 |
Center for Nonlinear Studies |
In analyses of spatially-referenced data, researchers often have two goals. One goal is to quantify the relationship between a response variable and a set of covariates while accounting for residual spatial dependence. Another goal is to predict the value of a response variable at unobserved locations. In this second case, when the response variable is categorical, prediction can be viewed as a classification problem. Existing classification methods for spatial data either ignore the response variable-covariate relationship and base predictions only on neighboring observations, or ignore neighboring observations and rely only on covariate information. The Bayesian spatial generalized linear (mixed) model offers a tool to accommodate both spatial and covariate sources of information in classification problems. In this talk, I formally define spatial classification rules based on these models in order to provide a principled empirical comparison to other spatial/non-spatial classification rules in the literature. I also take a close look at two different versions of these models that have been proposed in the literature, namely the Bayesian spatial generalized linear model (SGLM) and the Bayesian spatial generalized linear mixed model (SGLMM). I describe the implications of the seemingly slight differences between these models for spatial classification and explore the issue of robustness to model misspecification through a simulation study. I compare these SGLM-based classification methods to alternative classification methods in the literature using satellite-derived land cover data from Southeast Asia.
Host: Kary Myers