Improving Statistical Downscaling of General Circulation Models

Abstract

Credible projections of future local climate change are in demand. One way to accomplish this is to statistically downscale General Circulation Models (GCM’s). A new method for statistical downscaling is proposed in which the seasonal cycle is first removed, a physically based predictor selection process is employed and principal component regression is then used to train the regression. A regression model between daily maximum and minimum temperature at Shearwater, NS, and NCEP principal components in the 1961-2000 period is developed and validated and output from the CGCM3 is then used to make future projections. Projections suggest Shearwater’s mean temperature will be five degrees warmer by 2100.