Scientific Lectures //
Statistical Approaches for Calibration of Climate Models
Gabriel Huerta- Department of Mathematics and Statistics, University of New Mexico
Presented: August 19, 2014
ABSTRACT: We consider some recent developments to deal with climate models and that rely on various modern computational and statistical strategies. Firstly, we consider various posterior sampling strategies to study a surrogate model that approximates a climate response through the Earth’s orbital parameters. In particular, we show that for certain metrics of model skill, Adaptive/Delayed Rejection MCMC methods are effective to estimate parametric uncertainties and resolve inverse problems for climate models. We will also discuss some of the High Performance Computing efforts that are taking place to calibrate various inputs that correspond to the NCAR Community Atmosphere Model (CAM). Finally, we show how to characterize output from a Regional Climate Model through hierarchical modelling that combines Gauss Markov Random Fields (GMRF) with MCMC methods and that allows estimation of probability distribution that underlie phenomena represented by the climate output.