Synergy of Sentinel-1 and Sentinel-2 Time Series for Cloud-Free Vegetation Water Content Mapping with Multi-Output Gaussian Processes

Optical Earth Observation is often limited by weather conditions such as cloudiness. Radar sensors have the potential to overcome these limitations, however, due complex radar-surface interaction, retrieving of crop biophysical variables using this technology remains an open challenge. Aiming simultaneously benefit from optical domain background and all-weather imagery provided radar systems, we propose a data fusion approach focused on cross-correlation between streams. To do so, analyzed several multiple-output Gaussian processes (MOGP) models their ability fuse efficiently Sentinel-1 (S1) Vegetation Index (RVI) Sentinel-2 (S2) vegetation water content (VWC) time series over dry agri-environment in southern Argentina. MOGP not only exploit auto-correlations S1 S2 streams independently but also inter-channel cross-correlations. The RVI VWC at selected study sites being inputs proved be closely correlated. Regarding set assessed models, Convolutional model (CONV) delivered noteworthy accurate results winter wheat croplands belonging 2020 2021 campaigns (NRMSEwheat2020 = 16.1%; NRMSEwheat2021 10.1%). Posteriorly, removed observations & dataset corresponding complete phenological cycles September end December simulate presence clouds scenes applied CONV pixel level reconstruct spatiotemporally-latent maps. After applying strategy, phenology was successfully recovered absence data. Strong correlations were obtained reconstructed maps for assessment dates (R2¯wheat−2020 0.95, R2¯wheat−2021 0.96). Altogether, SAR EO with offers powerful innovative cropland trait monitoring cloudy high-latitude regions.