Ma et al. (2025) Runoff Forecast Model Integrating Time Series Decomposition and Deep Learning for the Short Term: A Case Study in the Weihe River Basin, China

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water
• Year: 2025
• Date: 2025-09-14
• Authors: R. Ma, Qiang An, Liu Liu, Yongming Cheng, Xingcai Liu
• DOI: 10.3390/w17182718

Research Groups

Not explicitly mentioned in the paper text, but the study focuses on Huaxian Station in China’s Weihe River Basin, suggesting involvement of research institutions or departments specializing in hydrology or water resources in China.

Short Summary

This paper introduces a novel framework integrating segmented decomposition sampling with a multi-input neural network to address forward data contamination in decomposition-based runoff prediction models, demonstrating improved accuracy and reliability for daily runoff estimation.

Objective

• To develop a novel framework for daily river runoff prediction that effectively avoids future data contamination in decomposition-based hybrid models while simultaneously improving prediction accuracy and computational efficiency.

Study Configuration

• Spatial Scale: Huaxian Station, Weihe River Basin, China.
• Temporal Scale: Daily runoff estimation, with predictions for 1 to 3 days ahead.

Methodology and Data

• Models used: Seasonal-Trend decomposition using Loess (STL), Convolutional Long Short-Term Memory (CNN-LSTM) network, multi-input neural network. Comparative models included VMD-CNN-LSTM and standalone LSTM.
• Data sources: Historical daily runoff data from Huaxian Station.

Main Results

• The proposed STL-CNN-LSTM framework effectively avoids future information leakage and improves prediction accuracy in daily runoff forecasting.
• For 1-day, 2-day, and 3-day ahead predictions at Huaxian Station, the STL-CNN-LSTM model achieved Nash-Sutcliffe efficiency (NSE) values of 0.96, 0.83, and 0.80, respectively.
• These NSE values represent improvements of 5.49%, 5.06%, and 12.68% over the VMD-CNN-LSTM model for 1-day, 2-day, and 3-day ahead predictions, respectively.
• The STL-based configuration also outperformed the standalone LSTM counterpart by 23.08%, 9.21%, and 17.65% in NSE for 1-day, 2-day, and 3-day ahead predictions, respectively.

Contributions

• Introduction of a novel framework integrating segmented decomposition sampling with a multi-input neural network to address the critical issue of forward data contamination in decomposition-based hybrid forecasting models.
• Development of a computationally efficient approach that avoids the high costs associated with stepwise decomposition methods.
• Demonstration of significant improvements in daily river runoff prediction accuracy and reliability compared to established single and hybrid models.

Funding

Not provided in the paper text.

Citation

@article{Ma2025Runoff,
  author = {Ma, R. and An, Qiang and Liu, Liu and Cheng, Yongming and Liu, Xingcai},
  title = {Runoff Forecast Model Integrating Time Series Decomposition and Deep Learning for the Short Term: A Case Study in the Weihe River Basin, China},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17182718},
  url = {https://doi.org/10.3390/w17182718}
}