Steiner etal 2006 resilienceecology
Upcoming SlideShare
Loading in...5

Steiner etal 2006 resilienceecology






Total Views
Views on SlideShare
Embed Views



0 Embeds 0

No embeds



Upload Details

Uploaded via as Adobe PDF

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
Post Comment
Edit your comment

Steiner etal 2006 resilienceecology Steiner etal 2006 resilienceecology Document Transcript

  • Ecology, 87(4), 2006, pp. 996–1007Ó 2006 by the Ecological Society of AmericaPOPULATION AND COMMUNITY RESILIENCEIN MULTITROPHIC COMMUNITIESCHRISTOPHER F. STEINER,1ZACHARY T. LONG,2JENNIFER A. KRUMINS, AND PETER J. MORINDepartment of Ecology, Evolution, and Natural Resources, Rutgers University, New Brunswick, New Jersey 08901 USAAbstract. Diversity–stability relationships have long been a topic of controversy inecology, but one whose importance has been re-highlighted by increasing large-scale threats toglobal biodiversity. The ability of a community to recover from a perturbation (or resilience) isa common measure of stability that has received a large amount of theoretical attention. Yet,general expectations regarding diversity–resilience relations remain elusive. Moreover, theeffects of productivity and its interaction with diversity on resilience are equally unclear. Weexamined the effects of species diversity, species composition, and productivity on population-and community-level resilience in experimental aquatic food webs composed of bacteria,algae, heterotrophic protozoa, and rotifers. Productivity manipulations were crossed withmanipulations of the number of species and species compositions within trophic groups.Resilience was measured by perturbing communities with a nonselective, density-independent,mortality event and comparing responses over time between perturbed communities andcontrols. We found evidence that species diversity can enhance resilience at the communitylevel (i.e., total community biomass), though this effect was more strongly expressed in low-productivity treatments. Diversity effects on resilience were driven by a sampling/selectioneffect, with resilient communities showing rapid response and dominance by a minority ofspecies (primarily unicellular algae). In contrast, diversity had no effect on mean population-level resilience. Instead, the ability of a community’s populations to recover fromperturbations was dependent on species composition. We found no evidence of an effect ofproductivity, either positive or negative, on community- or population-level resilience. Ourresults indicate that the role of diversity as an insurer of stability may depend on the level ofbiological organization at which stability is measured, with effects emerging only whenfocusing on aggregate community properties.Key words: biodiversity; biomass; composition; ecosystem functioning; microcosm; productivity;protists; resilience; stability.INTRODUCTIONThe impact of species diversity on emergent propertiesof communities and ecosystems is a central issue inecology that has recently gained renewed attention inlight of widespread, human-induced alterations of theearth’s biota (Tilman 1999, Kinzig et al. 2002, Loreau etal. 2002b). Much of this interest has been directedtoward understanding how altered species diversity mayaffect the stability of populations, communities, andecosystems (McCann 2000, Loreau et al. 2002a, Schmidet al. 2002). This topic has a long history in ecology andfew debates have been as contentious. Yet, littleconsensus has been gained regarding diversity–stabilityrelationships despite decades of attention replete withtheoretical consideration. A stumbling block towardadvancement is the paucity of direct experimentalinvestigation; few studies exist in which species diversityhas been directly manipulated and different facets ofstability examined. Fewer still are studies that haveconsidered how species diversity and composition mayinteract with environmental context (e.g., system enrich-ment and productivity) to determine stability.Confusion surrounding diversity–stability relation-ships is due in part to the many ways in which stabilitymay be defined and measured (Pimm 1984, McCann2000). In this study, we examine stability as resilience, orthe rate at which some attribute of a community returnsto its preperturbation state following a system-wideperturbation (Pimm 1984). Early theoretical studiesfocused primarily on population-level resilience: theability of all populations within a community to recoverfrom perturbations away from stable equilibria. In hiswell-known treatment of the subject, May (1973)analyzed eigenvalues of randomly generated Lotka-Volterra community matrices to examine population-level resilience and showed that the probability offinding a stable food web composition decreased as afunction of increasing species diversity. However,natural communities are likely not random in structure,and numerous community attributes may counter theManuscript received 11 May 2005; revised 17 October 2005;accepted 24 October 2005. Corresponding Editor: M. Holyoak.1Present address: W. K. Kellogg Biological Station,Michigan State University, 3700 E. Gull Lake Drive, HickoryCorners, Michigan 49060 USA; E-mail: steiner8@msu.edu2Present address: Institute of Marine Sciences, Universityof North Carolina, Chapel Hill, Morehead City, NorthCarolina 28557 USA.996
  • destabilizing effects of increased diversity such asreduced connectance, reduced average interactionstrength, self-limitation of populations, and donorcontrol (May 1973, DeAngelis 1975, Pimm 1982, Hay-don 1994). Consequently, natural environmental fluctu-ations and dynamic constraints could select forcommunity attributes that enhance system resilienceindependent of diversity (Pimm 1982, Fox andMcGrady-Steed 2002). This suggests that population-level resilience may show no consistent relationship withdiversity but may be more strongly dependent on speciescomposition.How diversity affects the resilience of aggregatecommunity properties such as total community biomass(i.e., community-level resilience) is unclear as well.Models of single trophic level communities suggest thatcommunity-level resilience may decrease with diversity,achieve maximum resilience at intermediate values, orshow no relationship with diversity (Loreau and Behera1999, Lehman and Tilman 2000). Moreover, the abilityto derive useful predictions of community-level resiliencefrom equilibrial analyses of community matrices andpopulation-level resilience (e.g., May 1973, Moore et al.1993, Loreau and Behera 1999) is problematic.Although theoretical predictions of population-levelresilience obtained from such analyses translate intocommunity-level resilience, the two may not correlate innature. For instance, it is conceivable that a subset ofspecies within a natural community may be able torapidly respond following a perturbation, dominatingcommunity biomass and giving rise to high community-level resilience but low mean population-level resilience.Diversity could enhance such effects by increasing theprobability of including species that respond rapidlyfollowing density reductions; what is commonly termeda ‘‘sampling’’ or ‘‘selection’’ effect (sensu Huston 1997,Tilman et al. 1997).While the impact of diversity on stability has enjoyedabundant attention, theoretical consideration of theinfluence of productivity on resilience is sparse andhighly equivocal concerning predictions. Though somestudies have found that productivity may enhancepopulation and community resilience (DeAngelis 1992,Moore et al. 1993), others have shown that no simplerelationship between productivity and resilience mayexist, with predictions being highly dependent on modelstructure and assumptions (Stone et al. 1996, Lundgrenand Frodin 1998, Xu and Li 2002). Even less clear ishow productivity may interact with diversity to deter-mine stability. While some studies have explored thecombined impact of variable food chain length andproductivity on resilience (DeAngelis 1992, Moore et al.1993), we know of no studies that have examined sucheffects within complex food webs.Despite ambiguous and seemingly contradictorymodel predictions, empirical explorations of resilienceemploying direct manipulations of species diversity haverecently emerged (Mulder et al. 2001, Pfisterer andSchmid 2002, Allison 2004). However, these studiessuffer from certain limitations. First, all have usedselective mortality agents (e.g., drought) to perturb theircommunities. Second, none have utilized multitrophiccommunities, focusing only on autotrophs, makingapplication to natural settings and existing theorytenuous (e.g., May 1973, Pimm 1982). We know of nostudies that have experimentally explored how produc-tivity may influence diversity–resilience relationships.Here we present results of an experiment in which weused multitrophic aquatic systems composed of bacteria,algae, heterotrophic protozoa, and rotifers to examinethe effects of species diversity, species composition, andproductivity on the resilience of community- andpopulation-level biomass. We predicted that popula-tion-level resilience would show no consistent relation-ship with diversity but would be dependent instead onspecies composition. We further predicted that com-munity-level diversity could enhance resilience byincreasing the incidence of species that exhibit rapidbiomass responses following disturbance (a selectioneffect). Given the ambiguity of the theoretical literatureon productivity effects, we formulated no hypothesesregarding the influence of productivity on population-or community-level resilience.METHODSExperimental designMicrocosms consisted of 200-mL, loosely cappedPyrex bottles. All experiments were conducted withinincubators at 228C under a 12:12 h light:dark cycle. Weassembled all experimental communities to include fivetrophic groups: decomposers (bacteria), primary pro-ducers (single-celled algae), bacterivores (protozoa),algivores/bacterivores (protozoa and rotifers), andomnivorous top predators (protozoa). All species inour source pool were maintained as laboratory stockcultures (for species lists and culture sources see digitalAppendix A). Each microcosm received one sterilizedwheat seed as a slow-release carbon/nutrient source and100 mL of nutrient medium consisting of distilled water,sieved soil (obtained from the grounds of RutgersUniversity), and Protist Pellet (Carolina BiologicalSupply, Burlington, North Carolina, USA) as a carbonand nutrient source. All materials were autoclavesterilized before use.We used a nested experimental design consisting ofthree diversity levels (low, medium, high) created bymanipulating the number of species within four of ourtrophic groups: primary producers, bacterivores, algi-vores/bacterivores, and omnivorous top consumers.Diversity treatments consisted of either one, two, orfour species per trophic group, respectively. Nestedwithin each diversity level were four unique speciescompositions created by randomly drawing species foreach trophic group from our laboratory source pool(Appendix B). We attempted to exclude species combi-nations that were known to be unviable based on knownApril 2006 997DIVERSITY–STABILITY RELATIONSHIPS
  • biology. Diversity/composition treatments were crossedwith two levels of productivity (low and high).Productivity was manipulated by varying soil andProtist Pellet concentrations, with low-productivitytreatments receiving 0.07 g pellet/L and 0.167 g soil/Land high-productivity treatments receiving 0.70 g pellet/L and 1.67 g soil/L. Medium concentrations equated tototal phosphorus concentrations of 25.4 lg/L for low-productivity treatments and 145.8 lg/L for high-productivity treatments—values spanning the mesotro-phic to hypereutrophic range (Wetzel 2001). Eachtreatment combination was replicated two times for atotal of 48 microcosms.Food web assembly and treatment impositionAll microcosms received a common bacterial com-munity and an assemblage of heterotrophic micro-flagellates. We added microflagellates because they area common experimental contaminant; thus, we equal-ized the probability of their inclusion in all replicates.Sterilized medium first received three species of bacteria(Serratia marcescens, Bacillus cereus, and Bacillussubtilis) known to be edible by all the bacterivores inour study and a low density inoculum of microflagellates(maintained as a laboratory culture). Because nonsterilestock cultures contained additional species of bacteria,we created a pooled bacterial inoculum by filtering ;1mL of medium from all stock cultures through a sterile1.2-lm filter to remove protists, algae, and rotifers. Thisisolate was then added to the experimental media. Tomonitor for contaminants that may have passed throughthe 1.2-lm filter, we added a small volume of theinoculum to bottles containing low- and high-produc-tivity sterile medium (three replicates of each concen-tration). We detected the following contaminants: twounknown species of unicellular green algae, Chrysopsis(an algal flagellate), and Uronema (a bacterivorousciliate). These taxa were detected in several experimentalreplicates. It is likely all contaminants had equalopportunity to invade all of our microcosms.Two days after addition of bacteria, primary pro-ducers were added to their respective treatments (;1 3105cells per species per microcosm). Bacterivores andalgivores were added four days later (10–50 individualsper species per microcosm). Primary consumers wereallowed to respond numerically for 8 d, at which timetop predators were isolated from stock cultures andadded to their respective treatments (10 individuals perspecies per microcosm). Hereafter, we refer to this as day0 of the experiment. Although total biomass initiallyvaried among our diversity/composition treatments, allpopulations that persisted in the microcosms exhibitedincreases in density in the first week of the experiment.Thus, effects of varying initial conditions were minimal.Because the majority of existing theory on resilience hasfocused on systems closed to immigration and emigra-tion, we maintained our microcosms as semiclosedsystems with no species dispersal. We performed weeklyreplacements of 10% of medium from each replicate withsterile medium to replenish nutrients.To sample microcosms, bottles were first gently mixedand a small volume of medium was removed andexamined with a dissecting microscope. We generallyremoved 900-1500 lL of medium; rare taxa wereenumerated by counting the entire sample volume whileabundant taxa were counted in smaller subsamples.Algae and microflagellates were enumerated using ahemacytometer and a compound microscope. Beginningon day 5, we sampled microcosms every three to fourdays up to day 22. On day 25 all experimentalcommunities were perturbed by imposing a nonselective,density-independent mortality event in the form of adilution. To perturb communities, each bottle wasthoroughly mixed and 10 mL of medium (10% of totalvolume) was transferred by pipette to a new bottlecontaining 90 mL of fresh, sterile medium. We retainedthe source microcosm as a control for its correspondingexperimental treatment. Two days after the perturbation(day 27) we sampled all microcosms every two days upto day 31 and then every three to six days up to day 53(the final date of the experiment). This duration waslong enough to encompass numerous generations of ourspecies which had generation times on the order of a fewhours (for some protists) to two days (for rotifers, ourlargest organisms). To measure biomass, we multipliedspecies densities by species-specific biomass constantsobtained from lab records and published accounts(Foissner and Berger 1996). Biomass and diversity ofbacteria are not considered in our analyses.Quantifying resilienceQuantifying resilience requires determination of areference state with which to compare the perturbedcommunity. While theory measures resilience as the rateof a population or community’s return to a stableequilibrium, population- and community-level biomassmeasures in our controls were highly variable over time(e.g., Fig. 1). Prior methods have used the limits of aconfidence interval around the mean of the control as areference (e.g., Cottingham et al. 2004). However,confidence limits may contain negative values for highlyvariable populations and communities (this was the casefor many of our populations). As a compromise, weused biomass measures of our controls averaged overthe post-perturbation experimental period (Fig. 1).To measure community-level resilience, we usedln(total biomass) of the control averaged over time.For each replicate and postperturbation sample, we thentook the difference between ln(biomass) measured in theperturbed community and the average ln(biomass)measured in its corresponding control (i.e., the natu-ral-log ratio). We used linear regression to examine therelationship between the natural-log ratio as the depend-ent variable and time as the independent variable. Theslope from the regression model was used as a measureof resilience, or how rapidly the natural-log ratioCHRISTOPHER F. STEINER ET AL.998 Ecology, Vol. 87, No. 4 View slide
  • approached zero (Fig. 1). To determine whetherdiversity–resilience patterns were caused by rapidresponse and dominance by a subset of species, weexamined the relationship between resilience and thechange in species evenness over time within ourperturbed communities. For each replicate and post-perturbation sample, we calculated evenness using amodified form of Simpson’s dominance index (equationE1/D in Smith and Wilson [1996]). We then used theslope of the linear regression between evenness and timeas a measure of the rate of change in evenness. Todetermine the degree to which community-level resil-ience was driven by producers versus consumers(algivores, bacterivores, and top predators), we alsocalculated resilience measures for these two groupsseparately (using the above method). We did notcalculate separate resilience measures for the threeconsumer trophic groups because some groups (espe-cially top consumers) commonly fell below the limits ofdetection following perturbations, making calculation ofln ratios impossible.To measure mean population-level resilience, we firstcalculated a similarity coefficient (S) based on theCanberra dissimilarity index (Legendre and Legendre1998) for each replicate and for each postperturbationsample,S ¼ 1 À1tðDÞD ¼Xtj¼1jyPj À yCj jðyPj þ yCjÞ where t was the total number of species, yPj was theuntransformed biomass of species j in the perturbedcommunity, and yCj was species j’s time-averagedbiomass in its corresponding control. We chose thismetric because it is not sensitive to differences in speciesrichness, total biomass, or evenness (C. F. Steiner,personal observation). Relationships between S and timewere commonly saturating. To linearize relationships,we log10 transformed time and used the slope from thelinear regression between similarity (S) and log10(time)as a relative measure of population-level resilience.Statistical analysisDue to methodological error, data from several low-and medium-diversity controls were lost on day 34. Wehave entirely removed this sample date from all stabilitycalculations and statistical analyses. We analyzedmeasures of resilience using a mixed model ANOVA,with composition (a random effect) nested withindiversity and crossed with productivity. Species compo-sitions diverged rapidly from their initial states; this andthe presence of contaminants caused realized diversity tovary within our diversity treatments. Consequently, wealso analyzed the effects of realized species richness onstability using ANCOVA, treating productivity as afixed effect and realized species richness as a continuouscovariate. To further explore potential determinants ofpopulation and community-level resilience, we per-formed stepwise multiple linear regressions for low andhigh-productivity treatments separately. The followingexplanatory variables were entered in the analyses:average realized species richness, change in speciesevenness over time, log10(mean total community bio-FIG. 1. Example dynamics of a perturbed system, itscorresponding control, and its community-level resilience.Perturbation was imposed on day 25. Results are for onereplicate of the medium-diversity, composition 5 treatment, athigh productivity. (A) Community-level biomass through timein the control and perturbed communities. The dashed linerepresents time-averaged biomass in the control over thepostperturbation period. (B) The difference between ln(com-munity-level biomass) in the perturbed community and thetime-averaged ln(biomass) of the control (i.e., the natural-logratio) through time. The linear regression line is shown; theslope of this relationship was used as a measure of community-level resilience.April 2006 999DIVERSITY–STABILITY RELATIONSHIPS View slide
  • mass), and log10(mean per capita biomass) (as a measureof average size per individual). Mean total communitybiomass for each perturbed community was averagedover the postperturbation period to obtain a singlemeasure. To calculate mean per capita biomass, for eachperturbed community and postperturbation sampledate, biomass was summed across species (excludingbacteria) and divided by the total number of individualspresent. Values were then averaged over the postpertur-bation period to obtain a time-average. All statisticswere performed using Systat Version 8 (Systat Software,Point Richmond, California, USA) and SAS Version 8(SAS Institute, Cary, North Carolina, USA).RESULTSRealized species richness initially declined and ap-peared to stabilize by day 8 of the experiment (Fig. 2).Though species richness diverged between productivitylevels and among composition treatments, significantdifferences in realized species richness were still present onday 22, prior to perturbations (Fig. 2; F2,42 ¼ 32.39, P, 0.0001, diversity effect, ANOVA; P , 0.005, allpairwise comparisons, Tukey’s hsd). No productivity orproductivity 3 diversity effect was detected (P . 0.13,ANOVA).Community-level resilienceWhen using nested ANOVA, we detected no effects ofdiversity, productivity or species composition on com-munity-level resilience (Fig. 3A, B; P . 0.14, all effectsand interactions). However, variation among composi-tion treatments was highly heterogeneous (Fig. 3B).Thus, some caution regarding analyses is warranted.Variances were especially high for low-productivitycompositions 2 and 4 (Fig. 3B). For both of thesetreatments, a contaminant species (Uronema) invadedone replicate but not the other, causing divergentresponses. However, when using nonparametric AN-OVA, no effect of composition was detected at eitherlow (P ¼ 0.38, Kruskal-Wallis test) or high productivity(P ¼ 0.15, Kruskal-Wallis test).Fig. 3A suggests a positive effect of diversity oncommunity-level resilience, especially at low productiv-ity. This was clearer when analyzing effects of realizedspecies richness (averaged over the postperturbationperiod) on community-level resilience (Fig. 3C). Asignificant effect of realized species richness was detectedwhen using ANCOVA (F1,44 ¼ 4.05, P ¼ 0.050, R2¼0.12); no effect of productivity or its interaction withrealized richness was present (P . 0.257, ANCOVA).Although no interaction with productivity was detectedusing ANCOVA, regressions showed that community-level resilience was related to realized species richness atlow productivity (R2¼ 0.163, P ¼ 0.050) but not in high-productivity treatments (R2¼ 0.022, P ¼ 0.487).Primary producers more strongly contributed tocommunity-level resilience compared to consumers(Fig. 4). Although significant positive relationships weredetected for primary producers (R2¼ 0.57, P , 0.0001,linear regression) and consumers (R2¼ 0.16, P ¼ 0.005,linear regression), resilience of total primary producerbiomass was greater than consumer resilience whenaveraging across all replicates (P , 0.001, t test).Moreover, consumer resilience measures fell near orbelow zero for several food webs (Fig. 4) indicating thatconsumer responses commonly weakened communityrecovery or had little influence.When analyzing community-level resilience usingforward and backward stepwise regressions, only changein species evenness was retained in the model. This wastrue of low-productivity treatments (R2¼ 0.297, P ¼0.006) and high-productivity treatments (R2¼ 0.207, P ¼0.025). Those communities that exhibited high commu-nity-level resilience were those that showed rapiddecreases in evenness following perturbations (Fig. 5).This effect was stronger in low-productivity treatments asindicated by a significant interaction between change inevenness and productivity when using ANCOVA (F1,44 ¼4.10, P¼ 0.049). Change in species evenness also covariedwith average realized species richness at low productivity,becoming more negative with increasing species richness(r ¼À0.40, P ¼ 0.053, Pearson correlation). No relation-ship was detected at high productivity (P ¼ 0.95).To determine which species were driving community-level resilience, for each replicate we performed correla-tions between species biomass responses (log10[x þ 1]transformed) in perturbed treatments and the natural logratio of community-level biomass; Bonferroni adjusted Pvalues were used to adjust for multiple comparisons. Apositive correlation indicates that an increase in theFIG. 2. Change in total species richness over time (means 6SE) in all six diversity and productivity treatment combinations.Dashed lines are low-productivity, and solid lines are high-productivity treatments. Results are up to the last sample datebefore the perturbation was imposed.CHRISTOPHER F. STEINER ET AL.1000 Ecology, Vol. 87, No. 4
  • biomass of a species was associated with a decreasingnegative value of the ln(ratio) and was thus contributingto community-level resilience. We found that commu-nity-level resilience was commonly associated with re-sponses by primary producers, consistent with Fig. 4. Of34 positive correlations significant at the P , 0.10 level,only five were attributable to consumer species. Twentycorrelations were due solely to the unicellular green algaeAnkistrodesmus and Chlorella, which were especiallyrapid in their responses following perturbations. Fig. 6displays the relationship between community-level resil-ience and the relative biomass of Ankistrodesmus and/orChlorella (results are averages grouped by compositionand productivity treatment). Increasing dominance byFIG. 3. (A) Resilience of community-level biomass as a function of low, medium, and high species diversity treatments at bothproductivity levels. Values shown are mean 6 SE of values that were first averaged across compositions. (B) Community-levelresilience as a function of species composition at low and high productivity. Values are mean 6 SE. Vertical lines demarcate low-,medium-, and high-diversity treatments (from left to right). (C) Community-level resilience as a function of average realized speciesrichness at low and high productivity. Linear regression lines for low-productivity treatments (solid line) and high-productivitytreatments (dashed line) are shown.April 2006 1001DIVERSITY–STABILITY RELATIONSHIPS
  • these algae was associated with increasing community-level resilience (r ¼ 0.58, P ¼ 0.003, Pearson correlation).Moreover, when using two-way ANOVA, presence ofthese two species (either alone or together) wassignificantly and positively related to community-levelresilience (F1,44 ¼ 5.65, P ¼ 0.02) regardless of diversitylevel; no interaction with productivity (P ¼ 0.69) or aproductivity main effect was detected (P ¼ 0.41).Mean population-level resilienceNo effect of diversity on population-level resiliencewas detected when using nested ANOVA (Fig. 7A; P ¼0.727, diversity effect; P¼0.362, diversity3productivityeffect). Population-level resilience varied significantlyamong composition treatments. However, the magni-tude of composition effects varied with productivitylevel, as indicated by a significant productivity 3composition interaction (Fig. 7B; F9,24 ¼ 2.33, P ,0.048); no main effect of composition was present (P ¼0.158). Much like community-level responses, varianceswere somewhat heterogeneous among compositiontreatments (Fig. 7B). However, nonparametric ANOVAalso revealed composition effects on population-levelresilience (P ¼ 0.057, Kruskal-Wallis test). Whenexamining the relationship between average realizedspecies richness, productivity, and population-levelresilience, no significant effects were present (Fig. 7C;P . 0.15, ANCOVA, all effects and interactions).Stepwise regressions with population-level stabilityproduced different results for low and high-productivitytreatments. At low productivity, only mean per capitabiomass was retained in forward and backward regres-sions (R2¼ 0.207, P ¼ 0.026). Population-level resiliencedecreased with increasing mean size of individuals (Fig.FIG. 4. The relationship between the resilience of primaryproducers (solid line) and consumers (dashed line) andcommunity-level resilience. Linear regression lines are shown.FIG. 5. Community-level resilience as a function of changein species evenness over time at low and high productivity.Regression lines for low-productivity treatments (solid line) andhigh-productivity treatments (dashed line) are shown.FIG. 6. The relationship between mean community-levelresilience and the mean (6SE) relative biomass of Ankistrodes-mus and/or Chlorella. Results have been grouped by composi-tion and productivity treatment. Algal relative biomass wasmeasured in the perturbation treatments on the final date of theexperiment (results were qualitatively similar when usingrelative biomass averaged over time). The linear regression lineis shown.CHRISTOPHER F. STEINER ET AL.1002 Ecology, Vol. 87, No. 4
  • 8). No explanatory variables were retained at the P ,0.05 level in high-productivity treatments.DISCUSSIONOf the numerous factors thought to influence thestability of populations and communities, diversityremains highly controversial. Healthy debate beganearly, impelled in large part by the mathematicalexplorations of May (1973), which appeared to contra-dict the prevailing wisdom that diversity should give riseto stability (reviewed in Goodman 1975, McNaughton1977). Theory since has done much to ascertain criticalmechanisms that may affect diversity–stability relation-ships but has arguably done less to clarify whatFIG. 7. (A) Population-level resilience as a function of low, medium, and high species diversity treatments at both productivitylevels. Values shown are the mean 6 SE of values that were first averaged across compositions. (B) Population-level resilience (mean6 SE) as a function of species composition at low and high productivity. Vertical lines demarcate low-, medium-, and high-diversitytreatments (from left to right).(C) Population-level resilience as a function of average realized species richness at low and highproductivity.April 2006 1003DIVERSITY–STABILITY RELATIONSHIPS
  • relationships are to be expected in nature and underwhat environmental contexts. Rigorous experimentalstudies of the impact of species diversity on stability area relatively recent development that may begin toprovide some resolution to this complex issue.Community-level resilienceOur experiment provided evidence that species diver-sity can enhance the stability of aggregate communityproperties. Resilience of total community biomassincreased as a function of increasing realized speciesrichness in low-productivity treatments. However, di-versity was clearly a weak explanatory variable in ourstudy, accounting for a relatively small percentage ofvariation in community-level resilience. Hence, speciesdiversity does not consistently ensure system stability.Community-level resilience was driven by rapid responseand dominance by a minority of species followingperturbations; more resilient communities were thosethat exhibited decreases in evenness over time (Fig. 5).More importantly, in low-productivity treatments morediverse communities showed a greater tendency forevenness to decrease following perturbations. Thus, thismechanism may explain, in part, our positive diversity-resilience relationship. Because our mortality agent wasnon-selective in nature (removing a set percentage of allpopulations), differences among communities in even-ness responses were not due to differences in species’abilities to resist initial perturbations. This pointdistinguishes our experiment from earlier diversity-perturbation studies (Mulder et al. 2001, Pfisterer andSchmid 2002, Allison 2004) in which the use of selectivemortality agents potentially confounds populationresponses (and resilience) with differential resistanceamong species to the initial perturbation. Our resultssuggest fundamental differences among communities inthe rate of species’ population responses; some com-munities contain species that respond quickly, dominat-ing biomass and lowering overall evenness.The tendency for more diverse communities to exhibithigher resilience and stronger decreases in evenness overtime indicates that these communities had a greaterprobability of harboring species with rapid populationresponses. Past studies that have sought mechanisticinsight into the positive effect of species diversity onecosystem functioning have commonly focused on twobroad classes of processes: complementarity effects andselection effects (Huston 1997, Tilman et al. 1997,Loreau and Hector 2001). The former occurs as a resultof niche differentiation and resource partitioning amongco-occurring species. The latter emerges when speciesthat have strong effects on ecosystem functioning inmonoculture occur in polycultures. Unfortunately,existing methods for mathematically partitioning com-plementarity and selection effects require measurementof the focal ecosystem response variable in both mono-and polycultures (Loreau and Hector 2001), an impos-sibility in a multitrophic design such as ours. Regardless,our study provided support for the selection effectsmodel. Diversity effects on community-level resilience inour experiment did not appear to be a result of the‘‘inherent’’ dynamic or structural properties of morediverse food webs per se. If such were the case, we wouldexpect the majority of populations within diversecommunities to rapidly rebound following perturba-tions. However, population-level resilience showed norelationship with diversity in our experiment. Rather,more diverse communities, by chance, included speciesthat were able to exhibit rapid biomass responsesfollowing mass mortality events. Selection effects areexpected to be especially strong when the number ofspecies combinations among high diversity communitiesare high relative to the number of species available in thespecies pool. This was especially true in our experimentin which our pool of available algal species was quitesmall. In fact, Ankistrodesmus and Chlorella (the twoalgal species that strongly dominated following pertur-bations) were present either alone or together in all ofour high diversity manipulations (Appendix B).While selection effects are a viable explanation for ourpositive diversity–community-resilience relationship,they may not be the only explanation for observedresponses. Though high resilience was commonlyassociated with high relative biomass by Ankistrodesmusand Chlorella, presence of these species did notinvariably translate into dominance by these algae orhigh community-level stability (Fig. 6). For example,Chlorella responses drove high resilience in composition10 at low productivity (Fig. 3B); this high diversitytreatment not only had high community-level resiliencebut the second strongest decrease in evenness over time.However, Chlorella was also present in low diversity,FIG. 8. The relationship between population-level resilienceand mean per capita biomass in low-productivity treatments.The linear regression line is shown.CHRISTOPHER F. STEINER ET AL.1004 Ecology, Vol. 87, No. 4
  • composition 3 (Fig. 3B; the black triangle below theregression line in Fig. 6); at low productivity, thistreatment had extremely low resilience, low meanChlorella relative biomass, and the strongest increasein evenness. This suggests that some aspect of compo-sition or diversity allowed strong Chlorella responses(and high resilience) in one treatment but not the other.Differential responses could be due to several factors.For instance, more diverse communities may have had ahigher probability of including predators that moreeffectively controlled algivores, indirectly benefitingdominant species such as Chlorella or Ankistrodesmus.Additionally, we observed that decomposition rateswithin our experimental communities (as measured bypercentage of decomposition of wheat seeds) increasedwith increasing realized species richness (J. A. Krumins,Z. T. Long, C. F. Steiner, and P. J. Morin, unpublishedmanuscript). Thus, enhanced nutrient regeneration couldhave further catalyzed algal responses in high diversitytreatments. Such effects could be driven by complemen-tarity among species and would be confounded withperceived selection effects of Ankistrodesmus and Chlor-ella as these taxa more frequently occurred in our highdiversity manipulations. Though we can only speculateon the operation of these mechanisms, they serve tohighlight the complex direct and indirect effects that areinherent in multitrophic settings.One point of concern is that our measures of resiliencerelied on natural-log ratios calculated using time-averaged biomasses of the control treatments. Thisobviously ignores temporal variability of the control.However, two lines of evidence point to the robustness ofour general conclusions. First, when calculating natural-log ratios by randomly pairing perturbed and controlbiomasses through time, resilience measures exhibited apositive trend with realized species diversity (P ¼ 0.07,ANCOVA; P . 0.20 for productivity and productivity3diversity effects). Second, when ignoring controls andmeasuring the slope of the relationship between ln(per-turbed biomass) and time (i.e., the rate of biomassaccrual in our perturbed communities), slopes werepositively related to realized species diversity (P ¼ 0.05,ANCOVA; P . 0.20 for productivity and productivity3diversity effects). Hence, recovery rates following per-turbations were higher in more speciose communities.Mean population-level resilienceUnlike community-level resilience in which responsescan be driven by any fraction of the resident community,our measure of population-level resilience was intended tocapture the ability of all component populations (exclud-ing bacteria) to return to their individual preperturbationstates. Of the many measures of stability, population-levelresilience has received a great amount of theoreticalattention. Although early theory suggested that diversitycould reduce population-level stability (May 1973), manystudies since have shown that this is not invariably true;complex communities can be highly resilient (e.g., Pimm1982, Haydon 1994). Our results indicate that population-level resilience may show no consistent relationship, eitherpositive or negative, with diversity. We found no strongeffects of diversity or realized species richness onpopulation-level stability. Instead, variation in thismeasure was dependent on species composition. At lowproductivity, population resilience was driven by meansize of individuals present in the community. Thosecommunities that were, on average, composed of individ-uals with lower per capita biomass exhibited higherpopulation-level resilience. It is possible that this effectwas mediated by body size effects on population growthrates; body size is known to scale strongly and negativelywith reproductive rate (Fenchel 1974). Moreover, smallerbody size and rapid reproductive rates are commonlythought to enhance the resilience of populations (Pimm1991). Deducing the causes of variable population-levelresilience inhigh-productivity treatmentsis moredifficult.Nested ANOVA exposed significant variation amongcompositions, indicating that variation in population-level resilience was present among high-productivitycommunities. However, regressions revealed no signifi-cant effects of the potential explanatory variables that wemeasured, including mean per capita biomass. Since arange of factors may theoretically affect population-levelresilience, comprehending our results may require muchmore detailed knowledge of community features such asconnectance and species interaction strength.As with community-level resilience, our measures ofpopulation-level resilience were calculated using time-averaged population biomasses in controls, ignoringtemporal variability in the control. When we calculatedsimilarity indices by randomly pairing each species’biomass in perturbed and control treatments, produc-tivity 3 composition effects on population-level resil-ience were weaker (P ¼ 0.13, nested ANOVA). This isnot surprising as many populations were highly variableover time in control treatments. Thus, some cautionregarding the strength of compositional effects onpopulation-level resilience is warranted.Productivity effectsMuch like studies of diversity–stability relations, theimpact of productivity on stability enjoyed earlytheoretical inquiry. However, this research focusedprimarily on the influence of enrichment on temporalstability of communities (Rosenzweig 1971, Gilpin1972); theoretical explorations of the effect of produc-tivity on resilience are a more recent development (e.g.,DeAngelis 1992, Moore et al. 1993, Stone et al. 1996).Some of these studies have found that productivity canenhance resilience (DeAngelis 1992, Moore et al. 1993).Yet, others suggest that productivity may have noconsistent effect (Stone et al. 1996, Lundgren andFrodin 1998). Our results lend support to the latterassertion; we found no consistent effect of productivityon either population or community-level resilience.When averaging across diversity and composition levels,April 2006 1005DIVERSITY–STABILITY RELATIONSHIPS
  • communities in high-productivity treatments were nomore resilient than those in low-productivity treatments.Nonetheless, productivity may interact with diversity todetermine community-level resilience; a significant rela-tionship with realized species richness was only detectedin low-productivity treatments. However, this resultmust be viewed cautiously as the effect was weak (nointeraction was detected in ANCOVA) and may havebeen due to the fact that realized species richnessreached a lower minimum in low-productivity treat-ments. We also detected a significant interaction ofproductivity and species composition when analyzingpopulation-level resilience. For some species composi-tions, productivity can enhance population resilience(e.g., compositions 5 and 6, Fig. 7B). However, whenexamining general trends, productivity clearly had weakeffects across the majority of compositions. Lack of aconsistent productivity effect on resilience does notnegate the possibility of observing stronger effects undermore strongly contrasting productivity conditions. Useof more widely varying productivity levels could haveyielded different results in our study.ConclusionIn the face of increasing threats to global biodiversity,ecologists confront the challenge of understanding andpredicting how such changes will impact community andecosystem properties. The ability of populations andcommunities to persist through time and withstandexternal perturbations is fundamental to this emergingissue. Fortunately, ecology is rich in theory regardingpredicted patterns and drivers of resilience. Yet,empirical explorations have arguably lagged behindtheir theoretical counterparts. As declared by McNaugh-ton (1977) during the height of the diversity–stabilitydebate, ‘‘continued assertions of the validity of one oranother conclusion about diversity–stability, in theabsence of empirical tests, are acts of faith, not science.’’The recent rise in experimental studies of stability willhopefully begin to shed light on this important problem.Our study adds to the growing body of evidence showingthat diversity can influence stability. Diversity canenhance a system’s ability to return from a perturbationby increasing the probability of including species thatcan respond quickly following mortality events. How-ever, this effect was only evident at the community-leveland in low-productivity treatments. These findings arequalitatively similar to those in a companion studywhich uncovered contrasting effects of species diversityand composition on temporal stability (Steiner et al.2005). Diversity effects only emerged when examiningtemporal stability at the community level while compo-sition effects outweighed diversity effects at the pop-ulation level (Steiner et al. 2005).Whether our conclusions extend beyond our simplemodel system is an important question that can only beanswered by careful experimentation in the field. Ourstudy was designed as a test of theory confined to ahighly localized scale and limited species pool. In themajority of natural systems, larger species source poolswill undoubtedly be available extending the array ofspecies compositions and diversity levels that a localcommunity may express. Moreover, species dispersalamong localities will be possible; the role that suchspatial dynamics play in ecosystem functioning andstability is potentially considerable, but little studiedempirically (Loreau et al. 2003, Cardinale et al. 2004,Loreau and Holt 2004). Our study also explored speciesdiversity effects by uniformly varying species numbers inall trophic groups. Yet, species loss is likely not randomin natural systems with extinction risk being appor-tioned unequally among trophic levels or among taxawithin trophic levels (Pimm et al. 1988, Petchey et al.2004). In theory, the impact of ordered species loss onthe functioning of ecosystems may differ from randomspecies extinction (Ives and Cardinale 2004, Petchey etal. 2004, Solan et al. 2004), but empirical tests of howsuch processes affect the stability of populations andcommunities remain unexplored. These points highlightseveral potentially fruitful and vital areas of futureresearch. Experimental studies of diversity–resiliencerelationships, especially within natural multitrophicsettings, are still in their infancy. Until further corrob-orative evidence accumulates, our work may be cau-tiously viewed as a prelude of patterns to come. Ourstudy shows that the value of diversity as a predictivemeasure or insurer of resilience may be highly contextdependent, being a function of both productivity and thelevel of biological organization at which stability ismeasured.ACKNOWLEDGMENTSWe thank Carla Ca´ ceres, Amy Downing, Spencer Hall, andCindy Hartway for thoughtful discussions; Jay Kelly for pondaccess (and brunch); and Kirsten Schwarz for lab assistance.Four anonymous reviewers provided valuable comments onthis manuscript. This research was supported by a NSFMicrobial Biology Postdoctoral Fellowship to C. F. Steiner.LITERATURE CITEDAllison, G. 2004. The influence of species diversity and stressintensity on community resistance and resilience. EcologicalMonographs 74:117–134.Cardinale, B. J., A. R. Ives, and P. Inchausti. 2004. Effects ofspecies diversity on the primary productivity of ecosystems:extending our spatial and temporal scales of inference. Oikos104:437–450.Cottingham, K. L., S. Glaholt, and A. C. Brown. 2004.Zooplankton community structure affects how phytoplank-ton respond to nutrient pulses. Ecology 85:158–171.DeAngelis, D. L. 1975. Stability and connectance in food webmodels. Ecology 56:238–243.DeAngelis, D. L. 1992. Dynamics of nutrient cycling and foodwebs. Chapman and Hall, New York, New York, USA.Fenchel, T. 1974. Intrinsic rate of natural increase: therelationship with body size. Oecologia 14:317–326.Foissner, W., and H. Berger. 1996. A user-friendly guide to theciliates (Protozoa, Ciliophora) commonly used by hydro-biologists as bioindicators in rivers, lakes, and waste waters,with notes on their ecology. Freshwater Biology 35:375–482.CHRISTOPHER F. STEINER ET AL.1006 Ecology, Vol. 87, No. 4
  • Fox, J. W., and J. McGrady-Steed. 2002. Stability andcomplexity in microcosm communities. Journal of AnimalEcology 71:749–756.Gilpin, M. E. 1972. Enriched predator–prey systems: theoreticalstability. Science 177:902–904.Goodman, D. 1975. The theory of diversity–stability relation-ships in ecology. Quarterly Review of Biology 50:237–266.Haydon, D. 1994. Pivotal assumptions determining therelationship between stability and complexity: an analyticalsynthesis of the stability–complexity debate. AmericanNaturalist 144:14–29.Huston, M. A. 1997. Hidden treatments in ecological experi-ments: re-evaluating the ecosystem function of biodiversity.Oecologia 110:449–460.Ives, A. R., and B. J. Cardinale. 2004. Food-web interactionsgovern the resistance of communities after non-randomextinctions. Nature 429:174–177.Kinzig, A. P., S. W. Pacala, and D. Tilman. 2002. Thefunctional consequences of biodiversity: empirical progressand theoretical extensions. Princeton University Press,Princeton, New Jersey, USA.Legendre, P., and L. Legendre. 1998. Numerical ecology.Elsevier, Amsterdam, The Netherlands.Lehman, C. L., and D. Tilman. 2000. Biodiversity, stability,and productivity in competitive communities. AmericanNaturalist 156:534–552.Loreau, M., and N. Behara. 1999. Phenotypic diversity andstability of ecosystem processes. Theoretical PopulationBiology 56:29–47.Loreau, M., A. Downing, M. Emmerson, A. Gonzalez, J.Hughes, P. Inchausti, J. Joshi, J. Norberg, and O. Sala. 2002a.A new look a the relationship between diversity and stability.Pages 79–91 in M. Loreau, S. Naeem, and P. Inchausti,editors. Biodiversity and ecosystem functioning: synthesis andperspectives. Oxford University Press, New York, USA.Loreau, M., and A. Hector. 2001. Partitioning selection andcomplementarity in biodiversity experiments. Nature 412:72–76.Loreau, M., and R. D. Holt. 2004. Spatial flows and theregulation of ecosystems. American Naturalist 163:606–615.Loreau, M., N. Mouquet, and A. Gonzalez. 2003. Biodiversity asspatial insurance in heterogeneous landscapes. Proceedings ofthe National Academy of Sciences (USA) 100:12765–12770.Loreau, M., S. Naeem, and P. Inchausti. 2002b. Biodiversityand ecosystem functioning: synthesis and perspectives.Oxford University Press, New York, New York, USA.Lundberg, P., and P. Frodin. 1998. Ecosystem resilience andproductivity: are predictions possible? Oikos 81:603–606.May, R. M. 1973. Stability and complexity in model ecosystems.Princeton University Press, Princeton, New Jersey, USA.McCann, K. S. 2000. The diversity–stability debate. Nature405:228–233.McNaughton, S. J. 1977. Diversity and stability of ecologicalcommunities: a comment on the role of empiricism inecology. American Naturalist 111:515–525.Moore, J. C., P. C. de Ruiter, and H. W. Hunt. 1993. Influenceof productivity on the stability of real and model ecosystems.Science 261:906–908.Mulder, C. P. H., D. D. Uliassi, and D. F. Doak. 2001. Physicalstress and diversity–productivity relationships: the role ofpositive interactions. Proceedings of the National Academyof Sciences (USA) 98:6704–6708.Petchey, O. L., A. L. Downing, G. G. Mittelbach, L. Persson,C. F. Steiner, P. H. Warren, and G. Woodward. 2004.Species loss and the structure and functioning of multitrophicaquatic systems. Oikos 104:467–478.Pfisterer, A. B., and B. Schmid. 2002. Diversity-dependentproduction can decrease the stability of ecosystem function-ing. Nature 416:84–86.Pimm, S. L. 1982. Food webs. Chapman and Hall, London, UK.Pimm, S. L. 1984. The complexity and stability of ecosystems.Nature 307:321–326.Pimm, S. L. 1991. The balance of nature? University of ChicagoPress, Chicago, Illinois, USA.Pimm, S. L., H. L. Jones, and J. Diamond. 1988. On the risk ofextinction. American Naturalist 132:757–785.Rosenzweig, M. L. 1971. Paradox of enrichment: destabiliza-tion of exploitation ecosystems in ecological time. Science171:385–387.Schmid, B., J. Joshi, and F. Schlapfer. 2002. Empirical evidencefor biodiversity–ecosystem functioning relationships. Pages120–150 in A. P. Kinzig, S. W. Pacala, and D. Tilman,editors. The functional consequences of biodiversity: empiri-cal progress and theoretical extensions. Princeton UniversityPress, Princeton, New Jersey, USA.Smith, B., and J. B. Wilson. 1996. A consumer’s guide toevenness indices. Oikos 76:70–82.Solan, M., B. J. Cardinale, A. L. Downing, K. A. M.Engelhardt, J. L. Ruesink, and D. S. Srivastava. 2004.Extinction and ecosystem function in the marine benthos.Science 306:1177–1180.Steiner, C. F., Z. T. Long, J. A. Krumins, and P. J. Morin.2005. Temporal stability of aquatic food webs: partitioningthe effects of species diversity, species composition andenrichment. Ecology Letters 8:819–828.Stone, L., A. Gabric, and T. Berman. 1996. Ecosystemresilience, stability, and productivity: seeking a relationship.American Naturalist 148:892–903.Tilman D. 1999. The ecological consequences of changes inbiodiversity: a search for general principles. Ecology 80:1455–1474.Tilman, D., C. L. Lehman, and K. T. Thomson. 1997. Plantdiversity and ecosystem productivity: theoretical consider-ations. Proceedings of the National Academy of Sciences(USA) 94:1857–1861.Wetzel, R. G. 2001. Limnology: lake and river ecosystems.Third edition. Academic Press, New York, New York, USA.Xu, C., and Z. Li. 2002. Stochastic ecosystem resilience andproductivity: seeking a relationship. Ecological Modeling156:143–152.APPENDIX AA list of species used in the experiment and culture sources (Ecological Archives E087-057-A1).APPENDIX BA table containing species combinations of all composition treatments (Ecological Archives E087-057-A2).April 2006 1007DIVERSITY–STABILITY RELATIONSHIPS