Hybrid Machine Learning with Hydrological and Hydraulic Models for Runoff Prediction and Flood Risk Assessment in Ghaghara Basin

India's low-lying floodplains and heavy monsoon rains cause frequent flooding in the Ghaghara River basin. Each year, these floods catastrophically damage infrastructure, agriculture, and human life. To effectively mitigate these impacts, understanding flood dynamics through precise and timely assessment techniques is crucial. Therefore, the study combines Machine Learning (ML) with hydrological and hydraulic models to create a strong modeling chain. Using historical hydrological and meteorological data, the LSTM model is trained to reconstruct continuous streamflow in an ungaged basin. The Soil and Water Assessment Tool (SWAT) then utilizes the ML-derived outflows to support model validation. The predicted runoff in SWAT is used in the HEC-RAS model to assess urban flood inundation and depth in the basin. The ML model achieved a good result for both training and testing. Similarly, the SWAT model demonstrated reliable performance, with a validation accuracy of 0.71 and a calibration accuracy of 0.75, making the model's results suitable for further analysis and interpretation. The available water level and flood depth were used to validate the HEC RAS flood result, which demonstrated satisfactory results. The hybrid ML, hydrological, and hydraulic approach effectively identifies vulnerable flood zones in the Ghaghara basin, thereby improving the accuracy of streamflow, runoff, and flood inundation predictions. The framework supports more efficient planning and mitigation efforts, offering a dependable method for flood assessment in areas with limited data.