Gu et al. (2026) MEDAL: a mechanism-embedded dual-physics attention learning framework for snowmelt-driven runoff prediction in cold-region basins

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-09-07
• Authors: Miao Gu, Wenchuan Wang, Kun-mei Luo, C. SHI, Dong-mei Xu
• DOI: 10.1016/j.jhydrol.2026.136376

Research Groups

• College of Water Resources, North China University of Water Resources and Electric Power

Short Summary

This study proposes a Mechanism-Embedded Dual-Physics Attention Learning (MEDAL) framework for daily runoff prediction in snowmelt-driven cold-region basins, achieving improved performance compared to existing physics-guided hybrid modeling approaches.

Objective

• Investigate the feasibility of using a mechanism-embedded dual-physics attention learning framework for predicting daily runoff in snowmelt-driven basins

Study Configuration

• Spatial Scale: Basin scale (five study basins)
• Temporal Scale: Daily time step, with evaluation metrics calculated over various temporal subsets (year, season, representative snowmelt event)

Methodology and Data

• Models used: Mechanism-Embedded Dual-Physics Attention Learning (MEDAL) framework, SIMHYD-Snow conceptual model as upstream physical feature extractor
• Data sources: Meteorological data, snow water equivalent, soil moisture, groundwater storage, and runoff observations

Main Results

• MEDAL achieved the best performance across all eight evaluation metrics in the five study basins, with reduced mean absolute error (MAE) and root mean square error (RMSE) by an average of 31.3%–51.1% and 24.2%–35.2%, respectively.
• The model demonstrated good robustness to data perturbations (Nash–Sutcliffe efficiency > 0.9 under scenarios with 10% noise or missing data).

Contributions

• This study provides a physics-guided hybrid modeling approach that balances predictive accuracy, physical consistency, process interpretability, and computational efficiency for snowmelt-driven runoff prediction in cold regions.

Funding

• This research was supported by the National Natural Science Foundation of China (Grant No. 52209262) and the Fundamental Research Funds for the Central Universities (Grant No. 2022MSM001).

Citation

@article{Gu2026MEDAL,
  author = {Gu, Miao and Wang, Wenchuan and Luo, Kun-mei and SHI, C. and Xu, Dong-mei},
  title = {MEDAL: a mechanism-embedded dual-physics attention learning framework for snowmelt-driven runoff prediction in cold-region basins},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136376},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136376}
}