Catchment-dependent runoff-partitioning uncertainty in the near-optimal parameter space: Insights from 62 catchments in Southeastern China

Study region This study was conducted in 62 small- to medium-sized catchments in southeastern China. Study focus Parameter equifinality can be especially problematic when hydrological models are used not only to reproduce streamflow, but also to infer internal runoff-generation processes. We used the semi-distributed THREW model and 10,000 Latin-hypercube parameter sets per catchment to examine how equifinality affects surface-subsurface runoff partitioning. The contribution of subsurface runoff to total runoff was evaluated within the near-optimal parameter space defined by performance rank and by an absolute Kling-Gupta efficiency threshold, and the resulting uncertainty was compared across catchments and related to a parsimonious set of catchment descriptors. The analysis therefore quantifies parameter-conditioned uncertainty in THREW-estimated runoff partitioning under the fixed model structure, input data, and streamflow-evaluation framework. New hydrological insights for the region Near-optimal parameter sets that reproduce streamflow similarly can imply markedly different surface-subsurface runoff partitioning, and the magnitude of this uncertainty varies systematically among catchments in southeastern China. Among the selected first-order catchment descriptors examined, runoff ratio showed the most consistent association with this variability. Higher-QR catchments exhibited a narrower spread in subsurface runoff contribution among the highest-ranked parameter sets, but broader parameter support and greater partitioning variability under a fixed KGE tolerance. Thus, runoff-partitioning equifinality has a catchment-dependent regional structure, and its apparent magnitude depends on how the near-optimal parameter space is defined. These findings identify runoff ratio as a regional signature of where runoff-partitioning inference from streamflow alone is more or less constrained.