摘要
为了研究贝叶斯（Bayes）方法对单站雷暴预报的应用效能,利用2003年8月~2006年8月WRF模式数值预报产品和单站观测资料,采用朴素贝叶斯分类器（N-Bayes,naive Bayes classifier）和贝叶斯判别准则（D-Bayes,Bayes discriminatory criterion）两种方法,结合多种强对流天气指数场、Fisher准则和相关系数法的预报因子选取技术,分别建立了漳平、广州和湛江3个单站的雷暴预报模型;利用2007年8月资料,检验了模型预报效果,并与Fisher模型的预报效果进行了比较试验。结果表明：N-Bayes和D-Bayes两种模型有较强的雷暴预报能力,24~48 h雷暴预报CSI评分均超过0.23,准确率在72%以上,两者CSI评分接近,趋势相同;两种Bayes分类方法在预报效果上要优于Fisher判别方法。
To test the application effect of Bayesian classification method in thunderstorm forecasting,WRF numerical forecast products and the corresponding observational data of a single station in August from 2003 to 2006 were used,and Naive Bayesian classifier（N-Bayes） and Bayesian discriminance criterion（D-Bayes） were adopted.Combined with predictors choosing technics using deep convectional weather indices,Fisher criterion and correlation coefficient method,thunderstorm forecasting models of the three stations were established in Zhangping,Guangzhou and Zhanjiang.By use of the data in August 2007,the forecasting models were not only experimented and tested but also compared with the results of the Fisher Discriminance criterion method.The results indicate that the critical success indexes（CSI） of the N-Bayes and D-Bayes model for 24 h to 48 h thunderstorm forecasting are more than 0.23,with more than 72% accuracy,that the 6 h interval CSIs of two Bayesian methods within the same forecasting period are close in magnitude and have similar variation;and that both the two models are superior to Fisher discriminance criterion method in forecasting efficiency.
出处
《解放军理工大学学报（自然科学版）》
EI
北大核心
2010年第5期578-584,共7页
Journal of PLA University of Science and Technology(Natural Science Edition)
基金
国家自然科学基金资助项目(40730953
40425009)
