Spatio-Temporal Modeling of Rain Rates Using Approximate Bayesian Computation

• Activity Number: 14
• Type: Topic Contributed
• Date/Time: Sunday, August 3, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Matthew Pratola*+ and Ying Sun
• Companies: Ohio State University and Ohio State University
• Keywords: Stochastic Weather Generator ; Likelihood-free ; Approximate Bayesian Computation ; Markov Chain Monte Carlo ; Gaussian Process

Abstract

Modeling spatial-temporal weather patterns using stochastic weather generators has received increasing research interest as a flexible approach to representing such complex weather patterns. Our interest is in modeling rain rates as measured using multiple rain gauges that are sparsely located in space and record high resolution measurements over time. One approach to modeling such data is to use a Bayesian latent process model conditional on parameters of interest, and to transform the latent process by a non-linear function to produce stochastically generated rain rates. However, calibrating such models to observed data is difficult since the likelihood is unavailable in closed-form. We apply a likelihood-free approach, Approximate Bayesian Computation, to construct an MCMC algorithm to fit a stochastic weather generator model to observations and sample the posterior distributions of the parameters of interest. Our method is applied to 15 minute rain rate data observed over 3 years at 12 rain gauge sites in the USA.