A Probabilistic Forecast Experiment of Short-duration Heavy Rainfall in Beijing Based on CMA-BJ

Based on numerical prediction products from China Meteorological Administration Beijing model (CMA-BJ), precipitation observation of ground weather stations in Beijing and ECMWF ERA5 dataset, the hourly rainfall samples from April to September during 2019-2021 are divided into short-duration heavy rainfall (SDHR, greater than 20 mm·h-1), ordinary rainfall (between 0.1 and 19.9 mm·h-1) and no rainfall (less than 0.1 mm·h-1). The probability density distribution characteristics of physical parameters are comparatively analyzed, including moisture conditions, thermal and energy conditions, and dynamic conditions for three categories. Monthly predictors are selected from these parameters by comparing their ability to discriminate among SDHR, ordinary rainfall and no rainfall weather. It is found that the distributions and thresholds of physical parameters differ between months to some extent. Among that, moisture conditions, thermal and energy conditions, and dynamic conditions are relatively stronger but with less discrimination degrees among SDHR, ordinary rainfall and no rainfall weather in July and August. The background circulation and the distributions of physical parameters show obvious monthly differences, so the forecast model is established for each different period. After that, forecast model of SDHR for 0-12 h at 1 h intervals in different periods is established by using the ingredients-based method and fuzzy logic algorithms. When probabilistic and composite reflectivity thresholds are 0.6 dBZ and 15 dBZ, threat score (TS) and bias of SDHR are 0.14 and 1.14 in Beijing from April to September during 2019-2021, showing relatively better forecasts. Therefore, the probabilistic and composite reflectivity thresholds corresponding to the optimal TS and bias for 2019-2021 are taken as the forecast probability and eliminating false thresholds of SDHR, and 0-12 h hourly forecast products of SDHR four times a day are tested from April to September of 2022. Results show that TS and bias of SDHR are 0.104 and 1.341, respectively, indicating that the probability prediction products are better than that of CMA-BJ. SDHR products achieve greater improvement, and balance hit rate and false alarm rate well in the piedmont and plain areas with high SDHR frequency. But performances in mountainous areas are not as good as that in plain areas, which may be related to less stations in the mountainous areas of the forecast model. In addition, the result based on case analysis show the predicted area of the products is relatively larger, but high probability area has a good indication for SDHR in Beijing.