… fails in my case and results in a blank or 1-class image. I tried all combinations: training polygons digitized on-screen or imported from .shp, Sentinel 2 or Landsat 7 and 8 images, different classifiers, different band combinations (resampled to uniform pixel size). Unsupervised classification runs perfectly on the same datasets. Any suggestions will be highly appreciated.
For SNAP I suggest you use the dedicated forum:
Thanks; I’ve found a hint for a solution there: all dataset (rasters and vectors) must be in lat-long CRS. Sentinel images come in WGS84 lat-long CRS, but despite that it is advisable to reproject them (writing it here just in a case that anybody else here stucks with the same problem).