10.07.2015 Views

Dispersal and transient dynamics in metapopulations - IEB

Dispersal and transient dynamics in metapopulations - IEB

Dispersal and transient dynamics in metapopulations - IEB

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Ecology Letters, (2003) 6: 197–204REPORT<strong>Dispersal</strong> <strong>and</strong> <strong>transient</strong> <strong>dynamics</strong> <strong>in</strong> <strong>metapopulations</strong>Fabio A. Labra, Nelson A. Lagos<strong>and</strong> Pablo A. Marquet*Center for Advanced Studies <strong>in</strong>Ecology <strong>and</strong> Biodiversity, <strong>and</strong>Departamento de Ecología,Facultad de Ciencias Biológicas,Pontificia Universidad Católicade Chile, Casilla 114-D,Santiago, Chile*Correspondence: E-mail:pmarquet@genes.bio.puc.clAbstractRecent studies of spatially explicit metapopulation models have shown the existence ofcomplex <strong>transient</strong> behaviour (super<strong>transient</strong>s <strong>and</strong> meso<strong>transient</strong>s) characterized byspontaneous changes <strong>in</strong> the system’s <strong>dynamics</strong> after thous<strong>and</strong>s or hundreds ofgenerations, respectively. Their detection <strong>in</strong> simple ecological models has been taken asevidence that <strong>transient</strong> <strong>dynamics</strong> may be common <strong>in</strong> nature. In this study, we explore thegenerality of these phenomena <strong>in</strong> a simple one-dimensional spatially explicit metapopulationmodel. We <strong>in</strong>vestigate how frequently super<strong>transient</strong> behaviour emerges <strong>in</strong>relation to the shape <strong>and</strong> type of the dispersal kernel used (normal <strong>and</strong> Laplace), systemsize, boundary conditions <strong>and</strong> how sensitive they are to <strong>in</strong>itial conditions. Our resultsshow that super<strong>transient</strong>s are rare, are heavily affected by <strong>in</strong>itial conditions <strong>and</strong> occur fora small set of dispersal parameter values, which vary accord<strong>in</strong>g to kernel type, systemsize, <strong>and</strong> boundary conditions. Similarly, meso<strong>transient</strong>s emerge over a very narrowrange of dispersal parameter values <strong>and</strong> are rare under all circumstances. Thus, <strong>transient</strong><strong>dynamics</strong> are not likely to be either common or widespread <strong>in</strong> simple models ofecological systems.KeywordsTransient <strong>dynamics</strong>, dispersal kernel, spatial model, coupled map lattice, <strong>metapopulations</strong>.Ecology Letters (2003) 6: 197–204INTRODUCTIONThe study of spatial population <strong>dynamics</strong> <strong>and</strong> persistencehas received renewed attention <strong>in</strong> the past few years (Maurer1994; Tilman & Kareiva 1997; Bascompte & Solé 1998a,b;Turch<strong>in</strong> 1998; Keymer et al. 2000), <strong>and</strong> has been the sourceof many new <strong>in</strong>sights to ecological phenomena (Paradis et al.2000; Williams & Liebhold 2000). One of the most popularforms of spatial population models considers <strong>metapopulations</strong>,or ensembles of local populations coupled bydispersal. These simple models, known as coupled maplattices (CML) (Kaneko 1989, 1998) predict a range ofspatiotemporal patterns, some of which have been found <strong>in</strong>nature (Maron & Harrison 1997; Ranta et al. 1997; Lamb<strong>in</strong>et al. 1998; Solé & Bascompte 1998; Blasius et al. 1999), <strong>and</strong>have been extensively used <strong>in</strong> the study of complex systems(Kaneko & Tsuda 2000). Theoretical studies of CMLs havesuggested that the population <strong>dynamics</strong> we observe <strong>in</strong>nature may well correspond to a <strong>transient</strong> dynamicalbehaviour, which shows spontaneous changes fromapparent chaotic <strong>dynamics</strong> to cyclic behaviour <strong>and</strong> viceversa occurr<strong>in</strong>g at time scales longer than a typical ecologicaltime series (Hast<strong>in</strong>gs & Higg<strong>in</strong>s 1994). Recently, Saraviaet al. (2000) noted that CML may show sudden changes <strong>in</strong>behaviour at shorter temporal scales, similar to that ofecological studies, <strong>and</strong> have suggested that these mesoscalechanges <strong>in</strong> <strong>dynamics</strong> (termed Ômeso<strong>transient</strong>sÕ by theauthors) may be of greater relevance than longer <strong>transient</strong>behaviour. These claims contrast with the traditionalemphasis on the analysis of long-term solutions (asymptoticstates) <strong>in</strong> the study of ecological models (Hast<strong>in</strong>gs &Higg<strong>in</strong>s 1994; Hast<strong>in</strong>gs 1998; Earn & Rohani 1999). If<strong>transient</strong> behaviour is common <strong>in</strong> nature, then ecologicalsystems might often be <strong>in</strong> a <strong>transient</strong> state, <strong>and</strong> changes <strong>in</strong>their <strong>dynamics</strong> might be compounded by <strong>in</strong>tr<strong>in</strong>sic dynamicaleffects <strong>and</strong> also by changes <strong>in</strong> the environment where<strong>in</strong> theyare embedded.The first step towards determ<strong>in</strong><strong>in</strong>g whether we can expectsuch complex <strong>transient</strong> behaviour to be common <strong>in</strong> natureis the assessment of their generality <strong>in</strong> model ecologicalsystems. It is known that super<strong>transient</strong> <strong>dynamics</strong> are likelyto appear when coupled local populations are <strong>in</strong> a chaoticregime (Kaneko 1989; Hast<strong>in</strong>gs & Higg<strong>in</strong>s 1994; Ruxton &Doebeli 1996) <strong>and</strong> that average <strong>transient</strong> length <strong>in</strong>creaseswith system size <strong>and</strong> the degree of global coupl<strong>in</strong>g betweenpopulations (Kaneko 1990, 1998; Hast<strong>in</strong>gs 1998). InÓ2003 Blackwell Publish<strong>in</strong>g Ltd/CNRS


198 F. A. Labra, N. A. Lagos <strong>and</strong> P. A. Marquetecological models, however, coupl<strong>in</strong>g has only beenmodelled assum<strong>in</strong>g a normal (gaussian) dispersal kernel<strong>and</strong> for a restricted set of dispersal <strong>in</strong>tensities <strong>and</strong> spatialranges (Hast<strong>in</strong>gs & Higg<strong>in</strong>s 1994; Ruxton & Doebeli 1996;but see Saravia et al. 2000). However, many organisms, suchas plants or aquatic <strong>in</strong>vertebrates <strong>and</strong> vertebrates, might bebetter represented by a fat-tailed, leptokurtic kernel such asLaplace dispersal kernel (Neubert et al. 1995; Clark et al.1999; Fraser et al. 2001), where settlement is probabilistic(i.e. there is a failure rate or hazard function).In this paper, we explore the generality of bothsuper<strong>transient</strong> <strong>and</strong> meso<strong>transient</strong> <strong>dynamics</strong> <strong>in</strong> terms ofdispersal (kernel type <strong>and</strong> coupl<strong>in</strong>g), system size, boundaryconditions <strong>and</strong> sensitivity to <strong>in</strong>itial conditions. We do this byanalys<strong>in</strong>g a model system for which both complex <strong>transient</strong>behaviours has been reported. Our results show thatsuper<strong>transient</strong> <strong>dynamics</strong> are a rare phenomenon emerg<strong>in</strong>gfor a restricted set of dispersal parameter values, whichnonetheless vary between kernels, <strong>and</strong> is affected by systemsize, boundary conditions <strong>and</strong> depends heavily on <strong>in</strong>itialconditions. Meso<strong>transient</strong>s on the other h<strong>and</strong> are seen tooccur for a smaller range of dispersal parameter values. Wediscuss the implications of these results both for theoretical<strong>and</strong> empirical population <strong>dynamics</strong>.MATERIALS AND METHODSFollow<strong>in</strong>g Hast<strong>in</strong>gs (1998), we modelled a small set of Mdiscrete local populations, of a semelparous species,distributed along a l<strong>in</strong>early structured environment, suchas may be found along a stretch of coastl<strong>in</strong>e. In each ofthese populations, reproduction occurs synchronously <strong>in</strong> adensity-dependent fashion. At the start of a given generationt, population size <strong>in</strong> the ith population is denoted N [t,i] . Thereproductive events generate a larval stage, L, whichdisperses among the local populations. Abundance of larvae<strong>in</strong> each population L [t,i] , after reproduction <strong>and</strong> before theydisperse, is given by (see Maynard-Smith & Slatk<strong>in</strong> 1973;Ruxton & Doebeli 1996),L ½t;iŠ ¼ N ½t;iŠ f ðN ½t;iŠ Þ¼kN ½t;iŠ1 þ aN½t;iŠbð1ÞThe parameter k is the growth rate, a is the <strong>in</strong>verse of theenvironmental carry<strong>in</strong>g capacity <strong>and</strong> b describes the <strong>in</strong>tensityof density dependence (for further discussion of this modelsee Hassell 1975; Bellows 1981; Doebeli 1995). All larvae areproduced synchronously <strong>and</strong> settlement is simultaneousthroughout the metapopulation. Larvae are redistributed <strong>in</strong> adensity-<strong>in</strong>dependent fashion accord<strong>in</strong>g either to Laplace ornormal kernel, which reflect different dispersal strategies oflarvae. In comparison with the normal kernel, larvae underLaplace kernel have at the same time a higher probability ofrema<strong>in</strong><strong>in</strong>g at the source <strong>and</strong> of dispers<strong>in</strong>g over a longerrange (Fig. 1). For the normal dispersal kernel (Neubertet al. 1995), the proportion of the larvae on patch i thatmove to patch j is:P ½i;jŠ ¼ p 12 ffiffiffiffiffi e4r 1 ði jÞ2 ð2ÞprIn the case of Laplace dispersal kernel (Neubert et al. 1995),the proportion of the larvae on patch i which moves topatch j is:P ½i;jŠ ¼ 12r e1 r jði jÞj ð3ÞFor both dispersal kernels the shape of the dispersal curve,which is controlled by the parameter r, gives the <strong>in</strong>tensity<strong>and</strong> extent of dispersal: the smaller the parameter, the largerthe proportion of <strong>in</strong>dividuals that will rema<strong>in</strong> <strong>in</strong> the parentalhabitat <strong>and</strong> more local populations will rema<strong>in</strong> unconnected(Fig. 1). High values of r will result <strong>in</strong> greater spatial spreadof the <strong>in</strong>dividuals with lower <strong>in</strong>tensity.The model assumes non-overlapp<strong>in</strong>g generations, <strong>and</strong> soall adult <strong>in</strong>dividuals die at the end of the reproductiveperiod. Local population size at the next generation is theresult of the recruitment of all the larvae dispersed from allFigure 1 <strong>Dispersal</strong> kernels used <strong>in</strong> this study. Note the leptokurticshape of the Laplace distribution. The figure shows the proportionof propagules dispersed to different distances from a central focalpopulation (site 25) for r ¼ 1, r ¼ 5 <strong>and</strong> r ¼ 10 (a) normaldispersal kernel <strong>and</strong> (b) Laplace dispersal kernel. Note that for asmall system size (M ¼ 10) large values of dispersal result <strong>in</strong> loss ofpropagules from the system under dissipative boundary conditionswhile periodic boundaries result <strong>in</strong> essentially a more platikurticredistribution kernel.σσσÓ2003 Blackwell Publish<strong>in</strong>g Ltd/CNRS


Transient <strong>dynamics</strong> <strong>in</strong> <strong>metapopulations</strong> 199subpopulations. The number of <strong>in</strong>dividuals <strong>in</strong> each population<strong>in</strong> the next generation is given by the sum of thelarvae dispersed over the entire ensemble to that particularpopulation.N ½tþ1;iŠ ¼ XMj¼0L ½t;jŠ p ½j;iŠð4ÞOn the other h<strong>and</strong>, total population abundance for themetapopulation at generation t is given by the sum of<strong>in</strong>dividuals present <strong>in</strong> the metapopulation after dispersal <strong>and</strong>prior to reproduction. Previous studies (Hast<strong>in</strong>gs & Higg<strong>in</strong>s1994; Ruxton & Doebeli 1996) have assumed dissipativeboundary conditions (but see Hast<strong>in</strong>gs 1998). This meansthat a fraction of the larvae disperse beyond the edges of theensemble <strong>and</strong> do not <strong>in</strong>fluence the <strong>dynamics</strong> of the system.It may be argued that this loss of <strong>in</strong>dividuals may lower thelocal growth rate, thus lower<strong>in</strong>g the likelihood of observ<strong>in</strong>g<strong>transient</strong> <strong>dynamics</strong>. To assess the importance of larval losswe compared, for each dispersal kernel, the effect ofdissipative vs. periodic boundary conditions. To allow easiercomparison of the results we set the parameters a ¼ 1,b ¼ 4.8 <strong>and</strong> k ¼ 7, as used by Ruxton & Doebeli (1996).Under these parameter values, isolated local populationsshow chaotic <strong>dynamics</strong>. Initial local population abundances(i.e. <strong>in</strong>itial conditions) were drawn <strong>in</strong>dependently from ar<strong>and</strong>om uniform distribution between 0 <strong>and</strong> 4. In eachsimulation, the model was iterated for 10 000 generations,<strong>and</strong> the <strong>dynamics</strong> exam<strong>in</strong>ed for the occurrence of <strong>transient</strong><strong>dynamics</strong> (Hast<strong>in</strong>gs & Higg<strong>in</strong>s 1994). This procedure wasperformed for both dispersal kernels us<strong>in</strong>g two ranges of thedispersal parameter (r), from 0.01 to 5 <strong>and</strong> from 5.01 to10.00 (with <strong>in</strong>crements of 0.01 with each range) <strong>and</strong> for twodifferent system sizes M ¼ 10 <strong>and</strong> M ¼ 50. To assess theeffect of each factor <strong>in</strong> isolation <strong>and</strong> the <strong>in</strong>teraction amongfactors (i.e. kernel, boundary, r, <strong>and</strong> lattice size) on theemergence of <strong>transient</strong> behaviour, we used a full factorialexperimental design (Qu<strong>in</strong>n & Keough 2002) consider<strong>in</strong>gthe two levels of r (r £ 5 <strong>and</strong> r > 5). All factorialcomb<strong>in</strong>ations of factors resulted <strong>in</strong> 16 different scenarios.Twenty replicates were run for each of them.To evaluate the effect of <strong>in</strong>itial conditions of localabundance on the observed results, we also carried out 50simulations with different <strong>in</strong>itial conditions for a subset ofparameter values (r), uniformly spread between 0.01 <strong>and</strong>10. We report the proportion of the 50 replicates for eachvalue of s, which exhibited <strong>transient</strong> behaviour undernormal <strong>and</strong> Laplace kernels <strong>and</strong> for both boundaryconditions.Follow<strong>in</strong>g Hast<strong>in</strong>gs & Higg<strong>in</strong>s (1994) <strong>and</strong> Saravia et al.(2000), <strong>transient</strong> <strong>dynamics</strong> were identified by not<strong>in</strong>g theexistence of abrupt <strong>and</strong> spontaneous changes betweendifferent dynamic regimes with time scales of hundreds ofiterations (meso<strong>transient</strong>s) <strong>and</strong> with a time scale greater thanone thous<strong>and</strong> iterations (super<strong>transient</strong>s). Detection of thesechanges was carried out by calculat<strong>in</strong>g the ratio of variances(RV) between two adjacent halves of a slid<strong>in</strong>g w<strong>in</strong>dow ofsize W (Saravia et al. 2000). Both RV <strong>and</strong> W weredeterm<strong>in</strong>ed empirically to allow detection of super<strong>transient</strong>s(RV ¼ 10, W ¼ 1000) <strong>and</strong> meso<strong>transient</strong>s (RV ¼ 20,W ¼ 100). For a given simulation, <strong>transient</strong> length wasdef<strong>in</strong>ed by the time <strong>in</strong>terval at which the maximum value ofRV was observed. The changes <strong>in</strong> metapopulation <strong>dynamics</strong>were corroborated by the <strong>in</strong>spection of the temporal<strong>dynamics</strong> <strong>and</strong> return maps (Hast<strong>in</strong>g & Higg<strong>in</strong>s 1994; Saraviaet al. 2000).RESULTSIn agreement with previous studies, we observed super<strong>transient</strong><strong>dynamics</strong> <strong>in</strong> this simple metapopulation model(Fig. 2a). These changes occurred once or several times <strong>in</strong>Figure 2 Example of <strong>dynamics</strong> for the total metapopulationabundance observed for: (a) normal dispersal kernel with periodicboundaries <strong>and</strong> r ¼ 9.01. (b) Laplace dispersal kernel withdissipative boundaries <strong>and</strong> r ¼ 8.44. (c) A subset of the <strong>dynamics</strong>shown <strong>in</strong> (b). In all simulations only the dispersal parameter value<strong>and</strong> boundary conditions were changed <strong>and</strong> the follow<strong>in</strong>gparameter values were used: a ¼ 1, b ¼ 4.8, k ¼ 7, M ¼ 50.Ó2003 Blackwell Publish<strong>in</strong>g Ltd/CNRS


200 F. A. Labra, N. A. Lagos <strong>and</strong> P. A. Marquetthe time series (Fig. 2b). Similarly, meso<strong>transient</strong>s were alsoobserved (Fig. 2c). However, despite us<strong>in</strong>g local chaoticconditions of density dependence (k ¼ 7), we observed<strong>transient</strong>s <strong>in</strong> a small proportion of the total number of thesimulations. For simulations us<strong>in</strong>g M ¼ 50 <strong>and</strong> a normaldispersal kernel, we found super<strong>transient</strong>s <strong>in</strong> 10.9 <strong>and</strong> 2.6%of the simulations across the range of r-values used <strong>and</strong> forperiodic <strong>and</strong> dissipative boundary conditions, respectively,while for Laplace kernel we observed 3.7 <strong>and</strong> 4.9% forperiodic <strong>and</strong> dissipative boundary conditions. For a smallersystem size (M ¼ 10), the normal kernel showed super<strong>transient</strong>s<strong>in</strong> 3 <strong>and</strong> 13% of the simulations (periodic <strong>and</strong>dissipative boundary conditions, respectively) while Laplacekernel showed 0.8 <strong>and</strong> 13% (periodic <strong>and</strong> dissipativeboundary conditions, respectively). In contrast, meso<strong>transient</strong>soccurred <strong>in</strong>


Transient <strong>dynamics</strong> <strong>in</strong> <strong>metapopulations</strong> 201A system is highly dependent on <strong>in</strong>itial conditions if for aspecific set of parameter values, simulations orig<strong>in</strong>at<strong>in</strong>g withdifferent <strong>in</strong>itial numbers of <strong>in</strong>dividuals at different placeslead the system to different types of <strong>dynamics</strong>. If this werethe case, then we would f<strong>in</strong>d that for the same set ofparameter values, some of the trajectories show <strong>transient</strong>swhile some do not, hence the results shown <strong>in</strong> Figs 3 <strong>and</strong> 4might not accurately reflect the effect of parameter valuesupon the emergence of <strong>transient</strong> <strong>dynamics</strong>. As shown <strong>in</strong>Fig. 5, <strong>in</strong> most cases the proportion of simulations thatshowed either super<strong>transient</strong>s or meso<strong>transient</strong>s, for thesame parameter values but for chang<strong>in</strong>g <strong>in</strong>itial conditions,was highly variable <strong>and</strong> heavily affected by the value of r.For a large system size (M ¼ 50) <strong>and</strong> for both dispersalkernels, super<strong>transient</strong>s seemed to be more robust underperiodic boundary conditions than under dissipative boundaryconditions (Figs 5a,b) follow<strong>in</strong>g a patterns similar to thatshown <strong>in</strong> Fig. 3(a,b). However, <strong>in</strong> general the averageproportion of simulations show<strong>in</strong>g super<strong>transient</strong>s was low,rang<strong>in</strong>g between 0.16 <strong>and</strong> 0.05 for the periodic boundaryconditions under the normal <strong>and</strong> Laplace dispersal kernels,respectively. Under dissipative boundary conditions thisrange was between 0.02 <strong>and</strong> 0.01 for normal <strong>and</strong> Laplacekernels, respectively. Meso<strong>transient</strong> <strong>dynamics</strong>, on the otherh<strong>and</strong>, are more likely to be observed for <strong>in</strong>termediate valuesof r rang<strong>in</strong>g between 2 <strong>and</strong> 6-under periodic boundaryconditions, <strong>and</strong> for values of r rang<strong>in</strong>g between 4 <strong>and</strong> 7(a)(b)Figure 5 Effect of <strong>in</strong>itial conditions on the occurrence ofsuper<strong>transient</strong>s <strong>and</strong> meso<strong>transient</strong>s as a function of r. (a) <strong>and</strong>(b) show the proportion of 50 simulations with different <strong>in</strong>itialconditions that show super<strong>transient</strong>s for normal <strong>and</strong> Laplacekernels, respectively. (c) <strong>and</strong> (d) show the proportion ofsimulations that show meso<strong>transient</strong>s <strong>dynamics</strong> for normal <strong>and</strong>Laplace kernels, respectively. Dissipative boundary conditions arerepresented by open symbols, while filled symbols representperiodic boundaries. Parameter value as <strong>in</strong> Figure 3.(c)(d)under dissipative boundary conditions <strong>and</strong> normal dispersal(Fig. 5c). On the other h<strong>and</strong>, for Laplace dispersalmeso<strong>transient</strong>s were more common for large values of r,particularly under dissipative boundary conditions (Fig. 5d).In general, the average proportion of simulations show<strong>in</strong>gmeso<strong>transient</strong>s was low (0.09) for the periodic boundaryconditions under the normal <strong>and</strong> Laplace dispersal kernels.Under dissipative boundary conditions, the average proportionof simulations show<strong>in</strong>g meso<strong>transient</strong>s ranged between0.12 <strong>and</strong> 0.05 for Laplace <strong>and</strong> normal kernels, respectively.Although these results cannot be readily applied to assessthe uncerta<strong>in</strong>ty of the results reported <strong>in</strong> Figs 3 <strong>and</strong> 4,because of the different values <strong>and</strong> <strong>in</strong>tervals of r used <strong>in</strong>both analysis, they consistently show that for a given set ofparameter values the emergence of <strong>transient</strong> behaviour ishighly dependent on <strong>in</strong>itial conditions. A similar result wasobserved for a smaller system size (M ¼ 10, not shown).In summary, our results show that both super<strong>transient</strong>s<strong>and</strong> meso<strong>transient</strong>s <strong>dynamics</strong> are rare; their emergencedepends strongly on the kernel type, dispersal, boundaryconditions <strong>and</strong> system size, <strong>and</strong> the <strong>in</strong>teraction among thesefactors. Furthermore, we show that <strong>transient</strong> <strong>dynamics</strong> arestrongly affected by <strong>in</strong>itial conditions.DISCUSSIONIt is known that long <strong>transient</strong>s can occur <strong>in</strong> spatialecological models with multiple attractors <strong>and</strong> hencechaotic saddles (Kaneko 1990; Hast<strong>in</strong>gs & Higg<strong>in</strong>s 1994;Hast<strong>in</strong>gs 1998) even <strong>in</strong> the presence of noise (Hast<strong>in</strong>gs1998). The existence of super<strong>transient</strong>s rema<strong>in</strong>ed amathematical eccentricity until Hast<strong>in</strong>gs <strong>and</strong> coworkers(Hast<strong>in</strong>gs & Higg<strong>in</strong>s 1994; Hast<strong>in</strong>gs 1998, 2001) arguedthat <strong>transient</strong> dynamic behaviour is widespread <strong>in</strong> nature,suggest<strong>in</strong>g that its emergence <strong>in</strong> spatial ecological models islikely to be the rule rather than the exception. Further, thefact that <strong>transient</strong> behaviour usually lasts for manygenerations implies that the <strong>transient</strong> state of the systemcould be potentially more important than its f<strong>in</strong>al attractor(Hast<strong>in</strong>gs & Higg<strong>in</strong>s 1994). Thus, if <strong>transient</strong>s are the rulerather than the exception for ecological systems, thentraditional theoretical models based on asymptotic equilibriumbehaviour might be of little relevance as tools forunderst<strong>and</strong><strong>in</strong>g.In this paper, we showed that the emergence ofsuper<strong>transient</strong> <strong>dynamics</strong> is a rare event, which dependsstrongly on the type, extent <strong>and</strong> <strong>in</strong>tensity of dispersal, as wellas on the system size <strong>and</strong> boundary conditions. Ruxton &Doebeli (1996) first made the po<strong>in</strong>t that the shape of thedispersal kernel (as measured by r) affected the emergenceof super<strong>transient</strong>s <strong>in</strong> simple spatial models, but providedno systematic assessment of its effect, <strong>and</strong> restrictedtheir analysis to a normal kernel. Similarly, Saravia et al.Ó2003 Blackwell Publish<strong>in</strong>g Ltd/CNRS


202 F. A. Labra, N. A. Lagos <strong>and</strong> P. A. Marquet(2000) po<strong>in</strong>t out that super<strong>transient</strong>s required a particularcomb<strong>in</strong>ation of k, b <strong>and</strong> r, to emerge but did not explicitlytest for the effect of dispersal <strong>and</strong> also restricted theiranalysis to a normal dispersal kernel. Our analysis showsthat, contrary to Ruxton & DoebeliÔs (1996) f<strong>in</strong>d<strong>in</strong>gs,super<strong>transient</strong>s can emerge for high values of the dispersalparameter r, depend<strong>in</strong>g on the size of the system (Figs 3<strong>and</strong> 4). However, these super<strong>transient</strong>s may be highlysensitive to <strong>in</strong>itial conditions (Fig. 5), strengthen<strong>in</strong>g ourconclusion that they are not the rule <strong>in</strong> simple ecologicalmodels.Our results also po<strong>in</strong>t out the importance of boundaryconditions <strong>and</strong> system size for the emergence of <strong>transient</strong><strong>dynamics</strong>. It may be expected that loss of <strong>in</strong>dividualsthrough dissipative boundaries would result <strong>in</strong> a lower<strong>in</strong>g ofthe local growth rates <strong>and</strong> hence <strong>in</strong> the likelihood ofobserv<strong>in</strong>g super<strong>transient</strong>s. For a small system size, however,super<strong>transient</strong>s were more common precisely under thisscenario. Thus, loss of <strong>in</strong>dividuals through dispersal may stillallow super<strong>transient</strong>s to emerge. However, system size,boundary conditions <strong>and</strong> <strong>in</strong>itial conditions <strong>in</strong>teract <strong>in</strong>complex ways such that for a large system size (M ¼ 50)<strong>transient</strong> <strong>dynamics</strong> tend to be less affected by <strong>in</strong>itialconditions under periodic boundaries, while the reverse istrue when the system is smaller (M ¼ 10, F. Labra,unpublished results). In a recent paper (Hast<strong>in</strong>gs 2001), itis reported that super<strong>transient</strong>s occur more frequently whenthe <strong>in</strong>itial condition <strong>in</strong>cludes empty patches. The <strong>in</strong>clusionof empty sites <strong>in</strong> our simulations agree with this result.However, the observed <strong>in</strong>crease <strong>in</strong> <strong>transient</strong>s only occursunder dissipative conditions, <strong>and</strong> only for super<strong>transient</strong>sbut not for meso<strong>transient</strong>s. To assess the relative importanceof each of these factors is beyond the scope of thepresent paper <strong>and</strong> deserves further <strong>in</strong>vestigation. However,this complex dependency strengthens our po<strong>in</strong>t; theemergence of <strong>transient</strong> dynamic behaviour <strong>in</strong> simpleecological models is not a common phenomenon butoccurs for a restricted set of conditions. Saravia et al. (2000),us<strong>in</strong>g the same model we used, reached a similar conclusionafter relax<strong>in</strong>g the assumption of global synchrony <strong>in</strong>reproduction <strong>and</strong> dispersal. This implies that the set ofconditions that lead to <strong>transient</strong> <strong>dynamics</strong> are likely to beeven more restricted than we report here. Although for thesake of comparability we have followed previous authors <strong>in</strong>study<strong>in</strong>g a one-dimensional cha<strong>in</strong> of <strong>in</strong>teract<strong>in</strong>g populations(e.g. Hast<strong>in</strong>gs & Higg<strong>in</strong>s 1994; Ruxton & Doebeli 1996;Saravia et al. 2000) it is likely that system dimensionality willalso have an effect on the emergence of <strong>transient</strong> behaviour.However, further <strong>in</strong>vestigations on two-dimensional systemsare necessary.Most previous studies have not reported an exhaustiveexploration of the parameters that may result <strong>in</strong> <strong>transient</strong><strong>dynamics</strong>. We found that under both dispersal kernels,super<strong>transient</strong> <strong>dynamics</strong> occur <strong>in</strong> very low frequency <strong>in</strong> spiteof strong local nonl<strong>in</strong>earity. From this perspective, <strong>and</strong>consider<strong>in</strong>g that its emergence also depends, as <strong>in</strong> our case,on the <strong>in</strong>itial conditions used <strong>in</strong> the simulations (see alsoRuxton & Doebeli 1996) <strong>and</strong> the value of the parameters k,b <strong>and</strong> r, it can be argued that super<strong>transient</strong>s, contrary toHast<strong>in</strong>gs & Higg<strong>in</strong>s (1994) <strong>and</strong> Hast<strong>in</strong>gs (1998, 2001), mightnot be a common feature of spatial models, let alone realecological systems <strong>and</strong> especially so, if chaotic <strong>dynamics</strong> arenot common <strong>in</strong> nature (see Rai & Schaffer 2001; below).The range of dispersal we have explored here ranges fromlocal coupl<strong>in</strong>g, when r is very low, to nearly global coupl<strong>in</strong>gwhen r is very high <strong>and</strong> periodic boundaries are imposed onthe system. In agreement with Hast<strong>in</strong>gs (1998), who used asystem of size M ¼ 10, we found that super<strong>transient</strong>s wereharder to f<strong>in</strong>d when dispersal conditions tended to besimilar to global coupl<strong>in</strong>g (large r-values). However, this istrue only for a small system size (M ¼ 10) (compare Figs 3<strong>and</strong> 4) <strong>and</strong> not all small <strong>and</strong> <strong>in</strong>termediate values of r result<strong>in</strong> super<strong>transient</strong>s, <strong>and</strong> of those that do, several are sensitiveto <strong>in</strong>itial conditions.It should be noted, however, that we observed super<strong>transient</strong>smuch more frequently than Saravia et al. (2000)did. This discrepancy arises because their study encompassesparameter comb<strong>in</strong>ations that span regions of stable,cyclic <strong>and</strong> chaotic <strong>dynamics</strong>. However, super<strong>transient</strong>s ariseonly <strong>in</strong> the latter regime <strong>and</strong> for strong nonl<strong>in</strong>earity(Kaneko 1989, 1990; Kaneko & Tsuda 2000). Simulationscarried out with a different set of parameters (a ¼ 0.5,b ¼ 30 <strong>and</strong> k ¼ 2), which produce chaotic <strong>dynamics</strong>, failedto produce either of the two complex behaviours understudy when simulations were carried out for a large systemsize (M ¼ 50). On the other h<strong>and</strong>, for a second parametersett<strong>in</strong>g (M ¼ 50, a ¼ 0.5, b ¼ 15 <strong>and</strong> k ¼ 7), the occurrenceof super<strong>transient</strong>s <strong>in</strong>creased under both kernels to 8.2<strong>and</strong> 19.5% for normal dispersal (periodic <strong>and</strong> dissipativeboundary conditions) while for Laplace kernel they occurred<strong>in</strong> 19.5 <strong>and</strong> 6.6% (periodic <strong>and</strong> dissipative boundaryconditions, respectively). If super<strong>transient</strong>s occur only forfully developed spatiotemporal chaos, then the questionshould be how common are they <strong>in</strong> this regime. As wediscussed above, they are not very common.Meso<strong>transient</strong>s, on the other h<strong>and</strong>, were seen <strong>in</strong> lowfrequency, <strong>and</strong> emerged over a narrower range of dispersalparameter values, especially for the large system. For thetwo additional parameter sett<strong>in</strong>gs we explored, meso<strong>transient</strong>swere completely absent <strong>in</strong> the first sett<strong>in</strong>g, while forthe second parameter sett<strong>in</strong>g they were found <strong>in</strong>


Transient <strong>dynamics</strong> <strong>in</strong> <strong>metapopulations</strong> 203<strong>in</strong>termittent chaos (Doebeli 1994; Cavalieri & Koçak 1995;Gavrilets & Hast<strong>in</strong>gs 1995; Strogatz 1997) <strong>and</strong> spatiotemporal<strong>in</strong>termittency (Kaneko 1985, 1989). We are not awareof any formal attempt to ascerta<strong>in</strong> if meso<strong>transient</strong> <strong>dynamics</strong>are <strong>in</strong> fact <strong>in</strong>termittent chaos or spatiotemporal <strong>in</strong>termittency.If it turns out to be the same to either of them, thenits prevalence <strong>and</strong> relevance <strong>in</strong> spatial ecological modelsmight be determ<strong>in</strong>ed more clearly.In conclusion, we show that super<strong>transient</strong> <strong>and</strong> meso<strong>transient</strong><strong>dynamics</strong> are not likely to be either common orwidespread <strong>in</strong> ecological systems be<strong>in</strong>g heavily affected bydispersal <strong>in</strong>tensity, boundary conditions, system size <strong>and</strong><strong>in</strong>itial conditions.ACKNOWLEDGEMENTSF.L. <strong>and</strong> N.L. acknowledge a doctoral scholarship providedby Conicyt. F.L. would like to thank I. Labra, I. Rodriguez<strong>and</strong> C. Huerta for their support <strong>and</strong> encouragement. P.M.acknowledges the f<strong>in</strong>ancial support of grants FONDECYT1990144 <strong>and</strong> FONDAP-FONDECYT 1501–0001. J. Velasco-Hernández,provided cogent criticisms, <strong>and</strong> suggestionsto an early draft of this manuscript. We speciallythank Michael Hochberg, Mercedes Pascual <strong>and</strong> fiveanonymous reviewers for their constructive criticisms <strong>and</strong>for mak<strong>in</strong>g us to realize that super<strong>transient</strong>s, far frombe<strong>in</strong>g a theoretical curiosity, can also be a real lifeexperience.REFERENCESBascompte, J. & Solé, R. (1998a). Spatiotemporal patterns <strong>in</strong> nature.Trends Ecol. Evol., 13, 173–174.Bascompte, J. & Solé, R. (1998b). Model<strong>in</strong>g Spatiotemporal Dynamics <strong>in</strong>Ecology. Spr<strong>in</strong>ger Verlag, Berl<strong>in</strong>.Bellows, T.S. (1981). The descriptive properties of some modelsfor density dependence. J. Anim. Ecol., 50, 139–156.Blasius, B., Huppert, A. & Stone, L. (1999). Complex <strong>dynamics</strong> <strong>and</strong>phase synchronization <strong>in</strong> spatially extended ecological systems.Nature, 399, 354–359.Cavalieri, L.F. & Koçak, H. (1995). Intermittent transition betweenorder <strong>and</strong> chaos <strong>in</strong> an <strong>in</strong>sect pest population. J. Theor. Biol., 175,231–234.Clark, J.S., Silman, M., Kern, R., Mackl<strong>in</strong>, E. & HilleRisLambers,J. (1999). Seed dispersal near <strong>and</strong> far: patterns across temperate<strong>and</strong> tropical forests. Ecology, 80, 1475–1494.Doebeli, M. (1994). Intermittent chaos <strong>in</strong> population <strong>dynamics</strong>.J. Theor. Biol., 166, 325–330.Doebeli, M. (1995). <strong>Dispersal</strong> <strong>and</strong> <strong>dynamics</strong>. Theor. Pop. Biol., 47,82–106.Earn, D.J.D. & Rohani, P. (1999). Complex <strong>dynamics</strong> <strong>in</strong> ecology.Trends Ecol. Evol., 14, 43–44.Fraser, D.F., William, J.F., Daley, M.J., Le, A.N. & Skalski, G.(2001). Expla<strong>in</strong><strong>in</strong>g leptokurtic movement distributions:Intrapopulation variation <strong>in</strong> boldness <strong>and</strong> exploration. Am. Nat.,158, 124–135.Gavrilets, S. & Hast<strong>in</strong>gs, A. (1995). Intermittency <strong>and</strong> <strong>transient</strong>chaos from simple frequency-dependent selection. Proc. R. Soc.Lond. B, 261, 233–238.Hassell, M. (1975). Density-dependence <strong>in</strong> s<strong>in</strong>gle-species populations.J. Anim. Ecol., 44, 283–295.Hast<strong>in</strong>gs, A. (1998). Transients <strong>in</strong> spatial ecological models. In:Model<strong>in</strong>g Spatiotemporal Dynamics <strong>in</strong> Ecology (eds Bascompte, J.& Solé, R.). Spr<strong>in</strong>ger Verlag, Berl<strong>in</strong>, pp. 189–198.Hast<strong>in</strong>gs, A. (2001). Transient <strong>dynamics</strong> <strong>and</strong> persistence of ecologicalsystems. Ecol. Lett., 4, 215–220.Hast<strong>in</strong>gs, A. & Higg<strong>in</strong>s, K. (1994). Persistence of <strong>transient</strong>s <strong>in</strong>spatially structured ecological models. Science, 263, 1133–1136.Kaneko, K. (1985). Spatiotemporal <strong>in</strong>termittency <strong>in</strong> Coupled MapLattices. Prog. Theor. Phys., 74, 1033–1044.Kaneko, K. (1989). Pattern <strong>dynamics</strong> <strong>in</strong> spatiotemporal chaos.Physica D, 34, 1–41.Kaneko, K. (1990). Super<strong>transient</strong>s, spatiotemporal <strong>in</strong>termittency<strong>and</strong> stability of fully developed spatiotemporal chaos. Phys. Lett.A, 149, 105–112.Kaneko, K. (1998). Diversity, stability <strong>and</strong> meta<strong>dynamics</strong>: remarksfrom coupled map studies. Transients <strong>in</strong> spatial ecologicalmodels. In: Model<strong>in</strong>g Spatiotemporal Dynamics <strong>in</strong> Ecology (eds Bascompte,J. & Solé, R.). Spr<strong>in</strong>ger Verlag, Berl<strong>in</strong>, pp. 27–46.Kaneko, K. & Tsuda, I. (2000). Complex Systems: Chaos <strong>and</strong> Beyond.Spr<strong>in</strong>ger-Verlag, Berl<strong>in</strong>.Keymer, J.E., Marquet, P.A., Velasco-Hernández, J.X. & Lev<strong>in</strong>,S.A. (2000). Ext<strong>in</strong>ction thresholds <strong>and</strong> metapopulation persistence<strong>in</strong> dynamic l<strong>and</strong>scapes. Am. Nat., 156, 478–494.Lamb<strong>in</strong>, X., Elston, D.A., Petty, S.J. & MacK<strong>in</strong>non, J.L. (1998).Spatial asynchrony <strong>and</strong> periodic travel<strong>in</strong>g waves <strong>in</strong> cyclicpopulations of field voles. Proc. R. Soc. Lond. B, 265, 1491–1496.Maron, J.L. & Harrison, S. (1997). Spatial pattern formation <strong>in</strong> an<strong>in</strong>sect host-parasitoid system. Science, 278, 1619–1621.Maurer, B.A. (1994). Geographical Population Analysis: Tools for theAnalysis of Biodiversity. Blackwell Scientific Publications, Oxford.Maynard-Smith, J. & Slatk<strong>in</strong>, J. (1973). The stability of predatorpreysystems. Ecology, 54, 384–391.Neubert, M., Kot, M., Lewis, M. et al. (1995). <strong>Dispersal</strong> <strong>and</strong> patternformation <strong>in</strong> a discrete-time predator-prey model. Theor. Popul.Biol., 48, 7–43.Paradis, E., Baillie, S.R., Sutherl<strong>and</strong>, W.J. & Gregory, R.D. (2000).Spatial synchrony <strong>in</strong> populations of birds: effects of habitat,population trend, <strong>and</strong> spatial scale. Ecology, 81, 2112–2125.Qu<strong>in</strong>n, G. & Keough, M.J. (2002). Experimental Design <strong>and</strong> DataAnalysis for Biologists. Cambridge University Press, Cambridge.Rai, V. & Schaffer, W.M. (2001). Chaos <strong>in</strong> ecology. Chaos, SolitonsFractals, 12, 197–203.Ranta, E., Kaitala, V. & Lundberg, P. (1997). The spatial dimension<strong>in</strong> populations fluctuations. Science, 278, 1621–1623.Ruxton, G. & Doebeli, M. (1996). Spatial self-organization <strong>and</strong>persistence of <strong>transient</strong>s <strong>in</strong> a metapopulation model. Proc. R. Soc.Lond. B, 263, 1153–1158.Saravia, L.A., Ruxton, G.M. & Coviella, C.E. (2000). Theimportance of <strong>transient</strong>sÕ <strong>dynamics</strong> <strong>in</strong> spatially extended populations.Proc. R. Soc. Lond. B, 267, 1781–1786.Solé, R. & Bascompte, J. (1998). Emergent phenomena <strong>in</strong> spatiallyextended model ecosystems. In: Model<strong>in</strong>g Spatiotemporal Dynamics<strong>in</strong> Ecology (eds Bascompte, J. & Solé, R.). Spr<strong>in</strong>ger Verlag, Berl<strong>in</strong>,pp. 1–26.Ó2003 Blackwell Publish<strong>in</strong>g Ltd/CNRS


204 F. A. Labra, N. A. Lagos <strong>and</strong> P. A. MarquetStrogatz, S.H. (1997). Nonl<strong>in</strong>ear Dynamics <strong>and</strong> Chaos. With Applicationsto Physics, Biology, Chemistry <strong>and</strong> Eng<strong>in</strong>eer<strong>in</strong>g. Addison-WesleyPublish<strong>in</strong>g Co., New York.Tilman, D. & Kareiva, P. (1997). Spatial Ecology: The Role of Space <strong>in</strong>Population Dynamics <strong>and</strong> Interspecific Interactions. Pr<strong>in</strong>ceton UniversityPress, Pr<strong>in</strong>ceton, NJ.Turch<strong>in</strong>, P. (1998). Quantitative Analysis of Movement: Measur<strong>in</strong>g <strong>and</strong>Model<strong>in</strong>g Population Redistribution <strong>in</strong> Plants <strong>and</strong> Animals. S<strong>in</strong>auerAssociates, Sunderl<strong>and</strong>, MA.Williams, D.W. & Liebhold, A.M. (2000). Spatial synchrony ofspruce budworm outbreaks <strong>in</strong> eastern North America. Ecology,81, 2753–2766.Manuscript received 11 September 2002First decision made 22 October 2002Manuscript accepted 11 November 2002Ó2003 Blackwell Publish<strong>in</strong>g Ltd/CNRS

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!