Kupzig et al. (2025) A controlled model experiment for the global hydrological model WaterGAP3: Understanding recent and new advances in the model structure

Identification

• Journal: Environmental Modelling & Software
• Year: 2025
• Date: 2025-09-25
• Authors: Jenny Kupzig, Martina Flörke
• DOI: 10.1016/j.envsoft.2025.106703

Research Groups

• Institute of Engineering Hydrology and Water Resources Management, Ruhr University Bochum, Germany

Short Summary

This study conducts a controlled model experiment on the global hydrological model WaterGAP3 to evaluate if increased model complexity in river routing, reservoir algorithms, and snow on wetlands leads to better process representation. It finds that volume-dependent river routing generally worsens performance, while reservoir algorithms and snow on wetlands show benefits but require further development.

Objective

• To assess whether recent advances in the WaterGAP3 model structure result in a more accurate and adequate representation of discharge.
• To evaluate the accuracy and adequacy of three previous modifications (volume-dependent river routing, irrigation/non-irrigation reservoirs, non-irrigation reservoir release management) and one newly integrated modification (snow processes on wetlands), and to provide insights for future model development.

Study Configuration

• Spatial Scale: Global hydrological model (WaterGAP3) applied at a 5 arcminute grid resolution across 226 unnested basins in North America, each with a minimum size of 3,000 square kilometers.
• Temporal Scale: Daily timestep simulations over a 16-year period (1979–1994), with a 10-year calibration period (1979–1988) and a 6-year evaluation period (1989–1994).

Methodology and Data

• Models used: WaterGAP3, a conceptual global hydrological model. The study utilized different structural versions of WaterGAP3, including:
  • Static vs. volume-dependent river routing.
  • Three reservoir algorithms: V0 (non-regulated lakes), V1 (constant release with reservoir-specific data), and V2 (V1 extended with flood/drought mitigation for non-irrigation reservoirs).
  • With vs. without snow processes on wetlands.
  • A Monte-Carlo approach was used for calibration of the parameter γ.
  • A full factorial design was employed for the controlled model experiment.
• Data sources:
  • Climate data: EWEMBI2B (Lange, 2019) reanalysis product.
  • Observed discharge data: Global Runoff Data Centre (GRDC, 2020).
  • Wetland and lake data: Global Lakes and Wetlands Database (Lehner and D¨oll, 2004).
  • Aridity Index: Zomer and Trabucco (2022).
  • Model evaluation metrics: Kling-Gupta-Efficiency (KGE), Nash-Sutcliffe Efficiency (NSE), and streamflow signatures (e.g., low/high flow magnitude, frequency, duration).

Main Results

• Volume-dependent river routing: This modification significantly reduced simulation quality, with 61% of behavioral basins showing a decrease in ΔKGE and 70% in ΔNSE. It adversely affected the timing of simulated discharge, leading to the rejection of the hypothesis that it improves accuracy and adequacy.
• Reservoir algorithm (V1): Increased model accuracy in 44% (ΔKGE) to 56% (ΔNSE) of basins. It reduced flow variability and enhanced low-flow magnitudes, but had a negligible impact on discharge timing. The improvement was primarily attributed to the use of more accurate reservoir-specific storage capacities rather than the algorithm's management rules. Hypothesis B was rejected due to insufficient improvement in timing.
• Reservoir algorithm (V2): Showed a general improvement in model performance (44% ΔKGE, 59% ΔNSE positive change) and a more pronounced positive impact on discharge timing compared to V1. It further mitigated variability. However, it exhibited unrealistic within-month fluctuations in reservoir releases due to numerical issues in its daily implementation of monthly-based inflow forecasts. The computational time also significantly increased. Despite deficiencies, hypothesis B was supported.
• Snow on wetlands: This newly integrated process improved discharge representation, particularly in non-behavioural basins (21% ΔKGE, 18% ΔNSE positive change). It mainly influenced the timing of discharge and improved seasonality by reducing low flow during the cold season. Identified deficiencies include the neglect of sublimation, suboptimal parameterization, and overestimated discharge during winter periods. Hypothesis B was not rejected, but further development is needed.
• Overall Model Performance: WaterGAP3 generally performed better in simulating coastal regions of North America, especially in more humid and less snow-influenced basins. Non-behavioural basins were characterized by greater snow influence and higher aridity.

Contributions

• First-time controlled model experiment for WaterGAP3 to systematically test the benefits of increased model complexity in specific components (river routing, reservoirs, snow on wetlands).
• Demonstrated the efficacy of model falsification through large-sample analysis for scientific progress in hydrological modeling.
• Highlighted that increased model complexity does not always lead to better process representation, emphasizing the need for rigorous testing of model modifications for both accuracy and adequacy.
• Provided specific insights for future development of WaterGAP3 components, including reversing the volume-dependent river routing, refining reservoir algorithms with more detailed data and management strategies, and improving snow processes on wetlands (e.g., by incorporating sublimation).
• Emphasized the importance of revisiting historical model development decisions, especially for global hydrological models used for extrapolation, as new data and methodologies can lead to different conclusions.

Funding

• This research received no specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
• The DEAL project partially funded the article processing charge.

Citation

@article{Kupzig2025controlled,
  author = {Kupzig, Jenny and Flörke, Martina},
  title = {A controlled model experiment for the global hydrological model WaterGAP3: Understanding recent and new advances in the model structure},
  journal = {Environmental Modelling & Software},
  year = {2025},
  doi = {10.1016/j.envsoft.2025.106703},
  url = {https://doi.org/10.1016/j.envsoft.2025.106703}
}