Scheller and Mladenoff (PDF) - Forest Landscape Ecology Lab ...

Landscape EcolDOI 10.1007/s10980-006-9048-4RESEARCH ARTICLEAn ecological classification of forest landscape simulationmodels: tools and strategies for understanding broad-scaleforested ecosystemsRobert M. Scheller Æ David J. MladenoffReceived: 3 November 2004 / Accepted: 19 September 2006Ó Springer Science+Business Media B.V. 2006Abstract Computer models are increasinglybeing used by forest ecologists and managers tosimulate long-term forest landscape change. Wereview models of forest landscape change from anecological rather than methodological perspective.We developed a classification based on therepresentation of three ecological criteria: spatialinteractions, tree species community dynamics,and ecosystem processes. Spatial interactions areprocesses that spread across a landscape anddepend upon spatial context and landscape configuration.Communities of tree species maychange over time or can be defined a priori.Ecosystem process representation may rangefrom no representation to a highly mechanistic,detailed representation. Our classification highlightsthe implicit assumptions of each modelgroup and helps define the problem set for whicheach model group is most appropriate. We alsoprovide a brief history of forest landscape simulationmodels, summarize the current trends inmethods, and consider how forest landscapemodels may evolve and continue to contributeto forest ecology and management. Our classificationand review can provide novice modelersR. M. Scheller (&) Æ D. J. MladenoffDepartment of Forest Ecology and Management,University of Wisconsin-Madison, 1630 Linden Dr.,Madison, WI 53706, USAe-mail: rmscheller@wisc.eduwith the ecological context for understanding orchoosing an appropriate model for their specifichypotheses. In addition, our review clarifies thechallenges and opportunities that confront practicingmodel users and model developers.Keywords Landscape ecology Æ Forest models ÆSimulation models Æ Gap models Æ Ecosystemprocess modelsIntroductionBroadly defined, forest landscape simulationmodels (FLSMs) are computer programs forprojecting landscape change over time. FLSMscan also be used to test hypotheses about theinteractions among processes and patterns acrossforested landscapes. Processes are the endogenousand exogenous forces that drive forestchange. Patterns are the spatial configuration,composition, and heterogeneity of landscapeelements, such as community types, tree speciesage classes, ecosystem process rates, or abovegroundbiomass.In this review, we will provide an introductionto FLSMs and outline the ‘hypothesis space’ forwhich they are suitable, as well as provide anupdate and synthesis for practicing modelers. Wefirst explain the theory, concepts, and technologythat have preceded and produced the current123

Landscape Ecol‘‘landscape’’ of FLSMs. We place FLSMs into thebroader context of ecological modeling andexamine what is unique about FLSMs. Next, wereview FLSMs using a classification that tiestogether distinct groups of FLSMs and providesan ecological perspective on landscape models.We discuss strategies for deploying FLSMs, particularlyfor users who are not model developers.Last, we speculate about the future challengesand opportunities for FLSMs.History and development of forest modelsThe history of FLSMs mirrors the history of bothforest and landscape ecologies. Within forestecology, there has been a long history of modeldevelopment and application, driven by managementimperatives and evolving ecological paradigms(Shugart 1998; Mladenoff and Baker1999a). Forest ecology in the United Statesinitially focused on the forest stand (typically10 3 ha) developed as forestmanagers and ecologists confronted natural disturbances(e.g., fire, hurricanes) and humaninduced stresses (e.g., land use change, climatechange) that required the context of the largerlandscape to be sufficiently understood. Theeffects of management practices, logging in particular,also motivated the shift in perspectivetowards broader spatial and temporal scales(Franklin and Forman 1987). Over time, thisemphasis on broader scales evolved into forestecosystem management (Levin 1999), managementof disturbance regimes (Engstrom et al.1999), and forest landscape planning (Barrett2001).In the past two decades, forest models havebenefited from ongoing improvements in technologyand data availability (Mladenoff 2004).Computer speed and memory have increaseddramatically, following Moore’s prediction of adoubling of transistors per integrated circuit every18 months (Moore 1965). Simultaneously, softwarehas eased model development and themanipulation of input and output data. Hardwareand software improvements have increased ourability to simulate many processes at multiplescales. The availability of data to parameterize,initialize, and validate forest models has alsoincreased significantly. Spatially extensive dataare now readily available from either satelliteclassifications (Wolter et al. 1995; Defries et al.2000), national surveys (Hansen et al. 1992;STATSGO 1994), and historical data sources(Schulte et al. 2002).Forest landscape simulation modelsWithin this broader context, FLSMs were designedspecifically to address management orresearch questions about spatially extensive landscapes.All FLSMs are spatially explicit: landscapeelements have map coordinates and areplaced within their geographic context. Landscapeelements must be assigned values beforesimulating a landscape, i.e., the landscape musthave an initial configuration. A geographic informationsystem (GIS) is typically used to input,store, and display data. FLSMs may also bespatially interactive, i.e., simulate lateral (horizontal)fluxes or processes that spread across thelandscape (Reiners and Driese 2001; Mladenoff2004; Peters et al. 2004).Another distinctive feature of FLSMs is theiremphasis on large-scale forcing, including disturbance(Baker 1989; Mladenoff and Baker 1999b).The types of disturbances that have been modeledare extensive, although wild fire has been thelargest focus of previous modeling (Keane andLong 1998; Keane et al. 2004). Human effects onlandscape change have also become a significantfocus, including harvesting (Wallin et al. 1994;Gustafson and Crow 1998; Gustafson et al. 2000),123

Landscape EcolForest Landscape Simulation Modelsexcluding spatialinteractionsincluding spatialinteractionsstaticcommunitiesdynamiccommunitiesstaticcommunitiesdynamiccommunitiesexcludingecosystemprocessesincludingecosystemprocessesexcludingecosystemprocessesGroup 1 Group 2 Group 3DOLY 2 TEM-LPJ 5MAPSS 1,2 IBIS 3 NoneBIOME2 2 MC1 4includingecosystemprocessesexcludingecosystemprocessesincludingecosystemprocessesexcludingecosystemprocessesincludingecosystemprocessesGroup 4 Group 5 Group 6 Group 7 Group 8ForClim 7 TELSA 8 MC1 4 LANDIS 13 FIRE-BGC 16EMBYR 10HARVEST 9MetaFor 14 LANDIS-II 17LINKAGES 6 VDDT/ SEM-LAND 11 LANDSIM 12 FACET 15Fig. 1 An example decision tree based on three ecologicalcriteria: inclusion of spatial interactions, static or dynamiccommunities, and inclusion of ecosystem processes. Theorder of the decision tree can be reconfigured, dependentupon the individual’s ranking of the three criteria. Groupnumbers correspond to the group labels in the text. Two orthree exemplar models are provided for each group.Subscripts.1 Neilson (1995);2 VEMAP Members (1995);3 Foley et al. (1996); 4 Bachelet et al. (2001a, b); 5 Pan et al.(2002);6 Pastor and Post (1986a);7 Bugmann (1996);8 Klenner et al. (2000);9 Gustafson and Crow (1998);10 Hargrove et al. (2000); 11 Li (2000); 12 Roberts (1996b);13 Mladenoff et al. (1996); 14 Urban et al. (1999); 15 Urbanand Shugart (1992); 16 Keane et al. (1996); 17 Scheller et al.(in review)patterning at multiple scales and thereforecontribute to the evolution of landscape patternand changes in spatial heterogeneity. Examplesof spatial interactions include the dispersalof seeds, a fire spreading, the movement ofherbivores, and neighboring trees competing forlight.Essential to the simulation of spatial interactionsis the representation of spatial informationand the neighborhood structure. Vector polygonsare contiguous areas that can have any size orshape within the maximum extent of the landscape(Fig. 2A). Vector polygons assume withinpolygonhomogeneity. Spatial interactions amongvector polygons are often limited to first-orderneighbors, defined by a shared edge betweenpolygons (Fig. 2A). A landscape can also bebroken into a grid of equal sized, typically squareor hexagonal, cells. Spatial interactions amongcells can be defined by either neighborhoods (e.g.,the 4, 8, or 12 nearest neighbors; Fig. 2B) and/orcan be a function of the distance between cellcenter points (Fig. 2C), thereby allowing moreflexibility in spatial interactions than vector polygons(Mladenoff and He 1999; Mladenoff andBaker 1999b).Every representation of spatial interactionsincurs a computational penalty. Finer resolutionspatial interactions (e.g., neighborhood shading)will incur a larger computational penalty thancoarse-scale interactions (e.g., the dispersion oflarge clear-cut patches). The computational costof a spatial interaction is sensitive to landscaperesolution and increases non-linearly withdecreasing cell size.Finally, the simulation of a particular spatialinteraction may not be appropriate for a givenhypothesis or scale (Peters et al. 2004). Forexample, at continental scales, most landscapevariation will be explained by patterns of temperatureand precipitation. Conversely, shorttermprojections may not require estimates ofcontinental tree species migrations. If spatialinteractions provide only a marginal increase inpredictive power, the additional complexity,computational overhead, and parameterizationrequired may not warrant their inclusion (Peterset al. 2004).123

Landscape EcolFig. 2 Sampleneighborhood structures:(A) polygons with firstand second orderneighborhoods; (B) a gridwith 4, 8, and 12 cellneighborhoods. Largerneighborhoods include allcells from smallerneighborhoods; (C) anunstructuredneighborhood wherespatial interaction is afunction of distance fromthe focal cell. Resolutionand extent are arbitraryand are not indicative ofneighborhood structure orinteractionsAFfocal polygon (F)1 st order neighbor2 nd order neighborBFfocal cell (F)4 nearest neighbors8 nearest neighbors12 nearest neighborsCFStatic or dynamic communitiesAll FLSMs include spatial and temporal changeof tree species composition—forests change overtime. How a community of tree species is representedvaries widely and will be explored below.The principle difference among FLSMs iswhether the community itself is static or dynamic.For a static community, tree species compositionand associated characters are defined a priori anddo not evolve during a simulation, although thespatial distribution of the community will changeover time. This property has been termed the‘invariance hypothesis’ (Logofet and Lesnaya2000). A dynamic community is not fixed andthe composition and character of simulated communitieswill evolve over time.Forest communities can be represented assuccessional stages or ‘seral community types’(Cattelino et al. 1979; Keane and Long 1998). Theterms ‘community state’, ‘vegetation classification’,and ‘land cover type’ also apply (Yemshanovand Perera 2002). A successional stage isassociated with species, age, and/or biogeochemicalinformation. Successional pathways are typical,likely, or potential transitions amongsuccessional stages. Without disturbance, thesepathways will converge on a single ‘‘climax’’community or potential vegetation type (Keaneand Long 1998; Logofet and Lesnaya 2000).Transitions occur after a certain time passes ordisturbance occurs (Klenner et al. 2000; Keaneet al. 2002). Alternatively, time dependent transitionsprobabilities (Markov chain) can be used(Balzter et al. 1998; Logofet and Lesnaya 2000;Yemshanov and Perera 2002). Successionalstages, pathways, and transition probabilitiesare often estimated from stand-scale models123

Landscape Ecol(Acevedo et al. 2001), multi-temporal data (Turneret al. 1996), field measurements (Logofet andLesnaya 2000; Yemshanov and Perera 2002), orsubjectively defined by managers and forestecologists.Since successional stage models have staticcommunities, their application is limited to stableecosystems or short time horizons. For example,climate change or species invasion will likely altersuccessional processes and cannot be modeledwith static communities. Another critical assumptionis that all successional pathways are knownand that novel combinations of species will notoccur. Finally, the use of static communitiestypically assumes that all species propagules areuniversally available.Alternatively, other FLSMs explicitly includediscrete tree species and their life history traits(Mladenoff et al. 1996; Roberts 1996a). Speciescan be recorded as present or absent, stratifiedinto classes (diameter and age classes being themost common), or recorded as individual trees.The inclusion of discrete species implicitlyassumes that forest change can only be adequatelypredicted by considering individual specieslife history attributes, physiology, behavior,and/or biochemistry. Communities representedby multiple species, each individually respondingto changing environmental drivers, are dynamicand will evolve over time. Although apparentlymore flexible, dynamic communities require thata greater number of demographic processes beparameterized and simulated, including at leastbirth, ageing, and death. The duration of themodel simulation and the estimated stability ofcommunity types will determine whether simulatinga dynamic community is justified, given theadditional parameterization required.Ecosystem processesEcosystem processes encompass a broad range ofbiophysical and biological processes that mediatethe exchange of energy and matter between bioticpools and the abiotic environment. Within forests,ecosystem ecology has traditionally focused onnutrient cycling, productivity, and decomposition(Whittaker et al. 1974; Pastor and Post 1986b;Rastetter et al. 1991; Saxe et al. 2001). FLSMsthat explicitly include ecosystem processes typicallysimulate the net growth of trees (photosyntheticcarbon fixation minus autotrophicrespiration) at a minimum. Complexity and detailincrease as biotic pools are added or furtherdivided, more processes are included (e.g., heterotrophicrespiration), and time scales are shortened.Inclusion of ecosystem processes may be particularlyrelevant when abiotic constraints are notexpected to remain constant, as may be the casewith climate change, changes in atmosphericchemistry, or nutrient deposition (Pitelka et al.2001; Saxe et al. 2001). Ecosystem processes arealso important when there are feedbacks betweenbiological processes (e.g., succession) and ecosystemfunction (e.g., water retention, nitrogencycling) (Pastor and Post 1986b; Hooper andVitousek 1997).Functional classification of forest landscapesimulation modelsFor each of the three ecological criteria, weclassified FLSMs on the basis of the inclusion orexclusion of these processes (spatial interactionsexcluded or included; static or dynamic communities;ecosystem processes excluded or included),providing a total of eight model groups. Thesethree ecological criteria were used to form abinary decision tree (Fig. 1). Since our classificationwas intended to highlight assumptions andperspectives, we provide only examples and manyexcellent models have not been listed here. Thereare many other reviews that together can providea more comprehensive list of available models(Baker 1989; Mladenoff and Baker 1999a; Gratzeret al. 2004; Keane et al. 2004; Perry andEnright 2006).Group 1. Excluding spatial interactions; staticcommunities; excluding ecosystem processesMany of the models in this group and the nextwere designed to project broad-scale (continentalto global) change. Species are amalgamatedinto plant functional types (PFTs, e.g., ‘temperateconifer’) with homogeneous physiological attributes(growth rates, structure, disturbance123

Landscape Ecoltolerance, etc.). PFTs are not limited to foresttypes but also include tundra, grasslands, savannas,shrub lands, and arid lands (Neilson 1995;Neilson and Drapek 1998; Bachelet et al. 2001a,b, 2003). Transitions between PFTs are dependentupon climate change, soils, disturbance, andthe physiological constraints for each PFT.Specific to Group 1 are Dynamic GlobalVegetation Models (DGVMs). DGVMs projectshifts in the location of vegetation types as afunction of climate. Disturbances are generallynot simulated. For example, the Mapped Atmosphere-Plant-SoilSystem (MAPSS) projectswhere different forest types are likely to persistusing current or projected climate and broadestimates of fundamental niches (VEMAP Members1995; Neilson 1995; Neilson and Drapek1998).DGVMs have been criticized because of theirassumption that PFTs can migrate rapidly andremain in equilibrium with climate (Pearson andDawson 2003). The validity of this assumptionwill depend on the functional diversity of thePFTs. Since these and similar models do notincorporate dispersal, disturbance, or humanfragmentation of the landscape, the results shouldnot be regarded as either predictions or projections.Instead, the results define a baseline estimateof where various forests types could exist,given the projected climate change. These modelsprovide valuable broad scale data and serve as animportant contrast to more regional, strictlyforest models.Group 2. Excluding spatial interactions; staticcommunities; including ecosystem processesThis group of models simulate changes in vegetationand various carbon pools at continental toglobal scales (Foley et al. 1996; Lenihan et al.1998; Mcguire et al. 2001; Cramer et al. 2001;Aber et al. 2001; Bachelet et al. 2003). Similar tothe first group, communities are not limited toforest tree species and seed dispersal is notlimiting. Disturbances are simulated to occur ator below the resolution of the grid cell (Thonickeet al. 2001) and therefore there are no explicitspatial interactions. Within a PFT, there is novariation in disturbance effects. For example, allspecies within the conifer plant functional type(Bachelet et al. 2001a, b) will react identically,with no variation due to serotiny, bark thickness,epicormic branching, etc. Finally, these models donot yet incorporate human influences, particularlylogging or fragmentation.Nevertheless, because of the very large spatialextents that can be simulated, such models arevaluable for evaluating broad-scale changes incarbon stocks or nutrient cycling due to interactionsbetween the terrestrial biome and theatmosphere. Similar to the first group, thesemodels can also define potential vegetation atbroad (continental) scales.Group 3. Excluding spatial interactions; dynamiccommunities; excluding ecosystem processesWhen simulating dynamic communities, modeldevelopers have usually made an implicit choicebetween including ecosystem processes or includingspatial interactions, both of which are computationallyexpensive. Consequently, currentlythere are no FLSMs with dynamic communitiesand neither spatial interactions nor ecosystemprocesses.Group 4. Excluding spatial interactions; dynamiccommunities; including ecosystem processesMany of the first models that included dynamicspecies composition and ecosystem processeswere gap models (Urban and Shugart 1992).Gap models operate at the scale of individualtrees or small forest gaps, typically an area lessthan 0.1 ha (Urban and Shugart 1992; Shugart1998). Gap models simulate individual treegrowth and mortality and may include decomposition.Therefore gap models represent at leastone ecosystem process and simulate dynamicspecies composition. Although individual treesare modeled, they are not given explicit spatialcoordinates. Nor is the larger landscape contextrepresented. Seeds are assumed to be universallydistributed and disturbances are not modeled asspatially dynamic processes.Although gap models have been linkedtogether to model larger landscapes (Urban et al.1991; Busing 1991; Easterling et al. 2001), the123

Landscape Ecolapplication of linked gap models at scales >10 2 hahas been limited due to high computationaldemands and the intensive input data required(Mladenoff 2004). The calculation of disturbanceeffects on individual trees is inherently complex(Hely et al. 2003) and spatial interactions withinlinked-gap models are typically limited to finescaleprocesses, including fire, seed dispersal, andcompetition (Miller and Urban 1999; Easterlinget al. 2001).Nevertheless, gap models have played a criticalrole forming the links between ecosystem processrates and dynamic communities. In addition, gapmodels have become an important source of inputdata for broader-scale FLSMs (He et al. 1998;Urban et al. 1999; Acevedo et al. 2001; Schelleret al. 2005).Group 5. Including spatial interactions; staticcommunities; excluding ecosystem processesWithin this model group, community types aresometimes extraordinarily broad with the landscapedivided only into forested (with forest age)and non-forested types. Such models commonlyserve as ‘null’ models for exploring interactionsamong landscape processes and a small number ofstate variables. Such null models have been usedextensively to simulate fires, including the effectsof landscape connectivity on fire spread (Turneret al. 1989), fires and fuel moisture (Turner et al.1995; Finney 1998), or time since last fire (Bakeret al. 1991; Baker 1992, 1993; Peterson 2002).Null models have also been used to simulatespecies dispersal and rate of spread (Hart andGardner 1997) and the effects of harvesting onlandscape pattern (Wallin et al. 1994; Gustafsonand Crow 1998).Other models within this group focus on thetransitions between community types due todisturbances (generally fire and logging) (Klenneret al. 2000). Successional stages (and any associatedecosystem properties) and transition probabilitiesamong stages are calculated a priori.Using relatively simple community types andlikely transitions has enabled rapid model deploymentbased on surveys of expert knowledge(Klenner et al. 2000).Models in Group 5 benefit from the emphasison spatial interactions. Limiting the array ofpossible interactions (with species, with ecosystemprocesses) has yielded important insights onthe interactions between disturbances and landscapepattern or between disturbances and thedistribution of different community types.Group 6. Including spatial interactions; staticcommunities; including ecosystem processesSeveral spatially interactive models simulate ecosystemprocesses and have static communities.Again, static communities dictate certain assumptions.For example, MC1, which operates withoutspatial interactions at continental scales, has beenapplied to a smaller landscape (1250 ha) throughthe inclusion of a fire spread module (Bacheletet al. 2000). The application of MC1 to a grassland-forestecotone assumed that all propaguleswere available throughout the landscape (Bacheletet al. 2000). Similarly, SEM-LAND simulatesforest growth and fire spread (Li 2000). Notably,SEM-LAND assumes that the climate is stableand forest type never changes, although age andbiomass accumulation are dynamic (Li 2000).These models can address local or regionalscale questions (

Landscape Ecoldisturbance, and landscape patterns (Roberts1996a). Later models extended species age categoriesto grids that allowed spatial interactionsbeyond the first-order neighborhood (Mladenoffet al. 1996; Urban et al. 1999). For example,LANDIS includes many types of spatially interactiveprocesses, each with its own neighborhoodstructure (He and Mladenoff 1999; Gustafsonet al. 2000; Sturtevant et al. 2004).However, the emphasis on individual speciesbehavior can introduce significant uncertainty.For example, mean and maximum seed dispersaldistances for many tree species remains unresolved(Clark 1998; Higgins et al. 2003). Assemblingthe necessary species data can be timeconsuming and may require significant estimation.All of these models generate estimates ofspecies age distributions across a forested landscape.Species age distribution data provide anopportunity to link to many other processescorrelated with age, including reproduction andmortality, harvesting, and the development ofuneven-aged stand characteristics (Scheller et al.2005). Significantly, they provide estimates oflandscape-scale demographics that are necessaryfor identifying local extinction risk (Syphard et al.2006; Scheller et al. 2005).Group 8. Including spatial interactions; dynamiccommunities; including ecosystem processesFLSMs that include spatial interactions, dynamiccommunities, and ecosystem processes are amore recent development. At a relatively smallextent (typically 10 4 ha) often represent ecosystem processesusing relatively simple growth, mortality, anddecay functions (Keane et al. 1996; Scheller andMladenoff 2004). By necessity, such modelsexclude many finer-scale interactions, such asshading caused by neighboring cells. These modelsbenefit from direct and indirect links torelatively simple ecosystem process models, suchas PnET-II (Aber and Federer 1992; Scheller andMladenoff 2004) or FOREST-BGC (Running andGower 1991; Keane et al. 1996) and simultaneouslyinclude broad-scale spatial interactionsand dynamic communities.DiscussionStrategies for deploying FLSMsIncreases in model availability and usabilityprovide an opportunity to open up the modelingprocess to more forest ecologists and managers.However, a friendly interface cannot overcomethe inherent limitations found in every model.Modeling requires many choices between extentand resolution, precision and generality, accuracyand meaningful prediction, parameterization andvalidation (Levins 1966; Baker and Mladenoff1999; Mladenoff 2004).Our classification can serve as a guide insofaras it can help elucidate what kind of model will beappropriate for a researcher or manager. Eachresearch question will require that differentweights be applied to our three criteria and theclassification system can be customized into aunique decision tree based on an individualranking of the three criteria. Although eachmodel group is diverse, there are commonassumptions within each group that will limithypothesis testing or the scope of the projections123

Landscape Ecolgenerated. Often, the scale or resolution ofpotential hypotheses is implicitly limited withina group. Whether a particular FLSM will beappropriate is dependent upon the model’s generality,availability, and the quality of documentation.If the creation of a new FLSM is required,our classification can serve as a guide to makingassumptions and balancing comprehensivenessand tractability.An equally important decision is the choice ofscenarios. Scenarios can produce projections oflandscape change or can be used to test hypotheses(Aber 1998; Dale and Winkle 1998). Dependentupon the confidence in the simulated processformulation, one approach or the other may beappropriate, but rarely both. Comparisons amongscenarios can provide meaningful indicators oftrends and likely changes in system behavior.Management decisions can immediately benefitfrom these general indicators of the consequencesof management choices (Carpenter 2000; Clarket al. 2001). Alternatively, scenarios can be usedto test hypotheses at temporal and spatial extentsfor which no other route to hypothesis testingexists (Pielke 2003). In this sense, FLSMs aresimilar to any tool in that deciding how to use thetool is as important as choosing the proper tool.Although the scope and application of FLSMscontinues to expand, FLSMs are never appropriatefor making predictions about the timing andlocation of events. The causes that precipitatesudden landscape change (including disturbanceand land use change) are driven by stochastic oridiosyncratic events that prevent site and timespecific prediction. Furthermore, the inherentcomplexity of FLSMs limits their predictivepower (Oreskes 2003) and no FLSM can representall relevant processes. Model results shouldtherefore always be presented as qualified ‘projections’(Dale and Winkle 1998). These projectionsare limited by our ability to formulatemeaningful questions and describe plausiblefutures through scenarios.Remaining challengesValidation remains a challenge at broad spatialand temporal scales (Rastetter 1996), particularlyunder unexpected environmental conditions, asmay be produced by climate change or speciesinvasions (Rykiel 1996; Rastetter 1996). Wedefine validation as the ‘quantitative comparisonof model results against observations’ (Prisley andMortimer 2004). Although tools for validatingknown landscape patterns exist (Gardner andUrban 2003), validation of many landscape-scalephenomena remains difficult. For example, U.S.Forest Inventory and Analysis (FIA) data arespatially extensive but the resolution of the data isrelatively coarse (Hansen et al. 1992). At best,FIA data can only provide bounds for modeloutput. Similarly, analyses of existing empiricaldata for corroboration can suffer from unevenspatial and temporal resolution, differing units,and varying motivations for data collection. Nevertheless,validation is not insurmountable. Asmore FLSMs generate increasingly quantitativeprojections and more empirical data becomeavailable, greater validation of individual ecologicalprocesses (represented as model components)is possible. After achieving such reductionistvalidation, the validation of the interactionsbetween processes will be the next significantchallenge.Another large contribution of FLSMs can be toour understanding of uncertainty. Uncertaintycan be divided into three components: modeluncertainty (internal representation of an ecologicalprocess), inherent uncertainty (stochasticvariation), and parameter uncertainty (the measuredand natural variation of model inputs)(Higgins et al. 2003; Peters et al. 2004). FLSMswith a flexible architecture can help quantifymodel uncertainty by allowing different algorithmsto be tested within the same modelingframework, thereby isolating the effects of processrepresentation. Model uncertainty can alsobe assessed through cross-model comparisons(e.g., VEMAP Members 1995; Badeck et al.2001; Burke et al. 2003). Inherent uncertaintycan be assessed within FLSMs by incorporatingstochastic variation when simulating many processes.Scenarios can address parameter uncertaintyby encompassing significant sources ofuncertainty. For example, if future disturbancerates are highly uncertain, scenarios could bedesigned to represent the highest and lowestexpected disturbance rates. New computational123

Landscape Ecolmethods have also been developed to separatethe relative contribution of sources of uncertaintyand increase inferential and predictive power(Clark 2005).Finally, many existing models risk becoming‘black boxes’ to model users and many implicitassumptions are often overlooked. Consequently,the risk of mis-application and misunderstandingof model results will increase. Therefore greatertransparency in model intentions, assumptions,and limitations will be required for furtheracceptance by managers and policy makers (Aberet al. 2003). In this regard, advances in modelarchitecture (Scheller et al. in press) and implementationof the emerging open-source paradigmcan increase model rigor while simultaneouslyenhancing comprehension.ConclusionsFLSMs reflect the changing focus of ecology andsociety (Mladenoff 2004). Placing FLSMs intothis broader context highlights the implicit andexplicit assumptions and the ecological choicesthat ultimately dictated their design and behavior.Model designers and users must acknowledgethese compromises, assumptions, and weaknessesif their results are to be used by policy or decisionmakers.FLSMs have made tremendous contributions tounderstanding forest landscape change and havebeen particularly valuable to forest management.Nevertheless, many management challenges remainto be substantially addressed at broad scales.Examples include the effects of nutrient deposition,changes in atmospheric chemistry, exotic orinvasive species, and landscape change caused byrural development and fragmentation. These andother issues confronting managers and policymakers can be expected to increase the need forFLSMs and will likely determine the future scopeand application of FLSMs.Acknowledgements Dean Urban and two anonymousreviewers provided an excellent review and contributedsignificantly to the final product. Numerous other peoplehave provided critical review, including Monica Turner,Stephen Carpenter, and Volker Radeloff.ReferencesAber JD (1998) Mostly a misunderstanding, I believe. BullEcol Soc Am 79:256–259Aber JD, Federer CA (1992) A generalized, lumpedparametermodel of photosynthesis, evapotranspirationand net primary production in temperate andboreal forest ecosystems. Oecologia 92:463–474Aber JD, Neilson RP, McNulty S, Lenihan JM, BacheletD, Drapek RJ (2001) Forest processes and globalenvironmental change: predicting the effects of individualand multiple stressors. Bioscience 51:735–751Aber JD, Bernhardt ES, Dijkstra FA, Gardner RH,Macneale KH, Parton WJ, Pickett STA, Urban DL,Weathers KC (2003) Standards of practice for reviewand publication of models: summary of discussion. In:Canham CD, Cole JJ, Lauenroth WK (eds) Models inecosystem science. Princeton University Press, Princeton,New Jersey, USA, pp 204–210Acevedo MF, Ablan M, Urban DL, Pamarti S (2001)Estimating parameters of forest patch transitionmodels from gap models. Environ Model Softw16:649–658Bachelet D, Lenihan JM, Daly C, Neilson RP (2000)Interactions between fire, grazing and climate changeat Wind Cave National Park, SD. Ecol Model134:229–244Bachelet D, Lenihan JM, Daly C, Neilson RP, Ojima DS,Parton WJ (2001a) MC1: a dynamic vegetationmodel for estimating the distribution of vegetationand associated ecosystem fluxes of carbon, nutrients,and water. PNW-GTR-508. USDA Forest ServicePacific Northwest Research Station, Portland, OR,USABachelet D, Neilson RP, Lenihan JM, Drapek RJ (2001b)Climate change effects on vegetation distribution andcarbon budget in the United States. Ecosystems4:164–185Bachelet D, Neilson RP, Hickler T, Drapek RJ, LenihanJM, Sykes MT, Smith B, Sitch S, Thonicke K (2003)Simulating past and future dynamics of naturalecosystems in the United States. Global BiogeochemCycles 17:1–21Badeck F, Lischke H, Bugmann H, Hickler T, HonningerK, Lasch P, Lexer MJ, Mouillot F, Schaber J, Smith B(2001) Tree species composition in European pristineforests: comparison of stand data to model predictions.Clim Change 51:307–347Baker WL (1989) A review of models of landscape change.Landsc Ecol 2:111–333Baker WL (1992) Effects of settlement and fire suppressionon landscape structure. Ecology 73:1879–1887Baker WL (1993) Spatially heterogeneous multi-scaleresponse of landscapes to fire suppression. Oikos66:66–71Baker WL, Mladenoff DJ (1999) Progress and futuredirections in spatial modeling of forest landscapes. In:Mladenoff DJ, Baker WL (eds) Spatial modeling offorest landscape change. Cambridge University Press,Cambridge, UK, pp 333–349123

Landscape EcolBaker WL, Egbert SL, Frazier GF (1991) A spatial modelfor studying the effects of climatic change on thestructure of landscapes subject to large disturbances.Ecol Model 56:109–125Balzter H, Braun PW, Kohler W (1998) Cellular automatamodels for vegetation dynamics. Ecol Model 107:113–125Barrett TM (2001) Models of vegetative change forlandscape planning: a comparison of FETM, LAND-SUM, SIMPPLLE, and VDDT. RMRS-GTR-76-WWW. USDA Forest Service Rocky MountainResearch Station, Ogden, UT, USABotkin DB, Janak JF, Wallis JR (1973) Some ecologicalconsequences of a computer model of forest growth. JEcol 60:849–872Bugmann HKM (1996) A simplified forest model to studyspecies composition along climate gradients. Ecology77:2055–2074Burke IC, Kaye JP, Bird SP, Hall SA, McCulley RL,Sommerville GL (2003) Evaluating and testing modelsof terrestrial biogeochemistry: the role of temperaturein controlling decomposition. In: Canham CD,Cole JJ, Lauenroth WK (eds) The role of models inecosystem science. Princeton University Press, Princeton,NJ, USA, pp 225–253Busing RT (1991) A spatial model of forest dynamics.Vegetatio 92:167–179Carpenter SR (2000) Ecological futures: building anecology of the long now. Ecology 83:2069–2083Cattelino PJ, Noble IR, Slatyer RO, Kessell SR (1979)Predicting the multiple pathways of plant succession.Environ Manage 3:41–50Clark JS (1998) Why trees migrate so fast: confrontingtheory with dispersal biology and the paleorecord.Am Nat 152:204–224Clark JS (2005) Why environmental scientists are becomingBayesians. Ecol Lett 8:2–14Clark JS, Carpenter SR, Barber M, Collins S, Dobson A,Foley JA, Lodge DM, Pascual M, Pielke R Jr, Pizer W,Pringle C, Reid WV, Rose KA, Sala O, SchlesingerWH, Wall DH, Wear D (2001) Ecological forecasts: anemerging imperative. Science 293:657–660Cramer W, Bondeau A, Woodward FI, Prentice IC, BettsRA, Brovkin V, Cox PM, Fisher V, Foley JA, FriendAD, Kucharik C, Lomas MR, Ramankutty N, Sitch S,Smith B, White A, Young-Molling C (2001) Globalresponse of terrestrial ecosystem structure and functionto CO 2 and climate change: results from sixdynamic global vegetation models. Glob Change Biol7:357–373Dale VH, Pearson SM (1999) Modeling the driving factorsand ecological consequences of deforestation in theBrazilian Amazon. In: Mladenoff DJ, Baker WL (eds)Spatial modeling of forest landscape change. CambridgeUniversity Press, Cambridge, UK, pp 256–276Dale VH, Winkle WV (1998) Models provide understanding,not belief. Bull Ecol Soc Am 79:169–170Defries RS, Hansen MC JR, Townshend G, Janetos AC,Loveland TR (2000) A new global 1-km dataset ofpercentage tree cover derived from remote sensing.Glob Change Biol 6:247–254Easterling WE, Brandle JR, Hays CJ, Guo QF, GuertinDS (2001) Simulating the impact of human land usechange on forest composition in the Great Plainsagroecosystems with the Seedscape model. EcolModel 140:163–176Ek AR, Monserud RA (1979) Performance and comparisonof stand growth-models based on individualtree and diameter-class growth. Can J For Res9:231–244Engstrom RT, Gilbert S, Hunter ML Jr, Merriwether D,Nowacki GJ, Spencer P (1999) Practical applicationsof disturbance ecology to natural resource management.In: Szaro RC, Johnson NC, Sexton WT, MalkAJ (eds) Ecological stewardship. A common referencefor ecosystem management, vol 2. ElsevierScience Ltd., Oxford, UK, pp 313–330Finney MA (1998) Farsite: fire area simulator—modeldevelopment and evaluation. Research Paper RMRS-RP-4 Revised. USDA Forest Service IntermountainFire Sciences Laboratory, Missoula, MT, USAFoley JA, Prentice IC, Ramankutty N, Levis S, Pollard D,Sitch S, Haxeltine A (1996) An integrated biospheremodel of land surface processes, terrestrial carbonbalance, and vegetation dynamics. Glob BiogeochemCycles 10:602–628Franklin JF, Forman RTT (1987) Creating landscapepatterns by forest cutting: Ecological consequencesand principles. Landsc Ecol 1(1):5–18Gardner RH, Urban DL (2003) Model validation andtesting: past lessons, present concerns, future prospects.In: Canham CD, Cole JJ, Lauenroth WK (eds)Models in ecosystem science. Princeton UniversityPress, Princeton, NJ, USA, pp 184–203Gardner RH, Milne BT, Turner MG, O’Neill RV (1987)Neutral models for the analysis of broad-scale landscapepattern. Landsc Ecol 1:19–28Gratzer G, Canham C, Dieckmann U, Fischer A, Iwasa Y,Law R, Lexer MJ, Sandmann H, Spies TA, SplechtnaBE, Szwagrzyk J (2004) Spatio-temporal developmentof forests—current trends in field methods and models.Oikos 107:3–15Gustafson EJ, Crow TR (1998) Simulating spatial andtemporal context of forest management using hypotheticallandscapes. Environ Manage 22:777–787Gustafson EJ, Shifley SR, Mladenoff DJ, Nimerfro KK,He HS (2000) Spatial simulation of forest successionand timber harvesting using LANDIS. Can J For Res30:32–43Hansen MH, Frieswyk T, Glover JF, Kelly JF (1992) Theeastwide forest inventory data base: users manual.GTR NC-151. USDA Forest Service North CentralForest Experiment Station, St. Paul, MN, USAHargrove WW, Gardner RH, Turner MG, Romme WH,Despain DG (2000) Simulating fire patterns in heterogeneouslandscapes. Ecol Model 135:243–263Hart DR, Gardner RH (1997) A spatial model for thespread of invading organisms subject to competition. JMath Biol 35:935–948He HS, Mladenoff DJ (1999) Spatially explicit andstochastic simulation of forest landscape fire disturbanceand succession. Ecology 80:81–99123

Landscape EcolHe HS, Mladenoff DJ, Crow TR (1998) Linking anecosystem model and a landscape model to studyforest species response to climate warming. EcolModel 114:213–233Hely C, Flannigan MD, Bergeron Y (2003) Modeling treemortality following wildfire in the southeastern Canadianmixed-wood boreal forest. For Sci 49:566–576Higgins SI, Clark JS, Nathan R, Hovestadt T, Schurr F,Fragoso JMV, Aguiar MR, Ribbens E, Lavorel S(2003) Forecasting plant migration rates: managinguncertainty for risk assessment. J Ecol 91:341–347Hooper DU, Vitousek PM (1997) The effects of plantcomposition and diversity on ecosystem processes.Science 277:1302–1305Keane RE, Finney MA (2003) The simulation of landscapefire, climate, and ecosystem dynamics. In: Veblen TT,Baker WL, Montenegro G, Swetnam TW (eds) Fireand climatic change in temperate ecosystems of theWestern Americas. Springer, New York, NY, USA,pp 32–68Keane RE, Long DG (1998) A comparison of coarse scalefire effects simulation strategies. Northwest Sci 72:76–89Keane RE, Ryan KC, Running SW (1996) Simulatingeffects of fire on northern Rocky Mountain landscapeswith the ecological process model FIRE-BGC.Tree Physiol 16:319–331Keane RE, Parsons RA, Hessburg PF (2002) Estimatinghistorical range and variation of landscape patchdynamics: limitations of the simulation approach.Ecol Model 151:29–49Keane RE, Cary GJ, Davies ID, Flannigan MD, GardnerRH, Lavorel S, Lenihan JM, Li C, Rupp TS (2004) Aclassification of landscape fire succession models:spatial simulations of fire and vegetation dynamics.Ecol Model 179:3–27Klenner W, Kurz W, Beukema S (2000) Habitat patternsin forested landscapes: management practices and theuncertainty associated with natural disturbances.Comput Electron Agric 27:243–262Lenihan JM, Daly C, Bachelet D, Neilson RP (1998)Simulating broad-scale fire severity in a dynamicglobal vegetation model. Northwest Sci 72:91–103Lett C, Silber C, Barret N (1999) Comparison of a cellularautomata network and an individual-based model forthe simulation of forest dynamics. Ecol Model121:277–293Levin SA (1992) The problem of pattern and scale inecology. Ecology 72:1943–1967Levin SA (1999) Towards a science of ecological management.Conserv Ecol [Online] 3:6Levins R (1966) The strategy of model building inpopulation biology. Am Sci 54:421–431Li C (2000) Reconstruction of natural fire regimes throughecological modelling. Ecol Model 134:129–144Li C, Flannigan MD, Corns IGW (2000) Influence ofpotential climate change on forest landscape dynamicsof west-central Alberta. Can J For Res 30:1905–1912Likens GE, Bormann FH, Johnson NM, Fisher DW,Pierce RS (1970) Effects of forest cutting and herbicidetreatment on nutrient budgets in the HubbardBrook watershed-ecosystem. Ecol Monogr 40:23–47Liu J, Ashton PS (1999) Simulating effects of landscapecontext and timber harvest on tree species diversity.Ecol Appl 9:186–201Logofet DO, Lesnaya EV (2000) The mathematics ofMarkov models: what Markov chains can reallypredict in forest successions. Ecol Model 126:285–298Maxwell T, Costanza R (1997) A language for modularspatio-temporal simulation. Ecol Model 103:105–113Mcguire AD, Sitch S, Clein JS, Dargaville R, Esser G,Foley J, Heimann M, Joos F, Kaplan J, KicklighterDW, Meier RA, Melillo JM, Moore B, Prentice IC,Ramankutty N, Reichenau T, Schloss A, Tian H,Williams LJ, Wittenberg U (2001) Carbon balance ofthe terrestrial biosphere in the twentieth century:analyses of Co2, climate and land use effects with fourprocess-based ecosystem models. Glob BiogeochemCycles 15:183–206Miller C. Urban DL (1999) Forest pattern, fire, andclimatic change in the Sierra Nevada. Ecosystems2:76–87Mladenoff DJ (2004) LANDIS and forest landscapemodels. Ecol Model 180:7–19Mladenoff DJ, Baker WL (1999a) Development of forestand landscape modeling approaches. In: MladenoffDJ, Baker WL (eds) Spatial modeling of forestlandscape change. Cambridge University Press, Cambridge,UK, pp 1–13Mladenoff DJ, Baker WL (eds) (1999b) Spatial modelingof forest landscape change. Approaches and applications.Cambridge University Press, Cambridge, UKMladenoff DJ, He HS (1999) Design, behavior andapplication of LANDIS, an object-oriented model offorest landscape disturbance and succession. In:Mladenoff DJ, Baker WL (eds) Spatial modeling offorest landscape change. Cambridge University Press,Cambridge, UK, pp 125–162Mladenoff DJ, Host GE, Boeder J, Crow TR (1996)LANDIS: a spatial model of forest landscape disturbance,succession, and management. In: GoodchildMF, Steyaert LT, Parks BO, Johnston CA, MaidmentD, Crane M, Glendinning S (eds) GIS andenvironmental modeling: progress and researchissues. GIS World Books, Fort Collins, CO, USA,pp 175–179Moore GE (1965) Cramming more components ontointegrated circuits. Electronics 38:114–117Neilson RP (1995) A model for predicting continentalscalevegetation distribution and water balance. EcolAppl 5:352–385Neilson RP, Drapek RJ (1998) Potentially complex biosphereresponses to transient global warming. GlobChange Biol 4:505–521Oreskes N (2003) The role of quantitative models inscience. In: Canham CD, Cole JJ, Lauenroth WK(eds) Models in ecosystem science. Princeton UniversityPress, Princeton, NJ, USA, pp 13–31Pacala SW, Canham CD, Silander JA Jr (1993) Forestmodels defined by field measurements. I. The design123

Landscape Ecolof a northeastern forest simulator. Can J For Res23:1980–1988Pan Y, Mcguire AD, Melillo JM, Kicklighter DW, Sitch S,Prentice IC (2002) A biogeochemistry-based dynamicvegetation model and its application along a moisturegradient in the continental United States. J Veg Sci13:369–382Pastor J, Post WM (1986a) Development of a linked forestproductivity-soil process model. ORNL/TM-(9519).Oak Ridge National Laboratory, Oak Ridge, TN,USAPastor J, Post WM (1986b) Influence of climate, soilmoisture, and succession on forest carbon and nitrogencycles. Biogeochemistry 2:3–27Pearson RG, Dawson TP (2003) Predicting the impacts ofclimate change on the distribution of species: arebioclimate envelope models useful? Glob Ecol Biogeogr12:361–371Perry GLW, Enright NJ (2006) Spatial modelling ofvegetation change in dynamic landscapes: a review ofmethods and applications. Prog Phys Geogr 30:47–72Peters DPC, Herrick JE, Urban DL, Gardner RH,Breshears DD (2004) Strategies for ecological extrapolation.Oikos 106:627–636Peterson GD (2002) Contagious disturbance, ecologicalmemory, and the emergence of landscape pattern.Ecosystems 5:329–338Pielke RA Jr (2003) The role of models in prediction fordecision. In: Canham CD, Cole JJ, Lauenroth WK(eds) Models in ecosystem science. Princeton UniversityPress, Princeton, NJ, USA, pp 111–138Pitelka LF, Bugmann H, Reynolds JF (2001) How muchphysiology is needed in forest gap models for simulatinglong-term vegetation response to globalchange? Introduction. Clim Change 51:251–257Prisley SP, Mortimer MJ (2004) A synthesis of literatureon evaluation of models for policy applications, withimplications for forest carbon accounting. For EcolManage 198:89–103Rastetter EB (1996) Validating models of ecosystemresponse to climate change. BioScience 46:190–198Rastetter EB, Ryan MG, Shaver GR, Melillo JM, NadelhofferKJ, Hobbie JE, Aber JD (1991) A generalbiogeochemical model describing the responses of theC-cycle and N-cycle in terrestrial ecosystems tochanges in CO 2 , climate, and N-deposition. TreePhysiol 9:101–126Reiners WA, Driese KL (2001) The propagation ofecological influences through heterogeneous environmentalspace. Bioscience 51:939–950Roberts DW (1996a) Landscape vegetation modelling withvital attributes and fuzzy systems theory. Ecol Model90:175–184Roberts DW (1996b) Modelling forest dynamics with vitalattributes and fuzzy systems theory. Ecol Model90:161–173Running SW, Gower ST (1991) FOREST-BGC, a generalmodel of forest ecosystem processes for regionalapplications. II. Dynamic carbon allocation andnitrogen budgets. Tree Physiol 9:147–160Rykiel EJ Jr (1996) Testing ecological models: the meaningof validation. Ecol Model 90:229–244Saxe H, Cannell MGR, Johnsen B, Ryan MG, Vourlitis G(2001) Tree and forest functioning in response toglobal warming. New Phytol 149:369–399Scheller RM, Mladenoff DJ (2004) A forest growth andbiomass module for a landscape simulation model,LANDIS: design, validation, and application. EcolModel 180:211–229Scheller RM, Mladenoff DJ (2005) A spatially interactivesimulation of climate change, harvesting, wind, andtree species migration and projected changes to forestcomposition and biomass in northern Wisconsin,USA. Glob Change Biol 11:307–321Scheller RM, Mladenoff DJ, Crow TR, Sickley TS (2005)Simulating the effects of fire reintroduction versuscontinued suppression on forest composition and landscapestructure in the Boundary Waters Canoe Area,northern Minnesota (USA). Ecosystems 8:396–411Scheller RM, Domingo JB, Sturtevant BR, Williams JS,Rudy A, Gustafson EJ, Mladenoff DJ (in press)Design, development, and application of LANDIS-II,a spatial landscape simulation model with flexible andtemporal resolution. Ecol ModelSchulte LA, Mladenoff DJ, Nordheim EV (2002) Quantitativeclassification of a historic northern Wisconsin(USA) landscape: mapping forests at regional scales.Can J For Res 32: 1616–1638Sequeira RA, Olson RL, Mckinion JM (1997) Implementinggeneric, object-oriented models in biology. EcolModel 94:17–31Shugart HH (1998) Terrestrial ecosystems in changingenvironments. Cambridge University Press, Cambridge,UKSoares-Filho BS, Cerqueira GC, Pennachin CL (2002)DINAMICA—a stochastic cellular automata modeldesigned to simulate the landscape dynamics in anAmazonian colonization frontier. Ecol Model154:217–235STATSGO (1994) State Soil Geographic (STATSGO)Data Base. Data use information. Report number1492. US Dept. of Agriculture National Cartographyand GIS Center, Fort Worth, TX, USASturtevant BR, Gustafson EJ, Li W, He HS (2004)Modeling biological disturbances in LANDIS: amodule description and demonstration using sprucebudworm. Ecol Model 180:153–174Syphard AD, Franklin J, Keeley JE (2006) Simulating theeffects of frequent fire on southern California coastalshrublands. Ecol Appl 16:1744–1756Thonicke K, Venevsky S, Sitch S, Cramer W (2001) Therole of fire disturbance for global vegetation dynamics:coupling fire into a dynamic global vegetationmodel. Glob Ecol Biogeogr 10:661–677Turner MG, Gardner RH, Dale VH, O’Neill RV (1989)Predicting the spread of disturbance across heterogeneouslandscapes. Oikos 55:121–129Turner MG, Gardner RH, O’Neill RV (1995) Ecologicaldynamics at broad scales. Ecosystems and landscapes.BioScience S-29:443–449123

Landscape EcolTurner MG, Wear DN, Flamm RO (1996) Land ownershipand land-cover change in the southern Appalachianhighlands and the Olympic peninsula. Ecol Appl6:1150–1172Urban DL, Shugart HH (1992) Individual-based models offorest succession. In: Glenn-Lewin DC, Peet RK,Veblen TT (eds) Plant succession: theory and prediction.Chapman & Hall, London, UK, pp 249–292Urban DL, Bonan GB, Smith TM, Shugart HH (1991)Spatial application of gap models. For Ecol Manage42: 95–110Urban DL, Acevedo MF, Garman SL (1999) Scaling finescaleprocesses to large-scale patterns using modelsderived from models: meta-models. In: Mladenoff DJ,Baker WL (eds) Spatial modeling of forest landscapechange. Cambridge University Press, Cambridge, UK,pp 70–98VEMAP Members (1995) Vegetation/ecosystem modelingand analysis project: comparing biogeography andbiogeochemistry models in a continental-scale study ofterrestrial ecosystem responses to climate change andCO 2 doubling. Glob Biogeochem Cycles 9:407–437Wallin DO, Swanson FJ, Marks B (1994) Landscapepattern response to changes in pattern generationrules: land-use legacies in forestry. Ecol Appl4(3):569–580Whittaker RH, Bormann FH, Likens GE, Siccama TG(1974) The Hubbard Brook ecosystem study: forestbiomass and production. Ecol Monogr 44:233–252Wolter PT, Mladenoff DJ, Host GE, Crow TR (1995)Improved forest classification in the northern LakeStates using multi-temporal landsat imagery. PhotogrammetricEng Remote Sens 61:1129–1143Woodbury PB, Beloin RM, Swaney DP, Gollands BE,Weinstein DA (2002) Using the ECLPSS softwareenvironment to build a spatially explicit componentbasedmodel of ozone effects on forest ecosystems.Ecol Model 150:211–238Yemshanov D, Perera AH (2002) A spatially explicitstochastic model to simulate boreal forest covertransitions: general structure and properties. EcolModel 150:189–209123