Remote sensing based classification of glacier surfaces has long been used to map glacier facies, debris cover, surface hydrology and different kinds of light absorbing impurities. Since the optical properties of the glacier surface affect the energy and mass balance, accurate characterizations of surface types are important for melt estimates that can take into account the impacts of impurities (e.g. mineral dust) on surface albedo. Surface classification in multispectral optical imagery is often based on empirical relationships between reflectance at particular wavelength bands or band combinations. In recent years, hyperspectral imagery has increasingly become available for environmental monitoring applications (e.g. PRISMA, EnMAP, EMIT satellite missions), although applications on ice and snow remain relatively rare. The higher spectral resolution and narrow bandwidths of hyperspectral data allow for classification and anomaly detection approaches that leverage the distinct spectral signatures of different surface types and impurities at much greater detail than in multispectral data. We use airborne hyperspectral imagery (VSWIR) of Mendenhall glacier, obtained during the 2020 melt season, and a Sentinel-2 acquisition from the same day to explore how spectral resolution affects surface classification results, focusing particularly on unsupervised classification. By varying the amount of spectral information supplied to the classifiers, we assess the sensitivity of the classification to spectral resolution and benchmark the output of several unsupervised methods against supervised classification. We present initial results from the Mendenhall Glacier case study and aim to discuss application-dependent strengths and limitations of different classification approaches.