Integration of Logistic Regression and Evidential Belief Function for Flood Risk Assessment in the West Bengal Plain, India

Periodic flooding has significantly impacted the livelihoods of communities in the West Bengal Plain, highlighting the need for detailed flood susceptibility maps to guide effective management strategies. This research assessed the effectiveness of multivariate logistic regression, bivariate Dempster–Shafer-based evidential belief function, and their combined approaches for flood risk mapping in the Lower Damodar Basin. The flood characteristics of the area are analysed using Gumbel's extreme value distribution, based on historical hydrological data from 1932 to 2023. A flood inventory map documenting 230 historical flood locations is used. After eliminating multicollinearity, spatial datasets for 12 hydrogeomorphic variables are compiled. The models utilized 190 of these flood sites for training and developing predictive flood susceptibility maps, while the remaining 40 are used to test model accuracy through metrics such as the receiver operating characteristic curve and the modified seed cell area index. Findings indicate that while the frequency of major floods has decreased after the construction of various flood control structures, the occurrence of minor floods persists. Flood vulnerability has increased downstream in the study area and is particularly high in the southeastern part of the basin due to favourable hydrogeomorphic conditions. The evidential belief function model has achieved the highest predictability rate with an area under the curve value of 0.82. The study highlights the necessity of integrating non-structural and ecosystem-based approaches with structural measures for effective flood mitigation and emphasizes the incorporation of real-time data monitoring and climate models into flood risk assessments for better future scenario predictions.