Multi-temporal Maximum Likelihood classification


I’m trying to do a MLC land cover classification over my study area using two level 1C Sentinel-2 images as input. The first one is from january 2017 and the second one from june 2017. Any suggestion how to proceed so that the algorithm takes into account both images?


I see no problem in doing this. When you open the ML tool you can add various rasters. If they have the same coordinate system and cover the same area the training areas can be used on both.
But I would recommend to do a radiometric calibration (sen2cor) so areas which are the same will truly have the same raster value.

When creating the training set, I did it only in one of the images. The other one does not have the training vectors on the “Vector Data” folder. Is this the correct way?

The two images are from the same tile (T29TNE) so they have the same coordinate system and they have been subset also the the same area of interest.

I run the tool and after 1 hour the output was an empty raster.

Could you please check that my procedure is okay? If so I will repeat it and post screenshots in case of failure


Technically this should work, training vectors only need to be in one of the products.
You can try stacking both products into one (Raster > Geometric Operations > Collocation or Radar > Coregistration > Stack Tools > Create Stack) and run again.

Can you please post a screenshot of your raster and the training samples?