The limits of direct community modeling approaches for broad-scale predictions of ecological assemblage structure

Two distinct modeling approaches are often used when predicting biodiversity patterns: stacking of species models (predict first, assemble later), and directly modeling a characteristic of a community such as species richness (assemble first, predict later). The relative utility of these two approaches for biogeographic, macroecological and global change analyses is uncertain. Here we compared the two approaches by predicting current-day avian dietary guild structure of assemblages worldwide. We found that the stacked-species modeling approach consistently predicted the geographic distribution of observed dietary guilds better than a direct community modeling approach. The exception was for plant-eating birds, especially frugivores, which are expected to have particularly strong climatic constraints on their diversity and distributions. Assemblage-level biodiversity patterns predicted by community-based modeling approaches, such as the stacked-species and direct community modeling approaches in this study, offer a means to help guide conservation decisions for determining environmental suitability and analyzing diversity hotspots. However, our results generally caution against the widespread use of direct community modeling approach at the large spatial extents for predicting species assemblages.