Daily Spatio-Temporal Stochastic Weather Simulation

• Activity Number: 643
• Type: Topic Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): William Kleiber*+ and Rick Katz and Balaji Rajagopalan
• Companies: National Center for Atmospheric Research and National Center for Atmospheric Research and University of Colorado at Boulder
• Keywords: Simulation ; Stochastic Weather Generator ; Gaussian Process ; Multivariate ; Precipitation ; Temperature

Abstract

Gridded daily weather simulations are key features of downscaling, hydrological and agricultural models, as well as climate impact studies. Often, spatially consistent weather simulations are required across a domain at locations even without observational data. We introduce an approach to daily weather simulation relying on multivariate latent and transformed Gaussian processes. In particular, we simulate daily maximum temperature, minimum temperature and precipitation, and discuss modeling approaches that allow the relationship between variables to vary across the simulation domain. A two-part model ensures locally accurate, as well as spatially and temporally correlated, simulations. The method is illustrated on a large domain in the Pampas region of Argentina where substantial seasonality in temperature and precipitation patterns yield challenging features for stochastic weather generators to replicate.