SVM方法在热带气旋降水预报中的应用

Application of SVM method in Tropical Cyclone Rainfall Forecast

应用国家气象中心下发的热带气旋报文资料(中心经纬度、中心气压、近中心最大风力),设计了诸如热带气旋中心与嘉兴市的最近距离等因子,并任选3个因子组成三维坐标(x,y,z),计算嘉兴市热带气旋降水的概率分布函数。利用SVM回归方法对选中的概率分布函数进行训练,建立嘉兴市热带气旋降水预报方法,并应用于业务试报。同时和多元回归方法进行历史回报和业务试验比较。结果表明,概率分布函数与热带气旋降水具有较好的相关性,SVM回归方法在嘉兴市热带气旋降水预报中的回归预报模型和业务化效果明显优于多元回归。

Using tropical cyclone data （position, pressure and wind） issued by NMC Beijing, the factors such as the distance between the Jiaxing and the tropical cyclone were calculated. Based on any three factors, the distribution functions of the tropical cyclone rainfall were constructed to establish and operationally run the SVM regression models for tropical cyclone rainfall forecast in Jiaxing. Compared to the multivariate regression model, it shows better correlations between the distribution function and the tropical cyclone rainfall and better forecast skills in Jiaxing.