Computer Model Calibration with High-Dimensional Multivariate Space-Time Observations

• Activity Number: 52
• Type: Invited
• Date/Time: Sunday, July 29, 2012 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Murali Haran*+ and Klaus Keller and Roman Olson and K. Sham Bhat and Won Chang
• Companies: Penn State University and Penn State University and Penn State University and Los Alamos National Laboratory and Penn State University
• Keywords: calibration ; Gaussian processes ; kernel mixing ; spatial data ; complex computer models

Abstract

Complex computer models are frequently used to make projections about climate. Calibrating unknown parameters of these models increasingly involves high dimensional multivariate space-time observations and computer model output. Extracting information from these data sources while accounting for various uncertainties appropriately poses considerable computational and inferential challenges. I will describe Gaussian process based modeling approaches that allow for flexible dependence structures while at the same time retaining computationally tractability by the use of dimension-reduction methods. The computational efficiency of these methods allow our scientific collaborators to learn about key climate model parameters like climate sensitivity and ocean vertical diffusivity, and to study important features related to the spatial scale of complex climate processes.