August 9, 2022, 1:12pm
Image S2A_MSIL1C_20220412T095031_N0400_R079_T34UCF_20220412T105419.SAFE doesn’t have 0-values on any band (let’s just focus on band8):
The same image after correction (S2A_MSIL2A_20220412T095031_N0300_R079_T34UCF_20220805T124252.SAFE) has lines of zeroes in the shape of land on the water some distance from the shore (and many more):
L1A doesn’t have negative values nor nodata in there. What is going on? It’s the same on all the images.
August 9, 2022, 1:40pm
If you check the same pixel in B12 of the L1C you will see they are zero.
Due to this it is possible the AC cannot be performed.
The scl band indicates those pixels as no-data.
In the upper row B12 and B8 of the L2A are shown. The bluish pixels mark the pixels which are zero. They are also defined by the scene classification data. In the lower part the B12 band of the L1C is shown and the zero pixels are highlighted in red.
August 9, 2022, 1:42pm
ok but what does it actually mean? sensor problem? why is it like that? looks like a processing problem
August 9, 2022, 2:01pm
If you check the B12 and trim the contrast stretch to low values [0, 0.003], you see those “echoes” or ghosts of the land.
On the left the L1C and on the right the L2A. For the L2A If colored the land in green. The no-data pixels are red. You can notice that most of the no-data pixels are located at the ghost-land.
@ABraun provided a possible explanation years ago.
Sentinel-2 MSI - B12 artifacts - s2tbx - STEP Forum (esa.int)
I’m not sure if this is really the cause. It looks a bit different as in the example.
@Jan do you know more about it?
August 10, 2022, 4:45am
Probably also this is related to the issue you see:
The problem is already known. It is due to the offset introduced in baseline 4.00 to keep negative reflectance pixels in the spectrum. In case a pixel has exactly the value of 1000 in an image with baseline 4.00, applying the offset results in a reflectance value of 0 which then is interpreted as NoData by the processor. It will be corrected.
August 10, 2022, 5:41am
Yes, reported in feb, not fixed for half a year?
August 16, 2022, 8:29am
This looks like gradient cross-talk to me. it affects B12. It is identified in Section 6.3 of the L1C DQR available from the Sentinel 2 Document Library ( Sentinel-2 MSI Document Library - User Guides - Sentinel Online - Sentinel Online).
OPT-MPC S2 Technical Manager
This issue has been fixed in Sen2Cor 2.11.
Please be aware of point 7) of
Sen2Cor FAQ if you plan to install Sen2Cor version 2.11 on Linux recent distributions.
The “official” L2A production included this fix within L2A Processing Baseline PB 05.09 starting on 6 December 2022.