Steyaert et al. (2025) Data derived reservoir operations simulated in a global hydrologic model

Identification

• Journal: Hydrology and earth system sciences
• Year: 2025
• Date: 2025-11-19
• Authors: Jennie C. Steyaert, Edwin H. Sutanudjaja, Marc F. P. Bierkens, Niko Wanders
• DOI: 10.5194/hess-29-6499-2025

Research Groups

• Department of Physical Geography, Utrecht University, Utrecht, the Netherlands
• Deltares, Unit D Subsurface and Groundwater Systems, Utrecht, the Netherlands

Short Summary

This study develops a workflow to implement data-derived reservoir operations in global hydrologic models using satellite altimetry and machine learning. It demonstrates that this approach significantly improves the accuracy of simulated global reservoir storage compared to generic operations, which tend to overestimate storage and water availability, while having modest impacts on downstream streamflow.

Objective

• To develop a workflow for implementing data-derived reservoir operations in large-scale hydrologic models using global satellite altimetry data and a data-driven framework.
• To evaluate if data-derived reservoir operations are more accurate than generic ones and what this implies for current data gaps in reservoir operational data.
• To assess the sensitivity of different reservoir operations to the size of the command area.
• To determine how the ability of global hydrologic models to reproduce streamflow and reservoir storage changes depending on the type of reservoir operation used (generic vs. data-driven).

Study Configuration

• Spatial Scale: Global, covering all landmasses except Greenland and Antarctica, at a 5 arcminute (approximately 10 km) spatial resolution. The study includes over 24,000 reservoirs and evaluates downstream command areas of 250 km, 600 km, and 1100 km.
• Temporal Scale: Weekly for reservoir operational bounds derivation (1980–2020), daily for hydrological model simulations (1980–2023), and long-term annual averages for analysis.

Methodology and Data

• Models used: PCR-GLOBWB 2 (global hydrological model), STARFIT (reservoir operational range derivation), Random Forest (extrapolation of operational parameters).
• Data sources: GeoDAR (global dam and reservoir dataset), GRanD (Global Reservoirs and Dams Dataset), ICOLD registry, GloLakes (remotely sensed global reservoir storage time series), ResOpsUS (historical reservoir operations for the contiguous United States), LandsatPlusICESat2/Sentinel2 (satellite altimetry for water levels), Global Runoff Data Center (GRDC) streamflow data, socioeconomic and hydroclimatic variables.

Main Results

• The random forest algorithm effectively captures reservoir storage dynamics, with primary errors stemming from the input satellite altimetry data.
• Implementing data-derived operational bounds significantly increases the accuracy of simulated global reservoir storage, showing better alignment with both remotely sensed (GloLakes RMSE: 0.28 vs. 0.32 for generic) and observed (ResOpsUS RMSE: 0.30 vs. 0.37 for generic) storage.
• Generic operational schemes tend to overestimate global reservoir storage and potentially water availability, as data-derived storages are consistently lower.
• The impact of data-derived operations on downstream streamflow is modest and localized, with minimal effects at basin outlets, though they show improved representation of streamflow variability.
• Including downstream demand (command areas) leads to slight improvements in model performance.
• Updating the reservoir dataset from GRanD to GeoDAR increases global modeled reservoir storage by 768 cubic kilometers (from 6355.72 cubic kilometers to 7123.66 cubic kilometers), with 95% of basins showing increases.

Contributions

• First global application of a data-derived methodology for inferring and implementing reservoir operations in a large-scale hydrologic model (PCR-GLOBWB 2).
• Provides a scalable workflow that leverages global satellite altimetry data and machine learning to derive dynamic operational bounds for over 24,000 reservoirs.
• Demonstrates that data-derived operations lead to more accurate global reservoir storage simulations, challenging the overestimation of water availability by generic schemes.
• Highlights the importance of validating reservoir models against storage observations in addition to streamflow, especially given the localized impact on streamflow.
• Categorizes dams into "irrigation-like" and "hydropower-like" operations based on main purpose, showing improved representation of observed dynamics.

Funding

• European Union's Horizon EUROPE Research and Innovation Programme (Grant Agreement No. 101059264, SOS-WATER – Water Resources System Safe Operating Space in a Changing Climate and Society).

Citation

@article{Steyaert2025Data,
  author = {Steyaert, Jennie C. and Sutanudjaja, Edwin H. and Bierkens, Marc F. P. and Wanders, Niko},
  title = {Data derived reservoir operations simulated in a global hydrologic model},
  journal = {Hydrology and earth system sciences},
  year = {2025},
  doi = {10.5194/hess-29-6499-2025},
  url = {https://doi.org/10.5194/hess-29-6499-2025}
}