Discovery of rare phases on Mars is important as they serve as indicators of the geochemistry of the Mars surface and facilitate understanding of mineral assemblages within a geologic unit. Identification of rare minerals in high spatial and spectral resolution Compact Reconnaissance Imaging Spectrometer for Mars (CRISM) visible/shortwave infrared (VSWIR) images has been a challenge due to the presence of both additive and multiplicative noise and other artifacts, affecting all collected images, in addition to the limited spatial extent of regions hosting these minerals. In an effort to automate this task we evaluate various clustering algorithms using the detection of rare jarosite, associated with spectrally similar minerals in CRISM imagery, as a case study. We compare nonparametric Bayesian and standard clustering algorithms and show that a recently developed doubly nonparametric Bayesian model could be effective for this task.