A regime-based framework for linking runoff initiation and event-scale sediment export in a small temperate agricultural loess catchment

Runoff initiation is a major uncertainty in event-scale erosion modelling, particularly in small agricultural catchments where sediment export is controlled by threshold-like hydrological responses. This study developed a regime-based framework to analyse how alternative representations of runoff initiation affect runoff volume, peak discharge, and sediment-yield prediction in a small temperate agricultural loess catchment in Wallonia, Belgium. The analysis included 154 rainfall–runoff events monitored between 2015 and 2023, of which 46 had complete suspended-sediment observations. Runoff initiation was represented by the initial abstraction ratio λ of the Soil Conservation Service Curve Number model. Event-specific optimised λ values were derived from observed rainfall–runoff responses and generalised using K-means regimes and continuous regression based on hydro-meteorological descriptors. These formulations were used to estimate runoff volume and peak discharge, which were then supplied as inputs to the Modified Universal Soil Loss Equation. Fixed and literature-based λ formulations performed poorly, with a best runoff-volume Nash–Sutcliffe efficiency (NSE) of 0.13. Five regimes provided a strong diagnostic summary, preserving runoff volume (NSE = 0.94) and peak discharge (NSE = 0.67). Under out-of-sample prediction, regime performance declined to NSE = 0.21 for runoff volume and −0.26 for peak discharge, whereas the continuous formulation achieved NSE = 0.40 and 0.34, respectively. Predictive sediment-yield performance remained poor for both approaches (NSE = −0.03 and 0.01). The framework distinguishes diagnostic structure from predictive performance and clarifies how hydrological uncertainty affects event-scale sediment-export modelling.