From classification to understanding of continuity: New expansions in petrology opened by machine learning and powder X-ray diffraction

Abstract

Machine learning (ML) techniques are powerful tools for extracting continuous trends in geological processes from numerical datasets. To date, however, ML use has been largely restricted to classification tasks for geochemical datasets, and few studies have attempted regression tasks. In addition, the general scarcity of numerical data on rock observations—such as whole-rock mineral modes, mineral shape and distribution, and fracture-pattern statistics—has prevented broader ML applications in geology. Here, I propose two approaches to capture continuous geological processes as data-driven geology. First, I present a regression-based Protolith Reconstruction Model trained on a compositional dataset of basalt. Second, I introduce a methodology for the rapid numerical representation of whole-rock mineral information via powder X-ray diffraction. Finally, I propose the integrated approach of ML and numerical datasets to effectively and objectively capture the continuity of multidimensional geological processes.