Conference Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 56 - Modern Bayesian Methods for Complex Spatial Data
Type: Topic Contributed
Date/Time: Sunday, August 7, 2022 : 4:00 PM to 5:50 PM
Sponsor: International Society for Bayesian Analysis (ISBA)
Abstract #322903
Title: Spatial Factor Models for High-Dimensional Binary Data Across Large Spatial Domains: A Case Study on Breeding Birds in the United States
Author(s): Jeffrey W. Doser* and Andrew O. Finley and Sudipto Banerjee
Companies: Michigan State University and Michigan State University and UCLA
Keywords: Nearest Neighbor Gaussian Process; latent factors; spatial prediction; joint species distribution model; binary data; Bayesian

Understanding spatial variation in individual species and community assemblages are fundamental goals of both applied and theoretical ecology. Bird communities globally have recently shown massive declines in abundance, leading to increased interest in quantifying variation in individual species distributions as well as biodiversity metrics of avian communities across large spatial domains. This requires modeling spatially dependent, binary detection-nondetection data for large community assemblages (e.g., 100 species) across vast spatial domains. With this as a motivating example, we develop a spatial factor Nearest Neighbor Gaussian Process model for high-dimensional binary outcomes. We leverage Polya-Gamma latent variables to yield an efficient Markov Chain Monte Carlo sampler and discuss its implementation in the spOccupancy R package. We estimate distributions of 98 bird species across the continental United States using our proposed model, in which we embed a detection sub-model to account for imperfect detection of bird species. We subsequently generate species richness maps of two bird communities with associated uncertainty across the continental United States.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2022 program