Performance evaluation of a 2D Hydro-dynamic Model for estimation of flood Inundation in the lower Semliki River Catchment.

Abstract

Rural Ugandan catchments suffer from limited gauge data and like any other, their model outputs are highly sensitive to choices of computational mesh size and Manning’s roughness coefficient. Among such catchments is the Lower Semliki catchment, prone to frequent, high-impact floods emphasizing the critical need for reliable 2D hydrodynamic models to support risk management. This study evaluates the performance of a coupled hydrologic-hydraulic modelling framework for the Lower Semliki River catchment, with particular emphasis on quantifying key parameter uncertainties on flood‐inundation simulations. A methodology was established first by developing and calibrating a continuous rainfall–runoff model in HEC-HMS, using 30 years of precipitation and evapotranspiration data to generate outflow hydrographs. These hydrographs were used to develop a 2D HEC-RAS model under unsteady simulations, systematically varying Manning’s n and mesh resolutions (30 m, 50 m, 100 m) to isolate their effects. Results indicate that after model development, the hydrodynamic model achieved a Root Mean Square Error (RMSE) within the “very good” performance range for flood depths—and an acceptable value of modified Kling–Gupta Efficiency (KGE′), reflecting moderate overall fit. Despite the low RMSE, the low KGE′ reveals misalignment in flood-peak timing and underestimation of variability across the hydrograph. Sensitivity tests showed that a finer 30 m mesh improved both RMSE and KGE′ without significant computational cost, whereas coarser meshes yielded acceptable magnitude errors yet exacerbated timing and variance discrepancies. Manning’s roughness coefficients were adjusted from their base literature until RMSE fell below a 0.5 threshold, demonstrating the strong sensitivity of model performance to roughness parameterization and producing a working range for the catchment and similar ones. Reducing parameter uncertainty—especially in roughness and mesh resolution—significantly enhances flood simulation fidelity in data-poor basins. The study provides practical guidance on balancing computational demands with accuracy needs, underscoring the importance of thorough calibration even where observational data is scarce.