From continent to coast: advances in HYPE hydrological model calibration and AI-based model enhancement

The Horizon Europe project FOCCUS (https://foccus-project.eu) aims to enhance Copernicus Marine Service's coastal dimension by developing innovative products as well as facilitating seamless ocean monitoring and forecasting. The scope of the project includes improvement of the estimation of water and matter runoff to the European coastal regions using large-scale hydrological models. Effort is directed on improving the pan-European HYPE (E-HYPE) hydrological model (see 1) both through a traditional calibration and a hybrid modelling approach through an AI-based enhancement.

The calibration framework was revised and updated, including changes in the application of common criteria (NSE, KGE, RE, R²). The previous E-HYPE calibration focused on improving domain-average model performance. Instead of relying on domain-average model performance, the proportion of calibration stations meeting acceptable performance thresholds was used to increase the number of well-calibrated stations. This approach reduces the influence of stations with highly unreliable data that may otherwise bias criteria-based parameter selection. The model validity was also assessed for key physical processes including snow accumulation, reservoir siltation, and sedimentation-resuspension dynamics. The final step involved manual inspection of time series and performance distributions for streamflow, nutrient and sediment concentrations, as well as snow water dynamics. Validation was conducted at the spatial extent across gauged catchments (not used for calibration), and at major coastal outlets. The updated E-HYPE model shows improved overall performance compared to its previous benchmark version, especially in streamflow, sediment concentration and evapotranspiration.

Finally, a hybrid modelling approach was applied, which included an AI-based post-processing to improve the streamflow predictions at coastal outlets (see 2). This effort involves transferring the knowledge learned from the upstream gauged locations, providing improved predictive performance at ungauged locations (not included in the training stations). Overall, the final dataset includes daily streamflow, sediment and nutrient concentration at 5,302 European coastal outlets for the period 2000-2024 and will be soon publicly available on Zenodo.

References:

1. Brendel, C., Capell, R., & Bartosova, A. (2023). To tame a land: Limiting factors in model performance for the multi-objective calibration of a pan-European, semi-distributed hydrological model for discharge and sediments. Journal of Hydrology: Regional Studies, 50, 101544. 

2. Du, Y., & Pechlivanidis, I. G. (2025). Hybrid approaches enhance hydrological model usability for local streamflow prediction. Communications Earth & Environment, 6(1), 334.