基于岭回归方法的华中地区闪电潜势预报

Experimental study on lightning potential forecast in central China based on the ridge regression method

为研究风云四号卫星的闪电产品在强对流天气预报中的应用,提高闪电潜势预报效果,利用全球预报系统(GFS)分析场数据,计算了华中地区2018年6-7月对流层内多种物理量和闪电数据之间单因素方差检验的F值.结果表明,闪电是否发生与低层大气温度、比湿、最大风速层风速U分量、对流层顶风速U分量、地面抬升指数、最优抬升指数显著相关;闪电活动多集中于低层大气温度和比湿的高值区,同时主要集中于抬升指数、最大风速层风速U分量和对流层顶风速U分量的低值区.由于岭回归方法在处理共线性数据方面效果较好,利用岭回归法得到闪电潜势预报模型,并对华中地区2018年8月共31 d的闪电进行了潜势预报,预报命中率达到0.75,虚假预警率为0.62,预报效果较好,对未来24 h闪电预报具有指导意义.

In order to study the application of FY-4 lightning products in convective weather prediction and improve the capability of lightning potential forecast, the data of global forecast system analysis field were used to calculate the ANOVA F-value between various physical quantities in the layer of troposphere with reference to the lightning data on central China from June to July 2018. The results showed that the occurrence of lightning was significantly related to the lower atmospheric temperature, specific humidity, U-component of wind of the max wind layer, u-component of wind of the tropopause, surface lifted index and best(4 layer) lifted index. The lightning activity was mainly concentrated in the high value area of the lower atmospheric temperature and specific humidity, while it was mainly concentrated in the low value area of the uplift index, U-component of wind of the max wind layer and U-component of wind of the tropopause. As the ridge regression method has a good effect in processing collinear data, it was also used to train the lightning potential prediction model. After 31 days of lightning potential prediction in central China in August 2018, the prediction hit rate was 0.75, the false alarm rate was 0.62, and the prediction result was effective, which has guiding significance for the accuracy of lightning potential prediction in the next 24 hours.