*2.4. Automated Mineralogy*

MLA is an automated mineralogy software package used in tandem with a field-emission gun scanning electron microscope equipped with electron dispersive spectrometer (EDS) detectors (FEG-SEM-EDS). The software uses high-resolution backscatter emission (BSE) images, image analysis, and elemental chemistry from EDS to create a mosaic image of an epoxy grain mount. The MLA software (version 3.1.4.686, FEI, Hillsboro, OR, USA) delineates grain boundaries based on BSE brightness contrast and subsequently collects a full X-ray spectrum (EDS) at the geometric center of each grain, comparing this to an established EDS mineral library assembled by the user, in order to identify the mineral [43]. The classified grains are subsequently combined into a false-color mineral image, and the program can quickly calculate for each grain, its size, mineral associations (occurrence and interlocking), particle properties (roundness, area, shape), and mineral liberation [44].

Carbon-coated grain mounts were analyzed using XBSE mode on a FEI Quanta 650 FEG SEM (Thermo Fisher Scientific, Waltham, MA, USA) operating under high vacuum at QFIR. The XBSE measurement mode collects a BSE image of each frame and uses variances in grey-level to define phase boundaries on a sample surface. This is followed by collection of a single EDS spot analysis at the calculated centroid of each identified phase. Operating conditions included a beam current of 10 nA, an accelerating voltage of 25 kV, and a spot size of 6. Backscattered electron image brightness and contrast were standardized to a gold (Au) imaging standard. Minimum feature size for image detection and minimum grain size for EDS analysis were both set to 4 μm. Magnification was set to 250× with a resolution of 1000, resulting in a pixel size of 2.14 μm and a horizontal field width of 2.14 mm. Each basal quarter contained between 75 and 210 frames for analysis, whereas each vertical section contained between 100 and 200 frames. Analysis of a single grain mount took between 35 min and 2.5 h, depending on the dimensions of the vertical slabs and the number of particles to analyze (Figure 3, slabs 1 and 3). Basal sections (Figure 3, slab 2) contain between 5000 and 50,000 particles.

Post-processing was performed using MLA Image Processing and Dataview, product version 3.1.4.686. Areas a ffected by charging e ffects were removed from the false-color map of each grain mount and each grain was classified using a mineral reference library constructed from spectra collected from the FEI reference library, augmented by spectra gathered manually over several years at QFIR.

Quality control evaluation of mineral identification, grain abundance analysis, and MLA error was carried out on one secondary mount of the 185–250 μm fraction and one of the <64 μm fraction. Both mounts were evaluated by performing a duplicate scan on a sub-section of each mount. Two scans were performed so that the di fference in generated modal mineralogy data could be evaluated between scans of the same area under identical operating conditions, within the same scanning routine. After this scanning routine was completed, the two mounts were left in the machine and it underwent routine in-house calibration. Keeping the mounts inside the machine ensured that their orientation would not change. Following calibration, a second scanning routine was performed using the same settings as

the first routine. This second scanning routine allowed for the evaluation of differences in generated modal mineralogy data prior to, and immediately following, routine calibration.

The time needed to complete an MLA scan varied depending on the operating parameters and the number of particles being analyzed. In general, decreasing particle size (and subsequently increasing particle abundance) increased the analysis time. Preparation of the mineral reference library was tailored to the needs of the study and the minerals/elements of interest, whereas the post-processing of data and generation of false-color grain maps and mineral abundance tables is a relatively routine process.
