Characterization of hydrothermal alteration along geothermal wells using unsupervised machine-learning analysis of X-ray powder diffraction data

Zonal distribution of hydrothermal alteration in and around geothermal fields is important for understanding the hydrothermal environment. In this study, we assessed the performance of three unsupervised classification algorithms—K-mean clustering, the Gaussian mixture model, and agglomerative clustering—in automated categorization of alteration minerals along wells. As quantitative data for classification, we focused on the quartz indices of alteration minerals obtained from rock cuttings, which were calculated from X-ray powder diffraction measurements. The classification algorithms were first examined by applying synthetic data and then applied to data on rock cuttings obtained from two wells in the Hachimantai geothermal field in Japan. Of the three algorithms, our results showed that the Gaussian mixture model provides classes that are reliable and relatively easy to interpret. Furthermore, an integrated interpretation of different classification results provided more detailed features buried within the quartz indices. Application to the Hachimantai geothermal field data showed that lithological boundaries underpin the data and revealed the lateral connection between wells. The method’s performance is underscored by its ability to interpret multi-component data related to quartz indices.