Quantitative relationships of near-surface spectra to Landsat radiometric data

The aim of our research has been to determine the quantitative relationship between the surface spectral character of a variety of geologic terrains and that sensed by the Landsat multispectral scanner. A spectral sampling and measurement program was conducted to accurately characterize the surface spectral reflectance of the Landsat resolution element and, for the first time, to establish statistically the degree of sampling required for a variety of natural terrains. Results from the study showed that for typical homogeneous and moderately heterogeneous terrains, the number of samples required to estimate the mean reflectance of a pixel is small. Only 9-20 samples are required to be within 2% reflectance at the 95% probability level. Coincident field measurements and satellite observations were used to test the equivalency and correlation of the reflectance data. Before the Landsat data could be compared with the surface measurements the satellite brightness values must be converted to absolute radiometric units, and corrected for atmospheric attenuation and scattering. A conversion method using a standard/target comparison, which indirectly compensated for atmospheric attenuation and scattering, produced a Landsat equivalent reflectance that exhibited a root-mean-square error of ± 4% reflectance, when compared with the surface measured value at 12 test sites. Although the equivalence of the surface and satellite data cannot be shown to be better than 4% reflectance, statistical study indicates that the surface and satellite data are highly correlated within defined contrast constraints. However, this correlation is present only after the satellite brightness values are corrected for between band gain differences and compensation is made for atmospheric attenuation and scattering.