Content uploaded by Rosie Blannin
Author content
All content in this area was uploaded by Rosie Blannin on Apr 21, 2021
Content may be subject to copyright.
Minerals Engineering 167 (2021) 106836
0892-6875/© 2021 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Uncertainties in quantitative mineralogical studies using scanning electron
microscope-based image analysis
Rosie Blannin
a
,
*
, Max Frenzel
a
, Laura Tus
¸a
a
, Sandra Birtel
a
, Paul Iv˘
as
¸canu
b
,
1
, Tim Baker
b
,
Jens Gutzmer
a
a
Helmholtz-Zentrum Dresden-Rossendorf, Helmholtz Institute Freiberg for Resource Technology, Freiberg, Germany
b
Eldorado Gold Corporation, Vancouver, Canada
ARTICLE INFO
Keywords:
Geometallurgy
Automated mineralogy
Nugget effect
Uncertainty estimation
Bootstrap resampling
ABSTRACT
Scanning electron microscope-based automated mineralogy studies are readily associated with quantitative re-
sults, providing one of the foundations of geometallurgical studies. Despite the importance of quantitative data
for such studies, and efforts to reduce statistical errors, the reporting of uncertainties is rare. This contribution
illustrates how bootstrap resampling can be used to provide robust estimates of statistical uncertainties for the
modal mineralogy, metal deportment and all relevant textural attributes of a sample or a series of samples. Based
on a case study of the Bolcana Au-Cu porphyry deposit in the South Apuseni Mountains, Romania, the impact of
insufcient sampling statistics on quantitative mineralogical studies is illustrated. Quantitative analyses of the
mineralogy and microfabric of milled ore samples from seven 40 m drill core intervals from the Bolcana Prospect
were conducted using a Mineral Liberation Analyser (MLA), complemented by electron probe micro-analysis.
Bootstrap resampling was then applied to assess how many grain mount surfaces should be analysed to ach-
ieve statistically robust results for both Cu and Au mineralogy, deportment and textural attributes. Despite
variable mineralogy, grades and mineralisation styles, estimated statistical uncertainties on Cu deportment are
consistently low. In contrast, uncertainties for Au deportment are so high that most reported values for important
characteristics are statistically meaningless. This is mainly attributed to the pronounced nugget effect for Au
mineralisation, exacerbated by the small sample size analysed by MLA. An unfeasible number of measurements
would be necessary to provide robust gures for the deportment of minor/trace elements and minerals, along
with other tangible mineralogical properties, such as mineral associations. The results of this case study
demonstrate that statistical uncertainties need to be carefully incorporated when considering the results of
automated mineralogical studies and their impact on geometallurgical models. This is particularly relevant for
studies of precious metal ores.
1. Introduction
Tangible ore characteristics, such as mineralogy, metal deportment
and microfabric constrain the efciency of mineral processing opera-
tions and, therefore, are important factors to consider during the
exploration and evaluation stages of a mining project (Core et al., 2006;
Cropp et al., 2013; Gregory et al., 2013; Kesler et al., 2002). Modern
analytical techniques, such as scanning electron microscope-based
automated mineralogy systems (e.g. Mineral Liberation Analyser,
QEMSCAN, Mineralogic, TIMA-X), allow quantitative data on these
characteristics to be acquired both quickly and relatively cheaply. Due
to the wealth of information these techniques provide, they are
commonly applied in geometallurgical studies (Bachmann et al., 2018;
Frenzel et al., 2018; Kern et al., 2018; Leichliter, 2013). They are also
frequently used to numerically constrain the deportment of rare and
precious metals (e.g. Goodall, 2008; Goodall and Butcher, 2012; Greg-
ory et al., 2013; Warlo et al., 2019; Zhou et al., 2009). However, such
studies do not typically report any statistical errors and thus, the
robustness of the results remains uncertain, as well as any technical
decisions taken based on these results.
Several methods have been employed to estimate uncertainties on
automated mineralogy-derived measurements. For example, Lastra and
* Corresponding author.
E-mail address: r.blannin@hzdr.de (R. Blannin).
1
Present address Dundee Precious Metals, Soa, Bulgaria
Contents lists available at ScienceDirect
Minerals Engineering
journal homepage: www.elsevier.com/locate/mineng
https://doi.org/10.1016/j.mineng.2021.106836
Received 11 September 2020; Received in revised form 9 February 2021; Accepted 11 February 2021
Minerals Engineering 167 (2021) 106836
2
Paktunc (2016) investigated variability in mineral quantity and mineral
liberation analyses through inter-laboratory testing of the +208 to -509
µm fraction of a sulphide otation rougher concentrate. They found a good
agreement in the mineral quantities, but less agreement in the liberation
and mineral association analyses. Therefore, correct mineral quantities do
not necessarily imply correct mineral liberation and association. In
another approach, Benvie et al. (2013) developed a statistical approach for
using automated mineralogy analyses in conjunction with diagnostic
leaching tests. It was determined that measurements of at least two grain
mounts were required for each head and leach residue sample, to derive
the background variance and standard deviation. Butcher et al. (2000)
analysed a sulphide otation concentrate from KCGM’s Super Pit opera-
tions in Kalgoorlie, Western Australia, using automated mineralogy to
quantify gold deportment. Twenty polished blocks of the same sample
were analysed, with an average of three gold grains identied per sample
replicate. Analyses of 13 of these blocks were required for 90 % condence
of capturing the mean number of grains i.e. to produce statistically reliable
results. More recently, Guseva et al. (2021) applied the point counting
method with the binomial distribution approximation (Chayes, 1945;
1944; Guseva et al., 2021; Van der Plas and Tobi, 1965) to evaluate
analytical errors on mineralogical measurements. The study suggested
that the binomial distribution approximation may not be adequate in all
cases, particularly for coarse materials, and that other methods should be
applied in such cases, e.g. bootstrap resampling or the estimation of con-
dence intervals based on the method developed by Leigh et al. (1993).
Bootstrap resampling has been identied as an effective tool for the
estimation of errors on textural characteristics measured by automated
mineralogy (Evans and Napier-Munn, 2013; Mariano and Evans, 2015).
For example, the bootstrap approach has been used to provide un-
certainties on particle properties measured by automated mineralogy for
the statistical modelling and simulation of mechanical separation pro-
cesses (Hannula et al., 2018), the evaluation of magnetic separation
efciency (Buchmann et al., 2018; Leißner et al., 2016) and density
separation processes (Schach et al., 2019).
Bootstrap resampling involves taking M random subsets of N samples
from a population, and replacing the randomly selected samples to
ensure that the entire population is available to be selected again
(Chernick, 1999; Efron, 1979). This method has been found to agree
with accepted statistical methods for assessing error in mineral grades
using point counting on polished sections (Chayes, 1945; 1944; Guseva
et al., 2021; Van der Plas and Tobi, 1965), whilst having the advantages
of being assumption-free, rather than assuming a binomial distribution,
and being applicable to a wide range of particle characteristics (Evans
and Napier-Munn, 2013). Based on these methods, the standard devia-
tion of mineral grades is proportional to the square root of the total area
of the particles measured, or the number of particles measured.
Following on from this, when comparing results from different methods
(X-ray powder diffraction, automated mineralogy, element to mineral
conversion), Parian et al. (2015) estimated the relative standard devi-
ation of measurements for any mineral grade using:
RSD =ax−0.5(1)
where x is the mineral grade, RSD is the relative standard deviation
and a is a coefcient, which implies that relative standard deviation is
proportional to the square root of the mineral grade.
The bootstrap method can also be used to determine how many
grains (i.e. total area) need to be measured in order to reach a given
uncertainty (Evans and Napier-Munn, 2013; Mariano and Evans, 2015).
Evans and Napier-Munn (2013) determined that measurements of only
two polished block surfaces are required to achieve a Relative Standard
Deviation (RSD) of 10 % for chalcopyrite grade in samples from a por-
phyry Cu deposit. In contrast, around 16 blocks would need to be
measured to achieve the same RSD for the chalcopyrite grain size dis-
tribution, similar to the nding of Lastra and Paktunc (2016), due to the
high uncertainties associated with the coarse size fractions.
Despite bootstrap resampling being a well-documented and proven
approach for the estimation of uncertainties for automated mineralogy
data, statistical uncertainties are rarely reported. Such uncertainties are
particularly important for process mineralogical and geometallurgical
studies of precious metal ores, as these are typically marked by
extremely low (mineral) grades. Without the reporting of uncertainties,
it cannot be assumed that enough precious mineral grains have been
analysed to ensure reliable data and robust statistics. Therefore, the
quantitative nature of such studies must be questioned.
This study demonstrates how the errors associated with automated
mineralogical studies can be estimated and reported, using the bootstrap
technique, to ensure that technical decisions take uncertainty of the
results into account. The Bolcana porphyry Au-Cu system in Romania
was used as a case study. As an early stage exploration project, Bolcana
provides an ideal environment for a geometallurgical assessment.
Automated mineralogy was applied, in combination with complemen-
tary analytical techniques, to characterise samples from a series of
metallurgical testing intervals, focused mainly on copper and gold
deportment, as well as mineral associations. The implications of the
ndings are assessed and recommendations provided for future process
mineralogical studies in terms of sampling strategy, analytical work and
reporting of statistical uncertainties.
2. Case study
The Bolcana porphyry gold-copper system is located in the Western
Tethyan magmatic belt, where the tectonic and geodynamic setting
evolved from Cretaceous subduction-related arc magmatism,
Fig. 1. Simplied geological map of the “Gold Quadrilateral”, South Apuseni
Mountains. Epithermal Au-Ag and porphyry Cu-Au systems are highlighted.
Modied after Ivǎs
¸canu et al. (2018).
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
3
transitioning from convergence to post-orogenic extension in the late
Eocene to early Oligocene, followed by widespread post-collisional
extension-related magmatism in the Miocene (Baker, 2019). The Bol-
cana deposit can be further localised to the Brad-Sacaramb metallogenic
district within the “Gold Quadrilateral” of the South Apuseni Mountains,
Romania, where the metallogenic endowment is related to Miocene
magmatism (Fig. 1). In comparison to the rest of the West Tethyan belt,
this region hosts signicant Au resources, forming Europe’s largest
epithermal Au-Ag-Te province. With twenty known porphyry copper
deposits, it is also one of Europe’s most important porphyry Cu-Au
provinces (e.g. Baker, 2019; Berbeleac et al., 2014; Cioacǎ, 2013; Milu
et al., 2003).
The porphyry mineralisation of the Bolcana system is being explored
by Eldorado Gold Corporation through its Romanian subsidiary, Deva
Gold S.A., who recently published a maiden resource estimate of 381 Mt
at 0.53 g/t gold and 0.18 % copper (Ivǎs¸ canu et al., 2019, 2018). The
Bolcana system comprises a sequence of complex cross-cutting porphyry
phases, breccias, alteration and veining. Early-stage crowded diorite
porphyries host copper–gold mineralisation and exhibit a classical
evolution of vein density and mineralisation decreasing from early to
intra to late mineral phases (Sillitoe, 2010), with potassic (biotite-feld-
spar-magnetite) and sodic (albite-actinolite-epidote-chlorite-magnetite)
alteration. Extensive phreatomagmatic brecciation developed in the roof
of the crowded porphyry system, which were subsequently intruded by
ne, uncrowded andesitic to microdioritic intermineral porphyry dykes
and breccias. Alteration of these ne intermineral porphyries is domi-
nantly potassic with a strong and pervasive magnetite-albite-chlorite-
epidote (MACE) overprint. The ne intermineral porphyries,
magmatic breccias and magmatic-hydrothermal breccias have higher
gold grades, but typically lower copper grades, than the earlier crowded
porphyries (Ivǎs¸canu et al., 2019, 2018). In the periphery of the deposit,
porphyry mineralisation is accompanied by Pb-ZnCuAuAg intermediate-
sulphidation epithermal veins (Cardon et al., 2008; Ivǎs¸canu et al.,
2018).
3. Materials and methods
3.1. Sampling and sample preparation
Seven drill core intervals, each 40 m in length, from two drill holes
were selected for this study: TRSD017 240–280 m, 346–386 m and
386–26 m; TRSD013 64–104 m, 152–192 m, 356–396 m and 758–798
m; as shown in the simplied cross section in Fig. 2. Details on the li-
thology, alteration, mineralisation, and location of the drill core in-
tervals within the deposit are provided in Table 1.
Representative splits, around 100 g, of crushed (~ 2 mm) drill core
material were provided by Deva Gold S.A for every 2 m section of the
studied testing intervals. Representative samples of the 40 m intervals
were constituted by mixing representative splits (25 g) of the respective
sub-samples for each interval, homogenising and milling to a 100 % pass
size of 600
μ
m in a Laarmann LMBM 2000 planetary mill at 200 revo-
lutions per minute (rpm), and with steel balls of 20 mm diameter. For
interval TRSD013 152–192 m, additional 4 m interval samples were
prepared to enable a more detailed assessment of the variable and
relatively high gold content detected in this interval. The samples were
prepared in the same way, by splitting and homogenisation, and then
milled in a Retsch RS 200 disc mill at 700 rpm. All 17 samples were then
split again to around 4 g (7 samples from 40 m intervals, 10 samples
from 4 m intervals of TRSD013 152–192 m). Sample preparation was
performed at the Helmholtz Institute Freiberg for Resource Technology
(HIF).
Fig. 2. Schematic cross section, derived from a 3D model, through the Bolcana
porphyry system, showing principal alteration domains gold. Gold-equivalent
grade shells of >0.4 g/t (red) and >1.0 g/t (purple) were calculated using
prices of 1,250 USD/oz Au and 3 USD/lb Cu. The studied drill holes, TRDS017
and TRSD013, are shown, with the studied intervals also highlighted
for TRSD013.
Table 1
Information on selected drill core intervals. Alteration types: Po =Potassic, So =Sodic, MACE =Magnetite-albite-chlorite-epidote.
Drill hole ID From-To
(m)
Zone Lithology Alteration Mineralisation
TRSD013
64–104 Shallow, central Intermineral porphyry ±
breccia
Moderate Po/So/MACE, moderate/strong clay
overprint
Ccp-Py, Cv-Cct replacement
152–192 Shallow, high grade,
central
Fine grained porphyry Strong Po/So, moderate clay overprint Ccp, Py, Bn, Cv-Cct
replacement
356–396 Intermediate, central Breccia, porphyry dykes/
veins
Strong Po/So, moderate clay alteration on veins Ccp and Py
758–798 Deep, high grade, West Fine grained porphyry Strong Po/So Ccp and Bn, low py
TRSD017
240–280 Shallow, low grade, South Breccia, porphyry dykes/
veins
Po/So/MACE, variable clay overprint Ccp, Py, local base metals
346–386 Moderate grade, South Intermineral porphyry ±
breccia
Po/So/MACE, weak clay overprint Ccp, Py
386–426 High grade, South Fine grained porphyry Po/So/MACE, weak clay overprint Ccp, Py, Bn
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
4
3.2. Chemical assays
Chemical assays were carried out by ALS Romania SRL for a total of
49 elements on every 2 m interval of drill core (140 samples). The
assaying procedure involved four acid digestion followed by inductively
coupled plasma mass spectrometry (ICP-MS) for all elements, with the
exception of Au, which was analysed by re assay with an atomic ab-
sorption spectroscopy nish. The results were provided by Deva Gold S.
A. and were used to check the validity of the mineralogical studies.
3.3. Mineral liberation Analyser
An MLA instrument was used to analyse the samples and provide
quantitative mineralogical and microstructural data (Fandrich et al.,
2007; Gu, 2003; Schulz et al., 2020). Polished grain mounts (30 mm) of
the milled samples (pass size of 600
μ
m) were prepared by mixing the
sample with pure graphite powder and embedding the mixture in epoxy
resin. The initial sample blocks were sliced vertically, rotated by 90◦and
remounted in epoxy resin before polishing, to reduce the effects of
gravity settling (Heinig et al., 2015). The polished grain mounts were
carbon coated in preparation for MLA. To improve measurement sta-
tistics, the grain mounts of the 40 m blend samples were re-ground and
polished, and the MLA analyses repeated, with a total of six surfaces
measured for each 40 m sample. Therefore a total of 52 analyses were
performed: 6 analyses for each 40 m samples (42) and one analysis for
each 4 m sample from interval TRSD013 152–192 m (10).
Analyses were performed at HIF on an FEI Quanta 650F eld emis-
sion SEM (FE-SEM) equipped with two Bruker Quantax X-Flash 5030
energy-dispersive X-ray (EDX) detectors and the MLA software suite
version 3.0. The GXMAP and SPL-DZ measurement modes were used,
with details of the modes found in Fandrich et al. (2007). The operating
conditions used for SEM and MLA are listed in Table 2.
The complex and ne-grained nature of the samples led to the gen-
eration of mixed spectra during MLA measurements, and the misclassi-
cation of ne gold grains. To remedy this, manually created mixed
spectra (cf. Bachmann et al. (2017); Kern et al. (2018) for this approach).
between gold and common host minerals (chalcopyrite, pyrite and
quartz) were added to the mineral reference list at ratios of 50:50, 60:40,
70:30, 80:20 and 90:100 (Au:Mineral). To prevent over-estimation of
copper and gold grades using MLA, Back Scatter Electron (BSE) Overlay
scripts were applied to remove pixels with BSE levels lower than 190 for
gold, 170 for the gold mixed spectra and 92 for chalcopyrite. Spectra
matching was set at 90 % for the GXMAP measurements, and 80 % for
the SPL measurements.
3.4. Electron probe microanalysis
Four crushed material samples were selected for electron probe
microanalysis (EPMA) to identify any sulphides that carry signicant
gold content. The TRSD013 64–104 m, TRSD017 386–426 m and
TRSD013 758–798 m samples were chosen to represent shallow, mod-
erate and high depths of the deposit. TRSD013 184–188 m was selected
based on the high Au content as reported by the chemical assay. See
Appendix A.1 for more information on the EPMA methodology, with the
detailed measurement settings in Table A1.
3.5. Estimation of analytical uncertainties
Uncertainties of MLA-derived data were estimated using the boot-
strap resampling method (Chernick, 1999; Efron, 1979), including
calculated assays of copper and gold, modal mineralogy, copper
deportment, grain size and mineral associations of relevant ore minerals
of copper and gold. Bootstrap resampling is an effective technique for
the estimation of errors associated with quantitative automated
mineralogy-derived measurements of the textural characteristics of
particulate materials (Evans and Napier-Munn, 2013; Mariano and
Fig. 3. Schematic diagram to illustrate: (A) an MLA false colour image of a whole grain mount surface; (B) how each grain mount surface was split into 5 strips; (C)
the process of bootstrap resampling to sample M sets of N sub-samples from an original population of n samples.
Table 2
SEM and MLA operating conditions (Acc. volt. =acceleration voltage; Probe cur.
=probe current; BSE cal. =Back Scatter Electron calibration; Res. =resolution;
pix. =pixels; Acq. time =acquisition time; Min. part. size =minimum particle
size).
SEM settings GXMAP SPL-
DZ
MLA settings GXMAP SPL-DZ
Acc. volt. (kV) 25 Pixel size (µm) 1.5 0.625
Probe cur.
(nA)
10 Res. (pix.) 1000 ×1000
Spot size 5.6 Step size (pix.) 6 ×6
HFW 1500 1000 Acq. time (ms) 55
Brightness 96.2 BSE trigger 26–255 225–255
Contrast 18.5 Min. part.. size
(pix.)
5 1
BSE cal. (Au) 254 Min. grain size
(pix.)
3 1
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
5
Evans, 2015). The bootstrap resampling method involves creating M
subsets of N samples, by randomly sampling the original sample set.
Using replacement, every sample from the original population is avail-
able for selection each time. A schematic diagram is provided in Fig. 3.C,
where n strips are available in the original population, and M samples
are constituted by taking N random samples.
In this study, composite ground material samples from seven 40 m
intervals were analysed with MLA. After repeated re-grinding and re-
polishing, six surfaces were measured for every grain mount, each
comprising 5 strips, resulting from the rotation of the original grain
mounts (see details in Section 3.4; Fig. 3.A-B). The bootstrap resampling
was performed by sampling and replacing from the total of 30 strips (5
strips in each of the 6 measured surfaces) for which data were obtained.
To estimate the errors on the textural characteristics of the particulate
samples, N was set as the number of strips (i.e. 30) and 1000 subsets of N
strips were randomly selected from the original set of 30 strips (i.e. M =
1000). For the 4 m interval samples from interval TRSD013 152–192 m,
only one grain mount surface was measured and so 5 strips were
available, and therefore N was 5 for these samples.
The area of each mineral group in the resampled strips was summed
and the overall modal mineralogy, copper and gold contents, and copper
deportment were calculated using the mineral densities and stoichio-
metric mineral compositions. Mineral associations for chalcopyrite were
calculated by dividing the summed contact lengths of chalcopyrite with
each mineral by the total length of the chalcopyrite grain boundaries
(Fandrich et al., 2007; Gu, 2003), according to the formula:
Mineral association =X/Y(2)
where X is the length of grain boundaries between two specic minerals
and Y is the total length of the grain boundaries of one of the minerals.
The MAMA ratio (cf. Kern et al., 2019) was subsequently calculated to
identify preferential associations using the following formula:
MAMA =Mineral Association of target mineral with MinX
Mineral Area (%)of MinX
(3)
where mineral association is calculated as given in equation (2), when
free perimeter is excluded, and mineral area is the area percent of the
associated mineral in the sample.
The mineral associations of gold were also investigated for intervals
with >10 gold grains (TRSD013 64–104 m, 758–798 m and TRSD017
386–426 m), to ensure reliable sample statistics. Due to the low number
of gold grains identied in each strip, bootstrap resampling of the gold
mineral associations was performed on individual grains rather than
using the strips. In this case, N was the number of gold grains present
and M remained 1000. For interval TRSD013 152–192 m, the gold
grains from both the 40 m and 4 m samples were combined and the
associations were bootstrapped for the whole 40 m interval. A ‘com-
posite’ of the gold grains from the 40 m samples was bootstrap resam-
pled to provide an overview of gold mineral associations throughout all
of the intervals. The gold grains from the 4 m samples from TRSD013
152–192 m were excluded to prevent over-representation of this
interval.
Following the estimation of errors on the measurements, bootstrap
resampling was applied to estimate how many grain mount surfaces
would need to be measured for an acceptable relative standard deviation
(see Section 5.1 for specic choices). RSD, also known as the coefcient
of variation, is calculated as follows:
RSD =100 ×
σ
μ
(4)
where
σ
is standard deviation and µ is the mean. N was increased in
increments of 5, to represent a whole grain mount surface, until the
required RSD was reached. The mineral association of gold was boot-
strap resampled for the composite of the gold grains from the 40 m
samples only, using individual grains rather than strips.
4. Results
The following subsections present the analytical results obtained by
MLA, in combination with EPMA. The mineralogical variability, Cu and
Au deportment, and mineral associations of chalcopyrite and gold grains
of the samples are subsequently assessed, including estimations of the
uncertainties related to the MLA measurements.
4.1. Mineralogical variability
A total of 44 minerals were identied in the samples using MLA.
Further, the 44 phases were divided into 19 mineral groups as shown in
Table 3. Note that feldspars and pyroxenes may well be both rock-
forming and alteration minerals, but were grouped as rock-forming
minerals for simplication. The area of each mineral was bootstrapped
for each sample, as detailed in Section 3.5, and the modal mineralogy,
with corresponding uncertainties, was subsequently calculated. The
median simulated modal mineralogies are plotted in Fig. 4 for both the
40 m samples and the 4 m samples.
The contents of both the rock-forming (quartz, feldspars, amphi-
boles) and alteration minerals (white micas, clay minerals, chlorite,
magnetite etc.) follow the well-established patterns of hydrothermal
alteration and lithological transitions with depth in the Bolcana por-
phyry system (D´
enes et al., 2015; Ivǎs¸canu et al., 2018; Ivǎs¸canu et al.,
2019). Quartz content varies from ~ 15–35 wt%, in contrast to feldspars
which increase from 10 wt% in the near-surface to around 50 wt% at the
centre of the system. Amphiboles are typically a minor constituent.
Clay minerals and white micas are dominant in the near-surface, at
Table 3
The list of minerals identied by MLA with mineral groupings and abbreviations
used for simplication purposes. The abbreviations are used in all following
tables and gures.
Group Minerals Abbreviation
4
Rock-forming
minerals
Quartz Quartz Qz
Feldspars K-feldspar, albite, labradorite Fsp
Amphiboles
(+Pyroxene)
Hornblende, actinolite,
wollastonite
Amp
Alteration
minerals
White micas Muscovite, illite Wm
Clay minerals Kaolinite Cly
Biotite Biotite Bt
Chlorite Chamosite, clinochlore,
other intermediate
compositions
Chl
Fe-oxides Magnetite/hematite
1
,
titanomagnetite
Fe-O
Carbonates Calcite, siderite, ankerite Cb
Other alteration
minerals
Epidote, allanite, gypsum/
anhydrite
1
, barite, uorite
OAM
Cu-bearing
minerals
Chalcopyrite Chalcopyrite Ccp
Bornite Bornite Bn
Chalcocite Chalcocite Cct
Covellite Covellite Cv
Sulphosalts Tetrahedrite, freibergite Ss
Sulphides
Pyrite Pyrite Py
Other sulphide
minerals
Sphalerite, galena,
arsensopyrite,
tellurobismuthite-pyrite
2
OSM
Trace/minor
minerals
Other minerals Titanite, rutile, apatite,
monazite, zircon
OM
Gold Gold Gold <20 % Ag, Gold <10 %
Ag, Gold, Gold mixed
spectra
3
Gold
1
Minerals cannot be distinguished by MLA.
2
A mixed-spectra between tellurobismuthite and pyrite.
3
Mixed spectra of gold and quartz, pyrite and chalcopyrite, as explained in
Section 3.4.
4
Mineral abbreviations according to Whitney and Evans (2010).
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
6
around 20 wt% and 38 wt% in TRSD013 64–104 m, respectively, in
relation to phyllic alteration. Biotite content is linked to sodic and
potassic alteration, occurring variably in the deeper intervals, up to
around 7 wt% in TRSD017 346–386 m. Chlorite content is particularly
associated with MACE (magnetite-albite-chlorite-epidote) alteration
(Ivǎs¸canu et al., 2019, 2018), remaining fairly constant at higher depths
(4–9 wt%). Fe-oxide content increases towards the centre of the system
with the transition from sodic and MACE alteration assemblages towards
the potassic core, reaching concentrations up to around 6 wt%. Other
alteration minerals are typically minor in abundance.
Pyrite content is most signicant in the phyllic alteration zone in the
near-surface environment, at ~ 3–5 wt%. Chalcopyrite varies from ~
0.3–1.5 wt%, with no clear trend with depth. Bornite, covellite, chal-
cocite, sulphosalts, sphalerite and galena are typically minor, at <0.1 wt
%, and also show no clear trend with depth. The lack of clear zoning in
the sulphides results from variable lithology and alteration styles,
combined with the complex architecture of cross-cutting porphyry
bodies e.g. ne intermineral porphyries are dominantly associated with
nely disseminated sulphides and/or C veins (chalcopyrite ±bornite ±
pyrite), whereas strong potassic alteration is coincident with B veins
hosting chalcopyrite (Ivǎs¸ canu et al., 2019).
In the individual 4 m samples from the TRSD013 152–192 m inter-
val, feldspar, clay, chlorite and pyrite contents vary considerably, while
chalcopyrite content reaches around 2 wt%. The TRSD013 152–156 m
interval is markedly different due to the high pyrite content (~ 11 wt%)
and absence of chlorite. The pyrite content is similarly high in TRSD013
156–160 m (~ 9 wt%) and then decreases to <2.5 wt% for the rest of the
interval.
To assess the uncertainties of the modal mineralogy measurements,
the median modal abundance values for different minerals were plotted
Fig. 4. Median modal mineralogy of the studied drill core intervals, from MLA GXMAP analyses of the 40 m crushed material samples (left) and the 4 m crushed
material samples from drill hole interval TRSD013 152–192 m (right).
Fig. 6. Distribution of copper between Cu-bearing minerals in the 40 m crushed material samples (left) and the 4 m crushed material samples from drill hole interval
TRSD013 152–192 m (right), calculated based on the median modal mineralogy from bootstrap resampling (Fig. 4).
Fig. 5. Median value of bootstrapped modal mineralogy of sample TRSD013
152–192 m, with error bars plotted at the 2.5th and 97.5th percentiles to
represent 95 % condence intervals.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
7
with error bars representing 95 % condence intervals. A graph for
sample TRSD013 152–192 m is given in Fig. 5, as an example, with
graphs for the remaining 40 m and 4 m intervals provided in Appendix B
(Figs. B1 and B2). Uncertainties for the abundances of the major min-
erals in the 40 m samples tend to be low, typically with an RSD of <10
%. For minerals with a modal content of around 0.1 wt% or less, the RSD
increases above 10 %. Despite its relatively low content, the RSD of
chalcopyrite is below 8 % in all 40 m samples. The highest RSDs are seen
for minor Cu-bearing minerals, such as bornite (5–31 %), covellite (9–25
%), sulphosalts (13–47 %) and particularly chalcocite (19–100 %) and
other sulphide minerals (9–39 %). High RSDs are also recorded for
alteration minerals, which occur variably throughout the intervals,
including clay minerals (4–47 %) and other alteration minerals (2–31
%). The uncertainties for mineral contents in the 4 m samples are rather
high and variable. Only for quartz, feldspars, amphiboles and white
micas are the RSDs below 10 % for each 4 m interval. Chalcopyrite has
an RSD below 11 % in all samples.
4.2. Copper deportment
A combination of MLA and EPMA was used to study copper
deportment in the Bolcana porphyry system. The copper deportment
was calculated from the bootstrap resampling of the grain areas using
the densities and mineral formulae provided by the standard MLA
mineral reference lists. EPMA analyses of Cu sulphide mineral grains
conrmed the general formulae used for the deportment calculations
(Appendix A.2).
The median calculated copper deportments (Fig. 6) indicate that
copper is hosted, rather variably, by chalcopyrite, bornite, chalcocite,
covellite and the sulphosalts tetrahedrite and freibergite. Chalcopyrite is
the most important Cu ore mineral in all samples, with between ~ 60–98
% of the overall Cu grade. The remaining Cu content is mainly hosted in
bornite (~ 2–22 % of Cu grade), and variably contributed by chalcocite,
covellite and sulphosalts. For instance, chalcocite is only signicant in
interval TRSD017 346–386 m (~ 16 %), while covellite contributes up
to ~ 7 %. In the TRSD013 152–192 m interval 4 m samples, chalcopyrite
remains the dominant Cu-bearing mineral, but sulphosalts (TRSD013
152–160 m), as well as bornite and covellite (TRSD013 180–192 m)
contribute signicantly to Cu-grade in some samples. The average
measured areas of each Cu-bearing minerals are reported in Table A3,
clearly showing again that chalcopyrite is the dominant Cu-bearing
mineral in all samples by one or two orders of magnitude.
As could be expected, the uncertainties for Cu deportment are closely
related to the uncertainties for the Cu-bearing mineral contents (Figs. 7,
B.3 and B.4). The RSDs for chalcopyrite (<7 %) and bornite (5–30 %)
are typically low, being the most abundant Cu-bearing minerals.
Conversely, chalcocite, covellite and sulphosalts contents are highly
variable and therefore the uncertainties of their contributions to the Cu
deportment are typically high, with RSDs from 19 to 102 % for chal-
cocite, 9–26 % for covellite and 11–46 % for sulphosalts. In the 4 m
samples from TRSD013 152–192 m, the RSDs of chalcopyrite are lower
than in the 40 m samples (0.1–5 %), while the other Cu-bearing minerals
are similarly variable, or higher in the case of bornite (RSDs of 7–55 %).
4.3. Chalcopyrite mineral associations
The mineral associations of chalcopyrite, as the most important Cu-
bearing mineral, were bootstrap resampled, based on the contact lengths
of chalcopyrite grains with other minerals (Fig. 8). The dominant min-
eral associations reect the modal mineralogy of the samples, with high
association with the rock-forming and alteration minerals. The propor-
tion of chalcopyrite potentially recoverable by froth otation, i.e. with
high free perimeter and association with sulphide minerals, is much
higher than that associated with rock-forming and alteration minerals.
However, the mineral associations do not provide information on the
actual liberation of the chalcopyrite grains, and the free perimeter
values should be treated with care, because the milling of the samples
was performed under laboratory conditions which do not necessarily
Fig. 8. Median bootstrapped chalcopyrite mineral associations (percentage of chalcopyrite grain perimeters) of the studied drill core intervals, from MLA GXMAP
analyses of the 40 m crushed material samples (left) and the 4 m crushed material samples from drill hole interval TRSD013 152–192 m (right). FP =free perimeter,
as also seen in some subsequent gures.
Fig. 7. Median value of boot strapped copper deportment of sample TRSD013
152–192 m, with 95 % condence intervals.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
8
correspond with industrial conditions.
The MAMA ratio was calculated to identify preferential associations,
as seen for the 40 m intervals in Fig. 9 and the 4 m samples in Fig. B.5.
There is a clear preferential association of chalcopyrite with other Cu-
bearing minerals, with MAMA ratio values mostly exceeding 10 and
reaching up to 800 for the 40 m samples (Fig. 9). Pyrite and other sul-
phide minerals have lower MAMA ratios, between 6 and 35 and 4–63,
respectively, but the preferential association remains clear when
compared to the rock-forming and alteration minerals which have
maximum MAMA ratios of 5.
Overall, the mineral association uncertainties are high for all sam-
ples, with RSDs usually exceeding 10 % for most minerals (Figs. 10, 11,
B.6). As could be expected, the uncertainties are lower for abundant
minerals or those with a close association to chalcopyrite, including the
main rock-forming and alteration minerals, and greater for those with
lower abundances and/or lower preferential association. Although the
Fig. 12. BSE images (A,C,E) and false colour images from MLA (B,D,F) showing
the preferential associations of Cu-bearing minerals and pyrite, from ground
material samples TRSD017 240–280 m (A,B), TRSD013 164–168 m (C,D) and
TRSD013 758–798 m (E,F). Chalcopyrite and pyrite are commonly intergrown
(A–D) while rims of bornite, covellite and chalcocite form around chalcopyrite
grains (A–D). Hypogene bornite occurring at greater depths is often associated
with chalcopyrite (E,F).
Fig. 10. Median value of bootstrapped chalcopyrite mineral associations
(percentage of chalcopyrite grain perimeters) for TRSD013 152–192 m, with 95
% condence intervals. The dashed line for chalcocite symbolises that the 2.5th
percentile was 0, as also seen in some following gures.
Fig. 11. Median value of MAMA ratio of chalcopyrite, calculated from the
bootstrap resampling results of chalcopyrite mineral associations and modal
mineralogy, for TRSD013 152–192 m, with 95 % condence intervals.
Fig. 9. Median MAMA ratio values of chalcopyrite for the 40 m interval sam-
ples, calculated from the bootstrap resampling results of chalcopyrite mineral
associations and modal mineralogy.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
9
association with bornite is rather variable (RSDs from 12 to 35 %), as-
sociations with the other Cu-bearing minerals are far more variable,
with RSDs from 25 to 94 % for chalcocite, 13–71 % for covellite and
22–63 % for sulphosalts. The RSDs for the 4 m interval samples (Fig. B.7)
are much higher than for the 40 m samples, with few values below 10 %.
As could be expected, the RSDs on the MAMA ratio values are virtually
the same as the equivalent mineral association RSDs (Fig. B.8-9).
The observed preferential associations between Cu-bearing minerals
are linked to the formation of rims of secondary bornite, covellite and
chalcocite around chalcopyrite grains (Fig. 12.A–D), particularly in the
shallow intervals. Hypogene bornite (Fig. 12.E–F) occurs at greater
depths in the system (Blannin et al., 2019). It could be expected that the
MAMA ratios of chalcopyrite with rim-forming Cu-sulphide minerals
should decrease with depth, and that the RSDs of these mineral associ-
ations be low in the shallow intervals due to the formation of clear rims,
and increase with depth. However, no clear trend is seen with depth for
the MAMA ratio of any Cu-bearing mineral, and the RSDs for chalco-
pyrite mineral associations with the Cu-bearing sulphides are typically
high as a result of the low and variable contents of these minerals.
Additionally, the milling process may have made the associations of
these minerals less apparent, and the resolution of the MLA cannot
clearly dene the rims in some cases, as seen by the incomplete rim
characterised by MLA in. Fig. 12.A–B.
4.4. Gold deportment
Gold deportment in the Bolcana porphyry system was investigated
using a combination of MLA and EPMA. A total of 116 gold grains were
identied by MLA, with compositions of <20 wt% Ag (i.e. electrum), <
10 wt% Ag and high purity Au (Fig. 13). The grains are predominantly
electrum, or contain some silver, while a signicant number (25) of the
identied grains were mixed spectra of gold with either quartz, chal-
copyrite or pyrite. For the purpose of simplication, all gold grain
compositions were grouped for further interpretation.
Detailed EPMA analyses indicate that Au concentrations in sulphide
minerals (pyrite, chalcopyrite, bornite, covellite and chalcocite) are low
(<100 ppm). Note that this is not to say that gold in solid solution in
sulphide minerals is negligible, as it may in fact contribute signicantly
to overall gold deportments (see Appendix A.3 for more details and
extended EPMA results). Nevertheless, for the purpose of the subsequent
bootstrap resampling calculations, it was assumed that gold hosted by
native gold is the dominant mode of occurrence in the Bolcana porphyry
system. This assumption was assumed valid as Kesler et al. (2002)
conrmed the importance of native gold grains as metal hosts in por-
phyry copper deposits.
The gold grain size distribution was calculated based on the equiv-
alent circle diameter (ECD) of the grains and is plotted in Fig. 14. Around
67 % of the gold grains are below 4 µm in size, and 90 % below 8 µm. The
coarsest gold grain found has an ECD of 22.5 µm. Fine gold is likely to
have been characterised reasonably well compared to coarse nuggety
gold, which is the hardest to accurately characterise as a result of the
enhanced nugget effect. Therefore, the grain size distribution may be
skewed to ner grain sizes.
4.5. Gold mineral associations
Due to the low number of gold grains, gold mineral associations were
investigated based on individual grains rather than grain mount strips.
To prevent over-interpretation of data from samples with few gold
grains, the gold mineral associations were only studied for intervals
with >10 gold grains, as well as for a composite of all grains in the 40 m
samples (Fig. 15). Gold grains are mainly associated with sulphide
minerals, with the remaining associations reecting the modal miner-
alogy of the samples.
Fig. 14. Gold grain size distribution of all gold grains in the 40 m interval
samples. The number of grains counted in each size fraction is given, based on
the ECD of the grains measured by MLA.
Fig. 13. Histogram of gold compositions for all gold grains identied in the
studied samples.
Fig. 15. Median bootstrapped gold mineral associations (percentage of gold
grain perimeters) of the studied drill core intervals with >10 gold grains. For
interval TRSD013 152–192 m, the gold grains from both the 40 m and 4 m
samples are included. The overall mineral associations for all gold grains are
shown in the composite bar, including all grains from the 40 m crushed ma-
terial samples.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
10
The MAMA ratio was calculated to identify preferential associations,
with the results plotted in Fig. 16. Gold is preferentially associated with
the Cu-sulphide minerals (Fig. 17.A–B,E–H), and pyrite in the near-
surface samples (Fig. 17.I–J), with MAMA ratios typically between 5
and 100, similar to the ndings of Arif and Baker (2004) and Gregory
et al. (2013). There is a high association with Fe-oxide in TRSD013
758–798 m, which is a result of the largest identied gold grain, with an
ECD of 22.5 µm, having a large contact area with Fe-oxide (Fig. 17.C–D).
The uncertainties for the mineral associations of gold are very high
for most minerals (Figs. 18 and B.10). The lowest RSD is typically seen
for either free perimeter (11–62 %) or chalcopyrite (14–29 %), as a
result of these being the largest and most consistent associations.
Therefore, although the uncertainties are quite high, the dominant as-
sociations observed are also the most statistically meaningful. The
association with pyrite is most notably seen in TRSD013 64–104 m and
TRSD013 152–192 m, with RSDs of 49 % and 34 % respectively. Overall,
the uncertainties for the composite of all gold grains, which includes 90
gold grains, are lower than the individual intervals. The MAMA ratio
uncertainties correspond to the mineral association uncertainties
(Figs. 19 and B.11).
5. Discussion
In the following, the quality of the MLA-derived quantitative data is
assessed. The number of MLA measurements required to provide robust
and statistically meaningful data was determined using bootstrap
resampling. The implications for process mineralogical and geo-
metallurgical studies are discussed accordingly.
Fig. 17. BSE images (A,C,E,G,I) and false colour images from MLA (B,D,F,H,J) of gold grains in ground samples TRSD017 386–426 m (A,B,G,H), TRSD013 758–798
m (C,D), TRSD013 152–192 m (E,F), and TRSD013 160–164 m (I,J). The gold grain is indicated by a purple arrow-head in A, B, E, F, I and J. The common association
of gold grains with chalcopyrite is shown in A–B and E–H, with pyrite in I and J, and with rock-forming and alteration minerals in C–H.
Fig. 16. Median MAMA ratio values of gold for the 40 m interval samples with
>10 gold grains and a composite of all gold grains identied in the 40 m in-
tervals, calculated from the bootstrap resampling results of chalcopyrite min-
eral associations and modal mineralogy.
Fig. 18. Median value of bootstrapped gold mineral associations of gold
(percentage of gold grain perimeters) for the composite of all 40 m intervals,
with 95 % condence intervals.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
11
5.1. Data quality
Limited examples of the application of estimated uncertainties to
automated mineralogy studies exist (e.g. Evans and Napier-Munn, 2013;
Mariano and Evans, 2015). Accordingly, there are as yet no universally
accepted guidelines for acceptable levels of uncertainty in such studies.
For the purpose of this study, we therefore decided that RSDs of 10 % for
modal mineralogy and copper-related measurements would be accept-
able, based on industry best-practice (e.g. Dominy et al., 2018) and Gy’s
formula (Gy, 1982). Of course, in different circumstances, higher or
lower uncertainties may be acceptable, depending on the deposit type,
the measurement data and the stage of the project. For instance, higher
uncertainties are expected for measurements of gold content and
textural properties as a results of the nugget effect. We therefore
assumed that a RSD of 20 % is sufcient for gold. The nugget effect is of
course partly a result of the inherent variability of gold grade in the
deposit, but sample size, preparation and analysis can contribute
signicantly to the total nugget effect (e.g. Clark, 2010; Dominy et al.,
2003; 2001; Dominy and Edgar, 2012).
The use of Gy’s formula (Gy, 1982) to guide sampling of particulate
material is well-established, particularly for trace elements, such as
gold, where the nugget effect may be very pronounced and detailed
planning is required to minimise errors during sampling and subsequent
sub-sampling and sample preparation steps. However, such error esti-
mates are often not considered when nally taking analytical samples,
such as for automated mineralogy purposes. Based on the material
properties of the studied samples, Gy’s formula suggests that around
1000 g of sample would be required for a RSD of 20 % for Au content, in
stark contrast to the 0.2 g required for an RSD of 10 % for chalcopyrite
grade. One grain mount contains around 1 g, or less,of sample material.
The RSD of taking 1 g of the material is around 4.5 % for chalcopyrite,
but around 600 % for gold. However, when only one surface of the grain
mount is measured, the equivalent mass measured by MLA will be only a
small fraction of this. This raises the question of the overall represen-
tativity of such samples.
To assess the quality and reliability of the MLA measurements,
calculated Cu and Au grades were compared with the chemical assays
provided by Deva Gold S.A. for the same samples (Fig. 20). Despite
different mineralisation styles and variable mineralogy, the general
agreement between the chemical and calculated assays for Cu is very
good (Fig. 20.A). The Cu contents calculated for the 40 m intervals show
no systematic over- or under-estimation of Cu grade, and all plot on, or
close to, the 1:1 line within error. The RSDs of the calculated Cu contents
in the 40 m samples vary between around 3 % and 8 % and are thus
deemed acceptable. All 4 m samples from TRSD013 152–192 m also lie
on the 1:1 line within error, although the uncertainties are larger than
for the 40 m interval samples, with RSDs between around 3 % and 11 %.
The higher variability recorded for the 4 m samples, relative to the 40 m
interval samples, is mostly related to fewer strips being available for re-
Fig. 20. Plots of calculated Cu (left) and Au (right) contents from MLA versus the Cu and Au contents measured by chemical assays. Both the 40 m crushed material
samples and the 4 m crushed material samples from TRSD013 152–192 m are plotted, with composite values for each. The errors of the MLA data were estimated by
bootstrap resampling, to give 95 % condence intervals. The errors on the chemical assay values were estimated based on blanks, standards and duplicate samples, to
give around ±2 % for Cu and ±5 % for Au.
Fig. 19. Median MAMA ratio values of gold for the 40 m interval samples with
>10 gold grains and a composite of all gold grains identied in the 40 m in-
terval, calculated from the bootstrap resampling results of chalcopyrite mineral
associations and modal mineralogy, with 95 % condence intervals.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
12
sampling during the bootstrapping procedure.
For Au, on the other hand, estimated uncertainties of the calculated
assays for individual samples are very high (Fig. 20.B). The RSDs for the
40 m intervals range from 36 to 88 %, and between 31 and 84 % for the
4 m samples. No clear systematic relationship is apparent between
calculated and chemical assays. However, it can be noted that median
calculated concentrations are often well below measured concentra-
tions. Only in rare cases do calculated Au contents exceed the chemical
assay results by a minor amount. The discrepancy in Au measurements is
likely a result of the low overall concentrations of Au and the low and
variable numbers of grains present in each section, leading to a
pronounced nugget effect. However, it is also possible that a consider-
able amount of gold may be hosted in sulphide minerals, as previously
discussed and shown in Appendix A.3.
5.2. Implications for analytical work
Uncertainties on all MLA-derived textural data estimated by boot-
strap resampling provide a valuable insight into the reliability of the
data (Evans and Napier-Munn, 2013; Mariano and Evans, 2015). In this
study, six measured grain mount surfaces were sufcient to give reliable
results for Cu content with MLA, but not for Au content (Fig. 20). For the
majority of the textural measurements of the 40 m samples, the RSDs
were acceptable for major minerals, such as the dominant minerals in
the modal mineralogy, the Cu deportment and the major chalcopyrite
mineral associations. The high uncertainties encountered for the
textural properties and contents of minor minerals could have been
expected. However, this information is typically not signicant for
geometallurgical modelling and ore processing, except in the case of
gold, where these measurements are of course important and the higher
uncertainties are problematic.
The question remains, how many sample surfaces would need to be
measured to yield reliable results for quantities, such as Au content that
Fig. 21. Bootstrap resampling of Cu content (A) and Au content (B) calculated from MLA data, for all 40 m interval samples. The resampling was carried out by
increasing N in increments of 5, to represent a single grain mount surface composed of 5 strips, until the RSD decreased to 10 % for Cu content and 20 % for Au
content, for all intervals.
Table 4
The number of grain mount surfaces required to characterise different properties with an RSD of 10 % (modal mineralogy, Cu content and chalcopyrite mineral
associations) or 20 % (Au content) for each 40 m interval sample, as estimated by bootstrap resampling.
TRSD013 TRSD017
64–104 m 152–192 m 356–396 m 758–798 m 240–280 m 346–386 m 386–426 m
Cu content 1 2 6 2 3 3 1
Au content 21 20 80 44 75 124 90
Modal mineralogy
*1
7 8 5 6 6 38
*2
5
Ccp mineral associations
*3
8 7 24 37
*4
15 29
*5
26
*1
Including minerals with an abundance of >5 wt% and Cu-bearing minerals which contribute >10 % of Cu deportment – Qz, Fsp and Ccp for all samples, Wm for
all samples except TRSD013 152–192 m, Cly for TRSD013 64–104 m and TRSD013 152–192 m, Chl for all samples except TRSD013 64–104 m, Bt for TRSD013
758–798 m and TRSD017 346–386 m, Fe-O for TRSD013 758–798 m, TRSD017 346–386 m and TRSD017 386–426 m, Py for TRSD017 240–280 m, Bn for TRSD013
64–104 m, TRSD013 758–798 m and TRSD017 346–486 m, Cv for TRSD013 64–104 m, and Cct for TRSD017 346–386 m.
*2
Bn requires 10 surfaces and Cct requires 38, but the remaining dominant minerals require only 5 surfaces.
*3
Including minerals with an association of over 5 % with chalcopyrite – Qz, Fsp, Wm, Py and FP for all samples, Cly for TRSD013 64–104 m and TRSD013 152–192
m, Chl for all samples except TRSD013 64–104 m and TRSD017 386–426 m, Fe–O TRSD017 346–386 m and TRSD017 386–426 m, Bn for TRSD013 64–104 m,
TRSD013 758–798 m and TRSD017 346–386 m and Cct for TRSD017 346–386 m.
*4
Excluding Py which required >50 surfaces.
*5
Excluding Cct which required >50 surfaces.
Table 5
Number of gold grains needed to reach a RSD of 20 % for mineral associations of
all gold in the 40 m interval crushed material samples, for minerals with asso-
ciations >5 % when free perimeter is excluded. N was increased in increments of
5 up to 700 for the bootstrap resampling, to reach the required RSD. The number
of grain mount surfaces required to reach the required RSD was calculated
assuming that two gold grains are present in each grain mount surface.
FP Qz Ccp Bn Py
No. Au grains 45 240 45 690 225
No. grain mount surfaces 23 120 23 345 113
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
13
were not reliably constrained by the current dataset. To answer this
question, further bootstrap resampling was performed to nd the
number of grain mount surfaces required to provide statistically robust
results for the important geometallurgical properties of Cu content, Au
content, modal mineralogy and chalcopyrite and gold mineral associa-
tions with the results summarised in Table 4.
Similar to the ndings of Evans and Napier-Munn (2013), who re-
ported the area of particles to be measured, the copper content calcu-
lated by MLA mainly required 1–3 grain mount surfaces to be analysed
for an RSD of 10 % (Fig. 21.A). However, for the 40 m sample with the
lowest Cu content (<0.2 % Cu), 6 surfaces are required (Table 5, Fig. 21.
A). The minimum number of grain mount surfaces required to charac-
terise Au content with an RSD of 20 % is 20, while the maximum is 124
surfaces (Table 5, Fig. 21.B). There is no clear relation between the
expected gold content and the number of surfaces to be measured.
The modal mineralogy of the 40 m interval samples was bootstrap
resampled to achieve acceptable results for minerals with an abundance
of >5 wt% and Cu-bearing minerals which contribute >10 % of Cu
deportment (Figs. 22 and C.1, Table C1). Overall, analyses of up to
around 3 surfaces are required to accurately characterise the dominant
mineral groups. When also considering the Cu-bearing minerals, the
number of surfaces required increases up to 6 for chalcopyrite, 15 for
bornite, and 38 for the other Cu-sulphides. When accounting for minor
Cu minerals, the number of surfaces required may almost be the same as
for gold due to their low abundance. In such cases, it may be more
appropriate to accept a higher RSD.
Mineral associations of the valuable minerals are important to
consider in geometallurgical studies. Measurements of between around
7 and 37 grain mounts surfaces are required to achieve RSDs of <10 %
for the dominant mineral associations, i.e. minerals with an association
of over 5 % with chalcopyrite (Figs. 23 and C2, Tables 5 and C2). Free
perimeter requires only a single surface for all samples. The number of
surfaces required tends to correlate with the chalcopyrite content, with
some exceptions, e.g. TRSD13 758–798 m, which may be a function of
texture and grain size. Due to their low contents, pyrite in TRSD013
Fig. 22. Bootstrap resampling of modal mineralogy of the TRSD013 152–192
m sample for minerals with an abundance of >5 wt% and Cu-bearing minerals
which contribute >5 % of Cu deportment. N was increased in increments of 5,
to represent a single grain mount surface composed of 5 strips, until the
required RSD of 10 % was reached for all minerals.
Fig. 23. Bootstrap resampling of mineral associations of chalcopyrite in the
TRSD013 152–192 m sample for minerals with an association of >5 % when
ignoring free perimeter. N was increased in increments of 5, to represent a
single grain mount surface composed of 5 strips, until the required RSD of 10 %
was reached for all minerals, or to a limit of 50 surfaces.
Fig. 24. Bootstrap resampling of the mineral associations of all gold grains in
the 40 m samples, for minerals with >5% association. The number of gold
grains was increased in increments of 5 up to 700 for the bootstrap resampling,
to reach the required RSD of 20 %.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
14
758–798 m and chalcocite in TRSD017 346–386 m would require >50
surfaces to be analysed for RSDs of <10 %. Again, it will be necessary to
determine which mineral associations may be most benecial or prob-
lematic, and ensure that these associations can be characterised to an
acceptable condence limit. MAMA ratio uncertainties are closely
related and therefore can be assumed to follow the same trends as the
mineral associations.
The dominant mineral associations, i.e. those with associations >5 %, of
all gold grains in the 40 m samples were bootstrap resampled (Table 5,
Fig. 24). The gold grains in the 4 m samples from TRSD013 152–192 m were
excluded to prevent over-representation of this interval. The uncertainties
for individual intervals were assumed too high to provide meaningful in-
sights. A very large number of gold grains are necessary to provide robust
statistics of the mineral associations for most minerals, even for the main
mineral associations which are assessed here. Free perimeter and chalco-
pyrite require around 45 grains for an RSD of 20 %, compared to 225 for
pyrite, 240 for quartz and up to almost 700 for bornite. Considering that 90
gold grains were identied in 42 grain mount surfaces, equating to around
two gold grains per surface, around 345 grain mount surfaces would be
needed to identify a sufcient number of gold grains to reach the required
RSD. When considering the individual 40 m samples, the number of surfaces
to be measured would increase even further.
The increase in uncertainties on textural measurements resulting
from low mineral grades, which was clearly observed in this study, can
be expected from theoretical principals e.g. Gy (1982). Mariano and
Evans (2015) analysed samples from rougher otation of an iron oxide
Cu-Au ore, sieved into 5 size fractions. Although it was illustrated that
RSDs on measurements reduce with decreasing particle size, the –75
+38 µm fraction of the low grade sample has RSDs from around 20 to
over 100 % for the distribution of pyrite liberation classes. Conversely,
for the same measurements of the same size fraction of a high grade
sample, not even a whole grain mount surface was required to reach an
acceptable RSD of 10 %. When combining the effects of low grade and
grain size, the measurement uncertainties are sure to increase. Assuming
that coarser size fractions are the major source of uncertainties,
measuring multiple size fractions may reduce measurement un-
certainties for coarser size ranges at least, but the effect of grade cannot
be ignored. Overall, the inuence of size fractions has not been clearly
shown in this study.
5.3. Consequences for geometallurgical programmes
Geometallurgical programs requiring the textural data provided by
automated mineralogy typically analyse large numbers of samples.
Although it is common practise to analyse duplicate samples, which
would allow estimations of measurement errors, duplicates may be
sacriced in order to investigate as many different ore types as possible.
Even if this was not the case, the present analysis showed that duplicate
measurements are generally not sufcient to characterise many impor-
tant parameters with an acceptable degree of accuracy. This is particu-
larly true for gold contents and mineral associations. In an industrial
setting, it would simply not be feasible to analyse >124 grain mount
surfaces for one single sample, as has been shown necessary to accu-
rately characterise gold content in some cases. This would require a
large amount of sample material, and time and money for sample
preparation, analysis, and data processing. As a consequence, it may be
necessary to concede that minerals present in minor amounts may never
be measured quantitatively by automated mineralogical techniques,
with the results being semi-quantitative at best, and perhaps only
qualitative.
Given these results, there must be a cost-benet analysis to decide
how many measurements should be performed to reduce uncertainties
to an acceptable value. Decisions must also be made on what textural
properties are most important for ore processing, and what uncertainties
are acceptable. For example, high RSDs should only be considered
critical if they are observed for combinations of ore minerals with
abundant gangue minerals. Any combination of rare ore mineral with
rare gangue mineral will yield elevated RSDs, but this is typically
irrelevant for processing decisions and can therefore be ignored.
Ore textures also inuence the complexity of accurate characterisa-
tion and therefore certain textures could be expected to contribute to
higher uncertainties. For instance, samples with nely inter-grown
minerals are much more challenging to characterise than coarse min-
eral grains. Strategies such as including mixed spectra in the mineral list
are commonly implemented to tackle the high abundance of mixed
spectra resulting from nely intergrown textures (e.g. Kern et al., 2018).
However, this approach is not always straightforward and requires
careful processing of the data to ensure that uncertainties remain low.
The measurement mode plays an important role in the quality of the
data obtained for complex textures. This becomes particularly important
when ore minerals of interest are very ne-grained, such as the gold
grains identied in this study. To improve the data quality it may be
necessary to use more accurate and time-consuming measurement
modes, or combinations of measurement modes, at high resolution (e.g.
GXMAP, Fandrich et al. (2007)) in order to identify ner grains and
reduce the generation of mixed spectra. However, this study encoun-
tered large uncertainties in the Au measurements, even when combining
two of the most accurate modes: GXMAP with SPL. Additionally, mea-
surements of some textural characteristics, such as mineral liberation,
require time-intensive measurement modes where entire particles are
scanned rather than simpler and faster line scans. As previously stated,
decisions must be made ahead of time about which textural properties
are the focus of a study, followed by a weigh-up of the costs and benets
of different measurement modes.
While this study only measured one size fraction, −600
μ
m, it would
perhaps be preferable to sieve each sample into several size fractions.
This strategy is often applied in geometallurgical programmes to reduce
uncertainties by measuring a higher number of particles for each size
fraction. This is particularly relevant for larger size fractions, where
uncertainties are typically signicant (Mariano and Evans, 2015). This
should also result in a reduction of uncertainties related to the cut-effect,
i.e. the fact that the true particle sizes in a grain mount are not known
and the observed sizes give only a lower bound. However, the separate
analysis of several size fractions also directly implies the preparation of
several grain mounts per sample. Unless some optimisation is done with
respect to the number of grain mounts to be measured for each size
fraction, this may in fact lead to greater preparation and measurement
costs of a geometallurgical programme than if samples are prepared
without prior sieving. Coarser size fractions may still require multiple
grain mounts to be prepared for reasonable measurement uncertainties,
increasing the number of analyses further. Unfortunately, the authors
are not aware of any rigorous comparative study assessing the relative
benets of both approaches to sample preparation. Therefore, it is
difcult to say which provides the greater benets for the improvement
of measurement statistics.
The contents, associations and uncertainties of the major minerals, as
well as those of chalcopyrite, the main Cu-bearing mineral, are generally
well-constrained. However, some samples have consistently higher un-
certainties than others (e.g. TRSD017 346–386 m as seen in Table 5). In
other cases, certain textural properties are less well constrained (e.g.
chalcopyrite mineral associations in TRSD013 758–798 m as seen in
Table 5). As can be seen in Table 1, the studied 40 m intervals comprise
different lithologies and styles of alteration and mineralisation.
Considering this, it is perhaps not surprising that some mineral contents
and textural properties exhibit variable uncertainties, albeit far less so
than the Au deportment. It is imperative for reliable results of process
mineralogical and geometallurgical studies, that sampling strategies
should be designed to reduce uncertainties at the earliest stage. For
instance, if large intervals are selected for analyses, where high vari-
ability can already be expected due to combinations of different lithol-
ogies, alteration and mineralisation styles, of course the uncertainties in
the following analyses will be large. Instead, sampling schemes should
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
15
be designed, with the theory of sampling in mind, to focus on domains of
lithology, mineralisation, alteration etc., which are relatively homoge-
neous within the interval.
Bootstrap resampling clearly provides an effective tool for the
determination of uncertainties (Evans and Napier-Munn, 2013; Mariano
and Evans, 2015). The routine application of bootstrapping error as-
sessments should become the norm, with estimated errors being clearly
stated to ensure that any technical decisions take uncertainty of the
results into account. However, bootstrap resampling alone cannot
replace duplicate analyses. Therefore, the sampling strategy, including
running of duplicates at regular intervals, should be carefully planned
and implemented at all stages.
6. Conclusions
Quantitative automated mineralogy methods have been, and will
continue to be, invaluable for many reasons. While great advances have
been made in the eld since their advent, the statistical uncertainties
accompanying the results, which were given much consideration during
the early development of these methods (e.g. Butcher et al., 2000;
Gottlieb et al., 2000), seem to have been somewhat sacriced for the
speed and cost of the measurements. An in-depth study into un-
certainties associated with automated mineralogy-based metal deport-
ment and mineral association studies has been conducted to illustrate
the need for the inclusion of error estimates, particularly for geo-
metallurgical studies where technical decisions may be made based on
the data. Such uncertainty assessments can be readily conducted on all
relevant textural parameters using the bootstrap resampling method.
Different minerals could also be studied, although caution must be taken
with minerals present in low concentrations, as clearly illustrated by the
high uncertainties for gold measurements in comparison to chalcopyrite.
Additionally, the method is not limited to the specic case study of a
porphyry Au-Cu deposit, but should be directly transferrable to any
deposit type, including both primary and secondary deposits such as
mine and metallurgical wastes.
Despite different mineralisation styles and variable mineralogy,
estimated statistical uncertainties on Cu content and mineral associa-
tions are low for the selected case study, a low-grade Cu-Au porphyry
system with concentrations of Cu well below 0.5 wt%. The same cannot
be said for Au content and mineral associations, which have very large
uncertainties. This is, of course, attributed to the very low concentration
of gold, the low number of gold-bearing mineral grains and the nugget
effect. Critically, our results show that the attainment of acceptable
uncertainties for metal grades does not necessarily imply that the same is
the case for other important parameters such as deportment or mineral
association. To constrain these characteristics, more data is typically
needed.
Although it was determined that up to 5 grain mount surfaces would
be sufcient to provide robust measurement statistics in most cases, this
is still a large number of measurements for a single sample for any
geometallurgical programme. In short, the selection of samples which
are representative of a whole ore deposit, while minimising time, cost
and importantly uncertainties on the textural parameters of interest is
still challenging. It should become the industry norm that such assess-
ments of uncertainties come hand-in-hand with automated mineralogy-
based studies, to ensure that meaningful interpretation is possible. Great
care should be taken rst when planning sampling schemes and auto-
mated mineralogy studies, and when subsequently reporting analytical
results, especially for precious metals and trace minerals which will
generally show higher uncertainties.
CRediT authorship contribution statement
Rosie Blannin: Conceptualization, Methodology, Software, Valida-
tion, Formal analysis, Investigation, Data curation, Writing - original
draft, Writing - review & editing, Visualization. Max Frenzel:
Conceptualization, Validation, Writing - review & editing. Laura Tusa:
Investigation, Writing - review & editing. Sandra Birtel: Writing - re-
view & editing, Supervision. Paul Iv˘
ascanu: Resources, Writing - re-
view & editing, Supervision. Tim Baker: Resources, Writing - review &
editing. Jens Gutzmer: Conceptualization, Writing - review & editing,
Supervision.
Declaration of Competing Interest
The authors declare that they have no known competing nancial
interests or personal relationships that could have appeared to inuence
the work reported in this paper.
Acknowledgements
We would like to acknowledge the anonymous reviewer for their
constructive comments and valuable insight. Sabine Gilbricht (TU Ber-
gakademie Freiberg), Kristine Trinks (HZDR) and Kai Bachmann
(HZDR) are thanked for their support during SEM-MLA data acquisition.
Joachim Krause (HZDR) is acknowledged for EPMA data acquisition and
processing. The sample preparation staff at the Helmholtz Institute
Freiberg for Resource Technology are thanked for the great help in
preparing the samples. The rst author would like to thank the European
Union’s Erasmus Mundus Program for a scholarship of the Emerald MSc
program. In addition, the completion of the paper was supported by the
SULTAN project. This project has received funding from the European
Union’s EU Framework Programme for Research and Innovation Hori-
zon 2020 under Grant Agreement No 812580.
Appendix A
Electron microprobe analysis method
EPMA analyses were carried out at HIF, using a JEOL JXA-8530F
EPMA equipped with ve wavelength dispersive spectrometers and a
eld emission electron gun. Twenty-one elements were analysed on 585
points (including standard blocks for major elements every ~ 160
points), each with a measurement time of eight and a half minutes, using
an acceleration voltage of 20 Kv and a probe current of 35 nA.
Mineral grains suitable for measurement were identied using the
MLA GXMAP results of samples TRSD013 64–104 m, TRSD017 386–426
m and TRSD013 758–798 m and TRSD013 184–188 m. Grains of gold,
pyrite, chalcopyrite, bornite, covellite and/or chalcocite were book-
marked. The measurement points were efciently located with the point
logger, the coordinates recorded and transferred to the EPMA (cf.
Osbahr et al. 2015). The grain mounts were coated twice with graphite
to reduce the likelihood of surface charging. The measurement settings
are found in Table A1.
During and after data acquisition, online and ofine corrections were
performed: (1) an ofine overlap correction method, based on weight
proportions of elements present (cf. Osbahr et al. 2015); (2) background
corrections to remove the contribution from the background to the
measured peak intensity (Lavrent’ev et al., 2015; Osbahr et al., 2015;
Reed, 2005); (3) a drift correction, carried out using the standard
measurements. The corrected measurements were ltered for values
below the quantication limit, and analytical errors exceeding 10 %.
Measurements with totals of 100 ±1.5 % were retained. The stoichi-
ometry of the minerals measured was calculated for S, Fe and Cu and
ltered for errors of >10 % for pyrite and chalcopyrite and >15 % for
bornite and chalcocite and covellite.
Copper deportment
The mineral chemistry of chalcopyrite, bornite, covellite and chal-
cocite were measured with EPMA in four grain mounts (TRSD013
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
16
64–104 m, 184–188 m, 758–798 m and TRSD017 386–426 m). The
stoichiometric compositions of the measured minerals were calculated,
based on the Fe, S and Cu contents (Table A2). It can be seen that there
are far fewer valid measurements for bornite, chalcocite and covellite
compared to chalcopyrite. This is partly because they occur at lower
abundances, but also because the surfaces of the minerals were often
altered and good measurements could not be taken. Despite this, no
systematic bias in measurements for a specic mineral composition was
detected. Therefore, the EPMA results validate the use of the standard
stoichiometric mineral formulae to calculate copper deportment.
The average areas of the copper-bearing minerals are given in
Table A3, for each 40 m and 4 m interval sample. The area of chalco-
pyrite is one or two orders of magnitude greater than bornite, chalcocite,
covellite and sulphosalts. This clearly illustrates the dominance of
chalcopyrite in the copper deportment of all samples.
Gold deportment
EPMA was used to analyse the composition of two gold grains in
sample TRSD013 184–188 m (Table A3). These have molar Au:Ag ratios
of 12.7 to 94.1, corresponding to the Ag contents detected by MLA
(Fig. 13).
In this study, gold is assumed to be mainly hosted by gold grains.
However, signicant amounts of gold could be hosted by sulphide
minerals in solid solution, as is common in porphyry deposits (e.g. Arif
and Baker, 2004; Gregory et al., 2013; Kesler et al., 2002; Simon et al.,
2000)). However, detailed EPMA analyses indicate that Au is absent in
concentrations close to, or above, the detection limit of around 100 ppm
in all measured sulphides (pyrite, chalcopyrite, bornite, covellite and
chalcocite). In contrast, Cioacǎ et al. (2014) estimated that samples from
the Bolcana system contained, on average, 46 ppm Au in pyrite, 108
ppm Au in chalcopyrite and 116 ppm Au in bornite, from EPMA studied.
Limited information can be inferred from this as a result of the samples
being from different locations in the Bolcana system.
To provide an estimate of how much gold may be hosted by sulphides
in the studied samples, it was assumed that pyrite, chalcopyrite and
bornite could host up to 50 ppm Au in solid solution, i.e. half the
detection limit of 100 ppm (Table A4). The estimated total content of
gold in solid solution varies between 0.5 and 2.9 ppm. However, when
compared to the ‘missing’ gold, i.e. the difference between the gold
calculated assay from MLA and the chemical assay, the estimate of solid
solution gold generally exceeds the ‘missing’ gold and therefore it is
likely an over-estimate. As a result, for the purpose of deportment cal-
culations, it was thus assumed that gold contents in these sulphide
minerals are negligible. Following this assumption, gold deportment in
all studied samples is limited to the presence of native gold grains.
However it is still likely that gold occurs in solid solution in some sul-
phide minerals, in variable amounts throughout the different intervals
due to varying styles of mineralisation, over-printing, alteration and
lithologies.
See Table A1, Table A2, Table A3, Table A4, Table A5
Table A1
Analytical conditions for EPMA analysis (Spectr =spectrometer; Peak pos. =peak position; Lower backgr. =lower background; Upper backgr. =upper background;
Meas. time peak =measurement time peak; Meas. time backgr. =measurement time background; Quant. limit =quantication limit). Standards supplied by ASTIMEX
Standards Ltd.
Element/
Line
Spectr./
Crystal
Peak pos.
(mm)
Lower backgr.
(mm)
Upper backgr.
(mm)
Meas. time peak
(s)
Meas. time backgr.
(s)
Quant. limit
(ppm)
Standards
Si K
α
1 1 TAP 77.311 / 5.565 20 10 99 Plagioclase_AST
Al K
α
1 2 TAP 90.584 6.858 9.337 20 5 129 Plagioclase_AST
As L
α
1 3 TAP 105.132 6.127 2.286 60 15 200 Arsenopyrite_AST
Se L
α
1 1 TAP 97.678 5.588 3.06 60 15 222 BismuthSelenite_AST
Sn L
α
1 2 PETJ 114.97 1.976 6.644 80 20 262 Tin_AST
Ag L
α
1 2 PETJ 132.697 5.219 1.874 120 30 300 Silver_AST
S K
α
1 2 PETJ 171.637 / 4.942 20 10 230 Sphalerite_AST
Hg M
α
1 2 PETJ 180.36 3.949 9.307 60 15 751 Cinnabar_AST
In L
α
1 3 PETL 121.114 16.693 8.014 160 15 192 IndiumPhospide_AST
Cd L
α
1 3 PETL 126.98 / 2.463 60 30 115 Cadmium_AST
Au M
α
1 3 PETL 187.002 10.418 9.281 180 45 304 Gold_AST
Zn K
α
1 4 LIFH 99.99 4.963 5.016 40 10.5 300 Sphalerite_AST
Cu K
α
1 4 LIFH 107.345 1.96 2.131 30 7.5 290 Copper_AST
Fe K
α
1 4 LIFH 134.885 5.1 3 30 7.5 217 Pentlandite_AST
Co K
α
1 4 LIFH 124.647 3.106 2.745 60 15 187 Cobalt_AST
Te K
α
1 4 PETH 105.15 10.989 7.96 80 20 107 Tellurium_AST
Sb Lβ1 4 PETH 103.143 8.361 / 40 20 343 Stibnite_AST
Ni K
α
1 5 LIFH 115.396 5.397 4.581 40 10 238 Pentlandite_AST
Mn K
α
1 5 LIFH 146.333 2.479 5.177 40 10 194 Rhodonite_AST
Bi Mβ1 5 PETH 157.205 5.125 18.723 60 15 325 BismuthSelenite_AST
Pb K
α
1 5 PETH 169.255 / 7.835 40 20 266 Galena_AST
Table A2
Stoichiometric compositions of Cu-bearing sulphide minerals, as calculated from EPMA results. n =number of valid measurements (total of 98.5–100.5 wt%, error of
<10 % in stoichiometry for chalcopyrite, and <15 % for bornite, chalcocite and covellite), Med. =median value and STD =standard deviation. Note that the low
number of valid measurements for bornite, chalcocite and particularly covellite mean that the standard deviation is not as statistically meaningful as for chalcopyrite.
Mineral n S Fe Cu
Med. STD Med. STD Med. STD
Ccp 92 2.2 0.02 1.2 0.01 1.2 0.01
Bn 7 1.6 0.04 0.4 0.03 2.2 0.05
Cct 5 1.4 0.09 0.1 0.04 2.6 0.07
Cv 2 1.5 0.02 0.1 0.05 1.3 0.03
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
17
Table A3
The average area (mm
2
) of Cu-bearing minerals identied by MLA in the six measured grain mount surfaces of each 40 m interval sample, and the one measured surface for the 4 m interval grain mounts from TRSD013
152–192 m. No value indicates that the mineral was not identied, or the average area of the given mineral was <0.001 mm
2
. The average area of gold is given in µm
2
rather than mm
2
, with 1 µm
2
equalling 1e-6 mm
2
.
TRSD0S13 TRSD0S17 TRSD013 152–192 m interval 4 m samples
64–104
m
152–192
m
356–396
m
758–798
m
240–280
m
346–386
m
386–426
m
152–156
m
156–160
m
160–164
m
164–168
m
168–172
m
172–176
m
176–180
m
180–184
m
184–188
m
188–192
m
Ccp 0.31 0.41 0.14 0.51 0.18 0.13 0.78 0.32 0.39 0.70 0.53 0.96 0.89 0.38 0.34 0.57 0.29
Bn 0.017 0.012 0.002 0.067 0.003 0.017 0.02 0.001 0.001 0.008 0.002 0.008 0.002 0.007 0.02 0.05 0.01
Cct 0.003 0.003 0.012 0.003
Cv 0.018 0.015 0.008 0.002 0.007 0.005 0.001 0.001 0.005 0.005 0.01 0.05 0.05 0.03
Ss 0.003 0.005 0.003 0.002 0.06 0.007 0.002 0.004 0.004 0.002
Au 11.20 39.13 5.99 147.27 6.18 16.41 96.42 1.24 7.42 0.78 8.72 7.49 0.91 8.72 1.95
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
18
Table A5
Potential concentrations of Au hosted by pyrite, chalcopyrite and bornite in each 40 m interval, assuming concentrations of 50 ppm Au in each mineral. The value of
‘missing’ gold (i.e. the difference between the MLA calculated assay and the chemical assay) is given for comparison, except for TRSD013 356–396 m and TRSD017
386–342 m, where the Au content calculated from the MLA results exceeds the chemical assay.
TRSD013 TRSD017
64–104 m 152–192 m 356–396 m 758–798 m 240–280 m 346–386 m 386–426 m
Pyrite 2.1 1.5 0.9 0.5 2.7 0.3 0.9
Chalcopyrite 0.4 0.5 0.8 0.5 0.2 0.2 0.8
Bornite 0.02 0.02 0.02 0.1 0.01 0.03 0.02
Total 2.5 2.0 1.7 1.0 2.9 0.5 1.7
‘Missing’ Au 0.4 1.7 – 1.9 0.2 1.4 –
Table A4
EPMA results of valid measurements of gold grains (total of 100 ±1.5 wt%). Elements with 0 content were removed from the table.
Measurement code Measured elemental composition (wt%) Molar Ratio
Au:Ag
Si Ag Cd Au Cu Fe Total
184_188 Gold 2 0.12 0.63 99.1 0.90 0.57 101.3 94.1
184_188 Gold 3 0.13 4.50 0.04 95.8 0.34 100.8 12.7
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
19
Appendix B
See Fig. B1, Fig. B2, Fig. B3, Fig. B4, Fig. B5, Fig. B6, Fig. B7, Fig. B8,
Fig. B9, Fig. B10, Fig. B11
Fig. B1. Median value of bootstrapped modal mineralogy, with error bars representing 95 % condence intervals. (A) TRSD013 64–104 m; (B) TRSD013 356–396 m;
(C) TRSD013 758–798 m; (D) TRSD017 240–280 m; (E) TRSD017 346–386 m; (F) TRSD017 386–426 m.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
20
Fig. B2. Median value of bootstrapped modal mineralogy, with error bars representing 95 % condence intervals. (A) TRSD013 152–156 m; (B) TRSD013 156–160
m; (C) TRSD013 160–164 m; (D) TRSD013 164–168 m; (E) TRSD013 168–172 m; (F) TRSD013 172–176 m; (G) TRSD013 176–180 m; (H) TRSD013 180–184 m; (I)
TRSD013 184–188 m; (J) TRSD013 188–192 m.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
21
Fig. B3. Median value of bootstrapped copper deportment, with error bars representing 95 % condence intervals. (A) TRSD013 64–104 m; (B) TRSD013 356–396
m; (C) TRSD013 758–798 m; (D) TRSD017 240–280 m; (E) TRSD017 346–386 m; (F) TRSD017 386–426 m.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
22
Fig. B4. Median value of bootstrapped copper deportment, with error bars representing 95 % condence intervals. (A) TRSD013 152–156 m; (B) TRSD013 156–160
m; (C) TRSD013 160–164 m; (D) TRSD013 164–168 m; (E) TRSD013 168–172 m; (F) TRSD013 172–176 m; (G) TRSD013 176–180 m; (H) TRSD013 180–184 m; (I)
TRSD013 184–188 m; (J) TRSD013 188–192 m. Chalcocite was only present in sample TRSD013 164–168 m, and is therefore zero and not plotted in the
other graphs.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
23
Fig. B5. Median MAMA ratio values of chalcopyrite for the 4 m interval samples, calculated from the bootstrap resampling results of chalcopyrite mineral asso-
ciations and modal mineralogy.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
24
Fig. B6. Median value of bootstrapped chalcopyrite mineral associations (percentage of chalcopyrite grain perimeters), with error bars representing 95 % condence
intervals. (A) TRSD013 64–104 m; (B) TRSD013 356–396 m; (C) TRSD013 758–798 m; (D) TRSD017 240–280 m; (E) TRSD017 346–386 m; (F) TRSD017 386–426 m.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
25
Fig. B7. Median value of bootstrapped chalcopyrite mineral associations (percentage of chalcopyrite grain perimeters), with error bars representing 95 % condence
intervals. The dashed lines in A, B, C, D, G, H, I and J symbolise that the 2.5th percentile of the particular mineral was 0. (A) TRSD013 152–156 m; (B) TRSD013
156–160 m; (C) TRSD013 160–164 m; (D) TRSD013 164–168 m; (E) TRSD013 168–172 m; (F) TRSD013 172–176 m; (G) TRSD013 176–180 m; (H) TRSD013
180–184 m; (I) TRSD013 184–188 m; (J) TRSD013 188–192 m.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
26
Fig. B8. Median chalcopyrite MAMA ratio values, calculated from bootstrap resampled mineral associations and mineral areas, with error bars representing 95 %
condence intervals. (A) TRSD013 64–104 m; (B) TRSD013 356–396 m; (C) TRSD013 758–798 m; (D) TRSD017 240–280 m; (E) TRSD017 346–386 m; (F) TRSD017
386–426 m.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
27
Fig. B9. Median chalcopyrite MAMA ratio values, calculated from bootstrap resampled mineral associations and mineral areas, with error bars representing 95 %
condence intervals. (A) TRSD013 152–156 m; (B) TRSD013 156–160 m; (C) TRSD013 160–164 m; (D) TRSD013 164–168 m; (E) TRSD013 168–172 m; (F) TRSD013
172–176 m; (G) TRSD013 176–180 m; (H) TRSD013 180–184 m; (I) TRSD013 184–188 m; (J) TRSD013 188–192 m.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
28
Fig. B10. Median value of bootstrapped gold mineral associations (percentage of chalcopyrite grain perimeters), with error bars representing 95 % condence
intervals. (A) TRSD013 64–104 m; (B) TRSD013 152–192 m, including gold grains from both the 40 m and 4 m samples; (C) TRSD013 758–798 m; (D) TRSD017
386–426 m.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
29
Fig. B11. Median gold MAMA ratio values, calculated from bootstrap resampled mineral associations and mineral areas, with error bars representing 95 % con-
dence intervals. (A) TRSD013 64–104 m; (B) TRSD013 152–192 m, including gold grains from both the 40 m and 4 m samples; (C) TRSD013 758–798 m; (D)
TRSD017 386–426 m.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
30
Appendix C
See Fig. C1 and Fig. C2
See Table C1 and Table C2
Fig. C1. Bootstrap resampling of modal mineralogy of the 40 m intervals for minerals with an abundance of >5 wt%, or Cu-bearing minerals which contribute >5 %
of Cu deportment. N was increased until the required RSD of 10 % was reached for all minerals. (A) TRSD013 64–104 m; (B) TRSD013 356–396 m; (C) TRSD013
758–798 m; (D) TRSD017 240–280 m; (E) TRSD017 346–386 m; (F) TRSD017 386–426 m.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
31
Fig. C2. Bootstrap resampling of mineral associations of chalcopyrite in the 40 m intervals for minerals with an association of >5 % when free perimeter is excluded.
N was increased until the required RSD of 10 % was reached for all minerals, or to a limit of 50 surfaces. (A) TRSD013 64–104 m; (B) TRSD013 356–396 m; (C)
TRSD013 758–798 m; (D) TRSD017 240–280 m; (E) TRSD017 346–386 m; (F) TRSD017 386–426 m.
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
32
References
Arif, J., Baker, T., 2004. Gold paragenesis and chemistry at Batu Hijau, Indoneisa:
Implications for gold-rich porphyry copper deposits. Miner. Depos. 39, 523–535.
https://doi.org/10.1007/s00126-004-0433-0.
Bachmann, K., Frenzel, M., Krause, J., Gutzmer, J., 2017. Advanced Identication and
Quantication of In-Bearing Minerals by Scanning Electron Microscope-Based Image
Analysis. Microsc. Microanal. 23, 527–537. https://doi.org/10.1017/
S1431927617000460.
Bachmann, K., Osbahr, I., Tolosana-Delgado, R., Chetty, D., Gutzmer, J., 2018. Variation
in Platinum Group Mineral and Base Metal Sulde Assemblages in the Lower Group
Chromitites of the Western Bushveld Complex, South Africa. Can. Mineral. 56,
723–743. https://doi.org/10.3749/canmin.1700094.
Baker, T., 2019. Gold ±copper endowment and deposit diversity in the western Tethyan
Magmatic Belt, Southeast Europe: Implications for exploration. Econ. Geol. 114,
1237–1250. https://doi.org/10.5382/econgeo.4643.
Benvie, Byron, Chapman, Nicole M., Robinson, David J., Kuhar, Laura L., 2013. A robust
statistical method for mineralogical analysis in geometallurgical diagnostic leaching.
Miner. Eng. 52, 178–183. https://doi.org/10.1016/j.mineng.2013.06.010.
Berbeleac, I., Udubas¸a, S.S., Iatan, E.-L., Vis¸an, M., 2014. Geological and Structural
Constraints on the Localization of Neogene Porphyry– Epithermal Related Cu-Au
(Mo), and Epigenetic Hydrothermal Deposits/Prospects from South Apuseni Mts.
Romania. Rom. J. Miner. Depos. 87, 47–52.
Blannin, R., Tusa, L., Birtel, S., Gutzmer, J., Gilbricht, S., Ivǎs
¸canu, P., 2019. Metal
deportment and ore variability of the Bolcana porphyry Au – Cu system (Apuseni Mts
, Romania) – Implications for ore processing. 15th SGA Bienn. Meet. 2019 Proc.
Buchmann, M., Schach, E., Tolosana-Delgado, R., Leißner, T., Astoveza, J., Kern, M.,
M¨
ockel, R., Ebert, D., Rudolph, M., van den Boogaart, K.G., Peuker, U.A., 2018.
Evaluation of magnetic separation efciency on a cassiterite-bearing skarn ore by
means of integrative SEM-based image and XRF–XRD data analysis. Minerals 8.
https://doi.org/10.3390/min8090390.
Butcher, A.R., Helms, T.A., Gottlieb, P., Bateman, R., Ellis, S., Johnson, N.W., 2000.
Advances in the quantication of gold deportment by QemSCAN. Seventh Mill Oper.
Conf. Kalgoorlie, WA, AusIMM 267–271.
Cardon, O., Reisberg, L., Andr´
e-Mayer, A.S., Leroy, J.L., Milu, V., Zimmermann, C., 2008.
Re-Os systematics of pyrite from the bolcana porphyry copper deposit, Apuseni
Mountains. Romania. Econ. Geol. 103, 1695–1702. https://doi.org/10.2113/
gsecongeo.103.8.1695.
Chayes, F., 1945. Petrographic analysis by fragment counting: Part 2. Precision of micro-
sampling and the combination error of sampling and counting. Econ. Geol. 40,
517–525. https://doi.org/10.2113/gsecongeo.40.8.517.
Chayes, F., 1944. Petrographic analysis by fragment counting: Part 1. The Counting
Error. Econ. Geol. 39, 484–505. https://doi.org/10.2113/gsecongeo.39.7.484.
Chernick, M.R., 1999. Bootstrap Methods: A Practitioner’s Guide. Wiley, New York.
Cioacǎ, M.E., 2013. Investigation of the rare elements from the porphyry copper deposits
in Romania.
Cioacǎ, M.E., Munteanu, M., Qi, L., Costin, G., 2014. Trace element concentrations in
porphyry copper deposits from Metaliferi Mountains, Romania: A reconnaissance
study. Ore Geol. Rev. 63, 22–39. https://doi.org/10.1016/j.oregeorev.2014.04.016.
Clark, I., 2010. Statistics or geostatistics? Sampling error or nugget effect? J. South.
African Inst. Min. Metall. 110, 307–312.
Core, D.P., Kesler, S.E., Essene, E.J., 2006. Unusually Cu-rich magmas associated with
giant porphyry copper deposits: Evidence from Bingham, Utah. Geology 34, 41–44.
Cropp, A.F., Goodall, W.R., Bradshaw, D.J., 2013. The Inuence of Textural Variation
and Gangue Mineralogy on Recovery of Copper by Flotation from Porphyry Ore – A
Review. Second AusIMM Int. Geometallurgy Conf. 30, 279–291.
D´
enes, R´
eka, Kiss, Gabriella B., Ivǎs¸canu, Paul, 2015. Petrographic and geochemical
study of the porphyry and epithermal mineralization in the bolcana magmatic-
hydrothermal system (Apuseni Mts, Romania). 13th SGA Bienn. Meet. 2015 Proc.
Dominy, S.C., Edgar, W.B., 2012. Approaches to reporting grade uncertainty in high
nugget gold veins. Trans. Institutions Min. Metall. Sect. B Appl. Earth Sci. 121,
29–42. https://doi.org/10.1179/1743275812Y.0000000013.
Table C1
Number of grain mount surfaces necessary to reach a RSD of 10 % for minerals with an abundance of >5 wt%, or Cu-bearing minerals which contribute >5 % of Cu
deportment in the 40 m interval crushed material samples. N was increased in increments of 5, to represent a whole grain mount surface composed of 5 strips, for the
bootstrap resampling until the required RSD was reached.
Mineral TRSD013 TRSD017
64–104 m 152–192 m 356–396 m 758–798 m 240–280 m 346–386 m 386–426 m
Quartz 1 1 1 2 2 2 2
Feldspars 1 1 1 1 1 1 1
White micas 1 1 3 – 1 2 3
Clay minerals 2 8 – – – – –
Biotite – – – 3 – 2 –
Chlorite – 3 2 3 2 2 2
Fe-oxides – – – 3 – 3 4
Chalcopyrite 3 3 5 4 6 5 5
Bornite 5 – – 6 – 10 –
Chalcocite – – – – – 38 –
Covellite 7 – – – – – –
Pyrite – – – – 3 – –
Table C2
Number of grain mount surfaces necessary to reach a RSD of 10 % for the mineral associations of chalcopyrite, with minerals with an association of >5 % when free
perimeter is excluded, in the 40 m interval crushed material samples. N was increased in increments of 5, to represent a whole grain mount surface composed of 5
strips, for the bootstrap resampling until the required RSD was reached, or until the number of surfaces reached 50.
Mineral TRSD013 TRSD017
64–104 m 152–192 m 356–396 m 758–798 m 240–280 m 346–386 m 386–426 m
Quartz 4 4 7 15 9 29 8
Feldspars 5 6 11 8 7 17 9
White micas 6 5 12 37 10 16 7
Clay minerals 7 6 – – – – –
Chlorite – 6 13 14 9 14 –
Fe-oxides – – – – – 26 26
Bornite 8 – – 20 – 21 –
Chalcocite – – – – – >50 –
Pyrite 7 7 24 >50 15 25 12
Free perimeter 1 1 1 1 1 1 1
Chalcopyrite content (wt%) 0.7 0.9 0.3 0.9 0.4 0.4 1.5
R. Blannin et al.
Minerals Engineering 167 (2021) 106836
33
Dominy, S.C., Johansen, G.F., Annels, A.E., 2001. Bulk sampling as a tool for the grade
evaluation of gold-quartz reefs. Appl. Earth Sci. Trans. Institutions Min. Metall. Sect.
B 110, B176–B191. https://doi.org/10.1179/aes.2001.110.3.176.
Dominy, S.C., O’Connor, L., Glass, H.J., Purevgerel, S., Xie, Y., 2018. Towards
representative metallurgical sampling and gold recovery testwork programmes.
Minerals 8. https://doi.org/10.3390/min8050193.
Dominy, S.C., Platten, I.M., Raine, M.D., 2003. Grade and geological continuity in high-
nugget effect gold-quartz reefs: Implications for resource estimation and reporting.
Appl. Earth Sci. Trans. Inst. Min. Metall. Sect. B 112, B239–B259. https://doi.org/
10.1179/037174503225003116.
Efron, B., 1979. Bootstrap methods: another look at the jacknife. Ann. Stat. 7, 1–26.
Evans, C.L., Napier-Munn, T.J., 2013. Estimating error in measurements of mineral grain
size distribution. Miner. Eng. 52, 198–203. https://doi.org/10.1016/j.
mineng.2013.09.005.
Fandrich, R., Gu, Y., Burrows, D., Moeller, K., 2007. Modern SEM-based mineral
liberation analysis. Int. J. Miner. Process. 84, 310–320. https://doi.org/10.1016/j.
minpro.2006.07.018.
Frenzel, M., Bachmann, K., Carvalho, J.R.S., Relvas, J.M.R.S., Pacheco, N., Gutzmer, J.,
2018. The geometallurgical assessment of by-products—geochemical proxies for the
complex mineralogical deportment of indium at Neves-Corvo, Portugal. Miner.
Depos. https://doi.org/10.1007/s00126-018-0849-6.
Goodall, W.R., 2008. Characterisation of mineralogy and gold deportment for complex
tailings deposits using QEMSCAN®. Miner. Eng. 21, 518–523. https://doi.org/
10.1016/j.mineng.2008.02.022.
Goodall, W.R., Butcher, A.R., 2012. The use of QEMSCAN in practical gold deportment
studies. Trans. Institutions Min. Metall. Sect. C Miner. Process. Extr. Metall. 121,
199–204. https://doi.org/10.1179/1743285512Y.0000000021.
Gottlieb, P., Wilkie, G., Sutherland, D., Ho-Tun, E., Suthers, S., Perera, K., Jenkins, B.,
Spencer, S., Butcher, A., Rayner, J., 2000. Using quantitative electron microscopy for
process mineralogy applications. Jom 52, 24–25. https://doi.org/10.1007/s11837-
000-0126-9.
Gregory, M.J., Lang, J.R., Gilbert, S., Hoal, K.O., 2013. Geometallurgy of the Pebble
porphyry copper-gold-molybdenum deposit, Alaska: Implications for gold
distribution and paragenesis. Econ. Geol. 108, 463–482. https://doi.org/10.2113/
econgeo.108.3.463.
Gu, Y., 2003. Automated Scanning Electron Microscope Based Mineral Liberation
Analysis, An Introduction to JKMRC/FEI Mineral Liberation Analyser. J. Miner.
Mater. Charact. Eng. 02, 33–41. https://doi.org/10.4236/jmmce.2003.21003.
Guseva, O., Opitz, A.K.B., Broadhurst, J.L., Harrison, S.T.L., Becker, M., 2021.
Characterisation and prediction of acid rock drainage potential in waste rock: Value
of integrating quantitative mineralogical and textural measurements. Miner. Eng.
163, 106750 https://doi.org/10.1016/j.mineng.2020.106750.
Gy, P., 1982. Sampling of Particulate Materials: Theory and Practice. Elsevier Scientic
Publ. Co., Amsterdam, New York.
Hannula, J., Kern, M., Luukkanen, S., Roine, A., van den Boogaart, K.G., Reuter, M.A.,
2018. Property-based modelling and simulation of mechanical separation processes
using dynamic binning and neural networks. Miner. Eng. 126, 52–63. https://doi.
org/10.1016/j.mineng.2018.06.017.
Heinig, T., Bachmann, K., Boogaart, G. Van Den, Gutzmer, J., 2015. Monitoring
gravitational and particle shape settling effects on MLA sampling preparation. IAMG
Conf. 2015, Freiberg, Ger.
Ivǎs¸canu, P., Baker, T., Lewis, P., Kulcsar, Z., D´
enes, R., Tamas, C., 2019. Bolcana,
Romania: Geology and discovery history of a gold rich porphyry deposit, in:
NewGenGold 2019 Conference Proceedings: Perth, Australia, 2019. pp. 69–82.
Ivǎs¸canu, P., Kulcsar, Z., D´
enes, R., Baker, T., Lewis, P., 2018. Bolcana Porphyry Au-Cu
Deposit, Apuseni Mountains, Romania. AME Roundup 2018 Core Shack Abstr. Guid.
31.
Kern, M., K¨
astner, J., Tolosana-Delgado, R., Jeske, T., Gutzmer, J., 2019. The inherent
link between ore formation and geometallurgy as documented by complex tin
mineralization at the H¨
ammerlein deposit (Erzgebirge, Germany). Miner. Depos. 54,
683–698. https://doi.org/10.1007/s00126-018-0832-2.
Kern, M., M¨
ockel, R., Krause, J., Teichmann, J., Gutzmer, J., 2018. Calculating the
deportment of a ne-grained and compositionally complex Sn skarn with a modied
approach for automated mineralogy. Miner. Eng. 116, 213–225. https://doi.org/
10.1016/j.mineng.2017.06.006.
Kesler, S.E., Chryssoulis, S.L., Simon, G., 2002. Gold in porphyry copper deposits: Its
abundance and fate. Ore Geol. Rev. 21, 103–124. https://doi.org/10.1016/S0169-
1368(02)00084-7.
Lastra, R., Paktunc, D., 2016. An estimation of the variability in automated quantitative
mineralogy measurements through inter-laboratory testing. Miner. Eng. https://doi.
org/10.1016/j.mineng.2016.06.025.
Lavrent’ev, Y.G., Karmanov, N.S., Usova, L. V., 2015. Electron probe microanalysis of
minerals: Microanalyzer or scanning electron microscope? Russ. Geol. Geophys. 56,
1154–1161. https://doi.org/10.1016/j.rgg.2015.07.006.
Leichliter, S.E., 2013. Gold Deportment and Geometallurgical Recovery Model for the La
Colosa, Porphyry Gold Deposit. University of Tasmania, Colombia.
Leigh, G.M., Sutherland, D.N., Gottlieb, P., 1993. Condence limits for liberation
measurements. Miner. Eng. 6, 155–161. https://doi.org/10.1016/0892-6875(93)
90129-B.
Leißner, T., Bachmann, K., Gutzmer, J., Peuker, U.A., 2016. MLA-based partition curves
for magnetic separation. Miner. Eng. 94, 94–103. https://doi.org/10.1016/j.
mineng.2016.05.015.
Mariano, R.A., Evans, C.L., 2015. Error analysis in ore particle composition distribution
measurements. Miner. Eng. 82, 36–44. https://doi.org/10.1016/j.
mineng.2015.06.001.
Milu, V., Leroy, J.L., Piantone, P., 2003. The Bolcana Cu-Au ore deposit (Metalifero
Mountains, Romania): rst data on the alteration and related mineralisation.
Comptes Rendus - Geosci. 335, 671–680. https://doi.org/10.1016/S1631-0713(03)
00120-2.
Osbahr, I., Krause, J., Bachmann, K., Gutzmer, J., 2015. Efcient and Accurate
Identication of Platinum-Group Minerals by a Combination of Mineral Liberation
and Electron Probe Microanalysis with a New Approach to the Ofine Overlap
Correction of Platinum-Group Element Concentrations. Microsc. Microanal. 21,
1080–1095. https://doi.org/10.1017/S1431927615000719.
Parian, M., Lamberg, P., M¨
ockel, R., Rosenkranz, J., 2015. Analysis of mineral grades for
geometallurgy: Combined element-to-mineral conversion and quantitative X-ray
diffraction. Miner. Eng. 82, 25–35. https://doi.org/10.1016/J.
MINENG.2015.04.023.
Reed, S.J.B., 2005. Electron Microprobe Analysis and Scanning Electron Microscopy in
Geology, Cambridge University Press. https://doi.org/https://doi.org/10.1017/
CBO9780511610561.
Schach, E., Buchmann, M., Tolosana-Delgado, R., Leißner, T., Kern, M., van den
Boogaart, G., Rudolph, M., Peuker, U.A., 2019. Multidimensional characterization of
separation processes – Part 1: Introducing kernel methods and entropy in the context
of mineral processing using SEM-based image analysis. Miner. Eng. 137, 78–86.
https://doi.org/10.1016/j.mineng.2019.03.026.
Schulz, B., Sandmann, D., Gilbricht, S., 2020. Sem-based automated mineralogy and its
application in geo-and material sciences. Minerals 10, 1–26. https://doi.org/
10.3390/min10111004.
Sillitoe, R.H., 2010. Porphyry copper systems. Econ. Geol. 105, 3–41. https://doi.org/
10.2113/gsecongeo.105.1.3.
Simon, G., Kesler, S.E., Essene, E.J., Chryssoulis, S.L., 2000. Gold in porphyry copper
deposits: Experimental determination of the distribution of gold in the Cu-Fe-S
system at 400 degrees to 700 degrees C. Econ. Geol. 95, 259–270.
Van der Plas, L., Tobi, A., 1965. A chart for judging the reliability of point counting
results. Am. J. Sci. 263, 87–90.
Warlo, M., Wanhainen, C., Bark, G., Butcher, A.R., McElroy, I., Brising, D., Rollinson, G.
K., 2019. Automated Quantitative Mineralogy Optimized for Simultaneous Detection
of (Precious/Critical) Rare Metals and Base Metals in A Production-Focused
Environment. Minerals 9, 440. https://doi.org/10.3390/min9070440.
Whitney, D.L., Evans, B.W., 2010. Abbreviations for names of rock-forming minerals.
Am. Mineral. 95, 185–187. https://doi.org/10.2138/am.2010.3371.
Zhou, J., Mermillod-Blondin, R., Cousin, P., 2009. Deportment of Gold and Silver in the
Pinos Altos Composite and Leach Residues: Implications for Recovery Improvement.
World Gold Conf. 2009, South. African Inst. Min. Metall. 75–83.
R. Blannin et al.
