基于潜势预报和雷达回波特征的雷电预报方法

针对现阶段仅依靠探空资料和潜势预报无法精准地预测雷电高发地区的现状,提出了一种基于潜势预报和雷达回波特征的雷电预报方法。首先采用L1正则化法筛选雷电预报因子,然后对3 h的潜势预报模型进行BP神经网络训练;在此基础上,利用潜势预报结果和雷达回波特征进行二次训练,构建30 min的临近预报模型;最后运用一次典型的雷暴过程进行方法检验,结果表明当概率阈值超过0.4时,临近预报的TS评分明显高于潜势预报。该雷电预报方法在空间精度和时间频率上均有明显提升,对提高雷电预报的准确率具有重要的作用。

Method of Lightning Prediction Based on Potential Prediction and Radar Echo Characteristics

According to the actuality that it is impossible to accurately predict the region of high lightning incidence only by sounding data and potential prediction, a method of lightning forecasting based on potential prediction and radar echo characteristics is proposed. Firstly, L1 regularization method is used to select the lightning prediction factors, and then the potential prediction model of 3 hours ahead is trained by BP neural network. On this basis, the neural network is retrained by using the potential prediction results and radar echo characteristics, and the nowcasting model of 30 minutes ahead is constructed. Finally, a typical thunderstorm process is used to test the method, and the result shows that threat score(TS)of nowcasting model is higher than that of potential prediction model when the probability threshold is more than 0.4. This lightning prediction method has a significant improvement in the spatial precision and time frequency, which plays an important role to improve the accuracy of lightning prediction.