COMPARISON OF SPECTRAL SIGNATURES AND DETERMINATION OF PSEUDO-INVARIANT FEATURES IN RADIOMETRICALLY CORRECTED IMAGES

Abstract

The prospect of making multitemporal studies with Landsat images is higher
as both data banks and facilities to obtain them increase. However, to get comparable data it is
necessary to perform atmospheric and geometric calibrations. Some features in these images
do not suffer radiometric alterations with time so they can be considered pseudo-invariant
features. The analysis of these features indicates that corrected data are comparable. In this
study, COST radiometric and geometric corrections were applied to 8 Landsat 5 TM images,
pathrow 226-82, from October 2008 to January 2010. Comparison of spectral signatures of
six classes and determination of pseudo-invariant pixels were carried out. We conclude that
in bands 1, 2 and 3, the categories native forest, water vegetation and, to a lesser extent, island
vegetation present a good similarity in the spectral signature and pseudo-invariant features. In
Band 4 they show greater variability and bands 5 and 7 are similar in different categories.