*2.4. Accuracy Assessment*

We visually compared the spectral classes with reference data derived from high spatial resolution true-colour aerial and satellite images in Google Earth and the Queensland Herbarium at corresponding dates to the Landsat images in order to verify land cover classification accuracy. A stratified random sampling design was appropriate for the accuracy assessment. The random point application in ArcGIS generated approximately 500 unbiased random points, and the validation points were projected onto the classified maps and compared to produce the error matrix. Subsequently, we determined the corresponding change detection thematic map's user's accuracies, producer's accuracies, overall accuracy, and Kappa coe fficient [61] (Table 4).


**Table 4.** Accuracy assessment of Landsat images captured in 2004 and 2017 at Rocky Dam Creek/Cape Palmerston National Park.
