Landslide inventory and susceptibility models considering the landslide typology using deep learning: Himalayas, India

Abstract

Landslide susceptibility modeling is complex as it involves different types of landslides and diverse interests of the end-user. Developing mitigation strategies for the landslides depends on their typology. Therefore, a landslide susceptibility based on different types should be more appealing than a susceptibility model based on a single inventory set. In this research, susceptibility models are generated considering the different types of landslides. Prior to the development of the model, we analyzed landslide inventory for understanding the complexity and scope of alternative landslide susceptibility mapping. We conducted this work by examining a case study of Kalimpong region (Himalayas), characterized by different types of landslides. The landslide inventory was analyzed based on its differential attributes, such as movement types, state of activity, material type, distribution, style, and failure mechanism. From the landslide inventory, debris slides, rockslides, and rockfalls were identified to generate two landslide susceptibility models using deep learning algorithms. The findings showed high accuracy for both models (above 0.90), although the spatial agreement is highly varied among the models.