The ordering of multi temporal fuzzy land-cover information derived from Landsat MSS data

Change detection from satellite imagery is fraught with problems of precision, and detection of incorrect changes at rates of up to 80% of pixels are not uncommon in the literature. In this paper, the Fuzzy c-Means classifier, which yields fuzzy membership values for each pixel in all land cover types, is applied to imagery of the same area at different dates where no actual change is know to have occurred. Analysis of this ordered data and the associated cover types shows that the rate of change, initially identified as 40.76% of pixels in hard classifications, which is clearly erroneous, can be reduced to about 0.32%; a rate that is considered to be very acceptable for most applications.