24:1117–24.Lewis AR. (1988) Buttress arrangement in Pterocarpus officinalis(Fabaceae): effects of crown asymmetry and wind. Biotropica 20:280–5.Mattheck C, Bethge K. (1990) Wind breakage of trees initiated byroot delamination. Trees-Struct Funct 4:225–7.Milliken W. (1998) Structure and composition of one hectare of centralAmazonian Terra Firme Forest. Biotropica 30:530–7.Moles, AT, Warton DL, Warman L, et al. (2009) Global patterns inplant height. J Ecol 97:923–32.Mori SA, Boom BM, de Carvalho AM, et al. (1983) Southern Bahianmoist forests. Bot Rev 49:155–232.Navez A. (1930) On the distribution of tabular roots in Ceiba(Bombacaceae). Proc Natl Acad Sci U S A 16:339–44.Newbery DM, Schwan S, Chuyong GB, et al. (2009) Buttress form ofthe central African rain forest tree Microberlinia bisulcata, and itspossible role in nutrient acquisition. Trees-Struct Funct 23:219–34.Nicoll BC, Ray D. (1996) Adaptive growth of tree root systems inresponse to wind action and site conditions. Tree Physiol 16:891–8.Pandey CB, Singh L, Singh SK. (2011) Buttresses induced habitatheterogeneity increases nitrogen availability in tropical rainforests.Forest Ecol Manag 262:1679–85.Porter DM. (1971) Buttressing in a tropical xerophyte. Biotropica 3:142–4.Richards PW. (1996) The Tropical Rain Forests. Cambridge, UK:Cambridge University Press.Richter W. (1984) A structural approach to the function of buttressesof Quararibea asterolepis. Ecology 65:1429–35.Slik JW, Eichhorn KA. (2003) Fire survival of lowland tropical rainforest trees in relation to stem diameter and topographic position.Oecologia 137:446–55.Smith AP. (1972) Buttressing of tropical trees a descriptive model andnew hypotheses. Am Nat 106:32–46.ter Steege Ht, Welle BJHt, Laming PB. (1997) The possibal functionof buttress in Caryocar nuciferum (Caryocaraceae) in Guyana:ecological and wood anatomical observations. Intl Assoc WoodAnatomists J 18:415–31.Tang Y, Yang X, Cao M, et al. (2011) Buttress trees elevate soil heterogeneityand regulate seedling diversity in a tropical rainforest.Plant Soil 338:301–9.The Angiosperm Phylogeny Group. (2009) An update of theAngiosperm Phylogeny Group classification for the orders andfamilies of flowering plants: APG III. Bot J Linn Soc 161:105–21.Thompson J, Proctor J, Viana V, et al. (1992) Ecological studies on alowland evergreen rain forest on Maracá Island, Roraima, Brazil.I. Physical environment, forest structure and leaf chemistry. J Ecol80:689–703.Voris HK. (1977) Comparison of herpetofaunal diversity in tree buttressesof evergreen tropical forests. Herpetologica 33:375–80.Warren SD, Eastmond DA, Whaley WH. (1988) Structural function ofbuttresses of Tachigalia Versicolor. Ecology 69:532–6.Whitfield SM, Pierce MSF. (2005) Tree buttress microhabitat use by aNeotropical leaf-litter herpetofauna. J Herpertol 39:192–8.Young TP, Perkocha V. (1994) Treefalls, crown asymmetry, and buttresses.J Ecol 82:319–24.366

Journal of Ecology 2012, 100, 905–914doi: 10.1111/j.1365-2745.2012.01964.xSeasonal differentiation in density-dependent seedlingsurvival in a tropical rain forestLuxiang Lin 1 *, Liza S. Comita 2,3 , Zheng Zheng 1 and Min Cao 11 Key Laboratory of Tropical Forest Ecology, Xishuangbanna Tropical Botanical Garden, Chinese Academy ofSciences, Kunming 650223, China; 2 Department of Evolution, Ecology, and Organismal Biology, The Ohio StateUniversity, Columbus, OH 43210, USA; and 3 Smithsonian Tropical Research Institute, PO Box 0843-03092, Balboa,Ancoń, Republic of Panama´Summary1. Density-dependent survival is prevalent in tropical forests and is recognized as a potentiallyimportant mechanism for maintaining tree species diversity. However, there is little knowledge ofhow density dependence changes in fluctuating environments.2. Across the 20-ha Xishuangbanna tropical seasonal rain forest dynamics plot in southwest China,which has distinct dry and wet seasons, we monitored seedling survival in 453 1-m 2 quadrats over2 years. Density dependence was assessed using generalized linear mixed models with crossed randomeffects.3. When pooling all species at the community level, there were strong negative effects of conspecifictree neighbours on seedling survival over the dry-season, wet-season and 2-year intervals. The proportionof conspecific seedling neighbours had a significant negative effect in the dry season, but notin the wet season.4. At the species level, the effects of conspecific tree and seedling neighbours varied widely amongspecies in the community and were significantly positively related to population basal area in thecommunity over the dry-season interval. In contrast, over the wet-season interval, the effects of conspecifictree and seedling neighbours did not significantly vary among species in the community.Overall community- and species-level results suggest that local-scale negative density dependence(NDD) tends to be stronger in the dry than wet season in the Xishuangbanna tropical forest.5. At the scale of the 20-ha plot, we found a community compensatory trend (CCT), in which rarespecies had relatively higher seedling survival than common species in both the wet and dry seasons.A positive association between potential NDD and population basal area suggests that the CCTresults from local-scale NDD, specifically because of negative effects of conspecific tree neighbours.6. Synthesis. Our results demonstrate that the strength of density-dependent seedling survival canvary between seasons and among species in tropical forests. Future research is needed to assess theunderlying mechanisms of this temporal and interspecific variation in NDD and its consequencesfor species coexistence and community composition.Key-words: compensatory trend, density dependence, Janzen–Connell hypothesis, mixedmodels, plant–plant interactions, tropical rain forest, water availabilityIntroductionNegative density dependence (NDD) has been recognized as adriving mechanism that may underpin the maintenance of treespecies diversity in tropical forests (Wills et al. 1997; Harmset al. 2000; Volkov et al. 2005; Freckleton & Lewis 2006). Oneexplanation for patterns of NDD in tropical forests is the*Correspondence author. E-mail: linluxa@xtbg.ac.cnJanzen–Connell hypothesis, which predicts that species-specificnatural enemies, such as pathogens and herbivores, canreduce survival, recruitment and growth at high local conspecificneighbour densities (LCD) (Janzen 1970; Connell 1971).In the past decades, many studies have provided evidence forthe Janzen–Connell hypothesis by testing for density dependence(e.g. Augspurger & Kelly 1984; Hubbell, Condit &Foster 1990; Condit, Hubbell & Foster 1994; Wills et al. 1997;Webb & Peart 1999; Hille Ris Lambers, Clark & BeckageÓ 2012 The Authors. Journal of Ecology Ó 2012 British Ecological Society367

906 L. Lin et al.2002; Blundell & Peart 2004; Swamy et al. 2011). However,most of these studies have only included common species(Carson et al. 2008). When pooling all species at the communitylevel, negative conspecific effects on survival have beendocumented at different developmental stages from seedlingsto larger saplings and adults, especially in tropical tree communities(Hubbell et al. 2001; Peters 2003; Comita & Hubbell2009; Metz, Sousa & Valencia 2010).Density-dependent survival tends to be prevalent amongspecies at the seedling stage (Harms et al. 2000; Hille RisLambers, Clark & Beckage 2002; Comita & Hubbell 2009;Chen et al. 2010; Comita et al. 2010). Seedlings may be morevulnerable to attack by natural enemies compared with adulttrees, and seedlings also experience strong asymmetric resourcecompetition with conspecific adults (Clark & Clark 1985;Wright 2002). The seedling stage therefore suffers high mortalityand represents a significant bottleneck in the life cycle oftrees (Harper 1977). According to the Janzen–Connell hypothesis,the negative effect of conspecific neighbours on seedlingsurvival should be stronger than the effect of heterospecifics, arequirement for density-dependent survival to promote speciesdiversity.If local-scale NDD (species’ sensitivity to conspecific neighbours)is similar among species in the community and commonspecies have higher LCD, common species are expected tohave higher overall seedling mortality. In this situation, localscaleNDD can result in a community compensatory trend(CCT; Connell, Tracey & Webb 1984) at larger spatial scales(Webb & Peart 1999; Queenborough et al. 2007; Chen et al.2010), in which commoner species will have lower survivalrates than rarer species, thereby promoting the maintenance ofspecies diversity (Connell, Tracey & Webb 1984). In addition,species’ sensitivity to conspecific neighbours has been found tovary among species (i.e. ‘asymmetric density dependence’) andto decrease with increased species abundance in a tropical treecommunity in central Panama, suggesting that local-scaleNDD is a potential mechanism shaping species commonnessand rarity (Comita et al. 2010). However, this negative associationbetween the strength of local-scale NDD and speciesabundance appears to run counter to a CCT (Kobe & Vriesendorp2011), because rare species could have higher mortalitythan common species if both have similar LCD or if rare specieshave higher LCD. However, a CCT could still result ifLCD increased with species community abundance (Kobe &Vriesendorp 2011). Thus, to evaluate the role of local-scaleNDD in generating a CCT, it is critical to consider two componentsto NDD that may vary across species: the per-neighbournegative effect of conspecifics (per-neighbour NDD) and theLCD that a species experiences (Kobe & Vriesendorp 2011).Previous studies have typically examined NDD over a singletime period. However, the two principal agents of densitydependentmortality, resource competition and host-specificnatural enemies, are both subject to environmental fluctuations,and so the strength of NDD may vary substantially overtime, particularly in variable environments. This is suggested,for example, by the finding of Comita et al. (2009) that theeffect of conspecific neighbours varied over different censusintervals following hurricane disturbance in Puerto Rico, likelydue to changes in resource availability and the dynamics ofnatural enemies. Additional studies are needed to assess temporalvariation in density dependence.When the intensities of density-dependent agents increase,resulting in stronger local-scale NDD, a stronger CCT mayalso result, depressing survival rates of common species andproviding opportunities for rare species. This would help preventrare species from going extinct and hence promote themaintenance of species diversity. Therefore, examiningchanges in NDD in fluctuating environments can help deepenour understanding of its role in maintaining species diversity.So far, studies examining the mechanisms responsible forNDD have focused on the role of host-specific naturalenemies, rather than resource competition (Gilbert 2005;Freckleton & Lewis 2006). Tropical seasonal forests experienceseasonal variation in water availability with distinct wet anddry seasons. Water availability may influence the strength ofNDD in several ways. NDD may be stronger in the wet seasonbecause of increased abundance of pathogens and insect herbivores(Coley & Barone 1996). Alternatively, species are understress during the dry season because of low water availability(Gerhardt 1996; Comita & Engelbrecht 2009) and thereforemay be more likely to die from density-dependent agents.Furthermore, species may partition soil water resources differentially(e.g. through differences in rooting depth) (Jacksonet al. 1995; Meinzer et al. 1999), which may increase intraspecificcompetition during periods of low water availability. Onthe other hand, under extremely stressful conditions, intraspecificcompetition may be weakened and facilitation may evenoccur (e.g. Fajardo & Mclntire 2011).In this study, we examine temporal variations in the effectsof local conspecific and heterospecific neighbours on seedlingsurvival in a tropical seasonal rain forest in southwesternChina. Using data on the survival of 7263 seedlings of 186 speciesin the 20-ha Xishuangbanna forest dynamics plot, we testfor seasonal differentiation in density dependence and testwhether local-scale NDD is related to community-level speciesabundance. Specifically, we answer the following questions: (i)Does the strength of density dependence differ between the dryseason and the wet season? (ii) Does the strength of local-scaleNDDvarysignificantlyamongspecies and relate to speciescommonness and rarity in this community? (iii) Is there evidenceof a CCT in the Xishuangbanna tropical seasonal rainforest and does CCT (NDD at larger spatial scales) result fromlocal-scale NDD?Materials and methodsDATA COLLECTIONWe conducted this study in the 20-ha Xishuangbanna tropical seasonalrain forest dynamics plot located in Mengla County, Yunnanprovince, southwest China (101°34¢E, 21°36¢N). The climate isstrongly seasonal with distinct alternations between the dry season(November–April) and the wet season (May–October), as this area issituated in the north edge of tropics and has a typical monsoonclimate. At the National Forest Ecosystem Research Station atÓ 2012 The Authors. Journal of Ecology Ó 2012 British Ecological Society, Journal of Ecology, 100, 905–914368

Seasonality in density dependence 907Xishuangbanna, annual precipitation averaged 1493 mm between1959 and 1998, of which 1256 mm (84%) occurred in the wet season(Cao et al. 2006). During the study period, there were 273 and131 mm of precipitation occurring in the first and second dry seasons,and 1525 and 1031 mm during the first and second wet seasons,respectively.All free-standing tree stems ‡1 cm diameter at breast height (dbh)(henceforth referred to as trees) in the plot have been measured,mapped and identified to species between November 2006 and April2007 (Lan et al. 2009). Trees in the Xishuangbanna forest exhibitmasting behaviour, with masting occurring most recently in 2004(Yan & Cao 2008) and 2007, based on field observations. We conductedthe first seedling census in November 2007 (at the beginning ofthe dry season) after abundant post-masting germination and monitoredseedling survival for the following two non-masting years. Inthe first census, we established 151 census stations in a stratified randomdesign to monitor seed rain and seedling dynamics in the plot.Each station consisted of a 0.5-m 2 seed trap and three 1-m 2 seedlingquadrats that were placed 2 m away from each of three sides of theseed trap. In each of 453 1-m 2 seedling quadrats, all free-standing treeseedlings and saplings 20 cm tall) over the dry-season, wet-season and 2-yearintervals (see Tables S1 and S2 in Supporting Information), to testwhether results differed with age and ⁄ or size class. Since the best-fitmodels for both the younger and older cohorts were not essentiallydifferent from the best-fit models when combining all seedlingstogether, we did not further analyse the models for the separatecohorts.Seedling survival can appear to be positively related to overall seedlingdensity because of habitat effects (Comita & Hubbell 2009), makingit difficult to detect negative effects of conspecific seedlingneighbours, therefore we also tried a model that included totalTable 1. Akaike’s information criterion (AIC) values for generalized linear mixed models of seedling survival in the 20-ha Xishuangbannatropical seasonal rain forest dynamics plotAICModel typeFixed effects*Dry- season interval Wet-season interval 2-year intervalUnconditional model (model 1) 9931.46 7541.91 6879.93Density-independent (model 2) Ht 9231.77 7058.27 6204.27Density-dependentEffect of conspecific Ht + Tots 9232.35 7046.88 6205.84=effect of Ht + Totb 9231.42 7059.55 6205.17Heterospecific (model 3) Ht + Tots + Totb 9231.34 7046.97 6206.99Effect of conspecific Ht + Cons + Hets 9223.69 7048.32 6193.47„ effect of Ht + Cons + Hets + Totb 9221.79 7048.28 6194.04Heterospecific (model 4) Ht + Conb + Hetb 9226.13 7059.24 6198.30Ht + Conb + Hetb + Tots 9224.69 7045.08 6200.30Ht + Cons + Hets + Conb + Hetb 9219.04 7046.97 6193.13*Fixed effects include: Ht (initial seedling height), Tots (total seedling density), Cons (conspecific seedling density), Hets (heterospecificseedling density), Totb (total tree basal area), Conb (conspecific tree basal area) and Hetb (heterospecific tree basal area). Bold valuesdenote the best-fit models based on the lowest AIC value; however, models within two AIC units of the minimum are also considered tohave strong support.Ó 2012 The Authors. Journal of Ecology Ó 2012 British Ecological Society, Journal of Ecology, 100, 905–914369

908 L. Lin et al.seedling neighbour density and the proportion of conspecific seedlingneighbours (instead of conspecific seedling neighbour density andheterospecific seedling neighbour density) (Chen et al. 2010).Seedling survival was analysed over two census intervals, thedry-season interval and the wet-season interval. We combined observationsof survival across two dry seasons in the analysis for thedry-season interval and across two wet seasons in the analysis for thewet-season interval. We also analysed survival over the entire 2-yearinterval for the seedlings present in the first census to look for densitydependence over a longer period.To quantify variation in density dependence among species in thecommunity, we constructed a varying-slope model by adding speciesspecificrandom slopes to the coefficients of local neighbourhood densityvariables (Chen et al. 2010; Comita et al. 2010). To test the relationshipbetween species-specific neighbourhood effects andpopulation abundance (total number of conspecific tree stems ‡1 cmdbh on the 20-ha plot) or basal area (total basal area of conspecifictree stems ‡1 cm dbh on the 20-ha plot) in the community, we modelledthese species-specific coefficients as a function of populationabundance or basal area using linear regression.To determine whether there was a CCT in the 20-ha plot, we usedGLMMs to examine seedling survival rates over the dry-season, wetseasonand 2-year intervals as a function of population abundance orbasal area in the community, using the method of Chen et al. (2010).Kobe & Vriesendorp (2011) suggested ‘potential NDD’ (the maximummortality attributable to density effects), which includes bothper-neighbour NDD and maximum local conspecific neighbour densities(m_LCD), be used to evaluate whether CCTs result from localscaleNDD. If potential NDD increases with population abundanceor basal area in the community, a CCT could result from local-scaleNDD even if per-neighbour NDD decreases with increasing populationabundance or basal area in the community (e.g. Comita et al.2010). Potential NDD for species j is calculated as the per-neighbourNDD for species j multiplied by the maximum local conspecific densityof species j.In the CCTs and crossed random effects models, the values of allcontinuous explanatory variables were standardized by subtractingthe mean value of the variable and dividing by 1 standard deviationbefore analyses (except when calculating per-neighbour NDD for thepotential NDD analysis). All analyses were conducted in R 2.11.1 (RDevelopment Core Team 2010). GLMMs were fitted by the lmer()function in the ‘lme4’ package with the Laplace approximationmethod (Bates, Maechler & Dai 2008). Likelihood ratio tests wereused to assess the significance of random effects, and Wald Z testswere used to assess the significance of fixed effects (Bolker et al. 2009).ResultsSEEDLING DYNAMICS IN THE XISHUANGBANNAFOREST PLOTA total of 7263 seedlings of 186 species were censused over thetwo years following the masting event in 2007. Of the initialseedling cohort (November 2007), 3722 out of 6628 seedlings(56%) had died by November 2009. Seedling survival ratesvaried widely among species, between dry and wet seasons andbetween years (see Table S3). Because of the masting in 2007,thereweremanymoreyoungseedlings at the start of the firstdry season (November 2007) than in any other census(Fig. S1), which led to higher mortality during the first dryseasoninterval than other census intervals (Fig. S2). Althoughseedling recruitment occurred in both the dry and wet seasons(Fig S2), recruitment rates were always much lower than seedlingmortality rates during the study period (Fig. S2), resultingin a continual decrease in total seedling abundance over time(Fig. S1). Conversely, average seedling height graduallyincreased through consecutive censuses (mean values of 20.0,24.0, 26.3 and 28.0 cm). In summary, in the 2 years followingthe masting event, seedling survival and growth weremore important drivers of seedling dynamics compared withrecruitment.LOCAL-SCALE DENSITY DEPENDENCEOver the dry-season interval, the probability of seedlingsurvival of all species combined was best described by the fulldensity-dependent model in which the effect of conspecificneighbours was different from that of heterospecific neighbours,for both seedling and tree neighbours (Table 1). Conspecifictree neighbours had a strong negative effect onseedling survival, while the effect of heterospecific tree neighbourswas near zero (Table 2). In contrast, heterospecific seedlingneighbours were positively associated with seedlingsurvival, while the effect of conspecific seedling neighbourswas near zero (Table 2). In contrast, over the wet-season interval,the best-fit model included separate effects of conspecificand heterospecific tree neighbours, but not separate effects ofTable 2. Coefficients (and standard errors) estimated by the best-fit models of neighbourhood effects on seedling survival of all species combinedfor the dry-season interval, wet-season interval and 2-year intervalIndependent variables Dry-season interval Wet-season interval 2-year intervalIntercept )1.676 (0.231)*** )1.868 (0.282)*** )3.480 (0.293)***Seedling height 1.461 (0.062)*** 1.525 (0.078)*** 1.609 (0.072)***Total seedling density – 0.015 (0.004)*** –Conspecific seedling density 0.002 (0.003) NS – )0.005 (0.004) NSHeterospecific seedling density 0.017 (0.005)** – 0.012 (0.006)*Total tree basal area – – –Conspecific tree basal area )0.248 (0.084)** )0.246 (0.104)* )0.265 (0.128)*Heterospecific tree basal area )0.052 (0.067) NS )0.040 (0.081) NS )0.048 (0.098) NS*P < 0.05; **P < 0.01; ***P < 0.001; NS, not significant.– Means the term was not included in the best-fit model.Ó 2012 The Authors. Journal of Ecology Ó 2012 British Ecological Society, Journal of Ecology, 100, 905–914370

Seasonality in density dependence 909conspecific and heterospecific seedling neighbours (Table 1).However, the full density-dependent model had an AIC withintwo units of the minimum (Table 1), indicating it had similarsupport. Over the wet-season interval, the effects of conspecificand heterospecific tree neighbours were similar to those overthe dry-season interval, while total seedling neighbours werepositively associated with seedling survival (Table 2). Over the2-year interval, the pattern of density dependence was similarto that over the dry-season interval (Table 2).In the model that included total seedling neighbour densityand the proportion of conspecific seedling neighbours (insteadof conspecific seedling neighbour density and heterospecificseedling neighbour density), total seedling neighbour densityhad a positive effect on seedling survival over both the dry-seasonand wet-season intervals, but was not different from zeroover the 2-year interval (Table 3). The proportion of conspecificseedling neighbours had a strong negative effect on seedlingsurvival over the dry-season and 2-year intervals, but amuch weaker effect over the wet-season interval. The negativeeffect of the proportion of conspecific seedling neighbours wassignificantly stronger over the dry-season interval than overthe wet-season interval (Table 3).VARIATION AMONG SPECIES IN THE STRENGTH OFDENSITY DEPENDENCEOver the dry-season interval, there was significant variationamong species in the effects of conspecific tree and seedlingneighbours on survival, as indicated by a likelihood ratio testcomparing models with and without variation among speciesin neighbour effects (Table 4). Despite the wide variation, conspecifictree and seedling neighbour densities showed negativeeffects on seedling survival for nearly all species (Fig. 1a,b,Table S4). The effect of heterospecific tree neighbours on seedlingsurvival was also negative for all species in the community,but values varied little across species (Fig. S3a). Similarly,the relationship between heterospecific seedling neighboursand seedling survival varied little across species (Fig. S3b).Over the wet-season interval, the effects of both conspecificand heterospecific tree neighbours were negative for all species,but, unlike in the dry season, values did not vary significantlyamong species (Table 4, Fig. 1c and Fig. S3c). In contrast,both conspecific and heterospecific seedling neighboursshowed a positive relationship with seedling survival for mostspecies, with little variation among species (Table 4, Fig. 1dand Fig. S3d).Over the 2-year interval, there was significant variation inthe effect of conspecific tree and seedling neighbours amongspecies, with survival being negatively related to conspecifictree neighbours for some species and positively related for others,and being negatively related to conspecific seedling neighboursfor nearly all species (Fig. 1e,f, Table S5). The effect ofheterospecific tree neighbours did not vary significantly amongspecies (Table 4) and values approached zero for most species(Fig. S3e). Heterospecific seedling neighbours did have variableeffects among species (Table 4), although, in contrast toconspecific seedling neighbour effects, the relationship withsurvival was positive for most species (Fig. S3f, Table S5).The effects of conspecific tree and seedling neighbours onsurvival were both significantly positively correlated with populationbasal area over the dry-season interval (Table 4). Inother words, rarer species experienced stronger local-scaleNDD than commoner species. We did not test this relationshipover the wet-season interval since there was not significant variationin the effects of conspecific tree and seedling neighboursamong species during that period. Over the 2-year interval, wefound no relationship between the effect of conspecific treeneighbours and population basal area, but did find a significantpositive relationship between the effects of conspecificseedling neighbours and population basal area, again indicatingthat rarer species experienced stronger local-scale NDDthan did commoner species over the 2-year interval (Table 4).We did not find a significant relationship between neighbourhoodeffects and population abundance (Table 4).COMMUNITY-LEVEL COMPENSATORY TRENDSSeedling survival rates were significantly negatively related topopulation basal area in the 20-ha plot over both the dry-seasonand wet-season intervals. Over the 2-year interval, thisrelationship was marginally significant (Table 5). This resultindicates a significant CCT in the 20-ha Xishuangbanna tropicalseasonal rain forest dynamics plot. When using populationabundance, no CCT was detected in any interval. Based onAIC values, the models using population basal area were betterthan those for population abundance (Table 5).Over the dry-season, wet-season and 2-year interval, them_LCD for both tree and seedling neighbours significantlyincreased with population basal area (Table S6). When multiplyingm_LCD by per-neighbour NDD to calculate potentialNDD, we found mixed results in terms of the relationshipbetween potential NDD and population basal area. Over theTable 3. Coefficients (and standard errors) estimated by the models of neighbourhood effects with total seedling density and the proportion ofconspecific seedling on seedling survival of all species combined for the dry-season interval, wet-season interval and 2-year intervalIndependent variables Dry-season interval Wet-season interval 2-year intervalIntercept )1.407 (0.205)*** )1.849 (0.253)*** )3.120 (0.247)***Seedling height 1.444 (0.062)*** 1.515 (0.078)*** 1.598 (0.072)***Total seedling density 0.006 (0.002)* 0.015 (0.004)*** 0.001 (0.004) NSProportion of conspecific seedling )0.835 (0.166)*** )0.316 (0.186)# )1.225 (0.206)****P < 0.05; **P < 0.01; ***P < 0.001; #P < 0.1; NS, not significant.Ó 2012 The Authors. Journal of Ecology Ó 2012 British Ecological Society, Journal of Ecology, 100, 905–914371

910 L. Lin et al.Table 4. Results of linear regression models used to test for a relationship between neighbourhood effects† and population abundance(conspecific tree abundance‡, model 1) or basal area (conspecific tree basal area§, model 2) in the community. The tests were only performedwhen the neighbourhood effects were found to vary significantly among species, based on likelihood ratio tests. Values of abundance and basalarea were standardized by subtracting the mean value and dividing by 1 standard deviation before analysesCensus intervals Neighbourhood effects Likelihood ratio test Model InterceptPopulation abundanceor basal areaDry-season interval Conspecific tree *** 1 )0.271*** 0.001 NS2 )0.271*** 0.013*Heterospecific tree NS – – –Conspecific seedling ** 1 )0.268*** 0.012#2 )0.268*** 0.037***Heterospecific seedling NS – – –Wet-season interval Conspecific tree NS – – –Heterospecific tree NS – – –Conspecific seedling NS – – –Heterospecific seedling NS – – –2-year interval Conspecific tree *** 1 )0.025*** )0.005 NS2 )0.025*** )0.009 NSHeterospecific tree NS – – –Conspecific seedling * 1 )0.330*** 0.011#2 )0.330*** 0.026***Heterospecific seedling ** 1 0.132*** )0.006 NS2 0.132*** 0.015***†Only includes those neighbourhood effects in the best-fit models.‡Total number of conspecific tree stems ‡1 cm dbh on the 20-ha plot.§total basal area (m 2 ) of conspecific tree stems ‡1 cm dbh on the 20-ha plot. *P < 0.05; **P < 0.01; ***P < 0.001; #P < 0.1; NS, notsignificant.Number of species0 20 40 60 80(a)Dry−season intervalConspecific trees(b)Dry−season intervalConspecific seedlingsNumber of species0 20 40 60 80(c)Wet−season intervalConspecific trees(d)Wet−season intervalConspecific seedlingsNumber of species0 20 40 60 80(e)Two−year intervalConspecific trees−0.6 −0.4 −0.2 0.0 0.2 0.4 0.6Neighborhood effects(f)Two−year intervalConspecific seedlings−0.6 −0.4 −0.2 0.0 0.2 0.4 0.6Neighborhood effectsFig. 1. Distribution of the neighbourhood effects on seedling survival. Histogram bars are based on coefficients of neighbourhood density variables.Bars to the left of the dashed zero line indicate species whose survival is reduced by increasing neighbour density.Ó 2012 The Authors. Journal of Ecology Ó 2012 British Ecological Society, Journal of Ecology, 100, 905–914372

Seasonality in density dependence 911Table 5. Coefficients (and standard errors) estimated in generalized linear mixed models used to test for a relationship between seedling survivaland population abundance (conspecific tree abundance†, model 1) or basal area (conspecific tree basal area‡, model 2) in the community, that is,a community compensatory trend. Akaike’s information criterion (AIC) values of the models are also givenCensus interval Model Seedling heightPopulation abundance orbasal areaAICDry-season interval Model 1 1.007 (0.042)*** )0.095 (0.184) NS 9233.512Model 2 1.007 (0.042)*** )0.556 (0.271)* 9229.869Wet-season interval Model 1 1.013 (0.052)*** )0.167 (0.196) NS 7059.555Model 2 1.012 (0.052)*** )0.638 (0.251)*** 7054.5412-year interval Model 1 1.100 (0.049)*** )0.081 (0.223) NS 6206.139Model 2 1.097 (0.049)*** )0.708 (0.377)# 6202.947†Total number of conspecific tree stems ‡1 cm dbh on the 20-ha plot.‡Total basal area (m 2 ) of conspecific tree stems ‡1 cm dbh on the 20-ha plot. *P < 0.05; **P < 0.01; ***P < 0.001; #P < 0.1; NS,not significant.dry-season and wet-season interval, potential NDD of conspecifictree neighbours was significantly positively correlated withpopulation basal area, indicating that more common speciessuffered more from NDD because of their higher conspecifictree densities. In contrast, potential NDD of conspecific seedlingneighbours showed a significant negative correlation withpopulation basal area (Table S7). This suggests that the CCTsresulted from local-scale NDD of conspecific tree neighbours,but not conspecific seedling neighbours. Over the 2-year interval,the relationship between potential NDD and populationbasal area was not significant for seedling neighbours and onlymarginally significant for tree neighbours (Table S7).DiscussionIn this study, community- and species-level results suggest thatlocal-scale NDD tends to be stronger in the dry than wet seasonin the Xishuangbanna tropical forest. In addition, wefound significant variation among species in the strength ofneighbour effects, with rarer species suffering stronger negativeeffects of conspecifics (on a per-neighbour basis). However,individuals of common species experienced higher local conspecificdensities. As a result, at the community-level, seedlingsurvival rates declined with population basal area, consistentwith a CCT. We expand on these findings below.SEASONAL DIFFERENCES IN NDD AT THE COMMUNITYLEVELNegative density dependence appears to be stronger in thedry season compared with the wet season in the Xishuangbannaforest. While conspecific tree neighbours had a strongnegative effect on seedling survival in both seasons, conspecificseedling neighbours had a stronger impact in the drycompared with the wet season. This was most evident in analysisthat examined the effect of the proportion of conspecificseedling neighbours on seedling survival and found a significantlystronger negative effect in the dry season (Table 3).This result is in contrast to the findings of Bunker & Carson(2005), who found that NDD was stronger in plots that hadbeen irrigated compared with control plots in seasonal forestin Panama, suggesting that increased soil moisture promotesstronger NDD.Why might conspecific seedling neighbours have a strongernegative effect in the dry season? Seedlings may be weakenedin the dry season because of drought stress, and as a result, bemore susceptible to, or likely to die from, attack by pathogensor herbivores. The low water availability during the dry seasonmay also lead species to compete with conspecifics more thanheterospecifics because species may partition soil waterresources differentially (Jackson et al. 1995; Meinzer et al.1999). Therefore, stronger NDD may result from strongerintraspecific competition during the dry season. This runs contraryto the stress gradient hypothesis, which proposes thatfacilitative interactions among plants become more dominantthan competitive interactions with increasing physical harshnessof the environment (e.g. Maestre et al. 2009; Fajardo &Mclntire 2011). However, this phenomena is unlikely to occurat our study site, where plants experience only moderatedrought stress, due in part to fog water that serves as an importantsource of moisture (Liu et al. 2004, 2005) and can be usedby seedlings in the dry season (Liu et al. 2010). In addition,other studies have demonstrated that the stress gradienthypothesis does not apply in all circumstances (Tielbörger &Kadmon 2000; Maestre, Valladares & Reynolds 2005).In contrast to effects of conspecific seedling neighbours, conspecifictree neighbours had similar negative effects on seedlingsurvival in the dry and wet seasons. Species-specific naturalenemies tend to be more numerous near mature trees (Janzen1970; Connell 1971), which could be the potential mechanismthat underpins negative effects of conspecific tree neighbours.As mentioned earlier, under drought stress in the dry season,seedlings may be more likely to die from natural enemy attack.However, pathogens and insect herbivores tend to be moreabundant in the wet season (Coley & Barone 1996). Thus,increased negative effects of conspecific tree neighboursbecause of drought stress in the dry season may be balanced byincreased natural enemy attack in the wet season.In contrast to conspecific neighbours, effects of heterospecificand total seedling neighbours on survival tended to be positive.The species herd protection hypothesis (Wills & Green1995) predicts that increased heterospecific crowding results inÓ 2012 The Authors. Journal of Ecology Ó 2012 British Ecological Society, Journal of Ecology, 100, 905–914373

912 L. Lin et al.fewer encounters between a host and its host-specific naturalenemies, which thereby increases host survival (Peters 2003).In this study, the pervasive positive effects of heterospecificseedling neighbours (Table 2) support the idea that heterospecificneighbours can enhance protection from natural enemies.However, there may be another cause for the positiveeffects of heterospecific seedling neighbours: habitats that arebeneficial for seedlings (e.g. small light gaps) will lead to bothhigh seedling densities and high seedling survival (Comita &Hubbell 2009). This would result in a spurious positive relationshipbetween heterospecific or total seedling neighboursand seedling survival.SEASONAL DIFFERENCES IN NDD AT THE SPECIESLEVELIn this study, we found wide variation among species in thestrength of NDD over the dry-season interval (Fig. 1a,b), butnot over the wet-season interval (Fig. 1c,d). This indicates thatseasonal differentiation in NDD can be more easily detected atthe species level than the community level. In the dry season, ifdrought stress strengthens NDD, species-specific variation inthe strength of NDD may result from the differences in speciessensitivities to drought. In that case, species-specific NDD inthe wet season, when drought stress is minimal or non-existent,would converge on the same level, as shown by our results(Fig. 1c,d). Over the 2-year interval of the study, species alsovaried significantly in the strength of NDD. However, in contrastto the dry season where nearly all species were negativelyaffected by conspecific neighbours, a number of speciesshowed positive effects of conspecific neighbours (i.e. positivedensity dependence) over the 2-year interval (Fig. 1). This maybe attributed to habitat filtering becoming a stronger driver ofseedling survival over longer time intervals.At this point, however, we can only speculate about theunderlying mechanisms driving the patterns of NDD observedat our site. There have been several experimental studies atother sites confirming that pathogens play a key role innegative effects of conspecific neighbours (Augspurger & Kelly1984; Bell, Freckleton & Lewis 2006; Bagchi et al. 2010;Mangan, Herre & Bever 2010; McCarthy-Neumann & Kobe2010). There have also been several experiments examiningseedling competition in tropical forests (Gerhardt 1996;Massey et al. 2006; Paine et al. 2008). However, further experimentalinvestigations are needed to determine the relativeimportance of natural enemies and resource competition forthe patterns of NDD observed here (e.g. Kobe & Vriesendorp2011).ASYMMETRIC DENSITY DEPENDENCE ANDPOPULATION BASAL AREAOver the dry-season and 2-year intervals, the strength of negativeconspecific effects decreased with increasing populationbasal area in the community (Table 4). This means that rarerspecies were more negatively affected by conspecifics thancommoner species, on a per-neighbour basis. Rarer speciescould therefore be prevented from increasing their abundancebecause of the stronger local-scale NDD they experience whentheir local abundance increases. In other words, less commonspecies cannot achieve dominance in the community understronger local-scale NDD. This kind of species-asymmetricdensity-dependent seedling survival might be an underlyingmechanism maintaining patterns of species commonness andrarity in the community, as suggested by Comita et al. (2010)and Mangan et al. (2010). If species-asymmetric densitydependence is related to differential drought sensitivity amongspecies at our study site, we would predict that common speciesshould have lower drought sensitivity than rare species. Infuture studies, the differentiation in NDD between droughtsensitivespecies and drought-tolerant species needs to beinvestigated further, to examine the mechanisms governingthe relationship between NDD and species abundance in thecommunity.COMMUNITY COMPENSATORY TRENDSPrevious studies have found mixed evidence for the existenceof CCTs in tropical forests (e.g. Connell, Tracey & Webb1984; Welden et al. 1991; He, Legendre & LaFrankie 1997;Webb & Peart 1999; Queenborough et al. 2007; Comita &Hubbell 2009; Chen et al. 2010). In the Xishuangbannaforest, we did find a compensatory trend at the 20-ha scaleduring both the wet and dry seasons, in which seedlingsurvival increased with decreasing population basal area inthe community. Under CCTs, rarer species have a greater percapita probability of survival than commoner species in thecommunity. The conventional interpretation of a CCT is thatper-neighbour NDD is similar among species in the communityand individuals of rarer species are less likely to be surroundedby conspecifics at local scales. Indeed, the m_LCDsignificantly increased with population basal area (Table S6).However, per-neighbour NDD showed a wide variationamong species (Fig. 1) and was negatively correlated withpopulation basal area (as discussed earlier). This would tendto offset a CCT, unless potential NDD (per-neighbourNDD · m_LCD) increases correspondingly with speciescommunity abundance (Kobe & Vriesendorp 2011). Ouranalysis of the relationship between potential NDD and populationbasal area suggest that over both the dry-season andwet-season interval, CCTs may have resulted from local-scaleNDD of conspecific tree neighbours, but not conspecificseedling neighbours.Under CCTs, mortality should be disproportionately lowfor rarer species, which could offset any disadvantage ofdrought stress for rarer species in the dry season (or otherstresses, such as pathogen attack, in the wet season) and couldpromote the persistence of rarer species. Thus, CCTs inducedby local-scale NDD prevailing in both the dry season and thewet season may promote the maintenance of rare species influctuating environments. However, the CCT was only marginallysignificant over the 2-year interval (Table 5). This maybe because, over longer intervals, factors such as species’ habitatpreferences or shade tolerance strongly influence seedlingÓ 2012 The Authors. Journal of Ecology Ó 2012 British Ecological Society, Journal of Ecology, 100, 905–914374

Seasonality in density dependence 913survival rates and may obscure the relationship between populationbasal area and seedling survival (Queenborough et al.2007; Comita & Hubbell 2009; Chen et al. 2010).ConclusionsOur results suggest that local-scaleNDDtendstobestrongerin the dry than wet season in the Xishuangbanna tropicalforest. This study provides further evidence that the densityand identity of neighbouring individuals influence juvenilesurvival at both local and larger spatial scales in forest communities,and lends support to the idea that NDD influencesthe diversity and abundances of tropical tree species. In addition,our results highlight the fact that simple models ofNDD, in which the strength of conspecific neighbour effectsis consistent over time and across species, do not capture thevariation observed in real-world communities. Nonetheless,few empirical or theoretical studies have examined this variation,the underlying mechanisms or the consequences of suchvariation for community structure (but see Chisholm & Muller-Landau2011). Thus, further empirical research is neededto better characterize patterns and drivers of variation indensity dependence, coupled with theoretical efforts toexplore how temporal and interspecific variation in thestrength of NDD influences species coexistence and abundancein diverse communities.AcknowledgementsThis study was funded by the National Natural Science Foundation of China(31000201) and the National Science & Technology Pillar Program from theMinistry of Environmental Protection of China (2008BAC39B02). LSCacknowledges the support of the National Center for Ecological Analysisand Synthesis, a centre funded by NSF (Grant #EF-0553768), the Universityof California, Santa Barbara, and the State of California. We thank XiaoliangLi, Jiu Ma, Wenfu Zhang Xiaobao Deng, Shishun Zhou and HongWang for their assistance with seedling censuses and species identification.We thank all the people who have contributed to the establishment of the20-ha Xishuangbanna tropical seasonal rain forest dynamics plot. Logisticalsupport was provided by Xishuangbanna Station of Tropical Rainforest EcosystemStudies (National Forest Ecosystem Research Station at Xishuangbanna),Chinese Academy of Sciences. We thank David Burslem, CalumBrown, Rich Kobe, Robin Chazdon and Danae Rozendaal for critical commentson the manuscript.ReferencesAugspurger, C.K. & Kelly, C.K. (1984) Pathogen mortality of tropical treeseedlings: experimental studies of the effects of dispersal distance, seedlingdensity, and light conditions. Oecologia, 61, 211–217.Bagchi, R., Swinfield, T., Gallery, R.E., Lewis, O.T., Gripenberg, S., Narayan,L. & Freckleton, R.P. (2010) Testing the Janzen-Connell mechanism: pathogenscause overcompensating density dependence in a tropical tree. EcologyLetters, 13, 1262–1269.Bates, D., Maechler, M. & Dai, B. (2008) lme4: Linear Mixed-Effects ModelsUsing S4 Classes. R Package Version 0.999375-28. Available at: http://lme4.r-forge.r-project.org/.Bell, T., Freckleton, R.P. & Lewis, O.T. (2006) Plant pathogens drive densitydependentseedling mortality in a tropical tree. Ecology Letters, 9, 569–574.Blundell, A.G. & Peart, D.R. (2004) Density-dependent population dynamicsof a dominant rain forest canopy tree. Ecology, 2004, 704–715.Bolker, B.M., Brooks, M.E., Clark, C.J., Geange, S.W., Poulsen, J.R.,Stevens, M.H.H. & White, J.S. (2009) Generalized linear mixed models:a practical guide for ecology and evolution. Trends in Ecology & Evolution,24, 127–135.Bunker, D.E. & Carson, W.P. (2005) Drought stress and tropical forest woodyseedlings: effect on community structure and composition. Journal of Ecology,93, 794–806.Burnham, K.P. & Anderson, D.R. (2002) Model Selection and MultimodelInference: A Practical Information-theoretic Approach, 2nd edn. Springer,New York.Cao, M., Zou, X., Warren, M. & Zhu, H. (2006) Tropical forests of Xishuangbanna,China. Biotropica, 38, 306–309.Carson, W.P., Anderson, J.T., Leigh Jr, E.G. & Schnitzer, S.A. (2008) Challengesassociated with testing and falsifying the Janzen-Connell hypothesis:a review and critique. Tropical Forest Community Ecology (eds W.P. Carson& S.A. Schnitzer), pp. 210–241. Wiley-Blackwell, Oxford.Chen, L., Mi, X., Comita, L.S., Zhang, L., Ren, H. & Ma, K. (2010) Community-levelconsequences of density dependence and habit association in a subtropicalbroad-leaved forest. Ecology Letters, 13, 695–704.Chisholm, R.A. & Muller-Landau, H.C. (2011) A theoretical model linkinginterspecific variation in density dependence to species abundances. TheoreticalEcology, 4, 241–253.Clark, D.B. & Clark, D.A. (1985) Seedling dynamics of a tropical tree: impactsof herbivory and meristem damage. Ecology, 66, 1884–1892.Coley, P.D. & Barone, J.A. (1996) Herbivory and plant defenses in tropical forests.Annual Review of Ecology and Systematics, 27, 305–335.Comita, L.S. & Engelbrecht, B.M.J. (2009) Seasonal and spatial variation inwater availability drive habitat associations in a tropical forest. Ecology, 90,2755–2765.Comita, L.S. & Hubbell, S.P. (2009) Local neighborhood and species’ shadetolerance influence survival in a diverse seedling bank. Ecology, 90, 328–334.Comita, L.S., Uriarte, M., Thompson, J., Jonckheere, I., Canham, C.D. &Zimmerman, J.K. (2009) Abiotic and biotic drivers of seedling survival in ahurricane-impacted tropical forest. Journal of Ecology, 97, 1346–1359.Comita, L.S., Muller-Landau, H.C., Aguilar, S. & Hubbell, S.P. (2010) Asymmetricdensity dependence shapes species abundances in a tropical tree community.Science, 329, 330–332.Condit, R., Hubbell, S.P. & Foster, R.B. (1994) Density dependence in twounderstory tree species in a neotropical forest. Ecology, 75, 671–680.Connell, J.H. (1971) On the role of natural enemies in preventing competitiveexclusion in some marine animals and in rain forest trees. Dynamics of Numbersin Populations (eds P.J. den Boer & G. Gradwell), pp. 298–312. Centerfor Agricultural Publication and Documentation, Wageningen.Connell, J.H., Tracey, J.G. & Webb, L.J. (1984) Compensatory recruitment,growth, and mortality as factors maintaining rain forest tree diversity. EcologicalMonograph, 54, 141–164.Fajardo, A. & Mclntire, E.J.B. (2011) Under strong niche overlap conspecificsdo not compete but help each other to survive: facilitation at the intraspecificlevel. Journal of Ecology, 99, 642–650.Freckleton, R.P. & Lewis, O.T. (2006) Pathogens, density dependence andthe coexistence of tropical trees. Proceedings of Royal Society B, 273,2909–2916.Gelman, A. & Hill, J. (2006) Data Analysis Using Regression and Multilevel⁄ hierarchical Models. Cambridge University Press, Cambridge.Gerhardt, K. (1996) Effects of root competition and canopy openness on survivaland growth of tree seedlings in a tropical seasonal dry forest. ForestEcology and Management, 82, 33–48.Gilbert, G.S. (2005) Dimensions of plant disease in tropical forests. Biotic Interactionsin the Tropics: Their Role in the Maintenance of Species Diversity (edsD.F.R.P. Burslem, M.A. Pinard & S.E. Hartley), pp. 141–164. CambridgeUniversity Press, Cambridge.Harms, K.E., Wright, S.J., Calderon, O., Hernandez, A. & Herre, E.A. (2000)Pervasive density-dependent recruitment enhances seedling diversity in atropical forest. Nature, 404, 493–495.Harper, J.L. (1977) Population Biology of Plants. Academic Press, London.He, F., Legendre, P. & LaFrankie, J.V. (1997) Distribution patterns of tree speciesin a Malaysian tropical rain forest. Journal of Vegetation Science, 8,105–114.Hille Ris Lambers, J., Clark, J.S. & Beckage, B. (2002) Density-dependent mortalityand the latitudinal gradient in species diversity. Nature, 417, 732–735.Hubbell, S.P., Condit, R. & Foster, R.B. (1990) Presence and absence of densitydependence in a neotropical tree community. Philosophical Transactions ofthe Royal Society B, 330, 269–281.Hubbell, S.P., Ahumada, J.A., Condit, R. & Foster, R.B. (2001) Local neighbourhoodeffects on long-term survival of individual trees in a neotropicalforest. Ecological Research, 16, 859–875.Jackson, P.C., Cavelier, J., Goldstein, G., Meinzer, F.C. & Holbrook, N.M.(1995) Partitioning of water resources among plants of a lowland tropicalforest. Oecologia, 101, 197–203.Ó 2012 The Authors. Journal of Ecology Ó 2012 British Ecological Society, Journal of Ecology, 100, 905–914375

914 L. Lin et al.Janzen, D.H. (1970) Herbivores and the number of tree species in tropical forests.The American Naturalist, 104, 501–528.Kobe, R.K. & Vriesendorp, C.F. (2011) Conspecific density dependence inseedlings varies with species shade tolerance in a wet tropical forest. EcologyLetters, 14, 503–510.Lan, G., Zhu, H., Cao, M., Hu, Y., Wang, H., Deng, X. et al. (2009) Spatialdispersion patterns of trees in a tropical rainforest in Xishuangbanna, southwestChina. Ecological Research, 24, 1117–1124.Liu, W.J., Meng, F.R., Zhang, Y.P., Liu, Y.H. & Li, H.M. (2004) Water inputfrom fog drip in the tropical seasonal rainforest of Xishuangbanna, SWChina. Journal of Tropical Ecology, 20, 517–524.Liu, W.J., Zhang, Y.P., Li, H.M. & Liu, Y.H. (2005) Fog drip and its relationto groundwater in the tropical seasonal rainforest of Xishuangbanna, SWChina: a preliminary study. Water Research, 39, 787–794.Liu, W.J., Liu, W.Y., Li, P.J., Duan, W.P. & Li, H.M. (2010) Dry season wateruptake by two dominant canopy tree species in a tropical seasonal rainforestof Xishuangbanna, SW China. Agriculture and Forest Meteorology, 150,380–388.Maestre, F.T., Valladares, F. & Reynolds, J.F. (2005) Is the change of plantplantinteractions with abiotic stress predictable? A meta-analysis of fieldresults in arid environments. Journal of Ecology, 93, 748–757.Maestre, F.T., Callaway, R.M., Valladares, F. & Lortie, C.J. (2009) Refiningthe stress-gradient hypothesis for competition and facilitation in plant communities.Journal of Ecology, 97, 199–205.Mangan, S.A., Herre, E.A. & Bever, J.D. (2010) Specificity between Neotropicaltree seedlings and their fungal mutualists leads to plant-soil feedback.Ecology, 91, 2594–2603.Mangan, S.A., Schnitzer, S.A., Herre, E.A., Mack, K.M.L., Valencia,M.C., Sanchez, E.I. & Bever, J.D. (2010) Negative plant-soil feedbackpredicts tree-species relative abundance in a tropical forest. Nature, 466,752–756.Massey, F.P., Massey, K., Press, M.C. & Hartley, S.E. (2006) Neighbourhoodcomposition determines growth, architecture and herbivory in tropical rainforest tree seedlings. Journal of Ecology, 94, 646–655.McCarthy-Neumann, S. & Kobe, R.K. (2010) Conspecific plant-soil feedbacksreduce survivorship and growth of tropical tree seedlings. Journal of Ecology,98, 396–407.Meinzer, F.C., Andrade, J.L., Goldstein, G., Holbrook, N.M., Cavelier, J. &Wright, S.J. (1999) Partitioning of soil water among canopy trees in a seasonallydry tropical forest. Oecologia, 121, 293–301.Metz, M.R., Sousa, W.P. & Valencia, R. (2010) Widespread density-dependentseedling mortality promotes species coexistence in a highly diverse Amazonianrain forest. Ecology, 91, 3675–3685.Paine, C.E.T., Harms, K.E., Schnitzer, S.A. & Carson, W.P. (2008) Weak competitionamong tropical tree seedlings: implications for species coexistence.Biotropica, 40, 432–440.Peters, H.A. (2003) Neighbour-regulated mortality: the influence of positiveand negative density dependence on tree populations in species-rich tropicalforests. Ecology Letters, 6, 757–765.Queenborough, S.A., Burslem, D.F.R.P., Garwood, N.C. & Valencia, R.(2007) Neighborhood and community interactions determine the spatial patternof tropical tree seedling survival. Ecology, 88, 2248–2258.R Development Core Team (2010) R: A language and environment for statisticalcomputing. R Foundation for Statistical Computing, Vienna, Austria.ISBN 3-900051-07-0, URL http://www.R-project.org.Swamy, V., Terborgh, J., Dexter, K.G., Best, B.D., Alvarez, P. & Cornejo, F.(2011) Are all seeds equal? Spatially explicit comparisons of seed fall and saplingrecruitment in a tropical forest. Ecology Letters, 14, 195–201.Tielbo¨rger, K. & Kadmon, R. (2000) Temporal environmental variation tipsthe balance between facilitation and interference in desert plants. Ecology,81, 1544–1553.Volkov, I., Banavar, J.R., He, F.L., Hubbell, S.P. & Maritan, A. (2005) Densitydependence explains tree species abundance and diversity in tropical forests.Nature, 438, 658–661.Webb, C.O. & Peart, D.R. (1999) Seedling density dependence promotes coexistenceof Bornean rain forest trees. Ecology, 80, 2006–2017.Welden, C.W., Hewett, S.W., Hubbell, S.P. & Foster, R.B. (1991) Sapling survival,growth, and recruitment: relationship to canopy height in a neotropicalforest. Ecology, 72, 35–50.Wills, C. & Green, D.R. (1995) A genetic herd-immunity model for themaintenance of MHC polymorphism. Immunological Reviews, 143, 263–292.Wills, C., Condit, R., Foster, R.B. & Hubbell, S.P. (1997) Strong density- anddiversity-related effects help to maintain tree species diversity in a neotropicalforest. Proceedings of the National Academy of Sciences of the UnitedStates of America, 94, 1252–1257.Wright, S.J. (2002) Plant diversity in tropical forests: a review of mechanisms ofspecies coexistence. Oecologia, 130,1–14.Yan, X.F. & Cao, M. (2008) Seedling growth and survival of the endangeredtree species Shorea wantianshuea after a mast-fruiting event. Journal of PlantEcology (Chinese Version), 32,55–64.Received 5 November 2011; accepted 10 February 2012Handling Editor: David CoomesSupporting InformationAdditional supporting information may be found in the onlineversion of this article:Figure S1. Total number of living seedlings, including younger andolder seedling cohorts, in the 453 1 m · 1 m quadrats in each of thefive censuses.Figure S2. Seedling recruitment rates and mortality in all 4531m· 1 m quadrats over the dry-season and wet-season intervals.Figure S3. Distribution of the neighbourhood effects on seedling survivalfor heterospecific tree and seedling neighbours.Table S1. Akaike’s information criterion (AIC) values for generalizedlinear mixed models for survival of the younger seedling cohort in the20-ha Xishuangbanna tropical seasonal rain forest dynamics plot.Table S2. Akaike’s information criterion (AIC) values for generalizedlinear mixed models for survival of the older seedling cohort in the20-ha Xishuangbanna tropical seasonal rain forest dynamics plot.Table S3. Seedling survival rates of 186 species in the 453 1 m · 1mquadrats over each dry-season, wet-season and one-year interval andthe 2-year interval.Table S4. The species-specific effects of conspecific tree and seedlingneighbours on seedling survival over the dry-season interval in generalizedlinear mixed models with crossed random effects.Table S5. The species-specific effects of conspecific tree and seedlingneighbours and heterospecific seedling neighbours on seedling survivalover the 2-year interval in generalized linear mixed models withcrossed random effects.Table S6. Parameter estimates in linear regression models used to testfor the relationship between the maximum local conspecific neighbourdensities and population basal area in the community.Table S7. Parameter estimates in linear regression models used to testfor the relationship between the potential NDD and population basalarea in the community.As a service to our authors and readers, this journal provides supportinginformation supplied by the authors. Such materials may bereorganized for online delivery, but are not copy-edited or typeset.Technical support issues arising from supporting information (otherthan missing files) should be addressed to the authors.Ó 2012 The Authors. Journal of Ecology Ó 2012 British Ecological Society, Journal of Ecology, 100, 905–914376

Oikos 121: 952–960, 2012doi: 10.1111/j.1600-0706.2011.19831.x© 2012 Xishuangbanna Tropical Botanical Garden. Oikos © 2012 Nordic Society OikosSubject Editor: Martin F. Qigley. Accepted 25 August 2011Dominant species and dispersal limitation regulate tree speciesdistributions in a 20-ha plot in Xishuangbanna, southwest ChinaYue-Hua Hu , Li-Qing Sha , F. Guillaume Blanchet , Jiao-Lin Zhang , Yong Tang ,Guo-Yu Lan and Min CaoY.-H. Hu, L.-Q. Sha (shalq@xtbg.ac.cn), J.-L. Zhang, Y. Tang and M. Cao, Key Laboratory of Tropical Forest Ecology, Xishuangbanna TropicalBotanical Garden, Chinese Academy of Sciences, Mengla 666303, Yunnan, PR China. – Y.-H. Hu, Graduate Univ. of Chinese Academy ofSciences, Beijing 100049, PR China. – F. G. Blanchet, Dept of Renewable Resources, Univ. of Alberta, Edmonton, Alberta, T6G 2H1, Canada.– G.-Y. Lan, Rubber Research Inst., the Chinese Academy of Tropical Agricultural Sciences, Danzhou 571737, Hainan, PR China.Habitat heterogeneity and dispersal limitation are widely considered to be the two major mechanisms in determining treespecies distributions. However, few studies have quantified the relative importance of these two mechanisms at different lifestages of trees. Moreover, rigorous quantification of the effects of dominant tree species in determining species distributionshas seldom been explored. In the present study, we tested the hypothesis that the distribution of tree species is regulated bydifferent mechanisms at different life history stages. In particular, we hypothesised that dispersal limitation regulates thedistribution of trees at early life stages and that environmental factors control the distribution of trees as they grow, becauseof niche differentiation resulting from environmental filtering. To test this, trees in 400-m 2 quadrats in a 20-ha plot inXishuangbanna, southwest China were grouped into four classes on the basis of the diameter at breast height (DBH) thatroughly represent different stages in the life history of trees. A neighbourhood index was computed to represent a neutralspatial autocorrelation effect. We used both biotic (dominant species) and abiotic (topography and soil) predictor variablesto model the distribution of each target species while controlling for spatial autocorrelation within each of the DBHclasses. To determine which factors played the largest role in regulating target species distribution, the simulated annealingmethod was used in model selection based on Akaike information criterion (AIC) values. The results showed that therelative importance of neutral and niche processes in regulating species distribution varied across life stages. The neutralneighbourhood index played the most important role in determining the distributions of small trees (1 cm DBH 10 cm), and dominant species, as biotic environmental predictor variables, were the next most important regulators fortrees of this size. Environmental predictor variables played the most important role in determining the distributions of largetrees (10 cm DBH). This finding builds on previous research into the relative importance of neutral and niche processesin determining species distributions regardless of life stages or DBH classes.Over the past century, numerous theories, including theJanzen – Connell hypothesis (Janzen 1970, Connell 1978),niche assembly theories (Hutchinson 1957, Wright 2002)and unified neutral theory (Hubbell 2001), have been proposedto explain species coexistence. Niche theory, and morespecifically niche-assembly theory, has been proposed asa major mechanism for the coexistence of tree species(Silvertown et al. 1999). The core concept of niche theorystems from Darwin ’ s theory of evolution by natural selection,in which species are predicted to be closely associatedwith particular environmental conditions (Losos and Ricklefs2009). Niche theory itself can be traced back to half a centuryago, when Hutchinson (1957) proposed that each specieshas its own niche. Niche-assembly theory has motivatednumerous studies to model and test the effects of habitat onthe spatial and temporal distributions of species. These studieshave contributed greatly to our understanding of speciescoexistence in various ecosystems, including tropical (trees:John et al. 2007, palms: Tuomisto et al. 2003), subtropical(evergreen forest plants: Legendre et al. 2009, Wang et al.2009), temperate (understory ferns: Gilbert and Lechowicz2004) and savanna ecosystems (tree-grass mixtures: Sankaranet al. 2004).Habitat variations are also important factors in the regulationof species distributions, and species can vary in theirresponses to environmental factors (Gilbert and Lechowicz2004, Legendre et al. 2009, Tuomisto et al. 2003). For example,some species may be positively correlated with particularexplanatory variables whereas others respond negatively. Tounderstand the habitat associations of individual species,precise field-oriented studies are typically required (Clark et al.1998). For example, Levine and HilleRisLambers (2009)conducted an experiment on serpentine annual plants to testthe stabilising effects of niche differences. However, this typeof experiment may not be suitable for study species that havelong life spans, such as trees.952377

Although various methods have been developed for studyingspecies – habitat relationships, few of them incorporatedthe effect of spatial autocorrelations in species distributionanalysis before Legendre and Fortin (1989). Legendre (1993)suggested that spatial autocorrelation should be treated properlyin the analysis of spatial processes. The spatial autocorrelationin tree species distribution is primarily caused byenvironmental factors and community processes (Legendre1993). Of the community processes, dispersal limitation hasbeen proposed as the most important mechanism for themaintenance of species diversity (Hubbell et al. 1999,Hubbell 2001). Recently, many empirical and theoreticalstudies have provided strong support for these hypotheses(Shen et al. 2009, Chen et al. 2010).Abiotic factors such as soil properties, topographic factorsand gap openness are widely considered to be importantfactors in the regulation of species distributions(Hubbell et al. 1999, Harms et al. 2001, John et al. 2007).At mesoscales ( ≈ 1 – 100 km 2 ), abiotic factors have beenverified to be important in regulating species distributions(Clark et al. 1998, 1999), but at the local scale (0 – 1 km 2 ),more studies are still required. In addition, biotic factorssuch as neighbourhood effects are also important in determiningspecies distributions (Hubbell 2001, Uriarte et al.2004, Canham and Uriarte 2006). Traditionally, ecologistshave tended to use all of the surrounding neighbour treesto analyse the performance of a target species (Uriarte et al.2004). The dominant species have rarely been consideredseparately as predictor variables in quantifying the distributionsof target tree species.Another issue that is often ignored in species–habitatstudies is that conspecific trees may not respond to thesame explanatory variables consistently across life stages.So far, few studies have taken individual tree size intoconsideration in the analysis of species spatial distribution(Comita et al. 2007, Lai et al. 2009). Many studieshave assumed that all trees respond similarly to theenvironment regardless of life stage. Using the entirepopulation of a species regardless of the size classes oftrees may obscure our understanding of species – habitatrelationships.Previous studies have shown that the contributionof demographic niches to species distribution is limited(Condit et al. 2006), and trees are more clumped thanwould be predicted by a random distribution (Condit et al.2000). We hypothesise that neutral processes play a greaterrole than niche processes in determining trees distributions,especially at early life stages but that the relative importanceof neutral processes is reduced and species – habitatassociation is enhanced throughout tree growth because ofthe effects of environmental filtering (Norden et al. 2009).Because the dominant species are the most abundant andwidely distributed and have the largest total basal area in thecommunity, they might play an important role in determiningthe distributions of trees. In this study, we integrateddominant species, edaphic factors and topographic factorsinto a regression model to fit the spatial distribution of treespecies while controlling for a spatial autocorrelation effectalong multiple life stages in a seasonal rain forest in southwestChina.MethodsStudy areaThe study was conducted in a 20-ha tropical seasonal rain forestplot in Xishuangbanna, southwest China (21 ° 37 ′ 08 ″ N,101 ° 35 ′ 07 ″ E). This region is located on the northern edgeof the Asian tropical rain forests and is identified as a partof the Indo – Burma biodiversity hotspot in the list of the 25top priorities in global biodiversity conservation (Myers et al.2000). The elevation of the plot ranges from 708.2 m to869.1 m (Fig. 1). Three perennial streams traverse the plotand merge together at the southeastern corner. The area isdominated by warm, wet air masses from the Indian Oceanin the summer and by continental air masses from subtropicalregions in the winter, resulting in the alternationbetween rainy (May to October) and dry (November toApril) seasons. The soil is derived from both igneous andsedimentary rocks (Cao et al. 2006).Data collectionThe 20-ha plot was divided into 500 quadrats of 400 m 2 each.All trees with diameter at breast height (DBH) 1 cm weretagged, identified and mapped. All branches with DBH 1 cm were also tagged and measured for multi-stemmedtrees (trees with more than one stem).We further divided each 400-m 2 quadrat into 16 subquadratsof 25 m 2 each. If 70% or more of the total area ofa sub-quadrat consisted of open canopy and the average treeheight was under 10 m, the sub-quadrat was classified as a‘ gap ’ . A gap openness value between 0 and 16 was assigned toeach 400-m 2 quadrat based on the number of sub-quadratswithin that quadrat that were classified as gaps.Soil was sampled using a regular grid of 30 30 mthroughout the 20-ha plot. Each of the 252 nodes in thisgrid was used as a ‘ base point ’ . Together with each baseFigure 1. Topographic map of the 20-ha study plot.378953

point, two additional sampling points were located at randomcombinations of 2 and 5 m, 2 and 15 m or 5 and 15m along a random compass bearing away from the associatedbase point. At each sample point, 500 g of topsoil wascollected from a depth between 0 and 10 cm. A total of756 soil samples were taken. Fresh soil samples were placedinto pre-labelled plastic bags and shipped to the BiogeochemistryLaboratory at the Xishuangbanna Tropical BotanicalGarden. In the laboratory, the pH values of the soil sampleswere measured as immediately as possible using a potentiometerin fresh soil after water extraction (soil/water 1/2.5weight/volume). The soil samples were then air-dried,smashed, sieved using 1-mm and 0.15-mm mesh and storedin plastic bags for later additional analysis (Liu et al. 1996).Soil bulk density was measured using the corer method,soil organic matter was measured in soil oxidised withH 2 SO 4 – K 2 Cr 2 O 7 , and carbon content was estimated byvolume. The micro-Kjeldahl method was used to evaluatetotal nitrogen (N) using a mixture of H 2 SO 4 and K 2 SO 4 –CuSO 4 – Se catalyst, and an automatic steam distilling unitwas used to determine the soil N content in the solution.Micro-diffusion was used to determine the available N inthe soil.The soil was digested in HNO 3 – HClO 4 solution, and thetotal phosphorus (P) and potassium (K) were determinedusing an inductively coupled plasma atomic emission spectrometer.Extractable P was released from the soil in a solutioncontaining 0.03 mol l 1 NH 4 F and 0.025 mol l 1 HCl andestimated colorimetrically. Exchangeable K was extracted ina neutral 1-mol l 1 CH 3 COONH 4 solution, and the total Kin the extract was determined using an inductively coupledplasma atomic emission spectrometer. Table 1 presents thebasic statistics for the soil measurements described here.Gamma regression analysisTree species distributions are commonly described accordingto the numbers of individuals of a target species in quadratsof a certain area (Harms et al. 2001). Other measures, however,may provide additional insight into the mechanismsregulating tree species distributions (Morlon et al. 2009).The lattice basal area, which reflects tree size and biomassaccumulation of the target species within each quadrat, wasused as the response variable in this study. Correspondingly,a gamma regression model was used for regression analysisof the data (Dobson 1990). The gamma regression modelis a special case of the generalised linear model in which theTable 1. Basic statistics calculated for 756 soil measurements fromthe study plot.Soil explanatory variables SD MeanOrganic matter (g kg 1 ) 5.30 18.4Total N (g kg 1 ) 0.40 1.83Ammonium N (mg kg 1 ) 41.2 180Total P (g kg 1 ) 0.10 0.34Extractable P (mg kg 1 ) 6.27 4.89Total K (g kg 1 ) 3.46 11.2Exchangeable K (mg kg 1 ) 89.8 181pH 0.64 4.91Soil bulk density (g cm 3 ) 0.12 1.13error term follows a gamma distribution. We chose to usethis approach to model the total basal area because this variableis not normally distributed and thus not suitable formultiple regression analysis. The flexibility of the gammaregression is more appropriate for modelling the total basalarea. To relate the response variable to the various explanatoryvariables in the gamma regression, we used a commonlyused inverse link function.Th e basal area of each target species in each 400-m 2 quadratwas summed to obtain a vector. To avoid a zero-inflatedeffect and thus meet the positive data requirement of thegamma regression, we further removed all zero data and usedthe final vector as the response variable. Gamma regressionmodels were calculated only for those tree species that werepresent in at least 30 quadrats and had at least one individualwith DBH 1 cm in each quadrat. Rare species generatedunstable models.The purpose of grouping trees into different DBH classeswas to categorise trees on the basis of life stage. For a treewith multiple stems, we computed the basal area of eachstem. We then summed all the basal areas to obtain a totalbasal area for the tree. We then assigned a DBH to the multistemmedtree using the equation:multi-stemmed DBH2totalbasal area of allstemsπFollowing this transformation, we treated multi-stemmedtrees as single-stemmed trees of the same total basal area.We then grouped all trees into five DBH classes followingHe et al. (1997):class 1 1 to 5 cm DBH,class 2 5 to 10 cm DBH,class 3 10 to 25 cm DBH, andclass 4 DBH 25 cm;We further defined class 0 DBH 1 cm, which encompassesthe previous four classes.After grouping, there were 191, 147, 61, 57 and 22 treespecies in classes 0 to 4, respectively (Supplementary materialAppendix 1, Table A1).To explain the spatial distributions of the tree species,we used three groups of explanatory variables. Each variablewas centred and divided by its standard deviation, and avalue of 4 was then added to each variable in order to transformall the explanatory variables into positive numbers.The first group of variables included soil properties. Usingthe original soil data, an ordinary kriging was performed togenerate a sub-quadrat grid map of 10 10 m for each soilvariable (Cressie 1993). The values of the soil variables foreach 400-m 2 quadrat were calculated as the mean of the valuesat each of the nine nodes of the 10 10 m sub-quadratswithin that quadrat, as soil data were originally sampled ata scale of 30 30 m. This was performed using the geoRpackage in the R statistical language (R Development CoreTeam 2009).The topographic variables and the presence or absence ofgaps comprised the second group of explanatory variables.The topographic variables included the mean elevation,mean convexity, mean aspect and mean slope in each quadrat(Legendre et al. 2009).(1)954379

The third group of explanatory variables consisted of thetotal basal areas of the five most dominant tree species ineach quadrat ( Castanopsis echidnocarpa , Garcinia cowa ,Mezzettiopsis creaghii , Parashorea chinensis and Pittosporopsiskerrii ). These five tree species were identified as the mostdominant according to their relative importance values,calculated using the method described by Cao et al. (1996).The neighbourhood index was calculated by averaging thetotal basal area of conspecifics of the target species in each ofthe adjacent neighbour cells. The definition of neighbourhoodindex used here is similar to that of Wang et al. (2009),but we used the total basal area per quadrat of a target speciesinstead of stem counts. This neighbourhood index representsthe spatial autocorrelation effect.The gamma regression model was expressed as:Y X β 1 β 2 NI ε (2)where Y is a response variable, in this case the quadratbasedtotal basal area vector of a target tree species withina given DBH class, X is the explanatory variable matrixconsisting of the five dominant neighbours and the edaphicand topographic variables, β 1 and β 2 represent the slopesassociated with the explanatory variables in the originalexplanatory variable matrix X , NI is the isotropic secondorderspatial autoregressive factor and ε is a random errorterm. For each quadrat, neighbours were defined as thosequadrats with which the target quadrat shared a commonedge or border.Parameters of the gamma regression models were estimatedusing maximum likelihood with an inverse linkfunction. Because not all of the explanatory variables wereimportant in structuring the distribution of each speciesof tree, we constructed more parsimonious models usingsimulated annealing on the set of explanatory variables(Kirkpatrick et al. 1983). This optimisation approach determinesthe best model by applying mutations to the model(i.e. adding, removing or changing a variable). A ‘ better ’model is selected if it has a lower Akaike information criterion(AIC, Akaike 1974) value than the one selected previously.If the model does not yield a lower AIC value thanthe previous one, a probability function is used to evaluatewhether the model should be kept. In the analyses, we usedthe probability function used by Kirkpatrick et al. (1983),which is based on an acceptance parameter that defines howoften a ‘ bad ’ model will be accepted. For all of the modelselection procedures, we used a parameter of mutationacceptance equal to 200, a start time equal to 10 and anannealing temperature equal to 0.5 (Supplementary materialAppendix 2, R language code A2). This approach wasused because it does not require normal distribution of thevariables.To assess the relative importance of each of the explanatoryvariables in determining species distributions, a principalcomponent analysis (PCA) was conducted on a transformedmatrix of the p-values generated by the gamma regressionmodels. Following model fitting and model selection, eachselected explanatory variable was assigned a coefficientand a corresponding p-value. Because the magnitudes ofthe coefficients can vary greatly across species, they are notsuitable for assessing the relative importance of the explanatoryvariables. The p-values associated with these coefficientsreflect the relationship between the response variable andeach explanatory variable, and these p-values do not differin magnitude across species and thus are suitable for PCA.The p-values were transformed by log(1/p) to generate thep-value matrix for the PCA. A value of 0 was assigned to anyexplanatory variable that was removed by the model selectionprocedure. Five transformed p-value matrices were thenconstructed, one for each of the five DBH classes. Using thescores of each of the explanatory variables on the first twoprincipal component axes, we drew biplots to indicatethe relative importance of each explanatory variable based onthe length of each associated vector.To demonstrate the relative importance of each ofthe three groups of explanatory variables in explainingthe distribution of tree species, we plotted Venn diagramsbased on the number of tree species that responded toeach of the seven combinations of the three groups ofexplanatory variables in the most parsimonious modelsfor each of the five DBH classes. When the explanatoryvariables in a most parsimonious model matched one of thecombinations, we added 1 to that combination. The procedurewas repeated for the most parsimonious models for allspecies, and the numbers of responses to each of the sevencombinations of variable groups were shown in the Venndiagram.ResultsThe neighbourhood index yielded the longest vectors in thePCA analyses for DBH classes 0 and 2 (Fig. 2a, 2c) and thethird longest vector in the analysis for DBH class 1 (Fig. 2b).This indicates that the neighbourhood index is the mostimportant factor in determining the distribution of smalltrees in the study plot. The lengths of the vectors representingthe abiotic environmental factors gradually increased as theDBH of the trees increased (Fig. 2d, 2e). This was consistentwith our prediction that the relative contributions of neutraland niche processes change across life stages. Of the abioticenvironmental predictors, elevation yielded the longest vectorsin the PCA analyses for DBH classes 1, 2 and 3 andthus had the strongest effect on the distribution of trees inthese classes. The edaphic variables only showed a significantimpact on the distribution of large trees: the total and extractableN, P and K levels in the soil were negatively correlatedwith the distribution of the trees in DBH class 4 (Fig. 2e).The five most dominant tree species together covered27.85% of the total basal area in the plot. Mezzettiopsiscreaghii , with the fourth highest importance value (Supplementarymaterial Appendix 3, Table A3), had significanteffects on the distributions of smaller trees: the vectorsrepresenting Mezzettiopsis creaghii were among the threelongest in the PCA analyses for DBH classes 0, 1 and 2(Fig. 2a – c). Mezzettiopsis creaghii did not have as prominentan effect on the distributions of trees in DBH classes 3 and4 (Fig. 2d – e).Th e median variance explained in the gamma regressionmodels became higher as tree size increased: for classes 1380955

Figure 2. The variance explained by the most parsimonious gamma models of all species in each of 5 classes: (a) class 0 (DBH 1 cm), (b)class 1 (1 cm DBH 5 cm), (c) class 2 (5 cm DBH 10 cm), (d) class 3 (10 cm DBH 25 cm) and (e) class 4 (DBH 25 cm).to 4, the explained variances were 0.21, 0.22, 0.29 and 0.49,respectively (Fig. 3). The median explained variance for class0 was 0.36. The median variance explained in the gammaregression model for class 4 was significantly higher thanfor classes 1 and 2 (Kruskal – Wallis rank-sum test, Supplementarymaterial Appendix 4, Table A4). Except for class 0,the amount of variation explained in the gamma regressionmodels became smaller with increasing total basal area of atree species (Fig. 4).The numbers of tree species responding to each of theseven combinations of the three groups of explanatoryvariables varied greatly across DBH classes (Fig. 5). Thecombination of all of the variables explained the mostresponses in each of the five DBH classes, but this patternwas less pronounced in the larger DBH classes, suggestingthat most of the factors that affect the distribution of treesexert this effect when trees are at younger life stages. Theresults of the Kruskal – Wallis rank sum test on the sevenparts of each of the five Venn diagrams showed that thenumbers in the central portions were significantly greaterthan the numbers in each of the other six portions. Therewas no difference among these six other numbers. Thisresult indicates that the joint effects of the three groups offactors predominate in regulating most of the tree speciesdistributions throughout all life stages.Figure 3. Boxplot with 95% confidence intervals indicatingthe distributions of the variance explained in each species by themost parsimonious gamma regression models for each of the5 DBH classes. A to E on the x-axis denote class 0 (DBH 1 cm),class 1 (1 cm DBH 5 cm), class 2 (5 cm DBH 10 cm),class 3 (10 cm DBH 25 cm) and class 4 (DBH 25 cm),respectively.956381

Figure 4. Principal component analysis ordinations of the explanatory variables for each of the five tree size classes: (a) class 0 (DBH 1 cm), (b) class 1 (1 cm DBH 5 cm), (c) class 2 (5 cm DBH 10 cm), (d) class 3 (10 cm DBH 25 cm) and (e) class 4(DBH 25 cm). Matrices of the transformed p-values from the gamma regression models were used to compute the ordinations. Scoreson the first two ordination axes were plotted for the following explanatory variables: AN (available nitrogen), AS (aspect), CA ( Castanopsisechidnocarp a), CO (convexity), EK (exchangeable potassium), EL (elevation), EP (extractable phosphorus), GA ( Garcinia cow a), GP (gap),ME ( Mezzettiopsis creaghii ), NI (neighbourhood index), OM (organic matter), PA ( Parashorea chinensi s), pH (soil pH), PI ( Pittosporopsiskerrii ), TK (total potassium), TN (total nitrogen), TP (total phosphorus), SB (soil bulk density) and SL (slope).DiscussionMechanisms regulating species distributionsacross life stagesOur study revealed that dispersal limitation, representedas a neighbourhood index, has the largest effect on the distributionof trees across life stages whereas environ mental factorsmainly affect the distribution of large trees. For example,Mezzettiopsis creaghii , the fourth most dominant tree species,was significantly correlated with its own neighbourhoodindex in all five size classes. However, the relative roles of neutraland niche processes in determining species distributionsare still controversial. Gilbert and Lechowicz (2004) reportedthat niche-structuring predominantly determines species distributionsin a temperate forest understory. Tuomisto et al.(2003) showed that although both environmental factors anddispersal limitation jointly contribute to floristic differencesin western Amazonian forests, environmental factors are moreimportant than dispersal limitation. Legendre et al. (2009)suggested that environmental factors and neutral processesperformed equally in partitioning the beta diversity of treespecies in a 24-ha subtropical forest plot in Gutian, China.In a tropical forest in Panama, recruitment limitation hasbeen reported as the predominant factor controlling tree speciesdiversity (Hubbell et al. 1999). He et al. (1997) suggestedthat the relative contribution of any factor to explaining speciescoexistence could change over time and space. The resultsof our study suggest that both neutral and niche process areimportant in determining tree species distribution but thatthese processes play different roles at different life stagesof the trees. Neutral processes are more important in regulatingthe distribution of smaller trees, and niche processesbecome dominant in shaping the distribution of larger trees.Most tropical rain forest tree species tend to display aggregateddistributions (Condit et al. 2000). In a similar studyby Wang et al. (2009), few negative associations between treespecies distributions and neighbourhood index were foundin a subtropical forest. In the plot studied here, dispersallimitation, indicated by the neighbourhood index, showedsimilar numbers of positive and negative effects (Supplementarymaterial Appendix 5, Table A5). According to theJanzen and Connell hypothesis (Connell 1978, Janzen 1970),density-dependent effects cause conspecific individuals toescape from maternal trees, and this provides a theoreticalexplanation for the negative effect of neighbourhood index.382957

Figure 5. Venn diagrams displaying the number of species responding to each of the seven combinations of the three groups of explanatoryvariables, according to the most parsimonious gamma regression models, for each tree size class: (a) class 0 (DBH 1 cm), (b) class 1(1 cm DBH 5 cm), (c) class 2 (5 cm DBH 10 cm), (d) class 3 (10 cm DBH 25 cm), (e) class 4 (DBH 25 cm). Explanatoryvariables are represented as follows: Group 1 represents soil parameters, Group 2 represents topography and gap and Group 3 representsdominant neighbours and neighbourhood index.Stoll and Newbery (2005) also found strongly negativeneighbourhood effects on larger trees of the genus Shorea inSabah, Malaysia.It is widely accepted that topographic and edaphic factorsaffect species distributions in both subtropical and tropicalforests (John et al. 2007). In the present study, environmentalpredictors differed greatly in explaining the distributionsof large trees, and edaphic factors contributed very littleto explaining species distributions for DBH classes 0 to 2.However, the edaphic factors did explain a large proportionof the variation in the distributions of larger trees. Topographyis generally correlated with many environmental factors,such as water regime (Daws et al. 2002) and the physicaland chemical properties of the soil (Bourgeron 1983). In ourstudy, elevation showed a strong effect on the distribution oftrees in most DBH classes.Our results suggest that failing to classify trees into multiplelife stages may lead to a biased interpretation of themechanisms contributing to tree distributions. We founddifferent results for trees at different stages of life history. Atearly stages, when trees are small, their distribution is largelydetermined by dispersal limitation. This result is consistentwith the finding that the distribution of seedlings is mainlyaffected by seed dispersal and the presence of heterospecificneighbours (Comita and Hubbell 2009). Thus, we assumethat the habitat preferences of many tree species are consistentacross the sapling and juvenile stages. However, as thehabitat associations of most tree species becoming strongeras trees reach the mature stage (Lai et al. 2009), the habitatpreferences of tree species change at mature stage. Our resultsdo indicate that environmental predictors affecting speciesdistributions shifting across life stages (Fig. 4). Thus, environmentalpredictors do have filtering effects on species distributions,as Norden et al. (2009) suggested. Consequently,analysing species distributions based on the assumption thatall individuals of a target species respond similarly across lifestages may obscure the true process by which species distributionpatterns are generated.The effects of dominant species on speciesdistributionsOur gamma regression model of tree species distributionsindicated that the dominant tree species were importantin regulating species distributions. This result is consistentwith the finding that forest composition is highly deterministic(Yu et al. 1998) and that the dominant speciesplay important roles in shaping community composition.Comita and Hubbell (2009) found that dominant tree speciesplay an important role in shaping the distributions of958383

non-dominant tree species across life stages and that crossspeciesneighbourhood effects have a prominent influenceon seedling survival. Our results extended the effects ofinter-species interactions to established trees (DBH 1 cm).One possible explanation for the effect of dominant treespecies on the distributions of other tree species is thatdominant species have the ability to restructure multipleaspects of their surrounding environment, including lightavailability, soil textures and temperature (Eviner 2004). Arecent study further hypothesised that plants alter competitionby modifying the bioavailability of nutrients in therhizosphere (Raynaud et al. 2007). Moreover, althoughdominant species often compete with other species,they may also facilitate the establishment of other nondominantspecies (Supplementary material Appendix 6,Table A6). The species herd hypothesis (Wills 1996) providesone possible explanation for the cooperation betweendominant and non-dominant tree species. This hypothesisstates that an increasing density of heterospecific treespecies can slow the rate or reduce the chance that naturalpredators, pathogens or viruses will encounter their hostspecies. We conclude that it is important to pay moreattention to the influence of dominant neighbours on speciesdistributions.ConclusionsThe results of our study support the concept that neutral andniche processes jointly affect species distributions (Cottenie2005, Leibold and McPeek 2006). We further develop thisconcept by showing that the relative importance of each ofthe two processes varies across life stages. This implies thatif species are not analysed at multiple life stages, a biasedconclusion may be reached regarding the mechanismsmaintaining species coexistence. The negative impact of theneighbour index on species distributions that we observedin some cases provides empirical support to the Janzen andConnell hypothesis. Our results also present a quantitativeperspective on how dominant species can regulate speciesdistributions.Acknowledgements – Th is research was supported by grants from theNational Science and Technology Pillar Program (2008BAC39B02),the Q-CAS Biotechnology Fund (grant no. GJHZ1130) and theNational Science Foundation of China (31061160188). We giveour thanks to the Biogeochemistry Laboratory and the XishuangbannaStation for Tropical Rain Forest Ecosystem Studies of theXishuangbanna Tropical Botanical Garden for the analyses of soilnutrient concentrations and assistance in the field. Our sincereappreciation is given to Lei-lei Shi, Carol C. Baskin, Jerry M.Baskin, Xiao Cheng, Bernard Rollet, Fangliang He, DouglasSchaefer and Sijun Meng for their comments on the paper. Themanuscript was prepared during a visit by the authors to the Deptof Renewable Resources, Univ. of Alberta, Canada.ReferencesAkaike, H. 1974. A new look at the statistical identification model.– IEEE T. Automat. Contr. 19: 716 – 723.Bourgeron, P. S. 1983. Spatial aspects of vegetation structure.– Elsevier.Canham, C. D. and Uriarte, M. 2006. Analysis of neighborhooddynamics of forest ecosystems using likelihood methods andmodeling. – Ecol. Appl. 16: 62 – 73.Cao, M. et al. 1996. Tree species composition of a seasonal rainforest in Xishuangbanna, southwest China. – Trop. Ecol. 37:183 – 192.Cao, M. et al. 2006. Tropical forests of Xishuangbanna, China.– Biotropica 38: 306 – 309.Chen, L. et al. 2010. Community-level consequences of densitydependence and habitat association in a subtropical broad –leaved forest. – Ecol. Lett. 13: 695 – 704.Clark, D. B. et al. 1998. Edaphic variation and the mesoscale distributionof tree species in a neotropical rain forest. – J. Ecol.86: 101 – 112.Clark, J. S. et al. 1999. Seed dispersal near and far: patterns acrosstemperate and tropical forests. – Ecology 80: 1475 – 1494.Comita, L. S. and Hubbell, S. P. 2009. Local neighborhood andspecies ’ shade tolerance influence survival in a diverse seedlingbank. – Ecology 90: 328 – 334.Comita, L. S. et al. 2007. Developmental changes in habitat associationsof tropical trees. – J. Ecol. 95: 482 – 492.Condit, R. et al. 2000. Spatial patterns in the distribution of tropicaltree species. – Science 288: 1414 – 1418.Condit, R. et al. 2006. The importance of demographic niches totree diversity. – Science 313: 98 – 101.Connell, J. H. 1978. Diversity in tropical rain forests and coralreefs: high diversity of trees and corals is maintained only in anon-equilibrium state. – Science 199: 1302 – 1310.Cottenie, K. 2005. Integrating environmental and spatial processesin ecological community dynamics. – Ecol. Lett. 8:1175 – 1182.Cressie, N. A. C. 1993. Statistics for spatial data, revised edition.– Wiley.Daws, M. I. et al. 2002. Topographic position affects the waterregime in a semideciduous tropical forest in Panama. – PlantSoil 238: 79 – 90.Dobson, A. J. 1990. An introduction to generalized linear models.– Chapman and Hall.Eviner, V. 2004. Plant traits that influence ecosystem processes varyindependently among species. – Ecology 85: 2215 – 2229.Gilbert, B. and Lechowicz, M. J. 2004. Neutrality, niches, anddispersal in a temperate forest understory. – Proc. Natl Acad.Sci. USA 101: 7651 – 7656.Harms, K. E. et al. 2001. Habitat associations of trees andshrubs in a 50-ha neotropical forest plot. – J. Ecol. 89:947 – 959.He, F. L. et al. 1997. Distribution patterns of tree species in aMalaysian tropical rain forest. – J. Veg. Sci. 8: 105 – 114.Hubbell, S. P. 2001. The unified neutral theory of biodiversity andbiogeography. – Princeton Univ. Press.Hubbell, S. et al. 1999. Light-gap disturbances, recruitment limitation,and tree diversity in a neotropical forest. – Science 283:554 – 557.Hutchinson, G. E. 1957. Concluding remarks. – Cold SpringHarbor Symp. Quant. Biol. 22: 415 – 427.Janzen, D. H. 1970. Herbivores and the number of tree species intropical forests. – Am. Nat. 104: 501 – 528.John, R. et al. 2007. Soil nutrients influence spatial distributionsof tropical tree species. – Proc. Natl Acad. Sci. USA 104:864 – 869.Kirkpatrick, S. et al. 1983. Optimization by simulated annealing.– Science 220: 671 – 680.Lai, J. S. et al. 2009. Species–habitat associations change in asubtropical forest of China. – J. Veg. Sci. 20: 415 – 423.Legendre, P. 1993. Spatial autocorrelation: trouble or newparadigm? – Ecology 74: 1659 – 1673.384959

Legendre, P. and Fortin, M. 1989. Spatial pattern and ecologicalanalysis. – Plant Ecol. 80: 107 – 138.Legendre, P. et al. 2009. Partitioning beta diversity in a subtropicalbroad-leaved forest of China. – Ecology 90: 663 – 674.Leibold, M. A. and McPeek, M. A. 2006. Coexistence of the nicheand neutral perspectives in community ecology. – Ecology 87:1399 – 1410.Levine, J. M. and HilleRisLambers, J. 2009. The importance ofniches for the maintenance of species diversity. – Nature 461:254 – 258.Liu, G. S. et al. 1996. Soil physical and chemical analysis and descriptionof soil profiles. – Standards Press of China, in Chinese.Losos, J. B. and Ricklefs, R. E. 2009. Adaptation and diversificationon islands. – Nature 457: 830 – 836.Morlon, H. et al. 2009. Taking species abundance distributionsbeyond individuals. – Ecol. Lett. 12: 488 – 501.Myers, N. et al. 2000. Biodiversity hotspots for conservationpriorities. – Nature 403: 853 – 858.Norden, N. et al. 2009. Interspecific variation in seedling responsesto seed limitation and habitat conditions for 14 Neotropicalwoody species. – J. Ecol. 97: 186 – 197.Raynaud, X. et al. 2007. Plants may alter competition by modifyingnutrient bioavailability in rhizosphere: a modelingapproach. – Am. Nat. 171: 44 – 58.Sankaran, M. et al. 2004. Tree-grass coexistence in savannas revisited- insights from an examination of assumptions and mechanismsinvoked in existing models. – Ecol. Lett. 7: 480 – 490.Shen, G. C. et al. 2009. Species – area relationships explained by thejoint effects of dispersal limitation and habitat heterogeneity.– Ecology 90: 3033 – 3041.Silvertown, J. et al. 1999. Hydrologically defined niches reveal abasis for species richness in plant communities. – Nature 400:61 – 63.Stoll, P. and Newbery, D. 2005. Evidence of species-specific neighborhoodeffects in the Dipterocarpaceae of a Bornean rainforest. – Ecology 86: 3048 – 3062.Tuomisto, H. et al. 2003. Dispersal, environment, and floristic variationof western Amazonian forests. – Science 299: 241 – 244.Uriarte, M. et al. 2004. A neighborhood analysis of tree growthand survival in a hurricane-driven tropical forest. – Ecol.Monogr. 74: 591 – 614.Wang, Z. et al. 2009. Species-topography association in a species-richsubtropical forest of China. – Basic Appl. Ecol. 10: 648 –655.Wills, C. 1996. Safety in diversity. – New Sci. 149: 38 – 42.Wright, S. J. 2002. Plant diversity in tropical forests: a review ofmechanisms of species coexistence. – Oecologia 130: 1 – 14.Yu, D. W. et al. 1998. Can high tree species richness be explainedby Hubbell’s null model? – Ecol. Lett. 1: 193 – 199.Supplementary material (available as Appendix O19831 at www.oikosoffice.lu.se/appendix ). Appendix 1 – 6.960385

Strong Neutral Spatial Effects Shape Tree SpeciesDistributions across Life Stages at Multiple ScalesYue-Hua Hu 1,2 , Guo-Yu Lan 3,4 , Li-Qing Sha 1 , Min Cao 1 *, Yong Tang 1 , Yi-De Li 2 , Da-Ping Xu 21 Key Laboratory of Tropical Forest Ecology, Xishuangbanna Tropical Botanical Garden, Chinese Academy of Sciences, Mengla, Yunnan, China, 2 Research Institute ofTropical Forestry, Chinese Academy of Forestry, Guangzhou, Guangdong, China, 3 Danzhou Key Field Station of Observation and Research for Tropical AgriculturalResources and Environments, Ministry of Agriculture, Danzhou, Hainan, China, 4 Rubber Research Institute, the Chinese Academy of Tropical Agricultural Sciences,Danzhou, Hainan, ChinaAbstractTraditionally, ecologists use lattice (regional summary) count data to simulate tree species distributions to explore speciescoexistence. However, no previous study has explicitly compared the difference between using lattice count and basal areadata and analyzed species distributions at both individual species and community levels while simultaneously consideringthe combined scenarios of life stage and scale. In this study, we hypothesized that basal area data are more closely relatedto environmental variables than are count data because of strong environmental filtering effects. We also address thecontribution of niche and the neutral (i.e., solely dependent on distance) factors to species distributions. Specifically, weseparately modeled count data and basal area data while considering life stage and scale effects at the two levels withsimultaneous autoregressive models and variation partitioning. A principal coordinates of neighbor matrix (PCNM) was usedto model neutral spatial effects at the community level. The explained variations of species distribution data did not differsignificantly between the two types of data at either the individual species level or the community level, indicating that thetwo types of data can be used nearly identically to model species distributions. Neutral spatial effects represented by spatialautoregressive parameters and the PCNM eigenfunctions drove species distributions on multiple scales, different life stagesand individual species and community levels in this plot. We concluded that strong neutral spatial effects are the principalmechanisms underlying the species distributions and thus shape biodiversity spatial patterns.Citation: Hu Y-H, Lan G-Y, Sha L-Q, Cao M, Tang Y, et al. (2012) Strong Neutral Spatial Effects Shape Tree Species Distributions across Life Stages at MultipleScales. PLoS ONE 7(5): e38247. doi:10.1371/journal.pone.0038247Editor: Ben Bond-Lamberty, DOE Pacific Northwest National Laboratory, United States of AmericaReceived January 16, 2012; Accepted May 2, 2012; Published May 29, 2012Copyright: ß 2012 Hu et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricteduse, distribution, and reproduction in any medium, provided the original author and source are credited.Funding: This study was supported by a grant from the National Science Foundation of China (31061160188) and a grant from the Queensland-ChineseAcademy of Science Joint Biotechnology Projects (GJHZ1130). The funders had no role in study design, data collection and analysis, decision to publish, orpreparation of the manuscript.Competing Interests: The authors have declared that no competing interests exist.* E-mail: caom@xtbg.ac.cnIntroductionHow large numbers of species coexist at a local scale (,1 km 2 )isachallenging question for ecologists. With the rapid improvement ofcomputertechnologyandstatisticaltools,itisnowfeasibletointegrateboth niche and neutral processes into models to analyze speciesdistribution data. Analytical methods, such as regression [1],ordination and machine learning, can be used to investigate themechanisms underlying species coexistence [2,3]. Traditionally,ecologists have used individual lattice count data to simulate speciesdistributions at the individual species or community levels [4,5]. Inthis method, trees are always counted as individuals regardless offactors such as age, size, branching and whether re-sprouting hasoccurred. However, the habitat associations of tree species may varyacross life stages [6,7], and thus, tree intensity variation across latticesmay be insufficient to reflect species distribution patterns.Many other traits of tree species can be used to simulate theirdistributions, such as percent cover, point quadrat frequency,biomass, basal area and energy and resource use [8]. These featuresmay provide novel insights for understanding species distributionsand their organizing mechanisms. We are unaware of previous workexplicitly comparing the results of using these features and the resultsof using individual count data to model species distributions. Basalarea, which represents tree size, plays a key role in determining thefunctional differences among species [9]. Basal area also correlateswith biomass accumulation and reflects the ability of trees tocompete for soil nutrients [10]. A comparative study in whichindividual count data and basal area data are examined separatelywill reveal the extent to which different results are generated by thetwo types of data.Most species tend to be clumped in their dispersion pattern [11],which may cause strong spatial autocorrelation, i.e., greater or lesssimilarity in variables located close to each other than would beexpected if species were distributed randomly across geographicspace [12]. This is commonly observed in species spatial distributiondata [1]. To control Type I error rates and obtain good parameterestimates, it is necessary to use spatially explicit models in spatialanalyses of species distributions [1,12,13]. In addition, environmentalfactors, such as topography and soil, are also widelyconsidered in models of species distributions [14,15]. Integratingspatial effects and environmental variables in species distributionmodels is generally accepted by ecologists [16,17,18].The effects of life stage and scale are critical for analyzing spatialdistributions of tree species. Physiological requirements, selectivepressures and distribution patterns can vary across the life stages ofplant species, which can lead to a shift in habitat preferencePLoS ONE | www.plosone.org 1 May 2012 | Volume 7 | Issue 5 | e38247386

Spatial Effects Shape Tree Species Distributionsthroughout its ontogeny [19,20,21]. In fact, numerous empiricalstudies have identified that the mechanisms underlying tree speciesdistributions do vary across different life stages in some forestdynamics plots [6,7,22,23]. Similarly, previous studies point out thatanalyses results can differ at different scales in ecological studies[24,25], indicating that the scale effect is important for tree speciesdistributions [4,5].In this study, we modeled lattice count data and basal area dataat the individual species level and at the community level whilesimultaneously considering scale and life stage effects. At theindividual species level, a simultaneous autoregressive (SAR)model was used. The spatially autocorrelated variation in theerror term of the SAR model is determined by cell connectivity,and the cell connectivity of the lattice basal area data and thecount data is exactly the same based on cell positions. Therefore,the spatial structure should be identical for the basal area data andthe count data. Under this premise, we hypothesized that basalarea data are more closely related to environmental variables andpredicted that the R-squared value of the fitted model based onbasal area data would be higher than that based on count databecause of strong environmental filtering effects. At the communitylevel, we partitioned the variation in community compositionbetween environmental variables and spatial effects for each of thetwo types of data. At this level, we also predicted that the variationexplained by environmental variables would be higher for basalarea data than for individual count data, also owing to strongenvironmental filtering effects.Materials and MethodsStudy Site and Data CollectionWe analyzed tree species distributions within a 20-ha tropicalforest dynamics plot (21u379080N, 101u359 070E) in Xishuangbanna,Southwest China [26]. The community was an old-growthnatural tropical seasonal rainforest tree community (more than200 years old), but a small portion of the plot, located on themountain ridge, was disturbed by humans approximately 40 yearsago. The tree community was dominated by Parashorea chinensis, anemergent canopy species with a maximum height of approximately60 m. Detailed descriptions of the climate, geology and flora ofXishuangbanna can be found in Cao et al. [27] and Zhu et al. [28].The 20-ha plot was established in 2007, and a topographic surveywas conducted of each node of a 10-m grid throughout the plot. Allstems with a DBH (diameter at breast height) $1 cm were tagged,mapped, measured and identified. There were 468 tree and shrubspecies with individuals of DBH $1 cm in this plot [26].To examine the mechanisms underlying any differencesbetween the results obtained from the basal area and count dataacross life stages, we defined trees with DBH $1 cm as class 0.This class was itself divided into four DBH classes, representingdifferent life stages of trees. This categorization of DBH classesfollowed He et al. [21]:For a tree with multiple stems, we computed a proxy DBH andthen classified the tree based on this proxy DBH. The calculationof proxy DBH followed Hu et al. [23].To evaluate the influence of scale on species distribution, wegrouped the trees within each DBH class using cells of 10610 m,20620 m, 25625 m and 50650 m in size. This generated 20combinations of DBH classes and cell sizes. Each DBH-cell sizecombination contained a group of tree species, and for each DBHcellsize combination, the tree species that occurred in at least 30 cellswere chosen for regression analysis and variation partitioning(Table 1).At each scale, the four topographic attributes of altitude,convexity, slope and aspect were calculated for each cell. Thesecalculations followed Legendre et al. [5]. Third-degree polynomialequations were constructed with altitude, convexity and slope. Thevariables sin(aspect) and cos(aspect) were calculated from theaspect and used as explanatory variables. Finally, we obtained 11expanded topographic variables. For 25625 m cells, the altitudinalvalues at all nodes were interpolated by kriging the raw datafrom the 10610 m cells.Because soil attributes are crucial to species distributions, wecollected 756 soil samples from throughout the 20-ha plot [23]. Ninesoil attributes were analyzed, including available nitrogen, exchangeablepotassium, extractable phosphorus, organic matter, soilpH, total potassium, total nitrogen, total phosphorus and soil bulkdensity, following the methods of Liu et al. [29]. For the soilattributes at each scale (cell size), the values at the four corners ofeach cell were interpolated by kriging from the 756 samples. Afterinterpolation, the mean value of each soil attribute at the fourcorners of each cell was assigned as the value for that cell. Thisprocedure was applied to each of the four scales of cell size. At eachscale, we calculated the principal components from the mean valuesof the nine soil attributes and used only the first three components.Together, these first three components explained 84.5%, 83.5%,86.9% and 89.1% of the total variation in soil attributes for the fourcell sizes from 10610 m to 50650 m, respectively.The Simultaneous Autoregressive (SAR) ModelGuisan et al. [30] suggested that regression and ordinationmethods are both suited for species-specific and multiple speciesmodels. We chose the SAR model for the regression analyses ofindividual species because SAR has commonly been used forlattice summary data [31]. Specifically, the SAR spatial errormodel was used in this study in the following form:Y~XbzlWmzeTable 1. Number of tree species in each of the 20combinations of DBH and cell size.ð1ÞClass0 : DBH§1cmClass1 : DBH~1tov5cmClass2 : DBH~5tov10cmClass3DBH~10tov25cmClass4 : DBH§25cmDBH class 10610 m 20620 m 25625 m 50650 mClass 0 206 192 187 153Class 1 163 147 147 111Class 2 70 61 56 33Class 3 62 58 54 30Class 4 25 22 21 10Note: class 0 (DBH $1 cm), class 1 (1 cm # DBH ,5 cm), class 2 (5 cm # DBH,10 cm), class 3 (10 cm# DBH ,25 cm), class 4 (DBH $25 cm).doi:10.1371/journal.pone.0038247.t001PLoS ONE | www.plosone.org 2 May 2012 | Volume 7 | Issue 5 | e38247387

Spatial Effects Shape Tree Species Distributionswhere Y is the response variable, in this case, the lattice count orbasal area vector of a focal tree species at a particular cell size in aparticular DBH class; X is the explanatory variable matrixconstituted by the first three principal components of the soilvariables and the 11 topographic variables at a particular cell size;b is a slope vector associated with the explanatory variables; l isthe spatial autoregressive coefficient; m is a spatially dependenterror term; e is a random error term; and W is the spatialweighting matrix that indicates whether the cells are neighbors ornot. The weight is defined as 1 if cells are immediate vertical andhorizontal neighbors and 0 otherwise. For each focal cell, the cellssharing a common edge (border) with it were defined as itsneighbor cells and were weighted by 1, and all other cells wereweighted by 0. To avoid zero-inflated effects on the regressionanalysis, cells containing no trees were removed for each species.We found that the R-squared values obtained by fitting only thenon-zero data were significantly higher than those obtained whenthe zero data were included.To evaluate the relative importance of all of the explanatoryvariables in determining species distributions for each of the 20combinations of DBH class and cell size for each type of data(lattice basal area and count), a principal component analysis(PCA) was used to analyze a transformed p-value matrix. TheSAR model yields a p-value for each of the explanatory variablesand l, and the p-values for all species can be formatted as amatrix. Two such matrices were generated for each of the 20DBH-cell size combinations: one that used basal area data and onethat used count data. Because the p-values reflected theassociations between responsible and explanatory variables in aninverse manner, the p-values themselves could not be used directlyfor the PCA. As a result, we performed a transformationprocedure on the p-values to obtain the transformed p-valuematrix, which positively reflected the association between responsibleand explanatory variables and were suitable for PCA. Themethod used to transform the p-values followed Hu et al. [23].Some of the p-values were small enough that a value of 0 wasreturned by the SAR model in the R statistical language [32], andthe transformation procedure could not be applied to these p-values. To address this issue, p-values smaller than 10 216 wereassigned a proxy value of 10 216 . In each analysis, we plotted thescores of all of the explanatory variables on the first two principalcomponent axes as arrows and assessed the relative importance ofthe explanatory variables based on their vector lengths.Community Composition Variation PartitioningTo quantify the contributions of the spatial and environmentalvariables to the variation in community composition observed foreach of the two types of data, variation partitioning based oncanonical redundancy analysis was applied [33]. The topographicand soil variables were grouped together as environmentalvariables for this analysis. To represent spatial variables, theprincipal coordinate neighbor matrix (PCNM) eigenfunctions werecomputed across all cells at each scale of cell size [5]. PCNMs withpositive eigenvalues were retained, and forward selection (using apermutation test with 999 permutations and a 5% significancelevel) was used to identify the significant PCNMs. These selectedPCNMs represented the spatial effects. We then partitioned thecontributions of the environmental variables and the PCNMs.This procedure was repeated for each of the 20 DBH-cell sizecombinations for basal area data and count data.To compare the R-squared values of the fitted regressive modelsas well as the total explained variation in community compositionbased on count data and basal area data, a Kruskal-Wallis rank-sumtest was performed. We conducted SAR analyses and variationpartitioning with the R (version 2.13.0) statistical language with the‘‘errorsarlm’’ function of the ‘‘spdep’’ package and the ‘‘varpart’’function of the ‘‘vegan’’ package, respectively [32].ResultsWe found no significant differences in the R-squared valuebetween the fitted SAR models based on the two types of data forany of the 20 DBH-cell size combinations, except for class 0 at thescales (cell sizes) of 20620 m and 50650 m (p-values of Kruskal-Wallis rank-sum test: 0.0087 and 0.0261, respectively). Among the20 DBH-cell size combinations, the median R-squared value ofthe fitted models based on count data were greater than that basedon basal area data in 14 cases, but only two of these cases werestatistically significant. By contrast, basal area data generatedgreater R-squared values than count data in only 6 cases, andnone of these differences were statistically significant. Fig. 1illustrates the distributions of the R-squared values generated bythe fitted SAR models based on each of the two types of data forthe 20 DBH-cell size combinations.There was a positive trend in the R-squared value withincreasing cell size (Fig. S1). However, there was no clearrelationship between the R-squared value and the DBH class(Fig. S2). There was a negative relationship between the R-squared value and the total abundance of the studied species,except at the 50-m scale (Fig. 2, Figs. S3, S4, S5).The spatial autoregressive parameter l of the SAR fitted modelhad the longest vectors in the 20 DBH-cell size combinations forboth count data and basal area data (Figs. 3 and 4, Figs. S6, S7,S8, S9, S10, S11), indicating that spatial effects played a moreimportant role than any of the environmental variables indetermining tree species distributions in this forest plot.The results of the community composition variation partitioningwere consistent with the results generated by the regression analysisperformed at the individual species level. Both analyses indicatedthat spatial effects are dominant in determining species distributions(Tables 2 and 3). There was no significant difference in the fraction ofvariation explained by the pure environmental variables when eitherthe count data or the basal area data were used. For 10 of the 20DBH-cell size combinations, the variation explained by theenvironmental variables was higher when basal area data wereused than when count data were used. In addition, the Kruskal-Wallis rank-sum test revealed no significant difference in the totalvariation explained by the combined effects of spatial andenvironmental variables when either the basal area data or thecount data were used. However, the count data yielded higher totalexplained variation than did the basal area data for 17 of the 20DBH-cell size combinations. For both types of data, the totalexplained variation tended to increase as the scale (cell size)increased (Fig. 5). By contrast, for both types of data, the totalexplained variation decreased with an increase in the DBH size class(Fig. S12).DiscussionThe Contributions of Environmental VariablesAlthough environmental variables constrained a portion of thevariations in the species distribution data, the variation partitioningresults demonstrate that the environmental variables arestrongly structured by PCNM eigenfunctions. In other words, thepure environmental variables play a limited role in determiningspecies distributions, and most of the variations in environmentalvariables are derived from distance limitation. Among theenvironmental variables, the nonlinear topographic variables andPLoS ONE | www.plosone.org 3 May 2012 | Volume 7 | Issue 5 | e38247388

Spatial Effects Shape Tree Species DistributionsFigure 1. Distributions of the R-squared values of the fitted SAR models based on count data and basal area data for each of the 20combinations of DBH and cell size. Each row represents a distinct scale of cell size; 0 to 4 in the x-axis labels represent DBH class 0 to 4,sequentially; ‘‘-CO’’ and ‘‘-BA’’ in the x-axis labels represent count data and basal area data, respectively. Classes 0 to 4 are defined as follows: class 0(DBH $1 cm), class 1 (1 cm # DBH ,5 cm), class 2 (5 cm# DBH ,10 cm), class 3 (10 cm # DBH ,25 cm) and class 4 (DBH $25 cm).doi:10.1371/journal.pone.0038247.g001the first two principal components of soil variables contribute moreto species distributions than other environmental variables (Figs. 3and 4, Figs. S6, S7, S8, S9, S10, S11). This implies that theoriginal topographic variables play little role in regulating speciesdistributions. In turn, this also is consistent with why Harms et al.[14] suggest that the original topographic variables contribute littleto the species distributions. The nonlinear effect has been reportedto work well for species habitat associations under the scenarios ofhabitat loss, patch size and isolation [34]. Because there is no suchdistinct abrupt change of environmental variables at this study site,the nonlinear effect does not dominantly contribute to speciesdistribution in this study.Strong Neutral Spatial EffectsOur analyses based on both count data and basal area dataindicate that neutral spatial effects, which are specificallyrepresented by spatial autoregressive parameters of SAR andPCNM eigenfunctions in this study, predominantly regulate treeFigure 2. Relationships between the R-squared values of the fitted SAR models and total species abundance for each of the 5 DBHclasses at the 20-m scale. Circles and triangles represent count data and basal area data, respectively. Classes 0 to 4 are defined as in Figure 1.doi:10.1371/journal.pone.0038247.g002PLoS ONE | www.plosone.org 4 May 2012 | Volume 7 | Issue 5 | e38247389

Spatial Effects Shape Tree Species DistributionsFigure 3. Principal component analysis ordinations (based on matrices of transformed p-values from the SAR models) of the 14explanatory variables and the spatial autoregressive parameter l for each of the 5 DBH classes at the 20-m scale of the count data.Classes 0 to 4 are defined as in Figure 1. The abbreviations in the third-degree polynomial equations of altitude, convexity and slope are as follows:altitude (AL), altitude 2 (AL2), altitude 3 (AL3), convexity (CO), convexity 2 (CO2), convexity 3 (CO3), slope (SL), slope 2 (SL2) and slope3 (SL3). Theabbreviations of the sine-cosine function of aspect and the spatial autoregressive parameter l are as follows: cos(aspect) (CA), sin(aspect) (SA) and l(LA). The abbreviations of the first three principal components of the soil variables are as follows: the first principal component (CP1), the secondprincipal component (CP2) and the third principal component (CP3).doi:10.1371/journal.pone.0038247.g003Figure 4. Principal component analysis ordinations (based on matrices of transformed p-values from the SAR models) of the 14explanatory variables and the spatial autoregressive parameter l for each of the 5 DBH classes at the 20-m scale, obtained withbasal area data. Classes 0 to 4 are defined as in Figure 1. The abbreviations are defined as in Figure 3.doi:10.1371/journal.pone.0038247.g004PLoS ONE | www.plosone.org 5 May 2012 | Volume 7 | Issue 5 | e38247390

Spatial Effects Shape Tree Species DistributionsFigure 5. Distribution patterns of the total explained variation in community composition for each of the four scales of cell sizebased on count data and basal area data. The reduplicate data at each scale consisting of the 5 total explained variations of the 5 DBH classes ofthe variation partitioning results at each of the four scales.doi:10.1371/journal.pone.0038247.g005species distributions across multiple life stages and scales at eitherindividual species or community levels. The spatial autoregressiveparameter and PCNM eigenfunctions are both distance-limitedfactors, while distance is a key concept of neutral theory [35]; thus,we conclude that neutral processes are essential to the tree speciesdistributions at the study site. In contrast to previous studies thathave focused on the scale at the individual species level [4] orcommunity level [5] or life stages at the individual species level [7]or community level [22], our study integrates all of the four scalesof analysis to conclude that neutral spatial effects play a dominantrole in determining species distributions. Furthermore, we verifythis conclusion with both count data and basal area data.In the present study, we extend the previously demonstratedcrucial role of neutral spatial effects in shaping species distributionsto multiple life stages for both basal area data and count data.Without categorizing trees into different DBH classes, manystudies have verified that neutral spatial effects are the principaldeterminants of species distribution patterns [18,36,37]. He et al.[21] demonstrate that tree species distributions maintain aggregatedpatterns at all life stages, and we demonstrate here thatneutral spatial effects are the dominant driver of tree speciesdistributions throughout life stages. Seidler and Plotkin [38] findthat seed dispersal modes are strongly correlated with the spatialaggregation of intra-species from saplings to mature trees in a 50-ha plot of Malaysian tropical forest, supporting our findings.However, this seems to vary between different forest dynamicsplots. Lai et al. [7] showed that there are strong tree specieshabitat associations at different life stages at the individual specieslevel. Kanagaraj et al. [22] demonstrated that habitat preferencestrongly determined species distributions at the juvenile stage, butneutral processes dominated the reproductive stage at thecommunity level. As far as our study is concerned, both basalarea and count data demonstrated that neutral processesoverwhelmingly regulated species distributions across life stagesat multiple scales at the individual species and community levels.We suggest analyzing data from multiple sites with one unifiedstatistical method to produce more comparable results.Legendre [12] suggests that either environmental variables orcommunity processes may result in spatial autocorrelation, whichrepresents the neutral spatial effect, of species distribution data.Because the two most-recognized environmental variables (topographyand soil) play a limited role in determining speciesdistributions in this study, community processes could be thecrucial reasons for the spatial autocorrelation of species distributiondata. Among the potential community processes, a distancelimiteddispersal process has been identified as a principal processfor producing tree species distributions in previous studies [18,38].Because both the spatial autoregressive parameters and PCNMeigenfunctions are distance-limited factors and the dispersalprocess is also distance-limited [39], we suggest that dispersallimitation serves as the major community process generating treespecies distributions in this plot. In turn, this explains why countdata and basal area data yield almost identical outcomes in thetwo-level analyses and is also consistent with previous studiesreporting that tree species distributions are more clumped thanrandom [11,21,40].Strong neutral spatial effects are also consistent with theargument that investigating species spatial distributions withoutconsidering spatial autocorrelation may bias the analysis results[1,13]. Kühn [41] even suggests that the analysis results may beinverted for the same data between analyses with and without theincorporation of spatial autocorrelation. Our results show thatenvironmental variables do contribute to the tree speciesdistributions to some extent, but both SAR and variationpartitioning analyses demonstrate that neutral spatial effects aredominant in this plot.Count and Basal Area DataContrary to our expectation that the environmental variablesmay be more closely related to the basal area data, the pureenvironmental variables were identically related to basal area andPLoS ONE | www.plosone.org 6 May 2012 | Volume 7 | Issue 5 | e38247391

Spatial Effects Shape Tree Species DistributionsTable 2. Results of the partitioning variation betweenenvironmental variables and spatial effects for each of the 20combinations of DBH and cell size using basal area data.Table 3. Results of partitioning variation betweenenvironmental variables and spatial effects for each of the 20combinations of DBH and cell size using count data.Cell size (m) DBH class [a] (%) [b] (%) [c] (%) [d] (%)10610 Class 0 0.16 1.720 11.80 86.31Class 1 0.44 15.30 30.71 53.55Class 2 0.41 10.80 21.90 66.89Class 3 0.25 4.41 13.58 81.76Class 4 0.16 1.57 14.89 83.3820620 Class 0 0.98 9.45 19.37 70.20Class 1 0.49 31.57 40.92 27.02Class 2 0.62 27.90 35.55 35.94Class 3 0.60 16.08 29.03 54.29Class 4 0.71 9.18 22.08 68.0425625 Class 0 0.58 10.07 21.60 67.75Class 1 0.51 25.47 50.94 23.08Class 2 1.02 20.97 45.09 32.92Class 3 0.38 15.28 32.47 51.88Class 4 0.63 9.83 19.42 70.1250650 Class 0 6.85 19.94 22.94 50.27Class 1 1.79 33.66 40.75 23.80Class 2 2.21 37.14 33.35 27.30Class 3 10.46 21.74 15.97 51.83Class 4 8.35 15.82 37.35 38.48Note: Adjusted R-squared statistics are shown. Fractions [a] – [d] are as follows:[a] = variation explained by the environmental variables and not spatiallystructured; [b] = variation explained by the environmental variables andspatially structured; [c] = spatially structured variation not explained by theenvironmental variables; [d] = residual variation. Fraction [b] is the intersectionof the variation explained by linear models of the two groups of explanatoryfactors. Topographic and edaphic variables were used to compute fractions [a]and [b]. Principal coordinates of neighbor matrix eigenfunctions were used asexplanatory variables to compute fractions [b] and [c]. class 0 (DBH $1 cm),class 1 (1 cm # DBH ,5 cm), class 2 (5 cm # DBH ,10 cm), class 3 (10 cm #DBH ,25 cm), class 4 (DBH $25 cm).doi:10.1371/journal.pone.0038247.t002Cell size (m) DBH class [a] (%) [b] (%) [c] (%) [d] (%)10610 Class 0 0.58 15.83 39.51 44.08Class 1 0.47 13.28 38.65 47.60Class 2 0.54 12.86 23.86 62.74Class 3 0.28 4.95 15.60 79.17Class 4 0.22 3.99 15.68 80.1220620 Class 0 0.45 31.95 45.30 22.31Class 1 0.29 28.42 47.47 23.82Class 2 0.45 30.91 36.56 32.08Class 3 0.69 16.93 31.79 50.58Class 4 0.53 17.82 28.26 53.3925625 Class 0 0.49 23.96 56.15 19.40Class 1 20.04 22.66 57.29 20.08Class 2 0.77 23.26 46.54 29.44Class 3 0.39 16.46 35.49 47.66Class 4 1.28 13.52 30.81 54.3950650 Class 0 2.66 31.52 39.54 26.27Class 1 2.26 28.89 40.78 28.07Class 2 5.43 35.39 31.38 27.80Class 3 3.41 29.35 26.47 40.76Class 4 6.60 20.79 18.88 53.73Note: Adjusted R-squared statistics are shown. Fractions [a] – [d] are as follows:[a] = variation explained by the environmental variables and not spatiallystructured; [b] = variation explained by the environmental variables andspatially structured; [c] = spatially structured variation not explained by theenvironmental variables; [d] = residual variation. Fraction [b] is the intersectionof the variation explained by linear models of the two groups of explanatoryfactors. Topographic and edaphic variables were used to compute fractions [a]and [b]. Principal coordinates of neighbor matrix eigenfunctions were used asexplanatory variables to compute fractions [b] and [c]. class 0 (DBH $1 cm),class 1 (1 cm # DBH ,5 cm), class 2 (5 cm # DBH ,10 cm), class 3 (10 cm #DBH ,25 cm), class 4 (DBH $25 cm).doi:10.1371/journal.pone.0038247.t003count data in terms of the community composition variationpartitioning results. This suggests that count data may be moreappropriate for analyzing species distributions than basal area datain this plot. However, count data may not be suitable forregression analyses when species are evenly distributed across allcells in which they are present, as may occur for species with smallpopulation sizes. For example, in this study, only one individual ofCanthium simile in DBH class 0 was counted in each cell at the 10-mspatial scale, and this resulted in an infinite value of R-squared inthe SAR models.Effects of Spatial Scale on Species DistributionsThe increase in cell connectivity with cell size observed in thisstudy may explain both the increases in the R-squared values andthe total explained variation in the results of the SAR models andvariation partitioning, repectively. As an example, cell connectivityclearly increased with increasing cell size for trees of Sloaneatomentosa in class 4 (Fig. 6). This results in decreasing p-values of lwith increasing cell size, except when basal area data were used atthe 10-m scale. The R-squared values of the fitted models for S.tomentosa tend to increase with increasing cell size, except whencount data are used at the 10-m scale (Fig. S13), consistent withprevious work demonstrating that the variation explained by auto-Poisson regressive models when count data were used was muchsmaller at the 10-m scale than at the 20-m and 25-m scales in a 20-ha subtropical forest plot in southern China [4]. By contrast, in astudy of the beta diversity of tree species in a 24-ha subtropicalforest plot, Legendre et al. [5] found that the total explainedvariations in species richness and community composition variedlittle across sampling scales. Here, we found that the R-squaredvalues decreased with increasing total abundance of species (Figs. 2,Figs. S3, S4, S5), in contrast to the finding of Wang et al. [4].Because we simultaneously considered spatial scale and life stage,our analyses generated more replicates than in previous studies[4,5], and our results may therefore more broadly reflect patternsat the individual species and community levels.ConclusionsIn conclusion, the present study demonstrates that both latticecount data and basal area data can be reliably used to simulate thespatial distribution of tree species. Neutral spatial effects, which arespecifically represented by the spatial autoregressive parametersand PCNM eigenfunctions, adequately explain the variations inboth count data and basal area data at the individual species andPLoS ONE | www.plosone.org 7 May 2012 | Volume 7 | Issue 5 | e38247392

Spatial Effects Shape Tree Species DistributionsFigure 6. Cell connectivity at each of the four scales of cell sizefor Sloanea tomentosa in DBH class 4.doi:10.1371/journal.pone.0038247.g006community levels. The community processes, especially distancelimiteddispersal process, may be the crucial mechanism underlyingclumped patterns of species distributions. We suggest grouping treesinto different DBH classes and analyzing their distributions atmultiple spatial scales to enhance the applicability of the results. Toachieve a broader understanding of the applicability of lattice countdata and basal area data in examining species spatial distributions atboth the individual species and community levels, further investigationsbased on large-scale plot data must be performed at additionaltropical, subtropical and temperate forest sites.Supporting InformationFigure S1 Patterns of median R-squared values from thefitted SAR models based on count data and basal areadata at four scales of cell size, controlling for DBH class.Circles and triangles connected by solid and dashed lines representcount data and basal area data, respectively. Bars indicatestandard deviations. Classes 0 to 4 are defined as in Figure 1.(TIF)Figure S2 Patterns of median R-squared values fromthe fitted SAR models based on count data and basalarea data for five DBH classes, controlling for scale.Circles and triangles connected by solid and dashed lines representcount data and basal area data, respectively. Classes 0 to 4 aredefined as in Figure 1.(TIF)Figure S3 Relationships between the R-squared valuesof the fitted SAR models and species total abundance foreach of the 5 DBH classes at the 10-m scale. Circles andtriangles represent count data and basal area data, respectively.Classes 0 to 4 are defined as in Figure 1.(TIF)Figure S4 Relationships between the R-squared valuesof the fitted SAR models and species total abundance foreach of the 5 DBH classes at the 25-m scale. Circles andtriangles represent count data and basal area data, respectively.Classes 0 to 4 are defined as in Figure 1.(TIF)Figure S5 Relationships between the R-squared valuesof the fitted SAR models and species total abundance foreach of the 5 DBH classes at the 50-m scale. Circles andtriangles represent count data and basal area data, respectively.Classes 0 to 4 are defined as in Figure 1.(TIF)Figure S6 Principal component analysis ordinations(based on matrices of transformed p-values from the SARmodels) of the 14 explanatory variables and the spatialautoregressive factor l for each of the 5 DBH classes at the10-m scale of the count data. Classes 0 to 4 are defined as inFigure 1. The abbreviations are defined as in Figure 3.(TIF)Figure S7 Principal component analysis ordinations(based on matrices of transformed p-values from the SARmodels) of the 14 explanatory variables and the spatialautoregressive factor l for each of the 5 DBH classes at the25-m scale of the count data. Classes 0 to 4 are defined as inFigure 1. The abbreviations are defined as in Figure 3.(TIF)Figure S8 Principal component analysis ordinations(based on matrices of transformed p-values from theSAR models) of the 14 explanatory variables and thespatial autoregressive factor l for each of the 5 DBHclasses at the 50-m scale of the count data. Classes 0 to 4are defined as in Figure 1. The abbreviations are defined as inFigure 3.(TIF)Figure S9 Principal component analysis ordinations(based on matrices of transformed p-values from the SARmodels) of the 14 explanatory variables and the spatialautoregressive factor l for each of the 5 DBH classes at the10-m scale of the basal area data. Classes 0 to 4 are defined asin Figure 1. The abbreviations are defined as in Figure 3.(TIF)Figure S10 Principal component analysis ordinations(based on matrices of transformed p-values from the SARmodels) of the 14 explanatory variables and the spatialautoregressive factor l for each of the 5 DBH classes at the25-m scale of the basal area data. Classes 0 to 4 are defined asin Figure 1. The abbreviations are defined as in Figure 3.(TIF)Figure S11 Principal component analysis ordinations(based on matrices of transformed p-values from the SARmodels) of the 14 explanatory variables and the spatialautoregressive factor l for each of the 5 DBH classes at the50-m scale of the basal area data. Classes 0 to 4 are defined asin Figure 1. The abbreviations are defined as in Figure 3.(TIF)PLoS ONE | www.plosone.org 8 May 2012 | Volume 7 | Issue 5 | e38247393

Spatial Effects Shape Tree Species DistributionsFigure S12 Patterns of total explained variation incommunity composition across life stages based oncount data and basal area data. The reduplicate data ateach DBH class consisted of the total explained variationsof the 4 scales of the variation partitioning results.Numerals 0 to 4 represent the five DBH classes whichare defined as in Figure 1.(TIF)Figure S13 The p-values of l and the R-squared valuesof the fitted SAR models for Sloanea tomentosa in DBHclass 4 at each of the four spatial scales.(TIF)References1. Beale CM, Lennon JJ, Yearsley JM, Brewer MJ, Elston DA (2010) Regressionanalysis of spatial data. Ecol Lett 13: 246–264.2. Legendre P, Gallagher ED (2001) Ecologically meaningful transformations forordination of species data. Oecologia 129: 271–280.3. Olden JD, Lawler JJ, Poff NL (2008) Machine learning methods without tears: aprimer for ecologists. Q Rev Biol 83: 171–193.4. Wang Z, Ye W, Cao H, Huang Z, Lian J, et al. (2009) Species-topographyassociation in a species-rich subtropical forest of China. Basic Appl Ecol 10:648–655.5. Legendre P, Mi XC, Ren HB, Ma KP, Yu MJ, et al. (2009) Partitioning betadiversity in a subtropical broad-leaved forest of China. Ecology 90: 663–674.6. Comita LS, Condit R, Hubbell SP (2007) Developmental changes in habitatassociations of tropical trees. J Ecol 95: 482–492.7. Lai J, Mi X, Ren H, Ma K (2009) Species-habitat associations change in asubtropical forest of China. J Veg Sci 20: 415–423.8. Morlon H, White EP, Etienne RS, Green JL, Ostling A, et al. (2009) Takingspecies abundance distributions beyond individuals. Ecol Lett 12: 488–501.9. Etienne RS, Olff H (2004) How dispersal limitation shapes species-body sizedistributions in local communities. Am Nat 163: 69–83.10. Paoli GD, Curran LM, Slik JWF (2008) Soil nutrients affect spatial patterns ofaboveground biomass and emergent tree density in southwestern Borneo.Oecologia 155: 287–299.11. Condit R, Ashton PS, Baker P, Bunyavejchewin S, Gunatilleke S, et al. (2000)Spatial patterns in the distribution of tropical tree species. Science 288:1414–1418.12. Legendre P (1993) Spatial autocorrelation: trouble or new paradigm? Ecology74: 1659–1673.13. Dormann CF, McPherson JM, Araujo MB, Bivand R, Bolliger J, et al. (2007)Methods to account for spatial autocorrelation in the analysis of speciesdistributional data: a review. Ecography 30: 609–628.14. Harms KE, Condit R, Hubbell SP, Foster RB (2001) Habitat associations oftrees and shrubs in a 50-ha neotropical forest plot. J Ecol 89: 947–959.15. John R, Dalling JW, Harms KE, Yavitt JB, Stallard RF, et al. (2007) Soilnutrients influence spatial distributions of tropical tree species. Proc Natl AcadSci USA 104: 864–869.16. Cottenie K (2005) Integrating environmental and spatial processes in ecologicalcommunity dynamics. Ecol Lett 8: 1175–1182.17. Leibold MA, McPeek MA (2006) Coexistence of the niche and neutralperspectives in community ecology. Ecology 87: 1399–1410.18. Shen G, Yu M, Hu XS, Mi X, Ren H, et al. (2009) Species-area relationshipsexplained by the joint effects of dispersal limitation and habitat heterogeneity.Ecology 90: 3033–3041.19. Werner E, Gilliam JF (1984) The ontogenetic niche and species interactions insize-structured populations. Annu Rev Ecol Syst 15: 393–425.20. Schupp EW (1995) Seed-seedling conflicts, habitat choice, and patterns of plantrecruitment. Am J Bot 82: 399–409.21. He F, Legendre P, LaFrankie JV (1997) Distribution patterns of tree species in aMalaysian tropical rain forest. J Veg Sci 8: 105–114.AcknowledgmentsWe are grateful to Prof. Fangliang He for his valuable comments on thepaper. We also thank the Biogeochemistry Laboratory of the XishuangbannaTropical Botanical Garden for the analysis of the soil nutrientconcentrations. We are also grateful to all of the field workers whocontributed to the tree species census of the 20-ha plot.Author ContributionsConceived and designed the experiments: Y-HH MC. Performed theexperiments: Y-HH G-YL L-QS. Analyzed the data: Y-HH MC.Contributed reagents/materials/analysis tools: Y-HH MC. Wrote thepaper: Y-HH G-YL L-QS MC YT Y-DL D-PX.22. Kanagaraj R, Wiegand T, Comita LS, Huth A (2011) Tropical tree speciesassemblages in topographical habitats change in time and with life stage. J Ecol99: 1441–1452.23. Hu YH, Sha LQ, Blanchet FG, Zhang JL, Tang Y, et al. (2011) Dominantspecies and dispersal limitation regulate tree species distributions in a 20-ha plotin Xishuangbanna, Southwest China. Oikos. In press.24. Dungan JL, Perry JN, Dale MRT, Legendre P, Citron-Pousty S, et al. (2002) Abalanced view of scale in spatial statistical analysis. Ecography 25: 626–640.25. He F, LaFrankie JV, Song B (2002) Scale dependence of tree abundance andrichness in a tropical rain forest, Malaysia. Landscape Ecol 17: 559–568.26. Cao M, Zhu H, Wang H, Lan GY, Hu YH, et al. (2008) Xishuangbannatropical seasonal rainforest dynamics plot: tree distribution maps, diametertables and species documentation. Kunming: Yunnan science and technologypress.27. Cao M, Zou XM, Warren M, Zhu H (2006) Tropical forests of Xishuangbanna,China. Biotropica 38: 306–309.28. Zhu H, Cao M, Hu HB (2006) Geological history, flora, and vegetation ofXishuangbanna, Southern Yunnan, China. Biotropica 38: 310–317.29. Liu GS, Jiang NH, Zhang LD (1996) Soil physical and chemical analysis &description of soil profiles; Liu GS, editor. Beijing: Standards Press of China, inChinese.30. Guisan A, Weiss SB, Weiss AD (1999) GLM versus CCA spatial modeling ofplant species distribution. Plant Ecol 143: 107–122.31. Wall MM (2004) A close look at the spatial structure implied by the CAR andSAR models. J Stat Plan Infer 121: 311–324.32. R Development Core Team (2011) R: A language and environment forstatistical computing. R Foundation for Statistical Computing, Vienna, Austria.33. Borcard D, Legendre P, Drapeau P (1992) Partialling out the spatial componentof ecological variation. Ecology 73: 1045–1055.34. Ficetola GF, Denoël M (2009) Ecological thresholds: an assessment of methodsto identify abrupt changes in species-habitat relationships. Ecography 32:1075–1084.35. Bell G (2001) Neutral macroecology. Science 293: 2413–2418.36. Hubbell SP, Foster RB, O’brien ST, Harms KE, Condit R, et al. (1999) Lightgapdisturbances, recruitment limitation, and tree diversity in a neotropicalforest. Science 283: 554–557.37. Hubbell SP (2001) The unified neutral theory of biodiversity and biogeography.Princeton: Princeton University Press.38. Seidler TG, Plotkin JB (2006) Seed dispersal and spatial pattern in tropical trees.Plos Biol 4: 2132–2137.39. Débarre F, Lenormand T (2011) Distance-limited dispersal promotes coexistenceat habitat boundaries: reconsidering the competitive exclusion principle.Ecol Lett 14: 260–266.40. Li L, Huang Z, Ye W, Cao H, Wei S, et al. (2009) Spatial distributions of treespecies in a subtropical forest of China. Oikos 118: 495–502.41. Kühn I (2007) Incorporating spatial autocorrelation may invert observedpatterns. Divers Distrib 13: 66–69.PLoS ONE | www.plosone.org 9 May 2012 | Volume 7 | Issue 5 | e38247394

J For Res (2012) 17:432–439DOI 10.1007/s10310-011-0309-yORIGINAL ARTICLETree species diversity of a 20-ha plot in a tropical seasonalrainforest in Xishuangbanna, southwest ChinaGuoyu Lan • Hua Zhu • Min CaoReceived: 20 March 2011 / Accepted: 18 May 2011 / Published online: 26 October 2011Ó The Japanese Forest Society and Springer 2011Abstract We censused all free-standing trees C1 cmdiameter at breast height (dbh) in a 20-ha plot establishedin a tropical seasonal rainforest in Xishuangbanna NationalNature Reserve, southwest China. A total of 95,834 freestandingtrees C1 cm dbh were recorded, and 95,498individuals (accounting for 99.65% of the total), including468 morphospecies in 213 genera and 70 families, wereidentified. Thirteen of 468 species (2.78%) had more than1,000 individual C1 cm dbh, which represented 56.36%individuals of the total. On the other hand, 230 of 468species (49.14%) had a mean density of B1 tree per ha, and69 of 468 species (14.74%) were singletons in the 20-haplot. The mean species richness, density and basal area perha were 216.50 species, 4,791.70 stems and 42.34 m 2 ,respectively. Pittosporopsis kerrii (20,918 stems, C1 cmdbh) of Icacinaceae and Parashorea chinensis (7,919stems, C1 cm dbh) of Dipterocarpaceae were the two mostabundant species dominating the emergent layer and treeletlayer, respectively. Compared with other 50-ha plotsestablished in other equatorial regions, tree species richnessper ha and tree abundance per ha of the plot were at themoderate level.Keywords Diversity Tropical seasonal rainforest XishuangbannaG. Lan H. Zhu M. Cao (&)Key Laboratory of Tropical Forest Ecology,Xishuangbanna Tropical Botanical Garden,Chinese Academy of Sciences, Kunming 650223, Chinae-mail: caom@xtbg.ac.cnG. LanRubber Research Institute, The Chinese Academy of TropicalAgricultural Sciences, Danzhou, Hainan 571737, ChinaIntroductionSince the late 1970s, the Center for Tropical Forest Science(CTFS) of the Smithsonian Tropical Research Institute hasestablished 40 large-sized (ranging from 16 to 52 ha) forestdynamics plots in the tropical forests of the five continents.To date, CTFS monitors more than 3 million individualtropical trees, representing about 8,500 tree species, nearly17% of the world’s entire tropical tree flora. However,there was no such tropical forest dynamics plot in themainland of China until 2007, although much of the oldgrowth high diversity lowland tropical rainforests in Chinaare located in the Xishuangbanna region of YunnanProvince (Cao et al. 2006). Due to its unique geographicallocation, Xishuangbanna is included in the Indo-Burmabiodiversity hotspots and contains over 5,000 species ofvascular plants, comprising 16% of China’s total plantdiversity (Cao et al. 2006). Naturally, the tropical rainforestof Xishuangbanna occurs at the limits in terms of latitudinaland altitudinal distribution of the southeast Asianrainforests (Wu 1987; Zhu et al. 1998). As a result of this,its flora composition is sensitive to climate change at alocal scale (Zhu 1993). In order to monitor long-termchanges in tree populations of tropical rainforest in thisregion and to test theories and hypotheses related to biodiversitymaintenance of tropical forests, a 20-ha dipterocarptropical seasonal rainforest dynamics plot wasestablished in Xishuangbanna Nature Reserve in 2007. Itsfield protocol standard followed the 50-ha plot located inBarro Colorado Island in Panama established by Hubbelland Foster (Condit 1995).This paper presented the results of the first census on thetree species diversity of the 20-ha plot in a tropical seasonalrainforest in Xishuangbanna, southwest China, trying toanswer following questions: (1) how many tree species are123395

J For Res (2012) 17:432–439 433Table 1 Temperature and precipitation distributions of the 20-ha plotof tropical seasonal rainforest in Xishuangbanna, southwest ChinaAMT(°C)MTH(°C)MTC(°C)AP(mm)PD(mm)PR(mm)21.0 24.6 15.2 1,531.9 281.6 1,250.3Dry season: November–April, Rainy season: May–OctoberAMT annual mean temperature, MTH mean temperature of the hottestmonth, MTC mean temperature of the coldest month, AP annualprecipitation, PD precipitation during dry season, PR precipitationduring rainy seasonthere in the 20-ha plot of tropical seasonal rainforest, andhow many are rare and how many are abundant, and(2) what are the similarities and differences betweenXishuangbanna tropical seasonal rainforest and othertropical forest dynamics plots in tree species diversity?Fig. 1 Location (star) of the 20-ha plot of tropical seasonal rainforestin Xishuangbanna, southwest ChinaMaterials and methodsStudy siteXishuangbanna is predominated by a typical monsoonclimate with an alternation between dry season and rainyseason. Taking Mengla County (14 km from the study sitewhere the 20-ha plot was established) as an example, theannual mean temperature is 21.0°C. Annual mean precipitationis 1,532 mm, of which about 80% occurs betweenMay and October (rainy season) (Table 1). The dry seasonis from November to April (Zhu 2006; Lan et al. 2009).Under such climatic conditions, tropical seasonal rainforestis developed in the lowland, valleys and hills with a goodwater supply (Wu 1987; Zhu 2006; Zhu et al. 2006). Thisforest type was believed to maintain the highest tree speciesin this region (Cao and Zhang 1997). The site for theestablishment of the 20-ha plot was chosen in a tract of thetropical seasonal rainforest at Bubeng village (101°34 0 26 00 –47 00 E and 21°36 0 42 00 –58 00 N), Mengla county, XishuangbannaNational Nature Reserve, southwest China (Fig. 1).It was dominated by Parashorea chinensis—a big staturetree species of Dipterocarpaceae.Data collectionA 20-ha plot (400 9 500 m 2 ) was established in a Parashoreachinensis forest. This plot ranges from 709 to869 m above sea level, indicating a heterogeneous habitat.It was sub-divided into 8,000 quadrats of 5 m 9 5 m. Alltrees C1 cm in diameter at breast height (dbh) in the 20-haplot were tagged with sequentially numbered aluminiumtags. Tree diameters were measured at 1.3 m from the ground(Condit et al. 1996; Ayyappan and Prthasarathy 1999).In the case of buttresses, the dbh was measured at thelowest point where the trunk was back to normal. Treeswith multiple stems were counted as a single individual,but each stem was also tagged and measured (Condit1998). All free-standing trees C1 cm dbh were identified tospecies. The nomenclature of the tree species followed theEnglish version of Flora of China, and the vouchers werestored at the herbarium of the Xishuangbanna TropicalBotanical Garden.Data analysisA six-capital letter code was assigned to the species with thefirst 4 letters denoting the generic epithet and the next 2letters the specific epithet. Species (genus and family)–areacurves were plotted for all trees (dbh C1 cm). Based on theirabundance, the tree species were grouped into 5 categories(Ayyappan and Prthasarathy 1999), viz.: (1) predominantspecies [those with abundance (A) C1,000 stems in the 20-haplot); (2) dominant species (A = 200–999), (3) commonspecies (A = 20–199), (4) rare species (A = 2–19), and(5) very rare species (A = 1, also singleton species].Fisher’s a was calculated for trees of C1, C10 andC30 cm dbh. a was defined according to Fisher et al.(1943):S ¼ a lnð1 þ N=aÞð1Þwhere a was the diversity index, N was the number of trees,S was the number of species.Relative density (FD), relative dominance (RA, usingbasal area) and relative frequency (RF) were calculated foreach species in order to estimate the importance value (IV).Importance value was defined as (Curtis and Mcintosh 1950,1951; Greig-Smith 1983; Linares-Palomino and Alvarez2005):123396

434 J For Res (2012) 17:432–439Table 2 Summary of theinventories of the 20-ha plot oftropical seasonal rainforest inXishuangbanna, southwestChina (336 unidentifiedindividual trees were exclued incalculating the number ofspecies, genera and families, butwere included when calculatingindividuals and basal area)dbh diameter at breast heightContentsNo. ofspeciesNo. ofgeneraNo. offamiliesNo. ofstemsBasalarea (m 2 )Fisher’s adbh C1 cmNo. in 20-ha plot 468 213 70 95,834 846.86 63.99Mean (ha -1 ) 216.50 131.80 53.60 4,791.70 42.34 46.64Range (ha -1 ) 187–239 116–144 49–59 3,160–6,181 29.04–55.16 41.03–55.23dbh C10 cmNo. in 20-ha plot 339 171 63 12,331 733.36 64.46Mean (ha -1 ) 123.50 84.80 43.50 616.60 36.66 46.54Range (ha -1 ) 110–139 68–91 39–48 452–810 23.85–51.04 32.62–54.94dbh C20 cmNo. in 20-ha plot 227 146 58 4,644 598.72 61.32Mean (ha -1 ) 74.3 56.05 32.3 232.2 29.93 33.65Range (ha -1 ) 64–96 49–66 28–41 168–338 20.03–44.07 18.35–50.12dbh C30 cmNo. in 20-ha plot 215 119 51 2,232 496.41 58.67Mean (ha -1 ) 42.70 35.00 23.70 111.60 24.82 25.28Range (ha -1 ) 22–64 17–50 14–35 70–159 17.07–41.13 8.87–46.38Importance value : IV j ¼ RF j þ RD j þ RA jRelative frequency : RF j ¼ 100 F j = X jRelative density : FD j ¼ 100 D j = X jRelative dominance : FA j ¼ 100 A j = X jD jF jA jð2Þð3Þð4Þð5Þwhere F j was the number of 1-ha subplots containingspecies j; D j the number of individuals of species j; A j wasthe total basal area of species j. For trees with multiplestems, basal areas of multiple stems and main stem werecalculated and summed as the basal area of a singleindividual.We classified the species into five strata according totheir estimated maximum height (Kenfack et al. 2007):treelets include all species with adults generally \10 mtall; understory trees are those with adults 10–20 m tall;lower canopy species have heights of 20–30 m; and uppercanopy species are those often 30–45 m in height andemergent above the main canopy ([45 m). Information onthe heights of the tree species was from Zhu (2006).ResultsSpecies richnessA total of 95,834 trees C1 cm dbh was enumerated in the20-ha plot (Table 2). Ultimately, 95,498 individuals(99.65%) were identified to species (93,410 individuals,97.47%), genera (2,079 individuals, 2.17%) and familylevels (9 individuals, 0.01%). Among the 95,498 stems,there were 468 morphospecies in 213 genera and 70 families.In addition, 336 individuals (0.35% of the total) havenot yet been assigned a morphospecies. Fisher’s a of treesC10 cm dbh in the whole 20-ha plot was slightly greaterthan that of trees C1 cm dbh indicating that species weredistributed more evenly among trees C10 cm dbh, becauseevenness among species would lead to higher diversity. Butamong trees C30 cm dbh, Fisher’s a of 20 1-ha subplotsvaried from 8.87 to 46.38 revealing that tree species wereunevenly distributed across the plot.Species–area curves and species–abundance curvesSpecies–area curves for all trees with dbh C1, C10 andC30 cm in 20-ha plot of tropical seasonal rainforest wereplotted (Fig. 2). In the range of 0–5 ha, the number ofspecies increased rapidly with increasing area, whereas theincreasing speed slowed down in the range of 5–10 ha.Approximately 80% of total species were included in thefirst 6 1-ha plots, 90% in the first 11 1-ha plots, and 99% inthe first 18 1-ha plots. Species–area curves for dbh C10 andC30 cm were quite similar to those of dbh C1 cm, roughlyparallel to one another.We compared the plot of actual number of speciesagainst the number of trees with that predicted by theequation S = a ln(1 ? N/a). There is little differencebetween the observed number of species and that of predictedfor tress with dbh C1 and C10 cm (Fig. 3). However,for trees of dbh C30 cm, the observed number wasmuch smaller than that of predicted. This may indicate that123397

J For Res (2012) 17:432–439 435Number of species0 100 200 300 400 500dbh ≥1cmdbh≥10cmdbh ≥30cmln (abundance)0 2 4 6 8 10Fisher’s a can only be used to measure diversity of smallsize of trees.Species abundance0 5 10 15 20Area (ha)Fig. 2 Species–area curves for three tree sizes in the 20-ha plot oftropical seasonal rainforest in Xishuangbanna, southwest China (opencircle for dbh C1 cm, open triangle for dbh C 10 cm, plus symbol fordbh C30 cm). dbh diameter at breast heightNumber of species0 100 200 300 400(a)0 50 100 150 200 250 300 350(b)(c)ObservedPredicted0 500 1000 1500 20000 2000 4000 6000 8000 100000 20000 40000 60000 80000Number of treesFig. 3 Comparing the plot of the observed number of species againstthe number of trees with that predicted by the equation[S = a 9 ln(1 ? N/a), where a is the whole-plot, a for the appropriatesize category; a diameter at breast height (dbh) C1 cm, b dbhC10 cm, c dbh C30 cm]There were a few dominant species ([1,000 individuals;Table 2) in the 20-ha plot (also see Fig. 4). Pittosporopsiskerrii, the most abundant species in the study site, had20,918 stems presenting 21.90% individuals of the total,0 50 100 150 200while P. chinensis was next, and had 7,919 individualsaccounting 8.29% (Table 3). At the other extreme, 230 of468 species (49.14%) had a mean density of B1 tree per ha.The ‘‘long tail’’ of the species sequence curve indicated thevery rare species (also singletons) in the plot (Fig. 4).Species composition0 100 200 300 400Species sequenceFig. 4 Species sequence curve of the 20-ha plot of tropical seasonalrainforest in Xishuangbanna, southwest ChinaThe Lauraceae is the richest family with 52 species in 11genera (Table 4). Icacinaceae is the most abundant familywith 21,769 stems (Table 5), but Dipterocarpaceae has thelargest basal area of 113.52 m 2 . On the species level, Pittosporopsiskerrii is the most abundant species with 20,918individuals, and its importance value ranked as first(Table 6). Parashorea chinensis has the largest basal area,but it has fewer individuals (7,919 individuals) thanPittosporopsis kerrii. The number of species in the tenrichest families and genera is shown in Table 4. Thenumber of species and basal area of top ten commonestfamilies, genera and species is shown in Table 5.The forest profile of the plot could be divided into fivetree layers. The emergent ([45 m) layer was dominatedsolely by P. chinensis of Dipterocarpaceae (Table 6).P. chinensis had the largest basal area with 5.68 m 2 per ha.The upper canopy (30–45 m) was dominated by Sloaneatomentosa, Pometia tomentosa, Semecarpus reticulata,and Barringtonia pendula (Table 6). The lower canopy(20–30 m) was mainly composed of Garcinia cowa, Knemafurfuracea, Nephelium chryseum, Cinnamomum bejolghota,Diospyros hasseltii, Ficus langkokensis and Pseuduvariaindochinensis. Baccaurea ramiflora and Dichapetalum gelonioidesdominated the understory (10–20 m). The treeletlayer (5–10 m) in the plot was mainly composed of P. kerriiand other representative species such as Mezzettiopsis123398

436 J For Res (2012) 17:432–439Table 3 Abundanceclassification in the 20-ha plotof tropical seasonal rainforest inXishuangbanna, southwestChina (336 unidentifiedindividual trees were excludedin calculating the number ofspecies, but were included whencalculating individuals)Type No. of species (%) No. of individuals (%)Predominant species (abundance (A) [1,000) 13 (2.78) 53,994 (56.34)Dominant species (200 B A \ 1,000) 64 (13.68) 27,503 (28.70)Common species (20 B A \ 200) 161 (34.40) 12,816 (13.37)Rare species (2 B A \ 20) 161 (34.40) 1,116 (1.17)Very rare species (A = 1) 69 (14.74) 69 (0.07)Total 468 (100.00) 95,498 (99.95)Table 4 Number of species inthe ten richest genera and theten richest families in the 20-haplot of tropical seasonalrainforest in Xishuangbanna,southwest ChinaRank The ten most diverse families The ten most diverse generaFamily No. of species No. of genus Genus No. of species1 Lauraceae 52 11 Ficus 222 Euphorbiaceae 38 19 Elaeocarpus 163 Moraceae 30 5 Litsea 144 Rubiaceae 28 19 Syzygium 135 Meliaceae 25 11 Beilschmiedia 96 Leguminosae 19 8 Castanopsis 97 Elaeocarpaceae 17 2 Cinnamomum 88 Annonaceae 15 7 Mallotus 79 Fagaceae 14 2 Phoebe 710 Myrtaceae 14 2 Albizia 5Total 252 86 110creaghii, Saprosma ternata, Leea compactiflora, Phoebelanceolata and Syzygium latilimbum.DiscussionThere was an obvious inflexion at the 5th hectare in thespecies–area curve for dbh C1 cm, and 363 species representingabout 78% of the total species were included inthe first 5 ha. Total of 99% species were included in 18thhectare. And in the last 2 ha, only 1% species were added,indicating that the plot is large enough to represent tropicalseasonal rainforest in Xishuangbanna, southwestChina.In comparison with two other large-sized plots establishedin China, the tree species richness in the Xishuangbanna’s20-ha plot (Xishuangbanna plot, 468 species) wasover two times larger than in the Dinghushan (23°09 0 21 00 –23°11 0 30 00 N, 112°30 0 39 00 –112°33 0 41 00 N) 20-ha plot in subtropicalevergreen broadleaved forest (210 species), andnearly three times as rich as that in the Gutianshan(29°15.102 0 –29°15.344 0 N, 118°07.010 0 –118°07.400 0 E) 24-haplot in mid-subtropical evergreen broadleaved forest (156species) (Lan et al. 2008; Ye et al. 2008; Zhu et al. 2008).Furthermore, species richness per ha of the present plot(216.5 species) was high as compared with some other50-ha plots (for trees C1 cm dbh) in Barro Colorado Island(BCI plot, 168 species per ha), Panama, and in HuaiKha Khaeng, Thailand (HKK plot, 104 species per ha)(Plotkin et al. 2000; Condit et al. 2005) (Table 4), whereasspecies richness per ha in our plot was relatively the sameas that of the Korup 50-plot (235 species per ha), Cameroon(Condit et al. 2005). However, the Xishuangbanna plotshowed lower tree species richness per ha in comparison tothose of plots in Asian equatorial tropical rainforests. Forexample, there were 497 tree species per ha with dbhC1 cm in the Pasoh plot (nearly two and half time as richas that in the Xishuangbanna plot), and 618 species per hain the Lambir plot (nearly three times as rich as that in theXishuangbanna plot), Malaysia (Condit et al. 2005)(Table 7).Compared to the BCI and HKK plots, the Xishuangbannaplot is located at a higher latitude, has a longer dryseason and receives less precipitation, but the tree diversityis higher than those of the other two plots. One of the mostimportant reasons is the dense fog which always existsduring the entire dry season on the lower hills and in thevalleys, averaging 146 foggy days per year and 1 mm123399

J For Res (2012) 17:432–439 437Table 5 Number of stems and basal area of the ten commonest families, genera and species in the 20-ha plot of tropical seasonal rainforest in Xishuangbanna, southwest ChinaRank Stems Basal area (m 2 )Species BasalareaGenus BasalareaFamily Stems Genus Stems Species Stems Family Basalarea1 Icacinaceae 21,769 Pittosporopsis 20,918 Pittosporopsis kerrii 20,918 Dipterocarpaceae 113.52 Parashorea 113.52 Parashorea chinensis 113.522 Euphorbiaceae 9,827 Parashorea 7,919 Parashorea chinensis 7,919 Fagaceae 106.33 Castanopsis 102.34 Castanopsis echinocarpa 48.603 Dipterocarpaceae 7,919 Garcinia 5,131 Garcinia cowa 4,333 Lauraceae 72.36 Ficus 42.07 Sloanea tomentosa 41.374 Lauraceae 7,302 Knema 3,778 Mezzettiopsis creaghii 3,300 Euphorbiaceae 59.27 Sloanea 41.37 Pittosporopsis kerrii 28.475 Guttiferae 5,150 Mezzettiopsis 3,300 Baccaurea ramiflora 3,212 Elaeocarpaceae 57.00 Pittosporopsis 28.47 Mezzettiopsis creaghii 25.286 Annonaceae 5,010 Baccaurea 3,212 Knema furfuracea 3,160 Moraceae 45.70 Mezzettiopsis 25.28 Pometia tomentosa 23.587 Rubiaceae 4,869 Ficus 3,088 Saprosma ternata 2,698 Sapindaceae 41.11 Pometia 23.58 Garcinia cowa 19.248 Myristicaceae 4,272 Phoebe 3,012 Phoebe lanceolata 2,409 Annonaceae 34.79 Garcinia 21.31 Castanopsis hystrix 18.909 Moraceae 3,315 Saprosma 2,698 Cinnamomum bejolghota 1,337 Icacinaceae 31.75 Litsea 18.05 Castanopsis megaphylla 18.2410 Meliaceae 2,990 Castanopsis 1,885 Ficus langkokensis 1,337 Meliaceae 27.59 Cinnamomum 15.93 Alseodaphne petiolaris 14.44Total 72,423 54,941 50,623 589.43 329.58 351.64precipitation per foggy day (Zhu 2006). This compensatesfor the insufficient precipitation so that a tropical moistclimate can form locally in spite of its relatively low meanannual precipitation.Stem density of Xishuangbanna plot (4,791.7 treesC1 cm dbh) was moderate compared with other 50-haplots, which ranged from 1,609 stems per ha in Huai KhaKhaeng, Thailand to 6,769 stems per ha in Pasoh,Malaysia (Bunyavejchewin et al. 2001; Kenfack et al.2007), but similar to that of the plot on Barro ColoradoIsland, Panama (4,844 stems per ha) (Bunyavejchewinet al. 2001).Hubbell and Foster (1986) defined rare species as thosewith a mean density of B1 tree per ha. According to thisdefinition, the Xishuangbanna plot had a large proportionof rare species (230 rare species representing 49.14% oftotal species but only 1.24% of total individuals). For thesake of comparison, we defined rare species here as havingfewer than 0.3 individuals per ha. In this case, the percentageof rare species (35.9%) of our plot was greater thanthose of the Pasoh plot (19.2%), Lambir plot (14.9%), BCIplot (25.6%) and Korup plot (29.2%) (Condit et al. 2005;Kenfack et al. 2007; Lan et al. 2008), but lower than thoseof the Ituri plot (48.40%) in Congo and the Huai KhaKhaeng plot (44.8%) in Thailand (Condit et al. 2005). Asregards species dominance, the Xishuangbanna plot alsohad the most abundant species, P. kerrii with a dominanceof 21.90%, which was much greater than those of the Pasohplot (2.70%), Lambir plot (2.60) and BCI plot (15.70%)(Condit et al. 2005). Furthermore, the emergent layer of theforest in our study site was dominated solely by P. chinensis,which is unusual for tropical rainforests in SoutheastAsia.ConclusionsOur study presents tree species diversity and flora compositionof a 20-ha plot in a tropical seasonal rainforest inXishuangbanna, southwest China. A total of 468 morphospecies,contributing 213 genera and 70 families, wasrecorded in the plot. Fisher’s a showed species among treesC10 dbh were distributed more evenly than were speciesamong trees C1 and C30 cm dbh. Fewer predominantspecies ([1,000 individuals) but relatively more very rarespecies (singletons in the 20-ha plot) were found in theplot. Monodominance, both in the emergent layer andtreelet layer, and a high percentage of rare species of thetropical seasonal rainforest in Xishuangbanna were unusualamong tropical rainforests in south-east Asia. Speciesrichness per ha and tree abundance per ha varied greatlyacross the plot. Tree species richness per ha of the plotwas relatively low when compared with the equatorial123400

438 J For Res (2012) 17:432–439Table 6 List of the top 20species with the greatestimportance values in the 20-haplot of tropical seasonalrainforest in Xishuangbanna,southwest China (336unidentified individual treeswere excluded in calculating theimportance value)IV importance value, treelet5–10 m, understory 10–20 m,lower canopy 20–30 m, uppercanopy 30–45 m, emergent[45 mRank Species Strata Abundance Stems per ha Basal area (m 2 ) IV1 Pittosporopsis kerrii Treelet layer 20,918 1,045.90 28.47 25.782 Parashorea chinensis Emergent 7,919 395.95 113.52 22.363 Garcinia cowa Lower canopy 4,333 216.65 19.24 7.304 Castanopsis echinocarpa Lower canopy 881 44.05 48.60 7.075 Mezzettiopsis creaghii Treelet layer 3,300 165.00 25.28 6.926 Sloanea tomentosa Upper canopy 502 25.10 41.37 5.907 Baccaurea ramiflora Understory 3,212 160.60 14.01 5.508 Knema furfuracea Lower canopy 3,160 158.00 11.24 5.129 Pometia tomentosa Upper canopy 480 24.00 23.58 3.7910 Phoebe lanceolata Treelet layer 2,409 120.45 4.43 3.5111 Saprosma ternata Treelet layer 2,698 134.90 1.01 3.4112 Nephelium chryseum Lower canopy 1,098 54.90 12.97 3.1613 Castanopsis hystrix Lower canopy 244 12.20 18.90 2.9314 Cinnamomum bejolghota Lower canopy 1,337 66.85 8.76 2.9115 Castanopsis megaphylla Lower canopy 255 12.75 18.24 2.8016 Diospyros hasseltii Lower canopy 815 40.75 12.47 2.7817 Ficus langkokensis Lower canopy 1,337 66.85 7.64 2.7818 Semecarpus reticulata Upper canopy 619 30.95 10.40 2.3519 Alseodaphne petiolaris Lower canopy 178 8.90 14.44 2.3320 Castanopsis indica Lower canopy 351 17.55 10.97 2.10Total 56,046 2,802.30 445.54 120.80Table 7 Comparison of species richness and density between Xishuangbanna forest dynamics plot and other forest dynamics plotsPlotsPlot size(ha)Precipitation(mm)Dry season(month)Elevation(m)Latitude(north)Stems per ha(C1 cm)Species per ha(C1 cm)Species in thewhole plot(C1 cm)Xishuangbanna, China 20 1,532 6 709–869 21.364° 4,792 216.50 468Lambir, Malaysia 52 2,664 0 104–244 4.187° 6,687 b 618.10 c 1,179 cHuai Kha Khaeng, Thailand 50 1,476 6 550–640 15.632° 1,609 b 103.90 c 251 aPasoh, Malaysia 50 1,788 0 70–90 2.982° 6,769 b 496.50 c 814 cKorup, Cameroon 50 5,272 3 150–240 5.065° 6,590 d 235.10 c 494 cBarro Colorado Island, Panama 50 2,551 3 110–140 9.152° 4,844 b 168.00 c 301 ca Plotkin et al. (2000)b Bunyavejchewin et al. (2001)c Condit et al. (2005)d Kenfack et al. (2007)rainforests of tropical Asia, but greater than tropical forestin BCI, Panama, and seasonal dry evergreen forest inThailand.Acknowledgments This project is supported by the KnowledgeInnovation Project of Chinese Academy of Sciences (KZCW2-YW-430) and the National Natural Science Foundation of China(30570128). The authors thank Dr. Fangliang He, Dr. I-Fang Sun andDr. Keping Ma for their assistance in data analysis and for valuablecomments on this manuscript.ReferencesAyyappan N, Prthasarathy N (1999) Biodiversity inventory of tree ina large-scale permanent plot of tropical evergreen forestat Varagalaiar, Anamalais, Western Ghats, India. BiodiversConserv 8:1533–1554Bunyavejchewin S, Baker PJ, LaFrankie JV, Ashton PS (2001) Standstructure of a seasonal dry evergreen forest at Huai Kha KhaengWildlife Sanctuary, western Thailand. Nat Hist Bull Siam Soc49:89–106123401

J For Res (2012) 17:432–439 439Cao M, Zhang JH (1997) Tree species diversity of tropical forestvegetation in Xishuangbanna, SW China. Biodivers Conserv6:995–1006Cao M, Zou XM, Warren M, Zhu H (2006) Tropical forests ofXishuangbanna, China. Biotropica 38:306–309Condit R (1995) Research in large, long-term tropical forest plot.Trends Ecol Evol 10:18–22Condit R (1998) Tropical forest census plots. Springer, New YorkCondit R, Hubbell SP, Foster RB (1996) Changes in tree speciesabundance in a Neotropical forest: impact of climate change.J Trop Ecol 12:231–256Condit R, Ashton P, Baslev H, Brokaw N, Bunyavejchewin S,Chuyong G, Co L, Dattaraja HS, Davies S, Esufali S, EwangoCEN, Foster R, Gunatileke N, Gunatileke S, Hernandez C,Hubbell S, John R, Kenfack D, Kirakiprayoon S, Hall P, Hart T,Itoh A, Lafrankie J, Liengola I, Lagunzad D, Lao S, Losos E,Magard E, Makana J, Manokaran N, Navarette H, MohammedNur S, Okhubo T, Pérez R, Smaper C, Hua Seng L, Sukumar R,Svenning JC, Tan S, Thomas D, Thomson J, Valleho M, VillaMunoz G, Valencia R, Yamakura T, Zimmerman J (2005)Tropical tree alpha-diversity: results from a worldwide networkof large plots. Biol Skr 55:565–582Curtis JT, Mcintosh RP (1950) The interrelations of certain analyticand synthetic phytosociological characters. Ecology 31:434–455Curtis JT, Mcintosh RP (1951) An upland forest continuum in theprairie-forest border region of Wisconsin. Ecology 32:476–496Fisher RA, Corbet AS, Williams CB (1943) The relation between thenumber of species and the number of individuals in randomsample of an animal population. J Anim Ecol 12:42–58Greig-Smith P (1983) Quantitative plant ecology. Blackwell, LondonHubbell SP, Foster RB (1986) Commonness and rarity in aNeotropical forest: implications for tropical tree conservation.In: Soulé M (ed) Conservation biology: science of scarcity anddiversity. Sinauer, Sunderland, pp 205–231Kenfack D, Thomas DW, Chuyong G, Condit R (2007) Rarity andabundance in a diverse African forest. Biodivers Conserv16:2045–2074Lan GY, Hu YH, Cao M, Zhu H, Wang H, Zhou SH, Deng XB, CuiJY, Huang JG, Liu LY, Xu HL, Song JP, He YC (2008)Establishment of Xishuangbanna tropical forest dynamics plot:species compositions and spatial distribution patterns. J PlantEcol 32:287–298 (in Chinese with English summary)Lan GY, Zhu H, Cao M, Hu YH, Wang H, Deng XB, Zhou SS, CuiJY, Huang JG, He YC, Liu LY, Xu HL, Song JP (2009) Spatialdispersion patterns of trees in a tropical rainforest in Xishuangbanna,southwest China. Ecol Res 24:1117–1124Linares-Palomino R, Alvarez SIP (2005) Tree community patterns inseasonally dry tropical forests in the Cerros de AmotapeCordillera, Tumbes, Peru. For Ecol Manag 209:261–272Plotkin JB, Potts MD, Lesliea N, Manokarane N, Lafrankieb J,Ashton PS (2000) Species–area curves, spatial aggregation, andhabitat specialization in tropical forests. J Theor Biol 207:81–99Wu ZY (1987) Vegetation of Yunnan. Science Press, Beijing (inChinese)Ye WH, Cao HL, Huang ZL, Lian JY, Wang ZG, Li L, Wei SG, WangZM (2008) Community structure of 20 hm 2 lower subtropicalevergreen broadleaved forest plot in Dinghushan, China. J PlantEcol 32:274–286 (in Chinese with English summary)Zhu H (1993) The floristic characteristics of the tropical rainforest inXishuangbanna. Trop Geogr 13:149–155Zhu H (2006) Forest vegetation of Xishuangbanna, south China. ForStud China 8(2):1–27Zhu H, Wang H, Li BG, Xu ZF (1998) Research on the tropicalseasonal rainforest of Xishuangbanna, south Yunnan. Guihaia18:371–384 (in Chinese with English summary)Zhu H, Cao M, Hu HB (2006) Geological history, flora, andvegetation of Xishuangbanna, Southern Yunnan, China. Biotropica38:310–317Zhu Y, Zhao GF, Zhang LW, Shen GC, Mi XC, Ren HB, Yu MJ,Chen JH, Chen SW, Fang T, Ma KP (2008) Communitycomposition and structure of Gutianshan forest dynamics plot ina mid-subtropical evergreen broadleaved forest, east China.J Plant Ecol 32:262–273 (in Chinese with English summary)123402

ReviewEcologyFebruary 2012 Vol.57 No.4: 307312doi: 10.1007/s11434-011-4690-xSPECIAL TOPICS:Lianas as structural parasites: A re-evaluationTANG Yong 1,2* , Roger L. KITCHING 2 & CAO Min 11Key Laboratory of Tropical Forest Ecology, Xishuangbanna Tropical Botanical Garden, Chinese Academy of Sciences, Mengla 666303, China;2Griffith School of the Environment, Griffith University, Nathan, Queensland 4111, AustraliaReceived April 28, 2011; accepted July 7, 2011; published online December 2, 2011Lianas are a principal physiognomic component of tropical and subtropical forests and are typically considered to be parasites oftrees. In contrast, the substantial contribution of lianas to rainforest leaf litter production (up to 40%) suggests that they play importantroles in nutrient cycles and may benefit their host trees. Lianas contribute disproportionately to total forest litter productionat least partially because lianas invest relatively little in support structures and proportionately much more to leaf productionwhen compared with trees. In contrast to tree leaves, liana leaves are higher in nutrient concentrations, relatively short-lived, anddecompose more rapidly. In addition, the special life form of lianas allows them to grow vertically and horizontally in the forestand relocate nutrients, mainly towards their host trees, through the production of leaf litter. Consequently, lianas may contributesubstantially to the high rainforest productivity, and the roles they play in liana/tree associations and rainforest dynamics needs tobe re-evaluated.Liana, compensatory effect, nutrient dynamics, leaf litter, rainforestCitation:Tang Y, Kitching R L, Cao M. Lianas as structural parasites: A re-evaluation. Chin Sci Bull, 2012, 57: 307312, doi: 10.1007/s11434-011-4690-xLianas are woody vines that climb other plants to ascendforest canopies [1]. The non-self-supporting life form oflianas has been considered as an evolutionary adaptationdriven by competition for light and allows lianas to investmore in transportation and photosynthetic organs, principallyleaves [2]. Lianas occur in almost all forest types butare more abundant in tropical forests [1]. The number ofspecies and the abundance of lianas decrease with an increasein latitude and show a sharp drop near the Tropic ofCancer where the climate changes northwards from tropicalto temperate. The low occurrence of lianas in temperateregions has been ascribed to frequent low winter temperaturesthat may cause freezing-induced embolisms in lianas’long vessel systems and lead to unrecoverable damage [3].As a principal physiognomic component of tropical andsubtropical rainforests, lianas comprise around 25% of thediversity of woody plants in these ecosystems [1,4,5]. Theirabundance and diversity increase along a low to high rainfallgradient, and peak in tropical rainforest sites near the*Corresponding author (email: tangy@xtbg.ac.cn)Equator, such as the Amazonian lowland rainforest, cooccurringwith the highest tree species diversity [1,6]. Additionalfactors, including soil type, seasonality of rainfall anddisturbance, have been proposed as being important in determininglocal liana distribution [6–8]. The high lianaabundance in seasonal rainforests reflects their ability tomaintain growth during dry seasons by accessing groundwater via deep roots and efficient vascular conducting systems[6]. This ability to obtain water during dry periodsallows lianas to take advantage of the high solar radiationenvironment associated with dry seasons. Thus, lianas growmuch faster than trees during the dry season in seasonalrainforests [6].Lianas play important roles in forest dynamics. They addsubstantially to baseline levels of plant diversity in forestsand maintain tree diversity through their role in gap dynamics[9]. A high liana load in tree canopies may cause hightree-fall rates, thus maintaining rainforests in a perpetualdisclimax and thereby potentially maintaining diversity byreducing dominance levels among tree species [10]. Manystudies have found that lianas increase in both abundance© The Author(s) 2011. This article is published with open access at Springerlink.com csb.scichina.com www.springer.com/scp403

308 Tang Y, et al. Chin Sci Bull February (2012) Vol.57 No.4and diversity following natural and anthropogenic disturbance[8,11,12]. Consequently, they play an important role inregulating forest regeneration processes [12]. The increasein the abundance of lianas reflects their ability to regeneratevegetatively and grow both vertically and horizontally inresponse to changes in the light environment. Thus, as rainforestfragmentation increases across the tropics, an increasein the dominance of lianas in rainforests is expected [8,11].Indeed, this effect may be enhanced through a positivefeedback loop driven by increases in atmospheric carbondioxide [13,14]. Finally, lianas, perhaps through their lowerlevels of chemical foliar defenses, are the preferred hosts ofmany herbivores, particularly insects, and liana occurrencein forests may substantially increase animal diversitythrough food-web effects in canopies and leaf litter [15,16].Lianas have long been considered as parasites of trees[11,17]. This hypothesis is based on the proposition thatthey have detrimental effects on the growth and reproductionof their host trees by causing physical damage and bycompeting for limited resources such as water, nutrients andlight [6,18,19]. Lianas may also significantly delay rainforestregeneration during the gap phase by competing withtrees, especially those of shade-tolerant species [20]. Themajority of studies on lianas, especially those conducted ina silvicultural context, have only focused on negative effectson their host trees [21]. However, many liana-bearingtrees are actually large canopy trees that live for many yearsand readily grow and produce seeds [22]. Presently, thereare still no explanations for how liana-bearing trees are ableto maintain their growth in highly competitive rainforestwhile incurring negative impacts from lianas. Although theimportance of lianas in forests has been widely accepted[23], the possibility that trees might benefit from the presenceof lianas has been overlooked. By reviewing publishedresearch, we demonstrate that lianas may play importantroles in nutrient dynamics by producing a large amount ofnutritious and easily decomposed leaf litter and can benefittheir host trees via nutrient transportation.1 Characteristics of liana leavesLianas produce a larger quantity of leaves than trees, andtheir leaves are structurally and nutritionally different fromthose of trees (Table 1). These differences have profoundimpacts on the litter decomposition process in rainforests.Liana leaves have significantly lower mass per unit leaf area(LMA) than tree leaves, because of the faster growth anddifferent resource allocation strategies of lianas [24–26].Furthermore, liana leaves have a much higher nitrogenconcentration than those of their host trees (Table 1). Althoughmost studies compared only green leaves, we expectthat leaf litter follows the same pattern because the nutrientcontent of green and senesced leaves are highly correlated[27]. Low LMA and high leaf nitrogen concentration havebeen suggested to be economical trade-offs [28] and maygenerate faster rates of litter decomposition than leaves withhigh LMAs and low nitrogen concentrations. Moreover,leaves with low LMA and high nitrogen contents, such asthose of lianas, tend to have a shorter life span than those ofmany trees [29–31], thereby maintaining a rapid nutrientcycling system.Liana leaves also have a high concentration of phosphorus[31–33], a major plant growth element generally limitedin many rainforests [34]. A recent study has shown that lianashave a lower phosphorus resorption rate from senescingleaves than do trees [32], which may be caused by the highnitrogen content [30] and the short lifespan of liana leaves.Thus, we expect that the leaf litter of lianas is higher inphosphorus concentration than that of trees. Given the pervasivephosphorus deficiency in many rainforests, lianaspotentially play an important role in phosphorus recyclingin rainforests by producing leaf litter with a high concentrationof this element [25,32,35]. Moreover, phosphorusavailability also has been considered to be an importantfactor regulating microbial decomposition processes inrainforests [36–38]. Leaf litter with a high phosphorus concentrationmay potentially promote microbial processes andthus make more of the element available to plants throughmicrobial phosphorus immobilization [36]. That is, phosphorusis taken up by microbes and then slowly releasedback into the system as they die, thereby preventing rapidloss of phosphorus.The microenvironment created by liana leaf litter couldpromote the decomposition of associated tree leaf litter, thusproviding an efficient nutrient recycling system. Plants andsoils nutrients interact in a manner such that correlationsbetween them may be caused by nutrient limitations to plantgrowth or by plant effects on soils [39,40]. The unresolvedassociation of lianas with richer soils [1,4], especially thoseTable 1Comparison of leaf nutrient contents of lianas and treesForest typeLianaTreeReferenceN (mg g –1 ) P (mg g –1 ) LMA (g m –2 ) N (mg g –1 ) P (mg g –1 ) LMA (g m –2 )Mountain rainforest 15.3 0.7 94.9 11.7 0.4 170.5 [25]Lowland rainforest 24.1 – 68.2 18.4 – 97.9 [24]Tropical mountain rainforest 24.6 1.24 101 22.8 1.21 115 [32]Tropical seasonal rainforest 28.79 1.58 49.1 22.22 1.12 63.2 [31]404

Tang Y, et al. Chin Sci Bull February (2012) Vol.57 No.4 309with high phosphorus concentrations [8,41], may be becauseof the high nutrient content of liana leaf litter and itshigh rate of decomposition, and not the inherent nutrientrichsoil per se.2 Leaf litter production by lianasLeaf litter production is an important part of forest nutrientcycling [42,43], and liana litter production is distinct fromthose of trees owing to their different life strategies [44].Lianas seldom exceed 5% of total forest biomass [5].However, their level of leaf production is relatively muchhigher than that of trees, contributing up to 36% of totalabove-ground leaf biomass (Table 2) and as much as 40%of total leaf area in tropical rainforests [5,45]. Comparedwith trees with the same diameter at breast height, lianasproduce much more leaf dry mass because, unlike trees,they do not invest heavily in mechanical support organs(stems, branches and buttresses). Accordingly, a largerproportion of their resources can be directed to the productionof leaves [44,46], and up to 16%–40% of all leaflitter within rainforests is derived from lianas (Table 2).Thus, liana leaves have a disproportionate importancein any consideration of nutrient dynamics, especially thosein the nutrient-poor rainforests [14,47,48]. Moreover, theproportion of liana leaves in forest leaf litter productionincreased from 10.1% in 1986 to 17.1% in 2002 in tropicalrainforest in Barro Colorado Island, Republic of Panama.This dramatic increase has been proposed to be becauseof increasing atmospheric levels of carbon dioxide [14]and may have important applications in rainforest management.3 Lianas as nutrient transportersLianas are able to grow horizontally from the places wherethey are rooted until they reach a suitable vertical structure(e.g., a trellis tree or an existing liana), which they can useto gain access to the canopy and light [11,50]. Some lianasare reported to grow horizontally as far as 100 m beforebeginning to grow vertically to gain access to the tree canopy[50]. Furthermore, lianas are capable of rooting at multipleplaces and may root multiple times at different placesduring their life cycle [11,51]. This makes them more flexiblein accessing resources than trees that are normally rootedat only one place during their life time. As a result, lianasmay relocate nutrients in the form of nutrient-rich and easilydecomposedleaf litter. Apart from this horizontal transportof nutrients, lianas usually have a deep root system that allowsthem to access nutrients and water beyond the reach oftrees [6,18]. Consequently, liana-supporting trees may receivea better supply of nutrients than liana-free treesthrough regular input of liana leaf litter around their rhizospere(Figure 1).Table 2 Annual leaf litter production (t hm −2 ) of trees and lianas in tropical and subtropical forests a)Study area Forest type Tree Liana ReferenceGabon Tropical seasonal rainforest 3.85 (62%) 2.34 (38%) [47]BCI * Tropical seasonal rainforest 5.91–7.80 (83%–87%) 0.87–1.59 (13%–17%) [14]Brisbane, Australia Subtropical rainforest 4.74 (76%) 1.47 (24%) [49]Coromandel coast, India Tropical dry evergreen forest 6.9 (71%) 2.8 (29%) [48]Coromandel coast, India Tropical dry evergreen forest 5.6 (62%) 3.4 (38%) [48]a) *, Data are ranges of annual leaf litter production over 17 years in a tropical seasonal rainforest on Barro Colorado Island, Panama. Data in parenthesesare the proportions of all leaf litter.Figure 1 Illustration of a compensatory liana and host-tree relationship in which lianas provide nutrients through large amounts of nutrient-rich and easily-decomposedleaf litter to compensate for the negative effect of using trees as support.405

310 Tang Y, et al. Chin Sci Bull February (2012) Vol.57 No.44 Liana and gap dynamicsIn rainforests, tree-fall gaps play a central role in tree dynamics[52]. The formation of many tree-fall gaps involvessubsequent vigorous growth of lianas, which may enlargethe opening by ‘locking-down’ neighboring trees and thuscausing more damage to the canopy [11,12]. Under thesecircumstances, a majority of fallen lianas can survive thetree-fall, and many lianas may grow into the gap horizontallyfrom nearby intact forest [11,12,50]. Although the highabundance of lianas in tree-fall gaps may delay redevelopmentof the forest canopy by competing directly with trees[9,20], the presence of lianas in the forest gaps may promotelocal nutrient cycling and thus contribute to the regenerationduring gap dynamics.Tree-fall gaps provide a high-light environment for bothlianas that survive the tree-fall and lianas in the adjacentundisturbed forest, leading to a temporary increase in thenumber of lianas in the forest gaps [4,12]. The increase inassemblages of small trees near forest gaps may facilitateaccess of lianas to the tree canopy [4]. The high number oflianas in tree-fall gaps may also act as nutrient transporterscontributing to redevelopment of the forest canopy. Manylianas that have their crown in forest gaps are actually rootedin the nearby forest interiors. High abundance of lianasin tree-fall gaps may lead to a high input of nutrient-richleaf litter, which may partly explain the increase in soil nutrientlevels, particularly phosphorus, within gaps [34]. Asimilar phenomenon also has been observed in Amazonianrainforests, where the distribution of lianas was significantlycorrelated with phosphorus concentration [8].5 Lianas and rainforestLianas are integral elements of rainforest which, togetherwith trees, form the rainforest canopies that foster the mostdiverse terrestrial ecosystem on earth. Many trees, especiallylarge trees, bear lianas. For example, Putz [4] found thathalf of the trees with DBH > 20 cm were host to at least oneliana in a Malaysian rainforest. Despite the wide recognitionof their receiving negative effects from lianas, host treesmay also benefit from the input of liana leaf litter and associatedecological processes. This extra source may play animportant role in balancing the growth of liana-bearing treesand liana-free trees, the latter depending mainly on localsupplies of nutrients. The benefits host trees receive, however,may not compensate for the negative effects once thetrees are climbed by excessive lianas over time. In addition,lianas may also produce roots close to the host trees andcompete with them both below and above ground. Underthese circumstances, trees may experience reduced growthand fecundity or even higher mortality rates [11,17,18].Rainforests are the most productive terrestrial ecosystemson Earth even though most tropical soils are among thepoorest in terms of their nutrient content [53]. The productivityof rainforest generally is limited by nutrients such asnitrogen [54] and phosphorus [43]; therefore an efficientnutrient cycling system is crucial for tropical rainforests tomaintain the high productivity [35]. Gentry [44] noticed thatthe litter to wood production ratio was higher in tropicalrainforests than in temperate forests and attributed the differenceto the presence of abundant lianas producing a largeamount of leaf litter in rainforest.An increasing number of studies have reported the highproduction of liana leaf litter in different rainforests (Table2). As discussed above, lianas contribute not only nutrient-richand easily-decomposed leaf litter, but are also ableto relocate nutrients within the rainforest. We hypothesizethat lianas are key elements in maintaining the productivityin highly dynamics rainforests. Thus, a comprehensive reevaluationof the interaction of lianas and trees is needed.6 Future researchIt is becoming clear that lianas may play important roles inthe maintenance of rainforest diversity through their interactionswith trees and other life forms [5,10]. Apart from afew reports on the positive impacts of lianas on rainforestsas physical and nutritional resources for animals [15,55],most research has focused on the negative impacts of lianason trees and the rainforest as a whole [8,17,18]. Althoughthe role of lianas in nutrient cycling, as discussed in thispaper, can be very important in maintaining the highproductivity of rainforests, it has received little attention.Lack of understanding of the roles that lianas play in rainforestdynamics may lead to inappropriate practices in rainforestconservation and management. The liana-tree associationproposed here provides new insights for further studyof lianas and contributes to a better understanding of themaintenance of high biodiversity and of the coexistence ofspecies in rainforest.Many aspects of the proposed new liana-tree associationneed to be tested by further investigation and long-termstudies. We suggest the following questions as focal pointsfor future studies.(i) What are the rates and amounts of nutrient transportfrom lianas to their host trees? The answer to this questionis central to formulate the compensatory liana-tree relationship.Isotope techniques can be used to monitor nutrientflow from lianas to trees and verify their function in relocatingnutrients in forests.(ii) What is the contribution of liana leaf litter to litterdecomposition in different kinds of forests? The disproportionateleaf litter production of lianas may contribute significantlyto nutrient cycling in forests, not only throughrapid decomposition of liana leaf litter per se, but also by anelevated decomposition rate for the whole forest, as lianaleaf litter may also have an additive effect and promote the406

Tang Y, et al. Chin Sci Bull February (2012) Vol.57 No.4 311decomposition of leaf litter from other plants.(iii) Are there changes in rooting locations of lianas andhost trees across latitude, altitude, and rainfall gradients andduring forest regeneration after disturbance? By rooting atdifferent positions, lianas and hosts are able to minimizebelow-ground competition for nutrients and water. The distancebetween the rooting positions of lianas and hosts,however, may be affected by environmental factors, such astemperature and water availability, that regulate the transportationefficiency of lianas. We expect this distance willbecome shorter from low to high latitudes and altitudes andwill also shift during forest regeneration through disturbances.We thank the Xishuangbanna Station for Tropical Rainforest EcosystemStudies; the Chinese Academy of Sciences for providing research facilities;Joseph Wright, Sarah Boulter, Cath Moran, Francis Putz, and Carol andJerry Baskins for their valuable comments on the manuscript. This workwas supported by a National Science and Technology Pillar Program fromthe Ministry of Environmental Protection of China (2008BAC39B02) and agrant from the QCAS Biotechnology Fund.1 Gentry A. The distribution and evolution of climbing plants. In: PutzF, Mooney H, eds. The Biology of Vines. Cambridge: CambridgeUniversity Press, 1991. 3–492 Putz F E, Holbrook N M. Biomechanical studies of vines. In: Putz F,Mooney H, eds. The Biology of Vines. Cambridge: Cambridge UniversityPress, 1991. 73–973 Ewers F W, Fisher J B, Chiu S T. A survey of vessel dimensions instems of tropical lianas and other growth forms. Oecologia, 1990, 84:544–5524 Putz F E, Chai P. Ecological studies of lianas in Lambir Nationalpark, Sarawak, Malaysia. J Ecol, 1987, 75: 523–5315 Schnitzer S A, Bongers F. The ecology of lianas and their role in forests.Trends Ecol Evol, 2002, 17: 223–2306 Schnitzer S A. A mechanistic explanation for global patterns of lianaabundance and distribution. Am Nat, 2005, 166: 262–2767 DeWalt S, Ickes K, Nilus R, et al. Liana habitat associations andcommunity structure in a Bornean lowland tropical forest. Plant Ecol,2006, 186: 203–2168 Laurance W F, Perez-Salicrup P, Delamonica P, et al. Rain forestfragmentation and the structure of Amazonian liana communities.Ecology, 2001, 82: 105–1169 Schnitzer S A, Carson W P. Treefall gaps and the maintenance ofspecies diversity in a tropical forest. Ecology, 2001, 82: 913–91910 Strong D R. Epiphyte loads, tree falls, and perennial forest disruption:A mechanism for maintaining higher tree species richness in thetropics without animals. J Biogeogr, 1977, 4: 5–2111 Putz F E. The natural history of lianas on Barro Colorado Island,Panama. Ecology, 1984, 65: 1713–172412 Schnitzer S A, Dalling L, Carson W, et al. The impact of lianas ontree regeneration in tropical forest canopy gaps: Evidence for an alternativepathway of gap-phase regeneration. J Ecol, 2000, 88:655–66613 Phillips O L, Martinez R V, Arroyo L, et al. Increasing dominance oflarge lianas in Amazonian forests. Nature, 2002, 418: 770–77414 Wright S J, Calderon O, Hernandez A, et al. Are lianas increasing inimportance in tropical forests? A 17-year record from Panama. Ecology,2004, 85: 484–48915 Aide T, Zimmerman J. Patterns of insect herbivory, growth, and survivorshipin juveniles of a Neotropical liana. Ecology, 1990, 71:1412–142116 Odegaard F. The relative importance of trees versus lianas as hostsfor phytophagous beetles (Coleoptera) in tropical forests. J Biogeogr,2000, 27: 283–29617 Stevens G. Lianas as structural parasites: The Bursera simaruba example.Ecology, 1987, 68: 77–8118 Schnitzer S, Kuzee M, Gongers F. Disentangling above- and below-groundcompetition between lianas and trees in a tropical forest.J Ecol, 2005, 93: 1115–112519 Wright S J, Jaramillo M A, Pavon J, et al. Reproductive size thresholdsin tropical trees: Variation among individuals, species and forests.J Trop Ecol, 2005, 21: 307–31520 Schnitzer S A, Parren M P E, Bongers F. Recruitment of lianas intologging gaps and the effects of pre-harvest climber cutting in a lowlandforest in Cameroon. Forest Ecol Manag, 2004, 190: 87–9821 Schnitzer S A, Carson W P. Lianas suppress tree regeneration anddiversity in treefall gaps. Ecol Lett, 2010, 13: 849–85722 Perez-Salicrup D R, de Meljere W. Number of lianas per tree andnumber of trees climbed by lianas at Los Tuxtlas, Mexico. Biotropica,2005, 37: 153–15623 Isnard S, Silk W K. Moving with climbing plants from Charles Darwin’stime into the 21st century. Am J Bot, 2009, 96: 120524 Kazda M, Salzer J. Leaves of lianas and self-supporting plants differin mass per unit area and in nitrogen content. Plant Biol, 2000, 2:268–27125 Salzer J, Matezki S, Kazda M. Nutritional differences and leaf acclimationof climbing plants and the associated vegetation in differenttypes of an Andean montane rainforest. Oecologia, 2006, 147:417–42526 Schimper A F W. Plant-Geography Upon A Physiological Basis.Oxford: Clarendon Press, 190327 Kobe R K, Lepczyk C A, Iyer M. Resorption efficiency decreaseswith increasing green leaf nutrients in a global data set. Ecology,2005, 86: 2780–279228 Wright I, Reich P, Cornelissen J, et al. Modulation of leaf economictraits and trait relationships by climate. Global Ecol Biogeogr, 2005,14: 411–42129 Hegarty E. Leaf life-span and leafing phenology of lianes and associatedtrees during a rainforest succession. J Ecol, 1990, 78: 300–31230 Reich P B, Walters M B, Ellsworth D S. Leaf life-span in relation toleaf, plant, and stand characteristics among diverse ecosystems. EcolMonogr, 1992, 62: 365–39231 Zhu S D, Cao K F. Contrasting cost-benefit strategy between lianasand trees in a tropical seasonal rain forest in southwestern China.Oecologia, 2010, 163: 591–59932 Cai Z, Bongers F. Contrasting nitrogen and phosphorus resorption efficienciesin trees and lianas from a tropical montane rain forest inXishuangbanna, south-west China. J Trop Ecol, 2007, 23: 115–11833 Lambert J, Arnazon J, Iyer M. Leaf-litter and changing nutrient levelsin a seasonally dry tropical hardwood forest, Belize, C.A. Plant Soil,1980, 55: 429–44334 Vitousek P M, Denslow J S. Nitrogen and phosphorus availability intreefall gaps of a lowland tropical rainforest. J Ecol, 1986, 74: 1167–117835 Vitousek P M. Litterfall, nutrient cycling, and nutrient limitation intropical forests. Ecology, 1984, 65: 285–29836 Cleveland C C, Townsend A R, Schmidt S K. Phosphorus limitationof microbial processes in moist tropical forests: evidence fromshort-term laboratory incubations and field Studies. Ecosystems,2002, 5: 680–69137 Kaspari M, Garcia M N, Harms K E, et al. Multiple nutrients limitlitterfall and decomposition in a tropical forest. Ecol Lett, 2008, 11:35–4338 McGlynn T P, Salinas D J, Dunn R R, et al. Phosphorus limits tropicalrain forest litter fauna. Biotropica, 2007, 39: 50–5339 Finzi A C, van Breemen N, Canham C D, et al. Canopy tree-soil interactionswithin temperate forests: Species effects on soil carbon andnitrogen. Ecol Appl, 2008, 8: 440–44640 Sollins P. Factors influencing species composition in tropical lowlandrain forest: does soil matter? Ecology, 1998, 79: 23–3041 van der Heijden G M F, Phillips O L. What controls liana success inNeotropical forests? Global Ecol Biogeogr, 2008, 17: 372–383407

312 Tang Y, et al. Chin Sci Bull February (2012) Vol.57 No.442 Vitousek P M, Sanford Jr R L. Nutrient cycling in moist tropical forest.Ann Rev Ecol Evol S, 1986, 17: 137–16743 Vitousek P M, Porder S, Houlton B Z, et al. Terrestrial phosphoruslimitation: Mechanisms, implications, and nitrogen phosphorus interactions.Ecol Appl, 2010, 20: 5–1544 Gentry A. Lianas and the “paradox” of contrasting latitudinal gradientsin wood and litter production. Trop Ecol, 1983, 24: 63–6745 Hladik A. Importance des lianes dans la production foliaire de la forêtequatoriale du Nord-Est du Gabon. C R Acad Sc, 1974, 278: 2527–253046 Putz F E. Liana biomass and leaf area of a “Tierra Firme” Forest inthe Rio Negro Basin, Venezuela. Biotropica, 1983, 15: 185–18947 Hladik A. Phenology of leaf production in rain forest of Gabon: Distributionand composition of food for folivores. In: Montgomery G G,ed. The Ecology of Arboreal Folivores. Washington: Smithsonian InstitutionPress, 1978. 51–7148 Pragasan L, Parthasarathy N. Litter production in tropical dry evergreenforests of south India in relation to season, plant life-forms andphysiognomic groups. Curr Sci India, 2005, 88: 125549 Hegarty E. Leaf litter production by lianes and trees in a sub-tropicalAustralian rain forest. J Trop Ecol, 1991, 7: 201–21450 Penalosa J. Basal branching and vegetative spread in two tropical rainforest lianas. Biotropica, 1984, 16: 1–951 Caballe G. Ramet proliferation by longitudinal splitting in the Gaboneserain forest liana Dalhousiea africana S. Moore (Papilionaceae).Biotropica, 1994, 26: 266–27552 Brokaw N, Busing R T. Niche versus chance and tree diversity inforest gaps. Trends Ecol Evol, 2000, 15: 183–18853 Huston M A, Wolverton S. The global distribution of net primaryproduction: Resolving the paradox. Ecol Monogr, 2009, 79: 343–37754 LeBauer D S, Treseder K K. Nitrogen limitation of net primaryproductivity in terrestrial ecosystems is globally distributed. Ecology,2008, 89: 371–37955 Emmons L H, Gentry A H. Tropical forest structure and the distributionof gliding and prehensile-tailed vertebrates. Am Nat, 1983, 121:513Open Access This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproductionin any medium, provided the original author(s) and source are credited.408

Forest Ecology and Management 262 (2011) 1507–1513Contents lists available at ScienceDirectForest Ecology and Managementjournal homepage: www.elsevier.com/locate/forecoTopography related spatial distribution of dominant tree species in a tropicalseasonal rain forest in ChinaGuoyu Lan a,b,c , Yuehua Hu a , Min Cao a,⇑ , Hua Zhu aa Key Laboratory of Tropical Forest Ecology, Xishuangbanna Tropical Botanical Garden, Chinese Academy of Sciences, 88 Xuefu Road, Kunming 650223, PR Chinab Danzhou Key Field Station of Observation and Research for Tropical Agricultural Resources and Environments, Ministry of Agriculture, Danzhou City,Hainan Province 571737, PR Chinac Rubber Research Institute, The Chinese Academy of Tropical Agricultural Sciences, Danzhou City, Hainan Province 571737, PR ChinaarticleinfoabstractArticle history:Received 13 March 2011Received in revised form 28 June 2011Accepted 29 June 2011Available online 31 July 2011Keywords:Canonical correspondence analysis (CCA)Principal coordinates of neighbour matrices(PCNM)Spatial distribution patternVariation partitioningThe degree to which variation in species distribution is predictable from topographic variation is of considerablecurrent interest. In this paper, canonical correspondence analysis (CCA), linear regression andprincipal coordinates of neighbour matrices (PCNM) models were used to explain the variation in the distributionsof the 13 dominant species in a 20-ha tropical rain-forest plot in China. The results showedthat: (1) Tree distribution maps show that some species are mainly found in the gullies of the plot,whereas others occur on the slopes. Which indicates topographic variables are important factors forthe distribution pattern of species. (2) Both linear regression and CCA results show that convexity andelevation are the most important variables effecting distribution of trees. For saplings, elevation, convexityand aspect explain 15.3%, 9.0% and 10.1% of the total variation of species abundance. For poles, elevationand convexity explain 19.3% and 11.4% respectively. However, only 5.3% of the total variation isexplained for adults. (3) The PCNM results showed that topography alone explained 20%, 24% and 5%of the total variation of species abundance for saplings, poles and adults, respectively. Overall evidencefor topographic control of the tropical tree distribution is strong, but the explanatory power of topographicvariables was a small part of the total of variation.Ó 2011 Elsevier B.V. All rights reserved.1. IntroductionThe theory of spatial dispersion of individuals in a species is acentral concept in ecological theory. Patchiness, or the degree towhich individuals are aggregated or dispersed, is crucial to howa species uses resources, to how it is used as a resource, and toits reproductive biology (Condit et al., 2000). Several studies relatedto distribution patterns have documented that clumpedspatial distributions amongst tropical tree species are extremelycommon (Condit et al., 2000; Bunyavejchewin et al., 2003; Lanet al., 2009). Dispersal, biotic interactions, and gap dynamicsare likely to produce spatial structure most evident at relativelyfine scales, whereas topographic variation may produce structureat different scales depending on underlying geomorphology(Jones et al., 2008). Associations between plant distributionsand environmental conditions in tropical forests have been studied(Harms et al., 2001; Gunatilleke et al., 2006). In Sinharajaforest, it was found that nearly four-fifths 79% of the speciesexamined were associated with topographically defined habitats⇑ Corresponding author. Fax: +86 871 5160916.E-mail address: caom@xtbg.ac.cn (M. Cao).(Gunatilleke et al., 2006). In a plot on Barro Colorado Island, 61%of species were significantly positively or negatively associatedwith at least one habitat type according to the torus-translationtests (110 out of 171) (Harms et al., 2001).The degree to which variation in plant distribution is predictablefrom topographic variation is of considerable current interest(Jones et al., 2008). We investigated this question byanalysing the data of the 13 dominant tree species in a 20-hadipterocarp tropical seasonal rain forest in Xishuangbanna Prefecture,China. Canonical correspondence analysis was used toanalyse the relationship between species distribution and environmentalvariables. We modelled the environmental componentusing mean elevation, convexity and mean slope of each 400-m 2quadrat and the spatial component with a more flexible model,principal coordinates of neighbour matrices (PCNM). We hypothesisedthat the spatial patterns of species correlate (negatively orpositively) with the topographic variables slope, elevation, convexityand aspect). We are further interested in the relative contributionsof topographic and spatial variables to the speciesdistribution and the degree to which variation in a species distributioncan be predicted from environmental and spatialvariables.0378-1127/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved.doi:10.1016/j.foreco.2011.06.052409

1508 G. Lan et al. / Forest Ecology and Management 262 (2011) 1507–15132. Methods2.1. Study siteThe study site was located in the Xishuangbanna National NatureReserve in south-western China (101°34 0 E, 21°36 0 N). Xishuangbannasituates in the southwestern Yunnan Province of China.It borders Myanmar in the southwest and Laos in the southeast,and has mountainous topography, with mountain ridges runningin a north–south direction, decreasing in elevation southward.The uplift of the Himalayas leads to the penetration of warm andmoist tropical air mass from the Indian Ocean to Xishuangbannain the summer, and forms a barrier preventing cold air mass fromthe north reaching the region in the winter, allowing for the existenceof tropical rain forest in its altitudinal and latitudinal northernlimits. The region is dominated by a typical monsoon climatewith an alternation between a dry season and a rainy season. As recordedby a weather station 14 km from the study site, the meanannual temperature is 21.0 °C, and the mean annual precipitationis 1532 mm, of which approximately 80% occurs between Mayand October. The dry season is from November to April (Zhangand Cao, 1995). Under these climatic conditions, the tropical seasonalrain forest grows in the lowlands, valleys and hills that havea good water supply (Wu, 1987; Cao and Zhang, 1997; Zhu, 2006;Zhu et al., 2006). The tropical seasonal rain forest has three or fourindistinct tree layers: the top layer consists largely of emergenttrees (more than 30 m tall the tallest up to 60 m); the second layer,the main canopy layer, contains trees up to 30 m high with almostcontinuous crowns; the third layer consists of small trees and juveniles5–18 m tall of species from the upper layers. This forest occursmainly on laterite and lateritic red soils with pH values of4.5–5.5. The study site had suffered human disturbance about40 years ago, some trees, especially trees on the slope had beencut down.A 20-ha permanent plot was established in the XishuangbannaNational Nature Reserve in 2007. The plot is roughly square inshape and measures 400 m (north–south) by 500 (east–west).The elevation of the plot ranges from 709 to 869 m above sea level;the highest elevation occurs in the north-west corner of the plot.Three perennial creeks wind through the plot and join togetherin the south-eastern corner of the plot (Fig. 1). The forest was dominatedby Parashorea chinensis, an emergent tree species ofDipterocarpaceae.2.2. Species data collectionThe plot was established following the field protocol of the Centerfor Tropical Forest Science (Condit et al., 1996; Condit, 1998;Ayyappan and Parthasarathy, 1999). All trees (P1 cm in diameterat breast height DBH) were mapped and tagged with unique numbers.Tree diameters were measured 1.3 m from the ground. Treeswith multiple stems were counted as a single individual, but eachstem was measured and assigned a tag number. All stemsP1 cm DBH were identified to species. There were in total95,498 stems belonging to 468 species. We selected the 13 mostabundant species, which had more than 1000 individuals in theplot (Table 1). These 13 species comprised 56% of the total stems(95,834) P1.0 cm DBH (Lan et al., 2008). Individuals of each ofthese species were classified into three life stages: saplings (1 to65 cm DBH), poles (5 to 6D950.1 cm DBH) and adults(>D950.1 cm DBH). Here, D950.1 is the 95th percentile of diameterof all trees P0.1 Dmax, and Dmax is the diameter of the thickesttree (King et al., 2006). For treelets (with maximum DBH no morethan 20 cm), stems between 1 and 3 cm in diameter were classifiedas saplings. And poles included stem diameters between 3 andD950.1 cm.2.3. Topographic dataTopographic attributes include mean elevation, convexity, slopeand aspect. We divided the 20-ha plot into 500 samples, each withan area of 400 m 2 . For each 400-m 2 quadrat, the mean elevation ofa cell was defined as the mean of the elevations of its four corners,convexity was the mean elevation of the cell of interest minus themean elevation of the eight surrounding cells, and slope was themean angular deviation from the horizontal of each of the fourFig. 1. Topographic map of the 20-ha permanent plot of tropical seasonal rain forest in Xishuangbanna Prefecture of China.410

G. Lan et al. / Forest Ecology and Management 262 (2011) 1507–1513 1509Table 1The 13 most abundant species (with >1000 individuals) in the 20-ha plot in a tropical seasonal rain forest in Xishuangbanna Prefecture.Rank Species SpeciescodeFamilyMaximumDBH ofsaplingsMaximumDBHMeanDBHMinimumDBH ofadultNo. ofsaplingNo.ofpolesNo. ofadultsTotalabundance1 Pittosporopsis kerrii PITTKE Icacinaceae 5 44.3 3.5 15 16,439 4453 26 20,918 21.83Craib2 Parashorea chinensis PARACH Dipterocarpaceae 5 191.1 5.2 50 6492 1308 119 7919 8.29Wang Hsie3 Garcinia cowa Roxb. GARCCO Guttiferae 5 45.7 5.3 21 2795 1448 90 4333 4.524 Mezzettiopsis creaghii MEZZCR Annonaceae 5 49.0 6.9 25 1744 1514 42 3300 3.44Ridl.5 Baccaurea ramiflora BACCRA Euphorbiaceae 5 31.9 5.8 21 1814 1365 33 3212 3.35Lour.6 Knema furfuracea KNEMFU Myristicaceae 5 47.8 4.1 27 2543 578 39 3160 3.30(Hook. f. andThomson) Warb.7 Saprosma ternata SAPRTE Rubiaceae 3 13.0 2.0 5 2332 345 21 2698 2.82Hook. f.8 Phoebe lanceolata PHOELA Lauraceae 3 18.1 4.0 12 895 1496 18 2409 2.51(Nees) Nees9 CinnamomumCINNBE Lauraceae 5 72.6 5.5 30 938 376 23 1337 1.40bejolghota (Buch.-Ham.) Sweet10 Ficus langkokensis FICULA Moraceae 5 37.7 6.3 21 761 537 39 1337 1.40Drake11 DichapetalumDICHGE Dichapetalaceae 3 20.1 3.8 12 704 473 45 1222 1.28gelonioides (Roxb.)Engl.12 Nephelium chryseum NEPHCH Sapindaceae 5 79.6 7.2 31 713 343 42 1098 1.15Blume13 Leea compactiflora LEEACO Vitaceae 3 24.2 1.8 4 974 64 13 1051 1.10KurzTotal 53,994 56.36Percentagetriangular planes formed by connecting three of its corners (Harmset al., 2001; Gunatilleke et al., 2006). Aspect in degrees from northwas calculated using the formulaAspect ¼ 180arctanðfy=fxÞð180=3:14Þþ90ðfx=jfxjÞ;where fx and fy are the changes in elevation from east to west andfrom north to south, respectively.2.4. Canonical correspondence analysis (CCA)We ran variance partitioning (Borcard et al., 1992) to quantifythe proportion of the variation in the species distribution explainedby variation in the topographic and spatial explanatory variables.We generated a sets of continuous spatial variables (x and y, x 2 ,y 2 , xy, x 3 , y 3 , x 2 y and xy 2 ) from the x and y coordinates in the R program.Topographic variables include mean elevation, convexity,slope and aspect as stated above. Then we adjusted the correlationcoefficients (R 2 values) to account for the numbers of samplingsites and explanatory variables, as unadjusted R 2- values are biased(Jones et al., 2008). We report the adjusted values (Ra 2 ) throughout.We recorded the proportion of variation explained (Ra 2 )inCCA analyses by either the significant spatial (PCNM) or the significanttopographic variables, or both simultaneously. Using theseRa 2 values, we calculated the purely topographic, purely spatial,and spatially structured environmental fractions of the total explainedvariation in species distribution (Borcard et al., 1992).We tested the significance of the purely spatial and purely environmentalfractions by means of 999 permutations under the reducedmodel (Jones et al., 2008).3. ResultsField surveys yielded a data matrix of 500 samples (each representingan area of 400 m 2 ) and the corresponding life-stage countsof 13 tree species. Then we performed the CCA by forward selectingthe independent variables and employing a Monte Carlo permutationtest to evaluate their significance level. Each variablewas tested at the 5% confidence level with 1000 random permutations.Scatter diagrams and simple linear regressions were used toexplore the correlation between environmental variables and speciesabundance.2.5. Variation partitioning4113.1. Tree distribution mapsMost of the dominant species appeared to be distributed nonrandomlyin the habitat. For instance, most individuals of Pittosporopsiskerrii grew on the slopes of the plot, whereas P. chinensis andMezzettiopsis creaghii occurred in the gullies. Knema furfuracea andPhoebe lanceolata preferred a higher-elevation habitat, but Saprosmaternata occurred at lower elevations. Most trees of Dichapetalumgelonioides were distributed on mid-elevation slopes.Distribution maps of the three abundant species across three lifestages were plotted (Fig. 2a–c). These maps clearly show that treesare not uniform across the plot: some species are found in the gullieswhereas others are found on the slope.3.2. Linear regression of abundance topographic variablesTo assess the influence of topographic variables on species distribution,linear regression analysis of abundance with elevation,convexity, slope and aspect was performed. The correlation coefficients(Table 2) of tree abundance with topographic variables indicatethat elevation and convexity are the two most importantfactors effecting the species distribution, and slope is not essential.For example, the correlation coefficients of six species P. kerrii,Garcinia cowa, M. creaghii, S. ternata, Nephelium chryseum and K.

1510 G. Lan et al. / Forest Ecology and Management 262 (2011) 1507–1513Fig. 2. Distribution maps of the three species (Parashorea chinensis, Garcinia cowa and Mezzettiopsis creaghii) and linear regression of tree abundance with elevation andconvexity in the tropical seasonal rain forest in Xishuangbanna Prefecture of China. (Grey cross: saplings; black circle: poles; back triangle: adults) Species codes are given inTable 1.Table 2The correlation coefficients of tree abundance with topographic variables (includingelevation, convexity, slope and slope aspect). Species codes are given in Table 1.Number in bold indicates the correlation coefficients are greater than 0.0800.Rank Species code Elevation Convexity Slope Aspect1 PITTKE 0.0897 0.3325 0.0212 0.11012 PARACH 0.0492 0.0018 0.0012 0.02443 GARCCO 0.1614 0.2013 0.0126 0.00274 MEZZCR 0.2718 0.3324 0.0009 0.00115 BACCRA 0.0011 0.0481 0.0042 0.02256 KNEMFU 0.2037 0.0019 0.1001 0.08727 SAPRTE 0.1095 0.0128 0.0278 0.00018 PHOELA 0.0676 0.2014 0.0157 0.02639 CINNBE 0.0200 0.0201 0.0018 0.000610 FICULA 0.0009 0.1004 0.0084 0.120511 DICHGE 0.0014 0.0336 0.0408 0.015612 NEPHCH 0.1047 0.0008 0.0553 0.026213 LEEACO 0.0013 0.0405 0.0072 0.0903412furfuracea) are greater than 0.08 indicates elevation is very importantvariable for these species. The scatter diagrams (Fig. 2d–i)show the correlation of three species abundance with elevationand convexity.3.3. Canonical correspondence analysisFour topographic variables (elevation, slope, convexity and aspect)and life-stage abundance data from 13 species in the 500quadrats were used for the redundancy analysis. This combinationof variables explained 25.3% of the total variance in species abundances.The first axis of ordination is strongly positive correlatedwith elevation, and explained 15.6% of the total varianceP = 0.001; Monte Carlo permutation test with 1000 permutations).However the second axis of ordination is positively correlated withconvexity but negatively correlated with slope and explained anadditional 4.2% (Fig. 3).

G. Lan et al. / Forest Ecology and Management 262 (2011) 1507–1513 151124%, respectively. However, topography alone explained only 5%for adults.Results of variation partitioning of the 13 species for differentcombinations of spatial data and topographic data are shown inFig. 5. For saplings and poles, topographic variables explained morevariation than for adults, especially in P. kerrii, G. cowa, M. creaghii,K. furfuracea, S. ternata, P. lanceolata, D. gelonioides and N. chryseum.Thus, we can conclude that the saplings and poles of these specieshave distribution patterns related to topography, which could beconfirmed by the distribution maps of these species. In contrast,topographic variables played a less important role in the distributionsof adults.4. DiscussionFig. 3. Canonical correspondence analysis diagram showing the relationship ofspecies with the topographic variables (elevation, slope, convexity and aspect).Species codes are given in Table 1.3.4. Variation partitioningVariation partitioning is shown in Fig. 4. For all life stages of the13 species, topography alone explained 26% of the total variation,and space explained 5%. Topography and space together explained1% and 68% of the variation could not be explained. For saplingsand poles of the 13 species, pure topography explained 20% andThe topography of our site was very diverse, with an elevationranging from 709 to 869 m above sea level and three perennialcreeks that joined together in the south-eastern corner of the plot.This heterogeneity explains why most of the studied species, especiallyM. creaghii and P. kerrii, showed a distribution pattern relatedto topography. The 13 dominant species in our plot also had aclumped distribution (Lan et al., 2009). This pattern supports theubiquity of clumped spatial distributions amongst tropical treespecies (He et al., 1997; Condit et al., 2000; Bunyavejchewinet al., 2003). In our plot, the diverse topography may have contributedto the clumped distribution of the 13 species. This result wasconfirmed by the CCA, linear regression and variation partitioninganalysis. For example, M. creaghii, P. chinensis, S. ternata and L. compactiflorawere mainly distributed in gullies with a relatively lowelevation; in the CCA ordination diagram, these four species hada negative score in the x-coordinates, which was positivelyFig. 4. Partitioning of the variation of the dominant species using topographic and spatial data. The numbers in the circles within the box indicate the fraction of explainedvariation attributable to environmental and spatial data. All of the testable model fractions (i.e. the unique contributions) were significant, with P = 0.001 after 999permutations. Species codes are the same as in Table 1.413

1512 G. Lan et al. / Forest Ecology and Management 262 (2011) 1507–1513Fig. 5. Variation explained using topographic (elevation, slope, convexity and aspect) and spatial data (PCNM variables) of the 13 dominant species. All of the testable modelfractions (i.e. purely spatial or purely environmental fractions) were significant, with P = 0.001 after 999 permutations. Species codes are given in Table 1.correlated with elevation. In contrast, G. cowa, K. furfuracea, P.lanceolata and N. chryseum had greater x-coordinates and mainlyoccurred on slopes with high elevation.It is interesting that topography alone explained 20% of the totalvariation for saplings and 24% for poles but only 5% for adults. Thispattern arises because seedlings or saplings were apparently dispersingand establishing widely, but suffering higher mortalityoutside the optimal habitat type; thus, topography could explainmore variation for poles than for saplings. The lower variation explainedfor adults may arise from high survival of rare recruits insuboptimal habitats and high density-dependent mortality ofabundant seedlings in optimal habitats. Most young trees (saplings)in our plot have a significantly clumped distribution, butadults have a random distribution (Lan et al., 2009). These differencesindicate that density-dependent mortality of the offspringis ubiquitous in our plot.Notably, topography (or environment) explained more variationthan did space in our plot. The purely spatial fraction of explainedvariation has sometimes been interpreted aspredominantly a dispersal effect (Gilbert and Lechowicz, 2004;Cottenie, 2005; Karst et al., 2005). Pure space explained only5% of the total variation, which may indicate that dispersal isnot strong in our plot. Seed dispersal by animals predominatesin tropical forest plant species (Willson et al., 1989) and involvesa tremendous diversity of animal species and behaviours. Animalsmay consume fruit and drop, spit or defecate the seeds,carry seeds in their coats or scatter hoarded seeds for later consumption.Most types of fruits, such as drupe, berry, capsula andsyncarp are dispersed by animals. The dominant canopy species,P. chinensis, has samaras nearly 60–70% of its seeds of fall withina circle of 1–8 m near the conspecific adults. Most studies wherespace explained more variation than environment used no dataon soil chemical for trees (Borcard et al., 1992; Svenning et al.,2004; Chust et al., 2006). In those studies where environmentexplained more variation than space, soil chemical data were included(Duivenvoorden, 1995; Gilbert and Lechowicz, 2004;Cottenie, 2005; Duque et al., 2005; Karst et al., 2005). Soil chemicaldata were not used in our study, which decreased the fractionof variation explained by the environment.The abundances of many species changed with topographywhich indicates topography predicts the composition of tree speciesin a rain-forest plot in the Xishuangbanna Prefecture, China.Valencia et al. (2004) found the same pattern in eastern Ecuador.However, a relatively large proportion (68%) of species variationin our dataset was unexplained by either environmental or414spatial data. Undoubtedly, this result is partly due to randomdispersal and mortality, but it may also include deterministicvariation caused by unmeasured environmental variables (Joneset al., 2008).We found strong evidence of topographic controlling of the distributionpattern of tree species. Amongst the four topographicvariables, the convexity and elevation were the two most importantfactors contributing the distribution patterns of tree speciesin the plot. It is not difficult to hypothesise that both convexityand elevation might be negatively associated with soil moisture.Thus, soil moisture may be the most important environmental variableaffecting species distribution. Soil moisture affects bothchemical and physical properties of soils. Menendez et al. (1995)demonstrated a strong dependence of soil solution compositionon soil water content. Increases in soil moisture content led tochange in the ion distribution, free hydrated metal concentrations,and complexation (Fotovat et al., 1997). Soil moisture variationmay thereby influence the field distribution of native plantsamongst habitats.It’s worth noting that the tropical seasonal rainforest in southernYunnan conspicuously decreased from a cover of 10.9% of thetotal area of the region in 1976 down to 3.6% in 2003, mainlydue to rubber planting. Tropical seasonal rain forests have been reducedto patches of protected zones in Xishuangbanna, one of thetop rubber-producing regions in China, as voracious economicgrowth has caused increasing encroachment on China’s last remnantsof uncultivated land (Zhu, 2008). Local people must be compensatedfor losses in order to avoid encroachment. As we allknown a successful management of conservation areas heavily dependson the behaviour of the local people. In words, local governmentshould pay enough attention on conserving and managingthe tropical seasonal rain forests.AcknowledgementsThis project was supported by a grant from the National Science& Technology Pillar Program from the Ministry of EnvironmentalProtection of PR China (2008BAC39B02) and a grant from the NationalScience Foundation of China (31061160188-03). This projectwas also funded by The National Natural Science Foundation ofChina (41071040). The authors thank Drs. Fangliang He, I-FangSun and Keping Ma for their assistance in data analysis and valuablecomments on this manuscript. We also thank the two reviewersfor their constructive and helpful comments.

G. Lan et al. / Forest Ecology and Management 262 (2011) 1507–1513 1513ReferencesAyyappan, N., Parthasarathy, N., 1999. Biodiversity inventory of tree in a large–scalepermanent plot of tropical evergreen forest at Varagalaiar, Anamalais, WesternGhats. India Biodivers. Conserv. 8, 1533–1554.Borcard, D., Legendre, P., Drapeau, P., 1992. Partialling out the spatial component ofecological variation. Ecology 73, 1045–1055.Bunyavejchewin, S., Lafrankie, J.V., Baker, P.J., Kanzaki, M., Ashton, P.S., Yamakura,T., 2003. Spatial distribution patterns of the dominant canopy dipterocarpspecies in a seasonal dry evergreen forest in western Thailand. Forest. Ecol.Manage. 175, 87–101.Cao, M., Zhang, J.H., 1997. Tree species diversity of tropical forest vegetation inXishuangbanna, SW China. Biodivers. Conserv. 6, 995–1006.Chust, G., Chave, J., Condit, R., Aguilar, S., Lao, S., Pérez, R., 2006. Determinants andspatial modeling of tree diversity in a tropical forest landscape in Panama. J.Veg. Sci. 17, 83–92.Condit, R., 1998. Tropical Forest Census Plots. Springer–Verlag, NewYork, 17p.Condit, R., Hubbell, S.P., Foster, R.B., 1996. Changes in tree species abundance in aNeotropical forest: impact of climate change. J. Trop. Ecol. 12, 231–256.Condit, R., Ashton, P.S., Baker, P., Bunyavejchewin, S., Gunatilleke, S., Gunatilleke, N.,Hubbell, S.P., Foster, R.B., Itoh, A., Lafrankie, J.V., Lee, S.L., Losos, E., Manokaran,N., Sukumar, R., Yamakura, T., 2000. Spatial patterns in the distribution oftropical tree species. Science, 1114–1118.Cottenie, K., 2005. Integrating environmental and spatial processes in ecologicalcommunity dynamics. Ecol. Lett. 8, 1175–1182.Duivenvoorden, J.F., 1995. Tree species composition and rain forest environmentrelationships in the middle Caquetá area Colombia NW Amazonia. Vegetatio120, 91–113.Duque, A.J., Duivenvoorden, J.F., Cavelier, J., Sanchez, M., Polanía, C., León, A., 2005.Ferns and Melastomataceae as indicators of vascular plant composition in rainforests of Colombian Amazonia. Plant Ecol. 178, 1–13.Fotovat, A., Naidu, R., Sumner, M.E., 1997. Water:soil ratio influences aqueousphase chemistry of indigenous Cu and Zn in soils. Aust. J. Soil Res. 35, 687–709.Gilbert, B., Lechowicz, M.J., 2004. Neutrality niches and dispersal in a temperateforest understory. Proc. Natl. Acad. Sci. USA 101, 7651–7656.Gunatilleke, C.V.S., Gunatilleke, I.A.U.N., Esufali, S., Harms, K.E., Ashton, P.M.S.,Burslem, D.F.R.P., Ashton, P.S., 2006. Species-habitat associations in a Sri Lankandipterocap forest. J. Trop. Ecol. 22, 371–378.Harms, K.E., Condit, R., Hubbell, S.P., Foster, R.B., 2001. Habitat association of treeand shrubs in a 50- ha neotropical forest plot. J. Ecol. 89, 947–959.Jones, M.M., Tuomisto, H., Borcard, D., Legendre, P., Clark, D.B., Olivas, P.C., 2008.Explaining variation in tropical plant community composition: influence ofenvironmental and spatial data quality. Oecologia 155, 593–604.Karst, J., Gilbert, B., Lechowicz, M.J., 2005. Fern community assembly: the roles ofchance and the environment at local and intermediate scales. Ecology,86:2473–2486.King, D.A., Davies, S.J., Noor, N.S., 2006. Growth and mortality are related to adulttree size in a Malaysian mixed dipterocarp forest. For. Eco. Manage. 206, 152–158.Lan, G.Y., Hu, Y.H., Cao, M., Zhu, H., Wang, H., Zhou, S.H., Deng, X.B., Cui, J.Y., Huang,J.G., Liu, L.Y., Xu, H.L., Song, J.P., He, Y.C., 2008. Establishment of Xishuangbannatropical forest dynamics plot: species compositions and spatial distributionpatterns in Chinese. J. Plant. Ecol. 32, 287–298.Lan, G.Y., Zhu, H., Cao, M., Hu, Y.H., Wang, H., Deng, X.B., Zhou, S.S., Cui, J.Y., Huang,J.G., He, Y.C., Liu, L.Y., Xu, H.L., Song, J.P., 2009. Spatial dispersion patterns oftrees in a tropical rainforest in Xishuangbanna, southwest China. Ecol. Res. 24,1117–1124.Menendez, I., Moreno, G., Gallardo, J.F., Saavedra, J., 1995. Soil solution compositionin forest soils of Sierra-de-gata Mountains, Central western Spain: relationshipwith soil–water content. Arid Soil Res. Rehabil. 9, 495–502.Svenning, J.C., Kinner, D.A., Stallard, R.F., Engelbrecht, B.M.J., Wright, S.J., 2004.Ecological determinism in plant community structure across a tropical forestlandscape. Ecology 85, 2526–2538.Valencia, R., Foster, R.B., Villa, G., Condit, R., Svenning, J.C., Hernández, C.,Romoleroux, K., Losos, E., Magård, E., Balslev, H., 2004. Tree speciesdistributions and local habitat variation in the Amazon: a large forest plot ineastern Ecuador. J. Ecol. 92, 214–229.Willson, M.F., Irvine, A.K., Walsh, N.G., 1989. Vertebrate dispersal syndromes insome Australian and New-Zealand plant-communities, with geographiccomparisons. Biotropica 21, 133–147.Wu, Z.Y., 1987. Vegetation of Yunnan. Science Press, Beijing, 33p.Zhang, J., Cao, M., 1995. Tropical forest vegetation of Xishuangbanna, SW China andits secondary changes, with special reference to some problems in local natureconservation. Biodvs. Consrv. 73, 229–238.Zhu, H., 2006. Forest vegetation of Xishuangbanna, south China. Froest. Stud. China8, 1–27.Zhu, H., 2008. Advances in biogeography of the tropical rainforest in southernYunnan, southwestern China. Trop. Conserv. Sci. 1, 34–42.Zhu, H., Cao, M., Hu, H.B., 2006. Geological history, flora, and vegetation ofXishuangbanna, Southern Yunnan, China. Biotropica 38, 310–317.415

附录 Ⅰ: 中文及其他期刊发表的文章目录1. 杜彦君 , 马克平 . 2012. 森林种子雨研究进展与展望 . 生物多样性 , 20:94- 107.2. 黄俞淞 , 刘晟源 , 彭日成 , 许为斌 . 2012. 中国鹤顶兰属 ( 兰科 ) 一新记录种 . 广西植物 , 32:143-145.3. 刘海丰 , 薛达元 , 桑卫国 . 2012. 地形因子对暖温带森林群落物种丰富度 - 地上生物量关系的影响 . 生态环境学报 . 21(8):1403-1407.4. 刘俊雁 , 郑征 . 2012. 西双版纳热带森林树洞丰富度及其分配特点 . 生态学杂志 ,31(2):271- 275.5. 刘志理 , 金光泽 *. 2012. 小兴安岭三种林型叶面积指数的测定 . 应用生态学报 , 23(9):2437-2444.6. 史宝库 , 金光泽 *, 汪兆洋 . 2012. 小兴安岭 5 种林型土壤呼吸时空变异 . 生态学报 , 32(17):5416-5428.7. 谢玉彬 , 马遵平 , 杨庆松 , 方晓峰 , 张志国 , 阎恩荣 , 王希华 . 2012. 基于地形因子的天童地区常绿树种和落叶树种共存机制研究 . 生物多样性 , 20 (2): 159-167.8. 徐丽娜 , 金光泽 *. 2012. 小兴安岭凉水典型阔叶红松林动态监测样地 : 物种组成与群落结构 . 生物多样性 , 20(4): 470-481.9. 闫慧 , 吴茜 , 丁佳 , 张守仁 . 2012. 不同降水及氮添加对浙江古田山 4 种树木幼苗光合生理生态特征与生物量的影响 . 生态学报 , DOI: 10.5846/stxb201204050477.10. 张娜 , 王希华 , 郑泽梅 , 马遵平 , 杨庆松 , 方晓峰 , 谢玉彬 . 2012. 浙江天童常绿阔叶林土壤的空间异质性及其与地形的关系 . 应用生态学报 , 23(9): 2361-2369.11. 钟军弟 , 李先琨 , 吕仕洪 , 刘晟源 , 陆茂新 , 陈燕 , 成夏岚 . 2012. 广西弄岗喀斯特区域不同群落的稳定性评价分析 . 中国岩溶 , 31: 22.12. 宾粤 , 叶万辉 , 曹洪麟 , 黄忠良 , 练琚愉 . 2011. 鼎湖山南亚热带常绿阔叶林 20ha 样地幼苗的分布 . 生物多样性 , 19:127- 133.13. 程佳佳 , 米湘成 , 马克平 , 张金屯 . 2011. 亚热带常绿阔叶林群落物种多度分布格局对取样尺度的响应 . 生物多样性 , 19: 168- 177.14. 丁佳 , 吴茜 , 闫慧 , 张守仁 . 2011. 地形和土壤特性对亚热带常绿阔叶林内植物功能性状的影响 . 生物多样性 , 19:158- 167.15. 宫贵权 , 黄忠良 , 黄建雄 , 叶万辉 , 曹洪麟 , 练琚愉 , 林国俊 . 2011. 鼎湖山 20 公顷森林样地单个物种对群落的构建 . 生态环境学报 , 20: 991- 995.16. 黄俞淞 , 陆茂新 , 杨金财 , 许为斌 . 2011. 中国双唇兰属 ( 兰科 ) 一新记录种 —— 中越双唇兰 . 广西植物 , 31: 578- 580.17. 李亮 , 刘海丰 , 白帆 , 祝燕 , 李广起 , 李文超 , 桑卫国 . 2011. 东灵山 4 种落叶阔叶次生林的物种组成与群落结构 . 生物多样性 , 19: 243- 251.18. 刘海丰 , 李亮 , 桑卫国 . 2011. 东灵山暖温带落叶阔叶次生林动态监测样地 : 物种组成与群落结构 . 生物多样性 , 19: 232- 242.19. 刘文平 , 曹洪麟 , 刘卫 , 练琚愉 , 吴林芳 . 2011. 鼎湖山季风常绿阔叶林不同生境物种多样性研究 . 安徽农业科学 , 39:16159- 16163.20. 曼兴兴 , 米湘成 , 马克平 . 2011. 雪灾对古田山常绿阔叶林群落结构的影响 . 生物多样性 ,19:197- 205.21. 牛红玉 , 王峥峰 , 练琚愉 , 叶万辉 , 沈浩 . 2011. 群落构建研究的新进展 : 进化和生态相结合的群落谱系结构研究 . 生物多样性 , 19:275- 283.22. 斯幸峰 , 丁平 . 2011. 欧美陆地鸟类监测的历史、现状与我国的对策 . 生物多样性 ,416

19:303- 310.23. 宋凯 , 米湘成 , 贾琪 , 任海保 , Dan Bebber, 马克平 . 2011. 不同程度人为干扰对古田山森林群落谱系结构的影响 . 生物多样性 , 19:190- 196.24. 唐华兴 , 陈天波 , 刘晟源 , 农登攀 , 蒙渊君 , 陆茂新 . 2011. 广西弄岗自然保护区黑叶猴的种群动态 . 四川动物 , 30: 136- 140.25. 王利伟 , 李步杭 , 叶吉 , 白雪娇 , 原作强 , 邢丁亮 , 蔺菲 , 师帅 , 王绪高 , 郝占庆 . 2011.山阔叶红松林树木短期死亡动态 . 生物多样性 , 19: 260- 270.26. 王伟 , 骆争荣 , 周荣飞 , 许大明 , 哀建国 , 丁炳扬 . 2011. 百山祖常绿阔叶林木本植物的生境相关性分析 . 生物多样性 , 19:134-142.27. 汪殷华 , 米湘成 , 陈声文 , 李铭红 , 于明坚 . 2011. 古田山常绿阔叶林主要树种 2002-2007 年间更新动态 . 生物多样性 , 19:178- 189.28. 吴茜 , 丁佳 , 闫慧 , 张守仁 , 方腾 , 马克平 . 2011. 模拟降水变化和土壤施氮对浙江古田山 5 个树种幼苗生长和生物量的影响 . 植物生态学报 , 35:256- 267.29. 邢丁亮 , 郝占庆 . 最大熵原理及其在生态学研究中的应用 . 2011. 生物多样性 , 19:295-302.30. 杨庆松 , 马遵平 , 谢玉彬 , 张志国 , 王樟华 , 刘何铭 . 李萍 , 张娜 , 王达力 , 杨海波 , 方晓峰 , 阎恩荣 , 王希华 . 2011. 浙江天童 20 ha 常绿阔叶林动态监测样地的群落特征 . 生物多样性 , 19 (2): 215–223.31. 袁志良 , 王婷 , 朱学灵 , 沙迎迎 , 叶永忠 . 2011. 宝天曼落叶阔叶林样地栓皮栎种群空间格局 . 生物多样性 , 19: 224- 231.32. 张磊 , 王晓荷 , 米湘成 , 陈建华 , 于明坚 . 2011. 古田山常绿阔叶林凋落量时间动态及冰雪灾害的影响 . 生物多样性 , 19:206- 214.33. 祝燕 , 白帆 , 刘海丰 , 李文超 , 李亮 , 李广起 , 王顺忠 , 桑卫国 . 2011. 北京暖温带次生林种群分布格局与种间空间关联性 . 生物多样性 , 19: 252- 259.34. Jin GZ, Liu L, Liu ZL and Kim JH*. 2010. Spatial pattern of Larix gmelini in a Spruce-firvalley forest of Xiaoxing’an Mountains, China. Journal of Korean Forest Society, 99(5):720-725.35. 陈金玲 , 金光泽 *, 赵凤霞 . 2010. 小兴安岭典型阔叶红松林不同演替阶段凋落物分解及养分变化 . 应用生态学报 , 21(9) : 2209-2216.36. 胡跃华 , 曹敏 , 林露湘 . 2010. 西双版纳热带季节雨林的树种组成和群落结构动态 . 生态学报 , 30(4):949- 957.37. 蒋子涵 , 金光泽 *. 2010. 择伐对阔叶红松林主要组成树种种内、种间竞争的影响 . 应用生态学报 , 21(9) : 2179-2186.38. 蒋子涵 , 金光泽 *. 2010. 择伐对阔叶红松林主要树种径向与纵向生长的影响 . 生态学报 ,30(21) : 5843-5852.39. 金光泽 *, 杨桂燕 , 马建章 , 李兰军 , 徐正刚 , 赵雪 , 洪美静 . 2010. 松果采摘对小兴安岭主要林型红松土壤种子库和幼苗库的影响 . 自然资源学报 , 25(11) : 1845-1854.40. 李超 , 李凤日 , 王胜蕾 , 岳树峰 , 王绪鹏 , 刘银帮 , 金光泽 . 2010. 基于大比例尺航片的单株立木坐标提取 . 东北林业大学学报 , 38(12) : 31-34.41. 林国俊 , 黄忠良 , 竺琳 , 欧阳学军 . 2010. 鼎湖山森林群落 β 多样性 . 生态学报 , 30:4875-4880.42. 刘妍妍 , 金光泽 *. 2010. 小兴安岭阔叶红松林粗木质残体空间分布的点格局分析 . 生态学报 , 30(22) : 6072-6081.417

43. 刘妍妍 , 金光泽 *. 2010. 小兴安岭阔叶红松林粗木质残体基础特征 . 林业科学 , 46(4):8-14.44. 刘妍妍 , 金光泽 *, 黎如 . 2010. 小兴安岭阔叶红松林粗木质残体的贮量特征 . 世界林业研究 , 23: 24-30.45. Jin GZ, Li ZH, Tang Y and Kim JH*. 2009. Spatial distribution pattern and association ofcrowns and saplings for major tree species in the mixed broadleaved-Korean pine forest ofXiaoxing'an Mountains, China. Journal of Korean Forest Society, 98(2):189-196.46. 陈征 , 朱华 . 2009. 西双版纳热带雨林草本植物区系初步分析 . 西北林学院学报 ,24(1):11- 15.47. 金光泽 *, 刘志理 , 蔡慧颖 , 台秉洋 , 蒋小兰 . 2009. 小兴安岭谷地云冷杉林粗木质残体的研究 . 自然资源学报 , 24 (7): 1256-1266.48. 刘妍妍 , 金光泽 *. 2009. 地形对小兴安岭阔叶红松林粗木质残体分布的影响 . 生态学报 ,29(3): 1398-1407.49. 祝燕 , 米湘成 , 马克平 . 2009. 植物群落物种共存机制 : 负密度制约假说 . 生物多样性 ,17:594- 604.50. Jin GZ, Zhao FX, Liu L and Kim JH*. 2008. The production and spatial heterogeneity oflitterfall in the mixed broadleaved-Korean pine forest of Xiaoxing'an Mountains, China.Journal of Korean Forest Society, 97(2): 165-170.51. 郝占庆 , 张健 , 李步杭 , 叶吉 , 王绪高 , 姚晓琳 . 2008. 长白山次生杨桦林样地 : 物种组成与群落结构 . 植物生态学报 , 32:251- 261.52. 刘双 , 金光泽 *. 2008. 小兴安岭阔叶红松林种子雨的时空动态 . 生态学报 , 28(11):5731-5740.53. Jin GZ, Li R, Li ZH and Kim JH*. 2007. Spatial pattern of Acer tegmentosum in the mixedbroadleaved-Korean pine forest of Xiaoxing'an Mountains, China. Journal of Korean ForestSociety, 96(6): 730-736.54. Jin GZ, Liu YY, Liu S and Kim JH*. 2007. Effect of gaps on species diversity in the naturallyregenerated mixed broadleaved-Korean pine forest of the Xiaoxing'an Mountains, China.Journal of Ecology and Field Biology, 30(4): 325-330.55. Jin GZ, Tian YY, Zhao FX and Kim JH. 2007. The pattern of natural regeneration by gap sizein the broadleaved-Korean pine mixed forest of Xiaoxingàn mountains, China. Journal ofKorean Forest Society, 96(2): 227-234.56. Jin GZ, Xie XC, Tian YY and Kim JH*. 2006. The pattern of seed rain in thebroadleaved-Korean pine mixed forest of Xiaoxing'an Mountains, China. Journal of KoreanForest Society, 95(5): 621-627.418

AppendixⅠ: Research Papers Published in Non-SCI journals1. Du YJ and Ma KP. 2012. Advancements and prospects in forest seed rain studies.Biodiversity Science, 20: 94-107.2. Huang YS, Liu SY, Peng RC and Xu WB. 2012. A newly recorded species of Phaius(Orchidaceae) from China. Guihaia, 32(2): 143-145.3. Liu HF, Xue DY and Sang WG. 2012. Effect of topographic factors on the relationshipbetween species richness and aboveground biomass in a warm temperate forest. Ecology andEnvironmental Sciences, 21(8): 1403-1407.4. Liu JY and Zheng Z, 2012, Abundance and distribution pattern of tree cavity in tropical forestin Xishuangbanna, Southwest China. Chinese Journal of Ecology, 31: 271-275.5. Liu ZL and Jin GZ *. 2012. Estimation of leaf area index of three forest types in Xiaoxing’anMountains of northeast China. Chinese Journal of Applied Ecology, 23 (9): 2437-2444.6. Shi BK, Jin GZ* and Wang ZY. 2012. Temporal and spatial variability in soil respiration infive temperate forests in Xiaoxing'an Mountains, China. Acta Ecologica Sinica, 32(17):5416-5428.7. Xie YB, Ma ZP, Yang QS, Fang XF, Zhang ZG, Yang ER and Wang XH*. 2012. Coexistencemechanisms of evergreen and deciduous trees based on topographic factors in Tiantongregion, Zhejiang Province, eastern China. Biodiversity Science, 20:159-167.8. Xu LN and Jin GZ *. 2012. Species composition and community structure of a typical mixedbroadleaved-Korean pine (Pinus koraiensis) forest plot in Liangshui Nature Reserve, northeastChina. Biodiversity Science, 20(4): 470-481.9. Yan H, Wu Q, Ding J and Zhang SR. 2012. Effects of precipitation and nitrogen addition onphotosynthetically eco-physiological characteristics and biomass of four tree seedlings inGutian Mountian, Zhejiang Province, China. Acta Ecologica Sinica, (Accepted forpublication). DOI: 10.5846/stxb201204050477.10. Zhang N, Wang XH, Zheng ZM*, Ma ZP, Yang QS, Fang XF and Xie YB. 2012. Spatialheterogeneity of soil properties and its relationships with terrain factors I broad-leaved forestin Tiantong of Zhejiang Province, east China. Chinese Journal of Applied Ecology. 23:2361-2369.11. Zhong JD, Li XK, Lv SH, Liu SY, Lu MX, Chen Y and Cheng XL. 2012. Studies on thestability of different communities in Nonggang karst region of Guangxi. Carsologica Sinica,31(1): 22.12. Bin Y, Ye WH*, Cao HL, Huang ZL and Lian JY. 2011. Seedling distribution in a subtropicalevergreen broad-leaved forest plot in the Dinghu Mountain. Biodiversity Science,19:127-133.13. Cheng JJ, Mi XC, Ma KP and Zhang JT. 2011. Responses of species-abundance distributionto varying sampling scales in a subtropical broad-leaved forest. Biodiversity Science,19:168-177.14. Ding J, Wu Q, Yan H and Zhang SR. 2011. Effects of topographic variations and soilcharacteristics on plant functional traits in a subtropical evergreen broad-leaved forest.Biodiversity Science, 19:158-167.419

15. Gong GQ, Huang ZL, Huang JX, Ye WH, Cao HL, Lian JY and Lin GJ. 2011. Howindividual species structure the community in Dinghushan 20 ha forest plot? Ecology andEnvironmental Sciences, 20(6-7): 991-995.16. Huang YS, Lu MX, Yang JC and Xu WB. 2011. Didymoplexis vietnamica, a newly recordedspecies of Didymoplexis (Orchidaceae) from China. Guihaia, 31(5): 578-580.17. Li L, Liu HF, Bai F, Zhu Y, Li GQ, Li WC and Sang WG*. 2011. Species composition andcommunity structure of four deciduous broadleaved secondary forest in Dongling Mountain.Biodiversity Science, 19: 243-251.18. Liu HF, Li L and Sang WG. 2011. Species composition and community structure of theDonglingshan forest dynamic plot in a warm temperate deciduous broad-leaved secondaryforest, China. Biodiversity Science, 19: 232-242.19. Liu WP, Cao HL, Liu W, Lian JY and Wu LF.2011. Study on diversity of monsoon evergreenbroad leaved forest in different kinds of habitat in Dinghushan. Jounrnal of AnhuiAgriculture Sciences, 39(26):16159-16163.20. Man XX, Mi XC and Ma KP. 2011. Effects of an ice storm on community structure of anevergreen broadleaved forest in Gutianshan National Nature Reserve, Zhejiang Province.Biodiversity Science, 19:197-205.21. Niu HY, Wang ZF, Lian JY, Ye WH* and Shen H. 2011. New progress in communityassembly: community phylogenetic structure combining evolution and ecology. BiodiversityScience, 19:275- 283.22. Si XF and Ding P. 2011. History, status of monitoring land birds in Europe and America andcountermeasures of China. Biodiversity Science, 19:303-310.23. Song K, Mi XC, Jia Q, Ren HB, Bebber D and Ma KP. 2011. Variation in phylogeneticstructure of forest communities along a human disturbance gradient in Gutianshan forest,China. Biodiversity Science, 19, 190-196.24. Tang HX, Chen TB, Liu SY, Nong DP, Meng YJ, Lu MX. 2011. The Population Dynam icsof Francois LangurTrachypithecus francoisi in Nonggang NatureR eserve, Guangxi, China.Sichuan Journal of Zoology, 30:136-140.25. Wang LW, Li BH, Ye J, Bai XJ, Yuan ZQ, Xing DL, Lin F, Shi S, Wang XG and Hao. ZQ2011. Dynamics of short-term tree mortality in broad-leaved Korean pine (Pinus koraiensis)mixed forest in the Changbai Mountains. Biodiversity Science, 19: 260- 270.26. Wang W, Luo ZR, Zhou RF, Xu DM, Ai JG and Ding BY. 2011. Habitat associations ofwoody plant species in Baishanzu subtropical broad-leaved evergreen forest. BiodiversitySciences, 19: 134-142.27. Wang YH, Mi XC, Chen SW, Li MH, Yu MJ. 2011. Regeneration dynamics of major treespecies during 2002–2007 in a subtropical evergreen broad-leaved forest in GutianshanNational Nature Reserve in East China. Biodiversity Science, 19:178-189.28. Wu Q, Ding J, Yan H, Zhang SR, Fang T and Ma KP. 2011. Effects of simulated precipitationand nitrogen addition on seedling growth and biomass in five tree species in Gutian Mountain,Zhejiang Province, China. Chinese Journal of Plant Ecology, 35: 256-267.29. Xing DL and Hao ZQ. 2011. The principle of maximum entropy and its applications inecology. Biodiversity Science, 19: 295- 302.420

30. Yang QS, Ma ZP, Xie YB, Zhang ZG, Wang ZH, Liu HM, Li P, Zhang N, Wang DL, YangHB, Fang XF, Yan ER and Wang XH*. 2011. Community structure and species compositionof an evergreen broad-leaved forest in Tiantong’s 20 ha dynamic plot, Zhejiang Province,eastern China. Biodiversity Science, 19: 215-223.31. Yuan ZL, Wang T, Zhu XL, Sha YY and Ye YZ. 2011. Patterns of spatial distribution ofQuercus variabilis in deciduous broadleaf forests in Baotianman Nature Reserve.Biodiversity Science, 19: 224-231.32. Zhang L, Wang XH, Mi XC, Chen JH and Yu MJ. 2011. Temporal dynamics of and effectsof an ice storm on litter production in an evergreen broad-leaved forest in GutianshanNational Nature Reserve. Biodiversity Science, 19: 206-214.33. Zhu Y, Bai F, Liu HF, Li WC, Li L, Li GQ, Wang SZ and Sang WG. 2011. Populationdistribution patterns and interspecific spatial associations in warm temperate secondaryforests, Beijing. Biodiversity Science, 19: 252-259.34. Chen JL, Jin GZ* and Zhao FX. 2010. Litter decomposition and nutrient dynamics at differentsuccession stages of typical mixed broadleaved-Korean pine forest in Xiaoxing’an Mountains,China. Chinese Journal of Applied Ecology, 21(9): 2209-2216.35. Hu YH, Cao M and Lin LX, 2010, Dynamics of tree species composition and communitystructure of a tropical seasonal rain forest in Xishuangbanna, Southwest China. ActaEcologica Sinica, 30: 949-957.36. Jiang ZH and Jin GZ*. 2010. Effects of selective cutting on intra-and interspecies competitionsamong major tree species in mixed broadleaved-Korean pine forest. Chinese Journal ofApplied Ecology, 21(9): 2179-2186.37. Jiang ZH and Jin GZ*. 2010. Effects of selection cutting on diameter growth and verticalgrowth among major tree species in the mixed broadleaved-Korean pine forest. Acta EcologicaSinica, 30(21): 5843-5852.38. Jin GZ, Liu L, Liu ZL and Kim JH*. 2010. Spatial pattern of Larix gmelini in a Spruce-firvalley forest of Xiaoxing’an Mountains, China. Journal of Korean Forest Society, 99(5):720-725.39. Jin GZ*, Yang GY, Ma JZ, Li LJ, Xu ZG, Zhao X and Hong MJ. 2010. Effect ofanthropogenic cone-picking on seed bank and seedling bank of Korean pine in the major foresttypes in Lesser Hingàn Mountains. Journal of Natural Resources, 25(11): 1845-1854.40. Li C, Li FR, Wang SL, Yue SF, Wang XP, Liu YB and Jin GZ. 2010. Stumpage coordinateextraction based on large-scale aerial photographs. Journal of Northeast Forestry University,138(12): 31-34.41. Lin GJ, Huang ZL, Zhu L and Ouyang XJ. 2010. Beta diversity of forest community onDinghushan. Acta Ecologica Sinica, 30(18):4875-4880.42. Liu YY and Jin GZ*. 2010. Spatial point pattern analysis for coarse woody debris in a mixedbroadleaved-Korean pine forest in Xiaoxing’an Mountains, China. Acta Ecologica Sinica,30(22): 6072-6081.43. Liu YY and Jin GZ*. 2010. Character of coarse woody debris in a mixed broad-leaved Koreanpine forest in Xiaoxing’an Mountains, China. Scientia Silvae Sinicae, 46(4): 8-14.44. Liu YY, Jin GZ* and Li R. 2010. Storage characteristics of coarse woody debris in a mixedbroadleaved-Korean pine forest in Xiaoxing'an Mountains, China. World Forestry Research,421

23(9): 24-30.45. Chen Z and Zhu H. 2009. Investigation on the Flora of herbaceous plants under the Tropicalrain forest of Xishuangbanna. Journal of Northwest Forestry University, 4:11-15.46. Jin GZ*, Liu ZL, Cai HY, Tai BY, Jiang XL and Liu YY. 2009. Coarse woody debris (CWD)in a Spruce-fir valley forest in Xiaoxing'an Mountains, China. Journal of Natural Resources,24(7): 1256-1266.47. Jin GZ, Li ZH, Tang Y and Kim JH*. 2009. Spatial distribution pattern and association ofcrowns and saplings for major tree species in the mixed broadleaved-Korean pine forest ofXiaoxing'an Mountains, China. Journal of Korean Forest Society, 98(2):189-196.48. Liu YY and Jin GZ*. 2009. Influence of topography on coarse woody debris in a mixedbroadleaved-Korean pine forest in Xiaoxing'an Mountains, China. Acta Ecologica Sinica,29(3): 1398-1407.49. Zhu Y, Mi XC and Ma KP. 2009. A mechanism of plant species coexistence: the negativedensity-dependent hypothesis. Biodiversity Science, 17 (6): 594-604.50. Hao ZQ, Zhang J, Li BH, Ye J, Wang X and Yao XL. 2008. Natural secondary poplar-birchforest in Changbai Mountain: Species composition and community structure. ChineseJournal of Plant Ecology, 32:251- 261.51. Jin GZ, Zhao FX, Liu L and Kim JH*. 2008. The production and spatial heterogeneity oflitterfall in the mixed broadleaved-Korean pine forest of Xiaoxing'an Mountains, China.Journal of Korean Forest Society, 97(2): 165-170.52. Liu S and Jin GZ*. 2008. Spatiotemporal dynamics of seed rain in a broadleaved-Koreanpine mixed forest in Xiaoxing'an Mountains, China. Acta Ecologica Sinica, 28(11):5731-5740.53. Jin GZ, Li R, Li ZH and Kim JH*. 2007. Spatial pattern of Acer tegmentosum in the mixedbroadleaved-Korean pine forest of Xiaoxing'an Mountains, China. Journal of KoreanForest Society, 96(6): 730-736.54. Jin GZ, Liu YY, Liu S and Kim JH*. 2007. Effect of gaps on species diversity in the naturallyregenerated mixed broadleaved-Korean pine forest of the Xiaoxing'an Mountains, China.Journal of Ecology and Field Biology, 30(4): 325-330.55. Jin GZ, Tian YY, Zhao FX and Kim JH. 2007. The pattern of natural regeneration by gap sizein the broadleaved-Korean pine mixed forest of Xiaoxingàn mountains, China. Journal ofKorean Forest Society, 96(2): 227-234.56. Jin GZ, Xie XC, Tian YY and Kim JH*. 2006. The pattern of seed rain in thebroadleaved-Korean pine mixed forest of Xiaoxing'an Mountains, China. Journal of KoreanForest Society, 95(5): 621-627.422

附录 Ⅱ:2006-2010 发表论文目录AppendixⅡ: Research papers published in 2006-2010SCI 刊物发表的文章目录Research papers published in SCI journals in 2007-20101. Bin Y, Wang ZG, Wang ZM, Ye WH, Cao HL and Lian JY. 2010. The effects of dispersallimitation and topographic heterogeneity on beta diversity and phylobetadiversity in asubtropical forest. Plant Ecology, 209:237-256. (IF:1.88)2. Chen L, Mi XC, Comita L, Zhang LW, Ren HB and Ma KP. 2010. Community-levelconsequences of density dependence and habitat association in a subtropical broad-leavedforest. Ecology Letters, 13:695-704. (IF:15.25)3. Dong L, Wang ZF, Zhu P and Ye WH. 2010. Isolation and characterization of microsatelliteloci in Castanopsis fissa in lower subtropical China. Silvae Genetica, 299. (IF: 0.69)4. Du J, Wang N, Alpert P, Yu MJ, Yu FH and Dong M. 2010. Clonal integration increasesperformance of ramets of the fern Diplopterygium glaucum in an evergreen forest insoutheastern China. Flora, 205:399-403. (IF:1.66)5. Lang A C., Härdtle W, Bruelheide H, Geibler C, Nadrowski K, Schuldt A, Yu MJ and vonOheimb G. 2010. Tree morphology responds to neighbourhood competition and slope inspecies-rich forests of subtropical China. Forest Ecology and Management, 1708-1715. (IF:1.99)6. Wang XG, Wiegand T, Hao ZQ *, Li BH, Ye J and Lin F. 2010. Species associations in anold-growth temperate forest in north-eastern China. Journal of Ecology, 98: 674- 686. (IF:5.26)7. Wang XG, Ye J, Li BH, Zhang J, Lin F and Hao ZQ*. 2010. Spatial distributions of speciesin an old-growth temperate forest, northeastern China. Canadian Journal of ForestResearch, 40: 1011-1019. (IF: 1.57)8. Wei SG, Li L, Walther B, Ye WH, Huang ZL, Cao HL, Lian JY, Wang ZG and Chen YY.2010. Comparative performance of species-richness estimators using data from a subtropicalforest tree community. Ecological Research, 25(1):93-101. (IF:1.28)9. Zhang J, Song B, Li BH, Ye J, Wang XG and Hao ZQ *. 2010. Spatial patterns andassociations of six congeneric species in an old-growth temperate forest. Acta Oecologica,36:29- 38. (IF:1.46)10. Zhu Y, Mi XC, Ren HB and Ma KP. 2010. Density dependence is prevalent in aheterogeneous subtropical forest. Oikos, 119:109-119. (IF:3.39)11. Chen GK, Kéry M, Zhang JL and Ma KP. 2009. Factors affecting detection probability inplant distribution studies. Journal of Ecology, 97:1383-1389. (IF: 5.26)12. Du YJ, Mi XC, Ren HB, Liu XJ, Chen L, Fang T, Zhu Y and Ma KP. 2009. Seed dispersalphenology and dispersal syndromes in a subtropical broad-leaved forest of China. ForestEcology and Management, 258:1147-1152. (IF: 1.99)423

13. Huang GM, Hong L, Ye WH, Shen H, Cao HL and Xiao W. 2009. Isolation andcharacterization of polymorphic microsatellite loci in Castanopsis chinensis Hance(Fagaceae). Conservation Genetics, 10:1069-1071. (IF: 1.26)14. Lai JS, Mi XC, Ren HB and Ma KP. 2009. Species-habitat associations change in asubtropical forest of China. Journal of Vegetation Science, 20:415-423.(IF: 2.46)15. Lan GY, Zhu H, Cao M, Hu YH, Wang H, Deng XB, Zhou SS, Cui JY, Huang JG and HeYC. 2009. Spatial dispersion patterns of trees in a tropical rainforest in Xishuangbanna,southwest China. Ecological Research, 24: 1117-1124. (IF: 1.28)16. Legendre P, Mi XC, Ren HB, Ma KP, Yu MJ, Sun I-F and He FL. 2009. Partitioning betadiversity in a subtropical broad-leaved forest of China. Ecology, 90:663-674. (IF:5.07)17. Li L, Wei SG, Huang ZL, Ye WH, Cao HL, Wang ZG, Lian JY, Sun I-F, Ma KP and He FL.2009. Spatial distributions of tree species in a subtropical forest of China. Oikos,118:495-502. (IF: 3.39)18. Shen GH, Yu MJ, Hu XS, Mi XC, Ren HB, Sun I-F and Ma KP. 2009. Species-arearelationships explained by the joint effects of dispersal limitation and habitat heterogeneity.Ecology, 90:3033-3041.(IF:5.07)19. Wang XG, Hao ZQ *, Zhang J, Lian JY, Li BH and Lin XY. 2009. Tree size distributions inan old-growth temperate forest. Oikos, 118: 25- 36. (IF: 3.39)20. Wang ZG, Ye WH, Cao HL, Huang ZL, Lian JY, Li L, Wei SG and Sun I-F. 2009.Species-Topography association in a species-rich subtropical forest of China. Basic andApplied Ecology, 10:648-655. (IF: 2.20)21. Zhang J, Hao ZQ *, Song B, Li BH, Wang XG and Ye J 2009. Fine-scale speciesco-occurrence patterns in an old-growth temperate forest. Forest Ecology and Management,257(10): 2115-2120. ( IF: 1.99)22. Zhang J, Hao ZQ *, Sun I-F, Song B, Ye J, Li BH and Wang XG. 2009. Density dependenceon tree survival in an old-growth temperate forest in northeastern China. Annuals of ForestScience, 66(2): 204p1-9. (IF: 1.44)23. Zhang X, Ye WH, Cao HL, Wang ZF, Shen H and Lian JY. 2009. Isolation a ndcharacterization of microsatellites in Chinese white olive (Canarium album) andcross-species amplification in Canarium pimela. Conservation Genetics, 10:1833-1835. (IF:1.26)24. Zhu P, Ye WH, Wang ZF, Cao HL, Zhang M, Li L and Xiao W. 2009. Isolation andcharacterization of ten polymorphic microsatellite in the endangered tree Erythrophleumfordii Oliv. Conservation Genetics, 10:1017-1019.(IF: 1.26)25. Li L, Wei SG, Huang ZL, Ye WH and Cao HL. 2008. Spatial patterns and interspecificassociations of three canopy species at different life stages in a subtropical forest, China.Journal of Integrative Plant Biology, 50(9):1140-1150. (IF: 1.60)26. Wang XG, Hao ZQ *, Ye J, Zhang J, Li BH and Yao XL. 2008. Spatial pattern of diversity inan old-growth temperate forest in Northeastern China. Acta Oecologica, 33: 345- 354. (IF:1.46)27. Wang XG, Hao ZQ *, Ye J, Zhang J, Li BH and Yao XL. 2008. Spatial variation of speciesdiversity across scales in an old-growth temperate forest of China. Ecological Research,23:709- 717. (IF: 1.28)424

28. Hao ZQ*, Zhang J, Song B, Ye J and Li BH. 2007. Vertical structure and spatial associationsof dominant tree species in an old-growth temperate forest. Forest Ecology andManagement, 252: 1-11. ( IF: 1.99)中文及其他刊物发表的文章目录 (2006-2010)1. Jin GZ, Liu L, Liu ZL and Kim JH *. 2010. Spatial pattern of Larix gmelini in a Spruce-firvalley forest of Xiaoxing’an Mountains, China. Journal of Korean Forest Society, 99(5):720-725.2. 白雪娇 , 李步杭 , 张健 , 王利伟 , 原作强 , 蔺菲 , 郝占庆 . 2010. 长白山阔叶红松林灌木物种组成、结构和空间分布 . 应用生态学报 , 21:1899-1906.3. 陈金玲 , 金光泽 *, 赵凤霞 . 2010. 小兴安岭典型阔叶红松林不同演替阶段凋落物分解及养分变化 . 应用生态学报 , 21(9) : 2209-2216.4. 胡跃华 , 曹敏 , 林露湘 . 2010. 西双版纳热带季节雨林的树种组成和群落结构动态 . 生态学报 , 30(4):0949- 0957.5. 黄建雄 , 郑凤英 , 米湘成 . 2010. 不同尺度上环境因子对常绿阔叶林群落的谱系结构的影响 . 植物生态学报 , 34:309-315.6. 蒋子涵 , 金光泽 *. 2010. 择伐对阔叶红松林主要组成树种种内、种间竞争的影响 . 应用生态学报 , 21(9) : 2179-2186.7. 蒋子涵 , 金光泽 *. 2010. 择伐对阔叶红松林主要树种径向与纵向生长的影响 . 生态学报 ,30(21) : 5843-5852.8. 金光泽 *, 杨桂燕 , 马建章 , 李兰军 , 徐正刚 , 赵雪 , 洪美静 . 2010. 松果采摘对小兴安岭主要林型红松土壤种子库和幼苗库的影响 . 自然资源学报 , 25(11) : 1845-1854.9. 赖江山 , 米湘成 , 任海保 , 马克平 . 2010. 基于多元回归树的常绿阔叶林群丛数量分类—— 以古田山 24 公顷森林样地为例 . 植物生态学报 , 34:761-769.10. 李超 , 李凤日 , 王胜蕾 , 岳树峰 , 王绪鹏 , 刘银帮 , 金光泽 . 2010. 基于大比例尺航片的单株立木坐标提取 . 东北林业大学学报 , 38(12) : 31-34.11. 李立 , 陈建华 , 任海保 , 米湘成 , 于明坚 , 杨波 . 2010. 古田山常绿阔叶林优势树种甜槠和木荷的空间格局分析 . 植物生态学报 , 34:241-252.12. 林国俊 , 黄忠良 , 竺琳 , 欧阳学军 . 2010. 鼎湖山森林群落 β 多样性 . 生态学报 , 30:4875-4880.13. 刘妍妍 , 金光泽 *. 2010. 小兴安岭阔叶红松林粗木质残体的贮量特征 . 世界林业研究 , 23:24-30.14. 刘妍妍 , 金光泽 *. 2010. 小兴安岭阔叶红松林粗木质残体空间分布的点格局分析 . 生态学报 , 30(22) : 6072-6081.15. 刘妍妍 , 金光泽 *. 2010. 小兴安岭阔叶红松林粗木质残体基础特征 . 林业科学 ,46(4):8-14.16. 杨小飞 , 唐勇 , 曹敏 . 2010. 西双版纳热带季节雨林 145 个树种繁殖体特征 . 云南植物研究 , 32:367-377.17. 原作强 , 李步杭 , 白雪娇 , 蔺菲 , 师帅 , 叶吉 , 王绪高 , 郝占庆 . 2010. 长白山阔叶红松林凋落物组成及其季节动态 . 应用生态学报 , 21:2171-2178.18. Jin GZ, Li Zhihong, Tang Y and Kim JH*. 2009. Spatial distribution pattern and associationof crowns and saplings for major tree species in the mixed broadleaved-Korean pine forest ofXiaoxing'an Mountains, China. Journal of Korean Forest Society, 98(2):189-196.425

19. Luo ZR, Ding BY, Mi XC, Yu JH, Wu YG. 2009. Distribution patterns of tree species in anevergreen broadleaved forest in eastern China. Frontiers of Biology in China, 4: 531-538.20. 陈英 . 2009. 常绿阔叶林谱系多样性对幼苗存活率的影响 . 植物生态学报 , 33:1084-1089.21. 陈征 , 朱华 . 2009. 西双版纳热带雨林草本植物区系初步分析 . 西北林学院学报 ,24(1):11- 15.22. 胡正华 , 钱海源 , 于明坚 . 2009. 古田山国家级自然保护区甜槠林优势种群生态位 . 生态学报 , 29:3670-3677.23. 金光泽 *, 刘志理 , 蔡慧颖 , 台秉洋 , 蒋小兰 . 2009. 小兴安岭谷地云冷杉林粗木质残体的研究 . 自然资源学报 , 24 (7): 1256-1266.24. 李晓亮 , 王洪 , 郑征 , 林露湘 , 邓晓保 , 曹敏 . 2009. 西双版纳热带森林树种幼苗的组成 ,空间分布和旱季存活 . 植物生态学报 , 33:658-671.25. 刘妍妍 , 金光泽 *. 2009. 地形对小兴安岭阔叶红松林粗木质残体分布的影响 . 生态学报 ,29(3): 1398-1407.26. 张健 , 李步杭 , 白雪娇 , 原作强 , 王绪高 , 叶吉 , 郝占庆 . 2009. 长白山阔叶红松林乔木树种幼苗组成及其年际动态 . 生物多样性 , 17:385-396.27. 祝燕 , 米湘成 , 马克平 . 2009. 植物群落物种共存机制 : 负密度制约假说 . 生物多样性 ,17:594- 604.28. Jin GZ, Zhao FX, Liu L and Kim JH*. 2008. The production and spatial heterogeneity oflitterfall in the mixed broadleaved-Korean pine forest of Xiaoxing'an Mountains, China.Journal of Korean Forest Society, 97(2): 165-170.29. 郝占庆 , 李步杭 , 张健 , 王绪高 , 叶吉 , 姚晓琳 . 2008. 长白山阔叶红松林样地 (CBS): 群落组成与结构 . 植物生态学报 , 32:238-250.30. 郝占庆 , 张健 , 李步杭 , 叶吉 , 王绪高 , 姚晓琳 . 2008. 长白山次生杨桦林样地 : 物种组成与群落结构 . 植物生态学报 , 32:251- 261.31. 兰国玉 , 胡跃华 , 曹敏 , 朱华 , 王洪 , 周仕顺 , 邓晓保 , 崔景云 , 黄建国 , 刘林云 , 许海龙 , 宋军平 , 何有才 . 2008. 西双版纳热带森林动态监测样地 - 树种组成与空间分布格局 .植物生态学报 , 32:287-298.32. 李步杭 , 张健 , 姚晓琳 , 叶吉 , 王绪高 , 郝占庆 . 2008. 长白山阔叶红松林草本植物多样性、季节动态及其空间分布格局 . 应用生态学报 , 19:467-473.33. 刘双 , 金光泽 *. 2008. 小兴安岭阔叶红松林种子雨的时空动态 . 生态学报 , 28(11):5731-5740.34. 王绪高 , 郝占庆 , 叶吉 , 张健 , 李步杭 , 姚晓琳 . 2008. 长白山阔叶红松林物种多度和空间分布格局的关系 . 生态学杂志 , 27:145-150.35. 王志高 , 叶万辉 , 曹洪麟 , 练琚愉 . 2008. 鼎湖山季风常绿阔叶林物种多样性指数空间分布特征 . 生物多样性 , 16:454-461.36. 魏识广 , 李林 , 刘海岗 , 杜彦君 , 黄忠良 . 2008. 鼎湖山格木种群动态分析 . 生态环境 ,17:285-290.37. 姚晓琳 , 朴正吉 , 李步杭 , 张健 , 王绪高 , 叶吉 , 郝占庆 . 2008. 啮齿动物和鸟类对红松种子的消耗 . 应用生态学报 , 19:1759-1763.38. 叶万辉 , 曹洪麟 , 黄忠良 , 练琚愉 , 王志高 , 李林 , 魏识广 , 王章明 . 2008. 鼎湖山南亚热带常绿阔叶林 20 公顷样地群落特征研究 . 植物生态学报 , 32:274-286.39. 张健 , 郝占庆 , 李步杭 , 叶吉 , 王绪高 , 姚晓琳 . 2008. 长白山阔叶红松 (Pinus koraiensis)林种子雨组成及其季节动态 . 生态学报 , 28:2245-2254.426

40. Jin GZ, Li R, Li ZH and Kim J H*. 2007. Spatial pattern of Acer tegmentosum in the mixedbroadleaved-Korean pine forest of Xiaoxing'an Mountains, China. Journal of Korean ForestSociety, 96(6): 730-736.41. Jin GZ, Liu YY, Liu S and Kim J H*. 2007. Effect of gaps on species diversity in thenaturally regenerated mixed broadleaved-Korean pine forest of the Xiaoxing'an Mountains,China. Journal of Ecology and Field Biology, 30(4): 325-330.42. Jin GZ, Tian YY, Zhao FX and Kim JH. 2007. The pattern of natural regeneration by gap sizein the broadleaved-Korean pine mixed forest of Xiaoxingàn mountains, China. Journal ofKorean Forest Society, 96(2): 227-234.43. 宫贵权 , 程积民 , 米湘成 , 陈声文 , 方腾 . 2007. 古田山常绿阔叶林木本植物与生境的相关性 . 中国水土保持科学 , 5(3):79-83.44. 徐敏 , 骆争荣 , 于明坚 , 丁炳扬 , 吴友贵 .2007. 百山祖北坡中山常绿阔叶林的物种组成和群落结构 . 浙江大学学报 ( 农业与生命科学版 ) ,33, 450~457.45. 张健 , 郝占庆 , 宋波 , 叶吉 , 李步杭 , 姚晓琳 . 2007. 长白山阔叶红松林中红松与紫椴的空间分布格局及其关联性 . 应用生态学报 , 18:1681-1687.46. Jin GZ, Xie XC, Tian YY and Kim JH*. 2006. The pattern of seed rain in thebroadleaved-Korean pine mixed forest of Xiaoxing'an Mountains, China. Journal of KoreanForest Society, 95(5): 621-627.Research papers published in non-SCI journals(2006-2010)1. Chen JL, Jin GZ* and Zhao FX. 2010. Litter decomposition and nutrient dynamics at differentsuccession stages of typical mixed broadleaved-Korean pine forest in Xiaoxing’an Mountains,China. Chinese Journal of Applied Ecology. 21(9): 2209-2216.2. Hu YH, Cao M and Lin LX, 2010, Dynamics of tree species composition and communitystructure of a tropical seasonal rain forest in Xishuangbanna, Southwest China. ActaEcologica Sinica, 30: 949-957.3. Huang JX, Zheng FY and Mi XC. 2010. Influence of environmental factors on phylogeneticstructure at multiple spatial scales in an evergreen broad-leaved forest of China. ChineseJournal of Plant Ecology, 34: 309-315.4. Jiang ZH and Jin GZ*. 2010. Effects of selective cutting on intra-and interspecies competitionsamong major tree species in mixed broadleaved-Korean pine forest. Chinese Journal of AppliedEcology. 21(9): 2179-2186.5. Jiang ZH and Jin GZ*. 2010. Effects of selection cutting on diameter growth and vertical growthamong major tree species in the mixed broadleaved-Korean pine forest. Acta Ecologica Sinica.30(21): 5843-5852.6. Jin GZ*, Yang GY, Ma JZ, Li LJ, Xu ZG, Zhao X and Hong MJ. 2010. Effect of anthropogeniccone-picking on seed bank and seedling bank of Korean pine in the major forest types in LesserHingàn Mountains. Journal of Natural Resources. 25(11): 1845-1854.7. Jin GZ, Liu L, Liu ZL and Kim JH*. 2010. Spatial pattern of Larix gmelini in a Spruce-fir valleyforest of Xiaoxing’an Mountains, China. Journal of Korean Forest Society. 99(5): 720-725.8. Lai JS, Mi XC, Ren HB and Ma KP*. 2010. Numerical classification of associations insubtropical evergreen broad-leaved forest based on multivariate regression trees―a case studyof 24 hm 2 Gutianshan forest plot in China. Chinese Journal of Plant Ecology, 34:761-769.427

9. Li C, Li FR, Wang SL, Yue SF, Wang XP, Liu YB and Jin GZ. 2010. Stumpage coordinateextraction based on large-scale aerial photographs. Journal of Northeast Forestry University.138(12): 31-3410. Li L, Chen JH, Ren HB, Mi XC, Yu MJ and Yang B. 2010. Spatial patterns of Castanopsiseyrei and Schima superba in mid-subtropical broadleaved evergreen forest in GutianshanNational Nature Reserve, China. Chinese Journal of Plant Ecology, 34: 241-252.11. Lin GJ, Huang ZL, Zhu L and Ouyang XJ. 2010. Beta diversity of forest community onDinghushan. Acta Ecologica Sinica, 30(18):4875-4880.12. Liu YY, Jin GZ* and Li R. 2010. Storage characteristics of coarse woody debris in a mixedbroadleaved-Korean pine forest in Xiaoxing'an Mountains, China. World Forestry Research.23(9): 24-30.13. Liu YY and Jin GZ*. 2010. Spatial point pattern analysis for coarse woody debris in a mixedbroadleaved-Korean pine forest in Xiaoxing’an Mountains, China. Acta Ecologica Sinica, 30(22):6072-6081.14. Liu YY and Jin GZ*. 2010. Character of coarse woody debris in a mixed broad-leaved Koreanpine forest in Xiaoxing’an Mountains, China. Scientia Silvae Sinicae, 46(4): 8-14.15. Yang XF, Tang Y and Cao M. 2010. Diaspore traits of 145 tree species from a tropical seasonalrain forest in Xishuangbanna, SW China. Acta Botanica Yunnanica, 32: 367-377.16. Chen Y. 2009. Detection effect of phylogenetic diversity on seedling mortality in anevergreen broad-leaved forest in China. Chinese Journal of Plant Ecology, 33:1084-1089.17. Chen Z and Zhu H. 2009. Investigation on the Flora of herbaceous plants under the Tropicalrain forest of Xishuangbanna. Journal of Northwest Forestry University, 4:11-15.18. Hu ZH, Qian HY and Yu MJ. 2009. The niche of dominant species populations inCastanopsis eyrei forest in Gutian Mounta in National Nature Reserve. Acta EcologicaSinica, 29: 3670-3677.19. Jin GZ*, Liu ZL, Cai HY, Tai BY, Jiang XL and Liu YY. 2009. Coarse woody debris (CWD) in aSpruce-fir valley forest in Xiaoxing'an Mountains, China. Journal of Natural Resources. 24(7):1256-1266.20. Jin GZ, Li ZH, Tang Y and Kim JH*. 2009. Spatial distribution pattern and association of crownsand saplings for major tree species in the mixed broadleaved-Korean pine forest of Xiaoxing'anMountains, China. Journal of Korean Forest Society. 98(2):189-196.21. Li XL, Wang H, Zheng Z, Lin LX, Deng XB and Cao M. 2009. Composition, spatialdistribution and survival during the dry season of tree seedlings in a tropical forest inXishuangbanna, SW China. Chinese Journal of Plant Ecology, 33: 658-671.22. Liu YY and Jin GZ*. 2009. Influence of topography on coarse woody debris in a mixedbroadleaved-Korean pine forest in Xiaoxing'an Mountains, China. Acta Ecologica Sinica. 29(3):1398-1407.23. Luo ZR, Ding BY, Mi XC, Yu JH, Wu YG. 2009. Distribution patterns of tree species in anevergreen broadleaved forest in eastern China. Frontiers of Biology in China, 4: 531-538.24. Zhang J, Li BH, Bai XJ, Yuan ZQ, Wang XG, Ye J and Hao ZQ. 2009. Composition andinterannual dynamics of tree seedings in broad-leaved Korean pine (Pinus koraiensis) mixedforest in Changbai Mountain. Biodiversity Sciences, 17:385-396.428

25. Zhu Y, Mi XC and Ma KP. 2009. A mechanism of plant species coexistence: the negativedensity-dependent hypothesis. Biodiversity Science, 17: 594–604.26. Hao ZQ, Li BH, Zhang J, Wang XG, Ye J and Yao XL. 2008. Broad-leaved Korean pine(Pinus koraiensis) mixed forest plot in Changbaishan (CBS) of China: Communitycomposition and structure. Chinese Journal of Plant Ecology, 32(2):238- 250.27. Hao ZQ, Zhang J, Li BH, Ye J, Wang XG and Yao XL. 2008. Natural secondary poplar-birchforest in Changbai Mountain: Species composition and community structure. ChineseJournal of Plant Ecology, 32(2):251- 261.28. Jin GZ, Zhao FX, Liu L and Kim JH*. 2008. The production and spatial heterogeneity of litterfallin the mixed broadleaved-Korean pine forest of Xiaoxing'an Mountains, China. Journal ofKorean Forest Society. 97(2): 165-170.29. Lan GY, Hu YH, Cao M*, Zhu H, Wang H, Zhou SH, Deng XB, Cui JY, Huang JG, LiuLY, Xu HL, Song JP and He YC. 2008. Establishment of Xishuangbanna tropical forestdynamics plot: Species compositions and spatial distribution patterns. Chinese Journal ofPlant Ecology, 32: 287-298.30. Li BH, Zhang J, Yao XL, Ye J, Wang XG and Hao ZQ. 2008. Seasonal dynamics andspatial distribution patterns of herbs diversity in broadleaved Korean pine (Pinus koraiensis)mixed forest in Changbai Mountains. Chinese Journal of Applied Ecology, 19(3): 467-473.31. Liu S and Jin GZ *. 2008. Spatiotemporal dynamics of seed rain in a broadleaved-Korean pinemixed forest in Xiaoxing'an Mountains, China. Acta Ecologica Sinica. 28(11): 5731-5740.32. Wang XG, Hao ZQ, Ye J, Zhang J, Li BH and Yao XL. 2008,Relationships between speciesabundance and spatial distribution pattern of broad-leavedKorean pine (Pinus koraiensis)mixed forest in Changbai Mountains of China. Chinese Journal of Ecology, 27(2): 145-150.33. Wang ZG, Ye WH*, Cao HL and Lian JY. 2008. Spatial distribution of species diversityindices in a monsoon evergreen broadleaved forest at Dinghushan Mountain. BiodiversityScience, 16:454-461.34. Wei SG, Li L, Liu HG, Du YJ and Huang ZL. 2008. Analyses of the dynamic state ofErythrophleum fordii population. Ecology and Environment, 17:285-290.35. Yao XL, Piao ZJ, Li BH, Zhang J, Wang XG, Ye J and Hao ZQ. 2008. Pinus koraiensis seedconsumption by rodents and birds. Chinese Journal of Applied Ecology, 19(8): 1759-1763.36. Ye WH, Cao HL, Huang ZL, Lian JY, Wang ZG, Li L, Wei SG and Wang ZM. 2008.Community structure of a 20 hm 2 lower subtropical evergreen broad-leaved forest plot inDinghushan, China. Chinese Journal of Plant Ecology, 32:274-286.37. Zhang J, Hao ZQ, Li BH, Ye J, Wang XG and Yao XL. 2008. Composition and seasonaldynamics of seed rain in broad-leaved Korean pine (Pinus koraiensis) mixed forest,Changbai Mountain. Acta Ecologica Sinica, 28(6):2445- 2454.38. Zhu Y, Zhao GF, Zhang LW, Shen GC, Mi XC, Ren HB, Yu MJ, Chen JH, Chen SW, FangT and Ma KP. 2008. Community composition and structure of Gutianshan forest dynamicplot in a mid-subtropical evergreen broad-leaved forest, east China. Chinese Journal ofPlant Ecology, 32: 262-273.39. Gong GQ, Chen JM, Mi XC, Chen SW and Fang T. 2007. Habitat associations of woodspecies in the Gutianshan subtropical broad-leaved evergreen forest. Science of Soil andWater Conservation, 5: 79-83.429

40. Jin GZ, Li R, Li ZH and Kim JH*. 2007. Spatial pattern of Acer tegmentosum in the mixedbroadleaved-Korean pine forest of Xiaoxing'an Mountains, China. Journal of Korean ForestSociety. 96(6): 730-736.41. Jin GZ, Liu YY, Liu S and Kim JH*. 2007. Effect of gaps on species diversity in the naturallyregenerated mixed broadleaved-Korean pine forest of the Xiaoxing'an Mountains, China. Journalof Ecology and Field Biology. 30(4): 325-330.42. Jin GZ, Tian YY, Zhao FX and Kim JH. 2007. The pattern of natural regeneration by gap size in thebroadleaved-Korean pine mixed forest of Xiaoxingàn mountains, China. Journal of Korean ForestSociety, 96(2): 227-234.43. Xu M, Zheng ZR, Yu MJ, Ding BY, Wu YG. 2007. Floristic composition and community structureof mid-montane evergreen broad-leaved forest in north slope. Journal of Zhejiang University( Agric & Life Sci), 33: 450~457.44. Zhang J, Hao ZQ, Song B, Ye J, Li BH and Yao XL. 2007. Spatial distribution patterns andassociations of Pinus koraiensis and Tilia amurensis in broad-leaved Korean pine mixed forest inChangbai Mountains. Chinese Journal of Applied Ecology, 18:1681-1687.45. Jin GZ, Xie XC, Tian YY and Kim JH*. 2006. The pattern of seed rain in the broadleaved-Koreanpine mixed forest of Xiaoxing'an Mountains, China. Journal of Korean Forest Society. 95(5):621-627.430

Contents - ä¸­å½æ£®æçç©å¤æ ·æ§çæµç½ç»

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?

Contents - ä¸å½æ£®æçç©å¤æ ·æ§çæµç½ç»