Spatial Multinomial Regression Models for the Analysis of Land Cover Data

• Activity Number: 161
• Type: Topic Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Jun Zhu*+ and Chongyang Jin and Stephan Sain and Michelle M. Steen-Adams
• Companies: Colorado State University and University of Wisconsin-Madison and National Center for Atmospheric Research and University of New England
• Keywords: Bayesian hierarchical model ; categorical data ; dimension reduction ; geographic information systems ; multivariate CAR model ; spatial statistics

Abstract

Land cover data in environmental and ecological studies are becoming increasingly abundant at the advances of geographic information systems, but statistical methods for analyzing such that data are limited. Here we aim at developing practical tools for regression analysis of spatially-referenced, multi-categorical data. We propose a spatial multinomial model in a generalized linear mixed model framework with a latent spatial random effect. The spatial random effect is multivariate at each spatial location and modeled by a multivariate conditional autoregressive model. Under a Bayesian hierarchical modeling framework, we devise Markov chain Monte Carlo algorithms to obtain posterior for statistical inference. The methodology is applied to analyze a land cover data set in northern Wisconsin.