基于支持向量机的雷暴潜势预报初探
投稿时间: 2012-08-22  最后修改时间: 2012-09-01  点此下载全文
引用本文:
摘要点击次数: 148
全文下载次数: 
作者单位E-mail
周明薇 南京信息工程大学 zhoumw2005@163.com 
肖稳安 南京信息工程大学大气物理学院  
张其林 南京信息工程大学大气物理学院  
彭双姿 湖南省邵阳市气象局  
宋喃喃 南京信息工程大学大气物理学院  
中文摘要:利用2008-2010年夏季邵阳地区NCEP (1°×1°)再分析资料和湖南省闪电定位资料,建立该地区夏季雷暴、非雷暴的支持向量机(SVM)分类预报模型,进行雷暴潜势预报,并用测试样本检验了该模型的预报能力,同时与Logistic回归模型和Bayes判别法的预报效果进行了比较。结果表明:SVM模型的预报准确率为86.89%,虚警率为13.79%,漏报率为8.20%,TSS评分为0.83,在邵阳地区雷暴6h的潜势预报中,SVM方法所建立的模型较其他两种方法具有更好的预报效果。
中文关键词:雷暴预报  支持向量机  Logistic回归  Bayes判别法
 
A Preliminary Study on Thunderstorm Forecast Based on SVM
Abstract:Based on the NCEP FNL Operational Global Analysis data on 1°×1° degree grids in Shaoyang and cloud-to-ground lightning data obtained by lightning detection network in Hunan Province, the predictors are selected from the NCEP data in summer from 2008 to 2010, and the predictand comes from the lightning location data. Based on these data, a SVM classification model of thunderstorm-forecasting in summer is set up. Meanwhile, preliminary study is compared with the forecast effect of the Logistic regression and the Bayesian discrimination. The result shows that the prediction accuracy of this model is 86.89%, false alarm rate is 13.79%, missing rate is 8.20% and the true skill statistic is 0.83. Therefore, compared to the other two methods, SVM is an effective mean to make 6-hour forecasts of thunderstorm in Shaoyang.
keywords:thunderstorm forecast  SVM  Logistic regression  Bayesian discrimination
查看全文  查看/发表评论  下载PDF阅读器