Machine Learning‐Based Landslide Hazard and Risk Assessment in the Darma Valley, NW Himalaya

ABSTRACT

Landslides have emerged as one of the most devastating geological hazards in the Himalayan belt, with a noticeable rise in both frequency and magnitude in the past few decades. It poses a critical threat to human life and public assets, creating an urgent need to assess landslide risk in these vulnerable mountainous areas. This study presents an in‐depth investigation of landslides hazard, vulnerability, and associated risk in Darma Valley, Kumaun Himalaya, India. To assess landslide susceptibility, four machine learning algorithms namely, Random Forest (RF), Multilayer Perceptron (MLP), Naïve Bayes (NB), and Bootstrap Aggregating (BA) were employed. Model performance was estimated by utilising ROC–AUC curve. Among them, the RF model showed the highest predictive accuracy and was used to integrate rainfall and seismic intensity maps to generate rainfall‐induced, seismic‐induced, and combined landslide hazard maps. The vulnerability was assessed using land‐use/land‐cover classes and their associated monetary values. The analysis revealed roads, followed by settlements and dam structures, as the most vulnerable elements due to their high reconstruction costs and exposure to landslides. The risk map is obtained by integrating the combined hazard with the vulnerability. It shows that about 9% of the area lies in high‐to‐very high risk zones, 22% in moderate risk zones, 26% in low risk zones, and 43% in very low risk zones. Societal risk assessment reveals that ~ 58% of the residents are settled in high to very high risk areas. These findings can support sustainable development and safer urban planning in the Himalayas by enhancing decision‐making processes.