For the current sen2Cor (up to Version 2.0.4) a scaling problem and additionally a mismatch between the 10m resolution output and the 20 / 60 m resolution and a strong distortion in the 10 m processing has been observed in the Level 2A output images and was reported by several users during the last weeks. This has now been analyzed in detail and led to the following modifications of the algorithm:
increasing the input sensitivity according to a new reflectance to radiance routine to be aligned with the current inputs. This now works also for the new planned 10000 scaling (instead of 1000).
improving the Dark Dense Vegetation classification, as the algorithm erroneously classified some water as dark pixels which contributed to the observed distortions.
For the 60 and 20 m case, a re-scaling of the path radiance was performed if scaled path radiance for the blue band differs more than 3 % from the path radiance of the standard aerosol.
This routine was however only called when a visibility calculation is performed, which is the case for the 60 and 20 m processing.
In the 10 m case this routine was not executed, as the visibility is resampled from 20 m.
Thus, for 10 m bands the scaled path radiance for the blue band (sc_lp_blue) always remained to 1.0 due to the different treatment of the visibility calculation (see above).
These findings explain:
why the observed differences only occur in some of the images: - only if sc_lp_blue differs more than 3% from the default 0f 1.0, or if sufficient amount of water pixels are found, a rescaling takes place.
why the effect is stronger in some of the images: - the weight of the sc_lp_blue goes into the correction.
why the differences also occur between 60 and 20m, when the 20 m visibility was resampled from the 60 m bands instead of being calculated new.
A modified version of Sen2Cor (release 2.0.5) was prepared, in which the fixes listed above had been applied and the modified scaling for the blue channel path radiance for 20 and 60 m is stored in the configuration and is then reused for rescaling also during the 10 m processing. With this modification the same results for the 10 m bands as for the 20 and 60 m ones can now be obtained.
This new version 2.0.5 is handed over to the Expert Support Laboratory (ESL) Team for final testing. When this tests pass, the version will be published soon.