Rapid analysis of drill core data for detection of geological boundaries

Regolith geochemical and geophysical patterns related to underlying mineralisation may be difficult to detect in regions where the regolith is thick, transported or has developed complex layering during its mineralogical and geochemical evolution. Detection and accurate logging of geochemical and mineralogical interfaces and horizons within drill samples obtained from such regolith profiles, including the boundary between regolith and underlying bedrock can be critical for detection and interpretation of geochemical patterns. In turn, this assists in selecting zones for systematic sampling and the optimum combination of geochemical, mineralogical and geophysical analysis and data processing to enhance signals or patterns associated with mineralisation. This study presents a multivariate data-driven approach to detecting boundaries within regolith and other profiles potentially incorporating near real-time, in-situ geochemical, mineralogical, and petrophysical data acquisition methods to aid decision-making during stratigraphic/exploration drilling campaigns. The approach is demonstrated using geochemical and mineralogical data from a drillhole at the McKinnons Au deposit in New South Wales (NSW), Australia. Embedding feature (variable) selection techniques to support vector machines and random forest approaches, followed by application of a wavelet tessellation technique encoded in Data Mosaic™ to the selected variables, delivered more detailed and refined identification of zonation and boundaries within the regolith compared with approaches using only visual core logging and geochemical assays. The method was subsequently applied to geochemical, petrophysical and spectral data acquired from two drillholes in the Delamerian Orogen of western NSW. Several subtle lithological boundaries were detected within the regolith and the interface between weathered profiles and basement rocks at different spatial scales. This included some zones displaying elevated Pb and Zn within the saprock part of the regolith profile. A critical zone above the saprock, highlighted by a high variance interval was also detected using the multivariate wavelet tessellation. This indicated shorter core sampling intervals may be needed to improve the likelihood of detecting mineralisation in this region. The methodology improves the identification of mineralised zones by enabling dynamic drilling adjustments through near real-time knowledge feedback, which also reduces costs and enhances field efficiency. • Multivariate and near real-time approach for detection of geological boundaries • Geochemical, spectral and petrophysical data used for regolith characterisation • Feature-selected Machine Learning techniques used for optimisation • The workflow facilitates decision-making during exploration drilling campaigns