TESTING POTENTIAL OF MULTIVARIATE ECOMETRICS FOR PALEOENVIRONMENTAL RECONSTRUCTION USING DISCRIMINANT FUNCTION ANALYSIS

Ecometrics is a statistical method which quantifies relationships between functional anatomy and environmental characteristics in order to estimate paleoenvironments. In the past, such studies have been exclusively univariate, comparing a single anatomical trait to a single environmental variable at a time. Because of the complexity of functional anatomy and its relationships with environment, a more holistic approach would be a multivariate study model comparing several anatomical traits to a single environmental variable. Here, we test two canonical variate analysis (CVA) models for modern ungulate taxa using three anatomical variables (body mass, hypsodonty, and calcaneal gear ratio) and vegetation cover type. We acquired spatial point coordinates from the Indiana University Polly Lab’s website for every continent except Antarctica at 50 km intervals. We downloaded geographic range from the IUCN Redlist database and generated genus-level faunal lists for each sample point. We collected anatomical data for each genus from the PanTHERIA and New and Old World (NOW) databases, as well as a recent published dataset. Environmental data were acquired from the WorldClim and Oak Ridge National Laboratory databases. We performed a quadratic discriminant function analysis in R to create a model which maximizes differences between ungulate communities belonging to different vegetation cover types. We tested the sorting accuracy of this model by cross-validation with jackknifing against the actual environmental categories for each site, and a used threshold of significance for sorting accuracy determined by calculation of a “chance-corrected” baseline with TAU index. We plan to use this method to generate vegetation cover and MAP estimates for Miocene fossil sites in North America.