Features for training (L4B processor)


I just want to make sure. Does L4B processor under its hood use exactly the same NDVI Time Features along with Brightness and NDWI statistical features as in L4A processor for the training procedure?

I can’t find this information in the papers and documentation.

Thank you!


No, there are a couple of differences:

  • the vegetation indices are computed on-the-fly, so there is no need to run the other processor beforehand
  • there is a data filling / temporal resampling step applied beforehand (on the reflectance data, not on the NDVI)
  • some reflectance bands are also included in the features used by the classifier.

Hello Mosin,

To complement Laurentiu’s answer,

  • S2 features extracted for the crop type classification are the surface reflectance values in the bands 3, 4, 5, 6, 7, 8, 11 and 12, NDVI, NDWI and Brightness calculated for each date of the resampled and gap-filled time series
  • S2 features extracted for the crop mask classification are spectral, statistical and temporal features related to resampled and gap-filled time series of NDVI, NDWI, Brightness (Valero et al, 2016) and S2 red-edge indices.

Thank you for the answers! It’s exactly what I wanted to know.