GIS 5265 Module 9: Remote Sensing

I was already looking forward to working with remotely sensed data next semester, so I found this module an enjoyable introduction.  Our task this week was to download a DEM of the Cahokia region and produce two raster classifications of that image.  The classification process divides the target image into categories that, ideally, reflect categories of land cover on the ground.  An unsupervised classification creates the classified raster automatically, with the user defining the number of categories and other limited variables.  A supervised classification, on the other hand, bases the output classified raster on a classification scheme provided by the user.  In this case, we created a new point shapefile composed of 30 points placed within our intended land cover categories.  This shapefile functioned as the input for the "Create Signatures" tool to create the classification file used with the "Maximum Likelihood Classification" tool.  If we were successful in our placement of control points, the output classified raster would expand the classification to cover the image.

Both my unsupervised and supervised classified rasters succeeded in some areas of classification while failing in others.  The biggest problem was the expansion of some categories to include areas I would rather be classified differently.  The unsupervised classification captured a wide range of land cover types in the two categories representing the most reflective objects, notably including Monk's Mound.  Perhaps additional classes would have teased out light grass, bare ground and the mound from roads and buildings.

I attempted to create a category specifically for Monk's Mound in the supervised classification; unfortunately, as can be seen above, the mound was still mostly classified along with roads, buildings, and light grass.  This isn't completely surprising since the mound is covered in grass, but I had hoped to capture some difference with the supervised classification.  As in the unsupervised classification, however, Monk's Mound does show up quite clearly as a distinct shape.  More tweaking of the input classification scheme would likely lead to a more accurate output.  Even as they are, both classified rasters could be useful if one were looking for new sites or interested in an overview of the land cover around Cahokia.