LIMB PROPORTIONS PREDICT AQUATIC HABITS IN EXTINCT TETRAPODS: A CASE STUDY FOR ASSESSING PREDICTIVE MODEL ACCURACY IN PALEONTOLOGY

Aquatic tetrapods provide striking examples of convergent evolution. They tell a story, repeated over two dozen times in the last 300 Myr, of how a terrestrial group adapts to life in the water, modifying its anatomy, behavior, and physiology to handle an aquatic medium. Reconstructing the adaptive trajectories of these groups can provide insight into the ecological drivers of recurring evolutionary radiations, the developmental processes underpinning convergence, and the factors promoting extinction in modern aquatic mammals and reptiles. However, to reconstruct the adaptive trajectories of these groups, we need to determine just how aquatic they were, and this effort continues to thwart paleontologists. Modern aquatic tetrapods span a behavioral continuum, from transiently aquatic species that seldom skirt the water to fully aquatic species that never near the shore. This continuum, and the variety of taxonomic affinities and locomotor modes displayed by aquatic tetrapods, makes it difficult to reliably predict just how aquatic different species were. Different modalities of data often disagree, and it can be challenging to assess which predictive models we should affirm in cases of conflict. Perhaps as a result, the aquatic affinities of many extinct tetrapods remain contentious. We addressed this issue by developing a new pipeline for constructing and assessing the accuracy of predictive models in paleontology. To do this, we generated the largest dataset of tetrapod limb measurements to date, fit phylogenetic binomial logistic regression models to the data, evaluated model accuracy with Receiver Operating Characteristic (ROC) curves, and fed contentious fossil taxa into our models to predict their aquatic affinities. Our workflow is distinguished by the use of ROC analysis—a statistical toolkit largely untouched by paleontologists—to assess the predictive accuracy of our models. Following this pipeline, we predict the aquatic habits and flipper phenotypes of extinct species. In doing so, we shed light on the ecomorphology of historically contentious taxa, and highlight the power of ROC analysis to evaluate predictions in Earth-system science.