*2.7. Accuracy Assessment*

The land cover map based on the classification using Sentinel-2 images (same extent as Worldview-2) was assessed using 30% of the reference data, which was excluded from classification and from cross-validation. To assess the accuracy of the land cover map based on the upscaling approach, a confusion matrix was constructed, consisting of 167 validation points collected during the field survey, for image interpretation using the Spot 6/7 images. We used a leave-one-out cross-validation [56] because our sample was relatively small (30% of the training data did not cover the land cover map to a large extent).

This matrix provides the overall accuracy, the kappa coefficient, and the user's and producer's accuracies for each class. The producer's accuracy represents how well reference pixels of the ground cover type are classified. The validation points were rasterized to the 10-m resolution of the Sentinel-2 image. Furthermore, a Wilcoxon test (non-parametric statistical test that compares two paired groups) was applied in order to estimate the significance difference between the user's and producer's accuracies for the two classification maps [57].
