Advanced statistical analysis of laser-induced breakdown spectroscopy data to discriminate sedimentary rocks based on Czerny–Turner and Echelle spectrometers

Abstract The correct identification of rock types is critical for understanding the origins and history of any particular rock body. Laser-induced breakdown spectroscopy (LIBS) has developed into an excellent analytical tool for geological materials research because of its numerous technical advantages compared with traditional methods. The coupling of LIBS with advanced multivariate analysis has received increasing attention because it facilitates the rapid processing of spectral information to differentiate and classify samples. In this study, we collected LIBS datasets for 16 sedimentary rocks from Triassic strata in Sichuan Basin. We compared the performance of two types of spectrometers (Czerny–Turner and Echelle) for classification of rocks using two advanced multivariate statistical techniques, i.e., partial least squares discriminant analysis (PLS-DA) and support vector machines (SVMs). Comparable levels of performance were achievable when using the two systems in the best signal reception conditions. Our results also suggest that SVM outperformed PLS-DA in classification performance. Then, we compared the results obtained when using pre-selected wavelength variables and broadband LIBS spectra as variable inputs. They provided approximately equivalent levels of performance. In addition, the rock slab samples were also analyzed directly after being polished. This minimized the analysis time greatly and showed improvement of classification performance compared with the pressed pellets.