Improving species distribution models for climate change studies: variable selection and scale

Statistical species distribution models (SDMs) are widely used to predict the potential changes in species distributions under climate change scenarios. We suggest that we need to revisit the conceptual framework and ecological assumptions on which the relationship between species distributions and environment is based. We present a simple conceptual framework to examine the selection of environmental predictors and data resolution scales. These vary widely in recent papers, with light inconsistently included in the models. Focusing on light as a necessary component of plant SDMs, we briefly review its dependence on aspect and slope and existing knowledge of its influence on plant distribution. Differences in light regimes between north‐ and south‐facing aspects in temperate latitudes can produce differences in temperature equivalent to moves 200 km polewards. Local topography may create refugia that are not recognized in many climate change SDMs using coarse‐scale data. We argue that current assumptions about the selection of predictors and data resolution need further testing. Application of these ideas can clarify many issues of scale, extent and choice of predictors, and potentially improve the use of SDMs for climate change modelling of biodiversity.     

Complementary strengths of spatially-explicit and multi-species distribution models

Species distribution models (SDMs) project the outcome of community assembly processes – dispersal, the abiotic environment and biotic interactions – onto geographic space. Recent advances in SDMs account for these processes by simultaneously modeling the species that comprise a community in a multivariate statistical framework or by incorporating residual spatial autocorrelation in SDMs. However, the effects of combining both multivariate and spatially-explicit model structures on the ecological inferences and the predictive abilities of a model are largely unknown. We used data on eastern hemlock Tsuga canadensis and five additional co-occurring overstory tree species in 35 569 forest stands across Michigan, USA to evaluate how the choice of model structure, including spatial and non-spatial forms of univariate and multivariate models, affects ecological inference about the processes that shape community composition as well as model predictive ability. Incorporating residual spatial autocorrelation via spatial random effects did not improve out-of-sample prediction for the six tree species, although in-sample model fit was higher in the spatial models. Spatial models attributed less variation in occurrence probability to environmental covariates than the non-spatial models for all six tree species, and estimated higher (more positive) residual co-occurrence values for most species pairs. The non-spatial multivariate model was better suited for evaluating habitat suitability and hypotheses about the processes that shape community composition. Environmental correlations and residual correlations among species pairs were positively related, perhaps indicating that residual correlations were due to shared responses to unmeasured environmental covariates. This work highlights the importance of choosing a non-spatial model formulation to address research questions about the species–environment relationship or residual co-occurrence patterns, and a spatial model formulation when within-sample prediction accuracy is the main goal.     

Niche variability and its consequences for species distribution modeling

When species distribution models (SDMs) are used to predict how a species will respond to environmental change, an important assumption is that the environmental niche of the species is conserved over evolutionary time-scales. Empirical studies conducted at ecological time-scales, however, demonstrate that the niche of some species can vary in response to environmental change. We use habitat and locality data of five species of stream fishes collected across seasons to examine the effects of niche variability on the accuracy of projections from Maxent, a popular SDM. We then compare these predictions to those from an alternate method of creating SDM projections in which a transformation of the environmental data to similar scales is applied. The niche of each species varied to some degree in response to seasonal variation in environmental variables, with most species shifting habitat use in response to changes in canopy cover or flow rate. SDMs constructed from the original environmental data accurately predicted the occurrences of one species across all seasons and a subset of seasons for two other species. A similar result was found for SDMs constructed from the transformed environmental data. However, the transformed SDMs produced better models in ten of the 14 total SDMs, as judged by ratios of mean probability values at known presences to mean probability values at all other locations. Niche variability should be an important consideration when using SDMs to predict future distributions of species because of its prevalence among natural populations. The framework we present here may potentially improve these predictions by accounting for such variability.     

Disentangling scale dependencies in species environmental niches and distributions

Understanding species’ responses to environmental conditions, and how these ­species–environment associations shape spatial distributions, are longstanding goals in ecology and biogeography. However, an essential component of species–environment relationships – the spatial unit, or grain, at which they operate – remains unresolved. We identify three components of scale‐dependence in analyses of species–environment associations: 1) response grain, the grain at which species respond most strongly to their environment; 2) environment spatial structure, the pattern of spatial autocorrelation intrinsic to an environmental factor; and 3) analysis grain, the grain at which analyses are conducted and ecological inferences are made. We introduce a novel conceptual framework that defines these scale components in the context of analyzing species–environment relationships, and provide theoretical examples of their interactions for species with various ecological attributes. We then use a virtual species approach to investigate the impacts of each component on common methods of measuring and predicting species–environment relationships. We find that environment spatial structure has a substantial impact on the ability of even simple, univariate species distribution models (SDMs) to recover known species–­environment associations at coarse analysis grains. For simulated environments with ‘fine’ and ‘intermediate’ spatial structure, model explanatory power, and the frequency with which simple SDMs correctly estimated a virtual species’ response to the simulated environment, dramatically declined as analysis grain increased. Informed by these results, we use a scaling analysis to identify maximum analysis grains for individual environmental factors, and a scale optimization procedure to determine the grain of maximum predictive accuracy. Implementing these analysis grain thresholds and model performance standards in an example east African study system yields more accurate distribution predictions, compared to SDMs independently constructed at arbitrary analysis grains. Finally, we integrate our conceptual framework with virtual and empirical results to provide practical recommendations for researchers asking common questions about species–environment relationships.     

Identifying the relevant spatial and temporal scales in plant species occurrence models: The case of arable weeds in landscape mosaic of crops

Species distribution models (SDMs) represent potential valuable tools to explore factors underlying species occurrence over a large range of spatial scales. However, a recurrent problem with this approach is identifying the appropriate spatial and temporal scales for modeling. This problem is reinforced in plant populations for which it is often difficult to evaluate the limits of habitat patches. In this study, we aimed at developing SDMs for 13 arable weeds in highly dynamic small agricultural region. Although weed dynamic is widely thought to result from local processes, we explored the spatial and temporal scales that would best explain species occurrence over the area. Models were developed using weed occurrence data in 58 fields over four consecutive years (2008–2011) and spatial organization of management practices over the landscape for eight consecutive years (2004–2011). We used a model selection approach based on the minimum AIC criteria to select the best SDMs. Results showed that SDMs can successfully be applied to model weed occurrence over a small region. The appropriate temporal scale to consider in weed SDMs should encompass several years to reflect the effect of management history while the relevant spatial scale should extend beyond the crop field itself and include the field border and neighboring fields. This study illustrates that adopting a multiple scale approach is successful to model plant occurrence over a highly dynamic landscape.     

Microclimate species distribution models estimate lower levels of climate-related habitat loss for salamanders

Species distribution models (SDMs) largely rely on free-air temperatures at coarse spatial resolutions to predict habitat suitability, potentially overlooking important microhabitat. Integrating microclimate data into SDMs may improve predictions of organismal responses to climate change and support targeting of conservation assets at biologically relevant scales, especially for small, dispersal-limited species vulnerable to climate-change-induced range loss. We integrated microclimate data that account for the buffering effects of forest vegetation into SDMs at a very high spatial resolution (3 m²) for three plethodontid salamander species in Great Smoky Mountains National Park (North Carolina and Tennessee). Microclimate SDMs were used to characterize potential changes to future plethodontid habitat, including habitat suitability and habitat spatial patterns. Additionally, we evaluated spatial discrepancies between predictions of habitat suitability developed with microclimate and coarse-resolution, free-air climate data. Microclimate SDMs indicated substantial losses to plethodontid ranges and highly suitable habitat by mid-century, but at much more conservative levels than coarse-resolution models. Coarse-resolution SDMs generally estimated higher mid-century losses to plethodontid habitat compared to microclimate models and consistently undervalued areas containing highly suitable microhabitat. Furthermore, microclimate SDMs revealed potential areas of future gain in highly suitable habitat within current species’ ranges, which may serve as climatic microrefugia. Taken together, this study highlights the need to develop microclimate SDMs that account for vegetation and its biophysical effects on near-surface temperatures. As microclimate datasets become increasingly available across the world, their integration into correlative and mechanistic SDMs will be imperative for accurately estimating organismal responses to climate change and helping environmental managers tasked with spatially prioritizing conservation assets.     

Pushing the limits in marine species distribution modelling: lessons from the land present challenges and opportunities

Aim Species distribution models (SDMs) have been used to address a wide range of theoretical and applied questions in the terrestrial realm, but marine‐based applications remain relatively scarce. In this review, we consider how conceptual and practical issues associated with terrestrial SDMs apply to a range of marine organisms and highlight the challenges relevant to improving marine SDMs. Location We include studies from both marine and terrestrial systems that encompass many geographic locations around the globe. Methods We first performed a literature search and analysis of marine and terrestrial SDMs in ISI Web of Science to assess trends and applications. Using knowledge from terrestrial applications, we critically evaluate the application of SDMs in marine systems in the context of ecological factors (dispersal, species interactions, aggregation and ontogenetic shifts) and practical considerations (data quality, alternative modelling approaches and model validation) that facilitate or create difficulties for model application. Results The relative importance of ecological factors to be considered when applying SDMs varies among terrestrial and marine organisms. Correctly incorporating dispersal is frequently considered an important issue for terrestrial models, but because there is greater potential for dispersal in the ocean, it is often less of a concern in marine SDMs. By contrast, ontogenetic shifts and feeding have received little attention in terrestrial SDM applications, but these factors are important to many marine SDMs. Opportunities also exist for applying more advanced SDM approaches in the marine realm, including mechanistic ecophysiological models, where water balance and heat transfer equations are simpler for some marine organisms relative to their terrestrial counterparts. Main conclusions SDMs have generally been under‐utilized in the marine realm relative to terrestrial applications. Correlative SDM methods should be tested on a range of marine organisms, and we suggest further development of methods that address ontogenetic shifts and feeding interactions. We anticipate developments in, and cross‐fertilization between, coupled correlative and process‐based SDMs, mechanistic eco‐physiological SDMs, and spatial population dynamic models for climate change and species invasion applications in particular. Comparisons of the outputs of different model types will provide insight that is useful for improved spatial management of marine species.     

Do species distribution models explain spatial structure within tree species ranges?

Aim To evaluate the ability of species distribution models (SDMs) to predict the spatial structure of tree species within their geographical ranges (how trees are distributed within their ranges). Location Continental Spain. Methods We used an extensive dataset consisting of c. 90,000 plots (1 plot km^?2) where presence/absence data for 23 common Mediterranean and Atlantic tree species had been surveyed. We first generated SDMs relating the presence or absence of each species to a set of 16 environmental predictors, following a stepwise modelling process based on maximum likelihood methods. Superimposing spatial correlograms generated from the predictions of the SDMs over those generated from the raw data allowed a model–observation comparison of the nature, scale and intensity (level of aggregation) of spatial structure with the species ranges. Results SDMs predicted accurately the nature and scale of the spatial structure of trees. However, for most species, the observed intensity of spatial structure (level of aggregation of species in space) was substantially greater than that predicted by the SDMs. On average, the intensity of spatial aggregation was twice that predicted by SDMs. In addition, we also found a negative correlation between intensity of aggregation and species range size. Main conclusions Standard SDM predictions of spatial structure patterns differ among species. SDMs are apparently able to reproduce both the scale and intensity of species spatial structure within their ranges. However, one or more missing processes not included in SDMs results in species being substantially more aggregated in space than can be captured by the SDMs. This result adds to recent calls for a new generation of more biologically realistic SDMs. In particular, future SDMs should incorporate ecological processes that are likely to increase the intensity of spatial aggregation, such as source–sink dynamics, fine‐scale environmental heterogeneity and disequilibrium.     

Using Species Distribution Models For Fungi

Species distribution models (SDMs) are an emerging tool in the study of fungi, and their use is expanding across species and research topics. To summarise progress to date and to highlight important considerations for future users, we review 283 studies that apply SDMs to fungi. We found that macrofungi, lichens, and pathogenic microfungi are most often studied. While many studies only aim to model species response to environmental covariates, the use of SDMs for explicitly predicting fungal occurrence in space and time is growing. Many studies collect fungal occurrence data, but the use of pre-collected records from reference collections and citizen science programs is increasing. Challenges of applying SDMs to fungi include detection and sampling biases, and uncertainties in identification and taxonomy. Further, finding environmental covariates at appropriate spatial and temporal scales is important, as fungi can respond to fine-scale environmental patterns. Fine-scale covariate data can be difficult to gather across space, but we show remote-sensing measurements are viable for fungi SDMs. For those fungi interacting with host species, host information is also important, and can be used as covariates in SDMs. We also highlight that competition among fungi, and dispersal, can affect observed distributions, with the latter particularly prominent for invasive fungi. We show how one can account for these processes in models, when suitable data are available. Finally, we note that environmental DNA records create new opportunities and challenges for future modelling efforts, and discuss the difficulties in predicting invasions and climate change impacts. The application of SDMs to fungi has already provided interesting lessons on how to adapt modelling tools for specific questions, and fungi will continue to be relevant test subjects for further technical development of SDMs.     

Spatial modelling of ecological indicator values improves predictions of plant distributions in complex landscapes

Ecologically meaningful predictors are often neglected in plant distribution studies, resulting in incomplete niche quantification and low predictive power of species distribution models (SDMs). Because environmental data are rare and expensive to collect, and because their relationship with local climatic and topographic conditions are complex, mapping them over large geographic extents and at high spatial resolution remains a major challenge. Here, we propose to derive environmental data layers by mapping ecological indicator values in space. We combined ~6 million plant occurrences with expert-based plant ecological indicator values (EIVs) of 3600 species in Switzerland. EIVs representing local soil properties (pH, moisture, moisture variability, aeration, humus and nutrients) and climatic conditions (continentality, light) were modelled at 93 m spatial resolution with the Random Forest algorithm and 16 predictors representing meso-climate, land use, topography and geology. Models were evaluated and predictions of EIVs were compared with soil inventory data. We mapped each EIV separately and evaluated EIV importance in explaining the distribution of 500 plant species using SDMs with a set of 30 environmental predictors. Finally, we tested how they improve an ensemble of SDMs compared to a standard set of predictors for ca 60 plant species. All EIV models showed excellent performance (|r| > 0.9) and predictions were correlated reasonably (|r| > 0.4) to soil properties measured in the field. Resulting EIV maps were among the most important predictors in SDMs. Also, in ensemble SDMs overall predictive performance increased, mainly through improved model specificity reducing species range overestimation. Combining large citizen science databases to expert-based EIVs is a powerful and cost–effective approach for generalizing local edaphic and climatic conditions over large areas. Producing ecologically meaningful predictors is a first step for generating better predictions of species distribution which is of main importance for decision makers in conservation and environmental management projects.