Predicting Meteorological Fields: Parallelization for Near Real-Time Forecasting Using Bayesian Hierarchical Models

• Activity Number: 342
• Type: Topic Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Quality and Productivity Section
• Author(s): Dorit Hammerling* and Matthias Katzfuss
• Companies: National Center for Atmospheric Research and Texas A&M University
• Keywords: Spatial statistics ; Bayesian Hierarchical Modeling ; Parallel computing

Abstract

Bayesian models have traditionally not been the method of choice for meteorological applications requiring near-real time predictions such as hourly weather forecasts. Mainly because such forecasts are often based on hundreds of thousands or millions of observations, which have so far rendered Bayesian model implementations computationally very challenging if not impossible. The increasing availability of high performance computing facilities might make the operational implementation of such models feasible though. Especially so, if the underlying inference and prediction algorithms are formulated in ways that allow for parallelization on large-scales. We will present results from such implementations of recently developed Bayesian hierarchical spatio-temporal models, which are conducted at the National Center for Atmospheric Research's super computing facilities.