Unsupervised classification.

OK. It's now time to start playing with unsupervised classification. We'll use the same data as the PCA lab - one from kitco, one from 80 Mile Beach. Run through this lab with the kitco image, then do the same at 80MB.

It's pretty straightforward, goto Raster - unsupervised - unsupervised classification. Provide an output cluster layer name. initialize from statistics using k-means. Set max iterations to 20 and a convergence threshold of .97. do NOT classify zeros.

Run it twice with 10 and 15 classes. Note, this might take awhile..... Bring these images up on screen and try to interpret! In short, what do you think each class represents on the ground?

Provide a table listing what you think each class is (a landuse or landcover). There will be two tables per image (the 10 and 15 class images). It can help to right click on the unsupervised image, display attribute table, and play with the colors. Also note, you can look at both these areas in higher detail in Google Earth. Throw in some pics if you think it will help.

Which of these two runs seems to best represent what is on the ground? Why?

Note, please recolor the classes so they are distinct and provide color images of your unsupervised images on what you turn in. So... two maps, two tables, and a writeup of which you prefer for the Kitco image. Ditto the 80MB image.

repeat for 80MB