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

Imaging spectrometers make possible remote detection of individual spectral features that can be attributed to specific physical characteristics of geologic targets. Spectral variability measured in the field or laboratory can be directly related to mineralogy, however, for airborne systems, variability is compounded by within-pixel mixing. The research described here evaluated the combined use of an absorption-feature-based analysis approach with linear spectral unmixing for analysis of Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data. The feature-based approach allowed direct identification of individual materials in mixed pixels based on comparison of AVIRIS absorption band feature characteristics with facts and rules compiled using a spectral library. Probability images were created using the AVIRIS data through simultaneous assessment of multiple absorption features for multiple materials. The areas with the highest probabilities for each potential endmember were used to generate image-based average endmember spectra. Average spectra were also extracted using n- dimensional geometric techniques from areas with the 'purest' pixels and the feature-based approach was used to identify the endmembers. Linear spectral unmixing of the AVIRIS data for each material of interest provided estimates of mineral abundances and their spatial distributions. Contour maps of individual absorption feature characteristics such as absorption band depth were overlain on the abundance images to compare spectral variability to estimated mineral abundances. These images showed a strong spatial correlation between the deepest absorption features for specific minerals and the highest mineral abundances for those mineral from the unmixing results. The results suggest a methodology for analysis of imaging spectrometer data, where rather that applying feature-based methods to the entire imaging spectrometer data set, these methods are used instead only to identify materials extracted using the unmixing concepts.