Chen et al. (2025) A physics-informed neural network for predicting depth-averaged velocities of flows with submerged vegetation: Integrating analytical formulas with deep learning

Identification

• Journal: Journal of Hydrology
• Year: 2025
• Date: 2025-12-17
• Authors: Z. Chen, Feng Luo, Yuan Li, Aifeng Tao, Chi Zhang, Jinhai Zheng
• DOI: 10.1016/j.jhydrol.2025.134801

Research Groups

• Key Laboratory of Ministry of Education for Coastal Disaster and Protection, Hohai University, Nanjing 210098, China
• Institute of Water Science and Technology, Hohai University, Nanjing 210098, China

Short Summary

This paper presents a physics-informed neural network (PINN) that integrates an algebraic momentum-balance relation with a data-driven framework to predict depth-averaged velocities in flows with submerged vegetation, demonstrating superior accuracy, generalization, and robustness compared to traditional analytical formulas and purely data-driven models.

Objective

• To develop and validate a physics-informed neural network (PINN) for accurately predicting depth-averaged velocities in flows with submerged vegetation, overcoming limitations of traditional analytical formulations and purely data-driven approaches.

Study Configuration

• Spatial Scale: Laboratory or experimental flume scale, focusing on local flow characteristics around submerged vegetation.
• Temporal Scale: Steady-state or quasi-steady-state flow conditions, as depth-averaged velocity is a primary descriptor.

Methodology and Data

• Models used: Physics-informed neural network (PINN) coupled with an algebraic momentum-balance relation.
• Data sources: Experimental datasets spanning a broad range of vegetation and flow conditions.

Main Results

• The proposed PINN demonstrates superior generalization and robustness, particularly under limited data availability, by enforcing physical constraints and reducing overfitting.
• It achieves higher predictive accuracy compared to several widely used traditional analytical formulas.
• The PINN produces physically consistent predictions across varied vegetation and flow conditions, retaining interpretability.

Contributions

• Introduces a novel physics-informed deep learning approach that effectively couples mechanistic physical laws with data-driven learning for ecohydraulic modeling.
• Provides an accurate, interpretable, and generalizable tool for modeling vegetation–flow interactions, extending classical hydraulic theory.
• Addresses the limitations of traditional analytical formulas (simplified assumptions, empirical parameterizations) and purely data-driven models (overfitting, poor generalization) in predicting depth-averaged velocities in complex vegetated flows.

Funding

• Not specified in the provided text.

Citation

@article{Chen2025physicsinformed,
  author = {Chen, Z. and Luo, Feng and Li, Yuan and Tao, Aifeng and Zhang, Chi and Zheng, Jinhai},
  title = {A physics-informed neural network for predicting depth-averaged velocities of flows with submerged vegetation: Integrating analytical formulas with deep learning},
  journal = {Journal of Hydrology},
  year = {2025},
  doi = {10.1016/j.jhydrol.2025.134801},
  url = {https://doi.org/10.1016/j.jhydrol.2025.134801}
}