 Sentinel-2 images have been used to improve global crop monitoring in smallholder agriculture zones, such as those in Madagascar. A combination of random forest, RF, classifiers and object-based approaches were used to reduce the number of variables and optimize the performance of the classifier. This resulted in a reduction of the data processing time and an increase in the overall accuracy of the classification. Spectral variables derived from an HSR time series were found to be the most discriminating, while VHSR data were only needed for segmentation purposes. This article was authored by Valentine Le Bourgeois, Stéphane Dupuis, Elodie Ventrue, and others.