loading page

Machine Learning as a Tool to Aid in the Interpretation of Spectroscopic Data: Applications to Lunar and Planetary Exploration
  • Prabhakar Misra,
  • Dina Bower,
  • Robert Coleman
Prabhakar Misra
NASA Goddard Space Flight Center

Corresponding Author:[email protected]

Author Profile
Dina Bower
University of Maryland College Park
Author Profile
Robert Coleman
Howard University
Author Profile


The precise spectroscopic identification of mineral polytypes and specific organic molecules is key to understanding planetary processes and the potential for life beyond Earth in the solar system. For in situ exploration, Raman spectroscopy has been chosen for the NASA Mars Perseverance Rover and upcoming ESA ExoMars missions because it is an information-rich, non-contact, non-destructive method for identifying and characterizing compounds. Misinterpretation of Raman spectra can result in the misidentification of key information used to reconstruct environmental regimes or the detection of potential biosignatures. Machine learning can provide a means to disentangle the mixed signatures that occur in spectra from heterogenous targets by building algorithms capable of discerning subtle differences. Here we discuss an approach that incorporates a Matlab-based machine learning algorithm to study individual mineral samples as a starting point for more complex algorithms targeted for rocks and sediments. The present study focuses on Raman spectroscopy using visible (VIS) excitation laser (514 nm and 532 nm) and a near IR (NIR) excitation laser (at 780 nm) of an assortment of mineral samples typical for rocks on Mars and the Moon, namely olivines, three types of plagioclase minerals (anorthite, bytownite, labradorite), and pyroxenes (augite and enstatite). We have also begun to study the effect of temperature on the vibrational modes for the same mineral samples over a temperature range 300 – 473 K under NIR excitation. Our preliminary data show, for example, that olivine samples from two different locations may exhibit the same typical symmetric and asymmetric stretch and bending vibrations for forsterite (Mg2SiO4); however, under increasing temperatures the peak intensities of ~ 820 cm-1 and ~ 845 cm-1 features exhibited by each sample differed. Our results also showed an enhancement of the Raman peak intensity for plagioclase samples as the temperature increased up to 373K, but a decrease at temperatures beyond that. *Acknowledgments: P. Misra and R. Coleman, Jr. acknowledge support from NASA Award # 80NCCS20M0019, NSF Award # PHY 1950379 & Howard University IDCR # U100189; and D. Bower would like to acknowledge the support of the Internal Research and Development and Fundamental Laboratory Research Programs at NASA Goddard.