Landslide susceptibility assessment using logistic regression and its comparison with a rock mass classification system, along a road section in the northern Himalayas (India)
Landslide susceptibility assessment using logistic regression and its comparison with a rock mass classification system, along a road section in the northern Himalayas (India)
复制标题
DOI：
10.1016/j.geomorph.2009.09.023
复制
发表时间：
2010-02-01
期刊：
影响因子：
3.9
通讯作者：
Hack, Robert
中科院分区：
文献类型：
--
作者：
Das, Iswar;Sahoo, Sashikant;Hack, Robert
滑坡研究通常由地面知识和岩石强度和边坡破坏标准的实地测量指导。然而，随着基于GIS的统计方法的日益成熟，滑坡易感性研究受益于从不同来源和方法收集的数据在不同尺度上的整合。本研究提出了一种逻辑回归方法的滑坡敏感性映射和验证的结果进行比较，以岩土工程为基础的边坡稳定性概率分类（SSPC）的方法。该研究在印度喜马拉雅山北方的一个易发生滑坡的国家公路路段进行。通过受试者操作特征（ROC）曲线评估Logistic回归模型性能，显示曲线下面积等于0.83。SSPC结果的现场验证表明，高和非常高的敏感性类与目前的滑坡发生率之间的对应关系为72%。两个磁化率图的空间比较揭示了基于岩土工程的SSPC方法的意义，因为90%的区域被逻辑回归方法归类为高和非常高的敏感区，对应于SSPC方法中的高和非常高的类。另一方面，只有34%的地区被SSPC方法归类为高和非常高的福尔斯落在逻辑回归方法的高和非常高的类。逻辑回归法的低估可归因于统计方法的概括，因此，存在于临界平衡条件下的一些斜坡可能不会被归类为高或非常高的敏感区。(C)2009爱思唯尔有限公司版权所有。
Landslide studies are commonly guided by ground knowledge and field measurements of rock strength and slope failure criteria. With increasing sophistication of GIS-based statistical methods, however, landslide susceptibility studies benefit from the integration of data collected from various sources and methods at different scales. This study presents a logistic regression method for landslide susceptibility mapping and verifies the result by comparing it with the geotechnical-based slope stability probability classification (SSPC) methodology. The study was carried out in a landslide-prone national highway road section in the northern Himalayas, India. Logistic regression model performance was assessed by the receiver operator characteristics (ROC) curve, showing an area under the curve equal to 0.83. Field validation of the SSPC results showed a correspondence of 72% between the high and very high susceptibility classes with present landslide occurrences. A spatial comparison of the two susceptibility maps revealed the significance of the geotechnical-based SSPC method as 90% of the area classified as high and very high susceptible zones by the logistic regression method corresponds to the high and very high class in the SSPC method. On the other hand, only 34% of the area classified as high and very high by the SSPC method falls in the high and very high classes of the logistic regression method. The underestimation by the logistic regression method can be attributed to the generalisation made by the statistical methods, so that a number of slopes existing in critical equilibrium condition might not be classified as high or very high susceptible zones. (C) 2009 Elsevier B.V. All rights reserved.