| To solve this problem I went back to the raw 6 bands of imagery. The hyperclusters were just too limited. I then used a supervised classification as opposed to the previously discussed unsupervised hypercluster aggregation. Supervised classification involves drawing training polygons over representative areas of a specific class. The computer then computes statistics for each class and puts each pixel into the class that is more statistically similar. Beside the usual 7 land cover classes, I added 2 more categories for each forest type to account for topographic variability. Each forest type (conifer, deciduous, mixed) had a north, south, and no slope class. The forest classes were the only ones concentrated on because of their domination of the steeper slopes (due to past land use history).
|
|
