SOSSF: Landsat-8 Image Synthesis on the Blending of Sentinel-1 and MODIS Data
Yu Xia,Wei He,Qi Huang,Hongyu Chen,He Huang,Hongyan Zhang
DOI: https://doi.org/10.1109/tgrs.2024.3352662
IF: 8.2
2024-02-02
IEEE Transactions on Geoscience and Remote Sensing
Abstract:Landsat optical sensor is crucial for long-term observations of the Earth's surface with a 30-m spatial resolution. However, the 16-day revisit cycle and severe atmospheric interference have impeded the monitoring of rapid surface changes. Spatiotemporal fusion (STF) is a classic method of predicting Landsat surface reflectance with multitemporal and multisource data, but it is limited by unpredictable temporal changes and cloudy Landsat–moderate resolution imaging spectroradiometer (MODIS) image pairs. Another emerging solution is synthetic aperture radar (SAR)-to-optical image translation (S2OIT), which always produces spectral distortions. To tackle these defects, we propose a new data-driven solution, SAR–optical data-based spatial–spectral fusion (SOSSF), which combines the high-spatial and cloud-free advantages of Sentinel-1 data and the high-spectral and high-temporal advantages of MODIS images to synthesize high-spatial and high-temporal Landsat-8 images. To achieve this solution, we first establish a worldwide benchmark dataset, namely, Sentinel-1, MODIS, and Landsat-8 triplets (SMILE), with various land cover types and all meteorological seasons, satisfying the big data requirements of deep learning (DL). Second, we design an attention-based dual-path fusion network (ADFNet) to respectively extract and fully fuse spatial and spectral information from SAR–optical data. Extensive experiments suggest that the proposed SOSSF solution outperforms the state-of-the-art (SOTA) STF and S2OIT solutions, robustly performing in the continuously changing and frequently cloudy regions. The proposed ADFNet model achieves the best visual effect and the highest accuracy in different scenes, seasons, and bands. Furthermore, the proposed SOSSF solution is proven to be a practical way to simulate time-series and large-scale Landsat-8 surface reflectance, considerably enriching raw Landsat-8 products.
imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geochemistry & geophysics