Bridging Extreme Data Gaps: SAR-Optical Fusion Guided by Dual-Temporal SAR Differences
Open Access
- 26 May 2026
- journal article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. PP (19391404) , 1-15
Abstract
Continuous and complete optical observations are essential for agricultural monitoring, phenology analysis, and land-surface assessment. In subtropical and tropical regions, persistent large-scale thick cloud cover frequently causes extended periods with few available optical image pixels, making traditional cloud removal and optical reconstruction methods that depend on spatially and temporally adjacent cloud-free data inapplicable. Similarly, existing SAR-optical fusion and multi-temporal optical approaches require partially cloud-free target imagery. As a result, they become ineffective in scenarios with extreme cloud coverage ($\geq$80%) or during prolonged periods of optical data absence. To address this challenge, we propose TDSAR-DR, a dual-temporal SAR-difference-guided optical data reconstruction framework for reconstructing complete optical imagery at the target time without requiring any usable target-time optical observation as model input. Modulated Fusion (TV-SMF) module integrates reference optical information with time labels and temporal change cues extracted by a Time Difference Enhancement Module (TDEM). These dual-temporal SAR derived cues explicitly encode the direction and magnitude of land-surface dynamics, providing physically grounded guidance for a directionally informed and quantitatively consistent transformation toward the target time. Then, a SAR Guided Structural Refinement Module (SGR-M) exploits detailed SAR spatial cues to improve the structural fidelity and spectral consistency of the reconstructed target-time optical image, thereby producing robust results even when optical observations are missing for extended periods. Extensive experiments on subtropical farmland datasets show that TDSAR-DR achieves better qualitative and quantitative results under extreme cloud coverage, outperforming current SOTA SAR-optical fusion models. TDSAR-DR achieves a PSNR of 27.134 and SSIM of 0.947, surpassing other the state-of-the-art methods. More importantly, the framework enables reconstruct continuous cloud free time series, producing temporally consistent optical imagery when optical data are fully missing for long periods.Keywords
Funding Information
- National Key Research and Development Program, China (2023YFC2809102)
- Major Key Project of PCL
This publication has 56 references indexed in Scilit:
- Iterative Global Mapping-Local Searching for Heterogeneous Change Detection with Unregistered ImagesInternational Journal of Computer Vision, 2026
- MT_GAN: A SAR-to-optical image translation method for cloud removalISPRS Journal of Photogrammetry and Remote Sensing, 2025
- Uncertainty-Based Thin Cloud Removal Network via Conditional Variational Autoencoders2023
- HS2P: Hierarchical spectral and structure-preserving fusion network for multimodal remote sensing image cloud and shadow removalInformation Fusion, 2023
- GLF-CR: SAR-enhanced cloud removal with global–local fusionISPRS Journal of Photogrammetry and Remote Sensing, 2022
- Siamese Neural Networks: An Overview2020
- Cloud removal in Sentinel-2 imagery using a deep residual neural network and SAR-optical data fusionISPRS Journal of Photogrammetry and Remote Sensing, 2020
- A commentary review on the use of normalized difference vegetation index (NDVI) in the era of popular remote sensingJournal of Forestry Research, 2020
- vPSNR: a visualization-aware image fidelity metric tailored for diagnostic imagingInternational Journal of Computer Assisted Radiology and Surgery, 2012
- An Introduction to Satellite Sensors, Observations and Techniques2007