I am trying to understand the different cloud flags contained in the L1B’s flags*.nc, specifically the ‘probability_cloud’ and ‘bayes’ ones. Can anyone recommend some documentation on them (the ATBD in the document library on sentinels.copernicus.eu unfortunately does not contain any)?

They don’t seem to be consistent with each other and with the ‘cloud’ flags and with the ‘confidence’ cloud summary bit. In order to judge, which one I want/need to use for my application, I need some further information. Like:

What are they supposed to mean, what’s their intended use? When are they good, when not? How are they derived?

The ATBD (Page 137) is indeed a bit sparse on this. The bayes flags are not explained at all. On this web page, the bayesian flags are at least mentioned, but it is outdated too.

I’ll try to contact some people in order to get something more recent.

Hello Jana,
ESA is aware of the incomplete documentation library and takes action to fill it with the missing documents.
So the library shall be updated soon.

Although this is an old post, I am still struggling with the same issue to find clear information on the SLSTR cloud flags in L1B. I am looking for the best possible cloud flag over land in the SLSTR L1B products. Hope someone can provide some insight.

According to the SLSTR Level 1 product notice, the Bayesian/probabilistic cloud flag outperforms the ‘basic’ summary cloud flag. So, instead of using the cloud_* I suppose I should switch to using another input.

Looking at this product S3B_SL_1_RBT____20220314T105220_20220314T105520_20220315T184413_0180_063_322_2160_LN2_O_NT_004.SEN3, I notice that the flags_*.nc file contains reference to ‘bayes’ but not to the probabilistic cloud flag. Does this mean this layers combines the output of the bayesian and probabilitic cloud masking?

I also notice that the bayes_* and probability_cloud_* layers are empty, see screenshot below. How is this possible? Am I looking at the wrong layers?

Hi Carolien, the bayes_* data include the Bayesian cloud mask over water, and the probabilistic cloud mask over land. bayes_a* data may be empty (I do not know why) but bayes_i* data include a mask named bayes_i*_single_moderate which is of Bayesian origin over water and of probabilistic origin over land. The Bayesian method uses thermal wavebands. Therefore the Bayesian mask is included with and associated with *_i* data. The different Bayesian and probabilistic methods are not reflected in the naming of the cloud masks.

Hello Ralph,
Thanks for your reply.
I am checking again the same product:
S3B_SL_1_RBT____20220314T105220_20220314T105520_20220315T184413_0180_063_322_2160_LN2_O_NT_004.SEN3

It seems that all the bayes_in_* layers are empty?

No idea then. Did you consider looking into other products to see if it is not empty? I have a database of about 60 products, where the masks are definitely not empty.

However, in some older data (e.g. S3A_SL_1_RBT____20200101T053619_20200101T053919_20200102T100530_0179_053_176_3960_LN2_O_NT_003.SEN3), the bayes_in layer does contain information.