**4. Discussion**

#### *4.1. Consideration for Working with Ultrafine-grained Heavy Minerals*

The size of mineral grains in till reflects the original grain size in bedrock, as well as the effects of glacial erosion, transportation, comminution and deposition on the grains. Dreimanis and Vagners [48] described a bimodal size distribution for specific minerals in till, and the "terminal" grade at which mineral grains become resistant to further comminution. The terminal grade size for most minerals is <250 μm, including garnets and other heavy minerals and, therefore, certain indicator minerals may be enriched in a specific size fraction that is <250 μm. The transport distance and distribution of this fine, terminal grade material in till is not currently known.

Previous work by Pickett [49] analyzing the fine (<63 μm) fraction HMC of till samples identified sample cross-contamination as a significant concern when dealing with fine-grained HMCs. Processing finer-grained sample material increases the difficulty in cleaning sieve surfaces as grains are more easily entrained in void spaces, adhered by electrostatic forces, or lost to aerosolization. Trapped grains may not be readily visible without magnification, making thorough sieve and related sieve equipment cleaning difficult and time consuming. Because low concentrations of indicator minerals (a couple of grains in a 10 kg sample) can constitute a significant 'anomaly', the potential for false-positive results stemming from sample cross-contamination, or false-negative results stemming from lost indicator grains, requires meticulous sample handling and sieving measures that mitigate mineral grain loss or cross contamination [3]. To address this need, this study utilized single-use nylon-screened sieves following the methods outlined in Lougheed et al. [50]. Single-use sieves eliminate the potential for sample cross-contamination and the need for time-consuming sieve cleaning.

#### *4.2. Mineral Liberation Analysis (MLA)*

The option to gather information only for grains above a specific brightness threshold is available in MLA. This option was successfully utilized by Hulkki et al. [47], using the grey level of hematite as a lower limit, to increase the detection of Cu-bearing mineral phases in stream sediment HMC samples. These techniques have the added benefit of decreasing the number of EDS analysis necessary during scanning routines, which decreases the overall analysis time for each sample. However, our study analyzed all grains on each polished grain mount.

The parameters that can be generated and queried by MLA (modal mineralogy, mineral associations, grain size, and grain angularity) would be impractical to measure manually for every grain in a sample, and thus automated SEM-based techniques offer the ability to capture both data that are indiscernible to the eye (spectral EDS imagery) and morphological data that would be impossible to measure for every grain with any consistency. These data are collected simultaneously during a single scan, and once collected can be manipulated by a user to examine individual parameters (e.g., prevalence of a single mineral phase) or to examine the relationships between them (e.g., grain size distributions for individual minerals, mineral association data). Because the number of grains available for analysis increases with decreasing size of the mineral grains being examined, the use of automated mineralogy on <250 μm HMC allows for rapid, efficient and cost-effective collection of large amounts of accurate, relatable quantitative data for thousands of grains.

The <250 μm fractions of processed HMC produced in mineral exploration and governmen<sup>t</sup> surveys are commonly archived immediately after sample processing that recovers the coarse (>250 μm) HMC fraction. The <250 μm of HMC may sometimes be pulverized and analyzed geochemically [51,52] but most commonly it is set aside and never examined. Utilizing this previously unused finer size fraction presents an opportunity to gain new insights without the added cost of revisiting the sampling areas or processing additional bulk samples. The false-color mineralogy maps generated by MLA can be used to identify rapidly and accurately indicator minerals that are already known to be in the samples, identify additional indicator minerals that were not previously reported, and identify new indicator minerals that could assist in characterizing the local bedrock geology. These minerals

can then be examined further using more precise elemental evaluation techniques like laser ablation inductively-coupled plasma mass spectrometry (LA-ICP-MS) or EMPA.

#### *4.3. Density Gradient and Grain Mounting*

The vertical cross-section slabs 1 and 3 of the primary mounts were prepared in order to evaluate the methods described by Blaskovich [41]. The two vertical slabs for each sub-sample displays a visible trend towards higher particle density (grains per unit area) at the bottom (basal surface) of each mount (Figure 4). They also display a greater proportion of the heaviest (red) particles towards the basal surface of the mount. These slabs demonstrate that a vertical density settling gradient exists through the vertical extent of cured epoxy grain mounts and that the number of grains per unit area is greatest at the base of the vertical slabs. The basal surface of a primary grain mount (Slab 2) consistently contains more heavy mineral grains per unit surface area than throughout the vertical slabs.

The basal surface area of a grain mount that is available for grains to be exposed and imaged is constant (5.07 × 10<sup>8</sup> μm2). However, if variable masses of minerals (e.g., 0.2 vs. 0.3 vs. 0.5 g) are mounted, the preferential settling of the heaviest minerals in the aliquot to be mounted will lead to over representation of the heaviest minerals on the basal surface. This concept is visualized in Figure 5. Therefore, it is important that grain mounts within an individual sampling program be prepared from a consistent mass of sample to ensure accurate comparison between normalized indicator abundances.

Lastra and Petruk [53] used 0.4 g of material to prepare epoxy grain mounts for automated mineralogy. In our study, 0.3 g was determined to be a su fficient mass to mount that allowed the most grains to settle on the basal surface of a circular 25 mm diameter epoxy mount for polishing. This determination was made by experimenting with Almonte till blank heavy mineral concentrate. Grains were added to a ring mount until the base was thinly covered and the mass was recorded. The mass of sample necessary to cover the base of a mount varied depending on the density of the medium being sampled as well as the size fraction being mounted. It is important for each sampling program to determine the appropriate mass of sample to prepare before mounting, and to ensure that this mass is used for all samples.

Automated mineralogical analysis of fine (<250 μm) HMC permits the counting of very small grains quickly and accurately. The data produced can be used for robust statistical analysis due to the large number of grains sampled and the consistency in the automated EDS collection method. Formerly qualitative assessments of grain morphology or mineral association can now be examined quantitatively using the grain association and angularity statistics calculated within the software. Reducing grain size and subsequently increasing the grains on a mount surface could result in a reduction in bulk sample size if future work finds that indicator mineral concentrations remain representative in a lower bulk sample volume.

It should be noted that the primary mount slabs examined in this study (Slab 2) were all prepared by quartering so there is the potential for biased representation of heavier minerals on the basal surface of the primary mount. The slab 2 surface analyzed only represents 1 4 of the total basal surface that would routinely be analyzed, thus the total number of grains counted and analyzed are 1 4 of what would be expected for analysis of the entire basal surface of a 25 mm epoxy grain mount using the recommended 0.3 g near-monolayer method described above.

#### *4.4. MLA Error Estimation*

The error in indicator mineral abundance (area%) between MLA runs was approximated using the di fference between the two values collected from scans of the same section of basal slab 2 on an epoxy grain mount. This is modeled after the work of Voordouw et al. [54], who performed similar repeat scans on thin sections of platinum group mineral (PGM)-bearing mineralized bedrock. Their findings indicate variation of ±7 wt. % for individual base metal sulfide minerals, and ±5 wt. % for individual silicate minerals. When using the MLA software, a section to be scanned in a sample run can be copied and re-applied to the exact same section, ensuring that all frames are analyzed

in the same location, with the same operating parameters. This overcomes the potential for nugge<sup>t</sup> effects skewing results as described in Hulkki et al. [47]. Their estimation of error had the caveat that their repeat scan could not be performed on precisely the same area, leaving the possibility that grains present along the edges of the scanned area could be missed between consecutive scans. This resulted in di ffering amounts of Cu minerals detected between their two runs. Their two runs were performed with di ffering operating parameters (BSE standards) and, therefore, are not directly comparable to the results of this study.

Error evaluation indicates that MLA reliably identifies and counts most indicator minerals, with negligible variation observed in consecutive runs within and outside of the same scanning routine (Table 2). Axinite-(Mn) (*Ca2Mn<sup>2</sup>*+*Al2(BO3)Si4O12(OH))*), an alteration mineral previously identified by Hicken [17] in the coarse (>250 μm) fraction of disaggregated mineralized bedrock from Izok Lake, was identified by MLA in the fine (<250 μm) fraction of till HMC by this study using MLA. However, the evaluation of error determined that the variability in axinite abundance calculated between runs was consistently higher than that of the other minerals examined (0.3–0.5%). The accelerating voltages used by this study are much higher than the optimal overvoltage for light elements like B (~5 keV) and, therefore, the peak to background ratio will be too low for e ffective detection of B [55]. The other elements comprising axinite are relatively common, and the high error evaluation by this study suggests that the MLA determination of axinite is being confused with other Ca-, Fe-, or Mn-containing aluminosilicates. The previous visual identification and subsequent EMPA confirmation of axinite in the coarse fraction of till from Izok lake indicates that at least some of the axinite identified by MLA is likely correctly attributed, but the variation in abundance between subsequent scans of the same surface is too high to use as an indicator mineral with this method. The di fficulty in consistently and accurately distinguishing axinite from other minerals with similar stoichiometry indicates that axinite evaluation with MLA is not an e ffective tool for exploration.

The error, as defined by this study, is greater in magnitude when analyzing the coarser (185–250 μm) fraction. The di fference in errors between the coarse (larger error) and fine size fractions (smaller error) is likely due to the greater number of grains/per unit area analyzed (8000 grains in the <64 μm fraction, compared to 4000 grains in the 185–250 μm fraction) and, therefore, the di fference in statistical sample size. The error is smaller for both size fractions studied after the machine underwent routine gun alignment calibration. These values are acceptable for this study, for two reasons: (1) no minerals of interest were identified in one scan but not in another, and (2) the minerals of interest are present in large enough amounts that these minor fluctuations would not impact the identification of an anomalous abundance. Further, this error evaluation emphasizes the need for regular maintenance on FEG-SEM systems, frequent in-house machine calibration and repeated evaluation of error to ensure consistent results.
