Flood Risk Assessment Using Machine Learning Approach in Mohana-Khutiya River of Nepal

Abstract

Nepal, known for its challenging topography and fragile geology is confronted with the constant threat of floods leading to substantial socio-economic losses annually. However, the country's efforts in planning and managing flood risks remain insufficient, especially in the vulnerable Mohana-Khutiya river. Therefore, this study focused on the Mohana-Khutiya river and utilizes the MaxEnt model to comprehensively map flood risk and fill crucial gaps in flood risk assessments. This study employed a combination of 10 geospatial environmental layers and field-based past flood inventory to implement the Maximum Entropy (MaxEnt) machine learning model for flood risk modeling. The available past flood data were divided into two sets, with 75% allocated for model construction and the remaining 25% for model validation. This study clearly demonstrated that the proximity of the river had a significant impact (33.1%) on the occurrence of the flood. Surprisingly, amount of annual precipitation throughout the year exhibited no detectable contribution to the flood event in the study site. About 4.9% area came under high flood risk zone followed by 12.75 % in moderate zone and 82.34% in low risk zone.The model exhibited excellent performance with an Area Under Curve (AUC) value of 0.935 and a low standard deviation of 0.018, indicating accurate predictions and consistent precision. These results highlight the model's reliability and its potential for practical use in flood risk assessment. Future research should refine the MaxEnt model by including more variables, validating against observed flood events, and exploring integration with other flood modeling approaches.

Keywords: Flood risk modeling, MaxEnt model, Environmental variables, AUC, LULC