INFLUENCE OF THE TOPOGRAPHIC CORRECTION STRATEGY ON THE LAND COVER CLASSIFICATION: A SPECTRAL APPROACH

Abstract

Topographic correction methods applied to orbital imagery have been evaluated by several authors. The evaluation criteria have been based on the correlation decreasing between shade and reflectance at different spectral bands or, in some cases, it has included the performance of digital classifiers trying to separate specifi c land cover types. Some topographic correction methods include sampling procedures based on the original image to be corrected and results are dependent on how they are conducted. The influence of the topographic correction on the land cover types spectral characterization is frequently neglected although it is relevant, specially in "quantitative remote sensing" in which empirical or physical relations between radiometric data and biophysical or geophysical data are explored. The objective here was to evaluate the classification results achieved from original and topographic corrected images using the C-correction method exploring two sampling procedures. The infl uence of the spectral characterization of some land cover types was also evaluated. Results indicated that pixels under high and low illumination conditions presented bigger spectral dynamics, specially in the infrared spectral region (near and SWIR). Topographic correction based on a single land cover class (vegetation) did not present better classifi cation result when compared to that classifi cation based on the original images. Sampling procedures based on diff erent land cover classes presented lower mapping accuracy and low matching level with the other classification results.