Yan et al. (2026) Advances in coupling machine learning with hydrological simulation: A review

Identification

• Journal: Water Science and Engineering
• Year: 2026
• Authors: Yu-fei Yan, Han-xiao Liu, Shu Qiong Xu, Qiong-lin Wang, Yu-hui Yang, Qingqing Chen, Chen-yang Wang, Tian-ling Qin
• DOI: 10.1016/j.wse.2026.01.002

Research Groups

• College of Resource Environment and Tourism, Capital Normal University, Beijing, China
• State Key Laboratory of Water Cycle and Water Security, China Institute of Water Resources and Hydropower Research (IWHR), Beijing, China
• School of Water Conservancy and Transportation, Zhengzhou University, Zhengzhou, China
• Henan Key Laboratory of Yellow Basin Ecological Protection and Restoration, Yellow River Institute of Hydraulic Research, Zhengzhou, China
• College of Hydrology and Water Resources, Hohai University, Nanjing, China

Short Summary

This review systematically synthesizes the evolution of hydrological modeling from traditional physical frameworks to data-driven machine learning (ML) approaches. It establishes that coupling physically-based mechanisms with ML architectures is the most effective pathway to enhance predictive accuracy, computational efficiency, and interpretability.

Objective

• To evaluate the evolutionary trajectory of hydrological simulation and identify the fundamental bottlenecks of traditional and ML-only models.
• To categorize and analyze contemporary paradigms for coupling physical hydrological models with machine learning.

Study Configuration

• Spatial Scale: Multi-scale review, covering applications from small catchments to large-scale global basins.
• Temporal Scale: Historical review of modeling advances from the 1930s to the present, focusing on the rapid development of ML since the 1990s.

Methodology and Data

• Models used:
  • Traditional: Xinanjiang, VIC, SWAT, TOPMODEL, HBV, SHE, WEP, and SAC models.
  • Machine Learning: ANN, SVM, Random Forests (RF).
  • Deep Learning: LSTM, CNN, RNN, and Transformer models.
  • Coupled Frameworks: Sequential coupling, Deep integration, and Physics-informed ML (PIML).
• Data sources: Synthesis of literature utilizing multi-source data including satellite remote sensing, ground-based observations, and reanalysis datasets.

Main Results

• Evolutionary Stages: Hydrological modeling has progressed through four ML phases: initial exploration (ANN/SVM), development (RF/multivariate data), deep learning (LSTM/CNN/Transformers), and the current stage of hybrid ML models.
• Coupling Paradigms: Three distinct patterns were identified:
  1. Sequential Coupling: ML acts as a post-processor to correct residuals and systematic errors from physical models.
  2. Deep Integration: ML modules replace computationally intensive sub-processes or dynamically optimize parameters within a physical framework.
  3. Physics-informed ML: Physical laws (e.g., mass conservation) are embedded directly into ML loss functions or architectures.
• Performance Gains: Hybrid models demonstrate superior performance in simulating extreme events (floods/droughts) and baseflow compared to standalone paradigms.
• Identified Bottlenecks: Current challenges include high data dependency, shallow bidirectional integration, and poor cross-basin generalization (regionalization).

Contributions

• Provides a rigorous classification of the degree of integration between physical mechanisms and data-driven techniques.
• Identifies the transition from "black-box" ML to "differentiable" and "physics-guided" architectures as a critical advancement.
• Proposes future research directions focusing on bidirectional feedback mechanisms, transfer learning for data-scarce regions, and the construction of hydrological knowledge graphs.

Funding

• National Key Research and Development Project (Grant No. 2022YFC3201700)
• National Natural Science Foundation of China (Grant No. U2443205)
• Anhui Provincial Key Research and Development Project (Grant No. 2408055US002)
• Five Major Excellent Talent Programs of the China Institute of Water Resources and Hydropower Research (IWHR) (Grant No. WR0199A012021)

Citation

@article{Yan2026Advances,
  author = {Yan, Yu-fei and Liu, Han-xiao and Xu, Shu Qiong and Wang, Qiong-lin and Yang, Yu-hui and Chen, Qingqing and Wang, Chen-yang and Qin, Tian-ling},
  title = {Advances in coupling machine learning with hydrological simulation: A review},
  journal = {Water Science and Engineering},
  year = {2026},
  doi = {10.1016/j.wse.2026.01.002},
  url = {https://doi.org/10.1016/j.wse.2026.01.002}
}

Generated by BiblioAssistant using gemini-3-flash-preview (Google API)