Hidden Markov Random Fields: Spatial Random Effects for Lattice Data

Abstract:

Abstract:
 

In the statistical modeling of an environmental process, it can be

useful to decompose the process into large-scale and small-scale

structure, although there can be a number of ways to accomplish this. In

many cases the small-scale structure should include spatial or temporal

dependence or both. In this work, the dependence is incorporated through

random effects with a conditional autoregressive (CAR) structure, and

the resulting model is a generalized linear mixed model (GLMM).

Pseudo-likelihood as well as Bayesian inference is addressed and two

particular examples in atmospheric science are illustrated.