Improved Performance of Simulated Japanese Climate with a Multi-Model Ensemble

Abstract

This paper analyzed the downscaling products for Japanese climate using five regional climate models (RCMs) with horizontal meshsize of 20 km with boundary conditions given from JRA-25 reanalysis. The RCMs successfully reproduced the temporal variations and geographic distribution of temperature and precipitation. Skill scores for the surface temperature were improved by the downscaling. The JRA-25 underestimated the precipitation amount for summer and winter seasons, and the RCMs reduced the error, especially in winter. The RCMs showed common features, such as a warm bias in the areas with a monthly-mean temperature lower than the freezing point, an overestimation of weak rainy days, and an underestimation of heavy rainy days. The comparison among five RCMs suggests that the warm bias is due to the lack of model resolution and the precipitation bias is related to the convective parameterization. The multi-RCM ensemble mean has considerable advantages over the individual RCM in regards to the bias and skill scores of surface temperature and precipitation, although it still showed the warm bias in snow areas in winter. The data set of the multi-model dynamical downscaling is expected to contribute to impact studies on the forthcoming climate change in Japan.