Hassanien et al. (2026) Coupling SWAT with recurrent and convolutional deep learning architectures for catchment-scale streamflow simulation and interpretability in the Upper Blue Nile River Basin

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-09-11
• Authors: Mahmoud M. Hassanien, Yangwen Jia, Chunfeng Hao, Yaqin Qiu, Xin Chen, Hussein A. Mohasseb, Mustafa Shaaban Ali, Shambel Yideg Arega
• DOI: 10.1016/j.ejrh.2026.103966

Research Groups

• State Key Laboratory of Water Cycle and Water Security, China Institute of Water Resources and Hydropower Research, Beijing, 100038, China
• Egyptian Ministry of Water Resources and Irrigation, Imbaba, 12666, Egypt
• State Key Laboratory of Earth System Numerical Modeling and Application, College of Earth and Planetary Sciences, University of Chinese Academy of Sciences, Beijing, 101408, China
• Environment and Climate Changes Research Institute, National Water Research Center, Cairo, Egypt
• Hydraulic and Water Resources Engineering Department, Debre Markos University Institute of Technology, Debre Markos, Ethiopia

Short Summary

This study coupled the Soil and Water Assessment Tool (SWAT) with three deep learning architectures LSTM, GRU, and TCN to improve streamflow simulation in the Upper Blue Nile River Basin. The results show that coupling benefits were station and metric dependent rather than universal.

Objective

• Compare the predictive performance of SWAT and three DL architectures during a common independent evaluation period.
• Determine whether the benefits of SWAT and DL coupling are consistent across architectures, stations, and performance metrics.
• Characterize STCN attribution patterns while explicitly evaluating predictor dependence, temporal attribution stability, and group level predictive reliance.

Study Configuration

• Spatial Scale: The Upper Blue Nile River Basin (UBNRB), Ethiopia.
• Temporal Scale: Monthly data from 2001 to 2014 were used for training, internal validation, and independent testing.

Methodology and Data

• Models used: LSTM, GRU, TCN, and their coupled counterparts with SWAT.
• Data sources: Meteorological data (precipitation, temperature, etc.) from the National Meteorological Agency of Ethiopia, streamflow records from the Ethiopian Ministry of Water and Energy, Digital Elevation Model (DEM), land use land cover (LULC) data, and soil data.

Main Results

• The results show that STCN achieved RMSE/NSE of 590.4 m³ s⁻¹ /0.921 at Al-Diem, compared with 723.8/0.881 for SWAT.
• At Kessie, STCN achieved 292.0/0.894 versus 294.1/0.893 for SWAT.
• The study found that coupling benefits were station and metric dependent rather than universal.

Contributions

• This study provides a controlled comparison of LSTM-, GRU-, and TCN-based standalone and SWAT-coupled configurations at two hydrologically contrasting stations, followed by a robustness-aware comparison of station-specific model attribution.
• The results highlight the importance of considering predictor dependence and temporal attribution stability when interpreting SHAP values.

Funding

• This research was funded by the State Key Laboratory of Water Cycle and Water Security, China Institute of Water Resources and Hydropower Research.

Citation

@article{Hassanien2026Coupling,
  author = {Hassanien, Mahmoud M. and Jia, Yangwen and Hao, Chunfeng and Qiu, Yaqin and Chen, Xin and Mohasseb, Hussein A. and Ali, Mustafa Shaaban and Arega, Shambel Yideg},
  title = {Coupling SWAT with recurrent and convolutional deep learning architectures for catchment-scale streamflow simulation and interpretability in the Upper Blue Nile River Basin},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2026.103966},
  url = {https://doi.org/10.1016/j.ejrh.2026.103966}
}