Skip to content

Amphibole machine learning thermobarometry and chemometry: a general model for igneous rocks

Martín Miranda-Muruzábal1, Oliver Higgins2, Luca Caricchi1

  • Affiliations:  1 Department of Earth Sciences, University of Geneva, Geneva, Switzerland; 2 School of Earth and Environmental Sciences, University of St Andrews, St Andrews, UK 

  • Presentation type: Poster

  • Presentation time: Monday 16:30 - 18:30, Room Poster Hall

  • Poster Board Number: 145

  • Programme No: 3.1.37

  • Theme 3 > Session 1

Abstract

Amphibole thermobarometry and chemometry models can provide quantitative information on the geometry and evolution of a broad variety of igneous systems. However, the chemical complexity of amphibole has hindered many previous attempts to calibrate robust, broadly applicable models; the ability of machine learning (ML) to unravel complexities in large datasets makes it a powerful tool for addressing this challenge. Here we develop a new ML amphibole-only thermobarometer and chemometer that incorporates the latest advancements in ML thermobarometry and is trained on a significantly expanded training dataset (>800 experimental amphibole-melt pairs) compared to the ML thermobarometer and chemometer developed by Higgins et al. (2022). Before calibration, this dataset was rigorously filtered using various statistical techniques, a stage often overlooked in earlier ML studies. The model includes both traditional and novel methods to describe uncertainty associated with each pressure (P), temperature (T) and melt composition (Xmelt) estimate. Additionally, it outputs the experimental amphibole and liquid compositionally closest to each natural amphibole, enabling users to consult the corresponding experimental paper to verify other geochemical features and critically assess the P-T-Xmelt estimate. Moreover, this allows users to reference the article/s with the best matching experimentally synthesised amphiboles. To validate the model, its performance is compared against widely applied amphibole thermometers and barometers (e.g., the amphibole-plagioclase thermometer and Al-in-hornblende barometer) as well as amphiboles from granites with known emplacement pressures. Finally, we apply our new calibration to amphibole-bearing intrusive fragments from the Lesser Antilles island arc, refining the previously reported P-T-Xmelt stratification of the crust.