Nonstationary Spatial Modeling of Environmental Data Using a Process Convolution Approach

• Activity Number: 373
• Type: Topic Contributed
• Date/Time: Wednesday, August 11, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Jenise Swall*+
• Companies: NOAA/EPA
• Keywords: Bayesian statistics ; spatial modeling ; environmental modeling

Abstract

Traditional approaches to modeling spatial processes involve the specification of the covariance structure of the field. Although such methods are straightforward to understand and effective in some situations, there are often problems in incorporating nonstationarity and in manipulating the large covariance matrices that result when dealing with large datasets. Our approach takes a different perspective, modeling a process as a convolution of a Gaussian white noise process and suitable kernels. Depending on the particular parameterization, this approach can allow flexibility in modeling nonstationary processes, while avoiding the task of working directly with the covariance matrix. We discuss some relevant approaches, and present an application involving environmental monitoring. In particular, we focus on such practical issues as computational efficiency and methods for assimilating data from differing sources.