基于对流参数的雷暴潜势预报方法对比分析

Comparative Analysis of Potential Thunderstorm Forecast Methods Based on the Convective Parameters

以南京站为中心、50km半径范围内的闪电定位资料为依据区分南京的雷暴与非雷暴，将2006～2008年6～8月的552个对流参数样本分别利用逐步选择法、逐步回归法和相关分析法进行因子筛选，将筛选出的与雷暴有密切关系的对流参数作为预报因子分别应用Bayes判别分析、Logistic回归判别和神经网络的雷暴潜势预报方法进行预报。结果表明，Logistic回归判别法对雷暴的预报准确率为最高的78％，Baye判别分析法的误报率为最低的7．6％，综合预报准确率和误报率来看，Logistic回归判别法最适合于南京地区的雷暴潜势预报；最后将筛选出的预报因子做主成分分析，给出预报因子与雷暴的关系的相应解释。

This article distinguished thunderstorms and non-thunderstorms on the basis of the lightning location information within a range of 50 km around Nanjing City and filtrated forecasting parameters during the 552 radio sound observations during June - August from 2006 to 2008 by the means of gradual selection, stepwise regression and correlation analysis. The convective parameters, which were closely related to thun- derstorms, were selected as the forecasting parameters and for the potential prediction of thunderstorms, Bayes discriminant analysis, Logistic regression discriminant and Neural network were deployed to forecast thunderstorms. The results of comparative analysis of the forecasting resuits by the three methods showed that the forecast accuracy of logistic regression discriminant was the highest with 78%, the false alarm rate of Bayes discriminant analysis is the least with 7.6% , logistic regression diseriminant is the most appropriate method to forecast thunderstorms of Nanjing considered both forecast accuracy and false alarm rate. Finally the forecasting parameters were used for principal component analysis so that an explanation on the relationship between the factors and thunderstorms was given.