基于集合预报极端天气预测指数的浙江分类强对流预报模型

利用2016—2021年ECWMF集合预报资料、浙江自动站实况资料等，计算浙江短时强降水、雷暴大风和冰雹等强对流天气相关物理量的极端天气预报指数（EFI：Extreme Forecast Index），分析EFI分布特征，并构建了分类强对流预报模型。结果表明：强对流天气与物理量的EFI有密切联系，发生短时强降水时，对流有效位能、整层可降水量、850 hPa与500 hPa温差和位温差的EFI较大，而垂直风切变的EFI为负值，因而较小的垂直风切变更有利于出现极端降水；发生雷暴大风和冰雹时，对流有效位能、850 hPa与500 hPa温差和位温差以及850 hPa温度露点差的EFI较大，700 hPa露点温度的EFI为负值，与上层干冷下层暖湿的有利层结条件有关。利用支持向量机多分类方法，将强对流天气相关物理量的EFI作为特征值开展训练，构建的预报模型对于非局地强对流天气有较好的预报效果，其中短时强降水的误判率明显低于雷暴大风。

Research on Classified Severe Convection Weather Forecast in Zhejiang Province Based on Extreme Forecast Index of Ensemble Prediction

In this paper, the Extreme Forecast Index (EFI) related to short time heavy precipitation, thunderstorm gale and hail in Zhejiang is calculated using ECWMF ensemble prediction data and Zhejiang automatic station observation data from 2016 to 2021. The characteristics of EFI are analyzed, and the forecast model is built. Results show that severe convective weather is closely related to the EFI of physical quantities. When short time heavy precipitation occurs, the physical quantities with larger EFI are convective effective potential energy, whole layer precipitable water, temperature difference and potential temperature difference between 850 hPa and 500 hPa. While the EFI of vertical wind shear is negative, indicating that the smaller vertical wind shear is more conducive to the occurrence of extreme precipitation. When thunderstorms and hailstorms occur, the physical quantities with a larger EFI index are convective effective potential energy, temperature difference and potential temperature difference between 850 hPa and 500 hPa, and temperature dew point difference of 850 hPa. EFI of dew point temperature of 700 hPa is negative, related to the favourable stratification condition of the dry and cold upper layer with the warm and wet lower layer. By using the multi classification method of the Support Vector Machine, the EFI of the physical quantities related to the strong convective weather are used as the characteristic value to carry out training. The prediction model is effective for non local severe convective weather, and the misjudgment rate of short term heavy precipitation is obviously lower than that of thunderstorm gale.