LIAISE Land Use CESBIO v1 2021

Landcover survey on a triangle between Ivars d'Urgell, Castelnou de Seana and Vila-Sana on July 25th 2021
The survey was carried (one driver and one surveyor) in two different cars, following the dirt roads around the two GLORI transect lines. The central coordinates of each plot were taken from the road whith the smartphone GPS and google maps application. Using QGIS, google maps and NDVI map, polygons have been drawn around the points of the original shape file (Emna Ayari , K. Dassas, V. Dehaye). Polygons have also been drawn on water and urbain areas. Classes have been attributed to the different land uses (OCS). 1 = maize, 2 = grass and alfalfa, 3 = maize and apple trees / apple trees /pear trees, 4 = wheat cut / dry grass / bare soil, 5 = water, 6 = urbain.

The shape files have been used to generate a lands map with Orfeo Tool box 8.0.1

Supervised classification is performed based on the Level-2A Sentinel-2 time series for the summer season of the 2021 cloud-free selected and ground-truth observations. To monitor the land use changes over time, the classification features include spectral indices and reflectance bands. Three indices are calculated, namely, the normalized difference vegetation index (NDVI), the normalized difference water index (NDWI) and the brightness index (BI). To better distinguish among the crop classes, the optical red edge bands are identified.
To perform supervised classification, the reference data for the land covers were divided randomly into training and validation sets with percentages of 70% and 30%, respectively. Based on the prepared datasets, supervised classification is performed using the random forest classifier. An excellent kappa index value of 0.90 characterizes the classification results. The overall accuracy of the land cover classification is approximately 0.93.
The final has been generated using all ground truth data.