Li et al. (2026) Physics-guided Kolmogorov–Arnold Network for extreme flood and drought flow prediction: A case study of the Upper Hanjiang River Basin

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-09-08
• Authors: Xiaodong Li, Jieliang Guo, Yicong Dai, Tiesheng Guan, Xin Yin, Feng Wang
• DOI: 10.1016/j.ejrh.2026.103895

Research Groups

• College of Computer Science and Software Engineering, Hohai University, Nanjing 211100, China
• Bureau of Hydrology (Information Centre) of Taihu Basin Authority, Shanghai 200434, China
• Nanjing Hydraulic Research Institute, Nanjing 210029, China
• Nanjing Automation Institute of Water Conservancy and Hydrology, Ministry of Water Resources, China
• School of Computer Science, Wuhan University, Wuhan 430072, China

Short Summary

This study proposes a novel deep learning framework, Deep Process Learning-Long Short-term Kolmogorov–Arnold Network (DPL-LSTKAN), for predicting extreme flood and drought flows in the Upper Hanjiang River Basin. The model integrates physical knowledge into data-driven methods to improve generalization and interpretability.

Objective

• Investigate the feasibility of using a deep learning framework to predict extreme hydrological events.
• Develop a physics-guided gating mechanism that dynamically assigns weights to different flow regimes based on dominant hydrological processes.

Study Configuration

• Spatial Scale: The study focuses on the Upper Hanjiang River Basin (UHRB), with a drainage area of 35,347 km2.
• Temporal Scale: Daily time scale, with data covering from August 1980 to December 2018.

Methodology and Data

• Models used: Deep Process Learning-Long Short-term Kolmogorov–Arnold Network (DPL-LSTKAN)
• Data sources: Meteorological forcing data from nine stations within the watershed, daily runoff observations at the Ankang Hydrological Station

Main Results

• The proposed DPL-LSTKAN model outperforms four baseline models in predicting extreme flood and drought flows.
• The physics-guided gating mechanism effectively assigns weights to different flow regimes based on dominant hydrological processes.

Contributions

• This study contributes to the development of a novel deep learning framework that integrates physical knowledge into data-driven methods for predicting extreme hydrological events.
• The proposed model improves generalization and interpretability by dynamically assigning weights to different flow regimes based on dominant hydrological processes.

Funding

• This research was funded by the National Natural Science Foundation of China (Grant No. 51979224), the Jiangsu Provincial Key Research and Development Program (Grant No. BE2019070100).

Citation

@article{Li2026Physicsguided,
  author = {Li, Xiaodong and Guo, Jieliang and Dai, Yicong and Guan, Tiesheng and Yin, Xin and Wang, Feng},
  title = {Physics-guided Kolmogorov–Arnold Network for extreme flood and drought flow prediction: A case study of the Upper Hanjiang River Basin},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2026.103895},
  url = {https://doi.org/10.1016/j.ejrh.2026.103895}
}