A Spatially Varying Coefficient Model for Climate Emulation

• Activity Number: 182
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Jingyu Bao* and Michael L. Stein
• Companies: The University of Chicago and The University of Chicago
• Keywords: spatio-temporal modeling ; climate emulation ; Matérn covariance

Abstract

The regression coefficients are assumed to be independent over the region in many environmental regression problems. However, in some problems, coefficients are highly correlated at the local level. We describe a new approach to emulate the output of a climate model, which is based on a spatially varying coefficient model with a spatio-temporal error. The approach captures the temperature well in transient climates based on the past trajectory of atmospheric CO2 concentrations. The flexibilities of this approach include modeling the spatio-temporal process using Matérn covariance function for the spatial patterns and structured matrix for the temporal patterns. Additionally, an approximate likelihood method is demonstrated to be useful and computationally efficient.