Classified Early-warning and Nowcasting of Hail Weather Based on Radar Products and Random Forest Algorithm

Abstract

Hail is a kind of severe convective weather with high possibility to cause serious disasters, but it is hard to be early-warned and nowcasted for meteorological operations. This study classified and nowcasted the hail disaster weather and its accompanying severe convective weather based on random forest (RF) algorithm and C-band radar products. The RF model was generated and processed using the observations of severe convective weather in 2008-2017 as the training set firstly. Evaluation results of RF model show a low generalization error of 4.2%. We then applied the model on hail weather observations in 2018-2019, and found that the false identical percentage of no severe convective weather is 15.6%, and the mean false identical percentage of four types of hail weather (the hail, hail with strong wind, hail with short-time heavy precipitation, hail with strong wind and short-time heavy precipitation) is 26.7%. For the nowcast of hail weather in 15 min to 1 h using the RF model and the storm identification and tracking products (SCIT), the mean probability of detection (POD) of four types of hail weather is 74.8%, the mean critical success index (CSI) is 60.8% and the mean false alarm ratio (FAR) is 24.4%. Therefore, the RF model can well classify and nowcast the hail disaster weather and its accompanying severe convective weather, and is appropriate to be applied in the automatic weather forecast operations.