A Parsimonious Semi-Distributed Framework for Event-Based Runoff Modeling

Event-based hydrologic modeling is typically governed by a fundamental trade-off: lumped models are straightforward to implement but neglect spatial variability, whereas fully distributed models require extensive parameterization, limiting their applicability. This study proposes a semi-distributed modeling framework coupled with data-driven parameter estimation, requiring minimal calibration. The studied basin is divided in sub-catchments, within which runoff generation is modeled using the Soil Conservation Service (SCS) Curve Number (CN) method.  Basin-specific CN relationships are developed for November–April and May–October, and used to rescale subbasin CNII values, preserving spatial heterogeneity. The effective precipitation is transformed to direct-runoff using the SCS Unit Hydrograph. This approach avoids over-parameterization while maintaining spatial detail and consistent performance at ungauged locations. In a case study over the Housatonic River Basin, the model reproduces observed storm peak discharges without calibration and performs consistently across gauges. Systematic and random error components, as well as CN uncertainty, are quantified to assess their effects on the simulated peak discharges. The findings show that the proposed modeling framework is well-suited for basin-scale applications, including integration into infrastructure risk assessment models.