基于maxEnt模型的哈尼梯田核心区滑坡易发性评价

(云南师范大学 旅游与地理科学学院, 昆明 650500)

滑坡; 易发性评价; ROC-AUC; GIS; maxEnt模型

Assessment on Landslide Susceptibility of the Core Area of Hani Race Terraces Heritage Site Maximum Entropy Model
ZHAO Dongmei, JIAO Yuanmei, QIU Yingmei, LIU Chengjing, XU Qiue, ZHANG Juan

(School of Tourism and Geography Sciences, Yunnan Normal University, Kunming 650500, China)

landslide; susceptibility assessment; ROC-AUC; GIS; maxEnt model

备注

有效精确地预测滑坡易发性对滑坡灾害风险管理与土地利用规划具有重要的现实意义。以往研究中,最大熵模型(maximum entropy model,maxEnt)因其能有效避免模型过拟合在生态学领域中运用广泛,但在滑坡易发性研究运用鲜见。以哈尼梯田世界遗产地为例,选取高程、坡度、坡向、地表粗糙度、平面和剖面曲率、地形湿度、曲率、水系、断层、道路、岩性、植被、土地利用、降雨量与居民点15种指标因子,基于GIS栅格分析法,采用maxEnt模型对研究区滑坡进行了易发性研究。应用受试者工作特征曲线下面积(ROC-AUC)对模拟结果进行检验,结果发现ROC-AUC值达到0.895,maxEnt该模型的预测效果较高,适用于滑坡易发性评价研究; 断层和道路、降雨量和居民点密度是影响遗产地滑坡发育的主要因子; 采用滑坡密度指标对分区结果检验发现滑坡密度值由易发性极低、低、中、高和极高依次递增,说明正态分布理论与专家经验法分区结果合理,且极高和高易发区占区域总面积的21.75%,其主要分布于人类居住较为集中区域。研究可为哈尼梯田核心区的遗产可持续发展、灾害监测和防治提供新的途径。

Effectively and accurately modelling landslide susceptibility is of significant and practical significance for landslide disaster risk management and land use planning. In previous studies, the maximum entropy model(maxEnt)has been widely used in the field of ecology due to avoiding model over-fitting, but is seldom used in landslide susceptibility research. We take the core area of Hani Race Terraces Heritage Site of Yuanyang County as the research site, and select 15 environmental variables including the altitude, slope, aspect, surface roughness, plane and section curvature, topographic humidity index, distance from water system, fault, road, average annual rainfall(AMR), residential density and normalized vegetation index and category data such as land use and lithology as indicators of landslide susceptibility. The landslide susceptibility in the heritage area was evaluated by combing with maxEnt and GIS technology. Furthermore, the accuracy of model was validated by the area under the receiver operating characteristic curve(ROC-AUC). The results show the maxEnt model has high accuracy(the ROC-AUC of the training and test data are 0.895 and 0.739, respectively); the main environmental factors affecting the landslide disaster are the distance from the road, AMR, distance from the settlement, and land use; additionally, landslide density analysis indicates that the zone of normal distribution theory and expert experience method are reasonable; the 21.75 percent of the total area is very high and high landslide susceptibility areas, which mainly distributes in the areas with relatively concentrated human settlements. These research results can provide new avenues for disaster monitoring and prevention in the heritage site.