基于随机森林权重的滑坡危险性评价:以东江流域为例

发布时间：2018-12-24 09:48

【摘要】：对滑坡潜在区域进行危险性评价是滑坡防治管理的一种有效手段。在评价过程中,指标权重关乎评价结果是否合理。基于随机森林(RF)算法提出一种客观赋权方法,选取8个指标变量构建随机森林权重(RFW)的滑坡危险性评价模型,并对东江流域进行评估,结果表明:增加RF分类树的数目可一定程度降低模型的训练和测试误差;RFW可根据RF分类过程产生的Gini指数减少值计算,其中高程(DEM)与坡度(SL)被识别为最重要的两个指标,而土壤可蚀性指数(SEF)和岩石抗剪能力(SRC)被识别为最不重要的2个指标;评估结果表明危险区域主要位于东江流域的东南部、西部及东北部山区,验证精度高达76.81%,分别高于熵权和CRITIC权的44.93%和42.03%;当去掉RFW中最重要和最不重要的2个指标并分别生成危险图,验证精度分别为75.36%和36.23%,进一步验证基于RF生成的权重科学合理。基于RF算法提出的客观赋权方法应用效果良好,可应用于滑坡危险性评价领域,也可为相关领域权重计算提供新思路。
[Abstract]:Risk assessment of potential landslide area is an effective means of landslide prevention and control management. In the evaluation process, the index weight is related to whether the evaluation results are reasonable or not. Based on the stochastic forest (RF) algorithm, an objective weighting method is proposed. Eight index variables are selected to construct the landslide risk assessment model of stochastic forest weight (RFW), and the Dongjiang river basin is evaluated. The results show that increasing the number of RF classification trees can reduce the model training and testing errors to some extent. RFW can be calculated on the basis of the Gini exponent reduction resulting from the RF classification process, in which elevation (DEM) and gradient (SL) are identified as the two most important indicators. Soil erodibility index (SEF) and rock shear resistance (SRC) were identified as the two least important indexes. The results show that the dangerous areas are mainly located in the southeast, west and northeast mountainous areas of Dongjiang River basin, and the accuracy of verification is as high as 76.81, which is higher than that of entropy weight and CRITIC weight by 44.93% and 42.03respectively. When the two most important and least important indexes in RFW are removed and the hazard maps are generated respectively, the verification accuracy is 75.36% and 36.23% respectively, which further verifies that the weights generated by RF are scientific and reasonable. The objective weighting method based on RF algorithm can be applied to landslide risk assessment and can provide a new idea for weight calculation in related fields.
【作者单位】：环境保护部华南环境科学研究所;中山大学地理科学与规划学院水资源与环境研究中心;华南理工大学土木与交通学院;广东省华南地区水安全调控工程技术研究中心;
【基金】：国家自然科学基金(51479216,51579105,91547202) 环保公益性行业科研专项经费项目(201509027)~~
【分类号】：P642.22

【相似文献】