I have been trying to run a number of the supervised land cover classifications on an L2A subset and resampled image. I have 5 vector data sets of training categories in a subtropical grazing environment. When I run them (RF or MDC or MLC) only one of the categories gets 100% frequency of the pixels as an end result - occasionally the output is completely blank… The unsupervised classification works as expected. I select the input S2A image for which the vector containers have been made, I use a subset of band B2, B5, B11 and B12 and I select all the training vectors. Is there something I might be missing?
Supervised classification only works with reprojected data - have you tried it: