Clerc-Schwarzenbach et al. (2026) Evaluating E-OBS forcing data for large-sample hydrology using model performance diagnostics

Identification

• Journal: Hydrology and earth system sciences
• Year: 2026
• Date: 2026-01-13
• Authors: Franziska Clerc-Schwarzenbach, Thiago V. M. do Nascimento
• DOI: 10.5194/hess-30-119-2026

Research Groups

• Department of Geography, University of Zurich, Zurich, Switzerland
• Eawag: Swiss Federal Institute of Aquatic Science and Technology, Dübendorf, Switzerland

Short Summary

This study evaluates the hydrological efficacy of E-OBS meteorological forcing data for 2682 European catchments by comparing it to eight national/regional CAMELS-like datasets using a bucket-type hydrological model. It finds that E-OBS data generally lead to slightly lower, but still good, model performance compared to national datasets, with performance strongly linked to E-OBS station density, making it a reasonable harmonized option for pan-European large-sample hydrological studies.

Objective

• To compare meteorological time series from the pan-European E-OBS product (as used in EStreams) with those from national/regional CAMELS-like datasets across 2682 European catchments.
• To assess how differences in these meteorological forcing data impact the performance of a bucket-type hydrological model for large-sample hydrological applications.

Study Configuration

• Spatial Scale: 2682 catchments across eight European countries (Czechia, Denmark, France, Germany, Great Britain, Spain, Sweden, Switzerland). Catchment areas are below 2000 square kilometres. The study covers a continental (pan-European) extent.
• Temporal Scale:
  • Data comparison period: October 1995 to September 2015 (20 years).
  • Hydrological model warming-up period: October 1990 to September 1995 (5 years).
  • Hydrological model simulation and calibration period: October 1995 to September 2015 (20 years).
  • Temporal resolution of data and simulations: Daily.

Methodology and Data

• Models used: HBV (Hydrologiska Byråns Vattenbalansavdelning) model, specifically the HBV-light version, calibrated using a genetic algorithm.
• Data sources:
  • Meteorological Forcing Data:
    • E-OBS (Ensemble Observation) product (version 1.4 from EStreams) for daily precipitation, potential evapotranspiration (Epot, calculated with Hargreaves formula using E-OBS temperature), and temperature. Spatial resolution: 0.1 degrees latitude/longitude.
    • Eight national/regional CAMELS-like datasets (CAMELS-CZ, CAMELS-DK, CAMELS-FR, CAMELS-DE, CAMELS-GB, BULL, CAMELS-SE, CAMELS-CH) for daily precipitation, Epot, and temperature. These are primarily station-based products with varying spatial resolutions (e.g., 1 km to 40 km).
  • Streamflow Data: Observed daily streamflow from the respective CAMELS datasets.
  • Catchment Attributes: EStreams dataset for catchment delineation, area, number of lakes, and normalized upstream reservoir capacity.
  • Digital Elevation Model: Copernicus DEM at 30 metres resolution for determining elevation zones for the HBV model.

Main Results

• Meteorological Data Differences:
  • Mean annual precipitation from E-OBS was systematically lower than CAMELS data for 88% of catchments, with deviations exceeding -10% for 28% of catchments. Largest differences were in Spain, smallest in Germany.
  • Mean annual potential evapotranspiration (Epot) from E-OBS was higher than CAMELS data for 94% of catchments, with deviations of at least 10% for 50% of catchments.
  • Average temperature from E-OBS was higher than CAMELS data for most catchments (median difference: 0.3 °C).
  • Aridity indices (Epot/P) were generally higher when calculated with E-OBS data due to lower precipitation and higher Epot.
• Hydrological Model Performance:
  • Both CAMELS (median Kling–Gupta efficiency (KGE) = 0.89) and E-OBS (median KGE = 0.87) forcing data resulted in high model performances (KGE > 0.70 for >90% of catchments).
  • However, model performance was statistically significantly higher when using CAMELS data for 62% of catchments.
  • Replacing E-OBS precipitation with CAMELS precipitation had the strongest positive impact on model performance, indicating precipitation as the primary driver of performance differences.
  • Replacing E-OBS Epot or temperature with CAMELS data had a very limited effect on model performance.
• Influence of Catchment Attributes:
  • Model performance with E-OBS forcing data was positively correlated with the density of E-OBS precipitation stations within or near a catchment.
  • Model performance tended to be lower in catchments with higher aridity indices (drier catchments) for both forcing datasets.

Contributions

• Provides the first continental-scale assessment of the E-OBS meteorological dataset specifically for hydrological modeling applications across Europe.
• Quantifies the systematic differences in precipitation, potential evapotranspiration, and temperature between the E-OBS product and various national/regional CAMELS-like datasets.
• Demonstrates the impact of these meteorological data differences on hydrological model performance, highlighting that while E-OBS generally yields slightly lower performance, it remains a reasonable basis for large-sample hydrological studies.
• Identifies key factors influencing model performance differences, such as station density, spatial resolution, and the nature of the forcing product (station-based vs. reanalysis components).
• Offers valuable insights for users of the EStreams dataset and E-OBS data, informing about potential data quality variations and suitability for pan-European hydrological synthesis.

Funding

• "Money Follows Cooperation" project (grant-no.: OCENW.M.21.230)
• Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO)
• Swiss National Science Foundation (SNSF)

Citation

@article{ClercSchwarzenbach2026Evaluating,
  author = {Clerc-Schwarzenbach, Franziska and Nascimento, Thiago V. M. do},
  title = {Evaluating E-OBS forcing data for large-sample hydrology using model performance diagnostics},
  journal = {Hydrology and earth system sciences},
  year = {2026},
  doi = {10.5194/hess-30-119-2026},
  url = {https://doi.org/10.5194/hess-30-119-2026}
}