基于随机森林赋权信息量的区域滑坡易发性评价—以三峡库区秭归至巴东段为例  

Regional Landslide Susceptibility Evaluation Based on Random Forest Weighting Information:a Case Study of Zigui to Badong Section in the Three Gorges Reservoir Area

作　　者：马敏 王江立 陈琦 李景富 MA Min;WANG Jiang-Li;CHEN Qi;LI Jing-Fu(Wuhan Center of China Geological Survey(Geosciences Innovation Center of Central South China),Wuhan 430205,Hubei,China;Guangdong Nonferrous Geological Environment Center,Guangzhou 510000,Guangdong,China)

机构地区：[1]中国地质调查局武汉地质调查中心(中南地质科技创新中心),湖北武汉430205 [2]广东省有色地质环境中心,广东广州510080

出　　处：《华南地质》2024年第4期749-763,共15页South China Geology

基　　金：中国地质调查局项目(DD20230706)。

摘　　要：在滑坡易发性评价研究中,常规的信息量模型通常将不同评价指标的信息量进行简单累加,而没有重视各评价指标间的权重差异,导致易发性分区结果不精确。本文以三峡库区秭归至巴东段为例,提出了一种基于随机森林赋权信息量的滑坡灾害易发性评价方法,以求提高评价精度。首先利用因子特征分析方法确定滑坡评价因子,接着运用随机森林模型确定各评价因子的权重,然后将权重与信息量模型融合,通过将评价因子信息量加权叠加的方法得到更准确的易发性评价结果。通过统计指标和ROC曲线对两个模型进行了分析和评估。结果显示,传统信息量模型和基于随机森林赋权的信息量模型在测试集的ROC曲线下面积(AUC)分别为0.778和0.855,确定随机森林赋权信息量的方法对传统信息量方法起到了优化作用,该方法为滑坡灾害风险评估提供了新的思路。In the study of landslide susceptibility assessment,conventional information models usually simply accumulate the information of different evaluation indicators,without paying attention to the weight differences between each evaluation indicator,resulting in inaccurate susceptibility zoning results.This article takes the Zigui to Badong section of the Three Gorges Reservoir area as an example and proposes a landslide hazard susceptibility evaluation method based on random forest weighting information to improve evaluation accuracy.This method first uses factor feature analysis to determine landslide evaluation factors,and then uses a random forest model to determine the weights of each evaluation factor.Then,the weights are fused with an information model,and a more accurate susceptibility evaluation result is obtained by weighting and superimposing the information of the evaluation factors.Two models were analyzed and evaluated using statistical indicators and ROC curves.The results showed that the area under the ROC curve(AUC)of the traditional information model and the information model based on random forest weighting were 0.778 and 0.855,respectively,in the test set.The method of determining the weighted information of the random forest optimized the traditional information method,providing a new approach for landslide disaster risk assessment.

关 键 词：滑坡 随机森林模型 信息量模型 易发性评价 三峡库区秭归至巴东段 

分 类 号：P642.22[天文地球—工程地质学]