A Bayesian semiparametric model for organism based environmental reconstruction

Reconstruction of past environment based on fossil data is an important scientific problem and much work has been done to address this. One important issue in this regard is the relationship between species and environment. The available statistical approaches, both classical and Bayesian, make the assumption that species abundances occur symmetrically around a single preferable environmental condition. In other words, it is generally assumed that the relationship between species and environment is unimodal. However, this may not be true in reality. In this paper, we propose a Bayesian hierarchical model, extending previous work, to incorporate multiple preferences (or, multiple modes) of each species. In particular, we propose to use a mixture of Gaussian response curves to relate species to environment. A main novelty of our approach is the use of Dirichlet process to learn about the number of preferences of each species. We demonstrate that our multimodal modeling approach performs better than unimodal model. For implementation of our approach, we use a recently developed efficient computational procedure. Copyright (c) 2006 John Wiley & Sons, Ltd.