Towards machine learning-driven estimation of crystal size distribution and morphology in plagioclase: a practical example from the Payún Matrú Volcanic Field, Argentina

Abstract

The textural characterization of minerals in thin sections of volcanic rocks allows the determination of the crystal size distribution (CSD). The study of this distribution provides information on the thermal and dynamic evolution of magma, as well as on nucleation and growth processes, cooling rates, and residence times. The classical analysis is based on logarithmic representations of the distribution as a function of crystal size and allows the recognition of slopes and curvatures sensitive to these magmatic processes. However, the manual obtention of these curves requires exhaustive work, limiting its analytical efficiency. In this context, machine learning methods, particularly deep learning (DL), offer an alternative to automatize the detection and characterization of crystal size in thin sections. This study presents a DL model that detects and classifies plagioclase phenocrysts in porphyritic volcanic rocks where this mineral is the main component. The model generates CSD curves automatically and semi-automatically, reducing processing times while preserving morphometric fidelity. As a preliminary application, the model was implemented on samples from the Payún Matrú Volcanic Field, located in the back-arc basaltic province of Payenia, west-central Argentina. Results show model performance consistent with CSD curves obtained manually and highlight the potential of the method as a complementary tool for textural studies in magmatic systems.