Hydrological non-stationarity is increasingly disrupting rainfall–runoff relationships in tropical agricultural watersheds, elevating flood risk and complicating water management in sub-Saharan Africa, where hydrometric data remain limited. Flood assessment in such basins often remains rainfall-centric, despite growing evidence that internal basin structure regulates streamflow response. This study applies continuous semi-distributed HEC-HMS modelling to diagnose how sub-basin configuration governs flood generation, flow partitioning, and spatial flood exposure in the Diamphwe watershed, central Malawi. Runoff–baseflow partitioning metrics, flow-duration characteristics, and peak-flow diagnostics were integrated into a composite Flood Contribution Index (FCI). These hydrological diagnostics were further linked with riparian geomorphological analysis. A targeted sensitivity experiment evaluated the influence of storage and flow-timing parameters in runoff-dominated sub-basins. Results indicate that runoff-dominated sub-basins disproportionately amplify downstream flood peaks, whereas storage-supported units sustain baseflow and moderate hydrological extremes. Increasing internal storage redistributes seasonal flow and attenuates wet-season peaks but does not increase total water yield. Integration with geomorphological analysis identifies river reaches where upstream hydrological dominance coincides with channel openness, concentrating flood exposure. These findings demonstrate that flood response is governed by internal basin configuration beyond rainfall forcing alone. The study provides a transferable, process-based framework for identifying dominant flood-generating areas and enabling spatially targeted flood management in data-limited tropical agricultural watersheds.