Clustering Supported Classification of ChemCam Data From Gale Crater, Mars. Issue 12 (29th November 2021)

• Record Type: Journal Article
• Authors: Rammelkamp, Kristin; Gasnault, Olivier; Forni, Olivier; Bedford, Candice C.; Dehouck, Erwin; Cousin, Agnès; Lasue, Jeremie; David, Gaël; Gabriel, Travis S. J.; Maurice, Sylvestre; Wiens, Roger C.
• Is Part Of: Earth and space science. Volume 8:Issue 12(2021)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-11-29
• Subjects: ChemCam -- Mars -- LIBS -- clustering -- classification; Space sciences -- Periodicals; Geophysics -- Periodicals; 500.5
• DOI: 10.1029/2021EA001903
• Languages: English
• ISSNs: 2333-5084

Abstract

Abstract: The Chemistry and Camera (ChemCam) instrument on board the Mars Science Laboratory (MSL) rover Curiosity has collected a very large and unique data set of in‐situ spectra and images of Mars since landing in August 2012. More than 800, 000 single shot laser‐induced breakdown spectroscopy (LIBS) spectra measured on more than 2, 500 individual targets were returned so far by ChemCam. Such a data set is ideally suited for the application of statistical methods for the recognition of patterns that are difficult to observe by humans. In this work, we develop an approach relying on the feature extraction method Non‐Negative Matrix Factorization (NMF) and the repetition of k‐means clustering to classify ChemCam spectra. A strong consistency of the clustering results among the repetitions were found, which allowed us to identify six clusters representing the dominant compositions measured by ChemCam in Gale crater so far. By tracking clusters across the rover traverse from landing to sol 2756, we are able to provide a chemostratigraphic overview of the Gale crater from the ChemCam perspective. Transitions between major geologic groups (such as the Bradbury and the Mt. Sharp groups) are identifiable demonstrating that they are compositionally distinct, consistent with previous work. Compositional differences between their members also appear in the results. Furthermore, a first approach in which a random forest classifier was trained and validated with the obtained cluster assignments, reveals promising results for predicting cluster memberships of new ChemCam LIBS data acquired after sol 2756. Key Points: With a combination of feature extraction and repeated k‐means clustering applied to ChemCam LIBS data, we identify six clusters representing dominant compositions in Gale crater The cluster detections are in agreement with the chemostratigraphy explored by Curiosity We successfully trained a random forest classifier with the obtained cluster labels and used it to classify recent ChemCam targets