2:00 PM - 2:15 PM
[SMP21-02] Cycling of groundwaters and deep-seated fluids beneath Kyushu: unsupervised machine learning of geochemical data
★Invited Papers
Keywords:Japan arc, Kyushu, subduction, fluid, machine learning
Geofluids and their cycling in the crust-mantle system are important in subduction zones, where subducting plates may supply aqueous fluids to cause metasomatism, ore formation, metamorphism, magmatism, and crustal deformation including earthquakes. The origin of geofluids and the processes during cycling can be recorded in geochemistry of geofluids, including groundwater and deep-seated fluid such as the Arima-type brine (Matsubaya et al., 1973; Kusuda et al., 2014). In turn, by analyzing the geochemistry of geofluids, subduction zone processes and their roots as well as circulation of groundwaters at shallower levels may be resolved. In this study we present such an attempt in Kyushu, southwest Japan, based on geochemistry of groundwater and deep-seated fluid.
Multivariate statistical analysis of geochemical data is useful for extracting independent factors that produce compositional variability of the mantle (Zindler et al., 1982; Allegre et al., 1987), arc magmas (Nakamura et al., 2019), and groundwater and deep-seated fluid (Iwamori et al., 2020). Principal Component Analysis (PCA) is one of the most popular methods of multivariate analysis, but when the data distribution does not follow a multivariate normal distribution (joint Gaussian distribution), it is not possible to extract independent features hidden in the data. Instead, other appropriate methods, e.g., unsupervised machine learning approaches of Independent Component Analysis (ICA) and whitened data-based K-means Cluster Analysis (whitened KCA), must be used (Iwamori et al., 2017). Iwamori et al. (2023) demonstrate that KCA are indeed useful for discriminating the origin and cycling processes of meteoric water, deep-seated fluid, and their reactions with the host rocks during cycling, based on 590 water samples and 12 major solutes with temperature, pH, O-H stable isotopic ratio, and He isotopic ratio. In addition, the geochemical data of 438 samples from the southern Kyushu, including several time-series data from the same locations, were analyzed to infer the origin and cycling processes of the groundwater and deep-seated fluid. In this study, we aim to integrate these data for a total of 1028 water samples and discuss the origin and cycling beneath the wide area from central to southern Kyushu, based on the appropriate multivariate statistical method. This research was supported by the Nuclear Regulation Authority "Research on Knowledge Development of Giant Eruption Processes, FY 2021-2023"
Multivariate statistical analysis of geochemical data is useful for extracting independent factors that produce compositional variability of the mantle (Zindler et al., 1982; Allegre et al., 1987), arc magmas (Nakamura et al., 2019), and groundwater and deep-seated fluid (Iwamori et al., 2020). Principal Component Analysis (PCA) is one of the most popular methods of multivariate analysis, but when the data distribution does not follow a multivariate normal distribution (joint Gaussian distribution), it is not possible to extract independent features hidden in the data. Instead, other appropriate methods, e.g., unsupervised machine learning approaches of Independent Component Analysis (ICA) and whitened data-based K-means Cluster Analysis (whitened KCA), must be used (Iwamori et al., 2017). Iwamori et al. (2023) demonstrate that KCA are indeed useful for discriminating the origin and cycling processes of meteoric water, deep-seated fluid, and their reactions with the host rocks during cycling, based on 590 water samples and 12 major solutes with temperature, pH, O-H stable isotopic ratio, and He isotopic ratio. In addition, the geochemical data of 438 samples from the southern Kyushu, including several time-series data from the same locations, were analyzed to infer the origin and cycling processes of the groundwater and deep-seated fluid. In this study, we aim to integrate these data for a total of 1028 water samples and discuss the origin and cycling beneath the wide area from central to southern Kyushu, based on the appropriate multivariate statistical method. This research was supported by the Nuclear Regulation Authority "Research on Knowledge Development of Giant Eruption Processes, FY 2021-2023"