Chen et al. (2025) Coupling Differentiable Modules of Reservoir Operation and Rainfall‐Runoff Processes for Streamflow Simulation

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water Resources Research
• Year: 2025
• Date: 2025-12-01
• Authors: Zexin Chen, Tongtiegang Zhao, Bingyao Zhang, Yu Li
• DOI: 10.1029/2025wr041684

Research Groups

Not explicitly stated in the abstract.

Short Summary

This paper tests the integration of reservoir operation and rainfall-runoff processes using differentiable parameter learning (dPL) within hydrological models. The study demonstrates that dPL significantly improves model efficiency, with a differentiable loosely coupled model (LCM) showing superior performance in simulating both inflow and outflow, particularly for ungauged catchments.

Objective

• To test the coupling of reservoir operation and rainfall-runoff processes under the framework of differentiable parameter learning (dPL) to improve hydrological modeling in human-regulated catchments.

Study Configuration

• Spatial Scale: 77 reservoirs across various catchments.
• Temporal Scale: Not explicitly defined in the abstract, but implied to be long-term time series for parameter calibration using LSTM networks.

Methodology and Data

• Models used:
  • Community Water Model (CWatM)'s reservoir operation module
  • Hydrologiska Byråns Vattenbalansavdelning (HBV) model
  • Differentiable Parameter Learning (dPL) framework
  • Differentiable Fully Coupled Model (FCM) using one Long Short-Term Memory (LSTM) network
  • Differentiable Loosely Coupled Model (LCM) using two LSTM networks
  • Differentiable HBV model
• Data sources:
  • Reservoir outflow observations
  • Reservoir inflow observations (for LCM and comparison)

Main Results

• Differentiable parameter learning (dPL) is effective in improving the efficiency of conventional hydrological models.
• The median Kling-Gupta efficiency (KGE) improved:
  • From 0.53 for HBV to 0.59 for differentiable HBV.
  • From 0.52 for FCM to 0.61 for differentiable FCM.
  • From 0.54 for LCM to 0.60 for differentiable LCM.
• Differentiable HBV fits reservoir outflow by underestimating recession coefficients and overestimating the baseflow index.
• Differentiable FCM fits outflow but not inflow, tending to overestimate the maximum storage of the upper soil layer.
• Differentiable LCM successfully fits both inflow and outflow, with one LSTM estimating HBV parameters and another estimating CWatM's reservoir operation module parameters.
• For ungauged catchments, the differentiable LCM outperforms differentiable FCM in reproducing both inflow and outflow.

Contributions

• Demonstrates the effectiveness of differentiable parameter learning (dPL) in enhancing the efficiency of hydrological models for human-regulated catchments.
• Introduces and evaluates two novel differentiable coupling strategies (fully coupled and loosely coupled models) for integrating reservoir operation and rainfall-runoff processes.
• Provides detailed insights into the hydrological process biases introduced by different differentiable model configurations (e.g., parameter estimation tendencies).
• Identifies the differentiable loosely coupled model (LCM) as a superior approach for simulating both inflow and outflow, particularly highlighting its robustness for ungauged catchments.

Funding

Not explicitly stated in the abstract.

Citation

@article{Chen2025Coupling,
  author = {Chen, Zexin and Zhao, Tongtiegang and Zhang, Bingyao and Li, Yu},
  title = {Coupling Differentiable Modules of Reservoir Operation and Rainfall‐Runoff Processes for Streamflow Simulation},
  journal = {Water Resources Research},
  year = {2025},
  doi = {10.1029/2025wr041684},
  url = {https://doi.org/10.1029/2025wr041684}
}