STATISTICAL DOWNSCALING OF PRECIPITATION WITH A FORMATTED REGRESSION FRAME

Abstract

5 Member of JSCE, Master of Eng., Staff, Nagoya Station, Central Japan Railway Company This paper illustrates a statistical downscaling technique considering the spatial correlation structure of precipitation. Downscaling target is 60-km resolution of daily precipitation for 20-km resolution data. We have considered a window having (3x60-km)x(3x60-km) of area, and the downscaling target is the 3x3 of 20-km resolution grids in the center of the downscaling window. For the evaluation of the proposed method, we have prepared 15 years (1979-1993) of observation data, and identify the parameters with the square root information filter scheme. We optimize the parameters on a monthly basis, and apply the regression model to 10 more years of testing period (1994-2004). The proposed regression model provides very effective and efficient results with a certain level of estimation error.