[M-GI34] Data-driven geosciences

[MGI34-04] Relationships between Al-Cr chemical zoning and deformation mechanisms of spinel: An approach applying machine learning analysis

*Tae-Hoon Uhmb¹, Katsuyoshi Michibayashi¹ (1.Department of Earth and Planetary Sciences, Research group of Petrology and Mineralogy, Nagoya University)

Keywords:Spinel, Chemical zoning, Lattice diffusion, Machine learning

Chemical zoning of mineral, which results from incomplete chemical reaction to keep chemical equilibrium, commonly used as evidence for interpreting compositions, cooling processes of magma, and reaction path of metamorphic rocks. However, Ozawa [1989] firstly reported Al-Cr chemical zoning of elongated spinel derived by deformation (lattice diffusion) from natural deformed peridotites. More recently, Suzuki et al. [2008] elucidated the processes of lattice diffusion induced Al-Cr chemical zoning of spinel by estimating self-diffusion coefficients of Cr and Al from Cr-Al interdiffusion experiment. In this study, we present relationships between the Al-Cr chemical zoning and geometrical properties of spinel grains by using machine learning algorithms. Moreover, we discuss connections between the relationships and deformation mechanisms considering results of the machine learning. To analyze Al-Cr chemical distributions within each spinel grain, we estimate Counts Per Second (cps) of EDS for 87 spinel grains within a dunite sample of the Horoman peridotite complex by using SEM-EDS line detection. For applying the estimated data sets to the machine learning algorithms, the following sequence is applied: analyzing chemical zoning, data pre-processing, data clustering, data splitting, classification and node analysis, and estimating feature importance. As analyzing result, the Al-Cr chemical distributions are clustered as three groups in accordance with intensity of the chemical zoning. The intensity is more importantly affected by grain size than aspect ratio of spinel grain and is much greater with increasing grain size. These results reflect that lattice diffusion is much actively contributed to deformation of spinel than grain boundary diffusion according to increasing grain size. And the connection between the deformation mechanisms and grain size can be explained by diffusivity ratio of spinel.