Unravelling the Chemistry of Environments Conducive to Life on Mars using Machine Learning
Lead Supervisor: Richard Harrison, Department of Earth Sciences
Co-supervisor: Nick Tosca, Department of Earth Sciences
Brief summary
PIXL datasets, which include multispectral micro-context images and fluorescence data (which include diffraction phenomena), will be analysed using a new automated machine-learning approach – SIGMA (Tung et al., 2023). SIGMA, a machine-learning workflow successfully applied to analyses of Energy-dispersive X-ray spectroscopy (EDS) data, is particularly suited to study the mixed signals on XRF data from Mars and should provide sizeable improvement to how things are currently done. The analysis will also include comparing results to existing methods and assessing errors and artefacts.
Importance of the area of research concerned
This is a cross-disciplinary project that aims to apply novel computational analyses to X-ray fluorescence (XRF) datasets recently acquired by the Mars 2020 Perseverance Rover at Jezero Crater. The Planetary Instrument for X-ray Lithochemistry (PIXL) has returned micro-focused XRF data of igneous and sedimentary rocks approximately 3.5-3.8 billion years old (Farley et al., 2022; Tice et al., 2022), and these data are being used to inform short- and long-term exploration strategies of Jezero Crater and the selection of drill core samples to be acquired and eventually returned to Earth. However, like other energy-dispersive spectroscopic datasets, PIXL data suffer from a problem of non-uniqueness; it is often difficult, if not impossible, to confidently determine which minerals may be present in a given analysis. The project will result in a family of solutions for the mineral species present in ancient rocks on Mars, as well as their textural relationships. This raises the possibility that the project can directly contribute to the major scientific goals of the Mars 2020 mission, which is to characterize ancient habitable environments on the surface of Mars, and select samples most likely to record evidence for ancient life, if present.
What will the student do?
Identification of unknown micro- and nano-sized mineral phases is commonly achieved by analyzing chemical maps generated from hyperspectral imaging data sets, particularly scanning electron microscope—energy dispersive X-ray spectroscopy (SEM-EDS). However, the accuracy and reliability of mineral identification are often limited by subjective human interpretation, non-ideal sample preparation, and the presence of mixed chemical signals generated within the electron-beam interaction volume. Machine learning has emerged as a powerful tool to overcome these problems. The student will apply a machine-learning approach to data being collected on Mars using the PIXL instrument, with the aim of performing automated quantitative analyses of the mineralogy within Jezero crater. The student will compare results to existing methods and assess errors and artefacts. The student will write a Python code to adapt SIGMA to load and process the XRF data sets. They will then use three different approaches: conventional methods, multivariate statistical analysis, and then the latest neural-network-based SIGMA workflow to provide a quantitative breakdown of the mineralogy. Benchmarking and comparison among these three approaches will also be performed. The student will then consider what the results of this analysis mean in terms of the core science questions of the mission.
References
- Farley, K. A. et al. Aqueously altered igneous rocks on the floor of Jezero crater, Mars. Science, 377, 6614, 2022.
- Tice M. M., Hurowitz J. A., Allwood A. C., Jones M. W. M., Orenstein B. J., Davidoff S., Wright A. P., Pedersen D. A. K., Henneke J., Tosca N. J., Moore K. R., Clark B. C., McLennan S. M., Flannery D. T., Steele A., Brown A. J., Zorzano M.-P., Hickman-Lewis K., Liu Y., VanBommel S. J., Schmidt M. E., Kizovski T. V., Treiman A. H., O’Neil L., Fairén A. G., Shuster D. L., Gupta S. and Team T. P. (2022) Alteration history of Séítah formation rocks inferred by PIXL x-ray fluorescence, X-ray diffraction, and multispectral imaging on Mars. Science Advances 8, eabp9084, 2022.
- Tung, P.-Y., Sheikh, H. A., Ball, M., Nabiei, F., & Harrison, R. J. (2023). SIGMA: Spectral interpretation using Gaussian mixtures and autoencoder. Geochemistry, Geophysics, Geosystems, 24, e2022GC010530. https://doi. org/10.1029/2022GC010530
- Proterozoic Belt Supergroup, Montana, USA. Sedimentary Geology, 120(1-4), pp.105-124.
Requirements as to the educational background of candidates that would be suitable for the project
Any background in the natural sciences will be suitable.
Applying
You can find out about applying for this project on the Leverhulme Centre for Life in the Universe widening participation PhD Studentships page.