Comparison of relative radiometric normalization techniques

Relative radiometric normalization (RRN) is a procedure used to prepare multitemporal image data sets for the detection of spectral changes associated with phenomena such as land cover change. This procedure reduces the numeric differences between two images that have been induced by disparities in the acquisition conditions (e.g. sensor performance, solar irradiance, atmospheric effects) rather than changes in surface reflectance. We have applied seven empirical multitemporal radiometric normalization techniques to 1973 and 1990 Landsat MSS images acquired of the Washington D.C. area. The results from the various techniques have been compared both visually and using a measure of the fit based on standard error statistic. Both methods of comparison indicate that a linear regression technique using pixels from the two images which did not undergo spectral change produces the best results.