Cloud-free compositing of Sentinel 2 time series

I have been examining spectral response curves for the period from March to November 2019 for an area in Senegal, and spectral trajectories from February to February for each spectral band.

The plots show that there are large spectral changes in the September image relative to those in June / July, and a reversal of those changes in late October. The trends are far from linear in this period. For this data set, if the September and October imagery were cloud-affected, then linear interpolation could give quite misleading results.

The patterns of the trajectories – with peaks or troughs around September for this area – raise concerns about whether linear interpolation is a sensible methodology for replacing cloud-affected values in the Sen2-Agri work flow.

I have prepared a document which shows the spectral response curves and spectral trajectories over time for the Sentinel 2 data, but I was unable to upload it. I will be happy to upload it if the moderator can send me a link.

Norm Campbell


We are of course very interested by your analysis! If you want, you can send this report to me ( and we will think how to share it with the other users. I am part of the team which is still working at UCLouvain and CS Romania on the development of Sen2-Agri.

Best regards,