Integrating processes and data types for predicting species distributions across spatial scales

Background/Question/Methods

The geographic distributions and environmental niche associations of species are determined by processes that vary in relevance over spatial scales. Similarly, our ability to capture species distributions - e.g. through museum records, citizen science observations, or expert knowledge – also varies with spatial scale, and so do the correlates of potential observation bias. Identifying the relative importance of key processes for mapping species or projecting distribution change thus requires an integration of both data and processes at their appropriate and relevant scales. Spatial hierarchical models in a Bayesian framework enable this integration. Through the development of such models and their application to a number of ecological examples, data types, and processes we demonstrate the promise of a new generation of distribution models that inherently address spatial scale.

Results/Conclusions

Using a variety of datasets from local to global scales and for terrestrial and freshwater species and communities we illustrate how multi-scale Hierarchical Bayesian models are able to more appropriately address ecological processes in species distribution modeling. For a North American bird species we show how the coarse-grain detections and non-detections can be downscaled to finer resolutions and used to study species-environment associations at finer grain that is limited only by the grain of the environmental data. We also introduce, and apply to South African birds, a method that enables the estimation of richness-environment association and prediction of geographic patterns of species richness at grains finer than the original grain of observation. The method is based on a hierarchical model that uses coarse-grain values of species richness and fine-grain environmental data as input. Further, we apply a similar, scale-and data-integrative set of models to North American freshwater fishes. Finally, we chart out new methodological advances in the field based on point process modeling and related approaches.