The Study Of Extreme Forecast Index Based On CMA T213Ensemble Prediction

The occurrence of extreme weather events in China is very frequent, so improvingthe forecasting quality of extreme weather events is of great significance to thenational disaster prevention and mitigation. As a new method of probabilistic forecast,ensemble prediction technology can provide a variety of possible future weatherpatterns according to a set of ensemble forecasting results, including the extremeweather pattern. Based on own ensemble prediction system, European Centre forMedium Range Weather Forecasts （ECMWF） designed a new ensemble forecastproducts to identify the extreme weather--extreme weather forecast index.In this paper, on the basis of analysis of the characteristic of the T213model dataand use the mathematical solution of EFI from ECMWF for reference, the EFI isestablished based on T213Ensemble prediction system. Then the recognition tests aredone for four extreme low temperature processes in2008and the recognition effect ofthe3rd process is analysed in detail. The major work and conclusion of this study issummarized as follows:（1） The difference between the observation and the model climate cumulativeprobability shows that the T213model data is more fit to generate the climatecumulative probability distribution. The thresholds of EFI₃in every lead time thatreleasing low temperature warning signal is determined by the TS score. Therecognition tests of extreme low temperature are done according to these thresholds,the warning signal maps shows, EFI₃do well in identifying the extreme lowtemperature, most regions of extreme low temperatures in analysis field can beidentified by EFI₃5days in advance.（2） Relative Operating Characteristic（ROC） curve is used to evaluate the skill ofidentifying extreme low temperature of EFI₃. The results show, the area of the ROC curves in every lead time are all more than0.5, denoting the skill of identifyingextreme low temperature of EFI₃is positive.（3） Two new set model climate cumulative probability distribution are used togenerate the EFI₃and the identity skills of EFI₃computed by the different modelclimate cumulative distribution are compared. The ROC curves show, the identityskill of EFI₃computed by the second set model climate cumulative probabilitydistribution is lower than that computed by the first set, the identity skill of EFI₃computed by the third set model climate cumulative probability distribution is lowerthan that computed by the first set in24h and48h lead time while is higher in72h,96h and120h lead time.（4） A new method computing the weight of every ensemble members ispresented. In this method, the T213ensemble members are grouped by the modelclimate equal probability interval, the weight of ensemble members is calculated bythe length of interval and the member number in every interval. Through this method,a new set EPS cumulative probability distribution is derived. Then the identity skill ofEFI₃computed by the different EPS cumulative probability distribution are compared.The ROC curves show the identity skill of EFI₃computed by the second set EPScumulative probability is lower than that computed by the first set, the difference isnot obvious.（5） The EFI calculation expression is revised on the basis of A-D test, the identityskill of EFI₃is compared with that of EFI_AD, then the identity skill of EFI_ADcomputedby the first model climate cumulative probability distribution is compared with thatcomputed by the first set. The ROC curves show the identity skill of EFI_ADis higherthat of EFI₃, the identity skill of EFI_ADcomputed by the third set model climatecumulative probability distribution is lower than that computed by the first set in24hand48h lead time while is higher in72h,96h and120h lead time.