Hyper-resolution hydrological modelling over Europe: results and emerging challenges

Modelling the terrestrial hydrological cycle at ‘hyper-resolution’, i.e., with a grid cell size of 1 km or below) was and still is a major quest in hydrological sciences. With an increase in computational power and the number of readily available and open datasets at useful spatial resolutions increasing as well, hyper-resolution modelling efforts have grown in number as well. We here present a first continental-scale application of the global hydrological model PCR-GLOBWB over Europe at 1 km spatial resolution, and offset it against runs with traditional resolutions of 10 km and 50 km, respectively. Model output was validated for more than 200 water provinces against observed discharge and the following remotely sensed data products: ESA-CCI soil moisture data and GRACE/GRACE-FO terrestrial water storage anomalies. Evaporation estimates were compared to GLEAM data. Evaluation metrics indicate good model performance over Europe and increased accuracy with finer spatial resolutions, particularly for discharge simulations. While the used validation products have the advantage of global coverage and long observational records, their spatial resolution is actually too coarse to fully assess the accuracy of models at hyper-resolution. At that scale, more recent satellite products can be of more use but at the cost of only short observation record. We thus additionally validated 1 km model output against Sentinel-1 surface soil moisture and compared it against results obtained for ESA-CCI soil moisture data. Besides challenges related to global-scale fine-resolution observational data, we also acknowledge that additional work needs to focus on model parameterization for hyper-resolution as well model improvements such as routing schemes better utilizing the available spatial detail. Another challenge we identified is required run time and computational power to analyze continental-scale 1 km data, even when using the state-of-the-art Dutch supercomputer. Here, efficient programming and use of latest parallelization techniques will become even more crucial. Despite these solvable challenges, our research shows that large-scale hyper-resolution modelling is now feasible and that further pursuing these efforts can eventually lead to more locally-relevant hydrological information and process understanding.