Cloud shadow detection problem

I face a problem with cloud shadows detection using quality_scene_classification in ESA SNAP during L2A products analysis.

  1. The opaque cloud shadows could have a lot of gaps and sometimes smaller than the actual shadow.
  2. The thin cirrus shadows are not detected at all.

The products used in the illustrations above:

  1. S2B_MSIL2A_20180505T094029_N0207_R036_T35VLF_20180505T115132.SAFE
  2. S2A_MSIL2A_20180530T094031_N0208_R036_T35VME_20180530T122531.SAFE

When can we wait for updates in the cloud shadow classification? What can be done to improve cloud shadow detection?

With all best,

Dear Marharyta,

as alternative for detecting clouds and cloud shadows in Sentinel-2 MSI images you can try the ‘fmask’ algorithm implemented in Python:


Another alternative is IdePix for Sentinel-2.
You can find it in the menu of SNAP at Optical / Pre-processing /Masking / IdePix / Sentinel-2 MSI.
But most cloud detection algorithms have their pros and cons. You just need to find the one which serves your needs best,