As part of the work conducted, a classification of grasslands was performed. The developed classification maps were prepared for the selected test areas, with the general division into high-productive grasslands and low-yielding unimproved grasslands. The classification approach was based on applying time series of high-resolution Sentinel-2 satellite data for discrimination of grassland types.
At first stage of the works, High Resolution Layer (HRL) existing within Copernicus service was applied, in order to delineate grassland areas for the selected test sites. Next, series of cloud-free Sentinel-2 images were collected for the regions of interest. It was decided to select the March 2020 and 2021 images. Grassland areas were extracted from S-2 images using masks prepared on the basis of HRL layer. The Random Forest classification method was applied using R programming language. The classification approach was based on information from farmers and advisory service regarding botanical composition and age of grassland as in situ data, then analyses of time series of vegetation indices calculated from high-resolution Sentinel-2 satellite data. There were over 500 test plots designated for each class. These were divided in a 50:50 ratio.
The aim of classification procedure was to discriminate two basic types of grasslands:
• High productive grasslands, which are situated on good or medium quality soils, used with more frequent defoliations, higher fertilization rates, higher stocking rates and producing higher yields than semi-natural grasslands
• Low-yielding unimproved permanent grasslands, which are dominated by indigenous, naturally occurring grass communities and other herbaceous species resulting in high biodiversity
The classification was made into Podlaskie and Wielkopolskie voivodships. The overall accuracy for Podlaskie was 97,6%, while for Wielkopolska was 86,7%. Percentage share of extensive and intensive grasslands in Podlaskie voivodship is 70:30 and Wielkopolskie voivodship is 75:25.