Open Collections

Abstract

Detrital porphyry indicator mineral (PIM) studies provide an effective means of targeting concealed porphyry Cu-Au systems. However, conventional heavy mineral programs rely on labour-intensive manual optical picking and costly automated scanning electron microscopy (ASEM) approaches, which constrain survey scale and limit their applicability for widespread exploration. This study investigates rapid, cost-effective analytical technologies for quantitative PIM identification and develops an industry-applicable workflow using benchtop micro X-ray fluorescence (µXRF). Heavy mineral concentrates derived from stream sediments proximal to the Lorraine Cu-Au porphyry deposit (British Columbia, Canada) were analyzed by manual visual methods at a commercial laboratory, and by ASEM and µXRF at the University of British Columbia. The effectiveness of each method was evaluated, including the analytical time and cost required to identify a suite of indicator, pathfinder, and gangue minerals. Porphyry indicator minerals were identified and characterized by ASEM, and µXRF-based supervised machine learning classification approaches were trained and evaluated on these datasets. Two µXRF PIM identification methods were developed: (1) visual interpretation of elemental maps to identify compositionally distinct PIMs, and (2) supervised random forest elemental map PIMs classification, trained on coregistered ASEM mineral maps and µXRF elemental counts data. µXRF elemental maps were effective for semi-quantitatively identifying chemically distinct indicator and ore minerals and for capturing fine grains that may be missed during subsequent mineral mapping at coarser resolutions. µXRF-random forest mineral mapping reproduced ASEM mineralogy with an overall mineral classification accuracy of 86% and a macro-F1 score of 0.66, based on 10-fold cross-validation, and identified a broad suite of indicator, pathfinder, and gangue minerals. Compositionally distinct PIMs such as apatite, zircon, titanite, and rutile were classified more reliably than compositionally similar or altered minerals (e.g., andradite versus other garnets; feldspar versus illite-muscovite and epidote). Model performance deteriorated for minority classes. These results demonstrate that benchtop µXRF, combined with data fusion and machine learning, provides a rapid, cost-effective, and scalable approach to quantitative PIM characterization that complements conventional indicator mineral exploration methods.