Evaluation of the Predictive Performance of a Statistical Climate Downscaling Model

• Activity Number: 633
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics and the Environment
• Author(s): Jenise Swall*+ and Benjamin Wells
• Companies: Virginia Commonwealth University (VCU) and Environmental Protection Agency
• Keywords: climate downscaling ; climate models ; temporal correlation ; climate research

Abstract

Statistical downscaling methods rely on correlations among observed and modeled meteorological variables to predict, based on low-resolution climate model simulations, probable future local weather conditions. As compared with traditional dynamical downscaling techniques, which use regional weather models guided by climate model output, the advantages of statistical downscaling methods lie in their speed, arbitrary resolution, and ability to quantify error and variability. However, it is unclear to what extent statistical models can reliably predict future conditions using only statistical associations observed in the recent past. Once a statistical downscaling model has been developed using data from a particular time period, it is unclear whether the relationships it incorporates will remain the same under future climatic conditions. This stationarity assumption is fundamental to statistical downscaling methods. This work utilizes simulations based on historical data to investigate the viability of this assumption for regression-based downscaling techniques, as well as the impact of other factors, such as the presence of substantial temporal correlation in the observations.