1:45 PM - 3:15 PM
[HQR03-P01] Study about correlation method of tephra based on chemical composition of volcanic glass: an example of Plio-Pleistocene tephras in the Hokuriku area
Keywords:correlation of tephra, multivariate analysis, log ratio analysis
1. Introduction
Correlation of tephra is based on a comprehensive review of stratigraphic and petrographic methods, and specific methods for the latter include total mineral composition analysis, chemical composition analysis of volcanic glasses and minerals, and refractive index measurements. Among these, the method of using chemical composition data of volcanic glass is reviewed in Lowe et al. (2017), especially the statistical analysis of chemical composition data is organized. The process is based on the examination, transformation, exploration, and verification of chemical composition data, and the comparison with stratigraphic and petrological characteristics. The examination is the removal of outliers, the transformation is the conversion of analytical values to other format, the exploration is the estimation of tephra combinations that may be correlated, and the verification is the test of whether the multivariate distribution of the chemical composition of volcanic glass is equal for the tephras that may be contrasted. However, there are several unclear points in this procedure. First, it is unclear whether a log ratio analysis is required or not in the transformation. There is also insufficient discussion of the chemical components that should be emphasized in the analysis. Furthermore, the methods recommended in the verification assume normality and equal variances of the data, and thus do not accommodate a wide variety of data. In this study, we examine these uncertainties using tephras distributed in the Pliocene-Pleistocene in the Hokuriku area, Japan.
2. Methods
In this study, we used principal chemical composition data (N≧100) of volcanic glasses from three Plio-Pleistocene tephra layers (Hb02, Hb04, and Hb12) of different horizon distributed along route at Hachibuse, Oyabe City, Toyama Prefecture to verify the above issues. Of the three tephra layers, Hb02 and Hb12 are clearly distinguishable by FeO* in both compositional/log ratio data, while the FeO* of Hb04 shows a bimodal distribution (one overlaps with Hb02 and the other is independent). Therefore, we divided Hb04 into a and b, respectively, and treated them as four groups in total, and searched for a method in which the multivariate distribution would be Hb02≈Hb04a≠Hb04b≠Hb12. The data sets were explored and verified by multivariate analysis using four combinations of composition data/log ratio data and total components/selected components. For the log ratio analysis, it was applied an additive log ratio transformation with Al2O3 as the normalized component. The selected components were SiO2, Al2O3, FeO*, and K2O, considering the lower limit of quantification and mobility due to weathering. Multivariate analysis was performed using principal component analysis for the exploration and discriminant analysis with decision trees for the verification.
3. Results and Discussion.
Multivariate analysis of the four data sets showed that the only combination that resulted in Hb02≈Hb04a≠Hb04b≠Hb12 was the log ratio/selected component data set. When all components were used, principal component analysis showed that Hb02 ≈ Hb04a ≈ Hb04b ≠ Hb12 for both composition and log ratio data. In the decision tree, Hb02 ≈ Hb04a ≠ Hb04b ≠ Hb12 in all data sets, but only in the log ratio/selected component data set was it clearly classified by FeO* alone. The results of this study suggest that log ratio analysis and component selection should be applied to tephra correlation using chemical composition data of volcanic glass. The results of this study suggest that log-ratio analysis and component selection should be applied to tephra contrast using chemical composition data of volcanic glass. The results of this study suggest that log ratio analysis and component selection should be applied to correlation of tephra using chemical composition data of volcanic glass. All of the multivariate analysis methods used in this study are easy to apply and do not require the consideration of stringent assumptions. The decision tree method is particularly useful in that it allows for quantitative and clear visualization of differences in chemical composition.
Correlation of tephra is based on a comprehensive review of stratigraphic and petrographic methods, and specific methods for the latter include total mineral composition analysis, chemical composition analysis of volcanic glasses and minerals, and refractive index measurements. Among these, the method of using chemical composition data of volcanic glass is reviewed in Lowe et al. (2017), especially the statistical analysis of chemical composition data is organized. The process is based on the examination, transformation, exploration, and verification of chemical composition data, and the comparison with stratigraphic and petrological characteristics. The examination is the removal of outliers, the transformation is the conversion of analytical values to other format, the exploration is the estimation of tephra combinations that may be correlated, and the verification is the test of whether the multivariate distribution of the chemical composition of volcanic glass is equal for the tephras that may be contrasted. However, there are several unclear points in this procedure. First, it is unclear whether a log ratio analysis is required or not in the transformation. There is also insufficient discussion of the chemical components that should be emphasized in the analysis. Furthermore, the methods recommended in the verification assume normality and equal variances of the data, and thus do not accommodate a wide variety of data. In this study, we examine these uncertainties using tephras distributed in the Pliocene-Pleistocene in the Hokuriku area, Japan.
2. Methods
In this study, we used principal chemical composition data (N≧100) of volcanic glasses from three Plio-Pleistocene tephra layers (Hb02, Hb04, and Hb12) of different horizon distributed along route at Hachibuse, Oyabe City, Toyama Prefecture to verify the above issues. Of the three tephra layers, Hb02 and Hb12 are clearly distinguishable by FeO* in both compositional/log ratio data, while the FeO* of Hb04 shows a bimodal distribution (one overlaps with Hb02 and the other is independent). Therefore, we divided Hb04 into a and b, respectively, and treated them as four groups in total, and searched for a method in which the multivariate distribution would be Hb02≈Hb04a≠Hb04b≠Hb12. The data sets were explored and verified by multivariate analysis using four combinations of composition data/log ratio data and total components/selected components. For the log ratio analysis, it was applied an additive log ratio transformation with Al2O3 as the normalized component. The selected components were SiO2, Al2O3, FeO*, and K2O, considering the lower limit of quantification and mobility due to weathering. Multivariate analysis was performed using principal component analysis for the exploration and discriminant analysis with decision trees for the verification.
3. Results and Discussion.
Multivariate analysis of the four data sets showed that the only combination that resulted in Hb02≈Hb04a≠Hb04b≠Hb12 was the log ratio/selected component data set. When all components were used, principal component analysis showed that Hb02 ≈ Hb04a ≈ Hb04b ≠ Hb12 for both composition and log ratio data. In the decision tree, Hb02 ≈ Hb04a ≠ Hb04b ≠ Hb12 in all data sets, but only in the log ratio/selected component data set was it clearly classified by FeO* alone. The results of this study suggest that log ratio analysis and component selection should be applied to tephra correlation using chemical composition data of volcanic glass. The results of this study suggest that log-ratio analysis and component selection should be applied to tephra contrast using chemical composition data of volcanic glass. The results of this study suggest that log ratio analysis and component selection should be applied to correlation of tephra using chemical composition data of volcanic glass. All of the multivariate analysis methods used in this study are easy to apply and do not require the consideration of stringent assumptions. The decision tree method is particularly useful in that it allows for quantitative and clear visualization of differences in chemical composition.