Statistical Downscaling Estimates of High Temperature in Summer in Japan

Abstract

The development of a quantitative method to predict regional extreme high temperature due to global warming is necessary. In this study, a statistical downscaling model for estimating monthly mean daily maximum temperature in August in Japan was proposed and examined. The model related the local variable with the principal components of large-scale climate variables over the region using stepwise multiple regression analysis. The principal components of air temperature and zonal wind at 850hPa level, sea level pressure, and geopotential height at 500hPa level were selected as eligible predictors. The statistical model was evaluated by the cross validation procedure. The correlation coefficients between the observation and the regression estimates were significant at most stations, and the model estimated the observation fairly well. Therefore it was confirmed that the method is applicable to the estimation of high temperature in the region. The method was then applied to the output of NCAR-CSM. The 1×CO₂ climate downscaled from the global model output was generally cooler than the observation due to the underestimate of 850hPa air temperature to the north of Japan showing that the downscaling model reflects deviations in the global model.