Application of spectral mixture analysis to hyperspectral imagery for lithological mapping

Hyperspectral reconnaissance mapping methods have become attractive to geological surveys to address the time intensive process of producing regional geological maps for Arctic regions. These sensors can discriminate spectral features associated with common rock-forming and alteration minerals. Three topics relevant to the analysis of such data are addressed in this thesis: (1) image endmember extraction; (2) spectral mixture analysis (SMA); and, (3) generation of a predictive lithological map.; Most of the previous work that deals with image endmember extraction discriminates between pixels using spectral characteristics alone, ignoring their spatial characteristics. High contrast endmembers are easy to identify, whereas low contrast endmembers are more problematic. Improving the potential for identification of low contrast endmembers can be accomplished by analyzing a scene using spatial subsets, thus, taking advantage of the relative spectral contrast between endmembers within a given subset region. Spatial spectral endmember extraction (SSEE) was applied to hyperspectral data and successfully extracted physically significant low contrast endmembers that were not selected by other well known spectral-based methods.; Fractional abundances predicted for a given pixel using SMA are most accurate when only the endmembers that comprise it are used, with larger errors occurring if inappropriate endmembers are included in the mixing process. To address this problem an iterative implementation of SMA (ISMA) was developed to optimize per-pixel endmember sets. ISMA was tested using simulated data with results showing lower abundance errors compared with that of published unmixing methods. ISMA was also effective at obtaining abundance fractions that are physically realistic for a real hyperspectral data set.; A remote predictive lithological map was generated for an area in southern Baffin Island, which comprises a diverse selection of rock types and represents a typical arctic environment with extensive lichen cover. This was accomplished by applying SSEE and ISMA to generate fractional abundance maps, which were subsequently combined into a single map. The predictive map correlates well with the existing published map, including more extensive exposures of potentially economic peridotite and carbonate units. This work also showed that lichen-rock mixtures could be used to map quartzofeldspathic units that have thick lichen coatings.