安全与环境工程2024，Vol.31Issue(3)：11-22,46,13.DOI:10.13578/j.cnki.issn.1671-1556.20231240
基于社交媒体数据的城市暴雨洪涝灾害风险评估
Urban storm flood disaster risk assessment based on social media data
摘要
Abstract
In recent years,the increasing occurrences of urban flood disasters triggered by heavy rainfall have severe-ly endangered people's lives,health,and property safety.Objective and accurate urban flood disaster risk assessment is crucial for effectively enhancing disaster prevention and reduction capabilities.However,the lack and lag of basic data for urban disaster points restrict the accuracy of urban storm flood disaster risk assessment results.With the development of mobile internet technology,the relevant disaster information posted by the public on social media gradually accumulates into a massive,timely,and thematically clear resource known as social media data.Introdu-cing this resource into urban storm flood disaster risk assessment work undoubtedly holds significant importance in accurately depicting the overall picture of urban storm flood disasters.Taking the"7·20"rainstorm event in Zhengzhou City in 2021 as an example,this study first selected thirteen influencing factors from meteorological fac-tors,basic geographic information,and socio-economic factors.Then,leveraging web crawling technology,it ob-tained information on waterlogging points from Weibo data.Finally,using four machine learning models,namely GBDT,XGB,RF,and AdaB,the study conducted a risk assessment of the rainstorm flood disaster in Zhengzhou"7·20".The results are as follows:① The four sets of indicator weights obtained based on the above models are statistically consistent.Among the influencing factors,road density,vegetation coverage index,maximum rainfall in half an hour,and maximum daily rainfall all rank in the top five in terms of importance in the four sets of indicator importance rankings,indicating that these factors are the main causes of the rainstorm flood disaster;② Based on the Pearson correlation coefficient test,it is found that the correlation between the evaluation results of the four models is relatively high.The areas with extremely high risk are concentrated in the central parts of the five main urban areas of Zhengzhou,the northeast part of Zhongmu City,Micun Town and Chengguan Town in Xinmi City,and the surrounding areas of Gongyi Station in Gongyi City;③ The AUC and ACC values of the four models are all above 0.7,confirming the effectiveness of machine learning models in urban flood risk assessment.Compared with the GBDT,XGB,and RF models,the AdaB model has the highest accuracy,and the sum of the Rei values of the high-risk and extremely high-risk areas obtained by it is the largest,indicating that its evaluation results are consistent with the actual situation.By introducing social media data into urban storm flood disaster risk assessment work,this study effectively enhances the accuracy of the assessment results,providing decision-making basis for risk warning and emergency response to urban flood disasters in Zhengzhou City and similar cities under heavy rainfall events.关键词
城市暴雨洪涝灾害/风险评估/机器学习模型/社交媒体数据/郑州市"7·20"暴雨事件Key words
urban flood disaster/risk assessment/machine learning model/social media data/"7·20"rain-storm event in Zhengzhou City分类
资源环境引用本文复制引用
王德运,张露丹,吴祈..基于社交媒体数据的城市暴雨洪涝灾害风险评估[J].安全与环境工程,2024,31(3):11-22,46,13.基金项目
国家自然科学基金项目(72274186) （72274186）
湖北省自然科学基金项目(2022CFD128) （2022CFD128）
湖北省社会科学基金一般项目(HBSKJJ20233263) （HBSKJJ20233263）
陕西省应急管理研究院科学研究项目(2024SXYY01) （2024SXYY01）
国家社会科学基金重点项目(23AZD072) （23AZD072）
