Machine Learning-assisted Investigation of Landslide Susceptibility for Aglar Watershed in the Lesser Himalaya Region

ABSTRACT Numerous susceptibility modelling and mapping studies have been conducted in the past, aiming to mitigate landslides. In recent years, Machine Learning-assisted predictive modelling has gained tremendous attention for landslide susceptibility analysis. The active tectonic activities, in compliment with the heavy annual monsoonal rainfall make the Himalayan region highly vulnerable and prone to landslides. Detailed studies have been conducted by the pioneer institutes, scientific community and stake holders including Geological Survey of India, etc. However, many landslide-prone areas remain under-explored owing to its vast high hazard zones. This work, pertinently, involves machine learning to identify the landslide susceptible zones in the Aglar watershed of Lesser Himalaya, using the data collected from satellite images and field investigations. Sixteen influential variables were selected based on the region’s topographical, geological, environmental, hydrological/meteorological, and anthropological characteristics. The six robust machine learning approaches have been employed to develop a reliable tool for landslide susceptibility prediction. The study indicated that the random forest model outperformed the other models with 92% Receiver Operating Characteristics- Area Under the Curve (ROC-AUC). The landslide susceptibility map was generated, and the zones were classified into very low (20.3%), low (20.3%), moderately high (20.2%), high (20.0%), and very high (19.1%) which hold promise to cater to the needs of landslide mitigation.