Impact of spectral curvature on at-surface reflectance accuracy and information extraction techniques

In Canada´s arctic, the landscape is a complex mixture of exposed rock with varying levels of lichen and tundra vegetation, illuminated by a low elevation sun. Imaging spectrometry (hyperspectral remote sensing) has been shown useful for mapping northern environments, with applications in geological exploration and habitat mapping. Use of hyperspectral imagery aims to exploit spectral characteristics unique to surface cover classes. This assumes pixels provide a linear combination of spectral reflectance from existing sub-pixel components, referred to as end-members. The level of success can be directly impacted by data quality. Insuring the highest quality data is being used requires a robust pre-processing system which evaluates known artefacts, such as spectral curvature (spectral smile), noise, and spectral gain error. In many systems, spectral smile is often the most overlooked source of uncertainty. To evaluate one aspect of hyperspectral imagery for northern environments, the impact of spectral smile is explored with relation to spectral indices which are used to discriminate between surface material types.