1Stromberg, Jessica, 1Schlegel, Tobias, 1Pejcic, Bobby, 1Birchall, Renee, 1Shelton, Tina
1CSIRO Mineral Resources, Kensington, Australia
Identifying alteration mineral zonation around hydrothermal ore systems is critical to the mineral exploration process. Hyperspectral methods are commonly used to map alteration because they are fast, inexpensive, and require little to no sample preparation compared to other mineralogical techniques such as scanning electron microscope (SEM) based mineral mapping, for example. The visible-near and shortwave infrared (VN-SWIR, 350-2500 nm) spectral regions are most used as they are sensitive to hydrated mineral phases including chlorite and white micas. However, in mineral systems which are iron oxide-rich or where key alteration assemblages include abundant anhydrous phases, such as in iron oxide-alkali-calcic alteration systems, using this spectral range can be problematic. In this case, the thermal-infrared (TIR) spectral range (6000-14500 nm) may be more appropriate as it is sensitive to anhydrous silicates such as quartz and feldspars. However, there are inherent challenges in unmixing hyperspectral data for deriving quantitative mineral abundances. In particular, when the quantification of minor or spectrally similar phases are key to the alteration assemblages, such as in the quantification of different feldspars. This can be overcome with the use of a calibration dataset such as quantitative X-ray diffraction or SEM-based quantitative mineralogy in combination with Partial-Least Square (PLS) regression methods. In developing such models, scale is of critical importance. Variability in the sampling area and volume between datasets is one of the greatest challenges in validating hyperspectral data with quantitative mineralogy, and in integrating any geoscience datasets. In this work, we used several hyperspectral and spectroscopic instruments (HyLogger, Agilent 4300 FTIR, Bruker Vertex FTIR, ASD Fieldspec Pro) to evaluate the impact of scale on hyperspectral data validation in IOCG systems. In this process we developed a methodology for creating the first scale-consistent dataset of VNIR-SWIR, TIR, and SEM-based quantitative mineralogy data on drill core samples. This dataset comprises 250 samples from a world-class IOCG deposit in which the key mineral phases, assemblages, and alteration patterns were identified using the SEM-based quantitative mineralogy. Hyperspectral data was processed using The Spectral Geologist Software (TSGTM) software and even with a scale consistent dataset, and a constrained mineral library based on the SEM-based mineralogy, conventional unmixing methods such as the The Spectral Assistant (TSATM) were unable to reproduce key alteration patterns for vectoring towards ore. Using the SEM-based mineralogy data, PLS modelling was applied to derive predictive models for key mineral phases from the TIR hyperspectral data. The resulting models produced quantitative mineralogy with r2 > 0.94 for key phases including quartz, K-feldspar, albite, calcite, and >0.80 for magnetite, biotite, and plagioclase. More importantly, the key mineral assemblages change with distance to ore and the relative abundance feldspar species (albite, K-feldspar, plagioclase) identified by the SEM-based mineral mapping were reproduced using PLS-derived mineral abundances in a validation drill core. This method and the models generated will provide a framework for improving the application of hyperspectral data for mapping alteration in IOGC systems.
Jessica is a Research Scientist in the Mineral Footprints team at CSIRO Mineral Resources where she leads Activity 5 of the NVCL and works on projects across multiple commodities applying the combined use of lab and field-based spectroscopic and geochemical techniques to address industry-specific challenges.