耦合模型视角下的湘西州土质滑坡易发性探讨

(1.西北师范大学 地理与环境科学学院, 兰州 730070; 2.中国建筑材料工业地质勘查中心广西总队, 广西 桂林 541002; 3.中国地质大学(北京)地球科学与资源学院, 北京 100083)

滑坡; 易发性; 信息量; BP神经网络; 湘西州

Analysis on Susceptibility Assessment of Soil Landslide in Xiangxi Prefecture from the Perspective of Coupling Model
SHI Ziyue1, ZHU Haiyan2, WANG Jingjing3, XIN Cunlin1

(1.College of Geography and Environmental Science, Northwest Normal University, Lanzhou 730070, China; 2.Guangxi Branch of China National Geological Exploration Center of Building Materials Industry, Guilin, Guangxi 541002, China; 3.School of Earth Sciences and Resources, China University of Geosciences, Beijing 100083, China)

landslide; susceptibility; information method; back propagation neural network; Xiangxi Prefecture

备注

在我国的滑坡灾害研究中,确定滑坡的易发性区域是重要的科学问题,也是进行灾害预警与管理的基础。以武陵山区滑坡灾害频发的湘西州为研究区,首先借助信息量法与BP神经网络建立土质滑坡易发性模型,通过量化9个影响因子对土质滑坡的贡献率,探讨土质滑坡易发性的主导因素和空间分布特征。结果表明:岩石坚硬程度、地形起伏度、年降雨量是土质滑坡的主控因子,地貌类型、坡度、土壤侵蚀强度对土质滑坡的发育产生较大影响; 超过90%的灾害点位于中、高易发区,主要分布于研究区的西北部、东南部; 灾害发生比率随易发性等级的升高而增加,与灾害等级的划分原则相符、与实际情况吻合; ROC检验曲线呈明显的“凸”型,AUC值为0.76。基于此,信息量法与BP神经网络的耦合模型可为相关部门制定有效的减灾措施提供科学参考。
The identification of landslide susceptibility spatial distribution is crucial for landslide warning and its management in China. We first focused on frequent landslide-prone of Xiangxi Prefecture residing in Wuling Mountains by applying the information method and back propagation neural network. The contributions of 9 impact factors were quantified, and the major factors and spatial distribution characteristics were studied. The results show that: hardness degree of rock, relief and annual rainfall are most closely correlated with soil landslide, whereas geomorphologic type, slope and soil erosion intensity play the minor role in soil landslide; more than 90% of the hazard sites fall into the medium and high susceptibility regions that are located in northwest and southeast of Xiangxi Prefecture; the occurrence ratio of hazards increases with the ascent of susceptibility grade, which is consistent with the principle of hazard classification and the actual situation. The further model evaluation of results had shown an accuracy with AUC value of 0.76, indicating that the information method and back propagation neural network were appropriate approaches to evaluate soil landslide susceptibility. This study demonstrates that further research on identifying and targeting landslide susceptibility assessment will be needed to improve geo-hazard risk management and thus to alleviate disaster.