Improved accuracy in quantitative laser-induced breakdown spectroscopy using sub-models

Spectrochimica Acta Part B: Atomic Spectroscopy
By: , and 



Accurate quantitative analysis of diverse geologic materials is one of the primary challenges faced by the Laser-Induced Breakdown Spectroscopy (LIBS)-based ChemCam instrument on the Mars Science Laboratory (MSL) rover. The SuperCam instrument on the Mars 2020 rover, as well as other LIBS instruments developed for geochemical analysis on Earth or other planets, will face the same challenge. Consequently, part of the ChemCam science team has focused on the development of improved multivariate analysis calibrations methods. Developing a single regression model capable of accurately determining the composition of very different target materials is difficult because the response of an element’s emission lines in LIBS spectra can vary with the concentration of other elements. We demonstrate a conceptually simple “sub-model” method for improving the accuracy of quantitative LIBS analysis of diverse target materials. The method is based on training several regression models on sets of targets with limited composition ranges and then “blending” these “sub-models” into a single final result. Tests of the sub-model method show improvement in test set root mean squared error of prediction (RMSEP) for almost all cases. The sub-model method, using partial least squares regression (PLS), is being used as part of the current ChemCam quantitative calibration, but the sub-model method is applicable to any multivariate regression method and may yield similar improvements.
Publication type Article
Publication Subtype Journal Article
Title Improved accuracy in quantitative laser-induced breakdown spectroscopy using sub-models
Series title Spectrochimica Acta Part B: Atomic Spectroscopy
DOI 10.1016/j.sab.2016.12.002
Volume 129
Year Published 2017
Language English
Publisher Elsevier
Publisher location Amsterdam
Contributing office(s) Astrogeology Science Center
Description 9 p.
First page 49
Last page 57
Google Analytic Metrics Metrics page
Additional publication details