PREDICTING MULTIVARIATE ECOLOGY FROM PHYLOGENETIC COMPARATIVE DATA YIELDS NOVEL INSIGHTS INTO THE NICHES OF FOSSIL TAXA

Many theories in macroevolution and ecology are predicated on the idea that taxa occupy discrete ecological niches and that competition for access to resources limits the number of taxa that can occupy the same niche in the same place at the same time. As a result, much effort has been expended on the quantification of ecologically important characteristics of living and extinct species. Neontologists frequently base their ecological classifications on direct observation of focal taxa, while paleontologists must predict ecological roles from fossilized remains, such as bones, teeth, and shells. However, these divergent approaches to the quantification of ecology mask a shared challenge, namely that while ecological variation is often continuously distributed, it is challenging to quantify in this form and is more typically codified using discrete classifications.

Diet is a fundamentally important component of a species' ecology and has been a particular focus of study in both modern and fossil systems. However, establishing metrics that effectively summarize dietary variability without excessive information loss remains challenging and most authors continue to reduce complex patterns of dietary variation into discrete classification systems. Here, we employ a dietary item relative importance coding scheme to derive multivariate dietary classifications for a sample of extant mammals and, using polychoric principal components analysis and using clustering algorithms, show that typical, discrete dietary coding strategies are inconsistent with this rich multivariate data . We then show how Bayesian phylogenetic multilevel modeling can be employed to predict the original item importance scores from a set of dental topographic metrics and use our models to provide novel insights into the dietary diversity of extinct species, namely the most probable composition of their diet and their closest extant analogues. Our approach need not be limited to diet as an ecological trait of interest, to these phenotypic traits, or to mammals. Rather, this framework serves as a general approach to predicting multivariate ecology from phylogenetic comparative data.