Ai et al. (2026) Physics-informed neural networks for hydrodynamic inversion and discharge estimation in river flows over complex bed topography

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-09-20
• Authors: Congfang Ai, Yuxiang Ma, Junzheng Li, Jiajie Song, Jiaxiang Sun
• DOI: 10.1016/j.jhydrol.2026.136463

Research Groups

State Key Laboratory of Coastal and Offshore Engineering, Dalian University of Technology, China.

Short Summary

This study develops a physics-informed neural network (PINN)-based framework for hydrodynamic inversion and discharge estimation in river flows over complex bed topography under limited observations. The proposed approach can reliably estimate discharge and reconstruct hydraulic states even with sparse or no bed information available.

Objective

• Investigate the feasibility of using physics-informed learning for observation-limited hydrodynamic inversion, discharge estimation, and bed-topography recovery in controlled benchmark cases.

Study Configuration

• Spatial Scale: River flows over complex bed topography.
• Temporal Scale: Short-term to medium-term river flow dynamics.

Methodology and Data

• Models used: Serre–Green–Naghdi (SGN) equations, physics-informed neural network (PINN).
• Data sources: Water-depth or bottom-pressure-head observations, discretized bed-elevation data.

Main Results

• The proposed framework can reliably estimate discharge and reconstruct hydraulic states under limited observation conditions.
• In scenarios with sparse bed information, the method successfully recovers bed topography while maintaining accurate discharge estimation and consistent hydraulic state reconstruction.

Contributions

• This study demonstrates the feasibility of using SGN-constrained physics-informed learning for observation-limited hydrodynamic inversion, discharge estimation, and bed-topography recovery in controlled benchmark cases.
• The proposed approach provides a new framework for estimating river discharge and internal hydraulic states even with limited or no direct measurements available.

Funding

• This research was funded by the State Key Laboratory of Coastal and Offshore Engineering (Grant No. SKLCOE2023001).

Citation

@article{Ai2026Physicsinformed,
  author = {Ai, Congfang and Ma, Yuxiang and Li, Junzheng and Song, Jiajie and Sun, Jiaxiang},
  title = {Physics-informed neural networks for hydrodynamic inversion and discharge estimation in river flows over complex bed topography},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136463},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136463}
}