Impact statement

I review different approaches for detecting extinction risk from climate change using species traits. This identified vital roles for process-explicit models in enriching knowledge of functional and life history traits that regulate species’ responses to shifting climates. More frequent application of these process-driven approaches to natural history and resurvey data will better identify species that are most vulnerable to climate change.

Trait-based vulnerability to extinction

Evidence suggests that measurable characteristics of organisms—so called species’ biological traits—regulate the ecological processes of population declines and loss of species distributions (Chichorro et al., Reference Chichorro, Juslén and Cardoso2019; Kotiaho et al., Reference Kotiaho, Kaitala, Komonen and Päivinen2005; Purvis et al., Reference Purvis, Gittleman, Cowlishaw and Mace2000). Ecological theory predicts that risks of extinction should be higher for animals with larger body sizes, smaller ranges, smaller population sizes and poorer dispersal abilities (McKinney, Reference McKinney1997; Pimm et al., Reference Pimm, Jones and Diamond1988). Larger-bodied species tend to have slower life histories, smaller population densities and bigger home ranges, making them particularly susceptible to increased human-driven mortality (Ripple et al., Reference Ripple, Wolf, Newsome, Hoffmann, Wirsing and McCauley2017). Plants and animals with smaller ranges and population sizes are often more vulnerable to stochastic events (Matthies et al., Reference Matthies, Bräuer, Maibom and Tscharntke2004), while species with reduced dispersal capacities are often more prone to climate and environmental disturbances because they are unable to move large distances (Chichorro et al., Reference Chichorro, Urbano, Teixeira, Väre, Pinto, Brummitt and Cardoso2022). Although these traits (supported by theory) are relatively easy to measure or derive, comparative analyses of extinction risk have shown that correlates of vulnerability to population declines and range contractions vary widely, often being idiosyncratic and context-dependent (Davidson et al., Reference Davidson, Hamilton, Boyer, Brown and Ceballos2009; Fisher and Owens, Reference Fisher and Owens2004; Owens and Bennett, Reference Owens and Bennett2000).

Accordingly, empirical studies have yielded opposing explanations for biological mechanisms underlying patterns of extinction risk (Beissinger, Reference Beissinger2000) and the central question of what ecological characteristics make certain species more vulnerable to population declines and extinctions has never been fully resolved. What is clear, however, is that the ecological mechanisms that put species at risk of extinction are complex, and that they vary depending on the threatening process or combination of processes, including habitat loss, overexploitation and invasive species (González-Suárez et al., Reference González-Suárez, Gómez and Revilla2013; Owens and Bennett, Reference Owens and Bennett2000). This makes predicting ecological outcomes (e.g., population declines, distributional shifts) from species traits difficult, particularly since threatening processes tend to operate in unison, often having synergistic effects (Brook et al., Reference Brook, Sodhi and Bradshaw2008).

Linking species traits to climate change responses

While it is estimated that one in six species is at risk of extinction from human-driven climate change (Urban, Reference Urban2015), the proximate causes of observed population declines and range contractions in shifting climates remain poorly understood. Consequently, knowledge of how species traits mediate climate-driven changes in the abundances and distributional boundaries of plants and animals is unclear (Angert et al., Reference Angert, Crozier, Rissler, Gilman, Tewksbury and Chunco2011; Beissinger and Riddell, Reference Beissinger, Riddell and Futuyma2021; Cahill et al., Reference Cahill, Aiello-Lammens, Fisher-Reid, Hua, Karanewsky, Yeong Ryu and Wiens2013). The reasons include a general lack of information on functional and life-history responses to climate change for most species (Garcia et al., Reference Garcia, Cabeza, Rahbek and Araújo2014b), and the mainly statistical methods being used to analyse the scarce amount of relevant data that is available (Green et al., Reference Green, Brookson, Hardy and Crowder2022).

Most predictions of vulnerability to climate change are done correlatively—not mechanistically—mainly using statistical-based ecological niche modelling approaches (Elith and Leathwick, Reference Elith and Leathwick2009). Here, statistical relationships between aspects of climate and species’ occurrences are used to estimate the climatic conditions needed to maintain viable populations of a species today and how these may shift in the future (Araujo and Peterson, Reference Araujo and Peterson2012). Modest data requirements, and simplifying assumptions, enable the potential impact of future climatic change on the distributions of thousands of species to be explored efficiently (Warren et al., Reference Warren, Price, Graham, Forstenhaeusler and VanDerWal2018), providing useful statistical tools for forecasting impacts and designing conservation interventions (Guisan et al., Reference Guisan, Tingley, Baumgartner, Naujokaitis-Lewis, Sutcliffe, Tulloch and Buckley2013). Spatial projections of climate change exposure from ecological niche models have also been used to match species’ traits to responses to shifting climates (Garcia et al., Reference Garcia, Araújo, Burgess, Foden, Gutsche, Rahbek and Cabeza2014a). However, the capacity of niche modelling approaches to directly establish species traits that increase extinction risk from future climate change is more limited (Fordham et al., Reference Fordham, Akçakaya, Araújo, Keith and Brook2013a). This is because they predict potential exposure to climate change—based on the availability of climatically suitable habitat—not extinction risk, with no direct limitations imposed by species traits. Although, new approaches, which enable phenotypic plasticity and local adaptation of fitness-related traits to be accounted for in ecological niche models, are enabling climate-survival responses to be better captured in projections of vulnerability to climate change (Benito Garzón et al., Reference Benito Garzón, Robson and Hampe2019).

An alternative way to establish trait-based vulnerabilities of species to future climate change is to quantify their past effects on local extinctions, and distributional shifts using statistical models (Pacifici et al., Reference Pacifici, Foden, Visconti, Watson, Butchart, Kovacs and Rondinini2015) (Table 1). Numerous studies have now used natural history collections and repeated surveys to statistically relate documented shifts in species geographic ranges to recent climatic change (Bradshaw et al., Reference Bradshaw, Brook, Delean, Fordham, Herrando-Pérez, Cassey and Araújo2014; Estrada et al., Reference Estrada, Meireles, Morales-Castilla, Poschlod, Vieites, Araújo and Early2015; Moritz et al., Reference Moritz, Patton, Conroy, Parra, White and Beissinger2008; Tingley et al., Reference Tingley, Koo, Moritz, Rush and Beissinger2012). When analysed together, using meta-analysis techniques, these studies of species range shifts can be linked to inter-specific variation in traits, allowing ecological function to be integrated into assessments of species’ vulnerability to climate change, and resultant conservation priority schemes (MacLean and Beissinger, Reference MacLean and Beissinger2017). This is done by asking, What are the ecological implications of species functional and life-history traits in changing climates for the fitness and persistence of species?

Table 1. Modelling approaches for identifying biological traits that predict vulnerability to climate change

Potential modelling approaches are statistical, experimental, biophysical, individual-based or population-based. Modelling approach determines capability to detect important functional as well as life-history traits, and establish whether links to climate shifts are based on causation or not. Most approaches are spatially explicit, while some approaches can disentangle trait-based responses at large (geographical) extents and in response to multiple (extrinsic) drivers of extinction risk that often interact synergistically. Note that “x” indicates that a function is present.

While these meta-analyses have detected negative impacts of human-induced climatic change on population declines and range shifts (Lenoir et al., Reference Lenoir, Bertrand, Comte, Bourgeaud, Hattab, Murienne and Grenouillet2020; Poloczanska et al., Reference Poloczanska, Brown, Sydeman, Kiessling, Schoeman, Moore and Richardson2013), they have not been able to consistently link variation in observed responses to climatic change to different functional and life history traits (Angert et al., Reference Angert, Crozier, Rissler, Gilman, Tewksbury and Chunco2011; Beissinger and Riddell, Reference Beissinger, Riddell and Futuyma2021; MacLean and Beissinger, Reference MacLean and Beissinger2017). This is probably because of a combination of (i) methodical issues when quantifying population declines and range shifts (Beissinger and Riddell, Reference Beissinger, Riddell and Futuyma2021); (ii) difficulties in obtaining precise measurements of a wide enough spectrum of species traits (Schleuning et al., Reference Schleuning, Neuschulz, Albrecht, Bender, Bowler, Dehling and Kissling2020); and (iii) problems with disentangling complex non-linear trait-based responses to climatic change, and their interplay with other anthropogenic impacts (Taheri et al., Reference Taheri, García-Callejas and Araújo2021).

Estimates of range shifts and changes in population trajectories generally have inherent uncertainties owing to imperfect detection and longitudinal changes in sampling effort and methods, which makes trait-based inferences often challenging (Hébert and Gravel, Reference Hébert and Gravel2023). Moreover, commonly available functional and life history traits used in attribution studies of changes in distribution and abundance—namely, body size, fecundity and habitat breadth—are likely to only partially account for processes underpinning observed responses of species to climatic change (Schleuning et al., Reference Schleuning, Neuschulz, Albrecht, Bender, Bowler, Dehling and Kissling2020). These issues, and a tendency for studies to use analytical techniques that assume linear relationships between species traits and demographic changes (Beissinger and Riddell, Reference Beissinger, Riddell and Futuyma2021), have meant that the biological traits of species identified as predictors of range shifts and local extinctions under climate change often differ widely between studies (Wheatley et al., Reference Wheatley, Beale, Bradbury, Pearce-Higgins, Critchlow and Thomas2017). This is seen even for assemblages of similar species in comparable ecosystems (Pinsky et al., Reference Pinsky, Worm, Fogarty, Sarmiento and Levin2013; Sunday et al., Reference Sunday, Pecl, Frusher, Hobday, Hill, Holbrook and Bates2015). Another important issue is that climatic changes over attribution time periods are generally small compared to the effects of other human pressures (habitat loss and degradation, over exploitation etc.), making it difficult to confidently pinpoint species’ traits associated with climate vulnerability (Fordham et al., Reference Fordham, Akcakaya, Alroy, Saltre, Wigley and Brook2016).

Alternatively, small-scale field experiments, laboratory microcosm and larger-scale mesocosm experiments allow direct testing of the importance of variation in functional and life history traits in fluctuating environments, potentially improving knowledge of how future climate change will affect species’ persistence, the structure and function of ecological communities, and the food webs that they are embedded in (Stewart et al., Reference Stewart, Dossena, Bohan, Jeppesen, Kordas, Ledger, Woodward, Woodward and O’Gorman2013). They do this by providing tractable yet ecologically realistic bridges between simplified experimental conditions and the real world (Fordham, Reference Fordham2015). These experimental approaches indicate that trophic position, behaviour and life-history characteristics can all influence responses to climatic warming (Bestion et al., Reference Bestion, Teyssier, Richard, Clobert and Cote2015; Ullah et al., Reference Ullah, Nagelkerken, Goldenberg and Fordham2018; Yvon-Durocher et al., Reference Yvon-Durocher, Allen, Cellemare, Dossena, Gaston, Leitao and Trimmer2015). While experiments can provide valuable information on trait-based responses to climate change that cannot be readily quantified from field-based surveys (Ullah et al., Reference Ullah, Fordham, Goldenberg and Nagelkerken2024), their insights are most relevant to the climate and environmental conditions under which the study was conducted, which can be vastly different to local and regional future conditions (De Boeck et al., Reference De Boeck, Vicca, Roy, Nijs, Milcu, Kreyling and Beier2015). Moreover, because experiments are difficult and expensive to construct and maintain, experimental inferences of climate change responses and vulnerabilities are available for only a narrow range of largely short-lived species, based on experiments that unavoidably simplify climatic forcing (Fordham, Reference Fordham2015).

While growth in experimental and observation data of species responses to climate change—and their synthesis using metanalysis—will undoubtably improve the taxonomic, spatial and temporal scale of trait-based vulnerability assessments, process-explicit simulation models provide an alternative and promising methodology for directly identifying ecological mechanisms of extinction and generating trait-based predictions of climate change vulnerability (Green et al., Reference Green, Brookson, Hardy and Crowder2022). By generating spatially and temporally explicit projections of range and abundance, these mechanistic approaches enable important interactions between spatiotemporal drivers and demographic, physiological and behavioural responses to be disentangled (Pilowsky et al., Reference Pilowsky, Colwell, Rahbek and Fordham2022a).

Process-explicit models

Species attributes that increase their risk of extinction from climate change can be dissected using process-driven models that simulate complex interplay between extrinsic (climatic and environmental) drivers and intrinsic (demographic, biophysical and behavioural) factors. For example, by simulating movement, mortality and reproduction in networks of populations across time, spatially- and temporally-explicit population models enable demographic responses to multiple and interacting anthropogenic impacts to be established (Fordham et al., Reference Fordham, Akcakaya, Brook, Rodriguez, Alves, Civantos and Araujo2013b). These process-explicit approaches have shown that vulnerability to extinction from future climate change can be predicted using life-history traits (e.g., generation length, growth rate, natal dispersal distance) and other ecological characteristics (e.g., range size, population size, niche breadth) that affect organismal performance (Pearson et al., Reference Pearson, Stanton, Shoemaker, Aiello-Lammens, Ersts, Horning and Akcakaya2014). Likewise, biophysical models that capture the exchange of energy and mass between an organism and its environment have identified functional traits (e.g., body surface area, basal metabolism, fur depth and density) and behavioural mechanisms (e.g., dispersal time, habitat use) that are likely to regulate species vulnerability to future climate change (Briscoe et al., Reference Briscoe, McGregor, Roshier, Carter, Wintle and Kearney2022; Kearney et al., Reference Kearney, Jusup, McGeoch, Kooijman and Chown2021). In biophysical approaches, the connection between trait and function is explicit. However, for population-based approaches the link is made without direct connections to species’ morphology, physiology, or behaviour (Table 1).

Biophysical models are particularly suited to providing robust predictions of range limits in new climates (Briscoe et al., Reference Briscoe, Elith, Salguero-Gómez, Lahoz-Monfort, Camac, Giljohann and Guillera-Arroita2019), and are now often used to establish functional traits that are likely to affect the distributional movements of species under future climate change (Briscoe et al., Reference Briscoe, Kearney, Taylor and Wintle2016; Kearney et al., Reference Kearney, Porter, Williams, Ritchie and Hoffmann2009; Levy et al., Reference Levy, Buckley, Keitt, Smith, Boateng, Kumar and Angilletta2015). These include the effects of shape, size, surface area and insulation on thermal tolerance (Kearney et al., Reference Kearney, Jusup, McGeoch, Kooijman and Chown2021), and the importance of different life stages for fitness (Levy et al., Reference Levy, Buckley, Keitt, Smith, Boateng, Kumar and Angilletta2015). Alternatively, spatially and temporally explicit population models provide computational frameworks that simulate geographic and demographic declines from multiple and interacting environmental threats (Fordham et al., Reference Fordham, Bertelsmeier, Brook, Early, Neto, Brown and Araújo2018). This makes them particularly suitable for establishing species attributes that increase risk of population decline and extinction from multiple drivers of global change (Green et al., Reference Green, Brookson, Hardy and Crowder2022). Individual-based modelling approaches are, in theory, better suited to identifying biological traits associated with vulnerability to climate change, because they simulate variation in trait values at the individual level (DeAngelis and Mooij, Reference DeAngelis and Mooij2005). However, these approaches are computationally demanding, making them challenging to run at large geographic extents (Table 1). Nevertheless, they have established the importance of life-history and behavioural traits (particularly those mediating dispersal) in response to climate and environmental disturbance (Bocedi et al., Reference Bocedi, Zurell, Reineking and Travis2014b).

Pattern-oriented detection of traits

A scarcity of ecological data on species’ responses to climate change has long been viewed as a barrier to a wider application of biophysical, population- and individual-based models in biodiversity and climate change research, including trait-based analyses (Urban et al., Reference Urban, Bocedi, Hendry, Mihoub, Pe’er, Singer and Travis2016).

Pattern-oriented modelling uses observations as filters for evaluating whether an ecological model is adequate in its structure and parameterisation to simulate biological processes (Pilowsky et al., Reference Pilowsky, Haythorne, Brown, Krapp, Armstrong, Brook and Fordham2022b). Shifts in species geographic ranges and abundances are often used as targets in pattern-oriented analysis, enabling population-based, individual-based, and biophysical models to be built and optimised using parameters with wide but plausible ranges (Fordham et al., Reference Fordham, Brown, Akçakaya, Brook, Haythorne, Manica and Nogues-Bravo2022; Grimm et al., Reference Grimm, Revilla, Berger, Jeltsch, Mooij, Railsback and DeAngelis2005; Strubbe et al., Reference Strubbe, Jiménez, Barbosa, Davis, Lens and Rahbek2023), including for data depauperate, vulnerable species (Pilowsky et al., Reference Pilowsky, Brown, Llamas, van Loenen, Kowalczyk, Hofman-Kamińska and Fordham2023). These studies, using process-explicit models and pattern oriented methods, have revealed the importance of life-history traits in shifting climates and environments, including dispersal distance, population growth rate and allee effect (Fordham et al., Reference Fordham, Brown, Canteri, Austin, Lomolino, Haythorne and Lorenzen2024; Pilowsky et al., Reference Pilowsky, Brown, Llamas, van Loenen, Kowalczyk, Hofman-Kamińska and Fordham2023).

Most recently, pattern-oriented modelling has emerged as a powerful tool for mechanistically reconstructing species’ demographic responses to multiple millennia of climate and environmental change, including periods when Earth’s climate warmed at rates similar to what is being forecast for the 21^st century (Fordham et al., Reference Fordham, Jackson, Brown, Huntley, Brook, Dahl-Jensen and Nogues-Bravo2020). This is being done using inferences of demographic and distributional change from fossils and ancient DNA (aDNA). These inferences are utilised as independent validation targets for identifying whether models have the structural complexity and parameterization needed to simulate species’ range shifts and population declines that happened hundreds to thousands of years ago (Pilowsky et al., Reference Pilowsky, Haythorne, Brown, Krapp, Armstrong, Brook and Fordham2022b). This multi-disciplinary approach—integrating the fields of macroecology, paleoecology, climatology, and genomics—is revealing species attributes that increase extinction risk during periods of rapid climatic change (Canteri et al., Reference Canteri, Brown, Schmidt, Heller, Nogués-Bravo and Fordham2022). Applying this general framework to a greater diversity of taxa and time frames, including more recent observations of species range movements and population changes, will likely provide a more thorough understanding of the demographic traits that regulate species’ responses to climate change, leading to better predictions of the species that are most vulnerable to future climate change (Fordham et al., Reference Fordham, Jackson, Brown, Huntley, Brook, Dahl-Jensen and Nogues-Bravo2020).

Looking ahead –

Climatologists have for decades been using process-explicit models and past observations to disentangle mechanisms of long-term climate forcing (Braconnot et al., Reference Braconnot, Harrison, Kageyama, Bartlein, Masson-Delmotte, Abe-Ouchi and Zhao2012; Sherwood et al., Reference Sherwood, Webb, Annan, Armour, Forster, Hargreaves and Zelinka2020). Accordingly, a wider usage of process-explicit models for quantifying and reporting climate-driven biodiversity dynamics is likely to uncover the functional and life history traits that make some species more prone to climate change, improving conservation management and policy.

Fortunately, the computer modelling platforms needed to unpack complex interactions between species traits and climate change now exist and are widely accessible (Bocedi et al., Reference Bocedi, Palmer, Pe’er, Heikkinen, Matsinos, Watts and Travis2014a; Fordham et al., Reference Fordham, Haythorne, Brown, Buettel and Brook2021; Kearney and Porter, Reference Kearney and Porter2020; Visintin et al., Reference Visintin, Briscoe, Woolley, Lentini, Tingley, Wintle and Golding2020). However, these process-driven modelling approaches continue to be primarily used to simulate the impact of forecast climate change on future species ranges and abundances (Briscoe et al., Reference Briscoe, Elith, Salguero-Gómez, Lahoz-Monfort, Camac, Giljohann and Guillera-Arroita2019). This is despite having capacities to directly identify species traits associated with demographic responses to observed climate change (Pilowsky et al., Reference Pilowsky, Colwell, Rahbek and Fordham2022a). Future research should now centre on using process-explicit models to analyse the growing volume of recent observations of spatiotemporal changes in species ranges and abundances in response to accelerating climate change (Lenoir et al., Reference Lenoir, Bertrand, Comte, Bourgeaud, Hattab, Murienne and Grenouillet2020; Rosenberg et al., Reference Rosenberg, Dokter, Blancher, Sauer, Smith, Smith and Marra2019). Particularly, if these observed demographic changes are used as targets in process-explicit and pattern-oriented models (Figure 1). These established methods and data promise to provide a deeper understanding of the traits that species use to respond to climate and environmental disturbances, improving indicators of the future vulnerability of biodiversity to worsening planetary change. Because process-explicit modelling frameworks can be timely to construct, particularly when applied to multiple species, the initial focus could be on developing these process-driven approaches for select species that enable testing of climate-sensitive traits previously identified with statistical analysis of repeat survey data.

Figure 1. Identifying biological traits using process-explicit models, pattern-oriented validation, and resurvey data. A) Spatially explicit population models are built using life history parameters drawn from wide but plausible ranges. B) These models simulate thousands of plausible patterns of spatiotemporal abundance. C) Pattern-oriented modelling methods are used to validate these reconstructions of spatiotemporal abundance using observations of demographic change (shifts in range and abundance) from resurvey data. D) This identifies parameter bounds for life history traits and establishes their relative importance for reconstructing observed shifts in species distributions and abundances in response to climate and environmental disturbances.

Longitudinal data on the abundances and ranges of a wide diversity of Earth’s plants and animals, living in terrestrial, marine and freshwater biomes, in open-access portals and databases has grown rapidly in recent decades (Edgar and Stuart-Smith, Reference Edgar and Stuart-Smith2014; Sabatini et al., Reference Sabatini, Lenoir, Hattab, Arnst, Chytrý, Dengler and Bruelheide2021). As has the diversity of species trait characteristics in online databases (Tobias et al., Reference Tobias, Sheard, Pigot, Devenish, Yang, Sayol and Schleuning2022; Wilman et al., Reference Wilman, Belmaker, Simpson, de la Rosa, Rivadeneira and Jetz2014). As human-induced climate change worsens in the coming decades, these resources will become invaluable, particularly if they are regularly updated and expanded, enabling pronounced climate-driven responses of biodiversity to be followed in-situ. Analyses of these repeat survey data with process-driven modelling approaches and pattern-oriented methods will overcome many of the problems that have so far limited the identification of functional and life history traits associated with extinction proneness to climate change using correlative analytical methods. These include problems with analysing data that is sparsely distributed in space and time, and issues with disentangling complex non-linear trait-based responses to climatic change and their interactions with other anthropogenic threats.

Temporally and spatially explicit population models have already been used to reconstruct the range and population dynamics underpinning observed snapshots of trajectories of abundances and distributions of birds in the UK at fine temporal and spatial scales, and across large geographical extents (Fordham et al., Reference Fordham, Bertelsmeier, Brook, Early, Neto, Brown and Araújo2018). Applying these and other process-explicit modelling methods to repeat survey data for species with diverse taxonomic coverage, is likely to reveal generalisable functional and life-history traits responsible for the fitness and persistence of species in shifting climates (Figure 1). These findings are needed immediately to strengthen the relevance of species traits in setting national and international conservation policy goals (Kissling et al., Reference Kissling, Walls, Bowser, Jones, Kattge, Agosti and Guralnick2018).

Given that it is now possible to use process-explicit models to establish ecological and demographic determinants of species range shifts and population declines that occurred hundreds to thousands of years ago, more emphasis should also be placed on identifying biological responses to rapid climatic events that occurred in the more distant past (Pilowsky et al., Reference Pilowsky, Colwell, Rahbek and Fordham2022a). Times and places where paces and magnitudes of past and future forecasts closely match, and where sufficient paleontological data on biotic responses to these warming events exists, will provide important opportunities to better anticipate and manage the responses of species to changing climates (Fordham et al., Reference Fordham, Jackson, Brown, Huntley, Brook, Dahl-Jensen and Nogues-Bravo2020).