Assessing Landslide Susceptibility in Chamoli, Uttarakhand (India) Through Artificial Neural Networks

Abstract

Chamoli, nestled within the Central Himalayas, presents a complex tapestry of geological, geomorphological, climatic, and hydrological dynamics that shape its vulnerability to natural hazards, particularly landslides. This study delves into the heart of this intricate landscape, combining meticulous methodology with robust results to unravel the nuances of landslide susceptibility. The study is initiated with a comprehensive landslide inventory map, meticulously curated through rigorous field inspections and satellite imagery analysis. This dataset, comprising 726 landslide incidents, forms the cornerstone of subsequent analysis and modeling efforts. An equivalent number of non-landslide points were randomly generated to ensure model integrity, creating a balanced dataset for training and validation. The conditioning factors, 13 in total, including geomorphological, hydrological, geological, and anthropogenic variables, were thoughtfully selected and prepared. Their classification, guided by environmental impact and existing literature, laid the groundwork for the modeling phase. Employing Artificial Neural Networks (ANNs), the model demonstrated exceptional performance. Validation metrics firmly established their reliability and discriminatory power, including sensitivity, specificity, accuracy, and an area under curve (AUC) value of 0.79. The landslide susceptibility map that emerged from this study painted a vivid picture of vulnerability across Chamoli. The northern Himalayan ranges, especially around Joshimath, emerged as high susceptibility zones, influenced by factors like heavy rainfall, fractured lithologies, and human interventions like road development. These findings underscore the urgency of strategic land-use planning and risk mitigation efforts, particularly in high and very high susceptibility areas. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2026.