Accuracies and detection limits of major, minor, and trace element quantification in rocks by portable laser-induced breakdown spectroscopy

The recent manufacture of portable laser-induced breakdown spectroscopy (pLIBS) instruments has permitted widespread applications of an established analytical technique for in situ geochemical quantifications. This novelty has created a need for investigations comparing the default univariate pLIBS calibration accuracies to custom instrument calibrations based on geological samples and to other modeling methods. Previous research has shown that multivariate models can mitigate some of the matrix effects that cause intensity and concentration non-linearity within LIBS plasmas of rocks. This study thus makes two pLIBS calibration comparisons: one compares partial least squares (PLS) multivariate regression to univariate instrument models, while the other compares the default instrument calibration, intended for general industrial use, to a custom instrument calibration based on rock standards. Both comparisons use root mean squared errors (RMSE) between model-predicted and true standard values of the major oxides Al2O3, CaO, Fe2O3, K2O, MgO, MnO, Na2O, P2O5, SiO2, and TiO2, and minor and trace elements Ba, Cr, Cu, H2O, Li, Ni, Pb, S, and Zn to test calibration accuracies. pLIBS spectra from 2007 diverse rock standards were used for the PLS versus univariate model comparison, while 89 publicly-available geochemical standards were used to compare the default and custom instrument calibrations. PLS models give test accuracies for major elements of +/- 8.61 wt% for SiO2, +/- 3.38 for Al2O3, +/- 0.77 for TiO2, +/- 3.84 for Fe2O3, +/- 4.41 wt% for MgO, +/- 3.08 wt% for CaO, +/- 1.09 wt% for Na2O, and +/- 1.52 wt% for K2O. Quantification of trace and minor elements was more accurate with multivariate regression than with univariate instrument models, and the new custom instrument calibration had lower prediction errors for these elements than those of the default models. However, minor and trace element errors from all three calibrations were often greater than the average concentration of the test standards, which suggests that these substances are not well quantified in rocks by pLIBS instruments.