Integrating evolutionary history with species distribution models significantly improves data-deficient species predictions

Background/Question/Methods

In recent decades, ecology has seen unprecedented growth in the amount and variety of biodiversity data collected. Species distribution models (SDMs) are a popular tool that relate occurrence data with environmental variables to characterize a species distribution. However, the accuracy with which SDMs are able to model species’ distributions is limited by the amount of data available- as much as 30% of known species continue to lack sufficient data for appropriate characterization of their geographic distributions. While SDMs have become increasingly sophisticated, they are ineffective for data-deficient species. Here we present a novel modelling framework that extends SDMs and allows data-deficient species to borrow predictive strength from data-rich species. Specifically, we demonstrate how shared evolutionary history among species can inform the likelihood of shared environmental distributions and in turn, be used to estimate the spatial distributions of data-deficient species. We quantify shared environmental history as ‘phylogenetic correlation’ – the degree to which species close together on the phylogeny share distribution characteristics - and leverage this to modulate information shared across species. Our model has two aims: i) to quantify the phylogenetic correlation across a given clade and ii) utilize phylogenetic information and data-rich species to estimate data-deficient species' distributions.

Results/Conclusions

Through simulations, we find that our model is able to accurately recover phylogenetic correlation both when true correlation is low (species distributions in a clade are effectively independent to each other i.e. no shared environmental distribution characteristics) and high (when species distributions across a clade are closely related to each other i.e. very similar environmental distributions). Given a positive phylogenetic correlation, our model significantly outperformed single SDM estimates in predicting data-deficient species’ distributions. As phylogenetic correlation decreases, our model’s estimates converge to a single SDM estimate. Therefore, when distributions across a clade are phylogenetically structured, the model borrows predictive strength from data-abundant species to inform data-deficient neighbors, however when there is no phylogenetic structure and distributions are sufficiently different, no information is shared between species. Here, we demonstrate phylogenetic information can be used where available data limits the use of traditional modelling approaches, i.e. in the cases of species lacking occurrence records. This work represents a concrete way forward for SDMs to integrate ancillary information and improve predictions of data-deficient species distributions. This modelling framework has wide applicability in helping us estimate how our biodiversity is distributed across geographic space.