Li et al. (2026) Physics-informed operator learning for rapid hydrograph-to-field prediction on an unstructured-mesh flood benchmark

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-07-15
• Authors: Yichen Li, Qiang Ye, Shanliang Zhang, Tiechui Liu, Lingzao Zeng
• DOI: 10.1016/j.jhydrol.2026.136017

Research Groups

• State Key Laboratory of Soil Pollution Control and Safety, Zhejiang University
• College of Environmental & Resource Sciences, Zhejiang University
• Powerchina Huadong Engineering Corporation Limited

Short Summary

This study develops a physics-informed Deep Operator Network (PI-DeepONet) for rapid hydrograph-to-field prediction on an unstructured-mesh flood benchmark. The model shows improved transient routing behavior and reduced localized errors compared to a data-driven baseline.

Objective

• Develop a physics-regularized operator learning strategy for rapid multi-scenario screening under fixed-domain flood-model settings

Study Configuration

• Spatial Scale: 9.27 km2 sub-basin of the Qingshui River Basin, China
• Temporal Scale: Steady and unsteady benchmark conditions

Methodology and Data

• Models used: PI-DeepONet (physics-informed Deep Operator Network)
• Data sources: HEC-RAS v6.6 simulations, high-resolution shallow-water equations

Main Results

• Improved transient routing behavior under steady and unsteady benchmark conditions
• Reduced localized errors near wet–dry fronts compared to a data-driven baseline
• Lower event-scale volume error relative to the baseline under an extreme event scenario

Contributions

• This study introduces physics-regularized operator learning as a viable benchmark-specific surrogate strategy for rapid multi-scenario screening under fixed-domain flood-model settings.

Funding

• This research was supported by [project/program name and reference code]

Citation

@article{Li2026Physicsinformed,
  author = {Li, Yichen and Ye, Qiang and Zhang, Shanliang and Liu, Tiechui and Zeng, Lingzao},
  title = {Physics-informed operator learning for rapid hydrograph-to-field prediction on an unstructured-mesh flood benchmark},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136017},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136017}
}