Comparison between two statistical downscaling methods for summer daily rainfall in Chongqing, China

Two circulation index-based methods were applied to downscale summer daily rainfall at four meteorological stations in Chongqing, China. One is a regression method (CIM) and the other a hybrid circulation classification plus regression method (CPM). Both methods used the same three circulation indices and surface specific humidity as predictors. In the first method (CIM), the indices and humidity were used directly as predictors and only one model was developed at each station. In the second method (CPM), the indices were further used to define circulation patterns using the objective Lamb-Jenkinson classification scheme, and one model was built using humidity as the sole predictor for each pattern. Logistic regression was used to define rainfall probability, and a gamma distribution was fit using observations to randomly generate daily rainfall amounts. The two downscaling methods were validated and compared. The results suggest that (1) both methods yield consistently reasonable results with respect to occurrence and amount of daily rainfall, although they perform poorly in reproducing interannual and interdecadal variability, (2) specific humidity should be used as a predictor. Both methods were forced by two future scenarios of a global climate model (GCM) to demonstrate the added value with two methods. The downscaled scenarios show a shift towards larger rainfall values, accompanied by more frequent dry days. This shift is mainly attributed to a change in specific humidity. Despite the similar performances of the two methods, CPM gives a higher frequency of dry days than CIM, whereas CIM produces stronger intensity of rainfall than CPM. This provides additional information about the uncertainty in the projections.