The L2A product S2A_MSIL2A_20180710T105031_N0208_R051_T31TDH_20180710T141216.zip displays no data information in the middle of the granule for the whole set of bands (see a band 3 window in the next figure) as noise following a straight line.
When observing the same data in L1C level, the noise is not there (see next figure)
When going into detail to analyze the different bands of the product, some strange artifact can be seen in the cirrus band (band 10 of the whole granule in the next figure)
Yes, I refer to the high variance between the different detectors sections in the band 10 of the MSI sensor for S2A. This variance is much greater compared with the previous images obtained by the same sensor (see the 20180630 image compared with the 20180710 one). The noise (non-data values) following a straight line is just in the border between two contrasted sections of detectors.
The reported problem is present in all the Sentinel-2A images reviewed to date (at least since 2018-07-10). In the next example we use a granule from south Italy. The problem looks like band 10 being out of calibration for detector3. See the next figure corresponding to band 10 of the L1C product S2A_MSIL1C_20180722T095031_N0206_R079_T33TXF_20180722T115552.SAFE
We have identified the issue. It appears that the issue stems from a lack of equalization in the precursor L1C. As you have noted, the issue is most apparent in B10, when stretching is applied (the impact on radiometry is limited to a few percent). We have identified that the cirrus cloud mask is also affected.
The issue is still under investigation, but we intend to have a fix in place in the coming days.
Thanks for your support.
We introduced new EQOB GIPPs yesterday to cure this problem. Our assessment indicates that the fix was successful, and that problem has been corrected. The GIPPs were inserted at the beginning of S2A Orbit 16215 (Observation time start 2018-07-31T00:16:04).
Since 18/07/2018, equalization issues have been observed on S2A products. The issue is especially visible on Band 10 if the contrast is enhanced. This issue also affects the cirrus cloud mask which may exhibit discrepancies between detectors. Note that the impact on the radiometry is limited to a few percent.
My investigations into the issue raised by @vicencpala identified the maximum discrepancy between detectors in B10 as being around 20DN. The issue and its effects on the data and data users is still being assessed.