Mapping spectral variability of geologic targets using Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data and a combined spectral feature/unmixing approach

摘要

Abstract: Imaging spectrometers make possible remote detection ofindividual spectral features that can be attributed tospecific physical characteristics of geologic targets.Spectral variability measured in the field orlaboratory can be directly related to mineralogy,however, for airborne systems, variability iscompounded by within-pixel mixing. The researchdescribed here evaluated the combined use of anabsorption-feature-based analysis approach with linearspectral unmixing for analysis of AirborneVisible/Infrared Imaging Spectrometer (AVIRIS) data.The feature-based approach allowed directidentification of individual materials in mixed pixelsbased on comparison of AVIRIS absorption band featurecharacteristics with facts and rules compiled using aspectral library. Probability images were created usingthe AVIRIS data through simultaneous assessment ofmultiple absorption features for multiple materials.The areas with the highest probabilities for eachpotential endmember were used to generate image-basedaverage endmember spectra. Average spectra were alsoextracted using n- dimensional geometric techniquesfrom areas with the 'purest' pixels and thefeature-based approach was used to identify theendmembers. Linear spectral unmixing of the AVIRIS datafor each material of interest provided estimates ofmineral abundances and their spatial distributions.Contour maps of individual absorption featurecharacteristics such as absorption band depth wereoverlain on the abundance images to compare spectralvariability to estimated mineral abundances. Theseimages showed a strong spatial correlation between thedeepest absorption features for specific minerals andthe highest mineral abundances for those mineral fromthe unmixing results. The results suggest a methodologyfor analysis of imaging spectrometer data, where ratherthat applying feature-based methods to the entireimaging spectrometer data set, these methods are usedinstead only to identify materials extracted using theunmixing concepts. !31