Mineralogical characterization of tropical soils by hyperspectral remote sensing

AVIRIS (Airborne Visible/Infrared Imaging Spectrometer) data collected in Brazil were used for the spectral characterization of a typical crop-pasture scene, and for the evaluation of the Spectral Feature Fitting (SFF) technique for clay mineral identification in the image. A six-endmember linear spectral unmixing model was applied, consisting of green and senescent vegetation, water, and the soils Alfisol, Oxisol and Entisol. For mineral identification of kaolinite, montmorillonite and gibbsite in the AVIRIS image, reference spectra from the JPL/NASA spectral library were selected. Pixel and reference spectra were normalized by the continuum removal method, in the 2,100-2,330 nm interval, and then compared by the use of the SFF technique. Kaolinite is the dominant mineral in the scene, whose identification is dependent on the soil type and on the spectral mixture at sub-pixel level. The best results were obtained for soils with intermediate to high overall reflectance and for pixels with soil abundance values greater than 70%, due, respectively, to the lower amount of opaque substances of these soils and to the reduction of spectral effects of the lignin-cellulose features. These factors tend to obliterate the absorption bands of clay minerals.