Magnetite is an important ore and gangue mineral in many economic deposits, and the ability to model the modal magnetite abundance from cost-effective data, such as hyperspectral reflectance spectra, has widespread applications. However, magnetite reflectance spectra collected with field and drill core sensors, which are increasingly used by the mineral resources industry, are characterised by very broad and poorly defined diagnostic features. Magnetite content could be modelled indirectly from mixed mineral spectra, but these efforts have been limited by the lack of quantified training datasets of mixed mineral assemblages. We created two hyperspectral libraries (n=104) of magnetite mixed with quartz, chlorite, and siderite collected from two different wavelength ranges to address this knowledge gap. Mineral ratios and particle size variation, which are important when modelling hyperspectral data, were determined using quantitative X-ray diffraction (QXRD) and scanning electron microscopy with energy-dispersive X-ray spectroscopy (SEM-EDS) analysis. Hyperspectral data were acquired across the visible-near, shortwave, mid to thermal infrared (VNIR-SWIR-MIR-TIR, 380-16669 nm) wavelength range, using HyLogger-3 and Bruker Vertex 80v Fourier transform infrared (FTIR) instruments. Predictive modelling was carried out using CSIRO's The Spectral Geologist (TSG) software partial least squares (PLS) modelling tool, which allows for modelling of one variable (e.g., magnetite wt%) from another (e.g., reflectance spectra) via calibration using a training dataset (magnetite mixture spectral library), and subsequent model validation using other hyperspectral data (e.g., from drillhole) of a similar wavelength range. Use of magnetite mixture PLS calibration enabled prediction of magnetite wt% from drill core VNIR-SWIR data that that matches the Fe2O3 assay and magnetic susceptibility detected from the core. We also noted several mineral diagnostic features in the MIR wavelength region, which can provide lower detection limits and an improvement in the accuracy of predictive modelling for magnetite and gangue minerals in the future.

Open-Access Online Publication: May 31, 2023