Locating adjacent Eucalypt and NonEucalypt polygons for training, inital tests with a Gaussian Maximum Likelihood Classifier approach.
Further work on the location selected above, including first plots of negative Log Likelihoods for classification of each pixel.
Examination of classifier behavior in Tinaroo, including visualization of low confidence pixels, and accuracy statistics as confidence threshold is increased.
Examination of classifier behavior in Kirrama, including visualization of low confidence pixels. Demonstrates that an optimal nLL certainty cutoff can be identified which gives 98% Eucalyptus classification.
As above, with some additional/supplemental plots.