基于多普勒天气雷达体扫资料的下击暴流预警方法研究  

Damaging downbursts warning algorithm using the Doppler weather radar scanning data

作　　者：孙京[1] 肖艳姣[1] 冷亮[1] SUN Jing;XIAO Yanjiao;LENG Liang(Hubei Key Laboratory for Heavy Rain Monitoring and Warning Research,Institute of Heavy Rain,China Meteorological Administration,Wuhan 430205,China)

机构地区：[1]中国气象局武汉暴雨研究所暴雨监测预警湖北省重点实验室

出　　处：《自然灾害学报》2019年第2期118-126,共9页Journal of Natural Disasters

基　　金：政府间国际科技创新合作重点专项(2016YFE0109400);国家自然科学基金项目(41275008);湖北省气象局年轻科技人员专项(2016Q03)~~

摘　　要：本文利用新一代多普勒天气雷达体扫数据、自动气象探空站和地面大风测站资料,对2009-2013年湖北省大风天气过程的风暴特征量进行相关统计分析,通过云模型和支持向量机(SVM)等方法确定了包含环境、反射率和径向速度特征的9个下击暴流雷达预警指标。基于已确定的预警指标,分别利用Bayes和BP神经网络两种方法建立了下击暴流预报模型,通过识别结果检验表明,两种算法均能有效区分大风与非大风。Bayes方法大风击中率(POD)可以达到81.8%,大风和非大风预报准确率为86.7%,虚警率(FAR)和失误率(FOM)分别为5.2%和18.1%,TS评分0.77;BP神经网络非线性预报方法对大风的识别准确率为84%。进一步证明了提出的下击暴流雷达指标的可预报性和实用性。Based on a Doppler radar volume data,meteorological sounding and wind data,the characteristic quantity of downbursts from 2009-2013 in Hubei province are statistically analyzed,and the nine main radar warning indices of downbursts are given by the cloud model and support vector machine,which includes the characteristics of environment,radar reflectivity and radar-velocity.On the basis,the downburst prediction models are established by the method of Bayes and BP neural network.The identification results show that the methods can differentiate the downburst and non-downburst,and the POD of Bayes is 81.8%,the forecast accuracy of downburst and non-downburst is 86.7%,and the FAR and FOM are 5.2%and 18.1%,respectively.The TS score of model is 0.77;the forecast accuracy of downburst based on the nonlinear prediction of BP neural network is 84%,which further proves the predictability and practicability of the downburst warning indices.

关 键 词：多普勒天气雷达 下击暴流 预警指标 云模型 BP神经网络 

分 类 号：P456.9[天文地球—大气科学及气象学] X43[环境科学与工程—灾害防治]