Mobarrat et al. (2026) Beyond Traditional Hydrological Models: Transformer‐Based Deep Learning for Future Streamflow and Flood Risk Assessment in the Brahmaputra River Basin

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

Identification

• Journal: International Journal of Climatology
• Year: 2026
• Date: 2026-09-18
• Authors: Md Mahin Mobarrat, Himel Moulik, Md. Mostafa Ali
• DOI: 10.1002/joc.70596

Research Groups

• Department of Hydrology, University of Dhaka
• Climate Change and Water Resources Laboratory, Bangladesh University of Engineering and Technology
• Institute of Atmospheric Sciences, Chinese Academy of Meteorological Sciences

Short Summary

This study presents an integrated framework combining process-based hydrological modelling with deep learning techniques to improve climate-driven streamflow prediction and flood risk assessment in the Brahmaputra River Basin. The Transformer model demonstrated superior predictive capability compared to traditional hydrological models.

Objective

• Investigate the added value of Transformer-based architectures against traditional hydrological modelling (HEC-HMS) and recurrent neural networks (LSTM and BiLSTM) for climate-driven streamflow prediction and flood risk assessment in the Brahmaputra River Basin.

Study Configuration

• Spatial Scale: The study focuses on the Brahmaputra River Basin at Bahadurabad, with a spatial scale of approximately 10,000 km².
• Temporal Scale: Historical discharge data from 1981 to 2014 were used for calibration and validation, while future projections were made for the 2030s, 2050s, and 2080s.

Methodology and Data

• Models used:
  • HEC-HMS (traditional hydrological modelling)
  • LSTM (recurrent neural network)
  • BiLSTM (bidirectional recurrent neural network)
  • Transformer (deep learning architecture)
• Data sources: Historical discharge data from 1981 to 2014, bias-corrected CMIP6 climate projections

Main Results

• The Transformer model demonstrated superior predictive capability, achieving an R² of 0.94 during testing with PBIAS within ±1%.
• HEC-HMS showed comparatively lower accuracy (R² = 0.72) and consistent underestimation exceeding 20%.
• Future projections reveal a systematic shift towards earlier monsoon onset and a 10%–25% increase in seasonal discharge.
• Mean annual flow is expected to rise from approximately 22,000 m³/s in the baseline period to nearly 30,000 m³/s by the late century.

Contributions

This study provides an original contribution to existing literature by evaluating the added value of Transformer-based architectures against traditional hydrological modelling and recurrent neural networks for climate-driven streamflow prediction and flood risk assessment. The findings highlight the enhanced capability of Transformer-based models to capture complex hydroclimatic dynamics.

Funding

• This research was funded by the Bangladesh University of Engineering and Technology (BUET) Research Fund (grant number: BUET-RF-2020-01)
• Supported by the National Science Foundation, China (grant number: 41875044)
• Partially supported by the International Atomic Energy Agency (IAEA) Coordinated Research Project on "Climate Change Impacts on Water Resources"

Citation

@article{Mobarrat2026Beyond,
  author = {Mobarrat, Md Mahin and Moulik, Himel and Ali, Md. Mostafa},
  title = {Beyond Traditional Hydrological Models: Transformer‐Based Deep Learning for Future Streamflow and Flood Risk Assessment in the Brahmaputra River Basin},
  journal = {International Journal of Climatology},
  year = {2026},
  doi = {10.1002/joc.70596},
  url = {https://doi.org/10.1002/joc.70596}
}