Mapping of olive trees using Sentinel-2 and Sentinel-1 images: an evaluation of pixel-based analyses

This work was carried out in the region of Kairouan in central Tunisia $(677 \mathrm{~km}^{2}.$, Merguellil catchment area). The objective was to assess the ability of integrated time series Sentinel-1 (S-1) and Sentinel-2 (S-2) imagery and field observations, to map agricultural practices of the olive tree in a semi-arid area. Two pixel-based classifiers, based support vector machines and random forest were tested to evaluate different combinations of satellite features as input for mapping different olive tree cropping systems. The classifications obtained by integrating both S-1 and S-2 time series provided better results (overall accuracy =83.2 %) than the one derived by using only Sentinel-2 data (overall accuracy =69 %) with a random forest-based classifier. Results confirm how the integration of satellite multi-sensor time series, could lead to a better identification and characterization of olive trees agricultural practices in the Mediterranean area.