A Data-Driven Approach for Predicting Riverine Flood Severity Index in the Transboundary Brahmaputra River Sub- Basin

 

Floods in large transboundary basins such as the Brahmaputra pose persistent threats to lives, livelihoods, and infrastructure. Flood event database is initially created utilizing gridded routed streamflow simulations. For each event occurrence, we extract a set of flood attributes including peak discharge, flood volume, duration metrices. These hydrologic characteristics are integrated with an inundation component, enabling the Flood Severity Index (FSI) to represent not only the intensity of flooding within the channel but also the amount and duration of inundation across the adjacent floodplain. Utilizing this index, we provide a data-driven machine learning framework to predict RFSI throughout the basin. Predictor variables include key hydro-climatic inputs such as temperature, precipitation, which collectively influence the generation and evolution of flood events. Multiple machine learning models were evaluated using performance metrics including R², RMSE, MAE, and cross-validation, all of which demonstrated strong predictive skill across diverse hydrologic regimes, establishing the proposed data-driven framework as a scalable and computationally efficient tool for forecasting flood severity. This approach offers a strong basis for evaluating future flood scenarios and understanding how climate change may alter flood risk, especially in large transboundary regions with limited observational data.