Landslide susceptibility modelling in a part of Himachal Pradesh, India: An integrated method based on machine learning and geospatial techniques

Abstract

Landslides are one of the most destructive natural hazards in the mountainous regions across the globe including the western Himalayas of India. Hence, it is essential to implement mitigation plans, evacuation measures, and an infrastructure plan based on precise, efficient landslide susceptibility models. Current methods of landslide susceptibility mapping are improving constantly, using geospatial techniques to incorporate visual representation of the environment. However, these current methods are often opinion driven, due to lack of consensus on which factors take precedence over others. This study aims to provide a different approach namely a machine learning based approach towards Landslide Susceptibility Mapping, integrating GIS to give an accurate visual representation of the surrounding areas ranked by order of susceptibility in/and around Kullu Valley of western Himalaya, India. The landslide conditioning factors used in the study involve both static and dynamic data such as slope, land use, land cover, and rainfall variables. The research found that although the Extremely Randomised Trees provide a considerably more accurate assessment of the study area’s vulnerability, the Random Forest Regressor has greater overall accuracy. There is a significant relationship between the model’s outputs and past landslides. According to the study, there would be significantly more regions with high susceptibility to the effects of climate change on landslides by 2030. The application can identify the geographical distribution of landscape risk and is significantly less time-consuming than current methods of susceptibility analysis. Machine learning models could be crucial in evacuation efforts and in preventing damage to life and property.