Katharine Masun

A comparative investigation of different remote sensing techniques was performed for lithological discrimination on north central Devon Island, Nunavut, Canada. Isodata, Maximum Likelihood and Spectral Angle Mapper algorithms were applied to Landsat Thematic Mapper (TM) and Hyperion hyperspectral data sets to determine the accuracy and utility of the classified maps without the benefit of field data acquisition or a priori knowledge. Results suggest that there is value in applying numerical classification techniques to define and map broad scale geological units. There is an overall agreement in the final classified maps and the Maximum Likelihood classifier applied to the Landsat TM data produced the most accurate overall results. However, the unsupervised Isodata classification of the Hyperion image returned the most accurate representation of lithology, whereas Isodata applied to TM data was unable to separate sedimentary formations. All classifiers delineate the formational boundaries of lithological units, but showed intra-formational confusion. Classification relies on the spectral characteristics of the rock surface, which are inherently spatially variable: rock surfaces are subjected to variable rates of weathering and erosion. Furthermore, the sedimentary formations within the study area comprise similar rock compositions and likely have considerable overlap in their spectral responses. Although the final Hyperion Isodata results provide the most accurate lithological classification, Landsat TM data with their greater land coverage and ease of data manipulation, have better utility for lithological mapping.