Classification and accuracies

The tool has been tested in the field (using mobile phones and tablet) by end-users and found to be a suitable help in the assessment of the status of the vegetation. Accuracy of the year classification is currently in the order of 80% while overall accuracy of single image classifications vary between 60 and 70%. A confusion matrix of the annual classification in the same year as the ground truth is shown in Table 3. The confusion occurs especially between classes in which vegetation is very much alike, such as the confusion between willow shrub and willow forest, and between rough natural grass lands and production grasslands.