Very high proportion of 0 values on MAJA L2A
I use MAJA executable, with MAJA v4.7.1.
I generated several L2A (the whole 2022 year over 4 tiles in Europe). One of the scientific partners from our Copernicus project used those L2A as input to benchmark some of their algorithm. They noticed a number of 0 value pixels in L2A scenes (especially bands B2, B3 and B4, mostly used by their algorithm) particularily high by comparison with other L2A (for example from Sen2Core).
Another partner of theirs also generated MAJA L2A scenes with the same issue (I don't know the version of MAJA used, nor the configuration; it is meant to be a recent one though).
We double checked internally at Magellium and noticed the same issue over all the L2A scenes we studied. We tried a comparison with L2A MAJA directly retrieved from THEIA. These have a number of 0 values similar to those from Sen2Core.
Two things we noticed :
- these possibly abnormal 0 values are sometimes aligned with potential topographic shadows
- on some tiles most of them are hidden behind the clouds found by MAJA Though it is only "impressions" and not proved facts, they may be a lead.
Is this kind of issue something you already noticed (from other users feedback or internal tests) ? Could you please investigate this (we already contacted Olivier Hagolle and it would not be an issue he knows about, hence maybe something linked to MAJA configuration or an issue with a version prior to 4.8) ?
The configuration I used to run MAJA is very close to the default one. MAJA was used with a coarse resolution of 120m, we did not correct cirruses, and allowed generation of L2A up to 100% of nodata (we aim to detect snow, the nodata parameter gather nodata, cloud and snow pixels). Other deviations are only system resources adaptations (RAM, number of threads).
Please find attached two files with images and some stats highlighting the issue, with the tiles and dates used for this analysis.