Accuracy Measurement in LU/LC Classification from Aerial Photos

While making educated interpretations of land use and cover from remotely-sensed images is all well and good, it is important to consider the fact that some classifications are bound to be erroneous.  User error is endemic to any variety of image interpretation, and to any application of human interpretation of anything, really.  How we classify and interpret these errors is of the utmost interest to any G.I.Scientist. 

The above aerial image, which was previously classified by land type and use, has been sampled for accuracy in the above map.  The sample points were placed within the squares of an overlaid grid, and allow for a spatially-systematic examination of the classification accuracy.  This (very simplified) error measurement is a "quick and dirty" method of examining the land classification, and forgoes a lot of the nuances and details produced by an error matrix with attendant measures of producer's and user's accuracy.  The relatively uniform spacing and placement of sample points in this assessment may also leave out some of the detailed differences in accuracy between classification categories.