Downscaling and probabilistic regional climate projection in Japan using a statistical method

Abstract:

We have developed a statistical downscaling method for estimating probabilistic climate projection using multi general circulation models (GCMs). A regression model is established so that the combination of weight of GCMs reflects the characteristics of the variation of observations at each grid point. Cross validation is conducted to select GCMs and to evaluate the regression model. Large warm biases of the surface air temperature of GCMs from CMIP3 and CMIP5 are successfully corrected by this method. The biases are much smaller than those of dynamically downscaled temperature in the present climate. When it is compared to one of the cumulative distribution function (CDF) based bias correction, it shows a comparable performance in estimating mean value of the present climate. Furthermore, this method generates probabilistic information in the future projection, such as exceedance probability of the temperature increase and several percentile values. Regarding the monthly mean of the surface air temperature, the dependency of future changes in the mean value and the standard deviation on the emission scenario are investigated. This probabilistic climate projection based on the statistical method can be expected to bring important information on the impact study and risk assessment.