Collignan et al. (2026) The added value of considering evapotranspiration fluxes in the calibration of a semi-distributed hydrological model

Identification

• Journal: International Journal of River Basin Management
• Year: 2026
• Date: 2026-09-19
• Authors: Julie Collignan, Alban de Lavenne, Maria-Helena Ramos
• DOI: 10.1080/15715124.2026.2711679

Research Groups

• INRAE, UR HYCAR, Université Paris-Saclay, Antony, France

Short Summary

This study investigates the added value of incorporating actual evapotranspiration (ET) fluxes into the calibration of a semi-distributed hydrological model (GRSD) to enhance its robustness and reliability for climate change projections. The results demonstrate that ET-constrained calibration leads to more spatially coherent fluxes and a more stable model response under non-stationary conditions, improving the trustworthiness of future hydrological projections.

Objective

• To investigate how hydrological model calibration choices influence the performance and robustness of a semi-distributed conceptual hydrological model under climate change.
• To assess the added value of introducing an additional constraint on actual evapotranspiration (ET) data during calibration to better represent the catchment’s water balance and enhance model transferability to future climate conditions.
• To test the hypothesis that imposing stronger constraints on multiple hydrological fluxes (discharge and ET) reduces equifinality and improves the representation of water balance components, thereby enhancing model robustness in both time and space.

Study Configuration

• Spatial Scale: Seine River catchment upstream of Vernon, France, covering 155 sub-catchments.
• Temporal Scale:
  • Historical period: 1996–2020 (climate data), 2001–2020 (discharge data).
  • Climate projection period: 2070–2098.
  • Daily time step for hydrological modeling and data.

Methodology and Data

• Models used:
  • Hydrological model: GRSD (Semi-distributed GR rainfall-runoff model), based on the lumped GR4J model.
  • Optimization algorithm: caRamel algorithm (multi-objective optimization).
  • Actual evapotranspiration (ET) estimation: Budyko-like Turc-Mezentev formula (ETTM) with a coefficient n = 2.5.
  • Potential evapotranspiration (PET) estimation: Oudin formula.
• Data sources:
  • Historical climate data (precipitation and temperature): SAFRAN meteorological reanalysis product from Météo France, downscaled to an 8 km × 8 km grid.
  • Climate projections: Explore2 national project, providing 17 global/regional circulation model pairs for 3 Representative Concentration Pathway (RCP) scenarios from CMIP5, downscaled to an 8 km × 8 km grid and bias-corrected. Four contrasted RCP8.5 scenarios were selected.
  • Discharge data: Daily time series from 155 gauging stations across the Seine River catchment, extracted from the French hydro database.
  • Reservoir data: Daily storage and release data, and theoretical rule curves, provided by Etablissement public territorial de bassin (EPTB) Seine Grands Lacs.

Main Results

• The ET-constrained calibration strategy (CalQE) resulted in a smaller and more homogeneous relative bias pattern for simulated actual evapotranspiration (ET) across the catchment compared to the discharge-only strategy (CalQ). The SPAtial EFficiency (SPAEF) metric for ET was 0.27 for CalQE versus −0.40 for CalQ.
• Cal_QE improved the reproduction of the discharge regime, particularly from September to December, and enhanced model performance during dryer months and for larger sub-catchments.
• Cal_QE demonstrated superior spatial robustness in the leave-one-station-out test (LOOT), especially for smaller sub-catchments (area < 100 km²).
• In split-sample tests, CalQE exhibited more consistent performance between calibration and validation periods, indicating greater temporal stability and less susceptibility to overfitting compared to CalQ, despite a slight reduction in overall KGE(Q) performance.
• The choice of calibration strategy significantly influenced the spatial distribution of anomalies for all discharge indicators (average discharge, annual minimum monthly discharge, annual maximum daily discharge) under future climate scenarios.
• Under drier climate scenarios, CalQ predicted a smaller decrease in average discharge in the eastern part of the catchment (an area with higher average ET) compared to CalQE, highlighting the impact of ET sensitivity on projected discharge anomalies.

Contributions

• This study provides robust evidence for the added value of multi-variate calibration, incorporating actual evapotranspiration (ET) as an additional constraint alongside discharge, for semi-distributed hydrological models.
• It demonstrates that this approach significantly enhances the spatial and temporal robustness of hydrological models, leading to more reliable and trustworthy projections under climate change conditions.
• The research extends previous findings by applying the ET-constrained calibration to a large, heterogeneous river basin (Seine River basin, France) and evaluating its impact across contrasted climate change scenarios.
• It highlights the critical importance of calibration strategy choices in shaping hydrological projections, particularly for extreme flows and spatial patterns, which is essential for water resources management and risk assessment.
• The use of a generalized Budyko-based ET constraint makes the proposed methodology easily replicable and applicable across diverse catchments without requiring specific remote sensing products.

Funding

• The STARS4Water project, funded through the European Union’s Horizon Europe research and innovation programme under Grant Agreement No 101059372.

Citation

@article{Collignan2026added,
  author = {Collignan, Julie and Lavenne, Alban de and Ramos, Maria-Helena},
  title = {The added value of considering evapotranspiration fluxes in the calibration of a semi-distributed hydrological model},
  journal = {International Journal of River Basin Management},
  year = {2026},
  doi = {10.1080/15715124.2026.2711679},
  url = {https://doi.org/10.1080/15715124.2026.2711679}
}