Journal
SPECTROCHIMICA ACTA PART B-ATOMIC SPECTROSCOPY
Volume 162, Issue -, Pages -Publisher
PERGAMON-ELSEVIER SCIENCE LTD
DOI: 10.1016/j.sab.2019.105715
Keywords
LIBS; Light elements; Atmospheric differences; Multivariate analysis; Univariate analysis
Categories
Funding
- Smith College's Praxis program
- NSF [IIS-1564083]
- NASA RIS4E SSERVI grant [NNA14AB04A]
- NASA [NNA14AB04A, 684981] Funding Source: Federal RePORTER
Ask authors/readers for more resources
Laser-induced breakdown spectroscopy (LIBS) is valued for its ability to remotely detect a wide range of elements, including light elements, under a variety of atmospheric conditions. This study uses LIBS spectra of 402 rock standards to quantify lithium (Li), boron (B), carbon/carbon dioxide (C/CO2), and sulfur (S) in Mars and Earth atmospheres and under vacuum. Two regression methods were tested: univariate analysis (UVA), here using peak areas to predict concentrations, and multivariate analysis (MVA). Partial least squares (PLS) and the least absolute shrinkage and selection operator (lasso) use information from larger regions of LIBS spectra. Rock powders were doped with up to 10 wt% of each light element to help identify strongly correlated peaks for UVA. UVA and MVA models were assessed using root mean square errors (RMSEs) of cross-validation (CV), calibration RMSEs, and R-2 correlation between predicted and true concentrations. Li had the most strongly correlated peaks, similar UVA and MVA model performance, and the lowest relative prediction errors. B and C had few weakly-correlated peaks, leading to extremely poor UVA R-2 correlations despite having similar RMSEs to MVA models with mediocre performance. S had no visible peaks in our LIBS setup and as a result, MVA models had extremely high prediction errors. Model performance was not significantly affected by atmospheric differences despite visible changes in peak appearance, as long as model and test data were acquired under identical conditions. PLS regression on the entire LIBS spectrum consistently created models with the lowest quantification errors and highest R-2 correlations. Light element predictions may be improved using higher resolution, gated spectrometers that cover a wider wavelength range than those used in our setup, which matches ChemCam.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available