Fully Bayesian Inference for Spatial Extremes Using Hierarchical Extreme Value Processes

• Activity Number: 400
• Type: Invited
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: IMS
• Author(s): Brian J. Reich and Ben Shaby*+
• Companies: North Carolina State University and UC - Berkeley
• Keywords: Max-stable process ; Spatial statistics ; Markov chain Monte Carlo

Abstract

We describe a an approach for constructing spatial max-stable models through a hierarchical representation that conditions on latent positive stable random variables. This class of models approximates and extends known spatial max-stable processes and, critically, is amenable to fully Bayesian inference through MCMC. Moreover, this hierarchical framework provides a foundation that can be extended in a fairly straightforward way to produce, for example, multivariate extreme value fields, or fields with more flexible spatial dependence structures.