Snow mask from S2

[christianbriese]

@Bart92 Del10: p29 could we use the snow mask from S2 (Level 2A Scene Classification) in order to retrieve separate class for permanent snow? otherwise in the map add white for no data (see SRR3: RID-86)

[Bart92]

To implement this we would need to :

• Store the result of predict_random_forest in a variable instead of writing it out to a netCDF
• Design a callback that defines whether a pixel contains permanent snow or not from the SCL layer (for example, if 80% of dates selected by the user is classified as snow in a certain pixel in the SCL layer, it should be classified as permanent snow. The output of this could be binary (0 for false, 1 for true)
• Store the SCL layer in a separate band and apply this callback using reduce_dimension on the time dimension
• Merge the output of that procedure with the output of the predict_random_forest method
• Do another reduce_dimension this time on the band dimension with an if / else, if array_element(1) == 1 then return value greater than any of the outputs of predict_random_forest (e.g. 99), else return array_element(0) (the predict_random_forest prediction)
• Write the final result out to file