Regional hydroclimatic projection using an coupled composite downscaling model with statistical bias corrector

The precipitation of General Circulation Model (GCM) output for Han River Basin, Korea was downscaled into a regional watershed using the Artificial Neural Network (ANN) model with multiple statistical processors of Nonstationary Quantile Mapping (NSQM) and a Stochastic Typhoon Model (STM). The stochastically generated typhoon rainfall was synthesized and added to the ANN-processed precipitation, and the summer precipitation underestimated by raw GCM was effectively recovered. The projection was evaluated in terms of the annual and seasonal quantities and annual daily maximum precipitation. The spatial dependency structure of the projection results were tested through the spatial autocorrelation indices of Moran's I and LISA (Local Indicators of Spatial Association). The results under baseline, B1 and A1B scenarios represent the effective reproduction of spatial autocorrelation, even though the hot spots of DJF under the A2 scenario shifted to an adjacent area. The seasonal analysis shows that the GCM output in JJA is relatively more reliable than other seasons in view of baseline biases (8.6%) and the conservation of spatial autocorrelation (Moran's I = 0.63). Under the 95% level of confidence, the annual precipitation by the year 2040 is projected to be 3.2%, 7.7% and 10.8% increase with respect to the baseline period under B1, A1B and A2 scenarios respectively. Under the same level of confidence, the JJA seasonal projection gives a -1.7%, 3.2% decrease and 7.5% increase under B1, A1B, and A2 scenarios, respectively. The rate of increase of JJA precipitation will be less than the annual total. Meanwhile the increasing ratio (13.2% mean increase for all scenarios) of the daily maximum precipitation is obviously higher than the annual or seasonal total precipitation, which means the impacts on hydrological extremes due to climate change would be more intense than one can sense.