基于遥感和社交媒体数据的城市洪涝灾害监测  

Urban Flood Disaster Monitoring Based on Remote Sensing and Social Media Data

作　　者：张会 潘晓宁 贾浩 许德合 ZHANG Hui;PAN Xiaoning;JIA Hao;XU Dehe(College of Surveying and Geo-informatics,North China University of Water Resources and Electric Power,Zhengzhou 450046,China)

机构地区：[1]华北水利水电大学测绘与地理信息学院,河南郑州450046

出　　处：《郑州大学学报（工学版）》2025年第1期82-89,共8页Journal of Zhengzhou University（Engineering Science）

基　　金：国家自然科学基金资助项目(42007423);河南省科技攻关项目(222102320021)。

摘　　要：城市洪涝监测对灾害管理与应急响应至关重要,单一数据源往往存在各自的缺陷,不利于多维度、全方位地分析城市复杂洪涝灾害灾情。结合不受云雨影响且覆盖范围大的GF-3 SAR影像与实时性较强的社交媒体数据,建立“7·20”特大暴雨郑州市中心城区洪涝监测方法,采用阈值分割、随机森林对GF-3 SAR影像进行多方案的灾前灾中水体提取,获取精度最高的水体提取与洪涝监测结果,分析典型区域的水体提取效果;利用Python工具获取城市内涝的社交媒体数据,对其进行处理、可视化与空间分析;结合GF-3 SAR影像与社交媒体数据洪涝监测结果探讨二者优势互补性。结果表明:SAR水体总体提取精度从高到低依次为随机森林、Otsu阈值法、水体指数法,但是在一些典型区域分析中,随机森林的提取效果低于其他方法;基于SAR影像提取的洪涝淹没范围主要集中在三环外的城市边缘地区与大型水体周围,基于社交媒体数据提取的洪涝信息主要集中在城市人口和建筑密集的三环内。Urban flood monitoring is crucial for disaster management and emergency response.Single data sources often have their own shortcomings,which are not conducive to the multi-dimensional and all-round analysis of complex urban flooding disasters.The GF-3 SAR images were combined,which were not affected by clouds and rain and had a large coverage,with real-time social media data to establish a flood monitoring method for the downtown area of Zhengzhou City during the"7·20"heavy rainstorm.Threshold segmentation and random forest were used to extract water bodies from GF-3 SAR images before and during the flood,the most accurate water body extraction results and flood monitoring results were obtained,and the effect of water body extraction in typical areas was analyzed.Using Python tools,the obtained social media data about urban waterlogging were processed,visualized and spatially analyzed.GF-3 SAR images and social media data flood monitoring results were combined to explore the complementary advantages of the two.The results showed that the overall extraction accuracy of SAR water bodies was in the order of random forest,Otsu threshold method,and water body index method.However,in some typical regional analyses,the extraction effect of random forest was lower than that of the other methods.The flood inundation range extracted based on SAR images was mainly concentrated in the urban fringe areas outside the Third Ring Road and around large water bodies.The flood information extracted based on social media data was mainly concentrated in the Third Ring Road,where the urban population was densely populated with buildings.

关 键 词：洪涝灾害 高分三号 阈值分割 随机森林 社交媒体数据 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]