Confusion matrix between the classes defined in this taxonomy in the visible-near-infrared and albedo input space. For each class in the taxonomic scheme, we give the average probability of its samples to be classified as any other class based on the complete dataset. The Ch-class is missing as it relies on the detection of the 0.7 µm h-feature and does not have an associated class probability. For better readability, the main matrix diagonal corresponding to the equal-class cases is left empty. These values are generally above 80% and lowest for K, L, M, and R.