摘要
甘肃庆阳地区董志塬覆盖着数十米厚的层状黄土,水土流失严重,沟壑纵横,滑坡灾害频繁发生,对人民群众生命财产安全和经济社会发展构成严重威胁。通过实际调查和资料收集,获取庆阳地区滑坡的分布数据,构建滑坡的基础数据库;在此基础上,选取坡向、坡形、坡度、高程、地形起伏度、地表粗糙度、距道路的距离、距河流的距离、归一化植被指数(NDVI)、年均降雨量和地层岩性作为滑坡易发性的影响因子,运用信息量模型、信息量-逻辑回归模型、随机森林对庆阳地区黄土滑坡进行易发性评价与区划,分析各影响因子与滑坡地质灾害分布规律的关联性。结果表明:(1)滑坡集中分布于高程1200~1400 m区域(滑坡数占比为47.08%)及年均降雨量540~590 mm区域(39.71%),滑坡极高易发区集中于董志塬地区,受地形起伏度、道路工程与降雨侵蚀协同驱动;(2)信息量-逻辑回归模型(受试者工作特征曲线下面积(AUC)为0.868)与随机森林(AUC值为0.864)预测精度显著优于单一信息量模型(AUC值为0.843),信息量-逻辑回归模型通过非线性校正量化因子权重差异,随机森林依托集成学习优势适应复杂非线性关系;(3)坡形、高程及距道路的距离是滑坡发育的核心驱动因子,凹形坡汇水侵蚀、道路工程扰动(距道路的距离近)与塬边高侵蚀带(高程1200~1400 m区域)交互作用加剧滑坡风险。
The Dongzhi tableland in Qingyang area of Gansu is covered by tens of meters of layered loess,suffering from severe soil erosion,crisscrossed gullies,and frequent landslide disasters,which pose a serious threat to the safety of people's lives and property and the development of economy and society.Through field investigations and data collection,the landslide distribution data in Qingyang area were obtained to construct a basic landslide database;on this basis,slope aspect,slope shape,slope gradient,elevation,topographic relief,surface roughness,distance to roads,distance to rivers,normalized difference vegetation index(NDVI),annual average rainfall,and stratum lithology were selected as influencing factors for landslide susceptibility;information value model,information value-logistic regression model,and random forest were used to evaluate and zone the susceptibility of loess landslides in Qingyang area,and the correlation between each influencing factor and the distribution law of landslide geological disasters was analyzed.The results show that①landslides are concentrated in areas with the elevation of 1200-1400 m(the proportion of landslide number is 47.08%)and areas with annual average rainfall of 540-590 mm(39.71%);the extremely high-susceptibility areas are concentrated in Dongzhi tableland,driven synergistically by topographic relief,road engineering,and rainfall erosion;②the information value-logistic regression model(area under the receiver operating characteristic curve(AUC)is 0.868)and random forest(AUC is 0.864)show significantly better prediction accuracy than the single information value model(AUC is 0.843);information value-logistic regression model quantifies the weight differences of factors through nonlinear correction,while random forest adapts to complex nonlinear relationships relying on the advantages of ensemble learning;③slope shape,elevation,and distance to roads are the core driving factors for landslide development;the interaction of water convergence erosion on concave slopes,road engineering disturbances(short distance to roads),and high-erosion zones at the tableland edges(the areas with the elevation of 1200-1400 m)exacerbates the landslide risks.
作者
薛一凡
马鹏辉
李泽坤
韩宁
陈立森
焦其宪
张永涛
王竟晔
XUE Yi-fan;MA Peng-hui;LI Ze-kun;HAN Ning;CHEN Li-sen;JIAO Qi-xian;ZHAHG Yong-tao;WANG Jing-ye(School of Geological Engineering and Geomatics,Chang'an University,Xi'an 710054,Shaanxi,China;Shaanxi Zhonghong Geotechnical Engineering Co.,Ltd.,Xi'an 710065,Shaanxi,China)
出处
《地球科学与环境学报》
北大核心
2025年第4期600-617,共18页
Journal of Earth Sciences and Environment
基金
国家自然科学基金项目(42477175)。
关键词
滑坡
易发性评价
信息量模型
信息量-逻辑回归模型
随机森林
模型对比
甘肃
landslide
susceptibility evaluation
information value model
information value-logistic regression model
random forest
model comparison
Gansu
