9:15 AM - 9:30 AM
[SMP25-02] High mobility of Sr and Ba in subduction-related metamorphism: Application of Machine-learning mass transfer analyses to mafic shists, the Sanbagawa belt
Keywords:Mass transfer, Metamorphism, Regional scale, Machine-learning, Subduction zone
Regional mass transfer in subduction-related metamorphic rocks records fluid activities in subduction zones and serves as a key to understand fluid-related processes such as seismic activity (e.g., Hacker et al., 2003) and magma generation in island arcs (e.g., Pearce et al., 2005). Although geophysical observations suggest that such fluid distributions are in regional scales (i.e., 1–10km), the scales of quantitative mass transfer analyses are usually limited to outcrop scales. Mass transfer analysis on the regional scale has been investigated as chemical trends (e.g., John et al., 2004; Bebout 2007; Masters and Ague 2005); however, protolith heterogeneity hinders quantitative analyses of mass transfer in scales larger than outcrops (i.e., >100 m; e.g., Moss et al., 1995; Bebout 2007; Uno et al., 2014). To understand regional scale metamorphic mass transfer, it is required to determine protolith composition from metamorphic rock sample by sample.
To evaluate metamorphic mass transfer, we have developed protolith compositional estimation models using machine learning. The models were built by learning protolith basalt compositional dataset (Ocean Island basalt, Mid-ocean ridge basalt, and Island arc basalt), and designed to estimate trace element compositions from limited numbers of input trace element concentrations (i.e., 2–9 elements). Gradient Boosting Decision Tree was used as a machine learning algorithm, and models were evaluated by the Root Mean Squared Error in the log unit. With Zr, Th, Ti, and Nb as input elements and Rb, Ba, U, K, La, Ce, Pb, Sr, Nd, Y, Yb, and Lu as output elements, the developed models can reproduce protolith basalt compositions within errors of 10%–25% (i.e., <0.1 in log10 unit). As Zr, Th, Ti, and Nb are the most immobile element in typical regional metamorphism (e.g., Ague et al., 2017), these elements are adopted as input elements. It is noted that models with other input elements are also available. These models were applied to seafloor altered basalt for validation. Compositions of fresh glass remained in the seafloor altered basalt are compared to the estimated protolith compositions, which ensured the model reliability.
The protolith compositional estimation models were applied to the Sanbagawa metamorphic belt, SW Japan. Samples were collected from Asemigawa-river, Central Shikoku, where it crosses chlorite zone (CHZ), garnet zone (GTZ), albite-biotite zone (ABZ), and oligoclase-biotite (OBZ), with increasing metamorphic grade. Mafic rocks are exposed as lenticular bodies or layers, and are collected at each grade for mass transfer analyses. It has been revealed that mafic rocks in CHZ are N-MORB-like compositions, and those in GTZ and OBZ are E-MORB-like compositions (Uno et al., 2014), and such protolith heterogeneity inhibited direct comparisons of trace element compositions among metamorphic grades. To enable direct comparisons of mass transfer among metamorphic grades, the protolith compositional estimation models were applied to the mafic rock samples. It is revealed that both enrichment and depletion had occurred for Rb, K, Sr, and Y compared to their protoliths; Ba is depleted for most of the samples; Not a significant mass transfer was detected for La, Ce, and Nd. Although the degree of mass transfer varies within a metamorphic grade, maximum depletion of Ba and Sr at each grade correlates with metamorphic grades: Ba has been depleted more than 90% at the maximum, Sr has been enriched more than 200% at the maximum, while some samples have been depleted 50% at the minimum. These results provide unique insights into elemental budgets and fluid activities in subduction zones.
To evaluate metamorphic mass transfer, we have developed protolith compositional estimation models using machine learning. The models were built by learning protolith basalt compositional dataset (Ocean Island basalt, Mid-ocean ridge basalt, and Island arc basalt), and designed to estimate trace element compositions from limited numbers of input trace element concentrations (i.e., 2–9 elements). Gradient Boosting Decision Tree was used as a machine learning algorithm, and models were evaluated by the Root Mean Squared Error in the log unit. With Zr, Th, Ti, and Nb as input elements and Rb, Ba, U, K, La, Ce, Pb, Sr, Nd, Y, Yb, and Lu as output elements, the developed models can reproduce protolith basalt compositions within errors of 10%–25% (i.e., <0.1 in log10 unit). As Zr, Th, Ti, and Nb are the most immobile element in typical regional metamorphism (e.g., Ague et al., 2017), these elements are adopted as input elements. It is noted that models with other input elements are also available. These models were applied to seafloor altered basalt for validation. Compositions of fresh glass remained in the seafloor altered basalt are compared to the estimated protolith compositions, which ensured the model reliability.
The protolith compositional estimation models were applied to the Sanbagawa metamorphic belt, SW Japan. Samples were collected from Asemigawa-river, Central Shikoku, where it crosses chlorite zone (CHZ), garnet zone (GTZ), albite-biotite zone (ABZ), and oligoclase-biotite (OBZ), with increasing metamorphic grade. Mafic rocks are exposed as lenticular bodies or layers, and are collected at each grade for mass transfer analyses. It has been revealed that mafic rocks in CHZ are N-MORB-like compositions, and those in GTZ and OBZ are E-MORB-like compositions (Uno et al., 2014), and such protolith heterogeneity inhibited direct comparisons of trace element compositions among metamorphic grades. To enable direct comparisons of mass transfer among metamorphic grades, the protolith compositional estimation models were applied to the mafic rock samples. It is revealed that both enrichment and depletion had occurred for Rb, K, Sr, and Y compared to their protoliths; Ba is depleted for most of the samples; Not a significant mass transfer was detected for La, Ce, and Nd. Although the degree of mass transfer varies within a metamorphic grade, maximum depletion of Ba and Sr at each grade correlates with metamorphic grades: Ba has been depleted more than 90% at the maximum, Sr has been enriched more than 200% at the maximum, while some samples have been depleted 50% at the minimum. These results provide unique insights into elemental budgets and fluid activities in subduction zones.