Bias Correction And Superensemble Method For Seasonal-Interannual Dynamical Climate Prediction

Numerical prediction is the most effective method for short-term climateprediction in future. But now, owing to the deficiency of the climate model, statisticalmethods are used to improve the performance of numerical model is an important wayto increase the accuracy of numerical prediction. Using secular ensemble hindcastresults of different versions of IAP climate model, seven air-sea coupled models of theDEMETER projection, and high resolution air-sea coupled model, SINTEX-F ofFrontier, the paper introduces and improves a kind of correction method based onmodal analysis, and verifies the effect of this kind of modal correction method on IAPclimate model's performance on predicting East Asian seasonal precipitation, and then,due to the unavoidable shortcoming of predicting by one single model, the paperintroduces a new superensemble method, effect of the new ensemble method ontropical and subtropical zone is verified respectively; At last, using the differentversion of IAP climate model, correction method and superensemble method, an IAPsuperensemble seasonal prediction system is established preliminarily. Main resultsare summarized as followed:1, The correction methods based on modal analysis can improve obviously theperformance of IAP climate model on each seasonal precipitation of East Asian area.The correction method can improve the predictability of numerical model notably,correction prediction shows remarkable higher skill score than not only the originalresults of model, but also those of old correction method. After correction, the averageof anomaly correction coefficient (ACC) between hindcast and observation of eachseason precipitation is bigger than 0.2, and in comparison, improvement effect is mostsignificant in Spring, the correction forecast has highest skill score in Winter, theaverage of ACC reaches 0.47 and there is also notable improvement on the predictionof summer precipitation. Secular correction effect also shows that the kind ofcorrection method has well stability and use value. Although the correction methodhas good correction effect on precipitation forecast, it does not well in the correctionon air temperature and geopotential height because of the difference in temporalcontinuity of modal analysis. Besides, the skill of precipitation forecast correction alsohas an upper limit.2, An new superensemble method is created in this paper. Because of the exist ofmodel system bias and random bias, there is some unavoidable deficiency for single model prediction. So that after the skill of single model is improved by modalcorrection method, Combining the merits of equal-weight ensemble methods(averageensemble) and unequal-weight ensemble methods(regression ensemble), a newsuperensemble prediction method is established based on Ensemble Kalmanfilter(assimilation ensemble). Ideal experiments show that, under the influence ofrandom error, assimilation ensemble can obtain more precise weights than regressionensemble, and then take good advantage of merits of each model. At the same time,by including spatial features of variables, assimilation ensemble can avoid the spatialdiscontinuation caused by ensemble in single point. Results of real ensemblepredictive experiments on summer rainfall in tropical and subtropical area also provethe superiority of assimilation ensemble, assimilation ensemble has increasedpredictive skill compared to the raw model output, ideal experiments and realensemble prediction both suggest that assimilation ensemble is superior to regressionensemble, and ensemble predictive experiments with different number modelsindicate that the dispersion of samples increases as the number of models grows, itwill improve the performance of both assimilation ensemble and average ensemble,but when models which have different merits are included, assimilation ensemble willhas stabilized higher skill than average ensemble.3, IAP short-term climate superensemble prediction system is establishedpreliminarily. According to good predictive skill of IAP atmospheric models in EastAsian area and the independence of real-time prediction system, the paper creates asuperensemble prediction system using three IAP models, modal correction method,and assimilation superensemble method. Utilizing secular hindcast results of the threeversion models, the performance of the system in East Asian area are verified, resultsalso show that assimilation ensemble has higher predictive skill, the ACC reaches0.33, and it is not only bigger than the raw model output and correction result ofsingle model, there is also some increase compare to regression ensemble based onregression and average.A modal correction method is investigated and improved, and it is the first timethat modal correction method is introduced in the real-time climate prediction inChina, and the prediction skill of IAP is increased obviously. But most important isthat, in according to the unavoidable deficiency of prediction made by one singlemodel, the paper attaches more attention to superensemble prediction, a new superensemble prediction method is introduce based on ensemble Kahnan filter, andthe superensemble shows notable improvement effect; at last, a superensembleclimate prediction system of China is established firstly.