Global change community ecology beyond species-sorting: a quantitative framework based on mediterranean-biome examples


  • Greg R. Guerin,

    Corresponding author
    1. School of Earth and Environmental Sciences, The Environment Institute, University of Adelaide, Adelaide, South Australia, Australia
    • Correspondence: Greg R. Guerin, School of Earth and Environmental Sciences, the Environment Institute, University of Adelaide, Adelaide, South Australia, Australia.


    Search for more papers by this author
  • Irene Martín-Forés,

    1. Department of Ecology, Universidad Complutense de Madrid, Madrid, Spain
    Search for more papers by this author
  • Ed Biffin,

    1. School of Earth and Environmental Sciences, The Environment Institute, University of Adelaide, Adelaide, South Australia, Australia
    Search for more papers by this author
  • Zdravko Baruch,

    1. School of Earth and Environmental Sciences, The Environment Institute, University of Adelaide, Adelaide, South Australia, Australia
    Search for more papers by this author
  • Martin F. Breed,

    1. School of Earth and Environmental Sciences, The Environment Institute, University of Adelaide, Adelaide, South Australia, Australia
    Search for more papers by this author
  • Matthew J. Christmas,

    1. School of Earth and Environmental Sciences, The Environment Institute, University of Adelaide, Adelaide, South Australia, Australia
    Search for more papers by this author
  • Hugh B. Cross,

    1. School of Earth and Environmental Sciences, The Environment Institute, University of Adelaide, Adelaide, South Australia, Australia
    2. Department of Environment, Water and Natural Resources, Adelaide, South Australia, Australia
    Search for more papers by this author
  • Andrew J. Lowe

    1. School of Earth and Environmental Sciences, The Environment Institute, University of Adelaide, Adelaide, South Australia, Australia
    2. Department of Environment, Water and Natural Resources, Adelaide, South Australia, Australia
    3. South Australian Museum, Adelaide, South Australia, Australia
    Search for more papers by this author

  • Editor: Josep Penuelas



Species-sorting predicts the influences of the environment on ecosystem composition across heterogeneous landscapes. It assumes that extinction and adaptation are negligible at ecological scales. Meanwhile, shifts associated with global change have been observed in metacommunity composition (species pools affected by extinctions and introductions) and in phenotypes. This suggests that predictions of future composition must move beyond re-sorting present-day species with fixed traits. We extend plant species-sorting concepts to consider biogeography and demography. We give an empirical context, highlighting the influences of biogeography, species-sorting and adaptation on community composition.


Global; case studies focus on the terrestrial mediterranean biome.


We review case studies of empirical approaches that have examined community composition at various scales. We develop a mathematical model based on community mechanics that incorporates species-sorting with shifting phenotypes and species pools.


As illustrated by real examples, community composition is influenced by factors such as history, modern extinction risk, species-sorting, biotic interactions, adaptation and ecological drift. There is ample evidence that species pools and phenotypes are not constant at ecological scales in the context of global change. Any implicit assumption in community analysis that they are constant should therefore be revisited. Our model breaks down shifting community constraints into intraspecific components – including genotype sorting, selection and plasticity – and interspecific components, including changes in relative abundance and species replacement from a shifting metacommunity.

Main conclusions

Predictions of community composition could benefit from extending species-sorting, to allow species pools and species traits to shift through time, as dealt with explicitly in our framework. The model predicts that responses to a shifting community constraint can be more diverse than deterministic species re-sorting. Consequently, the rate of species replacement depends on factors such as species adaptive capacity, competition, physical disturbance and habitat fragmentation.


A central aim in community ecology is to predict species occurrences within a habitat patch (‘community composition’) given a broader pool from which the species may be sourced (HilleRisLambers et al., 2012). Approaches that examine the landscape sorting of species have advanced beyond models involving just competition (Gilpin & Diamond, 1982) or environmental filters (Keddy, 1992). In parallel, biogeography has tested the diversification and persistence of assemblages within and across biomes (Byrne, 2008). Wiens & Donoghue (2004) identified a critical link between community ecology and biogeography, stating that ‘in community ecology, the composition of local-scale communities depends upon the regional species pool, and the composition of the regional species pool is, in turn, determined by large-scale biogeographical processes’.

There have been few attempts to explicitly link either adaptation or the development of regional species pools with species-sorting. In one such example, Prober et al. (2012) developed a framework for the management of a mediterranean-climate woodland in Western Australia. They considered ecological responses to climate change at different scales, from individuals to regions. Norberg et al. (2012) simulated species richness under climate change, a model in which both ecological sorting and evolutionary responses were allowed. Although quantitative empirical data in such cases are sparse (Norberg et al., 2012), empirical examples that link biogeography and community ecology show promise. For instance, Lee et al. (2012) showed that earlier-diverged lineages in the Southern Alps of New Zealand were more likely to be community dominants. Ackerly (2004) argued that limited leaf trait evolution in chaparral plants suggests ‘exaptation’ (i.e. essentially pre-adaptation) to mediterranean climates. Hui et al. (2013) showed that assemblages of recently introduced plant species were more stochastic than assemblages that had persisted through historical filters.

The major processes involved in sorting species across a landscape have been identified (de Bello et al., 2005; Fine & Kembel, 2011). Community composition has a context within a regional species pool, which has a biogeographical history (Wiens & Donoghue, 2004; HilleRisLambers et al., 2012; Lee et al., 2012; Hui et al., 2013), and this species pool undergoes environmental filtering. Species must be able to survive in a particular abiotic environment, given biotic interactions such as pollination, competition, grazing, parasitism and predation (Cornwell & Ackerly, 2009; Kleyer et al., 2012). These filters vary through space and time (Davis, 1986), and persistence further depends on population demography and physical disturbance regimes (Prior et al., 2003). Thus, community assembly is a complex set of interactions at multiple spatial and temporal scales (Fig. 1). In addition to today's environmental setting, historical factors that operate over tens to millions of years, such as phylogenetic biome conservatism and ecological drift (random changes in composition), contribute to present-day assemblages (Davis, 1986; Crisp et al., 2009; Hui et al., 2013).

Figure 1.

Community composition under global change. (1) Simplified conceptual model of key community-assembly processes presented as a pipeline from regional to local composition with accompanying reduction in spatial scale. Community composition (patch level) is a complex function of historical and neutral factors, present-day environmental filters, ecological interactions and demographic processes, resulting in communities that bear the signatures of these processes. (2) Examples of how global change is influencing community-assembly processes to divert future composition. The introduction of alien species and extirpation of indigenous species alters the regional pool of species available for sorting. Climate change shifts the broad environmental filter and affects biotic interactions and disturbance regimes. Global changes drive shifts in adaptive traits within species but also affect adaptive potential through changes to population sizes and landscape connectivity. (3) Examples of diverse approaches at different spatial and temporal scales that are used by ecologists to analyse the past, present and future impacts of global change on ecosystem composition. Species-sorting analysis sits in the middle of these scales and, by incorporating shifts in regional species pools and adaptation within species, has the potential for more realistic predictions of future composition that take advantage of these additional information streams.

Here, we develop a conceptual and mathematical model of extended species-sorting in plant communities that captures elements of biogeography, species-sorting and demographic processes. We argue that treating future community composition as the landscape re-sorting of a present-day pool of species, according to fixed species traits, ignores the potential for adaptation and significant local extinction rates.

First, we discuss empirical approaches to examining community composition under global change through case studies. The aim is to demonstrate how biogeography, species-sorting and adaptation are important in determining composition, and to provide concrete examples of the application of empirical methods relevant to the model. Although our model is applicable to any plant community, we focus on the mediterranean biome, because it is relatively well-studied, subject to multiple impacts and provides a coherent, real-world background.

Global change in the mediterranean biome

The mediterranean biome is known for its diverse flora, including numerous global biodiversity hotspots (Myers et al., 2000). The biome has been subject to impacts including altered land use, climate change and cross-introductions of alien species. These impacts threaten ecosystem function and the survival of endemic species (Van de Wouw et al., 2011). Future community composition in the mediterranean biome is a function of ‘disassembly’ (the non-random loss of biodiversity) in response to global change, but also of history and pre-existing drivers (de Bello et al., 2005; Byrne, 2008; Suding et al., 2008; Fig. 1).

Habitat fragmentation in mediterranean ecosystems, caused by land clearing, urbanization and land-use change, exceeds that in all other biomes (Wade et al., 2003). Habitat fragmentation generally affects the demographic trajectories of plant populations by reducing their population sizes, connectivity and genetic diversity (Bagaria et al., 2012; Schoville et al., 2012). It restricts the potential for the dispersal and re-sorting of species and the re-establishment of native ecosystems (Van de Wouw et al., 2011). It has been shown to cause non-random extinction with respect to functional groups (Lavergne et al., 2006; Lindo et al., 2012). Fragmentation has been a relatively recent disturbance in Australia and South Africa, although it began five centuries ago in Chile and California, and millennia ago in the Mediterranean Basin, which retains < 5% of its primary vegetation (Myers et al., 2000).

Invasive alien species can reduce the abundance and diversity of native species in mediterranean ecosystems. The impacts have been less significant in the Mediterranean Basin than in Australia and South Africa (Gaertner et al., 2009). In central Chile, sclerophyllous forests have been displaced by modified woodlands dominated by introduced species (Van de Wouw et al., 2011). Chilean woodlands were opened for grazing and cropping with the arrival of the Spanish in the 16th century, who transformed woodlands into grasslands dominated by species from the Mediterranean Basin (Martín-Forés et al., 2014). Successful exchanges of plants and pathogens within the mediterranean biome have resulted from cross-introductions of species, high human activity levels and the similar climate of formerly isolated regions (Fox, 1990).

Climate change is driving functional responses within species and changing community composition and biome boundaries (Peñuelas & Boada, 2003; Guerin et al., 2012). Although the global geographical extent of mediterranean climates is expected to increase with climate change, in Australia it is predicted to decrease by up to three-quarters this century (Klausmeyer & Shaw, 2009). A well-documented example of changes to the extent of the biome is the expansion of mediterranean vegetation into mountainous, cold-temperate forests in north-eastern Spain since 1945 (Peñuelas & Boada, 2003). Questions remain, however, about the influence of climate change versus land use, which has also changed. Changes have also been observed in individual species. In southern Australia, a decrease in leaf width across a latitudinal cline in the shrub Dodonaea viscosa has been associated with climate change (Guerin et al., 2012). A comprehensive assessment of phenology in the Mediterranean Basin found shifts across ecological communities linked to temperature increases after 1950 (Peñuelas et al., 2002). Not all species will be able to immediately adapt or migrate to track climate change, so that time lags and disequilibrium are expected before changes are realized (Davis, 1986).

Changes in land use, especially to the intensity and extent of agriculture, are particularly relevant for community composition in the Mediterranean Basin and Chile, where grazing pressure has existed for centuries (de Bello et al., 2005; Van de Wouw et al., 2011). Negative consequences have arisen from increased grazing pressure, land abandonment and urbanization (Aronson et al., 1998; Chauchard et al., 2013), such as the encroachment of woody species into grasslands (Bagaria et al., 2012).

The interacting effects of land-use changes and climate change on population demography and species traits (de Bello et al., 2005) are compounded by their effect on fire regimes (Mouillot et al., 2002). For long-lived woody vegetation, the influence of altered fire regimes on community composition may emerge slowly relative to short-term changes in visibility and abundance after fire (Davis, 1986). For example, Franklin et al. (2004) examined chaparral communities in California that had had contrasting fire regimes over 70 years. The relative cover of post-fire seeding versus sprouting species differed between the communities. The lack of species-level compositional change reported by Franklin et al. (2004) may relate to the longevity of the species examined. Although there are useful synergies across mediterranean regions, biogeographical history and land uses differ. Specific knowledge of historical factors and ecosystem drivers is needed for particular regions. There are a range of empirical tools for analysing community composition, including ecological modelling, community phylogenetics and ecological genomics. The application of these tools to species-sorting analysis for a region could improve predictions of future community composition (Table 1).

Table 1. Empirical approaches to examining community composition at different spatial and temporal scales
Assembly processesPossible approachesTarget processesSelected real exampleMain findingFuture challenges
Historical biome assemblyMolecular phylogenetics to examine genetic structure and divergence times with respect to geographical patternsBiogeographical barriers; radiation of lineages; persistence through environmental changeByrne (2008): synthesis of multiple-species molecular phylogeography to infer persistence of lineages through Pleistocene climate oscillations, southern AustraliaEvidence was found for persistent localized refugia generating complex contemporary spatial genetic structure.Harnessing genomic approaches to combine multispecies studies with intraspecific phylogeography; linking biome history explicitly to species re-sorting
Extinction riskModelling extinction risk on retrospective empirical dataNon-random extinction risk based on demographic or life history traitsLavergne et al. (2006): retrospective extinction risk for the Hérault département, mediterranean France; inference of species traits associated with extinctionOver 115 years, 5% of species became regionally extinct. Extinction was more likely in hydrophytic/water-dispersed species, annuals, particular taxonomic groups, and species at the southern margin of a broader geographical range.Linking correlation between functional groups and extinction risk to specific mechanisms, to make the links more predictive
Sorting of species within landscapesMeasuring community composition and traits along environmental/disturbance gradients; multitable ordination or ordination–regression, multivariate binomial models, permutation tests and statistical mechanicsInfluence of environment on species occurrences versus biotic interactions and neutral processesDe Bello et al. (2005): ordination–regression tree approach to test the interaction of climate filtering of functional traits and grazing intensity gradients on species composition in rangelands, north-east SpainTraits predicted responses of species and communities, but predictions depended on synergistic effects of climate and grazing regimes.Incorporating intraspecific phenotypic shifts and shifting species pools (e.g. extinction risk) into species-sorting analysis
Demographic responsesRetrospective analysis of shifts in adaptive traitsAdaptive gene migration, selection or plasticityThompson et al. (2013): retrospective comparison of genetic polymorphism controlling essential oil chemotype in Thymus vulgaris, Saint-Martin-de-Londres basin, mediterranean FranceA reduction in winter freezing over time and across spatial climate gradients was associated with decreased frequency of freezing tolerant phenotypes via gene flow among populations.Linking genome–environment associations to changes in functional phenotypic traits and fitness in non-model species and disassociating confounding factors

Assembly-Process Case Studies

Continental biogeographical and phylogeographical context

The context for global change community ecology is the set of species available for sorting. Phylogeography addresses the historical persistence of species based on the concept that historical range shifts create genetic signatures. Biogeography analyses taxa above the species level in an explicit spatial context in order to infer radiation and dispersal events (Buerki et al., 2012). High-throughput genomic approaches have the potential to link multiple-species biogeography with finer-scale phylogeography (Puritz et al., 2012). Studies on mediterranean ecosystems have mostly used older technologies but nevertheless provide useful insights.

Buerki et al. (2012) used chloroplast markers to examine the global radiation of Hyacinthaceae. Their methods accounted for uncertainties in phylogenetic reconstruction and molecular-clock divergence times, and they constrained their results according to palaeogeographical evidence. For Hyacinthaceae, Quaternary climate oscillations generated species diversity in situ in mediterranean regions of South Africa and the Mediterranean Basin. The reconstruction contrasted the role of adaptive radiation with recent dispersal.

Byrne (2008) synthesized the phylogeography of multiple species in southern Australia, to test for common responses to past climate change. With molecular clocks, she concluded that diversification occurred in the late Tertiary. During the Pleistocene, the species persisted through glacial cycles and a trend towards aridity. The phylogeographical studies revealed intraspecific genetic structure with strong geographical patterns. The emergence of the genetic structure coincided with Pleistocene climate oscillations. For example, geographical associations were present within the phylogeny of Acacia acuminata in Western Australia (Byrne et al., 2002). Evidence for past fragmentation of the species' range was consistent with climatic instability during the Pleistocene. Comparable patterns across species demonstrated the presence of multiple refugia occupied by different populations of the same species during glacial maxima. In contrast, a widespread genotype and a single localized centre of higher genetic diversity have been observed in Santalum spicatum in Western Australia (Byrne, 2008), consistent with a single refugial population.

Phylogeography relates history to present-day distributions and genetic diversity. Phylogeographical studies can locate areas where species are likely to be more resilient to climate change, based on historical persistence and genetic diversity.

Regional extinction risk

Global change may cause regional species loss. Lavergne et al. (2006) took an empirical, retrospective approach to assessing regional extinction risk. They compared historical and recent plant species records for the Hérault département of mediterranean France, to detect species with altered regional abundance, or that became extinct. Over 115 years, 5% of species became regionally extinct. Changes in occurrence were strongly biased by life history, taxonomic group, ecological niche and biogeographical affinity. The extinction of initially rare species was more likely in hydrophytic or water-dispersed species, annuals, and species at the southern margin of a broader range. Identifying the risk factors for species declines provides predictive power for non-random extinction (Lindo et al., 2012).

Demography and functional shifts

It has been proposed that stabilizing selection may result from species-sorting among habitat patches (Ackerly, 2003). Nevertheless, phenotypic shifts within species could limit the re-sorting of different species among habitat patches during environmental change (Suding et al., 2008; Fig. 1). The assumption of static species traits in sorting analyses ignores the potential of phenotypic shifts as a mechanism regulating species turnover. Approaches to exploring phenotypic responses to global change have involved retrospective or spatial analysis of functional or genetic changes and experimental manipulation.

Thompson et al. (2013) sampled essential-oil chemotypes and polymorphisms in chemotype genetic basis in Thymus vulgaris in the Saint-Martin-de-Londres basin (southern France), 36 years apart. During this time, the intensity and frequency of winter freezing events declined. The frequency of freezing tolerant and non-tolerant phenotypes was associated with both spatial and temporal changes in climate. Freezing-tolerant phenotypes decreased in frequency due to gene flow among populations and relaxed selection on freezing tolerance.

While it is informative to link known adaptive polymorphisms to climate change, a greater challenge is to find the genetic basis for shifts in important quantitative traits and fitness. Ecological genetics can explore adaptive potential and signatures of adaptation. Identifying adaptive genetic variation in non-model species can be technically difficult, especially if they are long-lived. Complexity reduction, such as mass sequencing of AFLP products or transcriptomes, is a current focus of landscape genomics that aims to overcome this hurdle (Schoville et al., 2012). There are, however, challenges in inferring the genetic signatures of selection from genomic data. For example, there is the potential for spatial autocorrelation of neutral genetic variants with environmental gradients. Space and environment can be dissociated with well-planned field sampling and the inclusion of spatial covariates in analysis (Schoville et al., 2012). In principle, loci under selection that have been identified from correlative landscape genetics should be linked to changes in phenotype or fitness to support adaptation (Schoville et al., 2012). In practice, such links are rarely demonstrated. Fournier-Level et al. (2011) tested the genomic association between climate and fitness of ecotypes in common gardens for a model species (Arabidopsis thaliana) in Europe. Phenotypes were linked to known genes and gene regions, providing a powerful example of local adaptation being linked directly to fitness. A broader application of ecological genomics requires methods that can be readily applied to non-model species.

Intraspecific phenotypic clines along environmental gradients and phenotypic responses to climate change have been documented (Peñuelas et al., 2002; Guerin et al., 2012; Thompson et al., 2013). The assumption of static species traits in sorting analysis may be an oversimplification and phenotypic variance and adaptive potential would ideally be considered.


Functional traits determine whether species can persist in particular environments (Lavergne et al., 2006; Soliveres et al., 2012). Communities distributed along environmental or disturbance gradients have patterns of species composition, associated with traits and phylogenetic relatedness (Fine & Kembel, 2011). Species-sorting analysis predicts species-level responses to changes in factors such as climate or grazing intensity. Shifts in composition, traits or phylogenetic groups have been demonstrated with respect to spatial changes in climate, habitat fragmentation and disturbance levels (de Bello et al., 2005; Cornwell & Ackerly, 2009; Bagaria et al., 2012; Guerin et al., 2013). Here, we review methods and case studies related to species-sorting. We show that species-level responses to global change are well documented, and give an overview of the analyses that are currently possible.

Ordination reduces the dimensionality of multivariate data (de Bello et al., 2005; Bagaria et al., 2012; Guerin et al., 2013) and can link species' traits to environmental variables (Bagaria et al., 2012). Using ordination, Bagaria et al. (2012) found that community composition in remnant grasslands in southern Catalonia was influenced by the contemporary size and isolation of habitat patches. The configuration of the same habitat patches 50 years earlier, assessed from aerial photographs, had no influence on community composition. These results suggest rapid community responses to fragmentation over 50 years, although they did not account for differences in soils and grazing intensity.

Ordination–regression approaches involve ordination of multivariate species occurrences, followed by regression of the ordination coordinates against environmental variables (Kleyer et al., 2012). De Bello et al. (2005) used canonical correspondence analysis (CCA) followed by regression trees to test which functional traits predicted the CCA coordinates. This tested whether traits independently predicted where species were located in environmental space. With this method, de Bello et al. (2005) found that traits predicted the response of species composition to grazing intensity in rangelands of north-eastern Spain. The traits that were most predictive varied, however, along elevational climate gradients. This highlights the need to consider synergies, and to collect empirical data for specific ecosystems. Guerin et al. (2013) used over 3500 vegetation inventory plots in southern Australia to predict species turnover under climate change using space-for-time substitution. They detected changes in the rate of species turnover along spatial climate gradients using CCA ordination, followed by cubic spline regression of the coordinates. They concluded that community composition was sensitive to climate, and that the ecotone between mesic and arid ecosystems was the most sensitive. One drawback of complexity reduction through ordination is that transformation of raw community composition data is subject to assumptions. For example, linear and unimodal species responses to gradients are assumed in RDA and CCA, respectively (Guerin et al., 2013). Jamil et al. (2013) proposed a model-based method for testing trait–environment–species occurrence patterns. Species' traits, environmental variables and their interactions were included as predictors in a community-level binomial generalized linear mixed model that predicted multivariate species occurrences. Although this method does not initially transform the input data, the results are less readily visualized and the model imposes assumptions around species responses. More flexible GLM approaches are possible, but may be limited by computational demands.

Statistical mechanical approaches to species-sorting begin with predictable, macroscopic properties of communities – such as constraints on community mean traits – and then calculate the most probable species composition (Shipley et al., 2006). To some degree, this by-passes the difficulty in making predictions for individual species. Shipley et al. (2006) applied statistical mechanics to a successional chronosequence of abandoned vineyards in southern France. They determined species compositions for habitat patches that maximized the Shannon entropy, subject to predicted abiotic constraints on community mean traits. In a statistical mechanics framework, this is the most likely composition in the absence of further information. Statistical mechanics can link shifting species pools with species-sorting by setting environmental limitations on communities before predicting species occurrences, but this needs further empirical testing (Shipley et al., 2006).

Current methods for inferring trait–environment relationships are able to predict with reasonable accuracy how a given set of species is sorted across a present-day landscape. To extend predictions to temporal change, the potential for extinctions, introductions and shifting species traits should also be considered, a framework which we outline below.

An Extended Species-Sorting Framework

Plant communities have well-documented environmental constraints on species richness, biomass, and the mean, variance and dispersion of community traits (Shipley et al., 2006; Cornwell & Ackerly, 2009). Community-level responses to global change provide a tractable means for framing complex responses among individual species (Shipley et al., 2006). We outline an extended species-sorting framework based on shifting community constraints. Notation is summarized in Table 2.

Table 2. Notation used in equations describing changes in a functional trait within a habitat patch
Parameters h2narrow sense heritability of z
i, jspecies and habitat patch, respectivelyδmean migration distance along χ per unit time
zvalue of a given trait; over-bar indicates community mean; over-hat indicates population meannpopulation size
θoptimum value of zTaverage generation time
prelative (proportional) abundanceMathematical operators 
aset of species that persistx Δ ySymmetrical difference between sets x & y (members of either set that are not members of both sets)
rset of species involved in replacement from the initial metacommunityx ∪ yUnion of sets x & y (all members of both sets)
uset of species involved in replacement due to introduction or extinctionx ∩ yIntersection of sets x & y (members of x and y that are members of both sets)
lset of all species occurring (combination of a, r and u)x ∈ yMembers of set x that are members of set y
χan environmental gradient/variableΣSum of specified elements
γstabilizing selection math formula Partial derivative: the instantaneous rate of change in x with respect to y, when x may also change with respect to other variables
σ2phenotypic variance of z math formula Second-order partial derivative

The abundance-weighted community mean of a trait is given by:

display math(1)

where math formula is the community mean value of trait zi, pi is the proportional abundance and zi is the trait value of the i-th of n species within a community (Ackerly & Cornwell, 2007). Let χ be an environmental gradient that constrains a community-level cline in z, so that math formula changes at rate math formula. Assuming an equilibrium start, a shift in χ will result in a new community constraint math formula and therefore a constraint differential math formula. We allow math formula to track math formula through time in response to math formula. The within-community contributions to a shifting math formula can be broken down into interspecific (changing relative species abundances) and intraspecific (changing species' traits) components (Norberg et al., 2012; Fig. 2):

display math(2)
Figure 2.

Expressions of shifting functional traits within a plant community. (a) The abundance-weighted community mean of a functional trait shifts with environment. There are three ways for species composition to adjust to the new constraint: (b) shifts in the trait values of individual species to allow stable relative abundance (intraspecific shifts can be achieved via plasticity, selection or migration of genes from a spatial cline); (c) changed in situ relative abundance of species; (d) species replacement: species that are unable to adapt or maintain viable populations become locally extinct and other pre-adapted species may disperse in. Species-sorting analyses that include (c) and (d) but exclude (b) may overestimate species turnover to achieve shifting community constraints. Extinction from and introduction into the metacommunity may be involved in species replacement but are rarely considered in species-sorting analysis.

The interspecific term of Eqn (2) (first term on right-hand side) can be further broken down into three components: changes in the proportional abundance of species that are already present; species replacement, involving the loss or gain of species from the existing metacommunity; and species replacement involving regional extinction or the gain of species newly introduced to the metacommunity. It is useful to break down the interspecific term to explicitly recognize that changes in relative abundance may limit species turnover and that species turnover can involve a shifting species pool and not just the re-sorting of existing species. We define interspecific components as follows.

Let S0 be the set of all species present in a metacommunity at time 0, and sj0 the subset of species in S0 present in the j-th habitat patch with proportional abundance of the i-th species pij, so that sj0 = {S0|pij0 > 0}. The set of species present in the metacommunity is allowed to change over time, as is the subset that occurs in the j-th habitat patch. Species extinct from the metacommunity at time t are the set Et = S0 −St and newly introduced species are the set It = St −S0. Species that persist in the metacommunity are the set Pt = S0 ∩ St. Similarly, the set of species that have persisted in the j-th patch is ajt = sj0 ∩ sjt, and their proportional abundance is allowed to differ between times 0 and t at rate ∂pij/∂t. Species replacement in patch j from among the initial metacommunity is the set: rjt = (sj0Δsjt) ∈ Pt, and replacement that is related to introduction or extinction at the metacommunity level is the set: ujt = (sj0Δsjt) ∈ (Et ∪ It). Finally, the complete set of species occurring in patch j at either time is: lj = sj0 ∪ sjt. Putting these components together, we can break down the species turnover term of Eqn (2) for patch j:

display math(3)

Without the intraspecific term (second term on right-hand side) of Eqn (2), only changes to species abundances and species replacement from a shifting metacommunity (Eqn (3)) are allowed to contribute to a changing community trait mean. To be more explicit about how intraspecific trait variation contributes to a shifting community mean, we break down the intraspecific term of Eqn (2) further.

A shifting intraspecific population mean of trait z can be described as a function of density-dependent genotype re-sorting among habitat patches, selection and plasticity (Pease et al., 1989; Chevin et al., 2010). Let the population mean of trait z for a given species be denoted math formula to avoid confusion with the community mean math formula. Let math formula be clinal with respect to our environmental gradient χ so that math formula changes at rate math formula. Let δ2 be the mean squared migration distance along χ per unit time, n the population size, h2 the narrow-sense heritability of z, σ2 the phenotypic variance of z, γ stabilizing selection and T generation time. Let θ be the optimal value of math formula (i.e. the value that maximizes fitness with respect to χ) and η the plastic response of z per unit of change in χ. We then give the rate of phenotypic change within a species as:

display math(4)

The first two terms on the right-hand side describe genotype sorting (see Pease et al., 1989, for derivation). The first term is the diffusion equation describing migration, and the second is the density-dependent (source–sink) mixing of populations along the cline. Greater dispersal velocity, larger neighbouring populations and larger clines generate larger trait shifts due to genotype sorting among habitat patches. The slope of the cline is here defined as the rate of change in the actual, rather than optimum, phenotype. The third term quantifies selection and is derived in Appendix S1. The trait changes faster with greater selection pressure (difference between math formula and θ and/or lower stabilizing selection) and additive genetic variance associated with z (product of phenotypic variance and heritability), and slower with generation time. The fourth term represents phenotypic plasticity. Here, η could be a constant (within certain limits of χ) or some function of χ.

Generalizing Eqn (4) to the multivariate (community) case, the intraspecific term of Eqn (2) can be replaced by:

display math(5)

Eqn (2) – and expanded terms in Eqns (3) & (5) – provide a community-mechanics perspective for including shifting species pools and species traits within an extended trait-based species-sorting framework (see Appendix S1 for the full equation). We have argued that the environment places community-level constraints on traits and that it is useful to break the contributions to a shifting constraint from species turnover and intraspecific change down into specific parts. Interspecific competition is implicit in the model, as it influences the species abundance terms. Intraspecific competition is implicit, in the trait evolution term. The relative contribution of interspecific and intraspecific processes is discussed in Appendix S1. The potential for metacommunity composition to shift through time is implicit in this approach. We discuss shifts in metacommunity composition due to extinction from a trait-based perspective in Appendix S1.

The model in practice

Useful traits in the context of our model must be ecologically relevant, measurable at the community level and variable within and between species. Candidates include leaf size, leaf nitrogen (N) and phosphorus (P) concentrations, specific leaf area (SLA), plant height, seed size/mass and reproductive phenology (Ackerly, 2004; de Bello et al., 2005; Shipley et al., 2006; Bagaria et al., 2012). We recommend SLA as a first consideration. SLA is frequently reported in the literature, feasible to measure at the community level and varies widely. SLA has the advantage of ecological interpretation in terms of growth strategy and resource investment (Ackerly, 2004). Leaf N and P concentrations also relate to growth strategy and resource investment (Ackerly, 2004); although they are more costly to measure than SLA, chemical concentrations can be measured efficiently from many individuals, including from dried field samples. Reproductive traits are informative but potentially more difficult to score than vegetative traits across communities with varied phenology.

For many regions, climate change is the obvious candidate for a ‘shifting gradient’ (i.e. χ) that drives changes in community composition. In the Mediterranean Basin, where the main impact on community composition has been altered land use, disturbance gradients may be more relevant. Other impacts to existing gradients include salinization and eutrophication. Habitat fragmentation is important to real-world applications of the model. Although not addressed explicitly in the mathematical formulation, it may directly influence several parameters via effects on genotype sorting, dispersal and genetic variance.

With the example of directional changes in climate or disturbance in fragmented landscapes, the model is relevant to predictions of future community composition. In the case of statistical (community) mechanics (Shipley et al., 2006), the prediction of community composition could reasonably allow for variance in species traits, and the likelihood of species extinction from the system over time. Our model is, however, intended to improve conceptual approaches and not to provide an analytical method for immediate application to real data. Empirical methods could incorporate elements of the model depending on data availability. Data for predicting the influence of traits and environment on community composition could include intraspecific variance (or multispecies phylogeography, clinal variation or fitness trials), habitat connectivity (e.g. spatial applications or landscape genetics) and extinction risk (e.g. retrospective analysis or modelling).


Much of the variation in plant community composition within mediterranean landscapes can be explained by the correlation between present-day environments and species traits or, alternatively, ecological drift. In our case studies, de Bello et al. (2005) found that 21%–27% of compositional variance was explained by grazing regime, while Bagaria et al. (2012) found that 45% of variance in species frequencies was explained by climate and habitat configuration. Such analyses address the sorting of a given set of species with fixed traits among habitat patches. However, the set of species available for sorting is determined by history as much as by today's environment. Additionally, intraspecific trait variation can account for 30% or more of community-level variance (Prior et al., 2003; Cornwell & Ackerly, 2009; Albert et al., 2010).

There have been two major assumptions in species-sorting analyses:

  1. Extinction and speciation (or long-distance immigration) occur over evolutionary time-scales and are negligible over ecological time-scales. This assumption should be questioned in the context of regional extinctions and the introduction of new species (Lavergne et al., 2006). A shifting species pool has broader implications for predicting community composition than the deterministic re-sorting of species that are present today.
  2. Adaptation within species is negligible compared to interspecific differences and can be ignored in community trait–environment analyses (Shipley et al., 2006). Despite theoretical recognition that the interspecific competition that limits niche overlap can involve both species replacement and trait evolution, and that traits vary within species across environments (Pease et al., 1989), species-sorting analyses almost exclusively consider only interspecific responses (Shipley et al., 2006; Albert et al., 2010; Jamil et al., 2013).

Environmental change promotes disequilibrium because of time lags between perturbation and composition shifts (Davis, 1986). Range shifts are limited by dispersal, competition and longevity. Because of time lags and historical effects, historical data, real-time observation and prediction together would make the most informed assessment (Davis, 1986; Chauchard et al., 2013). The synthesis of biogeography with landscape ecology could provide a more realistic context for global change, but remains a technical and conceptual challenge.

In conclusion, the prediction of plant community composition under global change could be improved by linking shifting species pools (extinctions and introductions), landscape species-sorting processes and phenotypic responses within species. Species-sorting analysis has the potential to be extended to accommodate the influence of adaptation (or at least trait variance) on species’ occurrences and regional persistence. Our model, based on shifting community constraints, outlines a conceptual framework that could underlie such a development in empirical ecology. The model predicts diverse responses to environmental change among ecosystems, depending on the relative contribution of species replacement to shifting community constraints. It therefore implicitly recognizes that changes to in situ species relative abundance, and trait evolution, are resilience mechanisms. Subsequently, the rate of species replacement will depend on the adaptive potential of species, competition between colonists and established individuals, physical disturbance and habitat fragmentation.


Funding and support were provided by the Australian Research Council (Linkage Project LP110100721; Super Science Fellowship FS110200051), South Australian Premier's Science and Research Fund and the Terrestrial Ecosystems Research Network.

Greg Guerin is a postdoctoral researcher at the University of Adelaide. His work focuses on spatial and temporal influences of climate and landscape on vegetation composition and species traits. Approaches to this research include retrospective analysis of historical datasets, spatial analysis and monitoring.

Author contributions: All authors contributed to the framework, content and text. G.R.G. drafted the manuscript and developed the model.