基于GIS和逻辑回归分析的山地城市洪涝灾害敏感性评估——以江西省吉安市为例

Susceptibility Assessment of Flood Disaster in Mountain Cities Based on GIS and Logistic Regression Analysis:A Case Study of Ji’an City,Jiangxi Province

近年来山区洪水暴发频繁,严重威胁着山区人民的生命财产安全。洪水灾害敏感性评价是预防和减轻洪灾的重要途径之一,而山区野外调查不足和资料的缺乏,是洪水敏感性制图中的重大挑战。大数据时代用户产生的数据为洪水风险管理提供了新的机遇。以吉安市为研究对象,利用互联网上用户产生的洪水灾害数据,随机选取70%的洪水事件作为训练区,选取高程、坡度、坡向、曲率、降雨量、河流距离、土地利用、归一化植被指数等8个洪水调节因子,采用Logistic回归模型进行洪涝灾害敏感性评价,并使用混淆矩阵、ROC曲线对评价结果进行检验。结果表明:(1)地势低、距水系近、降雨量较大且土地利用类型为建设用地的区域,洪水发生的概率较大;(2)混淆矩阵得出分类总体准确率为80.6%,训练区(70%)的AUC值为0.888,验证区(30%)的AUC值为0.980,AUC值均大于0.8,说明模型评估精度较高;(3)高风险和极高风险区所占比例为28.71%,包含研究区80.99%的洪水事件,说明这些区域洪水分布较为密集,易感性较高。根据2020年6月1日~6月8日的暴雨洪灾情况验证,该评价结果与实际情况相符。研究表明在数据不易获得的山区,利用互联网上获取的用户数据可行,该评价结果可为吉安市土地利用规划和洪灾风险管理提供参考。

In recent years,the frequent outbreaks of mountain floods have seriously threatened peoples’lives and property.Risk analysis such as flooding susceptibility assessment is one of the critical approaches to prevent and mitigate flooding disaster.However,the inadequate field survey and lack of data might become the significant challenges for the mapping of flood susceptibility.In the era of big data,user-generated data provides new opportunities for flood risk management.This paper takes Ji’an City as the focus area,using the flooding disaster data generated by users on the Internet.70%flood events were randomly selected as training sample and eight flood-conditioning factors including elevation,slope,aspect,curvature,rainfall,river distance,land use and normalized vegetation index were chosen to evaluate the flooding disaster by logistic regression model.The confusion matrix and ROC curve were used to verify the evaluation results.The results show that:(1)The area with low terrain,close to water system,large rainfall,and construction land have a higher probability of flood occurrence.(2)According to the confusion matrix,the overall accuracy rate of classification is 80.6%.Verified by ROC curves,the AUC value of the training sample and the validation sample is 0.888 and 0.980 respectively.The AUC values are both greater than 0.8,indicating that the evaluation accuracy of the model is relatively high.(3)The proportion of high-risk and extremely high-risk areas is 28.71%,including 80.99%of the flood events in the study area,which shows these areas are densely distributed and highly susceptible.The evaluation outcomes were consistent with the actual situation based on the verification of the flood events from June 1 to June 8,2020.It can be concluded from the results above that it is feasible to use the data generated by users on the Internet in mountainous areas where the data is not easy to obtain,and the evaluation results can be used to land use planning and flood risk management in Ji’an city.