Hashemi et al. (2026) High-resolution L-band VOD retrieval using GNSS-T–calibrated multi-sensor fusion and self-supervised learning

Identification

• Journal: Remote Sensing of Environment
• Year: 2026
• Date: 2026-07-01
• Authors: Mahya G.Z. Hashemi, John Bolten, Alexandra G. Konings, Vincent Humphrey, Andrew F. Feldman, Yitong Yao, Michael Niederberger, Alexander Damm
• DOI: 10.1016/j.rse.2026.115554

Research Groups

• Hydrological Science Laboratory, NASA Goddard Space Flight Center
• Department of Earth System Science, Stanford University
• Department of Geography, University of Zürich
• Federal Office of Meteorology and Climatology MeteoSwiss
• Biospheric Sciences Laboratory, NASA Goddard Space Flight Center
• Earth System Science Interdisciplinary Center, University of Maryland
• Department of Earth and Environmental Engineering, Columbia University
• Eawag, Swiss Federal Institute of Aquatic Science & Technology

Short Summary

This study develops a framework for estimating forest Global Navigation Satellite System Transmissometry (GNSS-T)-calibrated Vegetation Optical Depth (VOD) at 30 m resolution by fusing Sentinel-1A Synthetic Aperture Radar (SAR), Harmonized Landsat–Sentinel-2 optical reflectance, GEDI lidar canopy structure, and climate variables. The framework achieves high accuracy in predicting VOD using a self-supervised learning approach based on Vision Transformers.

Objective

• Estimate forest GNSS-T-calibrated VOD at 30 m resolution by fusing multi-modal satellite data and in situ GNSS-T VOD measurements.
• Develop a novel approach for monitoring vegetation water status and ecosystem dynamics at high spatial and temporal resolutions.

Study Configuration

• Spatial Scale: Local to regional scale, with study sites located in the conterminous United States and Switzerland.
• Temporal Scale: Sub-hourly to daily resolution, with GNSS-T VOD measurements available every 1–3 hours and satellite data acquired at 6–12 day revisit intervals.

Methodology and Data

• Models used: Self-supervised learning framework based on Vision Transformers (ViT) encoders pretrained on unlabeled SAR and optical imagery.
• Data sources:
  • Sentinel-1A C-band SAR data from the Alaska Satellite Facility (ASF)
  • Harmonized Landsat–Sentinel-2 (HLS) optical surface reflectance data
  • GEDI lidar canopy structure metrics (RH50, RH75, RH95)
  • Climate variables (precipitation and air temperature aggregated over same-day, 7-, 30-, and 60-day windows)

Main Results

• The framework achieves high accuracy in predicting VOD using a self-supervised learning approach based on Vision Transformers.
• Feature importance analysis reveals that medium-term climate drivers, particularly 60-day precipitation and 7-day temperature, consistently outrank SAR backscatter.
• The framework demonstrates robust temporal generalization across deciduous and evergreen forest types.

Contributions

• This study establishes GNSS-T-calibrated multi-sensor fusion with domain-specific self-supervised learning pretraining as a viable pathway toward operational 30 m L-band VOD estimation.
• The framework provides a novel approach for monitoring vegetation water status and ecosystem dynamics at high spatial and temporal resolutions.

Funding

• NASA Goddard Space Flight Center (Grant No. NNX17AC93G)
• National Science Foundation (Grant No. EAR-1552327)
• Swiss National Science Foundation (Grant No. 200020_172596)

Citation

@article{Hashemi2026Highresolution,
  author = {Hashemi, Mahya G.Z. and Bolten, John and Konings, Alexandra G. and Holmes, Thomas R. and Humphrey, Vincent and Feldman, Andrew F. and Yao, Yitong and Niederberger, Michael and Damm, Alexander},
  title = {High-resolution L-band VOD retrieval using GNSS-T–calibrated multi-sensor fusion and self-supervised learning},
  journal = {Remote Sensing of Environment},
  year = {2026},
  doi = {10.1016/j.rse.2026.115554},
  url = {https://doi.org/10.1016/j.rse.2026.115554}
}