Comparative analysis of runoff simulations using multiple hydrological models in the Ganjiang River Basin

ABSTRACT Hydrological models are critical for runoff forecasting. However, different models applied to the same basin often vary significantly due to uncertainties in the model structure, parameterization, and input data. To understand the applicability and limitations of different models, this study evaluated the performance of lumped Xin'anjiang (XAJ), semidistributed Soil and Water Assessment Tool (SWAT), and distributed time-variant gain model (DTVGM) at six hydrological stations in the Ganjiang River Basin, China, employing observed station precipitation and CN05.1 precipitation datasets. The results showed higher Nash–Sutcliffe efficiency (NSE) for XAJ (range: 0.56–0.92, mean: 0.82) and SWAT (range: 0.55–0.95, mean: 0.84) in daily and monthly runoff simulations using different datasets compared to DTVGM (range: 0.54–0.86, mean: 0.78) across six stations. For example, at Waizhou Station, the NSE of daily runoff simulations for XAJ and SWAT models with station precipitation was 0.86 and 0.87, respectively, larger than that for DTVGM with NSE 0.80. DTVGM exhibited the lowest relative error (5.87%), outperforming XAJ (22.65%) and SWAT (9.95%) models. In addition, the SWAT model performed better in simulating low-level runoff, while DTVGM tended to overestimate flood peaks and high-level runoff. Thus, selecting an appropriate model based on specific research is essential for achieving optimal results.