|
Abstract:
|
Due to lack of direct, scientific observations of climate over the last few hundred years, there is a need for development of statistical reconstructions of climate using proxy data. For many climate proxies including tree rings, ice cores, and pollen records, there has been development into relevant methods; for other proxy data, including testate amoebae counts, there has been less development. Testate amoebae are single celled organisms that live in peat bogs and grow a shell that leaves a record in peat cores. Historically, weighted averaging, partial least squares, and other "transfer function" methods that model each species' marginal response to climate through time are used to infer wetness of the bog. We propose a Bayesian approach inspired by multispecies modeling in ecology, wherein we model the response of the entire species assemblage to climate jointly while accounting for second-order interactions among species that change along gradients of the climate variable. Through simulation and cross-validation using real data, we demonstrate that our method achieves higher predictive skill while properly accounting for uncertainty.
|
