SOIL SPECTRAL EFFECTS ON VEGETATION DISCRIMINATION (REMOTE SENSING).

Abstract

The spectral behavior of a cotton canopy with four different soil types inserted underneath, respectively, was examined at various levels of vegetation density. Measured composite spectra, representing mixtures of vegetation with different soil backgrounds were compared with existing measures of greenness, including the NIR-Red band ratios, the perpendicular vegetation index and the green vegetation index. Observed spectral patterns involving constant vegetation amounts with different soil backgrounds could not be explained nor predicted by either the ratio or the orthogonal greenness measures. All greenness measures were found to be strongly dependent on soil brightness. Furthermore, soil-induced greenness changes became greater with increasing amounts of vegetation up to 60% green cover. Three versions of factor analysis were subsequently utilized to determine if soil background influences could be filtered from canopy spectral data sets. In R-mode factor analysis, canopy spectra were decomposed into orthogonal features called brightness and greenness. The greenness feature, however, was found to be dependent, not only on vegetation density, but on soil background spectral properties. Of most concern were soil brightness influences which resulted in lowered greenness values with wet or dark soil backgrounds and identical vegetation conditions. The Q-mode version of factor analysis decomposed canopy spectra into additive, soil and vegetation, reflectance components. Although soil spectral response was found to contribute and mix into the derived greenness measure, significant improvements in vegetation discrimination occurred, especially at low vegetation densities. Finally, the T-mode version of factor analysis successfully separated the spectral influences of soil background from the larger response due to vegetation canopy development. Canopy spectra were decomposed into soil-dependent and soil-independent canopy components. The soil-dependent component was found to resemble the spectral response of green vegetation due to the scattering and transmittance properties of the overlying vegetation canopy. Results showed how the soil-dependent signal mixed into various measures of greenness and hampered vegetation discrimination. The filtering of soil background response from spectral data sets significantly improved greenness indices and vegetation analyses.