A Weighted Basis Kernel Method for Non-stationary Spatial Models

• Activity Number: 354
• Type: Contributed
• Date/Time: Wednesday, August 14, 2002 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Stat. Sciences*
• Author(s): Jenise Swall*+ and David Higdon
• Affiliation(s): Kenyon College and Los Alamos National Laboratory
• Keywords: Bayesian ; spatial statistics ; MCMC

Abstract

Much current research in spatial statistics has centered around modeling strategies for non-stationary spatial processes, in contrast with more traditional approaches, which assume stationarity. Our previous work has focused on modeling a non-stationary process as a convolution of a Gaussian white noise process and a series of kernels. Instead of computing with a series of individual kernels, we used weighted functions of a set of "basis'' kernels. We continue to refine and extend this approach, particularly in the areas of improving computational efficiency and incorporating uncertainty in the choice of basis kernels. We utilize the fast Fourier transform to obtain the spatial realization using the weights and the "basis" kernels much more quickly, allowing use of the model for much larger data sets.