Different Value of Angle for Sentinel-2 from SNAP and FMASK(Python)

I noticed that the SNAP derived values (SA, SZ, VA, VZ) differ significantly from those derived from FMASK in the Python version, which is much impacts on LAINN (calculated with the two different datasets (SNAP and fmask)).
How can I adjust myself to this, who should I believe?
Thanks a lot for the answer.

I don’t know how FMASK handles the angles fo S2, but I know it for Landsat.
For Landsat, it generates only very rough angles and I’m not sure about the correctness. At least the results differ from what USGS provides.
For S2, the angles are also not absolutely correct in SNAP, especially not after resampling.
You might want to read these two pages on our wiki:
Influence of wrong angle resampling on water instituence
Intended resampling of S2 view angles
With the final release of SNAP 6, there will be a specific resampling tool which considers the issue described above.
Still, there will remain problems, but therefor an update of the specification S1 L1C products is necessary.
I’ve heard that this shall come in the future.