Logistic回归联合ROC曲线模型在雷电潜势预报中的应用

利用贵阳站2020年1-10月逐日探空资料和闪电监测资料，逐一选取72个物理量参数纳入单因素逻辑回归，筛选具有显著性统计学意义（P＜0.001）的指标进入多因素逻辑回归模型，选取满足检验条件P＜0.05的参数得到雷电潜势预报模型，通过logistic回归联合ROC曲线模型开展雷电潜势预报研究。结果表明：①多因素逻辑回归模型预警效果优于单因素模型，预警准确度从75.4%提高到79.5%；②联合ROC曲线确定预报模型的概率阈值为0.611，雷电潜势预报的命中率POD为84.01%％，虚假警报率FAR为26.05%，临界成功指数CSI为69.38%，准确率较高，雷电潜势预报具有较好的预报能力；③Logistic回归模型联合ROC曲线法在气象预测预报，特别是非线性预测中有一定的应用价值。

Application of Logistic Regression Combined with ROC Curve Model in Lightning Potential Prediction

Selects 72 physical parameters one by one into the single factor Logistic regression model based on the daily radiosonde data and lightning monitoring data of Guiyang station from January to October 2020, selects the indexes with significant statistical significance (P < 0.001) into the multi factor Logistic regression model, and selects the parameters that meet the test condition P < 0.05 to obtain the lightning potential forecast model, Logistic regression combined with ROC curve model is used to study the lightning potential prediction. The results show that: (1) the warning effect of multi factor Logistic regression model is better than that of single factor model, and the warning accuracy is improved from 75.4% to 79.5%; (2) the probability threshold of prediction model determined by joint ROC curve is 0.611, the hit rate pod of lightning potential prediction is 84.01%%, the false alarm rate far is 26.05%, and the critical success index CSI is 69.38% (3) logistic regression model combined with ROC curve method has a certain application value in meteorological forecasting, especially in nonlinear forecasting.