Abstract:Understanding the predictability of a numerical model is an essential step before that model is added to a super ensemble prediction system.It is also very important for the development of that model.But forecast noise easily hinders the thorough understanding of the predictability of a model.The forecast always consists of useful forecast information and forecast noises,but sometimes,the proportion of forecast noises are remarkable big,especially for the primary stage of the model development.The forecast will be far apart from the observation when the forecast noises are stronger than the useful forecast information.The predictability evaluation should focus on the useful forecast information which is only decided by the physical characteristics and dynamic properties of the model.Then a valuable statistical correction method is necessary for the reasonable evaluation of the model predictability.This paper evaluates the predictability of three Atmospheric General Circulation Models(AGCM) and seven Coupled General Circulation Models(CGCM) before and after the statistical correction.We selected three AGCMs of Institute of Atmospheric Physics(IAP),for the comparison,seven CGCMs were also selected from the Development of a European Multimodel Ensemble System for Seasonal-to-Interannual Prediction project (DEMETER),the potential predictability of these AGCMs and CGCMs on summer rainfall in tropical regions,especially in the tropical western Pacific region is evaluated.Results indicate that:All of these models are able to reproduce well the spatial features of mean rainfall of the past 20 years,but there are obviously differences between models for precipitation anomalies.In order to reduce the distraction of noises,the Singular Value Decomposition(SVD) method is used to estimate the predictability of models for main spatial features of observed precipitation anomalies.IAP 9-AGCM has the maximum similarity of the first SVD modes to the observed one,which means IAP 9-AGCM can reproduce the main feature of observed precipitation anomalies over the tropical western Pacific,the two IAP 2-level models also have similar spatial features,but the spatial characteristics of forecasted precipitation anomalies shift to the east about 20 degree.In the ocean area,the spatial features of CGCMs' anomalous precipitation are more similar to the observation than AGCMs',but for the west of 140E,CGCMs' anomalous precipitation are remarkable difference from those observed features,it may due to the complex topography,lots of small islands on this area could reduce the performance of the coupling between atmospheric models and ocean models.Considering the good ability of AGCMs,it reproduces the most important spatial feature of the observed precipitation anomalies which is a statistical correction method based on Empirical Orthogonal Function(EOF) analysis method is applied to hindcasts of the ten models.We calculate the Abnormal Correlation Coefficient(ACC) between observations and forecasts before and after the statistical correction.Due to the weak of rainfall intensity and the spatial shifting,the original ACCs of IAP AGCMs are remarkable lower than CGCMs',but after statistical corrections based on EOF,AGCMs show similar good ACCs of CGCM.The statistical correction method also improves ACCs of some of CGCMs,but it does not work when the original ACC is good enough.It implies that the improvement of predictability will ultimately depend on the progresses of researches on numerical models.With the successive improvement of IAP AGCM,the most obvious advance of summer rainfall appears in the tropical western Pacific,but in the eastern Pacific,hindcast precipitation anomaly remains weak.