基于Logistic回归模型的风云二号(G星)降雹识别技术研究  

作　　者：彭宇翔 刘涛 文继芬[1] 李皓 唐辟如 李怀志[1] 

机构地区：[1]贵州省人工影响天气办公室,贵阳550081

出　　处：《科技创新与应用》2024年第7期115-118,共4页Technology Innovation and Application

基　　金：贵州省科技计划项目(课题)(黔科合基础-ZK[2021]一般217);贵州省气象局科研业务项目(黔气科登[2021]08-03号);中国气象局人工影响天气中心业务项目。

摘　　要：以风云二号(G星)的7项卫星反演产品(云顶温度、云顶高度、液水路径、过冷层厚度、光学厚度、有效粒子半径以及黑体亮温)作为模型的输入参数建立Logistic回归冰雹识别模型,开展贵州降雹识别技术研究。收集了2020年3—5月期间11个降雹日共计136组风云二号(G星)卫星反演产品数据,其中包括了68个降雹点以及68个未降雹点数据,将每个降雹点降雹时刻前后15 min内的反演产品作为该点的数据。同时选取相应的未降雹点作为对比。将所建立的冰雹数据集分为训练集与检验集。其中,训练集为随机选取的116组数据,用于训练模型(分别包括58组降雹点与未降雹点数据),模型检验集为剩余20组数据(分别包括10组降雹点与未降雹点数据)。利用训练集完成Logistic回归冰雹识别模型建立,利用检验集验证模型识别效果。结果表明,Logistic回归冰雹识别模型对降雹的识别准确率为85%,其中对检验集中的10个降雹点识别准确率为90%,漏报率为10%;对10个未降雹点识别准确率为80%,空报率为20%。Taking the seven satellite retrieval products of Fengyun 2(G satellite)(cloud top temperature,cloud top height,liquid-water path,supercooled layer thickness,optical thickness,effective particle radius and blackbody brightness temperature) as the input parameters of the model,the Logistic regression hail recognition model was established to carry out the research on hail identification technology in Guizhou.A total of 136 sets of Fengyun 2(G satellite) satellite retrieval product data were collected from 11 hail days from March to May 2020,including 68 hail points and 68 non-hail points.The inversion products within 15minutes before and after the hail time of each hail point are taken as the data of this point.At the same time,the corresponding non-hail points are selected as the comparison.The established hail data set is divided into training set and test set.Among them,the training set is 116 groups of data randomly selected for the training model(including 58 groups of hail point data and non-hail point data respectively),and the model test set is the remaining 20 groups of data(including 10 groups of hail point data and non-hail point data respectively).The training set is used to establish the Logistic regression hail recognition model,and the test set is used to verify the recognition effect of the model.The results show that the recognition accuracy of Logistic regression hail recognition model is 85%,in which the recognition accuracy of 10 hail points in the test set is 90%,the missing report rate is 10%,the recognition accuracy rate of 10 unfallen hail points is 80%,and the false alarm rate is 20%.

关 键 词：LOGISTIC回归 分类 冰雹 识别 检验 

分 类 号：P458.1[天文地球—大气科学及气象学]