Sentinel 2 Geometric Accuracy

I’m trying to understand the geometric accuracy of Sentinel-2 and am hoping someone can provide additional information.

Based on the data quality report released in July 2018 (link below):

My questions:

  • Is there any more specific information on geometric accuracy for individual tiles? For instance, the Landsat 8 metadata file contains information on the geometric accuracy for each image. I’ve tried searching metadata files for sentinel-2 L1C tiles and all I can find is verification that the geometric correction was performed.
  • Any idea when the GRI correction procedure will be implemented? Will historical imagery be re-processed?

Thanks for your insights,

Hi Courtney,
I was expecting the S2 team to answer your interesting question, which concerns all S2 users, but it seems they are busy these days, maybe with the phi-week event.

  • Until now, and contrarily to Landsat 8, S2 uses no ground control points to ortho-rectify the images. As a result, all they have is the statistical assessment that you read in the report, but they have no per scene measurement. They will have the information when they start matching the data to the GRI.
  • Since the beginning of S2 mission, the start of orthorectification with GCP was announced within 6 months in the S2 mission status reports, but since this summer, the starting date is not announced any more, which worries me.

Following a question I asked in CESBIO’s blog six months ago, the S2 project manager told me they were planning a global reprocessing in 2019, of course after the ortho-rectification with GCP has already been set-up in real time. So let’s hope.

A confirmation and a schedule from ESA teams would be welcome indeed/

Hi Olivier,
Thanks for your reply and explanation, and for encouraging the reprocessing of historical imagery. I bookmarked your blog. Let’s see if someone from ESA gives us an update. Please keep us posted if you hear anything.


The Sentinel-2 geometric calibration team is still actively working on the preparation of the geometric refinement. The prototype processor has demonstrated long ago the efficiency of the geometric refinement for Sentinel-2 images. However the transfer to operations has proved more difficult than anticipated. First, the verification of the Global Reference Image (4 TB of data and millions of files) has taken more time than anticipated. Second, we had to rework the processing algorithm to make it more accurate and more robust. We have made some decisive progress recently and it looks like we will be able to activate the geometric refinement early next year.

Regarding the reprocessing of the archive, it is yet another story and ESA does not yet have any planning for this as far as I know.


Merci Sébastien,
we of course fully understand the difficulty of the task.

Thank you for the update!


I have a couple more follow up questions. Thanks for your feedback thus far.

According to this paper, differences between Landsat 8 and Sentinel 2 geolocation can be up to 38 meters. It was suggested in this paper that Landsat data is also corrected using the GRI - are there any plans to do this that you know of?

If I would like to improve the geometric accuracy, would you suggest attempting to manually correct images using control points tied to a high res image? Or do you think the spatial resolution of Sentinel/Landsat is too coarse and I would only increase the error? If I do use control points, I would think that I could use the same points for a time series of imagery, assuming that the error in geometric accuracy between images is much less than the absolute geometric accuracy - would you agree?

Thanks again for the helpful insights.


I know the Landsat team wants to use the Sentinel-2 GRI to co-register the LANDSAT images. They have done some preliminary tests but it’s not done on an operational basis yet.

It is possible of course to correct Sentinel-2 images a posteriori using ground control points. If you just care about relative (multi-temporal) co-registration, any Sentinel-2 image can be used as a reference. If you want absolute performance you will have to use other GCPs.

Generally speaking, correcting for a constant shift for each tile works well, but in mountainous areas there are local deformations which are not so straightforward to correct. This is where the geometric refinement with the GRI will make a big difference.


1 Like

Hi SC,
Thanks for your quick reply and suggestions.