Wang et al. (2025) An Enhanced CycleGAN to Derive Temporally Continuous NDVI from Sentinel-1 SAR Images

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

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-12-24
• Authors: Anqi Wang, Zhiqiang Xiao, Chunyu Zhao, Juan Li, Yunteng Zhang, Jinling Song, Hua Yang
• DOI: 10.3390/rs18010056

Research Groups

[Information not available in the provided text.]

Short Summary

This study developed an enhanced CycleGAN (SA-CycleGAN) to generate high-fidelity, temporally continuous Normalized Difference Vegetation Index (NDVI) from Synthetic Aperture Radar (SAR) imagery, demonstrating superior performance over other unsupervised models in overcoming optical remote sensing data gaps caused by cloud cover.

Objective

• To develop a robust and effective method for deriving a high-fidelity, temporally continuous Normalized Difference Vegetation Index (NDVI) from Synthetic Aperture Radar (SAR) imagery, particularly in cloudy regions where optical data is limited.

Study Configuration

• Spatial Scale: Four distinct sites with varying vegetation types (e.g., Zhangbei, Xishuangbanna), implying a regional or landscape scale.
• Temporal Scale: Time series analysis for continuous monitoring, aiming to overcome data gaps in optical time series.

Methodology and Data

• Models used: Enhanced CycleGAN (SA-CycleGAN) incorporating a spatiotemporal attention generator and a structural similarity (SSIM) loss function. Comparison models included DualGAN, GP-UNIT, and DCLGAN.
• Data sources: Sentinel-1 Synthetic Aperture Radar (SAR) images.

Main Results

• The SA-CycleGAN significantly outperformed the comparison models (DualGAN, GP-UNIT, DCLGAN) across all four evaluation sites.
• Quantitatively, the proposed SA-CycleGAN achieved the lowest Root Mean Square Error (RMSE) of 0.0502 at the Zhangbei site.
• The SA-CycleGAN achieved the highest Coefficient of Determination (R²) of 0.88 at the Xishuangbanna site.
• Ablation experiments confirmed that the novel spatiotemporal attention mechanism was crucial for capturing long-range spatial dependencies, and the SSIM loss function was essential for preserving structural and textural integrity in the synthesized images.

Contributions

• Introduction of SA-CycleGAN, a novel unsupervised deep learning model for SAR-to-optical equivalent translation, specifically for deriving continuous NDVI.
• Development of a spatiotemporal attention generator within CycleGAN to dynamically compute global and local feature relationships, capturing long-range spatial dependencies.
• Integration of a structural similarity (SSIM) loss function to enhance the preservation of structural and textural integrity in synthesized images.
• Provides a robust and effective solution for overcoming data gaps in optical time series caused by frequent cloud cover, addressing a significant limitation in continuous ecological monitoring.

Funding

[Information not available in the provided text.]

Citation

@article{Wang2025Enhanced,
  author = {Wang, Anqi and Xiao, Zhiqiang and Zhao, Chunyu and Li, Juan and Zhang, Yunteng and Song, Jinling and Yang, Hua},
  title = {An Enhanced CycleGAN to Derive Temporally Continuous NDVI from Sentinel-1 SAR Images},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs18010056},
  url = {https://doi.org/10.3390/rs18010056}
}