基于决策树算法的江苏省不同区域短时强降水预报研究

Study on short-term heavy precipitation forecast in different regions of Jiangsu based on decision tree algorithm

采用2000—2019年13个地级市气象站地面观测站点观测资料以及ERA5再分析资料,基于机器学习中的经典的C4.5算法对江苏省不同区域是否出现短时强降水建立气象要素预报模型。结果表明:基于C4.5算法的决策树预测模型能够较为直观准确的对江苏省不同区域是否发生短时强降水进行预测,并且该决策树模型具有较高的泛化能力。决策树模型利用各区域总样本的前15 a数据样本进行自学习,学习准确率在淮北地区为89.70%,在江淮之间地区为87.89%,在长江以南地区为87.88%,利用各区域剩余5 a样本对该决策树模型的泛化能力进行测试,测试准确率在淮北地区为85.73%,在江淮之间地区为83.39%,在长江以南地区为93.92%。

Based on the observation data from 13 prefecture-level meteorological stations from 2000 to 2019 and associated ERA5 reanalysis data,Based on the classic C4.5 algorithm in machine learning,established a meteorological element prediction model for short-term heavy precipitation forecast in different regions of Jiangsu.The results show that the decision tree prediction model based on C4.5 algorithm can intuitively and accurately predict the occurrence of short-term heavy precipitation in different regions of Jiangsu,and the decision tree model has a high generalization ability.The self-learning process is conducted based on the first 15 years of the total samples of each region.The learning accuracy is 89.70% in the Huaibei region,87.89% in the region between the Yangtze River and Huaihe River,and 87.88% in the region south of the Yangtze River,respectively.The generalization ability of the decision tree model is tested by using the remaining 5 years samples in each region,and the test accuracy is 85.73% in the Huaibei region,83.39% between the Yangtze River and Huaihe River,and 93.92% in the region south of the Yangtze River,respectively.