Xiao et al. (2026) Sentinel-1 SAR time series for forest phenology: multimodal deep learning with meteorological fusion across deciduous broadleaf and evergreen needleleaf forests

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

Identification

• Journal: Agricultural and Forest Meteorology
• Year: 2026
• Date: 2026-07-25
• Authors: Yi Xiao, Frédéric Frappart, Yongshuo H. Fu, Jean‐Pierre Wigneron, Marc Peaucelle
• DOI: 10.1016/j.agrformet.2026.111373

Research Groups

• Department of Forest Ecology and Management, Swedish University of Agricultural Sciences (SLU)
• Department of Physical Geography, Stockholm University
• NASA Goddard Space Flight Center

Short Summary

This study develops a multimodal deep learning framework that fuses Sentinel-1 time series with meteorological variables to estimate forest phenology in deciduous broadleaf and evergreen needleleaf forests. The results show improved retrievals compared to conventional methods, particularly when integrating meteorological information.

Objective

• Investigate the potential of synthetic aperture radar (SAR) for retrieving forest phenology across different forest types.
• Develop a multimodal deep learning framework that combines SAR data with meteorological variables to improve phenology estimation accuracy.

Study Configuration

• Spatial Scale: North America and Europe, covering various forest ecosystems.
• Temporal Scale: Annual time series of Sentinel-1 observations from 2016 to 2020.

Methodology and Data

• Models used: Convolutional Neural Network (CNN), Temporal Convolutional Network (TCN), Transformer, U-Net
• Data sources: Sentinel-1 SAR data, PhenoCam-derived reference dates, meteorological variables (temperature, soil moisture, radiation, precipitation)

Main Results

• The multimodal deep learning framework significantly improved phenology retrievals compared to conventional threshold methods.
• In deciduous broadleaf forests, the best model achieved RMSEs of 9.5 d for SOS and 10.8 d for EOS under leave-one-year-out temporal cross validation.
• Meteorological fusion further enhanced accuracy in DB forests, while improvements were more modest in evergreen needleleaf forests.

Contributions

• This study provides a novel approach to retrieving forest phenology using SAR data and deep learning techniques.
• The results demonstrate the potential of multimodal deep learning for improving phenology estimation accuracy, particularly when integrating meteorological information.

Funding

• NASA's Terrestrial Hydrology Program (NNH17ZDA001N-THP)
• Swedish Research Council (2016-06573)

Citation

@article{Xiao2026Sentinel1,
  author = {Xiao, Yi and Frappart, Frédéric and Fu, Yongshuo H. and Wigneron, Jean‐Pierre and Peaucelle, Marc},
  title = {Sentinel-1 SAR time series for forest phenology: multimodal deep learning with meteorological fusion across deciduous broadleaf and evergreen needleleaf forests},
  journal = {Agricultural and Forest Meteorology},
  year = {2026},
  doi = {10.1016/j.agrformet.2026.111373},
  url = {https://doi.org/10.1016/j.agrformet.2026.111373}
}