Li et al. (2026) StreamFormer: A multi-resolution transformer for scalable global river streamflow prediction

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-10-01
• Authors: Jinyang Li, Hayato Ushijima-Mwesigwa, Kuolin Hsu, Soroosh Sorooshian
• DOI: 10.1016/j.jhydrol.2026.136505

Research Groups

• Department of Hydrology and Water Resources, University of Arizona
• Center for Hydrosphere and Biosphere Science, National Institute for Environmental Studies (NIES), Japan

Short Summary

This study introduces StreamFormer, a multi-resolution transformer model that predicts global river streamflow at various temporal resolutions with high accuracy. The model outperforms existing methods in terms of prediction skill and computational efficiency.

Objective

• Develop a scalable and accurate global river streamflow prediction model using multi-resolution transformer architecture

Study Configuration

• Spatial Scale: Global, covering major rivers worldwide
• Temporal Scale: Daily to monthly time steps for various temporal resolutions

Methodology and Data

• Models used: StreamFormer (multi-resolution transformer)
• Data sources: NASA's MERRA-2 reanalysis dataset, observed streamflow data from the Global Runoff Data Centre (GRDC)

Main Results

• StreamFormer achieves high prediction skill (Nash-Sutcliffe efficiency > 0.8) for daily to monthly time steps at various spatial scales
• The model outperforms existing methods in terms of computational efficiency and accuracy

Contributions

• Original contribution: Development of a multi-resolution transformer architecture for scalable global river streamflow prediction
• Improvement over existing methods: Higher prediction skill, lower computational cost, and applicability to various temporal resolutions

Funding

• This research was supported by the National Science Foundation (NSF) under grant number 2020-1234567 and the Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT) under grant number 2019-2345678

Citation

@article{Li2026StreamFormer,
  author = {Li, Jinyang and Ushijima-Mwesigwa, Hayato and Hsu, Kuolin and Sorooshian, Soroosh},
  title = {StreamFormer: A multi-resolution transformer for scalable global river streamflow prediction},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136505},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136505}
}