Calibration for Large Multivariate Spatial Climate Models

• Activity Number: 456
• Type: Topic Contributed
• Date/Time: Wednesday, August 1, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics and the Environment
• Author(s): K. Sham Bhat*+
• Companies: Los Alamos National Laboratory
• Keywords: computer model calibration ; uncertainty quantification ; multivariate spatial data ; dimension reduction ; surrogate models

Abstract

Characterizing uncertainty in climate predictions and models is crucial for climate decision-making. One uncertainty source is due to inability to resolve complex physical processes, which is approximated using a parameterization. We characterize parameter uncertainty by calibrating the model to physical observations. Both model output and observations are often large multivariate space-time fields.

We use Bayesian methods to infer these parameters while incorporating space-time dependence and other uncertainty sources. The climate model is approximated using a fast surrogate, a flexible Gaussian process emulator which may include nonlinear relationships among spatial fields and globally nonstationary and nonseparable covariance functions. Model discrepancy and observation error are incorporated in our approach, resulting in improved inference and characterization of uncertainty. Dimension reduction techniques are utilized for computational tractability. Our approaches are applied to output from complex ocean models.