Sen2cor advice for processing a time-series

Hey! I have just recently started with snap, it is a very very cool tool so far but I’m still a bit lost on some guidelines.

So first, I am processing all my data to BOA reflectance - first question comes here - how does sen2cor behave in time series, am I required to do some image cross-normalization afterwards?

A more simple question is about my GIPP.xml file.

I basically forced it to use the 90m SRTM dem, it’s the best I have access to and made all the other parameters automatic. My study area is in Portugal so I do guess I could use the Rural/Summer(slash winter) but I’m a bit flabbergasted on how I could select the atmospheric profile beforehand. Therefore I just made the whole thing automatic.

Is this proper?

What about the other parameters , specially the visibility, altitude, etc? I am just using them by default but what are the recommendations here?

for analyzing time-series, the sen2three plugin could be interesting to you.


Yeah I looked into it but it seemed to be helpful for creating image composites (which is not my intent).

On the other hand, I’ve been exploring GEE and found that i could easily create a time-series of a point-geometry quite easily and fast but i am not yet able to process and store a full .tif file. The atmospheric correction in this case is a simplified 6S algorithm from python.

Hi Nunosa,
for your region of interest in Portugal, you might want to try the L2A images we produce at Theia. You just have to register freely and then download. We started processing Portugal to l2A from December 2015.

These images are atmospherically corrected and provided with a good quality cloud mask using the MAJA software developped by CNES/CESBIO and DLR. The MAJA processor uses the multi-temporal information to perform a better atmospheric correction and a better detection of clouds and shadows.

Of course, we do not only process Portugal, but also France and most of France neighbouring countries (more to come…), and a lot of scattered zones selected by French scientists :

Best regards,



Thanks OHagolle, I actually even shared your site recently in this FB group of GIS/RS. I’ll check it =)

Thanks, I’ll give feedback after =)

1 Like

Just checking it out, it has all the bands in 10m, did Theia ressample everything for that resolution in the end of the processing?

Maybe you have a publication you can point me towards so i can use it as a reference instead of asking.

(I am guessing it won’t make a huge difference for my intended application).

All the best,
Nuno Sá

Hi Nuno,
thanks for advertising on FB, I must be the only one not to have an account there.

We do not resample, we keep the native resolution of Sentinel-2 bands. You will find many information in our blog, in the Sentinel-2 menu. and this page contains references to our peer-reviewed papers.

We’re interested by your feedback (you will find my email easily)
Best regards

So far Olivier, I am pretty much sure I will end up using your data. It is saving me a lot of time, great work!

I especially enjoy the python script to download everything!

Are you guys considering having the data available in GEE? Could be very useful there also!


1 Like

Hi Nuno,
Thanks for your remarks and suggestions.

We would like to have our data published on the DIAS clouds Copernicus is setting up. We are also hoping the French collaborative center PEPS will make them accessible through a Jupyter interface, but it’s not ready yet.

And then, nothing in the data licence (“say thanks to Copernicus and CNES”) prevents any user to put the data on a cloud.
Best regards,

Hello OHagolle

I’ve been noticing a displacement between S2 images in different times. Maybe 1/2 pixel between each image. I noticed it also on the L1C products as well as in yours from Theia.

Do you have any input about that? is it me or something else?


Hi Nuno,
yes, it is a known issue, the quality control have measured a rmse error of registration accuracy above one pixel. A performance of 1 pixel without ground control points is excellent, I do not think many HR satellites can reach that, but the requirement is 0.3 pixel. It will be reached when ESA starts using ground control points in the geometrical processing, but it seems it is not operational yet. Does anyone here knows when it will become operational ?
Best regards,