Bo et al. (2026) Physics-constrained deep learning for evapotranspiration estimation by integrating process-model information and observations

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-10-01
• Authors: Yong Bo, Kai Liu, Shudong Wang, Long Li, Guoxu Li, Xueke Li
• DOI: 10.1016/j.jhydrol.2026.136557

Research Groups

• Department of Hydrology, Nanjing University
• Institute of Water Resources Science, Ministry of Water Resources
• Key Laboratory of Hydroscience and Engineering, Tongji University

Short Summary

This study proposes a physics-constrained deep learning approach for estimating evapotranspiration by integrating process-model information and observations. The method demonstrates improved accuracy in evapotranspiration estimation compared to traditional machine learning models.

Objective

• Investigate the feasibility of using physics-constrained deep learning for evapotranspiration estimation, incorporating both process-model information and observational data.

Study Configuration

• Spatial Scale: Continental scale (China)
• Temporal Scale: Daily time step

Methodology and Data

• Models used: Physics-Constrained Neural Network (PCNN) with the ISBA-AF-BE v1.3 model
• Data sources: Satellite data from MODIS, ground-based observations from the China Hydrology Dataset, and reanalysis data from ERA5

Main Results

• The PCNN-ISBA-AF-BE v1.3 model achieved a mean absolute error (MAE) of 0.45 mm d^-1 and a coefficient of determination (R^2) of 0.83 for daily evapotranspiration estimation.
• The results show that incorporating process-model information significantly improves the accuracy of evapotranspiration estimation compared to traditional machine learning models.

Contributions

• This study contributes to the development of physics-constrained deep learning methods for hydrological applications, particularly in evapotranspiration estimation.
• The proposed approach demonstrates improved accuracy and robustness compared to existing methods, providing a valuable tool for hydrologists and climate scientists.

Funding

• National Natural Science Foundation of China (Grant No. 52179215)
• Ministry of Water Resources' Special Fund for Scientific Research on Public Causes (Grant No. 201901003)

Citation

@article{Bo2026Physicsconstrained,
  author = {Bo, Yong and Liu, Kai and Wang, Shudong and Li, Long and Li, Guoxu and Li, Xueke},
  title = {Physics-constrained deep learning for evapotranspiration estimation by integrating process-model information and observations},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136557},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136557}
}