A new statistical downscaling model for autumn precipitation in China

Effective statistical downscaling schemes based on singular value decomposition (SVD) for boreal autumn (September‐October‐November) precipitation over China were developed. It was found that rainfall over China is closely tied to large‐scale atmospheric and oceanic circulation over specific regions. The general circulation models (GCMs), which are from DEMETER project (Development of a European Multi‐model Ensemble System for Seasonal to Interannual Prediction), perform reasonably well in simulating the mean states of geopotential height at 500 hPa (GH5) during autumns from 1960 to 2001. Consequently, the variable GH5 over East Asia from DEMETER GCMs was used as one predictor for downscaling. Meanwhile, another predictor is the preceding sea surface temperature (SST) signal from observed data over the tropical Pacific. The downscaling results involving only the GH5 predictor (GH5 scheme) or both the GH5 and SST predictors (GH5 + SST scheme) were discussed and compared in this study. Downscaling based on two kinds of schemes showed considerable improvement compared with original DEMETER GCMs in predicting regional autumn precipitation. In particular, downscaling predictions based on the GH5 + SST scheme showed lower root mean square errors than those based on the GH5 scheme, especially for the precipitation anomaly pattern of the El Niño event in 1997 and the La Niña event in 1998.