An analysis of radiometric correction effects on Landsat thematic mapper imagery
Waits, David Allan
MetadataShow full item record
Land-use classifications and spectral indices are commonly created from raw radiance satellite data. These data are known to be distorted due to sensor instrumentation errors and atmospheric contributions. The overall objective of this study was to evaluate different radiometric corrections of Thematic Mapper (TM) data on land-use classification results and the derivation of spectral indices. A Landsat-4 TM digital image of a diverse agricultural area in the High Plains region of eastern New Mexico was the primary data source. Ancillary data incorporated into the study included: extensive field verification data for a study area of approximately 1,820 square kilometers; ground-based radiometer derived spectral response data for commonly grown agricultural crops; and meteorological data used as input parameters for atmospheric modeling using the Lowtran-7 atmospheric correction program. Four different geometrically corrected image data sets were analyzed. The first was raw radiance data in radiometrically uncorrected form. The other three images were radiometrically corrected transforms created using procedures that adjusted the raw data for radiometric calibration and atmospheric correction. All four images were classified in terms of land-use using identical training fields. Supervised classifications were developed using ground truth data, and quantitative analyses were performed on all resulting classifications. Ground-based spectral response data for various land-use types were compared qualitatively to response data derived from the raw and radiometrically corrected image data for the same land-use types. Four spectral index models were applied to each of the four image data sets. The derived spectral indices were transforms that emphasized the quantitative differences among image data sets. The results showed no material differences in classification accuracy among the four image data sets. Thus, it does not appear necessary to perform radiometric corrections on raw radiance data to improve classification accuracy. Spectra derived from atmospherically corrected image data sets more closely approximated "true" spectral response patterns as obtained by a ground-based radiometer. Each of the various components of the radiometric correction process was found to contribute significantly to the derivation of spectral index values.