Inferring relationships between petrophysical data and hyperspectral-derived mineralogy using PLS-based modelling

The integration of borehole-focused datasets, such as hyperspectral drill-core data, with geophysical datasets offers significant potential for enhancing geophysical models. Drill cores from the Stavely region in the Grampians–Stavely Zone of western Victoria have been chosen as a case study to explore the applicability of hyperspectral data, including visible and near infrared (VNIR), short wavelength infrared (SWIR) and thermal infrared (TIR) reflectance spectra for predicting petrophysical properties of rocks using partial least-squares (PLS) regression methods. Hyperspectral data collected by Geoscience Australia from drill cores using a HyLogger-3 at the Geological Survey of South Australia were calibrated against measured density and gamma-ray data collected using a Multi-Sensor Core Logger. PLS modelling was applied to drill-core datasets to evaluate correlations between petrophysical and spectral data. Analysis of drill cores from the Stavely region revealed positive correlations between gamma-ray intensity and white mica abundance, with white mica vibrational modes contributing significantly to PLS-based gamma-ray predictions. Similarly, PLS-based density predictions were influenced by variation in the strength of spectral features related to quartz, carbonate and kaolinite, within the geological formations in the Grampians–Stavely Zone. In volcanic and sedimentary rocks, quartz abundance was linked to low-density values, whereas carbonate-related absorption features indicated higher densities. For sedimentary rocks, kaolinite signatures were associated with low densities, while higher densities correlated with quartz spectral features. These findings highlight the challenges of interpreting geological materials from physical properties owing to overlapping values and complex controls. By linking spectral data with mineralogical and petrophysical parameters, this study highlights the potential for modelling physical rock properties; however, the findings in the Stavely region are not necessarily applicable to other geological terranes owing to variation in lithological properties. Future work incorporating additional factors, such as porosity, grain arrangements and textural characteristics, could further enhance predictive modelling and our understanding of geological processes.