Abstract
Seasonal precipitation at the decadal time scale is predicted using the downscaling super ensemble (DSE) method, which is developed by combining the superensemble procedure with a statistical downscaling method in this study. The multi‐model data utilized are the long‐term integration of six atmosphere–ocean general circulation models (AOGCMs) and the downscaling method is based on the singular value decomposition with the empirical orthogonal function (EOF) truncation to correct the systematic bias in the dynamic models.
Interestingly, even though prediction skill in the training period is increased with increasing number of AOGCMs used, the skill is often decreased in the independent period. It is found that prediction skill in the independent period continues to rise when we use an optimal combination of predictors. The optimum combination used in constructing the superensemble model is the super‐3 ensemble, which is a combination of three AOGCMs (CCCma, CSIRO, and NCAR) among the six AOGCMs used in this study. In general, the first four EOFs of sea‐level pressure (SLP) in the super‐3 ensemble are very similar to those of the observed SLP. The dynamic link between Korean winter precipitation and East Asian monsoon circulation in the super‐3 ensemble is similar to that of the observed indicating that the super‐3 ensemble realistically simulates the circulations in the East Asian monsoon region. The cross‐validation for the prediction of the super‐3 ensemble shows that the correlation skill score is about 0.49, which is significant at the 5% level. The results provide hope for regional climate prediction in decadal time‐scales using superensemble methods together with statistical downscaling. Copyright © 2004 Royal Meteorological Society