基于不同机器学习算法的地震滑坡易发性评价——以鲁甸地震为例  

Evaluation of the Susceptibility of Earthquake Landslides Based on Different Machine Learning Algorithms——Taking Ludian Earthquake as an Example

作　　者：吉日伍呷 田宏岭[1,3] 韩继冲 JIRI Wuga;TIAN Hongling;HAN Jichong(Institute of Mountain Hazards and Environment,Chinese Academy of Sciences,Chengdu 610041,China;University of Chinese Academy of Sciences,Beijing 100049,China;Key Laboratory of Mountain Hazards and Surface Process,Chinese Academy of Sciences and Ministry of Water Conservancy,Chengdu 610041,China;Faculty of Geographical Science,Beijing Normal University,Beijing 100875,China)

机构地区：[1]中国科学院、水利部成都山地灾害与环境研究所,四川成都610041 [2]中国科学院大学,北京100049 [3]中国科学院地表过程与山地灾害重点实验室,四川成都610041 [4]北京师范大学地理科学学部,北京100875

出　　处：《昆明理工大学学报（自然科学版）》2022年第2期47-56,共10页Journal of Kunming University of Science and Technology(Natural Science)

基　　金：国家重点研发计划项目(2018YFE0100100)。

摘　　要：地震发生后,及时准确地评价灾区的滑坡易发性对应急救援,灾后重建具有重要意义.然而目前在地震滑坡易发性评价中对不同机器学习模型性能对比的研究较为缺乏.本文从地震动参数、植被、断层岩性、水文和地形方面构建了地震滑坡易发性评价指标体系.然后选取了逻辑回归、K近邻、朴素贝叶斯和随机森林算法,以鲁甸地震为研究案例分别构建了4种滑坡易发性评价模型,并对模型的预测精度进行了对比.结果显示:随机森林模型在测试数据集上的灵敏度(0.94)、精确度(0.94)和准确度(0.94)均高于另外3种机器学习模型,且该模型生成的滑坡易发性的空间分布与实际的地震滑坡分布较为一致,朴素贝叶斯模型的预测精度相对较差.此外,因子的重要性分析结果表明距河流距离、修正的麦加利地震烈度、距断层距离和坡度是影响滑坡易发性相对重要的评价指标.Accurate evaluation of landslide susceptibility caused by earthquake is of great significance for emergency rescue,but there is a lack of research on the performance comparison of different machine learning models in the evaluation of earthquake landslide susceptibility.The paper constructs an evaluation index system of earthquake triggered landslide susceptibility from the aspects of Modified Mercalli Intensity Scale,vegetation,fault lithology,hydrology and topography.Then,logistic regression model,K-nearest neighbors model,naive Bayes model and random forest algorithms model were selected,and four types of landslide susceptibility evaluation models were constructed using the Ludian earthquake as a research case,and the prediction accuracy of the models was compared.The results show that the sensitivity(0.94),precision(0.94),and accuracy(0.94) of the random forest model on the test data set are higher than the other three machine learning models,and the landslide susceptibility space distribution generated by this model is rather consistent with the actual earthquake landslide distribution,the prediction accuracy of the naive Bayes model is relatively poor.In addition,the analysis shows that the distance from the river,the corrected Mercury earthquake intensity,the distance from the fault and the slope are relatively important evaluation indicators that affect the susceptibility of landslides.

关 键 词：滑坡 地震 易发性评价 机器学习 鲁甸地震 

分 类 号：P642.22[天文地球—工程地质学] P315.9[天文地球—地质矿产勘探]