Machine learning thermobarometry and chemometry using amphibole and clinopyroxene: a window into the roots of an arc volcano (Mount Liamuiga, Saint Kitts)

Abstract

The physical and chemical properties of magma govern the eruptive style and behaviour of volcanoes. Many of these parameters are linked to the storage pressure and temperature of the erupted magma, and melt chemistry. However, reliable single-phase thermobarometers and chemometers which can recover this information, particularly using amphibole chemistry, remain elusive. We present a suite of single-phase amphibole and clinopyroxene thermobarometers and chemometers, calibrated using machine learning. This approach allows us to intimately track the range of pre-eruptive conditions over the course of a millennial eruptive cycle on an island arc volcano (Saint Kitts, Eastern Caribbean). We unpick the story of Mount Liamuiga, a stratovolcano that pops its upper-crustal (2 kbar), dacitic cork at the beginning of the Lower Mansion Series eruptive sequence. This permits a progressive increase in the thermal maturity of the magma arriving at the surface from the middle to upper crust (2 – 5.5 kbar) through time. The temperature increase correlates well with matrix plagioclase chemistry, which itself displays a remarkable progression to less evolved (more anorthitic) composition in time. We find that amphibole is a reliable themobarometer (SEE = 1.4 kbar; 40 ˚C), at odds with previous studies. We suggest it is the regression strategy, as opposed to the abject insensitivity to pressure, that has hindered previous calibrations of amphibole only thermobarometers. By recognising this, we have constructed a high-resolution, quantitative picture of the magma plumbing system beneath an arc volcano.

A Machine Learning‐Based Approach to Clinopyroxene Thermobarometry: Model Optimization and Distribution for Use in Earth Sciences

“…We have shown that machine learning is a powerful and versatile approach to thermobarometry in agreement with other studies (Higgins et al., 2022; Petrelli et al., 2020). Through detailed testing, we have determined models that have an SEE comparable to the leading clinopyroxene thermobarometers (SEE of 3.2 kbar, 47.6°C and 4.4 kbar, 76.0°C for the liquid and no liquid models, respectively, as compared to 3.4 kbar and 125°C for the alkaline only liquid‐cpx models of Masotta et al.…”

“…Thus, the twofold aim of this work is to (a) build and test the performance of a thermobarometer model for clinopyroxenes and (b) provide a comprehensive explanation of how to apply our thermobarometer for applications to natural data. Our regression strategy offers a generalized model that can be tailored for certain settings, applications, or other suitable mineral phases (e.g., amphibole; Higgins et al., 2022). We greatly expand the data set of clinopyroxene and liquid equilibria in our calibration data set compared to previous studies, allowing our calibrations to be as globally applicable and adaptable as possible for users.…”

“…(2020) and Higgins et al. (2022) have resulted in a machine learning random forest approach to thermobarometry. Both studies omitted several pertinent experimental data sets of clinopyroxene and liquid equilibria, which are now included in the model presented here.…”

Machine Learning Thermobarometry for Biotite‐Bearing Magmas

“…31 of Putirka (2008); T estimated using Cpx thermometry of Higgins et al. (2021)], Mgt‐Ilm thermometry (Evans et al., 2016), Ti‐in‐Qtz barometry (Zhang et al., 2020), rhyolite‐MELTS modeling of cotectic conditions (Gualda & Ghiorso, 2013) and fluid saturation barometry (Wallace et al., 1999). EBT: Early Bishop Tuff; LBT: Late Bishop Tuff.…”

“…In such a case, the traditional approach for establishing thermobarometry via a linear regression for retrieving temperature and/or pressure of a consistent experimental dataset does not work quite well (Figure 1). Alternatively, some recent studies have demonstrated that the data‐driven machine learning method is a powerful tool for solving the complexity of crystal chemistry (e.g., Li et al., 2020) and for establishing mineral‐based thermobarometry (e.g., Petrelli et al., 2020; Higgins et al., 2021; Thomson et al., 2021; Jorgenson et al., 2022). In this study, based on a collected experimental dataset associated with biotite, we evaluated different machine learning algorithms as potential thermometers and barometers based on biotite‐only or biotite + melt compositions, and the optimized algorithm using extremely randomized trees has been applied to three biotite‐bearing volcanic systems.…”