PREDICTING FLOOD RISK WITH MACHINE LEARNING AND GIS: INSIGHTS FROM THE SAKTOLA RIVER, ASSAM, INDIA

Abstract

In Assam, India, flooding is among the most frequent and devastating natural disasters, bringing about significant social and economic disturbances in the Brahmaputra Valley every year. Though floods occur quite frequently, a systematic assessment of flood susceptibility has not been extensively carried out in many secondary river basins, including the Saktola River Basin. This paper discusses the basin's first spatially defined flood susceptibility assessment using the Maximum Entropy (MaxEnt) machine learning method to facilitate informed management of flood risks. Field validation were combined with seven environmental variables that condition floods in order to model patterns of flood susceptibility.  The produced susceptibility map classifies the basin into zones with low, moderate, and high susceptibility to floods; high-risk areas mostly lie along river channels and near floodplains. Validation of models through Area Under the Receiver Operating Characteristic Curve (ROC) reflects high predictive accuracy, which will confirm the MaxEnt algorithm's efficiency in flood susceptibility modeling. The study revealed the applicability of machine learning-based flood susceptibility assessment for data-scarce river basins in Northeast India and provided valuable insights for management of hazard like flood and associated problems in the state of Assam.