The value of high model resolution for streamflow in small and large near-natural European river basins

Macroscale hydrological models typically operate at a coarse spatial resolution, which limits their ability to assess locally relevant impacts of climate change. Recent advances in available forcing data, computational resources, and calls for higher-resolution modelling have led to the development of models running at 5 arcmin or finer resolutions. However, evaluations of streamflow performance in several studies assessing whether higher resolution improves streamflow simulations remain inconclusive. While some studies report improved streamflow performance with increasing resolution, others find that higher spatial resolution does not necessarily translate into better model performance. To better understand these inconsistencies, we apply the 5 arcmin and 30 arcmin spatial resolutions of the latest version of the global hydrological model, WaterGAP, to investigate the impact of spatial resolution on streamflow performance in 131 near-natural European basins for the period 1950–2024. Basins are classified as small (3,000–9,000 km², 20 basins) or large (>9,000 km², 111 basins) to assess the added value of higher spatial resolution. Both model versions are run without calibration. Our results show comparable performance (KGE-based performance ratio is within ±20%) between model resolutions in 73% of large basins. The 5 arcmin version shows increased performance in 4.5% of large basins and decreased performance in 22.5% of large basins compared to 30 arcmin model resolution.   In small basins, the 5 arcmin version shows comparable performance to the 30 arcmin version in 55% of basins, performs worse in 30% of basins, and outperforms the 30 arcmin version in only 15% of cases, despite expectations that higher resolution should better capture spatial variability. Overall, median streamflow performance is lower in the 5 arcmin version compared to the 30 arcmin version, both in large (KGE_{5 arcmin} = 0.46, KGE_{30 arcmin} = 0.53) and small (KGE_{5 arcmin} = 0.41, KGE_{30 arcmin} = 0.45) basins.  This lower performance is primarily due to the overestimation of the long-term mean streamflow in both resolutions for both large and small basins, with the overestimation being slightly worse in the 5 arcmin version. It should be noted that median streamflow variability is underestimated in small river basins but is captured well in large river basins, with the 5 arcmin version showing better median streamflow variability than the 30 arcmin version. Also, correlation (timing) is improved as resolution increases. This limited ability to reproduce streamflow variability and long-term mean discharge, which has also been reported in the literature, may explain the inconsistencies in streamflow performance regarding the benefits of increased spatial resolution. These result underscores the need for targeted enhancements in model parameterization and forcing data to improve variability and bias performance .