We discussed a little bit about that during the webinar and I have to go back with what I said concerning the gap filling for the L4A crop type. So like Sophie said during the webinar, the system handles the download/access to the L8 data and the preprocessing using MAJA, to provide L8 L2A products. Then, if you trigger the L3B vegetation status processor, it will compute for each acquisition, the NDVI, LAI, FAPAR and FCOVER, separately for the S2 and L8 time series. And then, the L4B grassland mowing detection and L4C agricultural practices monitoring processors use by default all the available NDVI values for the parcels, both from S2 and L8 time series. However, for the L4A crop type processor, the L8 time series was not integrated in the processing:
Because we are working at a country scale and because of the numerous markers that are extracted from the S2 time series, it was already a challenge to optimize the gap filling and markers extraction steps (which are performed at the pixel level) and these steps are already quite long.
The availability of the S1 time series enables to limit the impact of a less dense S2 time series due to a higher cloud coverage so we preferred to put our effort to extract the most of the S1 time series than to integrate the L8 time series (knowing also the specs differences between S2 and L8).
Still, in the Sen2-Agri system, which is a pixel-based crop classification system, it combines S2 and L8 time series to create nice monthly composites, this is what I was confused with during the webinar.