基于遥感影像大数据、卷积神经网络的福建省有效致灾雷电分布模型及应用  

Study and Application of Effective Disaster-Causing Lightning Distribution in Fujian Province Based on Remote Sensing Image Sensing and Convolutional Neural Network

作　　者：张烨方 刘冰 冯真祯 朱彪[2,3] ZHANG Yefang;LIU Bing;FENG Zhenzhen;ZHU Biao(Fujian Key Laboratory of Severe Weather,Fujian Fuzhou 350001;State Key Laboratory of Severe Weather Chinese Academy of Meteorological Sciences,Beijing 100081;Fujian Meteorological Disaster Prevention Technology Center,Fujian Fuzhou 350001)

机构地区：[1]福建省灾害天气重点实验室,福州350001 [2]中国气象科学研究院灾害天气国家重点实验室,北京100081 [3]福建省气象灾害防御技术中心,福州350001

出　　处：《气象科技》2021年第6期953-959,共7页Meteorological Science and Technology

基　　金：福建省科技厅社会发展引导性(重点)项目(2019Y0063);灾害天气国家重点实验室开放课题(2021LASW-B07);福建省气象局研究型业务专项项目(2020YJ08);福建省气象局基层科技专项(2020J02)共同资助。

摘　　要：为了研究福建省有效致灾雷电的分布情况,基于福建省2004—2012年闪电定位数据及雷击人员伤亡数据、福建省L17级谷歌遥感影像瓦片,引入卷积神经网络模型对遥感影像所在区域是否属于人员活动的属性进行建模、训练和预测,得到福建省人员活动属性的格点产品,结合福建省历史雷电数据对有效致灾的雷电分布情况进行了分析,结果表明:(1)设计的遥感影像+CNN识别模型具有一定的可行性和准确率,通过显著性水平为0.01的假设检验;(2)福建省有63.55%的格点为无人员活动区域;(3)平均有45.36%的闪电落在无人员活动的区域,因地制宜地对其他致灾闪电进行预警是提高应急减灾服务效果的可行途径;(4)有效致灾雷电密度与历史雷击人员伤亡数据的相关性远大于常规雷电密度与历史雷击人员伤亡数据的相关性,有效致灾雷电分布在表征雷电灾害上比常规雷电分布更具有指示意义。In order to study the distribution of effective disaster-causing lightning in Fujian Province, based on the lightning location data and lightning casualty data of Fujian Province in 2004-2012, and the L17-class Google remote sensing image tiles of Fujian Province, the Convolutional Neural Network(CNN) model is introduced to model, train, and predicts for identifying whether the area where the remote sensing image belongs to is unpopulated. We obtained the grid products of the activity attribute of Fujian Province, combining with the historical lightning data of Fujian Province and analyzed the actual distribution of lightning. The results show that:(1) The designed remote sensing image and CNN identification model had certain feasibility and accuracy, passed the hypothesis test with a significance level of 0.01.(2) 63.55% of the grid points in Fujian Province were in unpopulated areas.(3) An average of 45.36% of lightning fell in unpopulated areas, and early warning and prediction of other disaster-affecting lightning was a feasible way to improve the effectiveness of emergency mitigation services according to local conditions.(4) The correlation between the effective lightning density and the historical lightning casualty data was much greater than that of the conventional lightning density and the historical lightning casualty data, and the distribution of effective lightning was more indicative than the regular lightning distribution.

关 键 词：遥感影像 卷积神经网络 有效雷电 应急减灾 雷电灾害 

分 类 号：X915.5[环境科学与工程—安全科学]