INFERRING EVOLUTIONARY PATTERNS FROM THE FOSSIL RECORD USING BAYESIAN INVERSION: AN APPLICATION TO SYNTHETIC STRATOPHENETIC DATA

This project formulates the inference of species-level evolutionary patterns from fossil data as an inverse problem: given morphological and stratigraphic data, how can we estimate the parameter values of models of evolution, ecophenotypy and preservation? A forward simulation, linking a high-resolution basin-fill model (SEDFLUX) to simple paleobiological models, is used to discover the statistical relationships that, for given values of the model parameters, allow predictions of values on observable parameters (simulated data). Probabilistic (Bayesian) inverse theory offers a coherent framework for incorporating uncertainty in both observed data and model, as well as information on their relationship obtained from the forward simulation. For high-dimensional nonlinear inverse problems where no analytical expression for the forward relation is available, the general solution requires Monte Carlo methods of sampling and optimization in the space of feasible solutions, providing measures of resolution and uncertainty of the parameter estimates. The Miocene sequences of the U.S. mid-Atlantic margin are well constrained in terms of sequence-, bio- and isotope stratigraphy, sedimentary facies, bathymetry and age, and available cores and outcrop contain abundant benthic microfossils. Sedimentological and stratigraphic information will be combined with morphometric measurements on microfossils to document stratophenetic series. Comparison of a common foraminifer and a less abundant ostracode will reveal the effects of stratigraphic architecture on observed patterns in groups having different ecological and taphonomic characteristics. As a means of demonstrating the method, an application of the Bayesian inversion procedure to synthetic data is presented.