A Statistical Framework for Downscaling Climate Model Information to Enhance Infrastructure Design

• Activity Number: 82 - Climate and Meteorology
• Type: Contributed
• Date/Time: Sunday, July 30, 2017 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Ernst Linder* and Meng Zhao and Yiming Liu
• Companies: Department of Mathematics and Statistics, University of New Hampshire and University of New Hampshire and University of New Hampshire
• Keywords: time series models ; rainfall models ; extremes ; peak flows ; climate change ; impact assessment

Abstract

Infrastructure design and climate impact assessment require projections of future weather scenarios at the local scale and a fine temporal resolution of days or hours. Global or regional climate models provide information at coarser spatial scale typically from 200 km to 20 km, that represents an average over a spatial grid box. The scale mismatch manifests itself as a difference in time series models between climate model output and local measurements, such as at a weather station. We propose a statistical model whose form is common to both sources of data and propose a downscaling procedure that translates parameters of the climate model output to those of the weather station data. The advantage of this approach over common approaches discussed in the climate literature is that it enables the calculation of the prediction variance and its propagation to an impact or design "end point", thus solves the quest for uncertainty quantification. Further it provides a framework for variance propagation when using multiple climate model ensembles. We provide examples of future daily temperature and precipitation as inputs for infrastructure design in the New England region.