Pathways to global-change effects on royalsocietypublishing.org/journal/rspb biodiversity: new opportunities for dynamically forecasting demography and species interactions Review Maria Paniw1,3, David García-Callejas2,4, Francisco Lloret5,6, Ronald D. Bassar7, Cite this article: Paniw M, García-Callejas D, Lloret F, Bassar RD, Travis J, Godoy O. 2023 Joseph Travis8 and Oscar Godoy4 Pathways to global-change effects on 1 Department of Conservation Biology and Global Change, Estación Biológica de Doñana (EBD-CSIC), Seville, biodiversity: new opportunities for dynamically 41001 Spain 2 Department of Integrative Ecology, Estación Biológica de Doñana (EBD-CSIC), Seville, 41001 Spain forecasting demography and species 3 Department of Evolutionary Biology and Environmental Studies, University of Zurich, Zurich 8057, Switzerland interactions. Proc. R. Soc. B 290: 20221494. 4 Instituto Universitario de Investigación Marina (INMAR), Departamento de Biología, Universidad de Cádiz, https://doi.org/10.1098/rspb.2022.1494 Campus Río San Pedro, 11510 Puerto Real, Spain 5 Center for Ecological Research and Forestry Applications (CREAF), Cerdanyola del Vallès 08193, Spain 6 Department Animal Biology, Plant Biology and Ecology, Universitat Autònoma Barcelona, Cerdanyola del Vallès 08193, Spain 7 Department of Biological Sciences, Auburn University, Auburn, AL 36849, USA Received: 7 August 2022 8 Department of Biological Science, Florida State University, Tallahassee, FL 32306, USA Accepted: 12 January 2023 MP, 0000-0002-1949-4448 In structured populations, persistence under environmental change may be particularly threatened when abiotic factors simultaneously negatively affect Subject Category: survival and reproduction of several life cycle stages, as opposed to a single Global Change and Conservation stage. Such effects can then be exacerbated when species interactions gener- ate reciprocal feedbacks between the demographic rates of the different Subject Areas: species. Despite the importance of such demographic feedbacks, forecasts that account for them are limited as individual-based data on interacting ecology species are perceived to be essential for such mechanistic forecasting—but are rarely available. Here, we first review the current shortcomings in asses- Keywords: sing demographic feedbacks in population and community dynamics. We size-dependent biotic interactions, forecast then present an overview of advances in statistical tools that provide an bias, Bayesian hierarchical models, latent state, opportunity to leverage population-level data on abundances of multiple integrated ecological forecasts, species to infer stage-specific demography. Lastly, we showcase a state-of- metapopulations the-art Bayesian method to infer and project stage-specific survival and reproduction for several interacting species in a Mediterranean shrub com- munity. This case study shows that climate change threatens populations most strongly by changing the interaction effects of conspecific and hetero- Author for correspondence: specific neighbours on both juvenile and adult survival. Thus, the Maria Paniw repurposing of multi-species abundance data for mechanistic forecasting e-mail: maria.paniw@ebd.csic.es can substantially improve our understanding of emerging threats on biodiversity. 1. Demographic determinants of species responses to environmental change We are living in an era of unprecedented human-driven perturbations affecting all levels of biological organization, from local populations to communities to entire ecosystems. Such perturbations are complex, often consisting of synergis- tic, nonlinear effects of multiple abiotic and biotic factors [1,2]. When this complexity of human impacts meets complex natural systems, where different Electronic supplementary material is available interacting species are differently affected by environmental drivers, it becomes online at https://doi.org/10.6084/m9.figshare. c.6403999. © 2023 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution License http://creativecommons.org/licenses/by/4.0/, which permits unrestricted use, provided the original author and source are credited. imperative to understand key pathways through which modelling perspective, this means that by parameterizing 2 environmental change can alter natural communities [3–5]. demographic models as functions of abiotic drivers and royalsocietypublishing.org/journal/rspb Understanding these pathways allows us to define more species interactions, we can assess how much abiotic drivers nuanced ecological forecasting, proposing different scenarios affect survival and reproduction directly as opposed to under which populations remain viable in the future, when indirectly, when abiotic drivers alter species interactions, they go locally extinct, or when they invade new habitats e.g. by changing abundances ([36]; figure 1). Only with [6–8]. such information can we understand how lower level demo- In communities that consist of age-, stage- or trait-struc- graphic effects scale up to affect population dynamics of tured species’ populations, a key pathway that needs to be several species simultaneously and consequently our fore- accounted for in robust forecasts are the nuanced effects of casts of changes in community composition ([33,37,38]; see global-change drivers across different life cycle stages ([9]; example in box 1 for interactions between size-dependent figure 1). For instance, climate-driven changes in the timing demographic responses and intraguild predation). of key life cycle events (i.e. phenology) can lead to substantial mismatches in the phenology between species, thus affecting Proc. R. Soc. B 290: 20221494 their survival and reproduction [9–12]. However, a lower reproductive output or higher offspring mortality can be 2. Shortcomings in forecasting species compensated by a higher survival of the remaining juveniles persistence and therefore have ultimately little effects on population Although the intrinsic relationship between demographic fitness [13]. These and other cases highlight that global- and community levels of organization has traditionally change drivers do not need to result in changed population been well recognized [52–56], its integration to forecasting dynamics if demographic tradeoff or compensation mechan- has remained elusive [57]. Prominent examples of multi- isms that buffer unfavourable environmental conditions are species demographic models exist in coexistence research in place [14–17] or environmental effects are concentrated [56,58], eco-evolutionary dynamics [59], trophic interactions on demographic rates with low contribution to population [60,61] and forest stand dynamics [35]. However, forecasting growth rates [18]. On the other hand, simultaneous negative applications that empirically assess the feedback between environmental effects on several life cycle stages may species demography and species interactions in a community exacerbate extinction risks [19]. context are largely missing (but see [35] for approaches to The example of mismatch in the phenology of interacting indirectly link demography and community composition species demonstrates that the complexity in species’ demo- via resources). Forecasts that scale from demographic rates graphic responses to environmental change is compounded to populations focus strongly on a single species and simplify by the complexity of the underlying environmental drivers, or omit interspecific interactions [62]. This is not surprising as where the effects of abiotic drivers on demography can be such forecasts rely almost exclusively on long-term individ- strongly mediated by biotic interactions (figure 1). That is, ual-based data. This is undoubtedly the most robust species do not exist in isolation in natural communities; both approach to quantify demographic processes across the life theoretical (e.g. [20,21]) and empirical studies (e.g. [22–24]) cycle and link them to emergent population properties [63– on coexistence and trophic interactions show that even rela- 65]. However, such datasets are rarely collected for several tively small changes in abiotic conditions can alter species interacting species. This limitation contrasts with ample evi- interactions and thereby community organization. For dence showing the role of such interactions in mediating instance, changes in competitive interactions can result in population fates under global-change drivers [31,66–68]. niche shifts for some species, thereby allowing for local popu- Therefore, a large knowledge gap remains in our understand- lation persistence even under adverse abiotic conditions ing of the pathways through which global-change drivers [25,26]. Predation pressures have also been shown to constrain affect the local persistence of multiple interacting species adaptive responses to climate change in local populations [27] within ecological communities. or to exacerbate adverse climate-change effects [28], while her- bivory can be important in maintaining plant community diversity under climate change [29], and the adverse effects of pathogens can be amplified by more extreme climatic 3. Repurposing abundance data to forecasts events [19]. The effects of species interactions are often specific feedbacks between demography and species to particular stages in the life cycles of the interacting species [4,30]. It has been shown that environmental perturbation on interactions key life cycle stages may amplify the effects on negative inter- An important step towards expanding the application of fore- actions among species, destabilizing entire ecological casts that integrate feedbacks between demography and communities [31,32]. Such feedbacks between demography species interactions is by repurposing existing long-term and species interactions mean that environmental effects on abundance datasets. While individual-based data on multiple any populations often scale nonlinearly (or non-additively) species in a community are rare, population-level data on to spatiot-emporal abundance changes of neighbouring stage-specific multi-species abundances are routinely col- species (figure 1) [33–35]. lected in studies focused on spatio-temporal changes in Collectively, the examples above suggest that our under- community composition [52,69–73]. Such studies assess rela- standing of changes in biodiversity would be improved if tive abundances of species [74–77] or changes in reproduction we assess how much feedbacks between demographic rates and survival at the species level [78–80]. However, stage- and species interactions can modify, under increasing abiotic specific demographic rates are mostly omitted, and studies pressures such as climate change, the population persistence typically do not explicitly forecast dynamic changes in of several species in natural communities (figure 1). From a species interactions. At the same time, forecasting feedbacks (a) 3 1 2 3 interspecific royalsocietypublishing.org/journal/rspb juveniles adults juveniles adults juveniles adults abundance survival survival survival low high abiotic driver abiotic driver abiotic driver abiotic driver abiotic driver abiotic driver (b) abundance at time t abundance at time t + 5 abiotic driver indirect effect of abiotic driver through changes in abundances at time t – 1 survivalstage,t a f (abiotic, biotic) Proc. R. Soc. B 290: 20221494 t direct effect of abiotic driver at time t Figure 1. Conceptual overview of how feedbacks between stage-specific demographic rates and species interactions can affect community dynamics under environ- mental change. As an example (a), stage-specific survival rates of three species respond differently to an abiotic driver and biotic interactions (approximated by interspecific abundances): 1—adult survival changes more strongly under abiotic ( positive) than biotic (negative) effects; 2—juvenile and adult survival decline more strongly under biotic effects; 3—strong effect of abiotic driver on adult survival, but juvenile survival benefits from high-interspecific abundances. A given value of the abiotic driver at time t can therefore (b) directly change demographic rates (e.g. survival, recruitment and stage transitions) from time t to t + 1. This then changes the abundances of species, which, in turn, affect demographic rates and abundances at time t + 1, representing an indirect effect of the abiotic driver. Such feedbacks mean that, while all species respond negatively to a decrease in the abiotic driver, unfavourable abiotic conditions may increase the abundance of those species that are more strongly affected by changes in interspecific abundances (e.g. release from competition, blue species here). between demography and species interactions using multi- challenge is the same as those studies inferring species inter- species abundance data may have advantages over using actions for highly diverse systems [98], namely, how to individual-based data, especially for applied management. circumvent the estimation of a large number of parameters This is because model outputs can be evaluated against [65], some of which may not be identifiable [99]. For instance, population-level data across broad spatial scales [81], which if we consider three interacting species (e.g. two herbivores is rarely done with population-level outputs from demo- which share a top predator) and three life stages ( juvenile, graphic models based on individual-based data, creating a non-reproductive adult and reproductive adult), we need to mismatch in scale [82]. estimate, at a minimum, 3 × 3 = 9 interaction effects, typically Recent methodological advances, often termed dynamic modelled as densities of con- and interspecific neighbours N-mixture models, have allowed us to empirically infer the [56,99], for each demographic stage. This equals to 27 pairwise demographic processes underlying species co-occurrences interactions determining the survival, development and or multi-species abundances (reviewed in [83], see also reproduction of the three species. We would also need to [84]). Generally, such methods use Bayesian latent-state add additional parameters that map the effect of environ- approaches to estimate gains (recruitment and immigration) mental change onto demographic rates and species and apparent survival (true survival minus emigration) as interaction coefficients, which can be common across species unobserved processes [85–87] from time series of counts, and life stages or specific to each combination. While such where such demographic rates can be structured by size, an effort is feasible in simulations [101], the accurate empirical age or stage [84,88], and modelled as a function of density estimation of so many parameters in a natural system can dependence ([89]; figure 2). These approaches offer great require an impossible level of effort [5]. Integrating these par- potential to link demography and species interactions for ameters also increases model complexity and can complicate multiple species [90–92]. Below, we first discuss some simpli- interpretation of modelling outcomes [102], in part because fying assumptions that need to be made to parameterize the likelihood of highly intercorrelated variables increases multi-species dynamics forecasts. We then illustrate with an with an increasing number of parameters, thus raising the example how to effectively estimate the demographic likelihood of spurious correlations [103] and issues with mechanisms of changes in abundances of several species in parameter identifiability [83]. Experimental manipulations of a community; we do not consider recent related advances densities are ultimately required to tease apart whether posi- in species range dynamics, which have been discussed in tive or negative associations between species indicate much detail elsewhere [93–96]. interactions [104,105], but these can also be difficult to execute in systems larger than two–three species. There are ways to reduce the dimensionality of parameter- 4. Modelling species interactions in structured ization for systems of multi-species interactions using discretization and grouping following general ecological demographic models rules and regularization approaches [106,107]. For one, species Inferring age- or stage-specific demographic rates from count do not interact randomly but follow, in most cases, specific data using latent-variable approaches has become increasingly rules determined by species traits such as body size or height more accessible [83,97]. However, parameterizing such models [3,37,108]. In addition, although interaction coefficients are for interacting species comes with some challenges. The main usually expressed as per capita effects (e.g. [56]), the relative 4 Box 1. Individual demographic interactions determine community dynamics. royalsocietypublishing.org/journal/rspb One well-studied example that highlights the complexities of species interactions in determining community structure occurs in the high-elevation streams on the Caribbean island of Trinidad. There, Hart’s killifish (Rivulus hartii) and Trinidadian gup- pies (Poecilia reticulata) compete with each other for resources and consume each other as part of a bidirectional intraguild predation community. Theory predicts that intraguild predation should result in an unstable community, but guppies and killifish coexist in these streams. This is, in part, because the magnitude of interactions between the two species depends strongly on their body sizes (Box Figure). Simple models of species interactions (e.g. Lotka–Volterra competition equations) describe the negative effects of compe- tition on population growth as simple functions of the numbers, densities, or biomass of competitors in the environment. However, when populations are structured by traits such as body size, simple models like these generate predictions that perform poorly when compared to models that incorporate size-specific interactions between the species [39,40]. For example, for both guppies and killifish, larger individuals are better competitors for a common resource than are smaller ones [41,42]. For guppies, incorporating these differences in projections of their long-term population dynamics produced Proc. R. Soc. B 290: 20221494 more accurate results than models assuming all individuals compete equally [30]. 60 0.90 recruitment (no. of embryos) 0.85 40 survival 0.80 0.75 20 0.70 0 20 30 40 50 initial body length (mm) Box Figure. Survival and recruitment in killifish vary depending on whether guppies are present (black) or absent (orange) in streams and whether guppies affect all or only certain sizes of killifish. Plot derived from [43]. Yet size often determines more than the competitive ability of the individuals. Larger killifish and guppies also consume resources at higher trophic levels than smaller individuals, which could decrease the negative effects of competition on the smaller fish. As intraguild predators, larger individuals of each species consume smaller individuals of the other species. Because the interactions between these species are so strongly dependent on body size—whether it be competitive inter- actions or predation—any factor that affects the size distributions of these species can generate markedly different population dynamics and different outcomes of their interaction, coexistence or deterministic competitive exclusion or exclusion of either species via priority effects [40]. Still, the size-based interactions between the two species tell an incomplete story. High-elevation Trinidadian streams con- tain different microhabitats that each of the species prefers [44]. Recent analyses show that each species may also respond to environmental variation in contrasting ways, with guppies doing well in drier periods and killifish performing better in wetter periods (RDB and JT unpublished data). Consider what can happen if climate change was to produce longer dry seasons and drier wet seasons. The pools favoured by guppies may be more productive because of longer periods of high light inten- sity while the riffles favoured by killifish may shrink in total area from the reduced water flow. As a result, guppies may attain larger body sizes, killifish may grow more slowly and display smaller body sizes, and the size-dependent interactions could result in the elimination of killifish. Both species also rapidly adapt to the presence of the other species [45,46] and display genetically based morphological, behavioural and life-history differences. These adaptations feedback to alter the rest of the community and ecosystem [47–51]. The outcome of these evolutionary effects is a change in way the species interact over time and for the resulting population dynamics to evolve [43]. These evolutionary changes are repeatable across multiple independent evolutionary instances of guppy/killifish communities, revealing that these evolutionary changes are predictable and hence can, in principle, be incorporated into predictive models. However, this is no small task. abundances of neighbouring species drive the frequency of neighbours such as fast-growing versus slow-growing species interactions between individuals of different species, and or native versus invasive species can facilitate the interpret- with it, the overall biotic effects on the focal species [3]. In the ation of model coefficients as true interactions [109]. Causal absence of individual monitoring to quantify per capita effects, inference methods also exist to better infer cause-effect relation- estimates based on trait-matching and abundance-based inter- ships from observational data [110,111]. Lastly, relatively action frequency can be used as a first-principles baseline for simple structured population models are able to capture a grouping interactions in empirical systems and across life wide range of demographic variation within and among stages (figure 2). Furthermore, life-history information of the populations [112–114]. This means that the step from collected data species under increasingly drier winter weather (an overview 5 (site i; time t) of the analyses steps in the case study is presented in royalsocietypublishing.org/journal/rspb species h species s figure 3). We use data on the abundance of individuals of juveniles juveniles common shrub species in Doñana National Park (Spain) adults adults that have been recorded across 18 sites since 2007. Details on the study and sampling design are described in electronic supplementary material, S1. The original aim of the monitor- (1) alternative formulation observed ing was to assess changes in community composition after a with detection probability counts (n) severe drought which led to a collapse of the shrubland in n a binomial(p, N [h,i,t] [h,i,t] 2005 [115]. Recent evidence suggests that resilience to n [h,i,t] a poisson(N [h,i,t] n [s,i,t] a poisson (N [s,i,t] drought is well described by community-weighted species- n [h,i,t] a poisson(N [h,i,t] n [s,i,t] a poisson (N [s,i,t] specific demographic traits (e.g. longevity, reproductive output) [116], but dynamic forecasts of multi-species popu- (2) estimated mean lations abundances have not been done thus far. Such Proc. R. Soc. B 290: 20221494 abudances (N) forecasts are important because rainfall has become increas- N [h,i,t+1] φ[s,i,t] N [h,i,t]  θ[h,i,t] N [h,i,t] ingly scarcer at the study site since the drought, a situation φ[s,i,t] N that is expected to continue under climate change [117,118]. N [s,i,t+1] [s,i,t]  θ[s,i,t] N [s,i,t] However, how rainfall scarcity affects plant demography N [h,i,t+1] G[h,i,t] N [h,i,t] N [s,i,t+1] G[s,i,t] N [s,i,t] directly or indirectly via changing species interactions (3) remains an open question, thus complicating assessments of demographic the fate of the critical shrublands. rates (e.g., φ) We used 8 years of data, collected across 18 plots between 2007 and 2021, on stage-specific abundances of the five most logit(φ[h,i,t]) = β0[h] + β1[h]rain[t] + β2[h] N [h,i,t] + β3[h] ¦ N [s,i,t] szh common shrub species in the study area (Halimium commuta- logit(φ[s,i,t]) = β0[s] + β1[s]rain[t] + β2[s] N [s,i,t] + β3[s] ¦ N [h,i,t] tum, Halimium halimifolium, Rosmarinus officinalis, Lavandula hzs stoechas and Cistus libanotis) to estimate stage-specific survival intraspecific interspecific (of saplings and reproductive adults), transitions of saplings species interactions to adults and gains of saplings to the population as latent, Figure 2. Overview of dynamic N-mixture models to infer demographic rates i.e. unobserved, states, inferred from the abundances from stage-specific count data. Using count data (n), stage-specific abun- (figure 3). We modelled these demographic rates jointly for dances (N ) for different species (here species h and s) can be modelled as all species using generalized linear models that included rain- a function of underlying survival (θ—juvenile survival and w—adult survi- fall and inter- and intraspecific densities (adult abundances in val) and recruitment or gains (G). Under imperfect detection, detection 5 × 5 m plots) as covariates. Recruitment of seedlings (ratio probabilities (р) can also be accounted for (shown for species h). Species’ seedlingst+1/adultst) was assumed to be fixed, as we only population dynamics can then be dynamically linked by including the effects had two years of data on recruitment (see electronic sup- of intra- and interspecific abundances in demographic rate models at time t, plementary material, S1). For each species, abundances at which, in turn, affect abundances at t + 1. Here, interspecific effects are rep- the beginning of time t were estimated from the demographic resented as the sum of adult abundances for all neighbour species (simplified transitions from (t − 1) to t. Interactions among the shrubs are to one neighbour for clarity) of a focal species. Species can then be jointly largely determined by size, i.e. the number of relatively large, projected under a common abiotic driver of demographic rates (e.g. rain). adult neighbouring shrubs [119,120]. In the absence of infor- mation on the spatial location of the shrubs, we considered all abundance-based data to demographic projections can potentially adult shrubs in a 5 × 5 m plot as neighbours. We summed the be achieved via relatively straightforward structured models, abundance of adult interspecific neighbours at the beginning of including as few as two stages ( juveniles and reproductive). time t and used this pooled measure of interspecific effects as We use the example of interactions among Mediterranean covariate in the demographic rate models at t for each species shrubs to highlight the minimum data required to do short-term (we included the abundance of intraspecific adults as a separate forecasts (i.e. out-of-sample predictions) and climate-change pro- covariate). Electronic supplementary material, S1 showcases the jections (i.e. using scenarios of rainfall change) at the interface of Bayesian dynamic approach used to parameterize models, test population and community ecology. Using a dynamic N-mixture their goodness of fit and test their ability to recover parameters model [85] (figure 2), we showcase how ecological forecasts can from simulated data. This methodology can be readily applied be improved through more biologically realistic parameteriza- to other community datasets [84] and to other drivers of global tions of the effects of environmental conditions on population change, but our study also highlights that sufficient spatio-tem- abundances, i.e. simultaneously projecting how biotic and abiotic poral replication of stage-specific counts is needed to avoid factors affect the demography of interacting species. overparameterization and parameter unidentifiability. In our particular case, replication in space compensated for the rela- tively low temporal replication of 8 years, but spatial 5. Example: inferring demography from replication of similar studies is typically much larger [83,84]. abundance datasets to project climate-change The low spatio-temporal resolution of the data can increase uncertainties of parameter estimates (electronic supplementary effects on interacting shrubs material, figures S18 and S19) and, in our case, did not give us Our case study focuses on understanding how feedbacks enough degrees of freedom to parameterize more complex and between demography and species interactions mediate realistic covariate effects (e.g. interactions between rainfall and changes in abundances of coexisting Mediterranean shrub densities or quadratic rainfall effects). (a) stage-structured population model (b) dynamic demographic rates 6 ρ 5 species (H), 18 sites no. adults no adults rain royalsocietypublishing.org/journal/rspb H1 H2 γ θ H3 NR τθ ϕ R S A H4 H5 no. adults no. adults rain θ (1 – τ)θ ϕ ϕ dispersal H3 – H5 (c) pathways of reduced-rainfall effects on communities scenario forecast dynamic projection (below-average rain) no dynamic (example S1) no dynamic fully dynamic abundance abundance Proc. R. Soc. B 290: 20221494 2022 only focal species all species affected S1 S2 S3 S4 S5 1 50 affected year observed Figure 3. Overview of analyses for the Mediterranean shrubs showcasing applications of dynamic multi-species N-mixture models. (a) The local demography of five shrub species (H1–H5) across 18 plots is described by a three-stage life cycle (from which a matrix population model is built), in which transitions among seedlings (R), saplings (S) and reproductive adults (A) are defined by demographic rates: g—gains of S, t—transition S to A; u, w—survival of S and A, respectively, and r—per capita number of R produced by A. The demographic rate in purple (u, w,t, g) are inferred as latent states using a Bayesian hierarchical (N-mixture) model. The local demography is connected across the 18 sites by dispersal of R (via seeds produced by A). (b) For all species (here shown for two species only), u, w are modelled as functions of rainfall and intra- and interspecific adult (A) abundances. (c) By assuming changes in rainfall in the demographic rate models, either for all species simultaneously (fully dynamic) or for focal species only (no dynamic), the demographic pathways of climate change effects on the entire community can be assessed in near-term forecasts (out-of-sample predictions) and longer-term projections based on climate-change scenarios. Forecast skill can be assessed by comparing correlations between observed and predicted data. Note that plots are fictitious, showing a schematic overview of potential results. We used predictions from the demographic rate models both intra- and interspecific densities) on adult survival— (electronic supplementary material, figures S2–S6) to build suggesting competition. Notably, however, for some species, stage-structured metapopulation models for each species such as H. commutatum and L. stoechas, such negative effects [120]. In these models, the local demography of every shrub are balanced by positive effects of interspecific densities on species in each of the 18 plots was described by a stage-struc- juvenile survival (figure 4). As expected, rainfall positively tured matrix population model (figure 3). The 18 local matrix affects populations, although the effect is not as strong as models were joined by a dispersal matrix, which described the effects of interspecific densities (figure 4). We performed adult plants producing seeds in a given plot that germinated additional sensitivity simulations in which we estimated the as seedlings in adjacent plots (see electronic supplementary indirect effects of rainfall on population growth via changes material, S1 for details). We validated how well the metapo- in interspecific densities (see Perturbation analyses in electronic pulation model predicted observed abundances by projecting supplementary material, S1). These simulations show that the abundances of each species until 2021, starting with the site- strength of such indirect effects differs among species and can and stage-specific population vector in 2010, and visually be substantial, most importantly for R. officinalis where a 10% comparing the results to observed abundances. This in- increase in rainfall affecting neighbours’ survival decreases sample validation showed a good fit to adult abundance population growth by 8% on average (electronic supplemen- data (observed abundances were within the 95% C.I. of pre- tary material, figure S15). This system of interacting shrubs dicted abundances for most species, sites and years; therefore suggests that changes in abiotic conditions can electronic supplementary material, figures S8–S13). Predicted strongly affect the community via interactions among neigh- abundances of saplings diverged more strongly from bouring species and contrasts with other systems of non- observed ones, likely because we estimated gains of saplings trophic interactions, where interspecific interactions affect to the population as a constant and did not account for seed demography far less than intraspecific interactions [36]. production when inferring demography from abundances. Intra- and interspecific interactions mediate population We simplified these processes as we lacked data for more responses of the shrubs when simulating a drier future. We complex models (electronic supplementary material, S1), jointly projected the metapopulations of the interacting highlighting that the lack of abundance data on several life shrubs using average posterior values of parameters (see cycle stages can be an important limiting factor on the [19,99] for examples of full uncertainty propagation in projec- detail of demographic inference from abundances. tions) and assuming a higher prevalence of below-average We then focused on local population dynamics of each rainfall in the next 50 years (figure 3; electronic supplemen- shrub species and determined the relative sensitivity of the tary material, S1). Under these projections, not all species local population growth (λ) of each shrub species at equili- decrease in abundance compared to baseline projections (ran- brium population density to changes in rainfall, intra- and domly sampling all rainfall values 2007–2021) (figure 5). interspecific densities. These sensitivity analyses demonstrate Abundances of C. libanotis decrease most strongly, followed that population dynamics of most species are strongly influ- by H. halimifolium, as adult survival in both species is directly enced by negative effects of species interactions (including positively affected by rainfall. Plants may also experience C. libanotis H. commutatum H. halimifolium 7 0.06 0.01 0 royalsocietypublishing.org/journal/rspb 0.04 0 –0.01 0.02 –0.01 0 –0.02 –0.02 –0.02 change λ (%) –0.03 sapling adult L. stoechas R. officinalis survival (θ) survival (ϕ) 0.025 0 0 –0.01 perturbed –0.02 rain –0.025 –0.03 intraspecific density –0.050 Proc. R. Soc. B 290: 20221494 –0.04 –0.05 interspecific density sapling adult sapling adult survival (θ) survival (ϕ) survival (θ) survival (ϕ) demographic rate Figure 4. Percentage of change in population growth (λ), measured as the dominant eigenvalue of stage-structured matrix population models describing the local demography of five shrub populations, in response to increasing rainfall or intra- and interspecific densities by 10% when predicting sapling survival (θ) and adult survival (w). dynamic no yes rainfall, a higher adult survival is countered by a lower juven- ile survival under lower interspecific densities (figure 4). By contrast, abundances of R. officinalis increase under climate change (figure 5), as lower abundances of neighbours under climate change decrease the negative effect of inter- change in mean abundance 6 specific density on adult and juvenile survival (figure 4). Projections that do not account for climate-change effects on interspecific densities (no dynamics in figure 5) tend to 3 show higher increases or decreases in abundances for most species compared with fully dynamic projections; they also increase the variation in abundance forecasts (electronic sup- plementary material, figure S17). Short-term forecasts of 0 most-recently collected abundances (2022) lend further sup- port to the importance of fully dynamic projections of abundance changes. Starting with stage-specific abundances in 2007, we forecast abundances to 2022, either assuming a fully dynamic model or a simplified one, i.e. assuming no is um m s is ha ot al liu at ec an in climate effects on interspecific densities (see electronic sup- ifo ut fic sto lib m im of om plementary material, S1 for details). We show that fully L. C. al R. .h .c dynamic forecasts provide an accurate estimate of observed H H Figure 5. Projected changes in abundance estimates of interacting Mediter- adult abundances in 2022 (but less so for saplings due to ranean shrub species. Boxplots show distributions of changes in mean adult the data limitations in the relevant models, as discussed abundances (number of individuals in 5 × 5 m plots, averaged over 50 time above; figure 6; electronic supplementary material, figure steps and 100 simulation runs) across 18 sites (for more detailed plots includ- S14). Simplified assumptions in the forecasts, meanwhile, ing different stages and changes in variance see electronic supplementary decrease forecast accuracy (figure 6). More years of indepen- material, figures S14 and S15). Changes are based on subtracting mean abun- dent data to evaluate the forecasts are needed for more dances under projections assuming climate change (sampling below-average conclusive results; this is particularly true for mechanistic rainfall with a probability of 0.8) from projections assuming past climate (ran- forecasts that include many parameter and lagged effects of domly sampling past observed rainfall values). We either projected feedbacks density (figure 1; [8,9]). Nevertheless, several lines of evi- between demography and species interactions (red) or sampled interspecific dence in our case study suggest that omitting a crucial densities from past observed values without dynamically projecting them pathway of climate change on populations in the forecast, (black; electronic supplementary material, S1 for details). i.e. interspecific interactions affecting stage-specific survival, can lead to an inaccurate interpretation of population dynamics under climate change. more competition from shrubs not directly negatively Winters in the Mediterranean have been getting drier over affected by scarcer rainfall, as is the case for the past 10 years; projections of our dynamic multi-species H. commutatum, L. stoechas and R. officinalis. Abundances of demographic models are in line with empirical evidence of H. commutatum and L. stoechas do not change under drier con- abundance trends (electronic supplementary material, S1) ditions, as, in addition to not being directly affected by and literature that shows a higher recent mortality of (a) (b) 8 MSE = 2781; r = 0.70 [0.58, 0.79] MSE = 1920; r = 0.29 [0.09, 0.47] 50 royalsocietypublishing.org/journal/rspb MSE = 2835; r = 0.65 [0.54, 0.79] MSE = 1923; r = 0.28 [0.08, 0.47] 40 10 30 predicted predicted 20 5 10 Proc. R. Soc. B 290: 20221494 0 0 0 20 40 0 5 10 15 20 observed observed Figure 6. Comparison of observed and predicted abundances of adults (a) and saplings (b) in 2022. Points show abundances of all five modelled shrub species for each of 18 study plots (n = 90). Predicted abundances show averages ± 95% credible interval (error bars) from 2100 posterior samples of the Bayesian multi-species demographic model. The points and error bars on the x-axis are displaced slightly to facilitate visualization. We either predicted how rainfall changes affect all interacting species simultaneously (red; fully dynamic) or perturbed the effect of rainfall change on interspecific densities (black; no dynamic). Black lines show the 11 line indicating perfect fit between predicted and observed abundances. For each of the two forecast scenarios, mean squared error (MSE) and Pearson’s correlation coefficients (r, mean [95% confidence interval]) were calculated on the average posterior abundance values to compare forecast skill. H. halimifolium and increases in abundances in L. stoechas [115]. case for the many communities where even small environ- Based on life-history theory, we would expect populations of mental changes can substantially change community relatively short-lived species such as L. stoechas [116] to be dynamics by altering the outcomes of species interactions more sensitive to changes in abiotic conditions than popu- [128]. Similarly, averaging these direct and indirect effects of lations of longer-lived species such as H. halimifolium [122]. environmental change over demographic stages and even However, indirect effects of such changes via species inter- populations can substantially obscure our ability to assess actions can have substantial effects on natural communities the capacity of communities to bounce back from pertur- across different life histories [28,60]. The models we developed bations [83,129]. This occurs when different demographic for the shrubs are relatively simple and likely do not incorpor- rates respond distinctly to intra- and interspecific densities, ate all relevant biological processes (all scripts and data can be so that decreases in one life cycle stage may be compensated found at: https://github.com/MariaPaniw/shrub_forecast; by increases in another. Such context-dependent demographic [123]). We will update our models iteratively [124] as we responses are likely very common in nature [62,130] and chal- gain a better understanding of size-mediated demographic lenge classic assumptions of what taxa are most sensitive to rates, spatially explicit interactions and hidden demographic global change [131]. Robustly capturing some of the pathways stages, such as seed banks [125]. However, these dynamic through which global change affects populations may allow us models highlight the importance of a better integration of to design robust alternative scenarios of outcomes under the pathways through which environmental change can global change, thus avoiding the ‘forecast trap’ (sensu affect communities. Perturbing these pathways creates differ- [132,133]) where management decisions rely too closely on ent forecasting scenarios and can result in more nuanced optimizing the forecast ability of a single best-fit model. decision making on the management of this community Our case study demonstrates that advances in statistical where the focus can vary between managing populations tools [83] and data integration [134], including life-history (e.g. of the cover of the most common species, H. halimifolium) information, known effects of heterospecific and conspecific and communities (e.g. introducing burning to entire patches neighbours, or dispersal, make it possible to explore demo- to promote seedling recruitment and thus higher food graphy–biotic interaction feedbacks in a wide range of availability for rodents) [126]. systems. However, more studies aimed at improving the parameterization and biological realism of multi-species demographic inference are needed. The methods presented 6. Conclusion and future directions here require that sufficient count data be available to infer Assessing individual, age- or stage-specific demographic demographic rates. This may preclude the inclusion of less responses to biotic and abiotic drivers for several species in a abundant and potentially more threatened species. Such chal- community simultaneously offers an essential perspective lenges can be partially overcome by the inclusion of into the fates of populations and communities under environ- informative priors into Bayesian model fitting [89], and mental change. Projections of population dynamics based on studies that integrate informative priors into multi-species abundance data have been demonstrated to be only as accurate demographic inference are needed. In addition, populations as projections based on demographic models in some systems of many species are structured by continuous traits, such as [127]. However, as our case study shows, this may not be the size or body mass, and many community-level studies record not only counts but also such key traits. Future studies writing—review and editing; R.D.B.: conceptualization, methodology, 9 that parameterize demographic rates as functions of continu- visualization, writing—review and editing; J.T.: conceptualization, resources and writing—review and editing; O.G.: conceptualization, royalsocietypublishing.org/journal/rspb ous traits, in turn affected by species interactions, may not writing—original draft and writing—review and editing. only address an important biological pathway of global- All authors gave final approval for publication and agreed to be change effects on populations and communities [84,135], held accountable for the work performed therein. but can also result in more efficient model fitting that requires Conflict of interest declaration. We declare we have no competing interests. fewer parameters [136]. Funding. This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Data accessibility. All data to replicate the analyses in this manuscript Sklodowska-Curie grant agreement No 894223. O.G. acknowledges have been deposited on Dryad: https://doi.org/10.5061/dryad. financial support provided by the Spanish Ministry of Economy and 8cz8w9gvc [137]. R scripts to process the data and run all analyses Competitiveness (MINECO) and by the European Social Fund presented in the manuscript are available at: https://github.com/ through the Ramón y Cajal Program (RYC-2017-23666). F.L. acknowl- MariaPaniw/shrub_forecast [123]. edges financial support provided by the Spanish Ministry of Science The data are provided in the electronic supplementary material and Innovation (MICIN, PID2020-115264RB-I00). R.D.B. and J.T. [138]. acknowledge financial support from the National Science Foundation, USA (2100163). Proc. R. Soc. B 290: 20221494 Authors’ contributions. M.P.: conceptualization, formal analysis, funding acquisition, investigation, methodology, writing—original draft and Acknowledgements. We are grateful for constructive feedback from two writing—review and editing; D.G.C.: conceptualization, resources anonymous reviewers as well as Prof. Tim Coulson on previous and writing—review and editing; F.L.: data curation, validation and versions of this manuscript. References 1. Zarnetske PL, Skelly DK, Urban MC. 2012 Biotic 11. Both C, van Asch M, Bijlsma RG, van den Burg AB, 21. Mathias A, Chesson P. 2013 Coexistence and multipliers of climate change. Science 336, Visser ME. 2009 Climate change and unequal evolutionary dynamics mediated by seasonal 1516–1518. (doi:10.1126/science.1222732) phenological changes across four trophic levels: environmental variation in annual plant 2. Zscheischler J et al. 2018 Future climate risk from constraints or adaptations? J. Anim. Ecol. 78, 73–83. communities. Theor. Popul. Biol. 84, 56–71. (doi:10. compound events. Nat. Clim. Chang. 8, 469–477. (doi:10.1111/j.1365-2656.2008.01458.x) 1016/j.tpb.2012.11.009) (doi:10.1038/s41558-018-0156-3) 12. Simmonds EG, Cole EF, Sheldon BC, Coulson T. 2020 22. Aung KMM, Chen HH, Segar ST, Miao BG, Peng YQ, 3. Dáttilo W, Marquitti FMD, Guimarães Jr PR, Izzo TJ. Phenological asynchrony: a ticking time-bomb for Liu C. 2022 Changes in temperature alter 2014 The structure of ant–plant ecological seemingly stable populations? Ecol. Lett. 23, competitive interactions and overall structure of fig networks: is abundance enough? Ecology 95, 1766–1775. (doi:10.1111/ele.13603) wasp communities. J. Anim. Ecol. 91, 1303–1315. 475–485. (doi:10.1890/12-1647.1) 13. Reed TE, Grøtan V, Jenouvrier S, Sæther BE, Visser (doi:10.1111/1365-2656.13701) 4. de Roos AM. 2021 Dynamic population stage ME. 2013 Population growth in a wild bird is 23. Hallett LM, Shoemaker LG, White CT, Suding KN. structure due to juvenile–adult asymmetry buffered against phenological mismatch. Science 2019 Rainfall variability maintains grass-forb species stabilizes complex ecological communities. Proc. 340, 488–491. (doi:10.1126/science.1232870) coexistence. Ecol. Lett. 22, 1658–1667. (doi:10. Natl Acad. Sci. USA 118, e2023709118. (doi:10. 14. Hansen BB et al. 2019 More frequent extreme 1111/ele.13341) 1073/pnas.2023709118) climate events stabilize reindeer population 24. Dunson WA, Travis J. 1991 The role of abiotic factors 5. Peterson ML, Doak DF, Morris WF. 2019 dynamics. Nat. Comm. 10, 1–17. (doi:10.1038/ in community organization. Am. Nat. 138, Incorporating local adaptation into forecasts of s41467-019-09332-5) 1067–1091. (doi:10.1086/285270) species’ distribution and abundance under climate 15. Ogutu JO, Piepho H-P, Kanga E. 2012 Dynamics of an 25. Román-Palacios C, Wiens JJ. 2020 Recent responses change. Glob. Chang Biol. 25, 775–793. (doi:10. insularized and compressed impala population: rainfall, to climate change reveal the drivers of species 1111/gcb.14562) temperature and density influences. Open Ecol. J. 5, extinction and survival. Proc. Natl Acad. 6. Neupane N, Zipkin EF, Saunders SP, Ries L. 2022 1–17. (doi:10.2174/1874213001205010001) Sci. USA 117, 4211–4217. (doi:10.1073/pnas. Grappling with uncertainty in ecological projections: 16. Compagnoni A et al. 2016 The effect of 1913007117) a case study using the migratory monarch butterfly. demographic correlations on the stochastic 26. Holt RD. 2009 Bringing the Hutchinsonian niche Ecosphere 13, e03874. (doi:10.1002/ecs2.3874) population dynamics of perennial plants. Ecol. into the 21st century: ecological and evolutionary 7. Lancaster LT, Morrison G, Fitt RN. 2017 Life history Monogr. 86, 480–494. (doi:10.1002/ecm.1228) perspectives. Proc. Natl Acad. Sci. USA 106, trade-offs, the intensity of competition, and 17. Villellas J, Doak DF, García MB, Morris WF. 2015 19 659–19 665. (doi:10.1073/pnas.0905137106) coexistence in novel and evolving communities Demographic compensation among populations: 27. Veldhuis MP, Hofmeester TR, Balme G, Druce DJ, under climate change. Phil. Trans. R. Soc. B 372, what is it, how does it arise and what are its Pitman RT, Cromsigt JPGM. 2020 Predation risk 20160046. (doi:10.1098/rstb.2016.0046) implications? Ecol. Lett. 18, 1139–1152. (doi:10. constrains herbivores’ adaptive capacity to warming. 8. Hefley TJ, Hooten MB, Russell RE, Walsh DP, Powell 1111/ele.12505) Nat. Ecol. Evol. 4, 1069–1074. (doi:10.1038/s41559- JA. 2017 When mechanism matters: Bayesian 18. Capdevila P, Stott I, Beger M, Salguero-Gómez R. 020-1218-2) forecasting using models of ecological diffusion. 2020 Towards a comparative framework of 28. Harley CDG. 2011 Climate change, keystone Ecol. Lett. 20, 640–650. (doi:10.1111/ele.12763) demographic resilience. Trends Ecol. Evol. 35, predation, and biodiversity loss. Science 334, 9. Petchey OL et al. 2015 The ecological forecast 776–786. (doi:10.1016/j.tree.2020.05.001) 1124–1127. (doi:10.1126/science.1210199) horizon, and examples of its uses and determinants. 19. Paniw M et al. 2022 Higher temperature extremes 29. Post E. 2013 Erosion of community diversity and Ecol. Lett. 18, 597–611. (doi:10.1111/ele.12443) exacerbate negative disease effects in a social stability by herbivore removal under warming. Proc. 10. Stenseth NChr, Mysterud A. 2002 Climate, changing mammal. Nat. Clim. Chang. 12, 284–290. (doi:10. R. Soc. B 280, 20122722. (doi:10.1098/rspb.2012. phenology, and other life history traits: nonlinearity 1038/s41558-022-01284-x) 2722) and match–mismatch to the environment. Proc. 20. Chesson P, Kuang JJ. 2008 The interaction between 30. Griffiths JI, Childs DZ, Bassar RD, Coulson T, Reznick Natl Acad. Sci. USA 99, 13 379–13 381. (doi:10. predation and competition. Nature 456, 235–238. DN, Rees M. 2020 Individual differences determine 1073/pnas.212519399) (doi:10.1038/nature07248) the strength of ecological interactions. Proc. Natl Acad. Sci. USA 117, 17 068–17 073. (doi:10.1073/ 45. Travis J, Reznick D, Bassar RD, López-Sepulcre A, 59. Coulson T, Benton TG, Lundberg P, Dall SRX, Kendall 10 pnas.2000635117) Ferriere R, Coulson T. 2014 Do eco-evo feedbacks BE. 2006 Putting evolutionary biology back in the royalsocietypublishing.org/journal/rspb 31. Rasher DB et al. 2020 Keystone predators govern help us understand nature? Answers from studies of ecological theatre: a demographic framework the pathway and pace of climate impacts in a the Trinidadian guppy. In Advances in ecological mapping genes to communities. Evol. Ecol. Res. 8, subarctic marine ecosystem. Science 369, research (eds J Moya-Laraño, J Rowntree, G 1155–1171. 1351–1354. (doi:10.1126/science.aav7515) Woodward), pp. 1–40. New York, NY: Academic 60. Peers MJL et al. 2020 Climate change increases 32. Miller LP, Matassa CM, Trussell GC. 2014 Press. predation risk for a keystone species of the boreal Climate change enhances the negative effects of 46. Walsh MR, Fraser DF, Bassar RD, Reznick DN. 2011 forest. Nat. Clim. Chang. 10, 1149–1153. (doi:10. predation risk on an intermediate consumer. The direct and indirect effects of guppies: 1038/s41558-020-00908-4) Glob. Change Biol. 20, 3834–3844. (doi:10.1111/ implications for life-history evolution in Rivulus 61. Quéroué M et al. 2021 Multispecies integrated gcb.12639) hartii. Funct. Ecol. 25, 227–237. (doi:10.1111/j. population model reveals bottom-up dynamics in a 33. Shoemaker LG et al. 2020 Integrating the 1365-2435.2010.01786.x) seabird predator–prey system. Ecol. Monogr. 91, underlying structure of stochasticity into community 47. Bassar RD et al. 2012 Direct and indirect ecosystem e01459. (doi:10.1002/ecm.1459) ecology. Ecology 101, e02922. (doi:10.1002/ effects of evolutionary adaptation in the Trinidadian 62. Paniw M et al. 2021 The myriad of complex Proc. R. Soc. B 290: 20221494 ecy.2922) guppy (Poecilia reticulata). Am. Nat. 180, 167–185. demographic responses of terrestrial mammals to 34. Pironon S et al. 2018 The ‘Hutchinsonian niche’ as (doi:10.1086/666611) climate change and gaps of knowledge: a global an assemblage of demographic niches: implications 48. Bassar RD, Heatherly 2nd T, Marshall MC, Thomas analysis. J. Anim. Ecol. 90, 1398–1407. (doi:10. for species geographic ranges. Ecography 41, SA, Flecker AS, Reznick DN. 2015 Population size- 1111/1365-2656.13467) 1103–1113. (doi:10.1111/ecog.03414) structure-dependent fitness and ecosystem 63. Stillman RA, Railsback SF, Giske J, Berger U, Grimm 35. Medvigy D, Wofsy SC, Munger JW, Hollinger DY, consequences in Trinidadian guppies. J. Anim. Ecol. V. 2015 Making predictions in a changing world: Moorcroft PR. 2009 Mechanistic scaling of ecosystem 84, 955–968. (doi:10.1111/1365-2656.12353) the benefits of individual-based ecology. Bioscience function and dynamics in space and time: 49. Bassar RD et al. 2010 Local adaptation in 65, 140–150. (doi:10.1093/biosci/biu192) ecosystem demography model version 2. J. Geophys. Trinidadian guppies alters ecosystem processes. 64. Clutton-Brock T, Sheldon BC. 2010 Individuals and Res. 114, G01002. (doi:10.1029/2008JG000812) Proc. Natl Acad. Sci. USA 107, 3616–3621. (doi:10. populations: the role of long-term, individual-based 36. Adler PB, Dalgleish HJ, Ellner SP. 2012 Forecasting 1073/pnas.0908023107) studies of animals in ecology and evolutionary plant community impacts of climate variability and 50. El-Sabaawi RW, Marshall MC, Bassar RD, López- biology. Trends Ecol. Evol. 25, 562–573. (doi:10. change: when do competitive interactions matter? J. Sepulcre A, Palkovacs EP, Dalton C. 2015 Assessing 1016/j.tree.2010.08.002) Ecol. 100, 478–487. (doi:10.1111/j.1365-2745.2011. the effects of guppy life history evolution on 65. Lasky JR, Hooten MB, Adler PB. 2020 What 01930.x) nutrient recycling: from experiments to the field. processes must we understand to forecast regional- 37. Coulson T. 2021 Environmental perturbations and Freshw. Biol. 60, 590–601. (doi:10.1111/fwb.12507) scale population dynamics? Proc. R. Soc. B 287, transitions between ecological and evolutionary 51. Marshall MC et al. 2012 Effects of consumer 20202219. (doi:10.1098/rspb.2020.2219) equilibria: an eco-evolutionary feedback framework. interactions on benthic resources and ecosystem 66. Morris WF, Ehrlén J, Dahlgren JP, Loomis AK, Peer Comm. J. 1, e6. (doi:10.24072/pcjournal.4) processes in a neotropical stream. PLoS ONE 7, Louthan AM. 2020 Biotic and anthropogenic forces 38. Evans MR, Norris KJ, Benton TG. 2012 Predictive e45230. (doi:10.1371/journal.pone.0045230) rival climatic/abiotic factors in determining global ecology: systems approaches. Phil. Trans. R. Soc. B 52. Abrego N et al. 2021 Accounting for species plant population growth and fitness. Proc. Natl 367, 163–169. (doi:10.1098/rstb.2011.0191) interactions is necessary for predicting how arctic Acad. Sci. USA 117, 1107–1112. (doi:10.1073/pnas. 39. Bassar RD, Childs DZ, Rees M, Tuljapurkar S, Reznick arthropod communities respond to climate change. 1918363117) DN, Coulson T. 2016 The effects of asymmetric Ecography 44, 885–896. (doi:10.1111/ecog.05547) 67. Cahill AE et al. 2013 How does climate change competition on the life history of Trinidadian 53. Suttle KB, Thomsen MA, Power ME. 2007 Species cause extinction? Proc. R. Soc. B 280, 20121890. guppies. Ecol. Lett. 19, 268–278. (doi:10.1111/ele. interactions reverse grassland responses to changing (doi:10.1098/rspb.2012.1890) 12563) climate. Science 315, 640–642. (doi:10.1126/ 68. Urban MC et al. 2016 Improving the forecast for 40. Bassar RD, Travis J, Coulson T. 2017 Predicting science.1136401) biodiversity under climate change. Science 353, coexistence in species with continuous ontogenetic 54. Descombes P et al. 2020 Novel trophic interactions aad8466. (doi:10.1126/science.aad8466) niche shifts and competitive asymmetry. Ecology under climate change promote alpine plant 69. Sabatini FM et al. 2021 sPlotOpen – an 98, 2823–2836. (doi:10.1002/ecy.1969) coexistence. Science 370, 1469–1473. (doi:10.1126/ environmentally balanced, open-access, global 41. Anaya-Rojas JM et al. 2021 The evolution of size- science.abd7015) dataset of vegetation plots. Glob. Ecol. Biogeogr. 30, dependent competitive interactions promotes 55. Blois JL, Zarnetske PL, Fitzpatrick MC, Finnegan S. 1740–1764. (doi:10.1111/geb.13346) species coexistence. J. Anim. Ecol. 90, 2704–2717. 2013 Climate change and the past, present, and 70. Magurran AE et al. 2010 Long-term datasets in (doi:10.1111/1365-2656.13577) future of biotic interactions. Science 341, 499–504. biodiversity research and monitoring: assessing change 42. Potter T, King L, Travis J, Bassar RD. 2019 (doi:10.1126/science.1237184) in ecological communities through time. Trends Ecol. Competitive asymmetry and local adaptation in 56. Adler PB, Ellner SP, Levine JM. 2010 Coexistence of Evol. 25, 574–582. (doi:10.1016/j.tree.2010.06.016) Trinidadian guppies. J. Anim. Ecol. 88, 330–342. perennial plants: an embarrassment of niches. Ecol. 71. White L, O’Connor NE, Yang Q, Emmerson MC, (doi:10.1111/1365-2656.12918) Lett. 172, 1019–1029. (doi:10.1111/j.1461-0248. Donohue I. 2020 Individual species provide 43. Bassar RD, Simon T, Roberts W, Travis J, Reznick DN. 2010.01496.x) multifaceted contributions to the stability of 2017 The evolution of coexistence: reciprocal 57. Travis J et al. 2014 Integrating the invisible fabric of ecosystems. Nat. Ecol. Evol. 4, 1–8. (doi:10.1038/ adaptation promotes the assembly of a simple nature into fisheries management. Proc. Natl Acad. Sci. s41559-020-01315-w) community. Evolution 71, 373–385. (doi:10.1111/ USA 111, 581–584. (doi:10.1073/pnas.1305853111) 72. Becker EA, Foley DG, Forney KA, Barlow J, Redfern evo.13086) 58. Bowler CH, Weiss-Lehman C, Towers IR, Mayfield JV, Gentemann CL. 2012 Forecasting cetacean 44. Goldberg JF, Fraser DF, Lamphere BA, Reznick DN. MM, Shoemaker LG. 2022 Accounting for abundance patterns to enhance management 2022 Differential habitat use and recruitment demographic uncertainty increases predictions for decisions. Endanger Species Res. 16, 97–112. facilitate coexistence in a community with intraguild species coexistence: a case study with annual (doi:10.3354/esr00390) predation. Ecology 103, e03558. (doi:10.1002/ecy. plants. Ecol. Lett. 25, 1618–1628. (doi:10.1111/ele. 73. de la Riva EG, Lloret F, Pérez-Ramos IM. 2017 The 3558) 14011) importance of functional diversity in the stability of Mediterranean shrubland communities after the 88. Zipkin EF, Sillett TS, Grant EHC, Chandler RB, Royle and non-parametric methods for short-term 11 impact of extreme climatic events. J. Plant Ecol. 10, JA. 2014 Inferences about population dynamics population forecasting. Oikos 123, 652–661. royalsocietypublishing.org/journal/rspb 281–293. (doi:10.1093/jpe/rtw027) from count data using multistate models: a (doi:10.1111/j.1600-0706.2014.00916.x) 74. Luna P, Corro EJ, Antoniazzi R, Dáttilo W. 2020 comparison to capture–recapture approaches. Ecol. 103. Detto M, Visser MD, Wright SJ, Pacala SW. 2019 Measuring and linking the missing part of Evol. 4, 417–426. (doi:10.1002/ece3.942) Bias in the detection of negative density biodiversity and ecosystem function: the diversity of 89. Bellier E, Kéry M, Schaub M. 2016 Simulation-based dependence in plant communities. Ecol. Lett. 22, biotic interactions. Diversity 12, 86. (doi:10.3390/ assessment of dynamic N-mixture models in the 1923–1939. (doi:10.1111/ele.13372) d12030086) presence of density dependence and environmental 104. Freilich MA, Wieters E, Broitman BR, Marquet PA, 75. Santamaría S, Galeano J, Pastor JM, Méndez M. stochasticity. Methods Ecol. Evol. 7, 1029–1040. Navarrete SA. 2018 Species co-occurrence networks: 2016 Removing interactions, rather than species, (doi:10.1111/2041-210X.12572) can they reveal trophic and non-trophic interactions casts doubt on the high robustness of pollination 90. Clare JDJ, Linden DW, Anderson EM, MacFarland in ecological communities? Ecology 99, 690–699. networks. Oikos 125, 526–534. (doi:10.1111/oik. DM. 2016 Do the antipredator strategies of shared (doi:10.1002/ecy.2142) 02921) prey mediate intraguild predation and mesopredator 105. Okuyama T, Bolker BM. 2012 Model-based, 76. García-Callejas D, Molowny-Horas R, Araújo MB. suppression? Ecol. Evol. 6, 3884–3897. (doi:10. response-surface approaches to quantifying indirect Proc. R. Soc. B 290: 20221494 2018 The effect of multiple biotic interaction types 1002/ece3.2170) interactions. In Trait-mediated indirect interactions: on species persistence. Ecology 99, 2327–2337. 91. Roth T, Bühler C, Amrhein V. 2016 Estimating effects ecological and evolutionary perspectives (eds T (doi:10.1002/ecy.2465) of species interactions on populations of Ohgushi, O Schmitz, RD Holt). Cambridge, UK: 77. Clark JS, Scher CL, Swift M. 2020 The emergent endangered species. Am. Nat. 187, 457–467. Cambridge University Press. interactions that govern biodiversity change. Proc. (doi:10.1086/685095) 106. Bartomeus I, Gravel D, Tylianakis JM, Aizen MA, Natl Acad. Sci. USA 117, 17 074–17 083. (doi:10. 92. Brodie JF et al. 2018 Models for assessing local-scale Dickie IA, Bernard-Verdier M. 2016 A common 1073/pnas.2003852117) co-abundance of animal species while accounting framework for identifying linkage rules across 78. Callaway RM et al. 2002 Positive interactions among for differential detectability and varied responses to different types of interactions. Funct. Ecol. 30, alpine plants increase with stress. Nature 417, the environment. Biotropica 50, 5–15. (doi:10. 1894–1903. (doi:10.1111/1365-2435.12666) 844–848. (doi:10.1038/nature00812) 1111/btp.12500) 107. Lim M, Hastie T. 2015 Learning interactions via 79. Matías L, Godoy O, Gómez-Aparicio L, Pérez-Ramos 93. Briscoe NJ et al. 2019 Forecasting species range hierarchical group-Lasso regularization. J. Comput. IM. 2018 An experimental extreme drought reduces dynamics with process-explicit models: matching Graph Stat. 24, 627–654. (doi:10.1080/10618600. the likelihood of species to coexist despite methods to applications. Ecol. Lett. 22, 1940–1956. 2014.938812) increasing intransitivity in competitive networks. (doi:10.1111/ele.13348) 108. Brose U et al. 2019 Predator traits determine food- J. Ecol. 106, 826–837. (doi:10.1111/1365-2745. 94. Thuiller W et al. 2013 A road map for integrating web architecture across ecosystems. Nat. Ecol. Evol. 12962) eco-evolutionary processes into biodiversity models. 3, 919–927. (doi:10.1038/s41559-019-0899-x) 80. Alexander JM, Diez JM, Levine JM. 2015 Novel Ecol. Lett. 16, 94–105. (doi:10.1111/ele.12104) 109. Martyn TE, Stouffer DB, Godoy O, Bartomeus I, competitors shape species’ responses to climate 95. Matthiopoulos J, Fieberg J, Aarts G, Beyer HL, Pastore AI, Mayfield MM. 2021 Identifying ‘useful’ change. Nature 525, 515–518. (doi:10.1038/ Morales JM, Haydon DT. 2015 Establishing the link fitness models: balancing the benefits of added nature14952) between habitat selection and animal population complexity with realistic data requirements in 81. Gelfand AE, Ghosh S, Clark JS. 2013 Scaling dynamics. Ecol. Monogr. 85, 413–436. (doi:10.1890/ models of individual plant fitness. Am. Nat. 197, integral projection models for analyzing size 14-2244.1) 415–433. (doi:10.1086/713082) demography. Stat. Sci. 28, 641–658. (doi:10.1214/ 96. Usinowicz J, Levine JM. 2018 Species persistence 110. Elwert F. 2013 Graphical causal models. In 13-STS444) under climate change: a geographical scale Handbook of causal analysis for social research (ed. 82. Ghosh S, Gelfand AE, Clark JS. 2012 Inference for coexistence problem. Ecol. Lett. 21, 1589–1603. SL Morgan), pp. 245–273. Dordrecht, The size demography from point pattern data using (doi:10.1111/ele.13108) Netherlands: Springer Netherlands. integral projection models. J. Agric. Biol. Environ. 97. Glennie R, Adam T, Leos-Barajas V, Michelot T, 111. Pearl J. 2000 Causality: models, reasoning and Stat. 17, 642–677. Photopoulou T, McClintock BT. 2022 Hidden Markov inference. Cambridge, UK: Cambridge University 83. Kery M, Royle JA. 2020 Applied hierarchical models: pitfalls and opportunities in ecology. Press. modeling in ecology: analysis of distribution, Methods Ecol. Evol. 14, 43–56. (doi:10.1111/2041- 112. Metcalf CJE, Koons DN. 2007 Environmental abundance and species richness in R and BUGS: 210X.13801) uncertainty, autocorrelation and the evolution of Volume 2: Dynamic and advanced models, p. 820. 98. García-Callejas D, Bartomeus I, Godoy O. 2021 The survival. Proc. R. Soc. B 274, 2153–2160. (doi:10. London, UK: Academic Press. spatial configuration of biotic interactions shapes 1098/rspb.2007.0561) 84. Shriver RK et al. 2019 Transient population coexistence-area relationships in an annual plant 113. Postuma M, Schmid M, Guillaume F, Ozgul A, Paniw M. dynamics impede restoration and may promote community. Nat. Commun. 12, 6192. (doi:10.1038/ 2020 The effect of temporal environmental ecosystem transformation after disturbance. Ecol. s41467-021-26487-2) autocorrelation on eco-evolutionary dynamics across life Lett. 22, 1357–1366. (doi:10.1111/ele.13291) 99. Paniw M et al. 2020 Assessing seasonal histories. Ecosphere 11, e03029. (doi:10.1002/ecs2.3029) 85. Dail D, Madsen L. 2011 Models for estimating demographic covariation to understand 114. Paniw M, Ozgul A, Salguero-Gómez R. 2018 abundance from repeated counts of an open environmental-change impacts on a hibernating Interactive life-history traits predict sensitivity of metapopulation. Biometrics 67, 577–587. (doi:10. mammal. Ecol. Lett. 23, 588–597. (doi:10.1111/ele. plants and animals to temporal autocorrelation. 1111/j.1541-0420.2010.01465.x) 13459) Ecol. Lett. 21, 275–286. (doi:10.1111/ele.12892) 86. Hostetler JA, Chandler RB. 2015 Improved state- 100. Hart SP, Freckleton RP, Levine JM. 2018 How to 115. Lloret F et al. 2016 Climatic events inducing die-off space models for inference about spatial and quantify competitive ability. J. Ecol. 106, in Mediterranean shrublands: are species’ responses temporal variation in abundance from count data. 1902–1909. (doi:10.1111/1365-2745.12954) related to their functional traits? Oecologia 180, Ecology 96, 1713–1723. (doi:10.1890/14-1487.1) 101. García-Callejas D, Godoy O. 2020 cxr: a toolbox for 961–973. (doi:10.1007/s00442-016-3550-4) 87. Brintz B, Fuentes C, Madsen L. 2018 An asymptotic modelling species coexistence in R. Methods Ecol. 116. Paniw M, Riva EG, Lloret F. 2021 Demographic traits approximation to the N-mixture model for the Evol. 11, 1221–1226. (doi:10.1111/2041-210X.13443) improve predictions of spatiotemporal changes in estimation of disease prevalence. Biometrics 74, 102. Ward EJ, Holmes EE, Thorson JT, Collen B. 2014 community resilience to drought. J. Ecol. 109, 1512–1518. (doi:10.1111/biom.12913) Complexity is costly: a meta-analysis of parametric 3233–3245. (doi:10.1111/1365-2745.13597) 117. Guardiola-Albert C, Jackson CR. 2011 Potential 124. Dietze MC et al. 2018 Iterative near-term ecological change. Authorea Preprints. (doi:10.22541/au. 12 impacts of climate change on groundwater supplies forecasting: needs, opportunities, and challenges. 164848872.26565315/v1) royalsocietypublishing.org/journal/rspb to the Doñana wetland, Spain. Wetlands 31, 907. Proc. Natl Acad. Sci. USA 115, 1424–1432. (doi:10. 132. Boettiger C. 2022 The forecast trap. Ecol. Lett. 25, (doi:10.1007/s13157-011-0205-4) 1073/pnas.1710231115) 1655–1664. (doi:10.1111/ele.14024) 118. Sergio F, Blas J, Tanferna A, Hiraldo F. 2021 125. Paniw M, Quintana-Ascencio PF, Ojeda F, Salguero- 133. Oreskes N, Shrader-Frechette K, Belitz K. 1994 Protected areas enter a new era of uncertain Gómez R. 2017 Accounting for uncertainty in dormant Verification, validation, and confirmation of challenges: extinction of a non-exigent falcon in life stages in stochastic demographic models. Oikos numerical models in the Earth sciences. Science Doñana National Park. Anim. Conserv. 25, 480–491. 126, 900–909. (doi:10.1111/oik.03696) 263, 641–646. (doi:10.1126/science.263.5147.641) (doi:10.1111/acv.12752) 126. Moreno S, Villafuerte R. 1995 Traditional 134. Zipkin EF et al. 2021 Addressing data integration 119. Vilà M, Sardans J. 1999 Plant competition in management of scrubland for the conservation of challenges to link ecological processes across scales. Mediterranean-type vegetation. J Veg. Sci. 10, rabbits Oryctolagus cuniculus and their predators in Front. Ecol. Environ. 19, 30–38. (doi:10.1002/fee. 281–294. (doi:10.2307/3237150) Doñana National Park, Spain. Biol. Conserv. 73, 2290) 120. Lloret F, Granzow-de la Cerda I. 2013 Plant 81–85. (doi:10.1016/0006-3207(95)90069-1) 135. Tredennick AT, Teller BJ, Adler PB, Hooker G, Ellner competition and facilitation after extreme drought 127. Tredennick AT, Hooten MB, Adler PB. 2017 Do we SP. 2018 Size-by-environment interactions: a Proc. R. Soc. B 290: 20221494 episodes in Mediterranean shrubland: does damage need demographic data to forecast plant population neglected dimension of species’ responses to to vegetation cover trigger replacement by juniper dynamics? Methods Ecol. Evol. 8, 541–551. (doi:10. environmental variation. Ecol. Lett. 21, 1757–1770. woodland? J. Veg. Sci. 24, 1020–1032. (doi:10. 1111/2041-210X.12686) (doi:10.1111/ele.13154) 1111/jvs.12030) 128. Sales LP, Galetti M, Carnaval A, Monsarrat S, 136. Ellner SP, Childs DZ, Rees M. 2016 Data-driven 121. Hunter CM, Caswell H. 2005 The use of the vec- Svenning JC, Pires MM. 2022 The effect of past modelling of structured populations: a practical guide permutation matrix in spatial matrix population defaunation on ranges, niches, and future to the integral projection model. Berlin, Germany: models. Ecol. Model. 188, 15–21. (doi:10.1016/j. biodiversity forecasts. Glob. Chang. Biol. 28, Springer. ecolmodel.2005.05.002) 3683–3693. (doi:10.1111/gcb.16145) 137. Paniw M, García-Callejas D, Lloret F, Bassar RD, 122. Compagnoni A et al. 2021 Herbaceous perennial 129. De Valpine P. 2003 Better inferences from Travis J, Godoy O. 2023 Data from: Pathways to plants with short generation time have stronger population-dynamics experiments using Monte global-change effects on biodiversity: new responses to climate anomalies than those with Carlo state-space likelihood methods. Ecology 84, opportunities for dynamically forecasting longer generation time. Nat. Commun. 12, 1824. 3064–3077. (doi:10.1890/02-0039) demography and species interactions. Dryad Digital (doi:10.1038/s41467-021-21977-9) 130. Fay R et al. 2022 Temporal correlations among Repository. (doi:10.5061/dryad.8cz8w9gvc) 123. Paniw M, Lloret F. 2023 Data from: Pathways demographic parameters are ubiquitous but highly 138. Paniw M, García-Callejas D, Lloret F, Bassar RD, to global-change effects on biodiversity: variable across species. Ecol. Lett. 25, 1640–1654. Travis J, Godoy O. 2023 Pathways to global-change new opportunities for dynamically (doi:10.1111/ele.14026) effects on biodiversity: new opportunities for forecasting demography and species interactions. 131. Rademaker M, van Leeuwen A, Smallegange I. 2022 dynamically forecasting demography and species GitHub. (https://github.com/MariaPaniw/shrub_ Why we should not necessarily expect life history interactions. Figshare. (doi:10.6084/m9.figshare.c. forecast) strategies to inform on sensitivity to environmental 6403999)