Composition and configuration of habitat elements are important determinants of animal movement and resource selection. However, calculating such metrics requires prior identification of habitat elements. This can be achieved using imagery.

In this map, I used two imagery-based classification methods to identify elements of sandhill habitat in central Florida. First, I conducted unsupervised classification, which groups pixels into classes without prior training. The second method involved image segmentation and model training. I segmented the image into super-pixels based on spectral values from the red, green, and near-infrared bands. I then created training samples for ground cover, tree-cover, and sand patches. Training samples were used to train a support vector machine classifier, and the resulting definition file was used to classify the segmented image.

From visual inspection, the segmentation method performed better than unsupervised classification. This map is meant to be a simple demonstration, but there are several quantitative methods for testing the validity of classifications.

In one of my next projects, I will demonstrate how landscape metrics can be derived from a classified raster layer, and why these metrics are relevant to wildlife ecology.