基于春季气候信号的“台风是否经过江苏”预测研究

利用中国气象局提供的1951—2018年台风最佳路径数据集,筛选经过江苏的热带气旋个例,发现历史上所有经过江苏的热带气旋都发生在夏秋季节。基于春季130种气候信号指数,利用C4.5决策树算法建立“台风是否经过江苏”的预测模型。结果表明:基于C4.5算法的决策树能够较为直观且准确地对每年“台风是否经过江苏”进行预测。利用1951—2000年(51 a)的数据样本进行训练,学习准确率达到82%,利用2001—2018年(18 a)的数据样本对模型进行泛化能力测试,测试准确率达到83.3%,最后利用2019年第9号热带气旋“利奇马”经过江苏的事实进行个例检验,符合决策树规则C。可以证明C4.5算法对“台风是否经过江苏”的预测具有较高的准确率和通用性。

Study on forecast model of tropical cyclones passing through Jiangsu Province based on spring climate signals

Based on the data set of the best track of typhoon from 1951 to 2018 provided by China Meteorological Administration, the data of tropical cyclones passing through Jiangsu Province are screened. It is found that all the tropical cyclones passing through Jiangsu in history occur in summer and autumn. Based on 130 climatic signal indices in spring, a forecast model of whether tropical cyclones pass through Jiangsu Province is established by using C4.5 decision tree algorithm. The results show that the decision tree based on C4.5 algorithm can be more intuitive and accurate to predict whether a tropical cyclone passes through Jiangsu in a year. The learning accuracy rate reaches 82% by using the data samples from 1951 to 2000 (50 years) for training. The testing accuracy rate reaches 83.3% by using the data samples from 2001 to 2018 (18 years) for testing the generalization ability of the model. Finally, an example is tested based on the fact that there is the known tropical cyclone "Lekima" passing through Jiangsu in 2019. It conforms to the decision tree rule C. It can be proved that the C4.5 algorithm has the high accuracy and robustness in predicting whether there is a tropical cyclone passing through Jiangsu.