Probabilistic multivariable calibration for major elements analysis of MarSCoDe Martian laser-induced breakdown spectroscopy instrument on Zhurong rover

Mars Surface Composition Detector (MarSCoDe) is one of the major payloads onboard the Zhurong rover. The laser-induced breakdown spectroscopy (LIBS) component within MarSCoDe is capable of acquiring LIBS emission from 240 to 850 nm at targets ranging from 1.6 to 7 m. To address the previous issue of the indirect uncertainty estimation on the calibrations of other Martian LIBS, a recently developed natural gradient boosting probabilistic prediction (NGBoost) ensemble learning method was adopted for the LIBS calibration of MarSCoDe. Probabilistic Major Element Composition (PMEC) models based on NGBoost were established with the capacity of predicting the concentrations of 8 major element oxides (SiO2, TiO2, Al2O3, Fe2O3T, MgO, CaO, Na2O, K2O) in the form of normal distributions. The models were trained based on a ground spectral repository from MarSCoDe LIBS re-cords of various reference samples. Optimizations for robustness and performance of the models were carried out with tunable max-pooling feature extraction and cross-validation-controlled hyperparameter tuning. The models were tested on an independent portion of the ground spectral repository and were checked with LIBS spectra of onboard calibration targets from Mars for their accuracy, error significancy relative to predicted uncertainty, consistency among single and averaged shots, and tolerance to mocked wavelength shifts. The PMEC models are presumably competent for basic elemental analysis of Martian silicate soil and rock along Zhurong's expedition. The limitation and interpretability of the models were also discussed to guide further works.

Automated summary

This article introduces the calibration method for the laser-induced breakdown spectroscopy (LIBS) of the Mars Surface Composition Detector (MarSCoDe). By using the natural gradient boosting probabilistic prediction (NGBoost) ensemble learning method, the probabilistic major element composition (PMEC) models were established and proved to be accurate and robust through testing and validation.