Data-Dependent Basis Vectors for Dimension-Reduced Modeling of Two Space-Time Processes

• Activity Number: 166
• Type: Topic Contributed
• Date/Time: Monday, July 30, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Jenny Brynjarsdottir*+ and L. Mark Berliner
• Companies: SAMSI and The Ohio State University

Abstract

Dimension reduced approaches to spatio-temporal modeling usually involve modeling the spatial structure in terms of a low number of specified basis functions. The temporal evolution of the space-time process is then modeled through the amplitudes of the basis functions. A common choice of basis are data-dependent basis vectors such as Empirical Orthogonal Functions. We introduce ways to extend these ideas to modeling of two spatio-temporal processes where the primary goal is to predict one process from the other, using Maximum Covariance Patterns. We incorporate these basis vectors in a Bayesian hierarchical and dimension reduced model and apply these methods to downscaling of temperatures over the Antarctic.