Abstract by Michael Christensen
Climate Model Validation via Local Approximate Gaussian Processes
Gaussian processes are a common tool in spatial statistics due to their flexibility and natural integration into the Bayesian statistical framework. In this presentation, we perform climate model validation by comparing simulated data for precipitation in High Mountain Asia (HMA) produced by the Weather Research and Forecasting (WRF) climate model to remote sensing data from the Tropical Rainfall Measuring Mission satellite. We accomplish this using local approximate Gaussian processes, which provide many of the benefits of traditional Gaussian process approaches while vastly improving on their computational speeds. We explore the discrepancy surface between these data sets and asses WRF’s performance. We also propose future work using a spatiotemporal process convolution model to better understand trends in HMA’s climate across time.