Automated lithological mapping using airborne hyperspectral thermal infrared data: A case study from Anchorage Island, Antarctica

• Published in: Remote Sensing of Environment
• Main Authors: Black, Martin, Riley, Teal R., Ferrier, Graham, Fleming, Andrew H., Fretwell, Peter T.
• Format: Article in Journal/Newspaper
• Language: English
• Published: Elsevier 2016
• Subjects: Earth Sciences; Antarctic; The Antarctic; West Antarctica; Anchorage; Anchorage Island; Antarc*; Antarctica
• Online Access: http://nora.nerc.ac.uk/id/eprint/511742/; https://nora.nerc.ac.uk/id/eprint/511742/1/Black_etal_2016_Automated_Lithological_Mapping_RSE%20%281%29.pdf; http://www.sciencedirect.com/science/article/pii/S0034425716300232

The thermal infrared portion of the electromagnetic spectrum has considerable potential for mineral and lithological mapping of the most abundant rock-forming silicates that do not display diagnostic features at visible and shortwave infrared wavelengths. Lithological mapping using visible and shortwave infrared hyperspectral data is well developed and established processing chains are available, however there is a paucity of such methodologies for hyperspectral thermal infrared data. Here we present a new fully automated processing chain for deriving lithological maps from hyperspectral thermal infrared data and test its applicability using the first ever airborne hyperspectral thermal data collected in the Antarctic. A combined airborne hyperspectral survey, targeted geological field mapping campaign and detailed mineralogical and geochemical datasets are applied to small test site in West Antarctica where the geological relationships are representative of continental margin arcs. The challenging environmental conditions and cold temperatures in the Antarctic meant that the data have a significantly lower signal to noise ratio than is usually attained from airborne hyperspectral sensors. We applied preprocessing techniques to improve the signal to noise ratio and convert the radiance images to ground leaving emissivity. Following preprocessing we developed and applied a fully automated processing chain to the hyperspectral imagery, which consists of the following six steps: (1) superpixel segmentation, (2) determine the number of endmembers, (3) extract endmembers from superpixels, (4) apply fully constrained linear unmixing, (5) generate a predictive classification map, and (6) automatically label the predictive classes to generate a lithological map. The results show that the image processing chain was successful, despite the low signal to noise ratio of the imagery; reconstruction of the hyperspectral image from the endmembers and their fractional abundances yielded a root mean square error of 0.58%. The ...