Study on the optimal neighborhood area to generate probabilistic prediction of heavy rainfall based on deterministic convection-allowing model

The ARPS3DVAR+WRF (Advanced Regional Prediction and 3-dimensional variational System)rapid assimilation model is used to simulate several heavy rainfall events in Sichuan and Chongqing areas occurred in recent years. Focusing on the strongest precipitation within 12 h. Neighborhood approach is adopted to the SSRAFS (Storm-Scale Rapid Assimilation and Forecast System)products to perform Neighborhood Mean(NM) forecast, Station Probability(SP) forecast and Neighborhood Probability(NP) forecast in the ranges of different upscale radius. Then the characteristics and effects are respectively analyzed, and the effect of increasing upscale window area to the precipitation forecast is particularly discussed. Finally, the optimum radius of the operational forecast is found by combining traditional and spatial verification results. The results show that the performance of the NM forecast is not stable in light rain and downpour. The improvementof the moderate rain is not obvious, However, it has a good effect on the prediction of heavy rainfall. The singlestation probability may be misleading, but NP forecast could serve as a remedy, by giving better classification information on the uncertainty of heavy rainfall prediction, and provide better reference to improve the capability of short-term operational forecast. FSS and AROC verification results based on NP prediction has a better consistency guidance than TS scores of NM prediction. It reveals that the size of 36 km upscale could eliminate the uncertainty of heavy precipitation to a certain extent while retaining the characteristics of convective feature, which should be selected as the optimal window region.