Multi-model Downscaling Ensemble Prediction in National Climate Center

Dynamic model is the dominant tool for the seasonal prediction operation in most climate prediction centers of the world. But now, for any single model, the predictability to seasonal precipitation and temperature is quite limited. Therefore, two kinds of techniques (i.e., multi-model ensemble and downscaling) are developed efficiently to access better prediction ability. Multi-model ensemble can reduce model error and then bring higher prediction skills. Meanwhile, as the model predictability of circulation is better than that of precipitation and temperature, downscaling improves the prediction of temperature and precipitation via regional model or statistic methods.Due to the complex physical mechanism, the seasonal prediction to China climate is much a challenge. China National Climate Center (NCC) develops a new kind of prediction technique combining multi-model ensemble and downscaling. At present, the output variables from four seasonal models from WMO GPCs (including ECMWF, TCC, NCEP and NCC) are used as predictors and four statistic downscaling methods (EOF-ITE, BP-CCA, Optical Subset Regression, Regress Ensemble of High Correlation Factors) are used to set prediction model. Every model output and every downscaling method are used so that 16 model-downscaling components are available. Besides, two methods (equal-weighted average, classic super-ensemble) are employed to access the ensemble result, respectively. As an index showing the prediction ability, the mean PS scores are computed for the reforecast of recent five years for every model-downscaling and ensemble component. The component with highest mean PS score is chosen as the best prediction result.In NCC, the Multi-model Downscaling Ensemble Prediction System (MODES) are set up to realize the above ideas and the operational application of monthly and seasonal temperature with precipitation over China. Reforecast and operational application are carried out. The present reforecast and operational application for seasonal climate indicates that MODES has achieved quite good prediction skills for temperature and also improved precipitation prediction. The real-time application for monthly climate prediction for from September 2012 to July 2013 is assessed with NCC traditional PS methods. For monthly mean temperature, MODES holds the mean and median PS score of 76 and 81, respectively, showing much good prediction ability. Meanwhile, for the monthly precipitation of MODES, the mean and median PS scores are both 68, higher than mean scores of the operational prediction product of NCC. The reforecast and operational application indicate that MODES is a useful tool for the short-term climate operation prediction.