Domaining in Mineral Resource Estimation: A Stock-Take of 2019 Com-

mon Practice

René Sterk1, Koos de Jong1, Greg Partington2, Sebastian Kerkvliet1, Mathijs van de Ven1
1
RSC Consulting Ltd, Level 2, 109 Princes Street, Dunedin 9016, New Zealand
2
Kenex, PO Box 775, Dongara, Western Australia, 6525

ABSTRACT

Keywords: Resource blocks estimated within a particular domain should only be informed by sample
domaining practice points from within that domain. If this fundamental principle of mineral resource
resource estimation estimation is not adhered to it may severely compromise the quality of the final resource
estimate. Nonetheless, the repeated reminder of resource downgrades or even complete
modelling techniques
project devaluations as the consequence of poor domaining practice suggests that this
implicit modelling principle is still not well-entrenched in industry practice. As many practitioners have
grade-shell warned and as documented in books, course materials and online blogs over the years:
boundary analysis geology is important.

It is good practice to base domains primarily on geological information derived from

geological logging of drill core or chips, underpinned by an understanding of the
structural geometry of the ore body and a well-understood genetic model of
mineralisation. However, a sound statistical analysis of geochemical data that are accurate
and precise is also required to evaluate the domains. Multi-element geochemistry from
laboratory or portable XRF instruments combined with multivariate data analysis and
machine-learning (ML), as well as core scanners and down-hole optical televiewers, and
other recent technological advances are powerful tools that enable the proper delineation
of domains. However, these tools are often underutilised as they require additional
investment at a time in a project when their value can be hard to demonstrate.

Unfortunately, the use of geological information to create domains is the exception rather
than the rule. Our review of mineral resource estimates published since 2017 suggests
that more than half of all estimation-constraining wireframes are built using grade cut-
offs and are not informed by any primary geological information such as lithology or
alteration. While using a grade cut-off may seem perfectly logical to delineate domains
when detailed geological information is not available, if not treated with caution, this can
lead to poor domain integrity.

Books and courses on resource estimation clearly express the importance of domaining
but offer few practical solutions or rules of thumb. There appears to be a lack of clear
standards, a lack of a framework to distinguish good from bad domaining practice and
there is a perpetuation of bad practice masked as common practice. Here we offer some
recommendations to raise domaining standards across the industry and present a rules-
based approach to improve domaining practices at the individual level.

INTRODUCTION resource estimates (MRE). Popular authors in this

space such as Coombes (2008), Parker (2004) and
For several decades, the mining and exploration Stephenson & Vann (1999) have memorable one-
industry has been warned about the detrimental liners on offer in their books to make it clear that we
effects of getting the geology wrong in mineral cannot afford to get this wrong. Others have devoted

*
[email protected], +64 211 78 55 00

Page | 1
their professional careers to the process of In this paper we present a review of current
domaining and take every opportunity to point out domaining practice by analysing the last three years
bad practice on social media or other platforms (e.g. of publicly-available MRE reports and identify
Reid & Cowan, 2019). It seems that as an industry trends in the definition of domains in MRE. Are new
we do not need further warning. However, there are techniques being adopted? How were resource
still peer-reviewed papers published where the main, domains constrained and what data were used to
and arguably foregone, conclusion is that “…better create the three-dimensional (3-D) wireframes?
geological input leads to better and more realistic What validation processes were followed? We
models” (Gutierrez & Ortiz, 2019, p. 6). present an insight in current practices and suggest
what future domaining workflows may look like.
Over the last five years, ‘big data’ has changed the We also propose an improved workflow to generate
way we approach data acquisition from exploration grade-shell boundaries.
programmes. With the availability of core scanners,
portable XRF instruments mobile analytical
facilities such as Lab-at-Rig™ (Hillis et al., 2014)
and integrated sample photography solutions such as WHY IS DOMAINING IMPORTANT?
Imago™, to name a few, we now have more and
The basic principles of the estimation techniques
better information than ever to constrain our
used to estimate the average grades of deposits rely
domains using geological data. The introduction of
on the rules of statistics. If we agree to use statistics
‘implicit modelling’ (Cowan et al., 2003) has
to predict grades at unknown points (often practi-
probably been the most significant recent innovation
cally represented as discretisation points that repre-
in the geological modelling and domaining space
sent ‘blocks’), based on the data surrounding it, then
and provides a fast alternative to traditional, more
we must conclude that a model based on inference
labour-intensive, ‘explicit modelling’. All major
cannot predict an event that is not represented in the
software packages now provide implicit modelling
data itself. This leads to the requirement of station-
functionality in some shape or form. The potential
arity within the domains in which we estimate,
of Machine Learning (ML) in this context is also
which is the most important underlying condition of
increasingly recognised (e.g. Oliver & Wilingham,
any domains we build. As Dunham (2017) puts it:
2016, Gazley et al., 2019, Webb et al., 2019,
“we need to ensure the data in our domains are ap-
Whaanga et al., 2019) and the integration of this
ples and apples, not apples and oranges while at the
technique into the space of domaining is the topic of
same time allowing for the possibility that those ap-
numerous recent peer-reviewed papers and
ples may actually be plums”. Glacken & Snowden
presentations (e.g. Green et al., 2019; McManus et
(2001, p. 190) state that “a geological domain repre-
al., 2019, Wedge et al., 2019).
sents an area or volume within which the character-
The combination of big data, machine learning and istics of the mineralisation are more similar than out-
implicit modelling techniques should be the catalyst side the domain”. Therefore, the practice of domain-
ing can be summarised as the generation of 3-D
for a major step change in domaining practices. The
days of creating domains purely based on grade due shapes (often called wireframes, constraints or do-
to a lack of any other data should be over. However, mains) in which we assume that the controls on min-
the mineral resource estimation discipline is eralisation are monogenetic and within which the
sometimes labelled the most conservative in the data have no trends and are comprised of a mono-
modal distribution.
general mining industry (Gleeson, 2015). For
example, 98% of all 161 initial resource estimations
Failure to generate domains that adhere to the prin-
publicised since January 2017 were estimated using
ciple of stationarity may lead to significant bias in
a mix of inverse distance (ID) and ordinary kriging
block grades and hence to flawed estimates. Some-
(OK) (opaxe, 2019), both 60-year old linear
times this is inevitable, such as in early-stage explo-
estimation techniques. Change in practice takes time
ration projects where there is insufficient data about
and is not aided by the perpetual display in the public
geological- and grade-continuity. In these circum-
arena of poor or long-superseded domaining
stances, it is up to the competent person(s) reporting
practices.
and classifying these estimates to be transparent and
to carefully identify the inherent risks. However,
even in advanced exploration projects where data

Page | 2
Figure 1 Statistics of the 161 MREs considered in this report showing all reports classed by: a) mineral deposit class; b) reporting
code; and c) the type of company issuing the report.

are in abundance, there are too many examples of reviewed 161 public reports of mineral resource es-
how a failure to create sound domains has led to sub- timations announced to global markets since Janu-
stantial downgrades or project failures (e.g. Rubi- ary 2017, available through www.opaxe.com. Only
con’s Phoenix (Rubicon Minerals, 2016); Midway initial mineral resources (sometimes called ‘maiden
Gold’s Pan (Midway Gold, 2015); Torex Gold’s El resources’) were reviewed, as resource upgrades or
Limón Guajes (Torex Gold, 2016); Golden Queen’s updates typically build on existing domains and in-
Soledad (Golden Queen, 2015); Goldcorp’s formation from these reports is biased towards old
Cochenour (Koven, 2016); and Albidon’s Munali techniques.
(Albidon Minerals, 2013)).
The 161 MREs were classified into 37 different min-
eral system classes using a customised classification
based on the Geoscience Australia system (Geosci-
REVIEW OF LAST THREE YEARS OF DO- ence Australia, 2019 and references therein), with
MAINING PRACTICE
the top-three most-reported classes being lode-gold
To better understand the uptake of new techniques (24%), pegmatite lithium (11%), and epithermal
and to get a view of current domaining practices, we gold-silver (7%) (Figure 1a). A total of 81 reports
(50%) cited compliance with NI43-101 and 78
(48%) with the JORC Code (2012) (Figure 1b). A

Figure 2 Statistics of the MRE reports showing a) the type of data used for constraining domains; b) a qualitative assessment of the
degree to which geological information used in the estimates that used a grade-shell; and c) the type of information used to con-
strain the cut-off if a grade-shell was used for domaining. In c), ‘nominal’ captures all instances where cut-off grades were selected
without statistical justification.

Page | 3
Figure 3 Statistics for the MRE reports of deposits classified as lode-Au (N=39), showing: a) the type of data used for constraining
the domains; b) how cut-offs were determined for the models based on grade-shells; and c) the types of domain boundaries for the
domains determined based on geology and grade-shells.

total of 40% of MREs were estimated by Compe- no information was provided on the geological rele-
tent/Qualified Persons working for small consultan- vance of the resulting wireframes (i.e. no comment
cies, 34% by Competent Persons working for large made on whether mineralisation was mapped
consultancies, and 9% by one-man consulting com- properly and if the resource geometry reflected the
panies (Figure 1c). An additional 14% of MREs geological interpretation) (Figure 2b). In 32% of
were carried out internally without involvement of MRE reports, some form of statistical analysis was
external consultants, whereas 3% of MREs were re- discussed to justify the cut-off grade (Figure 2c).
ported internally with the estimation work done by However, the majority of these discussions centred
consultants (Figure 1c). around the (subjective) appraisal of an inflection
point on a cumulative probability grade graph and
Eighteen of the 161 reports did not document how lacked statistical justification. For nearly half of the
domains were constrained. The lack of such infor- estimates using grade-shells (49%), no statistical
mation may constitute a breach of the various report- analysis was carried out to justify the cut-off grade
ing codes that require this information to be in- and grades were picked either nominally or by refer-
cluded. Important information on geology and con- ring to “common practice”, or “analogous settings”
trols on mineralisation, deposit style are mandatory (Figure 2c). For 20% of MRE reports it was unclear
headers that are addressed under specific chapter how the cut-off grade was determined. In 55% of all
numbers in NI43-101 and were therefore easy to ex- MRE reports, no comment was made on the nature
tract. By contrast, for MREs reported in accordance of the contacts or boundaries (i.e. if they were ‘hard’
with JORC, this information was comparatively or ‘soft’). For the 100 MREs that were domained us-
more difficult to find or insufficiently documented. ing grade-shells, 69% were modelled by explicit
Based on the 161 reports reviewed here, such short- techniques, while 31% were modelled using implicit
comings in MRE reports are largely limited to the techniques.
estimates carried out by small consultancies and in-
ternal resource staff; larger consultancies typically The ability to rely on geological information to sup-
report this information sufficiently. port domains may depend on deposit style. Building
good domains in structurally complex gold deposits
In 108 of the 161 MREs (67%), domains were con- with various overprinting mineralisation events
strained using grade-shells, based primarily or ex- hosted in otherwise undifferentiated rocks may be
clusively on a grade cut-off (Figure 2a). The cut-off more difficult than for instance in pegmatite-hosted
grade was either applied to a single element (85%), deposits.
or to a metal equivalent of more than one element
(15% of reports, equivalents calculated based on In a breakdown of domaining practices by style
metal prices at the time of reporting). For 38% of the class, of the 39 projects classified as lode-Au depos-
MREs that constrained domains using grade-shells, its, which includes a mix of alteration-, lithology-,

Page | 4
Figure 4 Statistics for the MRE reports of deposits classified as pegmatite (N=17), showing: a) the type of data used for constraining
the domains; b) how cut-offs were determined for the models based on grade-shells; and c) the types of domain boundaries for the
domains determined based on geology and grade-shells (if known).

vein-, and shear-hosted deposit types, 10 were do- more obvious geological contrast present in pegma-
mained based on geology and 25 were domained us- tite-hosted deposits, grade-based domains are as
ing grade-shells (Figure 3a). In the remaining four common as they are in the more complex lode-gold
reports, domaining was not discussed or domains deposits.
were not used. Of the 25 MREs domained using
grade-shells, cut-off grades ranged from 0.05 g/t Au Of the 161 initial resource estimates, 67% were
to 1.0 g/t Au. Five reports did not specify what cut- modelled with explicit techniques and 33% with im-
off grade was used. The majority of the estimates plicit techniques. Additionally, 98% of MREs were
(60%) were based on cut-off grades for the grade- estimated using a mix of inverse distance (20%) and
shells higher than 0.25 g/t Au. Nine reports sup- ordinary kriging (78%) while 2% were estimated by
ported the chosen cut-off grade with a statistical Multi-Indicator Kriging. Machine learning was not
analysis (Figure 3b), generally an appraisal of the discussed in any of the reports to have been used as
cumulative-probability grade plot. Of the 35 reports a tool to generate or support domains. Only one re-
that were domained using either geology or grade- port specifically referenced multi-element geochem-
shells, 18 defined the ore-waste contacts as hard, istry and down-hole geophysical data as important
two as a combination of soft and hard, and 15 did proxies to support geology-driven domains. Gener-
not report the nature of the contact (Figure 3c). ally, where grade-shells were used to constrain do-
mains, this was done so without reference to- or val-
Of the 17 projects classified as pegmatite deposits, idation against geological observations or interpre-
five were domained using geology and nine were tations. For most reports, the geological justification
domained using grade-shells (Figure 4a). Three re- of the domain-building process is limited to state-
ports provided insufficient information to under- ments such as: “domains were generated using
stand what domaining approach had been applied. grade-shells, whilst still honouring the geology”.
Of the nine MREs domained using grade-shells, cut-
off grades ranged from 0.2% to 0.5% Li2O. One re-
port did not specify the cut-off grade. Two reports
DISCUSSION: BEST PRACTICE VS COM-
justified the chosen cut-off by a statistical analysis,
MON PRACTICE
generally an appraisal of the cumulative probability
grade plot (Figure 4b). Four MRE reports deter- The review of the last three years of initial MRE re-
mined a hard contact for the domain, while nine ports suggests that despite recent technological ad-
MRE reports did not describe the nature of the con- vancements and warnings about the importance of
tact and one reported a soft/gradational boundary high-quality domaining, there is little unity in the ap-
(Figure 4c). The data suggest that regardless of the proach to domain definition. In principle, one might

Page | 5
argue, this seems perfectly reasonable, as no two de- practitioners can adopt in their daily domaining
posits are the same. Moreover, the public reporting workflows.
codes we adopt put considerable faith and trust in
the competence of the Competent Person and allow High-Level Recommendations and Suggestions
the estimator the freedom to apply their experience
to the process of MRE. However, the clear lack of 1) The AusIMM and kindred groups that have “to
geology as a first-order driver for the domaining set and maintain high professional standards for
process, the lack of statistical validations for cut-off their members” as one of their key purposes
grades, the lack of contact or boundary analysis, the should take ownership and generate an industry
widespread use of unsubstantiated and often too- best practice standard, similar to the AusIMM’s
high grade-shell cut-offs, as well as the repeated re- Monograph 30 on Mineral Resource Estima-
minders of high-profile resource downgrades due to tion, with specific focus on domaining prac-
poor domaining practice, suggest that warnings tices.
about the importance of geology have not yet lost
2) Improve accountability for CPs carrying out
their purpose.
sub-standard work.
At the root of the issue is a lack of best-practice ex- 3) Highlight, and make more accessible, examples
amples, case studies and other educational material. of good practice as well as bad practice; for ex-
The authors are not aware of any books on domain- ample, information services such as
ing, and popular consultant-driven MRE courses www.opaxe.com can play a role in providing
merely state that ‘good domaining is important’ practitioners with a centralised repository of
without covering the practice in any detail. There is technical information.
simply no standard; and thus there is no clear dis-
4) Software developers should take ownership and
tinction to incoming practitioners between best prac-
create educational material on how to best use
tice and common practice. How practitioners apply
their software; it is in their own interest for their
domaining is to a large degree determined by who
software to be used consistent with best prac-
they learn from, which can lead to the perpetuation
tice.
of bad practice. Over time, due to a lack of good ed-
ucation and an absence of regulation or control, bad 5) Course material provided by consultants and
practice turns into common practice. In addition, other 3rd parties should have a stronger focus
with the onset of user-friendly implicit modelling on domaining best practice, rather than just the
software with intuitive user interfaces that make it nuts and bolts of geostatistics.
easier to create blobs from drill hole data, there is
6) Address the fundamental skill issue at deeper
now a further deepening of the chasm (e.g. Reid &
levels and create educational materials at entry-
Cowan, 2019). At least with the explicit-modelling
level for students and young professionals to
approach, someone had to show a new practitioner
learn about modelling, stationarity and includ-
the functionality of the (hundreds of) different but-
ing structural controls on mineralisation into
tons and complex workflows to create the desired
models. Geological mapping skills are funda-
outcome and there was some knowledge transfer in
mental to building realistic resource models and
the process. As an industry, we seem obsessed with
should receive stronger emphasis in university
our competent persons’ freedom to make just about
programmes.
any technical decision as they see fit, with no system
in place to call them to account.
Rules for Good Domaining Practice

Rule #1: Use geological information as primary in-

IMPROVEMENTS TO DOMAINING PRAC- put for the domains, complemented by proxies for
TICE geology (such as geophysical, geochemical, petro-
To improve the standard of domaining, we provide physical and photographic data)
a number of high-level recommendations and sug- Only 31 of the 161 MREs published since 2017 used
gestions, directed to the industry, as well as some geological parameters as the primary source for its
easily implementable rules of good practice that domaining, and only one study had included proxies
for geology. Paragenesis, plunge orientations, small-

Page | 6
Figure 5 Example of an unrealistic ‘blobby’ model lacking any structural control. From: Reid, 2017

scale as well as deposit-scale structural architecture mapping of geological units will often allow inter-
are critical ingredients to any resource estimation pretation of primary structural controls and con-
process. Invest in getting specialist advice on your strain the geometries of mineralisation. Addition-
deposit. Even deposits drilled via reverse circulation ally, automatic core scanning facilities can now pro-
are no excuse for lack of structural data. Down-hole duce images where pixels can be classified to a 1
optical techniques have been available for several mm resolution. These classifications can be im-
years and allow the same structural interpretations ported straight into modelling software where some
as from diamond core for only a few dollars per me- (such as Seequent’s Leapfrog™) allow immediate
tre extra (Nielsen, 2017). implicit modelling of each classification. Depending
on the style of mineralisation this could also allow
Proxies for geological information are a signifi- identification and implicit modelling of alteration
cantly underutilised resource that can provide im- minerals and zones, quickly and cost-efficiently.
portant information on structure, mineralisation ge-
ometry and other relationships (e.g. King, 2019; Rule #2: Appreciate the difference between geolog-
Browning et al., 2019; Reid et al., 2019; Webb et al., ical domains and estimation domains.
2019). For instance, multi-element geochemistry
Regardless of Rule #1, it is important to note that
collected using portable XRF instruments provides
estimation domains are not the same as geological
elemental concentrations of typically around 25 ele-
domains. Domains based on primary geological data
ments. In combination with data reduction tech-
(e.g. “Lith1”) often lack the resolution to separate
niques such as principal component analysis, it is of-
different phases of mineralisation, leading to multi-
ten possible to identify geochemical fingerprints of
modal distributions or trends in the numerical data
units with different geological characteristics (e.g.
within a single geological domain. By contrast,
Gazley et al, 2015; Caciagli, 2015; Brauhart et al.
models based exclusively on numerical data (e.g.
2016; Ordóñez-Calderón & Gelcich, 2018). Many of
grade-shells) often do not adequately represent real-
these units are often difficult to consistently log vis-
ity either and in the hands of the inexperienced op-
ually by geologists, who will inevitably introduce
erator, often turn a deposit where mineralisation is
subjective bias in the logging, regardless of the qual-
controlled by various different structural trends into
ity assurance procedures in place. Following this,

Page | 7
a messy mix of ‘blobs’ (Figure 5). This is far from a within each domain, and then readjusting the do-
representative model of the true geology and such mains, is critical to getting meaningful and repre-
models should never underpin any subsequent feasi- sentative estimates of mineral deposits.
bility studies and are the result of widespread igno-
rance of how structural geological architecture plays Rule #4: Check for statistical stationarity
a key role in the geometry of mineralisation (Cowan, At the end of each iteration (Rule #3) there should
2017; Reid & Cowan, 2019). always be a test of statistical stationarity. The issue
of stationarity was only addressed in 2 of 161 re-
Rule #3: Regard the process of domaining as an it-
ports. Although this may be a reporting issue (it was
erative process
carried out but not reported), since statistical station-
Following Rule #2, geological and numerical data arity is critical to the integrity of the entire MRE pro-
need to be reviewed together in order to end up with cess, it should be a standard procedure to include in
robust estimation domains. The domain building any technical report. We recommend the simple
process is typically iterative and geological and nu- grid-based variance review approach (Coombes,
merical data are both used to build, validate and ad- 2016), or similar, as a minimum requirement.
just any domains. Geology should always be the
main control and the final model should represent as Rule #5: Treat grade-based domains appropriately
best as possible ‘the rocks as they occur’. Grade-based domains may be perfectly acceptable if
they have been created honouring the geology, their
Sound, iterative, workflows that integrate estimation
geometries have been verified against the structural
domains with geological domains in sensible ways
data, and the statistics within the domains have been
are part of the modern resource geologist’s toolkit.
assessed for stationarity. However, there are too
The process of building domains, analysing the data
many estimates published where the domains have
been established by economic parameters, or where

Figure 6 Phinar X-10 allows for testing of different cut-off grades by comparing the average grades of the first metre inside and the
first metre outside the grade-shell. a) the maximum contrast for the waste-ore top boundary is found at a cut-off of 0.22 g/t Au; b)
the maximum contrast for the ore-waste bottom boundary is found at a cut-off of 0.13 g/t Au.

Page | 8
a lack of understanding of implicit modelling tech- tively, this process is the same as looking at the max-
niques has resulted in short grade intervals being imum inflection point in a cumulative probability
modelled into a surreal cluster of blobs (Figure 6, plot, but it has the important advantage that it is ge-
c.f. Reid & Cowan, 2019). The use of economic ospatially relevant and is not just based on a total
grade cut-offs or combined multi-element grade cut- average distribution curve. To the authors’
offs based on element conversions (e.g. using a ‘Au- knowledge, the only software currently providing
equivalent’ cut-off grade for an Au-Ag-Mo deposit) this capability is Phinar X-10. In Pinar X-10, the
should never be used. The distribution of the ele- user sets a maximum internal dilution and maximum
ment of interest within the stationary domain is consecutive dilution and the software creates unmin-
simply not dependent on dollars or the market. Re- eralised and mineralised intervals for a range of cut-
member that if we agree to use statistics to predict off grades, specified by the user. The software also
grades at unknown points, based on the data sur- allows grade cap settings. It then calculates the av-
rounding it, then we must conclude that a model erage contrast between unmineralized and mineral-
based on inference cannot predict an event that is not ised zones for each grade cut-off and visualises this
represented in the data itself, or with data that is cur- in two graphs (Figure 6). The user can then select
tailed by artificial grade cut-offs. the maximum contrast and generate a contact analy-
sis for that grade cut-off to check the nature of the
Take gold as an example: the average grade-shell contact (e.g. hard/abrupt or soft boundaries) (Figure
cut-off used by practitioners in the last three years 7).
was 0.5 g/t Au. By effectively excluding material
between 0.1 and 0.5 g/t Au from the MRE, the esti- Although the grade cut-off may not have a large im-
mation is going to result in fewer tonnes at a higher pact on the grade-tonnage curve or the quality of the
grade. The practice of setting high grade cut-offs for estimate, this must always be verified and the sensi-
resource grade-shells may be premeditatively, yet tivity of the model to different grade cut-offs should
often unknowingly, used by practitioners to offset be reported in the public report. It is clear from this
the well-known volume-variance effect (the obser- work that this rarely happens.
vation that when mining the resource, lower ton-
nages and higher grades are achieved when the cut-
off grade is lower than the resource mean grade).
However, more realistic grade-tonnage curves are
achieved by using far lower grade-shell cut-offs that
appropriately mark the boundary of a mineralised
domain and isolate a statistical population. The vol-
ume-variance effect can be dealt with afterwards
through a process of variance correction (Rossi &
Deutsch, 2013).

Grade cut-offs should be based on statistical evalua-

tion of the difference between background concen-
tration and enrichment and should be underpinned
by geological evidence. A statistical evaluation
could be a visual evaluation of an inflection point on
a cumulative probability plot. However, a much
more effective workflow is to review the maximum
difference (expressed by the coefficient of variation)
between the average grade of all first metres outside
a zone of mineralisation and all the first metres in-
side a zone of mineralisation for a range of different
cut-offs. The grade cut-off where this difference is
largest presents the maximum geological contrast
between waste and mineralisation and represents a
geologically justifiable domain boundary. Effec- Figure 7 Domain boundary analysis carried out in a) Phinar
X-10 and b) Leapfrog Edge.

Page | 9
Figure 8 Narrow vein modelling workflow in Leapfrog Geo (Reid 2019, Pers. Comm.).

Rule #6: Carry out a boundary analysis. implicit modelling a powerful tool to resolve do-
maining and other sensitivity issues. Through ad-
Boundary analysis is an important tool to investigate
vances in computer power and implicit modelling al-
whether ore-waste contacts are abrupt ‘hard’ or gra-
gorithms, this process can now be executed far more
dational ‘soft’ boundaries. This analysis not only
efficiently than it could 15 years ago, but an under-
provides important input for the estimation process
standing of the software is important to avoid poor-
but can also reveal important trends around the con-
quality outcomes.
tact that could invalidate the supposition of station-
arity in the domain. For those working in a hard-boundary and narrow-
vein environment who are still on the fence, See-
Phinar X-10 software automatically creates contact
quent’s Leapfrog has excellent workflows that han-
plots once a grade cut-off with maximum contrast
dle such deposits and honour the geology and create
has been selected. Seequent’s Leapfrog software au-
models far superior to those generated via explicit
tomatically creates contact analysis plots once
methods (Figure 8).
grade-shells have been generated and an estimation
workflow has been added. The Leapfrog boundary Rule #8: Report what you have done.
analysis has the advantage that it is calculated per-
pendicular to the actual domain boundary. Our analysis suggests that important aspects of the
MRE process are commonly not sufficiently docu-
Rule #7: Use implicit techniques where possible mented. Common issues are:
but use them wisely.
1) a description of the controls on mineralisation is
As an industry we need to come to terms with the missing, or the geological model is absent, or
fact that the benefits of implicit techniques outweigh only available in minimal format. This issue is
their disadvantages (Gleeson, 2015). However, note more common in documents that state adher-
that the boundary between implicit and explicit tech- ence to the JORC Code (2012);
niques has become fuzzy (Kentwell, 2019) and that
2) the extent to which geological information was
many of the modern implicit workflows are heavily
used in the generation of estimation domains is
dependent on many explicit user settings. Most im-
not discussed;
portantly, implicit modelling is superior in that it is
much faster, and facilitates the sensitivity testing of 3) the rationale for the choice of cut-off grades for
different model assumptions. The ability to test dif- domains is not presented;
ferent hypotheses simultaneously by running mod-
4) boundary conditions are not discussed, or plots
els on multiple computer processing cores, makes
are not presented; and

Page | 10
5) no discussion on the sensitivity to different ge- Coombes, J, 2008. The Art and Science of Resource
ological models or domaining approach is pre- Estimation: A Practical Guide for Geologists and
sented. Engineers (Coombes Capability: Perth, Australia).
Since project failures often stem from inadequate Coombes, J, 2016. I'd Like to be OK with MIK, UC?
domaining at the MRE stage, the above information A Critique of Mineral Resource Estimation Tech-
should always be transparently reported so that in- niques, 261 p (Coombes Capability: Perth, Aus-
vestors can appropriately ascertain the risk in the tralia).
project.
Cowan, E, 2017. The fundamental reason why your
geological models may be completely wrong [blog-
post], Available from: <https://www.ore-
CONCLUSIONS find.com/blog/orefind_blog/2017/10/23/the-funda-
mental-reason-why-your-geological-models-may-
The clear lack of geological input in the domaining be-completely-wrong> [Accessed: 11-11-2019].
process for MREs published since 2017 suggests
that regardless of the doctrine that geology is im- Cowan, E, Beatson, R, Ross, H, Fright, W, McLen-
portant, the mining and exploration industry is not nan, T, Evans, T, Carr, J, Lane, R, Bright, D, Gill-
doing a very good job of adhering to its basic prin- man, A, Oshust, P and Titley, M, 2003. Practical Im-
ciples. Moreover, while having abandoned geology plicit Geological Modelling, Proceedings 5th Inter-
as a primary control on domains in the estimation national Mining Geology Conference 2003, pp 89-
process, it is clear that our collective approach to us- 99 (Australian Institute of Mining and Metallurgy:
ing grade-shells is flawed as well. At the root of this Carlton, Victoria).
issue is a lack of good education and an absence of Dunham, S, 2017. The Truth About Estimation #6 –
clear standards, a lack of framework in which good The Domain Dilemma. [blogpost], Available from:
and bad domaining practices can be identified, and <https://sd2.com.au/2017/11/28/the-truth-about-es-
a perpetuation of bad practice masked as common timation-6-the-domain-dilemma/> [Accessed: 08-
practice. 11-2019].
Gazley, M F, Collins, K S, Robertson, J, Hines, B R,
Fisher, L A and McFarlane, A, 2015. Application of
REFERENCES principal component analysis and cluster analysis to
mineral exploration and mine geology, Proceedings
Albidon Minerals, 2012. Material Resource Down-
AusIMM New Zealand Branch Annual Conference
grade [press release], Available from:
2015, pp 131-139 (Australian Institute of Mining
<https://www.asx.com.au/asxpdf/20121108/pdf/42
and Metallurgy: Dunedin, New Zealand).
b25mlj3c41md.pdf> [Accessed: 11-11-2019].
Gazley, M F, Hood, S H and Sterk, R, 2019. Ma-
Brauhart, C, Grunsky, E and Hageman, S, 2016.
chine Learning and Resource Geology, presented to
Magmato-hydrothermal space: A new metric for ge-
Mining Geology Conference, 25-26 November
ochemical characterisation of metallic ore deposits,
2019, Perth, Australia.
Ore Geology Reviews, 86(2017):867-895.
Geoscience Australia, 2019. Mineral Deposits Min-
Browning, F, Clark, H, Ulrich, S and Clark, F, 2019.
eral Systems and Mineralising Events [online],
Developing the Geological Model for Sunrise Dam
Available from: <https://www.ga.gov.au/scientific-
Gold Mine – Insights from Exploration to Mining
topics/minerals/mineral-exploration/deposits-
presented to Mining Geology Conference, 25-26
events> [Accessed: 12-11-2019].
November 2019, Perth, Australia.
Glacken, I and Snowden, D, 2001. Mineral Resource
Caciagli, N, 2015. Multielement Geochemical Mod-
Estimation, in Mineral Resource and Ore Reserve
elling for Mine Planning: Case Study from an Epi-
Estimation - The AusIMM Guide to Good Practice.
thermal Gold Deposit, Proceedings International
(ed: A Edwards), pp 189-198 (The Australasian In-
Workshop on Compositional Data Analysis 2015, pp
stitute of Mining and Metallurgy: Melbourne, Aus-
45-61 (Springer: Cham, Switzerland).
tralia).

Page | 11
Gleeson, P, 2015. "Is Implicit Modelling Reliable McManus, S, Rahman, A, Horta, A and Coombes, J,
Enough For Use in Resource Domaining?" [blog- 2019. Applied Bayesian Modeling for Assessment
post], Available from: <https://www.see- of Interpretation Uncertainty in Spatial Domains,
quent.com/is-implicit-modelling-reliable-enough- presented to Applied Statistics and Policy Analysis
for-use-in-resource-domaining/> [Accessed: 11-11- Conference, 5-6 September, 2019, Wagga Wagga,
2019]. Australia.
Golden Queen, 2015. Golden Queen Mining Com- Midway Gold, 2015. Midway Gold Updates Pan
pany Ltd. Provides Update on Resource Estimates, Modelling [press release], Available from:
Reserve Estimates and Construction Progress at <http://www.midwaygold.com/_re-
Soledad Mountain [press release], Available from: sources/news/2015.05.11.pdf> [Accessed: 11-11-
<https://www.goldenqueen.com/assets/docs/2015- 2019].
02-10_NR.pdf> [Accessed: 11-11-2019].
Nielsen, S H H and Franey, D, 2017. Downhole
Green, C, Sanderson, H, Carter, D and Sullivan, S, Logging in 3D Geology and Mineral Potential Mod-
2019. Deep Learning - A New Paradigm for Domain elling, presented to AusIMM NZ Branch Confer-
Modelling presented to Mining Geology Confer- ence, 11-13 September 2017, Christchurch, New
ence, 25-26 November 2019, Perth, Australia. Zealand.
Gutierrez, R and Ortiz, J, 2019. Sequential Indicator Oliver, S and Willingham, D, 2016. Maximise Ore-
Simulation with Locally Varying Anisotropy–Sim- body Value through the Automation of Resource
ulating Mineralized Units in a Porphyry Copper De- Model Development Using Machine Learning, Pro-
posit, Journal of Mining Engineering and Research, ceedings 3rd International Geometallurgy Confer-
1(1):1-7. ence 2016 2016, pp 295-301 (Australian Institute of
Mining and Metallurgy: Perth, Australia).
Hillis, R R, Giles, D, Van der Wielen, S E, Baensch,
A, Cleverley, J S, Fabris, A, Halley, S W, Harris, B opaxe, 2019. Resource Reports Database [online],
D, Hill, S H, Kanck, P A, Kepic, A, Soe, S P, Stew- Available from: <www.opaxe.com> [Accessed: 11-
art, G and Uvarova, Y, 2014. Coiled Tubing Drilling 09-2019].
and Real-Time Sensing - Enabling Prospecting
Ordóñez-Calderón, J C and Gelcich, S, 2018. Ma-
Drilling in the 21st Century?, Society of Economic
chine learning strategies for classification and pre-
Geologists Special Publication, 18:243-259.
diction of alteration facies: Examples from the Rose-
Joint Ore Reserves Committee of The Australasian mont Cu-Mo-Ag skarn deposit, SE Tucson Arizona,
Institute of Mining and Metallurgy Australian Insti- Journal of Geochemical Exploration, 194:167-188.
tute of Geoscientists and Minerals Council of Aus-
Parker, H M, 2004. Harry M. Parker GAA Honorary
tralia, 2012. Australasian Code for Reporting of Ex-
Life Member [online], Available from:
ploration Results, Mineral Resources and Ore Re-
<http://www.gaa.org.au/pdf/GAA_Hon_Life_Mem
serves. http://www.jorc.org.
ber_harry_parker.pdf> [Accessed: 12-11-2019].
Kentwell, D, 2019. Destroying the distinction be-
Reid, D, Harvey, G and Glasson, M, 2019. Regolith
tween implicit and explicit geological modelling,
domain modelling using multivariate cluster analy-
presented to Mining Geology Conference, 25-26
sis at Mt Thirsty Co-Ni Deposit presented to Mining
November 2019, Perth, Australia.
Geology Conference, 25-26 November 2019, Perth,
King, A R, 2019. Applying Geophysics to Mining Australia.
Geology Problems, presented to Mining Geology
Reid, R, 2017. Implicit Modelling disasters in the
Conference, 25-26 November 2019, Perth, Aus-
making – Part 1 [blogpost], Available from:
tralia.
<https://www.linkedin.com/pulse/implicit-model-
Koven, P, 2016. Goldcorp Inc cuts dividend, lowers ling-disasters-making-part-1-ron-reid/> [Accessed:
production guidance for next three years, Financial 22-11-2019].
Post, 25 February 2016, Available from:
Reid, R and Cowan, E, 2019. Toward Robust and
<https://business.financialpost.com/commodi-
Reliable Implicit Geological Models, presented to
ties/mining/goldcorp-inc-cuts-dividend-posts-big-
Mining Geology Conference, 25-26 November
ger-quarterly-loss-on-us3-9-billion-charge>.
2019, Perth, Australia.

Page | 12
Reid, Z, 2019. Narrow Vein Modelling Workflow [press release], Available from:
Discussion, 18 November 2019. <https://www.torexgold.com/news/torex-an-
nounces-q1-2016-results-and-updated-life-of-mine-
Rossi, M E and Deutsch, C V, 2013. 7.3 Volume
plan> [Accessed: 24-11-2019].
Variance Correction, in Mineral Resource Estima-
tion, pp 122-128 (Springer Science & Business Me- Webb, L, Grujic, M and Truong, D T, 2019. The Ap-
dia: Dordrecht, The Netherlands). plication of Deep Learning to Extract Geological In-
formation from Drill Core, presented to Mining Ge-
Rubicon Minerals, 2016. Underground Activities
ology Conference, 25-26 November 2019, Perth,
Temporarily Suspended at the Phoenix Gold Pro-
Australia.
ject; Rubicon to Enhance Its Geological Model and
Develop an Implementation Plan [press release], Wedge, D, Hartley, O, McMickan, A, Green, T and
Available from: <https://rubiconminerals.com/In- Eun-Jung, H, 2019. Machine learning assisted geo-
vestor-News/News/Press-release-details/2015/Un- logical interpretation of drillhole data: Examples
derground-Activities-Temporarily-Suspended-at- from the Pilbara Region, Western Australia, Ore Ge-
the-Phoenix-Gold-Project-Rubicon-to-Enhance-Its- ology Reviews, 114.
Geological-Model-and-Develop-an-Implementa-
Whaanga, A, Vigor-Brown, W and Nowland, S,
tion-Plan/default.aspx> [Accessed: 11-11-2019].
2019. The implementation of photogrammetry and
Stephenson, P and Vann, J, 1999. Common Sense automated data analysis functions at the Waihi Un-
and Good Communication in Mineral Resource and derground Gold Mine presented to Mining Geology
Ore Reserve Estimation, Proceedings Pacrim 1999, Conference, 25-26 November 2019, Perth, Aus-
pp 435-444 (Australian Institute of Mining and Met- tralia.
allurgy: Melbourne, Victoria).
Torex Gold Resources Inc., 2016. Torex Announces
Q1 2016 Results And Updated Life Of Mine Plan

Page | 13