Wang et al. (2026) A deep learning model for runoff prediction via Adaptive Reinforced Feature Mining

Identification

• Journal: Applied Soft Computing
• Year: 2026
• Date: 2026-07-17
• Authors: Wenchuan Wang, H. Ren, Wei-can Tian, Qi-Qi Zeng
• DOI: 10.1016/j.asoc.2026.115991

Research Groups

• College of Water Resources, North China University of Water Resources and Electric Power

Short Summary

This study proposes a deep learning model called DRRA (Deep Learning Model for Runoff Prediction via Adaptive Reinforced Feature Mining) that improves daily runoff prediction accuracy by adaptively extracting global and local features from runoff series.

Objective

• Investigate the effectiveness of a novel deep learning architecture in predicting daily runoff, focusing on capturing long-term trends and short-term fluctuations.

Study Configuration

• Spatial Scale: River basin scale
• Temporal Scale: Daily time step

Methodology and Data

• Models used: DRRA (Deep Learning Model for Runoff Prediction via Adaptive Reinforced Feature Mining)
• Data sources: Three river basins with distinct climatic and hydrological conditions, including satellite and observation data.

Main Results

• The DRRA model significantly reduces prediction errors and improves fitting performance compared to existing models.
• The Kling–Gupta efficiency reached 0.9868, 0.9698, and 0.9343 for the three basins, respectively.
• Peak flow prediction error was reduced by approximately 25.4% on average.

Contributions

• This study introduces a novel deep learning architecture that improves daily runoff prediction accuracy by adaptively extracting global and local features from runoff series.
• The proposed model enhances both overall prediction accuracy and peak flow prediction, providing reliable support for flood forecasting and water resource management.

Funding

• This research was funded by the National Natural Science Foundation of China (Grant No. 52109201) and the Key Research and Development Program of Henan Province (Grant No. 212102310104).

Citation

@article{Wang2026deep,
  author = {Wang, Wenchuan and Ren, H. and Tian, Wei-can and Zhang, Xu-tong and Zeng, Qi-Qi},
  title = {A deep learning model for runoff prediction via Adaptive Reinforced Feature Mining},
  journal = {Applied Soft Computing},
  year = {2026},
  doi = {10.1016/j.asoc.2026.115991},
  url = {https://doi.org/10.1016/j.asoc.2026.115991}
}