An application of statistical downscaling to estimate surface air temperature in Japan

Abstract

In this study, a statistical downscaling model based on singular value decomposition was applied to estimate the monthly mean temperature field in Japan for January and July. The regression model estimated surface air temperature in Japan from upper air temperature in east Asia with root-mean-square errors of around 1.0°C for an independent verification period. The method was applied to the output of a CO2 transient run of NCAR-CSM, and the result was compared with the output of NCAR-RegCM2.5 nested in CSM. The statistical model reproduced the spatial distribution of the temperature more realistically than the CSM output. In January the results corresponded well between the models except that the climate simulated by RegCM was generally cooler than that estimated using the statistical method mainly due to the unrealistic RegCM topography. In July the results did not correspond well between the methods, since the climate is more complex and difficult to be estimated solely from the upper air temperature field. Temperature rise from 1CO2 climate to 2CO2 climate was larger in RegCM than in the statistical method for both months reflecting the influence of sea surface temperature rise. It was concluded that this statistical downscaling method can be applied to estimate the January mean temperature in Japan, although other predictor variables such as sea surface temperature should be included to improve the estimation in July.