50km以内雷暴系统的分类识别方法研究  

Method Study of Classification and Recognition of Thunderstorm System Less than 50 km

作　　者：王萍[1] 高毅[1] 李聪[1] 

机构地区：[1]天津大学电气与自动化工程学院,天津300072

出　　处：《气象》2016年第2期230-237,共8页Meteorological Monthly

基　　金：天津市科学基金项目(14JCYBJC21800)资助

摘　　要：为解决水平尺度在50 km以内的雷暴系统自动分类识别问题,提出了系统结构疏密性特征、移出率、液态水含量及累加液态水含量特征的构建算法;这些特征既在冰雹和短时强降水之间具有显著性差异,又可以共同描述冰雹和强降水同时发生的具有双重性特质的复合性系统。为配合累加液态水含量这种与时间相关特征的使用,在所实现的分类树中引进了迭代机制。实验表明,本文方法对50 km以内雷暴系统引发的短时强降水击中率达到89.1%,误报率9.5%;对其引发的冰雹的击中率为79.8%,误报率3.5%;平均临界成功指数达到80.0%。To solve the problem of automatic classification and recognition of thunderstorm system with horizontal scale less than 50 km, four features are constructed, including the density feature of thunder- storm system, emigation rate, liquid water content and cumulative liquid water content. There are signifi cant differences in these features between hail and short-time intense precipitation. In addition, these fea- tures can be used to describe complex system that generates hail and short-time heavy precipitation at the same time. In view of the relationship between cumulative liquid water content and time series, the itera- tive mechanism is introduced in the classification tree. Experiments show that the recognition rate of sys- tems generating short-time heavy precipitation is as high as 89. 1 % and the false positive rate is 9. 5%, while the recognition rate of systems generating hail is 79.8% and the false positive rate is 3.5%. The av erage critical success index （CSI） reaches 80.0%.

关 键 词：50 km以内雷暴系统 结构疏密性特征 移出率 液态水含量 决策树 

分 类 号：P457[天文地球—大气科学及气象学]