Abstract by Brent Mabey
Shane Reese, William Christensen
Gaussian Processes to Validate Precipitation Models
Precipitation around High Mountain Asia can vary wildly in both space and time, and so it is of interest to develop a computer model which captures the precipitation patterns. We utilize Gaussian Processes to emulate and validate a computer model simulation of rainfall in High Mountain Asia. First, we model the computer simulation as a Gaussian Process to produce estimates of the simulation at arbitrary points. Next, we model the discrepancy between the simulation and observed data as a separate Gaussian Process, allowing us to compute the bias of the computer simulation at arbitrary points. This model is put into a Bayesian framework to estimate the unknown parameters within it. Discrepancy surfaces are produced and used to determine the accuracy and validity of the computer model.