Hybrid FAPAR and FVC retrieval from Sentinel-3 SYNERGY with Gaussian processes: Development, validation, and cloud-readiness

作者：Dávid Kovács, Emma De Clerck, Luke A. Brown, Pablo Reyes-Muñoz, Jochem Verrelst · 发表于：Science of Remote Sensing · 年份：2026 · DOI：10.1016/j.srs.2026.100462 · 研究领域：Adversarial Robustness in Machine Learning、Gaussian Processes and Bayesian Inference、Space Satellite Systems and Control

This study addresses the need for robust methods to retrieve essential vegetation traits (EVTs) from the Sentinel-3 SYNERGY 300 m reflectance product, named SY_2_SYN. SY_2_SYN surface reflectance is produced from a combination of Ocean and Land Colour Instrument (OLCI) and Sea and Land Surface Temperature Radiometer (SLSTR) observations, available on cloud platforms. To routinely retrieve the fraction of absorbed photosynthetically active radiation (FAPAR) and fraction of vegetation cover (FVC) from SY_2_SYN imagery at continental scales, we developed and evaluated Gaussian Process Regression (GPR) models trained using Soil Canopy Observation of Photosynthesis and Evapotranspiration (SCOPE) simulations. For validation across multiple vegetation types, the Ground Based Observations for Validation (GBOV) service was used, a component of the Copernicus Land Monitoring Service (CLMS), which provides multi-year, upscaled in situ reference measurements. The GPR-SYN FAPAR/FVC products were also intercompared with established CLMS satellite-based estimates over the same GBOV validation sites. Overall, the GPR-SYN and CLMS products demonstrated strong agreement when validated against GBOV data, with GPR-SYN slightly outperforming CLMS. For FAPAR, GPR-SYN achieved an R 2 of 0.89 (RMSE: 0.13) compared to CLMS’s R 2 of 0.87 (RMSE: 0.13). For FVC, GPR-SYN demonstrated an R 2 of 0.83 (RMSE: 0.13) while CLMS yielded an R 2 of 0.81 (RMSE: 0.15). GPR-SYN FAPAR/FVC aligns with CLMS in intercom...