Species distribution models in conservation biogeography: developments and challenges

Species distribution modelling (SDM) associates georeferenced observations of a biotic response variable – typically species occurrence or abundance – with multiple environmental predictors using a broad array of statistical learning methods (Elith Franklin, 2010b; Elith Kessell, 1976; Strahler, 1981), weed or pest species risk assessment (Sutherst Busby, 1991) and studies of climate impacts on the biota (Busby, 1986; Nix this classic literature is directly related to species distribution modelling. Diversity and Distributions is a journal of conservation biogeography. Its mission is to publish papers that apply biogeographical principles, theories and methods (those addressing the distributional dynamics of taxa and assemblages) to problems concerning the conservation of biodiversity. The study of biological invasions is considered a key component of conservation biogeography, and the journal is an important forum for research on biogeographical aspects of biological invasions (Richardson, 2004; Richardson Appendix S2). Papers selected for the virtual issue include contributions that both address pressing conceptual and methodological issues and provide key examples of the use of SDM for biodiversity assessment, conservation planning, risk analysis for invasive species and forecasting global change impacts. I selected those papers that have a high rate of citation relative to time since publication (empirical evidence that they are influential and useful; Table 1) or more recently published papers that are particularly creative in their use of SDM to support conservation biogeography (my subjective judgment or prediction that they will become influential). Another aim of the editorial is to suggest some profitable avenues of research relating to SDMs, both ‘nuts and bolts’ work on the philosophical underpinnings and technical aspects of such modelling, but also how SDMs could and should be used in advancing the aims of conservation biogeography. The following sections describe the articles in the Virtual Issue and their linkages by grouping them into three areas: (1) those that address vexing methodological issues in SDM ranging from variability among modelling methods to sample size and sample design, (2) those that use SDM in innovative and rigorous ways to ‘interpolate’ species distributions in space, for example for biodiversity inventory, prospecting and conservation planning and (3) those that combine SDM with other data and methods in thoughtful ways to ‘extrapolate’ species distributions to different places or time periods to forecast impacts of environmental change on species distributions or risk of biological invasions. I conclude with prospects and priorities for future research on modelling species distributions in support of conservation biogeography research. Methodological papers included in the virtual issue tended to focus on challenges that face modellers who must rely on presence-only observations of species occurrences, such as those available from natural history collections and increasingly from global databases that compile those collections information and other observations. This group of papers also addressed methodological issues of spatial dependence and non-stationarity, sample design, data resolution, sample size and consensus forecasting. Tsoar et al. (2007) compared six presence-only modelling methods, and while they did find systematic differences in performance among methods, also found that differences among species tended to be consistent across models. Osborne et al. (2007) showed that local regression methods are appropriate for interpolation of species distributions in space, while global methods are more appropriate for extrapolation to different places or time periods. Dark (2004) also demonstrated that spatial (auto-)regression models were more effective at identifying correlates of distribution of invasive species than non-spatial models. Elith there was only a slight trend towards lower performance at the coarser scale. Wisz et al. (2008) described how SDM performance degrades with smaller sample sizes. Consensus forecasting is one way of dealing with SDM uncertainty (Araújo this information could be used for spatial conservation planning aimed at preserving amphibian diversity. Species distribution modellings are no longer enough on their own when we want to extrapolate, for example, the effects of future global (climate, land use) change on the biota, and risk of invasive species, although extrapolation has come to be their primary mode of application and has included studies remarkably broad in scope (Warren et al., ). Species distribution modellings are limited in their ability to forecast to novel environments by their empirical nature and equilibrium assumption, especially if naively applied with inadequate data. If used with explicit consideration of these limitations, however, they can be an important part of a methodological toolkit used to address pressing forecasting needs (Franklin, 2010a). There are three ways SDMs can be more effectively used for extrapolation. (1) Data or information from more mechanistic or process-based studies or models (population, ecophysiology, community dynamics) can be incorporated during conceptual and statistical formulation (see Table 9.1 in Franklin, 2010b), for example deriving explanatory variables, variable selection, model estimation, specifying interactions and response curve shape (Elith et al., 2010). (2) SDMs can be linked with process models (Franklin, 2010a). This is sometimes called hybrid modelling (Dormann et al., 2012), but often the output from one model is used as the input to another, without feedback, so ‘linked’ or ‘coupled’ modelling is more descriptive. (3) Predictions from SDMs can be compared with process-based models and much can be learned from where and how they agree and disagree, in light of their respective assumptions (for example Kearney et al., 2010; Serra-Diaz et al., In Press). In the virtual issue, Hof et al. (2012) used the first approach to extrapolation, informing their SDM with information about important biotic interactions affecting the distribution and abundance of the focal species (predator–prey dynamics). Naujokaitis-Lewis et al. (2013) used the second approach, linking SDM and population models to forecast climate change impacts in a study of Hooded Warbler to examine uncertainty due to different Global Climate Models (GCMs). Population viability estimates were sensitive to GCM effect on vital rates, but more sensitive to direct habitat loss projected from SDM. Thuiller et al. (2006) contrasted the predicted changes in plant species richness under climate change when assuming no dispersal versus unlimited dispersal from current distributions to bracket the range of outcomes likely to be generated using process models that more explicitly simulated dispersal. The problem of extrapolation to predict risk of invasive species was addressed by Beaumont et al. (2009) who found that including data from the entire (native and non-native) distribution of invasive species may better characterize its fundamental niche and better forecast potential for invasion in space and time, for example, under climate change (but see Webber et al., 2011). Václavík she found the same factors to be important in both cases (lower elevations, higher road density and higher native plant species richness), pointing to the importance of species traits in determining whether an alien species is invasive or not. Finally, a broad-reaching study recently published in Diversity and Distributions by Junker et al. (2012) modelled habitat suitability for African great ape taxa using environmental and human impact variables representing conditions in the 1990s. They projected these models to the 2000s based on updated human impact variables (population density, proximity to roads, etc.) and estimated losses of suitable habitat ranging from 11% to 59% for different taxa. While they cautioned that the coarse spatial scale of the analysis meant that it is informative to broad-, but not fine-scale conservation planning, their temporal extrapolation was short-term and well justified and was based on actual, observed changes in the driving variables (rather than modelled projections). This is an exemplary use of SDM for extrapolation over a limited time horizon in support of conservation biogeography. Because forecasting species distributions in novel or non-analogue environments is so central to conservation biogeography in an era of rapid global change (Sala et al., 2000), research that develops and tests innovative ways of forecasting impacts of global change – climate change, land use change, invasive species including emerging infectious diseases, altered disturbance regimes – on biodiversity should be of great interest to Diversity and Distributions. Hindcasting distributions to address historical and phylogeographical questions can also inform conservation biogeography (e.g. Porto et al., 2013; Smith et al., 2013). Molecular methods can provide genetic information about historical demography and dispersal dynamics of taxa (Scoble Duckett et al., 2013). Incorporating information about diversity below the species level may be particularly important for identifying genetically and geographically structured populations that may differ in their potential for genetic adaptation to environmental change (Hamann Kearney Dubuis et al., 2011; Syphard et al., 2013). Shifts in disturbance regimes that play out at large spatial scales may have important implications for conservation biogeography (Reside et al., 2012; Syphard et al., 2013). Moving forward, Diversity and Distributions is interested in publishing those studies that use insights from phylogeography and palaeodistribution dynamics, draw in cutting-edge work on genetics, and build on key developments in invasion, population and community ecology, to address critical information needs in conservation biogeography. These studies are likely to be multiscale and multidisciplinary, interfacing with climate and land change science, so that drivers of species distributions can be characterized at relevant scales. Species distribution modelling can be part of a methodological toolkit to address these information needs. Its limitations are well known, but solutions to those limitations are also being described in the growing literature on this topic. Often overcoming those limitations involves collecting additional data about species, ecological communities and habitat (Elith & Franklin, 2013). Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.