Multiphysics property prediction from hyperspectral drill core data

Abstract. Hyperspectral data provides rich quantitative information on both the mineralogical and fine-scale textural properties of rocks, which, in turn, largely control their petrophysical characteristics. We therefore developed a deep learning model to predict petrophysical properties directly from hyperspectral drill core data. Our model learns relevant features from high-dimensional hyperspectral data and co-registered sonic, gamma-gamma density and gamma-ray logs to infer slowness, density, and gamma-ray counts. We demonstrated the performance of this approach on data acquired in the Spremberg region of Germany. Our results demonstrate that with meticulous pre-processing steps and thorough data cleaning, one can overcome the difference in capturing resolution and learn the relationship between hyperspectral data and petrophysics. Using a test dataset from a spatially independent borehole, we generate a pixel-resolution (≈ 1 mm²) model of the petrophysical properties and resample it to match the measured logs. This test indicates substantial accuracy, with R² scores and root-mean-squared errors (RMSE) of 0.7 and 16.55 μs.m^-1, 0.86 and 0.06 g.cm^-3 and 0.90 and 15.29 API for the slowness, density and gamma-ray readings respectively. Overall, our findings lay the groundwork for building deep learning models that can learn to predict physical and mechanical rock properties from hyperspectral data. Such models could provide the high-resolution but large-extent data needed to bridge the different scales of mechanical and petrophysical characterisation.