5:15 PM - 7:15 PM
[SCG60-P14] Evaluation of estimation methods for physicochemical information of equilibrium melt by multivariate analysis of high-T and high-P experimental data of clinopyroxene
Keywords:clinopyroxene, high-temperature and high-pressure experimental data, machine learning, Sakurajima Volcano, equilibrium melt
It has been considered that most andesitic magmas are generated through the mixing of different magmas (Sakuyama, 1981). In this case, it is difficult to quantitatively constrain the origin of the end-member magmas and their mutual relationships from the whole-rock chemical composition. On the other hand, phenocrysts record the evolution process of magma from crystallization to eruption. To estimate the temperature and pressure of magma from the chemical composition of the phenocryst, geothermometers such as pyroxene thermometer (eg., Lindsley, 1983; Putirka, 2008), amphibole single-phase thermometer (e.g., Putirka, 2016), and amphibole single-phase barometer (e.g., Ridolfi & Renzulli, 2012) have been used. However, there are many constraints on the prerequisites for accurate temperature estimation (e.g., pyroxene thermometer). In addition, andesite does not necessarily contain amphibole. Therefore, investigations using the above-mentioned methods have limitations in understanding magma activity in subduction zones where andesite is the most dominant. Recently, the chemical properties of clinopyroxene has attracted attention, which is more widely found in volcanic rocks than amphibole, and empirical formulas have been proposed to estimate the temperature and pressure of the equilibrium melt (e.g., Nimis, 1995; Nimis & Ulmer, 1998; Nimis, 1999; Nimis & Taylor, 2000; Wang et al., 2021). In addition, a method has been introduced to estimate not only the temperature, pressure but also major element composition of the equilibrium melt by multivariate analysis of high-T and high-P experimental data of clinopyroxene using machine learning (Petrelli et al., 2020; Higgins et al., 2021; Jorgenson et al., 2022; Chicchi et al., 2023). If these methods can be applied to natural andesite, it will be possible to estimate the physicochemical information of the equilibrium melt using the major element composition of clinopyroxene.
In this study, we applied the above-mentioned empirical formulas and machine learning estimation methods to estimate the physicochemical information of equilibrium melt from the major element compositions of clinopyroxene in volcanic rocks of Sakurajima Volcano, a Quaternary volcano on the Ryukyu Arc. Sakurajima Volcano is still active, and many geophysicochemical investigations are being conducted on it. Therefore, the purpose of this study is to evaluate whether the estimation methods function as a tool to estimate the temperature and pressure and the major element composition of the equilibrium melt using clinopyroxene from Sakurajima Volcano by comparing whether the results of the estimation using the above methods are consistent with the physical and chemical information of magma reported in previous investigations.
The major chemical compositions of approximately 200 clinopyroxenes were measured by EPMA, and the temperature and pressure of the equilibrium melt were estimated using machine learning, and it was found that all clinopyroxenes crystallized at temperatures of 956-1147℃ and pressures of up to 654 MPa (depth of ≃29 km). However, even for a single clinopyroxene, the estimated temperature and pressure differed depending on the method, and the uncertainty and error of the estimation results were large. In addition, the pressure estimation results using the empirical formulas were unnatural, such as negative values in some cases and pressure estimates that greatly exceeded the magma accumulation depth reported in previous investigations (eg., Araya et al., 2019). Meanwhile, machine learning estimation of the major element composition of equilibrium melt showed similar trends to the changes in chemical composition generally observed in natural samples, except for Na2O. From these results, we concluded that estimating the temperature and pressure of the equilibrium melt from clinopyroxenes of Sakurajima Volcano is not practical at present, due to the heterogeneity of the estimation methods, which show different results for each method. In this study, we consider that the unnatural trends in temperature and pressure estimation and the differences in estimation results between methods are due to the differences in the number and types of high-T and high-P experimental data used in each estimation method, the mechanism of the Random-forest algorithm, and the use of data on clinopyroxenes with chemical compositions that are not applicable to each method, and we evaluate the reliability, accuracy, and limitations of each estimation method.
In this study, we applied the above-mentioned empirical formulas and machine learning estimation methods to estimate the physicochemical information of equilibrium melt from the major element compositions of clinopyroxene in volcanic rocks of Sakurajima Volcano, a Quaternary volcano on the Ryukyu Arc. Sakurajima Volcano is still active, and many geophysicochemical investigations are being conducted on it. Therefore, the purpose of this study is to evaluate whether the estimation methods function as a tool to estimate the temperature and pressure and the major element composition of the equilibrium melt using clinopyroxene from Sakurajima Volcano by comparing whether the results of the estimation using the above methods are consistent with the physical and chemical information of magma reported in previous investigations.
The major chemical compositions of approximately 200 clinopyroxenes were measured by EPMA, and the temperature and pressure of the equilibrium melt were estimated using machine learning, and it was found that all clinopyroxenes crystallized at temperatures of 956-1147℃ and pressures of up to 654 MPa (depth of ≃29 km). However, even for a single clinopyroxene, the estimated temperature and pressure differed depending on the method, and the uncertainty and error of the estimation results were large. In addition, the pressure estimation results using the empirical formulas were unnatural, such as negative values in some cases and pressure estimates that greatly exceeded the magma accumulation depth reported in previous investigations (eg., Araya et al., 2019). Meanwhile, machine learning estimation of the major element composition of equilibrium melt showed similar trends to the changes in chemical composition generally observed in natural samples, except for Na2O. From these results, we concluded that estimating the temperature and pressure of the equilibrium melt from clinopyroxenes of Sakurajima Volcano is not practical at present, due to the heterogeneity of the estimation methods, which show different results for each method. In this study, we consider that the unnatural trends in temperature and pressure estimation and the differences in estimation results between methods are due to the differences in the number and types of high-T and high-P experimental data used in each estimation method, the mechanism of the Random-forest algorithm, and the use of data on clinopyroxenes with chemical compositions that are not applicable to each method, and we evaluate the reliability, accuracy, and limitations of each estimation method.