Liu et al. (2026) Geospatial Foundation Embeddings as Transferable Catchment Descriptors

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

Identification

• Journal: Geophysical Research Letters
• Year: 2026
• Date: 2026-08-12
• Authors: Jiangtao Liu, Chaopeng Shen, Yuan Yang, Haoyu Ji, Kathryn Lawson, Nicholas Kraabel
• DOI: 10.1029/2026gl122814

Research Groups

• Department of Hydrology, University of California, Los Angeles (UCLA)
• Department of Computer Science, University of Oxford

Short Summary

This study evaluates the use of geospatial foundation model embeddings as transferable catchment descriptors for rainfall-runoff modeling, achieving comparable performance to traditional attributes in a global prediction scenario.

Objective

• Investigate the effectiveness of geospatial foundation model embeddings as transferable catchment descriptors for deep learning hydrological models.

Study Configuration

• Spatial Scale: Global (531 CAMELS basins and 3,434 ungauged basins)
• Temporal Scale: Long-term temporal validation (multiple years)

Methodology and Data

• Models used: Deep learning hydrological models with dual-path gated architecture
• Data sources: CAMELS dataset, AlphaEarth and StefaLand geospatial foundation model embeddings

Main Results

• Geospatial foundation model embeddings achieved comparable performance to traditional attributes in rainfall-runoff modeling (median Nash-Sutcliffe Efficiency 0.779 vs. 0.781)
• Combining embeddings with traditional attributes improved median NSE from 0.683 to 0.720 in a global prediction scenario

Contributions

• This study provides evidence for the use of learned embeddings as transferable catchment descriptors, improving the performance and applicability of deep learning hydrological models.

Funding

• Not specified

Citation

@article{Liu2026Geospatial,
  author = {Liu, Jiangtao and Shen, Chaopeng and Yang, Yuan and Ji, Haoyu and Lawson, Kathryn and Kraabel, Nicholas},
  title = {Geospatial Foundation Embeddings as Transferable Catchment Descriptors},
  journal = {Geophysical Research Letters},
  year = {2026},
  doi = {10.1029/2026gl122814},
  url = {https://doi.org/10.1029/2026gl122814}
}