Abstract In mountainous areas, landslides are highly destructive natural and anthropogenic hazards leading to massive-scale damages and losses of properties and lives. The core objective of this study is to understand the geo-environmental drivers and their potentiality on landslide susceptibility (LS) in the Lish-Gish-Chel River basin of Darjeeling Himalayas, India, using a frequency ratio (FR) model integrating geospatial techniques. A total of 510 (100%) landslide polygons have been recognized to develop a landslide inventory (LI) map combining Google Earth and satellite imageries with proper field verification. The LI was split into training 70% (357) and testing 30% (153) for the model. Twenty-five drivers and LI have been considered for LS mapping. Multicollinearity analysis (MA) signified no collinearity issue among the drivers. The FR value and receiver operating characteristic (ROC) curve techniques have been used to authenticate the model. The jack-knife test has been used to understand the relative contribution (RC) of each driver for the FR model. The LS map has been classified into six landslide susceptibility zones (LSZs). The findings showed that the LSZs from very low to very high recorded by the FR values of 0.003, 0.156, 0.433, 0.809, 1.896, and 8.140. The ROC curve denoted the precision of the LS map was 94.80%. The jack-knife test revealed that the lineament density driver had the maximum RC for the FR model. The outcomes of this research will help planners and decision-makers in planning, development, and hazard/disaster management strategies in mountainous environments. Highlights <UnorderedList Mark="Bullet"> <ItemContent> The relationships between LCDs and LI were assessed. </ItemContent> <ItemContent> MA signified that there is no collinearity issue among the LCDs. </ItemContent> <ItemContent> More than 64% areas to total landslide-affected areas exist in high and very high LSZs. </ItemContent> <ItemContent> The ROC curve denoted 94.80% prediction accuracy of the LS map by the FR model. </ItemContent> <ItemContent> The maximum FR values were recorded in high (1.896) and very high (8.140) LSZs. </ItemContent> <ItemContent> The results of jack-knife test denoted that lineament density has the maximum RC for LS mapping. </ItemContent> </UnorderedList> Graphical Abstract

Understanding the geo-environmental drivers and their potentiality to landslide susceptibility in the Lish-Gish-Chel River basin of Darjeeling Himalayas, India

期刊论文
发表日期：

2025年11月29日

查看全文

Biplab Mandal,
Somen Das

摘要

Abstract

In mountainous areas, landslides are highly destructive natural and anthropogenic hazards leading to massive-scale damages and losses of properties and lives. The core objective of this study is to understand the geo-environmental drivers and their potentiality on landslide susceptibility (LS) in the Lish-Gish-Chel River basin of Darjeeling Himalayas, India, using a frequency ratio (FR) model integrating geospatial techniques. A total of 510 (100%) landslide polygons have been recognized to develop a landslide inventory (LI) map combining Google Earth and satellite imageries with proper field verification. The LI was split into training 70% (357) and testing 30% (153) for the model. Twenty-five drivers and LI have been considered for LS mapping. Multicollinearity analysis (MA) signified no collinearity issue among the drivers. The FR value and receiver operating characteristic (ROC) curve techniques have been used to authenticate the model. The jack-knife test has been used to understand the relative contribution (RC) of each driver for the FR model. The LS map has been classified into six landslide susceptibility zones (LSZs). The findings showed that the LSZs from very low to very high recorded by the FR values of 0.003, 0.156, 0.433, 0.809, 1.896, and 8.140. The ROC curve denoted the precision of the LS map was 94.80%. The jack-knife test revealed that the lineament density driver had the maximum RC for the FR model. The outcomes of this research will help planners and decision-makers in planning, development, and hazard/disaster management strategies in mountainous environments.

Highlights

The relationships between LCDs and LI were assessed.

MA signified that there is no collinearity issue among the LCDs.

More than 64% areas to total landslide-affected areas exist in high and very high LSZs.

The ROC curve denoted 94.80% prediction accuracy of the LS map by the FR model.

The maximum FR values were recorded in high (1.896) and very high (8.140) LSZs.

The results of jack-knife test denoted that lineament density has the maximum RC for LS mapping.

Graphical Abstract