Assessing the impact of the spatial scale of Representative Elementary Watershed (REW) delineation for hydrological modeling

The spatial scale and delineation of Representative Elementary Watersheds (REWs) are fundamental determinants of fidelity in physically based hydrological modeling. The REW framework facilitates a semi-distributed representation of dominant processes—such as surface runoff, subsurface flow, and channel routing—by averaging conservation equations of mass and momentum over discretized sub-units. Leveraging the Tsinghua Hydrological REW model (THREW), this study investigates the sensitivity of hydrological simulations to REW spatial scale across a diverse set of over 80 catchments. By systematically varying drainage area thresholds for REW delineation, we observed distinct scale-dependent behaviors: for larger basins, higher spatial resolution generally enhances model accuracy with relatively low sensitivity to the specific delineation threshold. Conversely, in smaller catchments, excessive discretization often degrades performance and exhibits heightened sensitivity to threshold selection. In the context of daily time-step simulations, we found that for smaller catchments, the detriments of increased data noise and parameter uncertainty often outweigh the marginal gains derived from resolving spatial heterogeneity. In contrast, the explicit characterization of this heterogeneity is critical for optimizing model performance in larger basins.