Uysal et al. (2026) A data-assimilated SEAS5 forecasting framework for seasonal hydropower inflows in a snow-dominated basin

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-03-10
• Authors: Gökçen Uysal, Marie-Amélie Boucher, Charles Mathieu, Rodolfo Alvarado-Montero, Ali Arda Şorman
• DOI: 10.1016/j.jhydrol.2026.135279

Research Groups

• Universit´e de Sherbrooke, Department of Civil and Building Engineering, Faculty of Engineering, Sherbrooke (Qu´ebec), Canada
• Eskis¸ehir Technical University, Department of Civil Engineering, Faculty of Engineering, Eskis¸ehir, Türkiye
• Hydro-Qu´ebec - Groupe Exploitation et infrastructure, Montr´eal (Qu´ebec), Canada
• University of Calgary, Department of Civil Engineering, Schulich School of Engineering, Calgary (Alberta), Canada

Short Summary

This study developed a seasonal hydropower inflow forecasting framework for a snow-dominated basin by integrating a variational data assimilation (VarDA) scheme into the HBV hydrological model, demonstrating significant improvements in inflow and snow water equivalent predictions, especially at short lead times.

Objective

• To enhance seasonal predictions of inflow, snow cover area, and snow water equivalent in the Manic-5 basin, Qu´ebec, Canada.
• To assess the reliability of SEAS5 seasonal forecasts in simulating key climate drivers in snow-dominated regions.
• To quantify the extent to which seasonal ensemble forecasts can improve inflow and snow state predictions.
• To evaluate the effectiveness of the Variational Data Assimilation (VarDA) technique in enhancing the skill of long-range hydrological forecasts.

Study Configuration

• Spatial Scale: Manicouagan 5 (Manic-5) sub-basin, Qu´ebec, Canada, with a draining area of 24,717 km². Elevation ranges from 346 m to 1072 m, divided into four elevation bands.
• Temporal Scale:
  • Calibration period: 1980–2007 (28 years)
  • Validation period: 2008–2016 (9 years)
  • Forecast period: 2017–2022 (6 years)
  • Forecast horizon: 215 days (approximately 7 months)
  • Data assimilation window: 6 months (retrospective)

Methodology and Data

• Models used:
  • Hydrological model: HBV-96 (semi-distributed conceptual hydrological model)
  • Data Assimilation: Variational Data Assimilation (VarDA) using Moving Horizon Estimation (MHE), implemented via RTC-Tools.
  • Meteorological forecasting system: ECMWF Seasonal Forecasting System version 5 (SEAS5)
  • Benchmark forecasting method: Extended Streamflow Prediction (ESP)
  • Operational platform: Delft-FEWS
• Data sources:
  • Hydro-meteorological data: Hydro-Qu´ebec (daily precipitation, daily mean air temperature, naturalized inflow, in-situ snow water equivalent (SWE) measurements).
  • Satellite-based snow cover area (SCA): IMS (Interactive Multisensor Snow and Ice Mapping System) version 1.2 (daily, 4 km spatial resolution).
  • Reanalysis snow water equivalent (SWE): ERA5-Land (daily, 9 km spatial resolution).
  • Seasonal meteorological forecasts: ECMWF SEAS5 (215-day horizon, 36 km horizontal resolution, 51 ensemble members).

Main Results

• The HBV model demonstrated robust performance in simulating inflow (Nash–Sutcliffe Efficiency (NSE): 0.77-0.80, Coefficient of Determination (R²): 0.78-0.84), snow cover area (R²: 0.89-0.93), and snow water equivalent (R²: 0.86-0.94) across calibration, validation, and forecast periods.
• Variational Data Assimilation (VarDA) significantly improved both inflow and snow water equivalent (SWE) forecasts across all meteorological forcing scenarios (Perfect, SEAS5, ESP), with the most notable gains observed during the first 1–3 months of lead time.
• The SEAS5-VarDA configuration consistently achieved the lowest Continuous Ranked Probability Score (CRPS) and highest Brier Skill Score (BSS) for inflow, SCA, and SWE, outperforming SEAS5-noDA and ESP-based systems.
• The benefits of VarDA were most pronounced during the melting season (April–June), enhancing the predictability of snow-dependent hydrological processes.
• VarDA improved forecast reliability, aligning forecast probabilities more closely with observed frequencies, particularly for inflow and SWE at short to medium lead times.
• SEAS5 demonstrated superior skill in capturing climate signals compared to ESP, while ESP maintained skill through its reliance on historical analogs.

Contributions

• Developed and evaluated a novel seasonal hydropower inflow forecasting system for a snow-dominated basin, integrating Variational Data Assimilation (VarDA) with ECMWF SEAS5 ensemble forecasts and the HBV hydrological model.
• Quantified the critical role of accurate snow state (SWE and SCA) initialization via VarDA in significantly enhancing seasonal inflow and snowpack prediction skill, particularly at short lead times and during the melting season.
• Provided a comprehensive comparative analysis of SEAS5-driven forecasts against climatology-based Extended Streamflow Prediction (ESP) and perfect forecasts, demonstrating the added value of dynamical seasonal forecasts augmented with data assimilation.
• Showcased the robustness and applicability of the VarDA approach in data-scarce environments, effectively utilizing sparse in-situ SWE data complemented by reanalysis and satellite products.
• Offered practical implications for climate-resilient hydropower planning and adaptive water management strategies in cold-region basins facing non-stationary climate dynamics.

Funding

• The Scientific and Technological Research Council of Türkiye (TÜB˙ITAK) under grant number 1059B192202922 (2219 International Postdoctoral Research Fellowship Programme).
• Programme de professeurs invit´es of the Universit´e de Sherbrooke.
• Hydro-Qu´ebec (for providing access to hydro-meteorological data).

Citation

@article{Uysal2026dataassimilated,
  author = {Uysal, Gökçen and Boucher, Marie-Amélie and Mathieu, Charles and Alvarado-Montero, Rodolfo and Şorman, Ali Arda},
  title = {A data-assimilated SEAS5 forecasting framework for seasonal hydropower inflows in a snow-dominated basin},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.135279},
  url = {https://doi.org/10.1016/j.jhydrol.2026.135279}
}