Comparative Study of Machine Learning Vs. BIS Approach for Landslide Hazard Zonation in Kashmir Himalayas, India

Abstract

ABSTRACT Presently slope stability analysis and landslide hazard monitoring are the most challenging tasks in mountainous regions like the Himalayas. These events which were earlier considered as a process of attaining equilibrium in the topographic surface of the earth by nature, with the increase in population and onset of industrial revolution in the past few decades, changed the scenario of this natural phenomenon and transformed it into a disaster. The main reason for this is the utilization of inaccessible terrains for engineering mega projects and urbanization. Today, landslides are considered as the worst natural disasters and have become the objects of mass destruction and thereby new approaches for landslide study are being constantly developed with time to understand this natural phenomenon as much as possible. Witnessing the severity of this disaster in India and globally; this study is aimed to study landslide hazard in the domain of landslide susceptibility mapping and hazard zonation. This approach makes use of different instability causing parameters prevalent in the area to demarcate the region into different probable hazard zones. The resulting maps thus prepared are called Landslide Hazard Zonation (LHZ). For the study, a sub-part of Karewa Basin is selected as an area of study in Anantnag district, Jammu and Kashmir, India. Two different methods, the established Bureau of Indian Standards (BIS) method and the new machine learning based classification method using decision tree classification algorithm, are selected for the study of landslide hazard in the area. The resulting maps from both methods are analyzed and validated using the landslide inventory data. The results from both the methods are also used to perform a comparative analysis to examine which method yields better results. It is concluded that machine learning based methods, provided accurate training dataset can yield better results than the traditional methods which require a makeover to incorporate more data, made available with the advancement of technology.