Abstract: Combining the 552 radio sound observations during June and August from 2015 to 2017 with the lightning location information within a range of 50 km around Tangshan, this article applies dynamic clustering method to classify thunderstorm occurance in Tangshan district into three categories （as category 1, 2, 3）. The reliability of the categories is verified with overlay of lightning location information onto radar echo PPI. Ten convective parameters associated with each radio sound observation are filtrated by the means of gradual selection and stepwise regression method. The convective parameters, which are closely related to thunderstorms, are selected as the forecasting parameters and are used for principal component analysis so that an explanation on the relationship between the factors and thunderstorm is given. Finally for the potential prediction of thunderstorms, Bayes discriminant and Logistic regression discriminant are deployed to establish prediction models and Logistic regression discriminant can also provide the probability of the three categories. The results show that the forecast accuracy of the methods mentioned above are better than 70 percent and might be used as a reference in the specified district.

Key words: thunderstorm, convective parameters, dynamic clustering, Bayes discriminant, Logistic regression discriminant, Tangshan city