A coupled hydrological- hydraulic and data-driven modeling framework for flood modelling in ungauged urban catchments

Flooding is among the most recurrent natural hazards globally, with its impacts intensifying in urban areas due to rapid urban expansion, land use transformation, and the increasing occurrence of high intensity rainfall events. Flood modeling in ungauged urban catchments remains particularly challenging because of limited hydrological observations and the dominant role of impervious surfaces on runoff generation. This study presents a coupled hydrological- hydraulic and data-driven modeling framework to simulate flood inundation in an ungauged urban region of Hyderabad, India with a specific focus on Zone 5 of the Greater Hyderabad Municipal Corporation. Two sets of modelling scenarios were employed. In the first scenario, flood inundation mapping was simulated coupling HEC-HMS and HEC-RAS. For the hydrograph generation, SCS Curve number method was used with rainfall of finer temporal resolution and ward wise land use land cover data. These hydrographs were used as boundary conditions in the HEC-RAS model. In the second scenario, an Artificial Neural Network (ANN) model was developed using rainfall intensity and other meteorological variables along with lagged simulated discharges from the HEC-HMS model. The ANN-predicted discharges were then coupled with HEC-RAS to generate inundation depths. For validation with ground truth data, both the scenarios were validated using geotagged crowd sourced flood images. The second modelling scenario integrating data driven and hydraulic-hydrologic modelling, performed better than the conventional HEC-HMS HEC-RAS approach, as it showed closer agreement  with inundated flood depth. Overall, the findings demonstrate that coupling data driven techniques with hydrologic and hydraulic models significantly enhances urban flood simulation capabilities in data scarce environments.