A new method for atmospheric correction and de-noising of CRISM hyperspectral data

We propose a new method to perform atmospheric correction and de-noising on hyperspectral image cubes acquired by the Compact Reconnaissance Imaging Spectrometer for Mars (CRISM) on board NASA's Mars Reconnaissance Orbiter (MRO). The CRISM imager has had an important role in advancing our understanding of many aspects of Martian mineralogy. Many mineral detections from CRISM data have been facilitated by significant efforts in the development of the CRISM data processing pipeline to retrieve surface reflectance. However, some residuals remain in CRISM spectra after atmospheric correction, causing difficulty in the interpretation of processed reflectance spectra. In addition, CRISM images are occasionally corrupted with high noise levels exhibiting heterogeneous statistical properties. This paper identifies the cause of such spectral distortions and describe a technique that simultaneously performs both atmospheric correction and de-noising for each image cube individually. In particular, our method focuses on the 1.0-2.6 mu m wavelength region of CRISM images and is applicable to images of non-icy surfaces. Experimental results show that our technique is able to significantly mitigate noise and distortions from various sources like gaseous absorptions, detector temperature, and water ice aerosols, compared with the atmospheric correction method in the CRISM official processing pipeline called volcano scan correction, for a variety of scenes. Careful validations that include the qualitative examination of noise and artifacts both on ratioed and non-ratioed spectra and comparison using multiple overlapping images strengthen confidence in our approach.