The script have to be used in the following order:

1. randomForest_CLEANING_TRAININGDATA:

(neccessary input: Sentinel-2 data and training polygons) rasterizes the training polygons and creates clean training data pixels and saves it as a tif. Only pixels that are covered by more than 80% of a polygon of one class get used, everything else will be set to NA.

2. randomForest_newIndices:

(neccessary input: results from 1. randomForest_CLEANING_TRAININGDATA) caculates a number of spectral indices and texture on basis of the Sentinel-2 images. Saves a multiband tiff with all the original Sentinel-2 bands, the indices and textures and the rasterized training data.

3. randomForest_Training

   (neccessary input: results from 2. randomForest_newIndices) trains a random Forest to classify the forest into "clear", "deadwood" and "undisturbed". Its a very flexibel script, which can do oversampling and undersampling, use different regions as an training and validation basis, use different bands and indices.