Hydrogeomorphological approaches to understanding flood hazards and flood-risk management in northwest Rwanda

Flood hazards and flood-risk management in tropical mountainous regions are constrained by scarce or missing hydrometeorological data, a limitation that hampers understanding of the factors governing rainfall-runoff responses. This thesis investigates the role of topography, soil characteristics, and land use/land cover in shaping flood dynamics, using short-term rainfall-streamflow observations and statistical and numerical approaches in two contiguous agricultural catchments of the Mukungwa watershed in northwest Rwanda: Nyamutera (44 km²) and Gaseke (109 km²).
Firstly, cost-effective river monitoring systems were deployed to generate high-resolution rainfall-runoff datasets, enabling detailed analysis of storm-events variability. Event-based assessments revealed contrasting runoff coefficients and peak discharges between the steep Nyamutera and the floodplain-connected Gaseke catchments, reflecting the role of topography, soil properties, land use/land cover, and water storage connectivity in shaping hydrological responses. The findings highlight clear contrasts in the hydrogeomorphic behaviour of the two catchments, which likely reflect differences in topography, soil properties, and storage capacity. Together, these results advance understanding of flood hazards and rainfall-streamflow dynamics in tropical mountainous environments and underscore a shift away from purely empirical approaches toward process-based hydrological interpretation. . These insights contribute to the development and application of hydrological models in similar data-limited mountainous catchments.
Secondarily, the applicability of the Natural Resources Conservation Service- Curve Number method was tested through storm-based simulations using a developed custom Python-based hydrological modelling framework. Event-based calibration and validation reveal contrasting model behavior between the two catchments rather than uniformly poor performance. In Nyamutera, NSE values are generally low during both calibration and validation, indicating limited skill in reproducing peak magnitudes and runoff volumes, even though KGE often remains acceptable and hydrograph timing is reasonably captured. In Gaseke, calibration results are more stable, with better agreement between observed and simulated peaks for many events; however, validation shows increased scatter, highlighting event-specific deviations likely linked to variable floodplain storage and routing dynamics rather than systematic model failure. Peak-flow and runoff volume errors were frequently substantial, highlighting challenges in capturing flood magnitudes and hydrograph dynamics. These results emphasize the importance of locally calibrated, event-based, process-informed models for flood hazard assessment and their potential to support rainfall-driven early warning systems in mountainous tropical catchments.
Thirdly, the “Génie Rural à 4 paramètres Journaliers” (GR4J)-based approach was applied to reconstruct historical streamflows and derive flood early-warning thresholds. Catchment-specific calibration revealed a rapid rainfall response in Nyamutera and dampened peaks in Gaseke, attributable to floodplain buffering. These contrasting behaviours were consistent with expectations and linked to differences in topography, soil properties, land use/land cover, and water storage connectivity. Logistic regression fused with physical criteria identified antecedent rainfall windows of 5–6 days for Nyamutera and 7–11 days for Gaseke as critical for forecasting flood risks.
Overall, this research advances understanding of rainfall-runoff dynamics in tropical mountainous catchments, demonstrates the value of short-term monitoring and tailored modelling frameworks, and delivers practical tools for flood hazard management and early warning in data-limited regions worldwide.