不同采样策略下的区域滑坡易发性评价  

Susceptibility assessment of regional landslides under different sampling strategies

作　　者：于海坤 欧阳九发 王丙千 贾雨霏 徐光黎[1] YU Haikun;OUYANG Jiufa;WANG Bingqian;JIA Yufei;XU Guangli(Faculty of Engineering,China University of Geosciences(Wuhan),Wuhan 430074,China)

机构地区：[1]中国地质大学(武汉)工程学院,湖北武汉430074

出　　处：《安全与环境工程》2024年第5期122-134,162,共14页Safety and Environmental Engineering

基　　金：湖北省重点研发计划项目(2021BCA219)。

摘　　要：滑坡易发性评价作为滑坡风险评价的第一步,为地质灾害预警预测提供了数据基础。在易发性评价过程中,样本集的质量和数量是决定滑坡易发性建模准确性的关键因素。为了探索如何获得高质量的滑坡样本集,以湖北省十堰市南部山区为研究区,选取坡度、坡向等与研究区滑坡发育特征相关的评价因子;采用随机采样、滑坡缓冲采样、信息量约束采样、合成少数类过采样技术(synthetic minority oversampling technique,SMOTE)4种采样策略建立训练和测试数据集;分别选取逻辑回归模型和支持向量机模型进行了易发性评价,并通过ROC曲线对4种滑坡样本采样策略的精度进行了对比分析。结果表明:SMOTE采样策略得到了最高的评价精度,在该采样策略下支持向量机模型精度(AUC值为0.9304)优于逻辑回归模型(AUC值为0.8663),得到了更精确的评价结果。通过混合采样技术对区域内滑坡样本进行优化,可为滑坡易发性评价样本的选择提供新的思路。Landslide susceptibility assessment is the first step in risk assessment and provides a data basis for early warning and prediction of geological disasters.In the process of susceptibility evaluation,the quality and quantity of the sample set are the key factors that determine the accuracy of landslide susceptibility modeling.In order to explore how to obtain a high-quality landslide sample set,this paper takes the southern mountainous area of Shiyan City,Hubei Province as the study area,and selects evaluation factors such as slope and aspect that are related to the development characteristics of landslides in the study area.The paper establishes training and test data sets using four sampling strategies,i.e.,random sampling,landslide buffering sampling,information value constrained sampling,and synthetic minority oversampling technique(SMOTE).Then,the paper select logistic regression model and support vector machine model for susceptibility evaluation,respectively,and compares and analyzes the accuracy of the four sampling strategies by using the ROC curves.The results show that the SMOTE sampling strategy has obtained the highest evaluation accuracy.Under this sampling strategy,the accuracy of the support vector machine model(0.9304)is better than that of the logistic regression model(0.8663),and more accurate evaluation results are obtained.This paper optimizes the landslide samples in the area through the mixed sampling technology,which provides a new idea for the selection of landslide susceptibility evaluation samples.

关 键 词：滑坡 易发性评价 SMOTE采样 支持向量机 

分 类 号：X43[环境科学与工程—灾害防治]