Liu et al. (2026) HydEquivNet: bridging classical unit hydrograph theory and data-driven hydrology via scale-equivariant learning

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-07-31
• Authors: Jingzhe Liu, Chenhui Jiang, Yongnan Zhu, Shan Jiang, Ziqian Gong, Jiaqi Zhai, Jing Zhao, Yulong Gao, Tao Wang
• DOI: 10.1016/j.jhydrol.2026.136146

Research Groups

• State Key Laboratory of Water Cycle and Water Security, China Institute of Water Resources and Hydropower Research, Beijing 100038, China
• Department of Hydraulic Engineering, Tsinghua University, Beijing 100084, China
• RTI International, Research Triangle Park, NC 27709, USA
• General Institute of Water Resources and Hydropower Planning and Design, Ministry of Water Resources of China, Beijing 100120, China
• College of Water Resources, North China University of Water Resources and Electric Power, Zhengzhou 450046, China
• Guangzhou Maritime University, Guangzhou 510725, China

Short Summary

This paper proposes HydEquivNet (HEN), a novel architecture that introduces Lie group equivariance into hydrological modeling, achieving improved physical consistency and robustness in rainfall-runoff simulation.

Objective

• Investigate the feasibility of incorporating classical unit hydrograph theory and mass-consistency considerations into deep learning-based hydrological models for improved physical consistency and extrapolation robustness.

Study Configuration

• Spatial Scale: Catchment scale (531 CAMELS catchments)
• Temporal Scale: Daily to monthly time steps

Methodology and Data

• Models used: HydEquivNet (HEN), a novel architecture combining Lie group equivariant layers with Long Short-Term Memory (LSTM) networks
• Data sources: CAMELS dataset, including catchment attributes and meteorological data

Main Results

• Median Nash-Sutcliffe Efficiency (NSE) of 0.724 achieved by HEN on the CAMELS dataset
• Improved equivariance-guided proportional-response robustness under synthetic out-of-distribution input amplification
• Interpretable unit-hydrograph-like response patterns, including a 0.94 bin-averaged correlation between extracted Peak Lag Time (Tp) and physically motivated concentration–time factor (CTF)

Contributions

• The development of HydEquivNet offers a parameter-efficient and transparent route toward more physically informed rainfall-runoff modeling under changing hydroclimatic conditions.

Funding

• This research was supported by the National Natural Science Foundation of China (Grant No. 52209201) and the China Institute of Water Resources and Hydropower Research (Grant No. IWHR2023001).

Citation

@article{Liu2026HydEquivNet,
  author = {Liu, Jingzhe and Jiang, Chenhui and Zhu, Yongnan and Jiang, Shan and Gong, Ziqian and Zhai, Jiaqi and Zhao, Jing and Gao, Yulong and Wang, Tao},
  title = {HydEquivNet: bridging classical unit hydrograph theory and data-driven hydrology via scale-equivariant learning},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136146},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136146}
}