SVM 方法在北京地区夏季晴雨预报中的初步应用

Application of SVM method to summer clear-rain forecast in Beijing region

摘要: 以常规观测降水资料、T639 模式输出产品为基础，利用支持向量机（SVM）方法分别建立了北京地区延庆、密云、房山和观象台4 个站点的夏季晴雨预报模型，并用独立的样本对预报模型进行了业务试运行检验。检验结果表明，经过SVM 方法学习后建立的站点晴雨预报效果优于T639 模式直接输出的晴雨预报效果，08—20 时段和20—08 时段4 站晴雨预报平均准确率分别提高了20.06%和9.45%。SVM 方法对T639 的晴雨预报效果有显著的改进，对北京地区夏季晴雨预报有一定的客观参考价值。

Abstract: Based on precipitation data from conventional observations and the T639 Numerical Weather Prediction (NWP) data, the clear-rainy forecast models are established on Yanqing, Miyun, Fangshan and Guanxiangtai in Beijing by the Support Vector Machine (SVM) statistical method. Independent samples are used to validate the clear-rainy forecast models. The results show that, the clear-rainy forecasted by SVM are superior to that of T639 DMO (direct model output). For the periods from 08 to 20 and 20 to 08, the accuracy of clear-rainy forecast is improved by 20.06% and 9.45%, respectively. The accuracy of clear-rainy forecast is remarkably improved by SVM, which provides an objective way in clear-rainy forecast in Beijing region.