Comparison of surface reflectance derived by relative radiometric normalization versus atmospheric correction for generating large-scale Landsat mosaics

Abstract

Generating large-scale Landsat mosaics of surface reflectance is challenging because of the tediousness arising from atmospheric correction for a large number of scenes. To find out an alternative approach, we conducted an empirical investigation to compare the surface reflectance derived by relative radiometric normalization versus atmospheric correction using four pairs of adjoining Landsat Thematic Mapper/Enhanced Thematic Mapper Plus scenes in northern Canada. Each image was first atmospherically corrected to convert top-of-atmosphere radiance to surface reflectance. One of the converted images in each pair was then respectively used as a reference to radiometrically normalize the other original one for deriving surface reflectance. Comparison of the surface reflectance derived by these two different approaches indicates that they can match reasonably well for different landscapes, atmospheric conditions, and sensors, and the difference measured by root mean square error is no more than 0.0098 for the visible band (Band 3), 0.0271 for the near-infrared band (Band 4), and 0.022 for the middle-infrared band (Band 5). Given such a small difference, we would expect that relative radiometric normalization may be used as an alternative approach for reliable and fast retrieval of surface reflectance from Landsat data for generating mosaics of surface reflectance over large areas, overcoming the tediousness arising from atmospheric correction for a large number of scenes.