Application Of A Three-step Forecasting Method In Flood Season Precipitation Forecasting In China

A three-tiered prediction scheme based on the North American multi-model ensemble for predicting SSTs is proposed from the practical operation of flood precipitation prediction.The scheme uses SST predictions from three advanced coupled Earth system models,Can CM4,GFDL-CM2p5,and GEOS-S2 S.After statistical revision,predicted SSTs are used to force BCC?AGCM for uncoupled simulations.Thereafter,dynamical downscalings are conducted with lateral boundary conditions from previous simulations using CWRF.A set of hindcast experiments(from 1991 to 2013)were conducted then analyzed for verification of the threetiered method.A set of coupled predictions using BCC?CSM was also used for better comparisons.Analysis showed that,The CWRF downscaling significantly improves pattern correlation coefficients(PCC)of distribution of the predicted climatology surface air temperature & daily precipitation and reduces root mean square error(RMSE)based on the global model predictions.Difference within coupled and uncoupled driven predictions is relatively small,which coincides the design of three-tiered prediction scheme to superimpose predicted SST anomaly on same observed SST climate state.Moreover,a stable climate state is beneficial to the subsequent model parameter tuning.Extreme precipitation indices including PCT95 and SDII are more sensitive to the physical processes within the model,and the coarse-grid global model cannot correctly capture the extreme precipitation features while the CWRF model shows a more realistic spatial distribution of rainbands and extreme precipitation magnitudes,although the precipitation magnitudes are higher than the interpolated results of ground-based station observations.The three-tiered prediction method significantly increases both the statistical mean and the spread range of interannual correlation coefficients of surface air temperature and precipitation prediction compared with the coupled prediction.Monthly anomaly correlation coefficients(ACC)show a noticeable decrease after initial Month(March)for all schemes.But two members of the three-tiered predictions still show advantages compared to the coupled scheme.By extracting the spatial and temporal variability characteristics from observed precipitation and performing the correlation analysis with atmospheric circulation and SST,this study first identifies the key circulation regions and oceanic regions that influence spatial and temporal distribution of precipitation in China,and then explains the logical route of threetiered prediction method to reduce the precipitation prediction error on this basis.Owing to the lower SST prediction error,three-tiered prediction simulates the high-level U-winds and lowlevel jets that are closer to the real atmospheric conditions driven by the same atmospheric circulation module,thus improving the precipitation prediction skill in China.The vertical profiles of spatial correlation coefficients at different vertical levels predicted by different schemes show that CWRF has stronger skill in predicting low-level circulation and water vapor distribution which helps the prediction of precipitation.Finally,this study applies the support vector regression method to ensemble the results of different members of the three-tiered scheme predictions and evaluates them over a fifteen-year validation period.The results show that the ensemble support vector regression predictions demonstrate significant improvements in predicting both climate regimes and interannual anomalies.