Study on neighborhood ensemble probability forecasting method of precipitation for CMA-REPS system

Abstract

In this paper, based on the regional ensemble CMA-REPS V3.1 system, the algorithm for the neighborhood ensemble probability method of precipitation is optimized. The daily 24-hour accumulated precipitation data from May to July 2021 are selected to calculate the neighborhood probability of precipitation. The grid precipitation product combined from three sources developed by the National Meteorological Information Center is selected as the observational data. The optimized method is evaluated using the area scoring method with the relative operating characteristic curve, and is compared with the scoring results of original neighborhood ensemble probability method and ensemble mean neighborhood probability method. At the same time, a typical precipitation case is selected to evaluate these three methods. It is found that the optimized method has the highest score, and its predicted information of precipitation is more consistent with the observations. In this paper, the three precipitation neighborhood probability prediction results are also used to calculate the FSS (Fractions Skill Score) of ensemble precipitation. It is found that the FSS score based on the optimized method is higher than that of ensemble mean neighborhood probability method. Both the optimized and original methods have some advantages in terms of the FSS score. The former one has better scoring for small precipitation, especially for light rain and moderate rain, and the latter one for large precipitation, especially, the scores of rainstorm is better. FSS score based on the optimized method is relatively more objective.

Keywords

• cma-reps v3.1 system
• optimized neighborhood ensemble probability method
• ensemble mean neighborhood probability method
• neighborhood ensemble probability method
• fss score
• roc curve area score