Correlating tephras and cryptotephras using glass compositional analyses and numerical and statistical methods: Review and evaluation. (1st November 2017)

• Record Type: Journal Article
• Authors: Lowe, David J.; Pearce, Nicholas J.G.; Jorgensen, Murray A.; Kuehn, Stephen C.; Tryon, Christian A.; Hayward, Chris L.
• Is Part Of: Quaternary science reviews. Volume 175(2017)
• Page Start: 1
• Page End: 44
• Publication Date: 2017-11-01
• Subjects: Tephra -- Cryptotephra -- Tephrochronology -- Tephrostratigraphy -- Glass-shard analysis -- Microshard -- Volcanic glass -- Crystal -- Microlite -- Melt inclusion -- Electron probe -- Laser ablation -- LA-ICP-MS -- EPMA -- Multivariate statistics -- Similarity coefficients -- Machine learning -- Cluster analysis -- Statistical distance -- Bivariate plot; ANOVA analysis of variance -- AVF Auckland Volcanic Field -- BSE back-scattered electron (image) -- CV coefficient of variation -- CVA canonical variates analysis -- DFA discriminant function analysis -- EPMA electron probe microanalysis -- FMAZ Faroe Marine Ash Zone -- HFSE high field strength elements -- INTAV International Focus Group (IFG) on Tephrochronology and Volcanism -- IS internal standard -- LA-ICP-MS laser ablation inductively coupled plasma-mass spectrometry -- LLQ lower limit of quantitation -- MANOVA multivariate analysis of variance -- μ-XRF micro-X-ray fluorescence -- MTT Middle Toba Tuff -- OTT Oldest Toba Tuff -- PAM partitioning around medoids -- PC principal component -- PCA principal components analysis -- REEs rare earth elements -- SC similarity coefficient -- SEM scanning electron microscope -- SN-ICP-MS solution nebulisation inductively coupled plasma-mass spectrometry -- SVM support vector machines -- YTT Youngest Toba Tuff -- XRF X-ray fluorescence; Geology, Stratigraphic -- Quaternary -- Periodicals; Stratigraphie -- Quaternaire -- Périodiques; 551.79
• DOI: 10.1016/j.quascirev.2017.08.003
• Languages: English
• ISSNs: 0277-3791

Abstract

Abstract: We define tephras and cryptotephras and their components (mainly ash-sized particles of glass ± crystals in distal deposits) and summarize the basis of tephrochronology as a chronostratigraphic correlational and dating tool for palaeoenvironmental, geological, and archaeological research. We then document and appraise recent advances in analytical methods used to determine the major, minor, and trace elements of individual glass shards from tephra or cryptotephra deposits to aid their correlation and application. Protocols developed recently for the electron probe microanalysis of major elements in individual glass shards help to improve data quality and standardize reporting procedures. A narrow electron beam (diameter ∼3–5 μm) can now be used to analyze smaller glass shards than previously attainable. Reliable analyses of 'microshards' (defined here as glass shards <32 μm in diameter) using narrow beams are useful for fine-grained samples from distal or ultra-distal geographic locations, and for vesicular or microlite-rich glass shards or small melt inclusions. Caveats apply, however, in the microprobe analysis of very small microshards (≤∼5 μm in diameter), where particle geometry becomes important, and of microlite-rich glass shards where the potential problem of secondary fluorescence across phase boundaries needs to be recognised. Trace element analyses of individual glass shards using laser ablation inductively coupled plasma-mass spectrometry (LA-ICP-MS), with crater diameters of 20 μm and 10 μm, are now effectively routine, giving detection limits well below 1 ppm. Smaller ablation craters (<10 μm) can be subject to significant element fractionation during analysis, but the systematic relationship of such fractionation with glass composition suggests that analyses for some elements at these resolutions may be quantifiable. In undertaking analyses, either by microprobe or LA-ICP-MS, reference material data acquired using the same procedure, and preferably from the same analytical session, should be presented alongside new analytical data. In part 2 of the review, we describe, critically assess, and recommend ways in which tephras or cryptotephras can be correlated (in conjunction with other information) using numerical or statistical analyses of compositional data. Statistical methods provide a less subjective means of dealing with analytical data pertaining to tephra components (usually glass or crystals/phenocrysts) than heuristic alternatives. They enable a better understanding of relationships among the data from multiple viewpoints to be developed and help quantify the degree of uncertainty in establishing correlations. In common with other scientific hypothesis testing, it is easier to infer using such analysis that two or more tephras are different rather than the same. Adding stratigraphic, chronological, spatial, or palaeoenvironmental data (i.e. multiple criteria) is usually necessary and allows for more robust correlations to be made. A two-stage approach is useful, the first focussed on differences in the mean composition of samples, or their range, which can be visualised graphically via scatterplot matrices or bivariate plots coupled with the use of statistical tools such as distance measures, similarity coefficients, hierarchical cluster analysis (informed by distance measures or similarity or cophenetic coefficients), and principal components analysis (PCA). Some statistical methods (cluster analysis, discriminant analysis) are referred to as 'machine learning' in the computing literature. The second stage examines sample variance and the degree of compositional similarity so that sample equivalence or otherwise can be established on a statistical basis. This stage may involve discriminant function analysis (DFA), support vector machines (SVMs), canonical variates analysis (CVA), and ANOVA or MANOVA (or its two-sample special case, the Hotelling two-sample T 2 test). Randomization tests can be used where distributional assumptions such as multivariate normality underlying parametric tests are doubtful. Compositional data may be transformed and scaled before being subjected to multivariate statistical procedures including calculation of distance matrices, hierarchical cluster analysis, and PCA. Such transformations may make the assumption of multivariate normality more appropriate. A sequential procedure using Mahalanobis distance and the Hotelling two-sample T 2 test is illustrated using glass major element data from trachytic to phonolitic Kenyan tephras. All these methods require a broad range of high-quality compositional data which can be used to compare 'unknowns' with reference (training) sets that are sufficiently complete to account for all possible correlatives, including tephras with heterogeneous glasses that contain multiple compositional groups. Currently, incomplete databases are tending to limit correlation efficacy. The development of an open, online global database to facilitate progress towards integrated, high-quality tephrostratigraphic frameworks for different regions is encouraged. Highlights: Advances in the microanalysis of major, minor, and trace elements of glass shards are reviewed. We evaluate numerical and statistical methods for tephra correlation via glass/crystal analyses. We focus on (1) differences in mean composition of samples or their range; and. (2) sample variance and degree of compositional similarity to establish equivalence or not. We illustrate various statistical methods and data transformations using case studies.