Adaptive Full-Scale Approximation Approach for Modeling Large Spatio-Temporal Data Sets

• Activity Number: 254
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Bohai Zhang*+ and Huiyan Sang and Jianhua Z. Huang
• Companies: Texas A&M and Texas A&M and Texas A&M
• Keywords: Spatio-temporal process ; covariance function ; Bayesian Treed Gaussian process ; Geostatistics ; RJMCMC

Abstract

Spatio-temporal data sets arising from geology, meteorology and other disciplines have received lots of interests in recent years. Although a variety of spatio-temporal covariance models have been proposed, the statistical modeling faces tremendous computational challenges when the data size is large. The full-scale approximation (FSA) with block modulating function provides an effective way in modeling large spatio-temporal data sets, where the block partition is a tuning parameter and needs to be pre-specified. Motivated by the idea of Bayesian Treed Gaussian process, we will discuss how to automatically select block partitions for the FSA method using the tree generating process. Then the following Bayesian model averaging yields a relatively smooth predictive surface. The method is illustrated through simulation studies and a real data example.