Spatio-Temporal Downscaling of Daily Precipitation Extremes in New England from Regional Climate Model Outputs

• Activity Number: 241
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Yibin Pan*+ and Ernst Linder and Eric Laflamme
• Companies: and University of New Hampshire and University of New Hampshire
• Keywords: NARCCAP ; Extreme value analysis ; GPD ; Precipitation trends ; Return levels ; Bayesian hierarchical models

Abstract

Projected future precipitation extremes are a primary concern since they indicate potential changes in frequency of high-impact events, such as floods and hurricanes. We develop a spatial-temporal Bayesian model for downscaling daily precipitation extremes from regional climate model outputs and for related uncertainty quantification. This extends a station by station application of a semiparametric GPD mixture extension of the XCDFt method of Kallache et al. 2011 (see companion abstract by Laflamme et al.), by embedding the downscaling methodology in a hierarchical space-time framework. The main goal is to gain precision in return level estimation through ensemble fitting using spatial prior random fields for the extreme value parameters of station data, and of current and future regional climate model outputs. The method is applied to 58 station data of the National Climatic Data Center (NCDC) network New England and to NARCCAP model outputs. We examine improvements in return level estimation relative to single station analyses and provide regional assessments of projected precipitation trends.