CROP TYPE MAPPING USING MULTI-DATE IMAGERY FROM THE SENTINEL-2 SATELLITES

This paper presents the results of a crop type mapping exercise conducted in two study areas in Bulgaria and based on data from the Sentinel-2 (S2) satellites. A multi-date maximum likelihood classification approach was used in which nine spectral bands from three cloud-free images, well distributed across the growing season, were used. Validation was performed using field data collected as part of the study and data from the Integrated Administration and Control System (IACS) dataset. Depending on the validation dataset and the study area, an overall accuracy of 74-95% was achieved after the crop type maps were post-processed by mode filtering. Further increase in accuracy may be obtained if parcel boundaries, as defined in the IACS dataset, are used to aggregate the per-pixel classification to a parcel level.