Effects of univariate and multivariate regression on the accuracy of hydrogen quantification with laser-induced breakdown spectroscopy

被引:20
作者
Ytsma, Cai R. [1 ]
Dyar, M. Darby [2 ]
机构
[1] Smith Coll, Dept Chem, Northampton, MA 01063 USA
[2] Mt Holyoke Coll, Dept Astron, S Hadley, MA 01075 USA
关键词
Libs; ChemCam; Hydrogen; Partial least-squares; Lasso; CHEMCAM INSTRUMENT SUITE; NEUTRONS DAN EXPERIMENT; GALE CRATER; DYNAMIC ALBEDO; MARS; SYSTEM; WATER; ROVER; CLASSIFICATION; PREDICTION;
D O I
10.1016/j.sab.2017.11.010
中图分类号
O433 [光谱学];
学科分类号
0703 ; 070302 ;
摘要
Hydrogen (H) is a critical element to measure on the surface of Mars because its presence in mineral structures is indicative of past hydrous conditions. The Curiosity rover uses the laser-induced breakdown spectrometer (LIBS) on the ChemCam instrument to analyze rocks for their H emission signal at 656.6 nm, from which H can be quantified. Previous LIBS calibrations for H used small data sets measured on standards and/or manufactured mixtures of hydrous minerals and rocks and applied univariate regression to spectra normalized in a variety of ways. However, matrix effects common to LIBS make these calibrations of limited usefulness when applied to the broad range of compositions on the Martian surface. In this study, 198 naturally-occurring hydrous geological samples covering a broad range of bulk compositions with directly-measured H content are used to create more robust prediction models for measuring H in LIBS data acquired under Mars conditions. Both univariate and multivariate prediction models, including partial least square (PLS) and the least absolute shrinkage and selection operator (Lasso), are compared using several different methods for normalization of H peak intensities. Data from the ChemLIBS Mars-analog spectrometer at Mount Holyoke College are compared against spectra from the same samples acquired using a ChemCam-like instrument at Los Alamos National Laboratory and the ChemCam instrument on Mars. Results show that all current normalization and data preprocessing variations for quantifying H result in models with statistically indistinguishable prediction errors (accuracies) ca +/- 1.5 weight percent (wt%) H2O, limiting the applications of LIBS in these implementations for geological studies. This error is too large to allow distinctions among the most common hydrous phases (basalts, amphiboles, micas) to be made, though some clays (e.g., chlorites with approximate to 12 wt% H2O, smectites with 15-20 wt% H2O) and hydrated phases ( e.g., gypsum with 20 wt% H2O) may be differentiated from lower-H phases within the known errors. Analyses of the H emission peak in Curiosity calibration targets and rock and soil targets on the Martian surface suggest that shot-to-shot variations of the ChemCam laser on Mars lead to variations in intensity that are comparable to those represented by the breadth of H standards tested in this study. (C) 2017 Elsevier B.V. All rights reserved.
引用
收藏
页码:27 / 37
页数:11
相关论文
共 73 条
  • [1] Aharonson O, 2006, J GEOPHYS RES-PLANET, V111, DOI [10.1029/2005JE002636, 10.1029/2005JE002636.]
  • [2] Anderson D.E., 2015, LUNAR PLANET SCI, VXLVI
  • [3] Improved accuracy in quantitative laser-induced breakdown spectroscopy using sub-models
    Anderson, Ryan B.
    Clegg, Samuel M.
    Frydenvang, Jens
    Wiens, Roger C.
    McLennan, Scott
    Morris, Richard V.
    Ehlmann, Bethany
    Dyar, M. Darby
    [J]. SPECTROCHIMICA ACTA PART B-ATOMIC SPECTROSCOPY, 2017, 129 : 49 - 57
  • [4] Correlating multispectral imaging and compositional data from the Mars Exploration Rovers and implications for Mars Science Laboratory
    Anderson, Ryan B.
    Bell, James F., III
    [J]. ICARUS, 2013, 223 (01) : 157 - 180
  • [5] Clustering and training set selection methods for improving the accuracy of quantitative laser induced breakdown spectroscopy
    Anderson, Ryan B.
    Bell, James F., III
    Wiens, Roger C.
    Morris, Richard V.
    Clegg, Samuel M.
    [J]. SPECTROCHIMICA ACTA PART B-ATOMIC SPECTROSCOPY, 2012, 70 : 24 - 32
  • [6] The influence of multivariate analysis methods and target grain size on the accuracy of remote quantitative chemical analysis of rocks using laser induced breakdown spectroscopy
    Anderson, Ryan B.
    Morris, Richard V.
    Clegg, Samuel M.
    Bell, James F., III
    Wiens, Roger C.
    Humphries, Seth D.
    Mertzman, Stanley A.
    Graff, Trevor G.
    McInroy, Rhonda
    [J]. ICARUS, 2011, 215 (02) : 608 - 627
  • [7] Sparse models by iteratively reweighted feature scaling: a framework for wavelength and sample selection
    Andries, Erik
    [J]. JOURNAL OF CHEMOMETRICS, 2013, 27 (3-4) : 50 - 62
  • [8] Blank J.G., 2015, LUNAR PLANET SCI, VXLVI
  • [9] Boucher T.F., 2016, LUNAR PLANET SCI, VXLVII
  • [10] A study of machine learning regression methods for major elemental analysis of rocks using laser-induced breakdown spectroscopy
    Boucher, Thomas F.
    Ozanne, Marie V.
    Carmosino, Marco L.
    Dyar, M. Darby
    Mahadevan, Sridhar
    Breves, Elly A.
    Lepore, Kate H.
    Clegg, Samuel M.
    [J]. SPECTROCHIMICA ACTA PART B-ATOMIC SPECTROSCOPY, 2015, 107 : 1 - 10