Speaker: Hsin-Hsiung Bill Huang (UCF, Department of Statistics & Data Science)
Title: Bayesian Spatiotemporal Modeling of Deer Mouse Populations: Gaussian Processes and Scalable Approximations
Abstract: The North American deer mouse (Peromyscus maniculatus) serves as a critical ecological indicator, with its population dynamics influenced by environmental factors such as precipitation and temperature. This talk presents a Bayesian hierarchical framework to model spatiotemporal variations in deer mouse capture data from the National Ecological Observatory Network (NEON), spanning 2013–2022 across 46 U.S. sites. We employ Gaussian Processes (GP) to capture nonlinear temporal correlations and introduce the Nearest Neighbor Gaussian Process (NNGP) as a scalable approximation for large spatial datasets. Our model integrates region-specific effects, weather covariates, and random effects, estimated via Markov Chain Monte Carlo (MCMC) methods. Results reveal significant precipitation and temperature impacts, with distinct regional and temporal patterns, validated through cross-validation against simpler models. We discuss computational challenges, ecological interpretations and future directions including multi-species extensions, demonstrating the power of Bayesian methods in biodiversity research.
Read More