NaNs appear while GPT graph processing + writing to NetCDF4

I’ve created a graph for batch processing SAR images in Snap and saved it as a XML. This graph
reads an input Sentinel-1 file and then performs: radiometric calibration, applies orbit file, deburst and multilook. The output is converted to dB and saved as NetCDF4. The graph runs very well from inside Snap, but when I used the GPT tool for cli processing, some NaNs appear in the resulting data. Why do you think it could happen?

My graph file is attatched here:s1_iw_slc_2_nc.xml (3.0 KB)

Below is an image of the output data. White pixels are NaNs.

After more research, I created a better description of the problem here.

As a correction to this post, the pixels were actually not NaNs but masked zeros.