Zhang et al. (2025) Machine Learning Prediction of River Freeze-Up Dates Under Human Interventions: Insights from the Ningxia–Inner Mongolia Reach of the Yellow River

Identification

• Journal: Water
• Year: 2025
• Date: 2025-11-24
• Authors: Lu Zhang, Suyu Liu, M. Fan, Dongling Chen, Ze Yuan, Xiuwei Zhang
• DOI: 10.3390/w17233357

Research Groups

• Hydrology Bureau of Yellow River Conservancy Commission, Zhengzhou, China
• School of Civil Engineering, Sun Yat-sen University, Zhuhai, China
• School of Computer Science and Technology, Northwestern Polytechnical University, Xi’an, China
• National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology, Xi’an, China

Short Summary

This study developed a systematic machine learning framework to predict river freeze-up dates in the Ningxia–Inner Mongolia reach of the Yellow River, explicitly incorporating stage-specific human interventions. It found that tailored predictor selection, hyperparameter optimization, and a stage-specific cumulative temperature predictor significantly improved accuracy, with XGBoost demonstrating the best overall performance (Mean Absolute Error = 2.95 days).

Objective

• To address the challenges posed by the increasing complexity of ice process evolution driven by human interventions, develop a robust forecasting model for Yellow River freeze-up dates, and provide more reliable technical support for ice disaster control, reservoir operation, and early warning.

Study Configuration

• Spatial Scale: Ningxia–Inner Mongolia reach of the Yellow River (NIMRYR), stretching 1203.8 km. Key locations include Toudaoguai Hydrological Station, Baotou Meteorological Station, and Bayangaole station.
• Temporal Scale: 1960–2024 (65 years) of observational data. Training set: 1960–2020 (61 years). Testing set: 2021–2024 (4 years).

Methodology and Data

• Models used: Multiple Linear Regression (MLR), Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), Multilayer Perceptron (MLP).
• Data sources: Long-term observation records (1960–2024) from the Hydrological Bureau of the Yellow River Conservancy Commission, including:
  • Ice conditions: Freeze-up date, drift-ice date.
  • Hydrological conditions: Daily flow data (Toudaoguai Hydrological Station), bankfull discharge.
  • Meteorological observations (Baotou Meteorological Station): Cumulative temperature indices, extreme value indices.

Main Results

• Predictor selection improved prediction accuracy by 7.6–23% compared to using all candidate predictors.
• Hyperparameter optimization further enhanced model accuracy by 4.5–46%.
• Incorporating a "Stage-Specific Predictor" (redefined cumulative temperature thresholds based on reservoir operation stages) improved model accuracy by 10–22%, proving more effective than simply discarding early data.
• During the 2021–2024 test period, optimal prediction errors were 0.16 days (MLP), -0.99 days (XGBoost), -7.61 days (MLP), and 0.07 days (XGBoost) for each respective year.
• XGBoost achieved the best overall performance with a Mean Absolute Error (MAE) of 2.95 days across the entire dataset.
• SVR demonstrated the best prediction stability with the narrowest average 90% confidence interval width (9.34 days).
• MLP performed best for late freeze-ups and in complex years (e.g., 2023), suggesting better handling of non-meteorological factors.
• Human interventions, particularly intensive multi-reservoir regulation (2014–2024), generally increased prediction errors for most models (2.05–2.85 days).

Contributions

• Established a systematic machine learning framework for Yellow River freeze-up date prediction, addressing previous limitations in model design and human activity consideration.
• Demonstrated the critical importance of tailored predictor selection and hyperparameter optimization for different machine learning models in hydrological forecasting.
• Introduced and validated a novel "Stage-Specific Predictor" approach to effectively integrate the evolving impact of human interventions (reservoir operations) on river ice processes.
• Provided a scientific basis for operational freeze-up prediction in the Yellow River basin and enhanced the understanding of freeze-up mechanisms in seasonally ice-covered rivers under significant anthropogenic influence.

Funding

• National Natural Science Foundation of China [Grant U2243221]
• Natural Science Basic Research Program of Shaanxi [Grant 2024JC-DXWT-07]

Citation

@article{Zhang2025Machine,
  author = {Zhang, Lu and Liu, Suyu and Fan, M. and Chen, Dongling and Yuan, Ze and Zhang, Xiuwei},
  title = {Machine Learning Prediction of River Freeze-Up Dates Under Human Interventions: Insights from the Ningxia–Inner Mongolia Reach of the Yellow River},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17233357},
  url = {https://doi.org/10.3390/w17233357}
}