Wu et al. (2025) Hybrid Variational Modal Decomposition-Extreme Learning Machine-Adaptive Boosting Model for Monthly Runoff Prediction

Identification

• Journal: Water
• Year: 2025
• Date: 2025-10-31
• Authors: Li Wu, Junfeng Tian, Zhongfeng Jiang, Yong Wang
• DOI: 10.3390/w17213129

Research Groups

• School of Municipal and Environment Engineering, Henan University of Urban Construction, Pingdingshan, China
• State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering, Nanjing Hydraulic Research Institute, Nanjing, China

Short Summary

This study developed a novel hybrid VMD-ELM-AdaBoost model for monthly runoff prediction, demonstrating superior accuracy and efficiency compared to benchmark models, particularly in data-limited basins, by effectively addressing non-stationarity and improving generalization.

Objective

• To develop and validate a novel hybrid Variational Modal Decomposition-Extreme Learning Machine-Adaptive Boosting (VMD-ELM-AdaBoost) model for accurate and efficient monthly runoff prediction, especially in data-scarce basins, by addressing non-stationarity and improving generalization capabilities.

Study Configuration

• Spatial Scale: Two hydrological stations (Yanshan and Baiguishan) located on the Shahe River and Lihe River, respectively, within the Shaying River Basin in Pingdingshan City, Henan Province, China.
• Temporal Scale: Monthly runoff data spanning 67 years (from 1956 to 2022). The dataset was split into 80% for training and 20% for validation.

Methodology and Data

• Models used:
  • Proposed: Hybrid Variational Modal Decomposition-Extreme Learning Machine-Adaptive Boosting (VMD-ELM-AdaBoost)
  • Components: Variational Modal Decomposition (VMD) with Particle Swarm Optimisation (PSO) for parameter tuning, Extreme Learning Machine (ELM), Adaptive Boosting (AdaBoost).
  • Comparative models: VMD-ELM, ELM-AdaBoost, Long Short-Term Memory (LSTM), VMD-TPE-LSTM.
• Data sources: Monthly runoff depth observations from the Baiguishan Hydrological Station and Yanshan Hydrological Station.

Main Results

• The VMD-ELM-AdaBoost model consistently outperformed all comparative models in monthly runoff prediction.
• At Yanshan Station, the model achieved a Root Mean Square Error (RMSE) of 2.521 mm and a Mean Absolute Percentage Error (MAPE) of 8.56%. This represents a 34.8–45.1% lower RMSE compared to VMD-ELM, ELM-AdaBoost, and LSTM.
• At Baiguishan Station, the model yielded an RMSE of 2.906 mm and a MAPE of 9.02%, showing a 22.3–42.6% lower RMSE than VMD-TPE-LSTM and other alternatives.
• The complete VMD-ELM-AdaBoost model demonstrated substantial accuracy gains over the baseline LSTM model (RMSE: 12.3309 mm and 11.2203 mm; MAPE: 19.75% and 18.76%).
• The integration of AdaBoost significantly improved accuracy from VMD-ELM to VMD-ELM-AdaBoost, reducing RMSE by approximately 4.7 mm and 3.3 mm, and MAPE by over 10% at Yanshan and Baiguishan Stations, respectively.
• The model effectively predicts extreme runoff events (>100 mm), providing reliable support for water resource management.

Contributions

• Proposed a novel three-stage decomposition-ensemble-revision hybrid framework (VMD-ELM-AdaBoost) specifically designed to address the dual challenge of non-stationarity and data scarcity in hydrological forecasting.
• Introduced PSO-optimised VMD for adaptive tuning of modal number (K) and penalty factor (α), enhancing decomposition accuracy for runoff data compared to fixed-parameter VMD.
• Integrated an ELM-AdaBoost ensemble for decomposed subsequences, leveraging ELM's data efficiency and AdaBoost's robustness to improve generalization without requiring large datasets, unlike deep learning models.
• Incorporated a lightweight residual feedback correction loop (secondary VMD-ELM prediction if MAPE > 10%) to refine results, a feature rarely included in similar hybrid models.
• Demonstrated high accuracy using only historical runoff data, making it a practical tool for monthly runoff forecasting in ungauged or data-limited basins.

Funding

• National Key Points of Hydrology, Water Resources and Hydraulic Engineering Science Laboratory Open Project Fund (grant number: 2017490611)
• Natural Science Foundation of Henan (grant number: 242300420012)
• The Key Scientific Research Project of Henan Higher Education Institutions (grant number: 24B410001)

Citation

@article{Wu2025Hybrid,
  author = {Wu, Li and Tian, Junfeng and Jiang, Zhongfeng and Wang, Yong},
  title = {Hybrid Variational Modal Decomposition-Extreme Learning Machine-Adaptive Boosting Model for Monthly Runoff Prediction},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17213129},
  url = {https://doi.org/10.3390/w17213129}
}