Understanding flood susceptibility in a monsoon-influenced catchment through interpretable deep learning and hydro-climatic controls

Abstract

Flooding is one of the most common and severe hydro-meteorological hazards, posing a significant risk to water resources, infrastructure, and socio-economic security, especially in monsoon-prone river catchments. Accurate identification of flood-prone areas is critical for effective water risk management and flood-prone catchment planning. This study proposes a data-driven framework for flood susceptibility mapping in the Baitarani River Basin, eastern India, by incorporating hydro-geomorphic, hydro-climatic, and hydro-soil factors, which are critical for understanding the complex flood mechanisms in the catchment. The proposed framework incorporates a novel feature selection method, the μ-Relief algorithm, which selects critical flood predictors and removes redundancy in high-dimensional flood datasets. A novel interpretable deep-learning model specifically designed for tabular data has been developed and improved by evolutionary optimisation for improved performance and reliability. The performance of the proposed framework was evaluated using multiple performance metrics, including AUROC, F1-score, Jaccard index, and Hamming loss, in multiple validation experiments to improve the reliability and robustness of the proposed framework. The proposed framework achieved promising predictive performance, with an AUROC of 0.93 and an F1-score of 0.888, outperforming the baseline models. Spatial analysis revealed that 29% of the basin is classified as moderately to highly flood-susceptible, predominantly occurring in very-low lying floodplains, river confluences, and rapidly urbanising areas. The interpretability analysis has shown that elevation and rainfall are the dominant flood predictors. In contrast, wetness, land use, and soil factors also contribute significantly to flood susceptibility, which is critical for understanding flood mechanisms in the Baitarani River Basin. The proposed framework has demonstrated its applicability in other catchments and provides a tool to enhance flood resilience in flood-prone catchments in response to changing hydro-climatic and hydro-geomorphic conditions.