Testing statistical downscaling methods in simulated climates

Abstract

The consistency of two statistical downscaling methods and two different predictors to estimate past (last millennium) and future (21st century) precipitation in the Iberian and Scandinavian Peninsulas is assessed in the surrogate climate of a coupled climate model simulation. The methods are based on canonical correlation analysis and the search for analogs, with sea level pressure (SLP) and 500 mb geopotential height as predictors. The precipitation downscaled by the statistical methods within the simulated climate is compared with the direct model output. The estimation based on SLP alone agrees with the modeled precipitation on the Iberian Peninsula. However, the estimates for Scandinavia in the 21st century are drier than the target precipitation. Geopotential height as predictor performs worse than SLP in both regions yielding too dry future climate. Differences in both regions are analyzed in terms of relative humidity, which contributes to future precipitation changes of Scandinavian precipitation.