minerals

Article
Integrating the Theory of Sampling into
Underground Mine Grade Control Strategies
Simon C. Dominy 1,2, * ID , Hylke J. Glass 1 ID , Louisa O’Connor 2 ID
, Chloe K. Lam 1 ID
,
Saranchimeg Purevgerel 3 and Richard C.A. Minnitt 4 ID
1 Minerals Engineering Research Group, Camborne School of Mines, University of Exeter, Penryn,
Cornwall TR10 9FE, UK; [email protected] (H.J.G.); [email protected] (C.K.L.)
2 Department of Mining and Metallurgical Engineering, Western Australian School of Mines,
Curtin University, Bentley, WA 6102, Australia; [email protected]
3 Department of Mineral and Energy Economics, Western Australian School of Mines, Curtin University,
Murray Street, Perth, WA 6000, Australia; [email protected]
4 School of Mining Engineering, University of the Witwatersrand, Johannesburg 2050, South Africa;
[email protected]
* Correspondence: [email protected]




Received: 7 May 2018; Accepted: 25 May 2018; Published: 29 May 2018


Abstract: Grade control in underground mines aims to deliver quality tonnes to the process plant
via the accurate definition of ore and waste. It comprises a decision-making process including data
collection and interpretation; local estimation; development and mining supervision; ore and waste
destination tracking; and stockpile management. The foundation of any grade control programme is
that of high-quality samples collected in a geological context. The requirement for quality samples has
long been recognised, where they should be representative and fit-for-purpose. Once a sampling error
is introduced, it propagates through all subsequent processes contributing to data uncertainty, which
leads to poor decisions and financial loss. Proper application of the Theory of Sampling reduces
errors during sample collection, preparation, and assaying. To achieve quality, sampling techniques
must minimise delimitation, extraction, and preparation errors. Underground sampling methods
include linear (chip and channel), grab (broken rock), and drill-based samples. Grade control staff
should be well-trained and motivated, and operating staff should understand the critical need for
grade control. Sampling must always be undertaken with a strong focus on safety and alternatives
sought if the risk to humans is high. A quality control/quality assurance programme must be
implemented, particularly when samples contribute to a reserve estimate. This paper assesses grade
control sampling with emphasis on underground gold operations and presents recommendations for
optimal practice through the application of the Theory of Sampling.

Keywords: underground mine grade control; Theory of Sampling; sampling errors; representative
sampling; sampling protocol optimisation; quality assurance/quality control

1. Introduction

1.1. Overview
Sampling is a critical component throughout the mine value chain; it includes the sampling of
both in-situ and broken material for geological (resource and grade control), geoenvironmental, and
geometallurgical purposes [1–7]. Sampling errors generate both monetary and intangible losses [3,8].
The data produced must be fit-for-purpose in order to contribute to Mineral Resources/Ore Reserves
reported in accordance with the 2012 Joint Ore Reserves Committee (JORC) [9] or other codes.
Quality assurance/quality control (QAQC) is critical to maintain data integrity through documented

Minerals 2018, 8, 232; doi:10.3390/min8060232 www.mdpi.com/journal/minerals

Minerals 2018, 8, 232 2 of 45

procedures, sample security, and monitoring of precision, accuracy and contamination. Samples and
their associated assays are key inputs into important decisions throughout the mine value chain.
The Theory of Sampling (TOS) defines sampling errors, where actions may lead to uncertainty
and create a total measurement error [1,10]. Theory of Sampling attempts to break down this error
into a series of contributions along the sampling value chain (e.g., the planning to assay-measurement
process) (Table 1).

Table 1. Sampling value chain.

Planning Collection Transport Laboratory Preparation Assaying

Stage
1 2 3 4 5
Observe
Equipment
Collect Equipment operation
Scope Integrity/security operation
Bag and tag Equipment clean
Activity Develop Chain of Equipment clean
QAQC QAQC
Execute custody QAQC
Integrity/security Integrity/security
Integrity/security
Chain of custody
In-situ nugget effect
FSE, GSE PE
Sampling errors FSE, GSE PE
DE, EE, WE, PE AE
DE, EE, WE
Precision Precision (if splitting)
Dominant effect on results Bias Bias
Bias Bias
FSE: fundamental sampling error; GSE: grouping and segregation error; DE: delimitation error; EE: extraction error;
PE: preparation error; WE: weighting error; AE: analytical error; QAQC: Quality assurance/quality control.

Errors are generally additive throughout the sampling value chain and generate both monetary
and intangible losses. The aim is to collect representative samples to accurately describe the material
in question. Sample collection is followed by reduction in both mass and fragment size to provide
a sub-sample for assay. An assay is the quantitative measurement of the concentration (e.g., mass
fraction such as g/t gold) of a metal by a given methodology, for example a 30 g fire assay followed by
measurement of gold using an instrumental method (e.g., inductively coupled plasma spectroscopy).

1.2. Grade Control Paradigm

Underground mines focus on the maximisation of ore and minimisation of waste recovery.
The traditional role of grade control samples relates to: development ore/waste decisions, investigation
of ore limits and grade deportment within (for example) vein versus wall rocks, identification of grade
trends and/or local continuity along development or in stopes, and local estimation and input into the
short- to medium-term mine plan.
The process of grade control broadly comprises: data collection, integration and interpretation,
supervision of development and mining, local estimation and input into stope design, stockpile and
mill feed management, and leading ultimately to reconciliation. In many cases, grade control samples
constitute just-in-time samples as the assay results should be returned within 24 h or less to allow
ore/waste decisions to be made. Where samples feed into ore/waste decisions, the risks of potential
economic impact must be understood. It may be the case that the results are biased and that time
would be better spent on detailed geological mapping and/or more drilling. In other cases, the
decisions based on grade control samples are not critical. Once a face or stockpile sample is collected,
the operation has already spent money and committed to develop infrastructure and development.

1.3. Sampling for Grade Control

Underground grade control sampling methods have remained the same for many years [11,12],
though sample preparation and assaying techniques have improved greatly. Methods are dominated
by hand-collected chip and channel in-situ samples and grab broken-rock samples. In many mines,
diamond core drilling and/or reverse circulation (RC) drilling has become important in the grade
Minerals 2018, 8, 232 3 of 45

control process. A best practice grade control programme will provide information on grade and its
Minerals 2018, 8, x FOR PEER REVIEW 3 of 43
relationship to geology (Figure 1).
Grade control samples such as chip or channel samples are relatively cheap to collect, and apart
Grade control samples such as chip or channel samples are relatively cheap to collect, and apart
from the effort involved, easy to take. The potentially large numbers of samples used in estimation
from the effort involved, easy to take. The potentially large numbers of samples used in estimation
smooth out individual assay variability. However, the consequence is that the sampling distributions
smooth out individual assay variability. However, the consequence is that the sampling distributions
drawn from these data may show large numbers of low grade assays, positive disjointed tails, and
drawn from these data may show large numbers of low grade assays, positive disjointed tails, and
high skewness. This makes the estimation process more challenging and generally indicates that the
high skewness. This makes the estimation process more challenging and generally indicates that the
samples are not representative.
samples are not representative.

(A) (B)

Figure 1. Face photographs showing sample lines based on geology: (A) A single chip sample line
Figure 1. Face photographs showing sample lines based on geology: (A) A single chip sample line
divided into a central vein sample with associated hanging and footwall samples; (B) Dual face chip
divided into a central vein sample with associated hanging and footwall samples; (B) Dual face chip
sample lines across a sheeted vein zone.
sample lines across a sheeted vein zone.

1.4. Mineralisation Characteristics

1.4. Mineralisation Characteristics
Gold deposits often pose challenges during sampling because of their apparently erratic grade
Gold deposits often pose challenges during sampling because of their apparently erratic
distribution. Consideration should be given to the nature of the ore with respect to in-situ gold
grade distribution. Consideration should be given to the nature of the ore with respect to in-situ
particle sizing and distribution; the partition of gold between sulphide-locked and free categories;
gold particle sizing and distribution; the partition of gold between sulphide-locked and free
geological versus assay cut-offs; and stringers/disseminations that require sampling beyond vein
categories; geological versus assay cut-offs; and stringers/disseminations that require sampling
margins. Coarse gold-bearing mineralisation is generally characterised by a high geological or in-situ
beyond vein margins. Coarse gold-bearing mineralisation is generally characterised by a high
nugget effect and prone to high sampling nugget effect if sampling activities are not optimised [13,14].
geological or in-situ nugget effect and prone to high sampling nugget effect if sampling activities are
Gold particles greater than 100 µm in size are generally considered to be coarse, where very coarse
not optimised [13,14]. Gold particles greater than 100 µm in size are generally considered to be coarse,
gold is greater than 2000 µm in size.
where very coarse gold is greater than 2000 µm in size.
In some gold deposits, there is evidence to show the existence of both fine- and coarse-gold
In some gold deposits, there is evidence to show the existence of both fine- and coarse-gold
particles that may be part of separate paragenetic stages or coeval and overprinted [15]. The fine gold
particles that may be part of separate paragenetic stages or coeval and overprinted [15]. The fine gold
particles are more likely to be disseminated through the ore zone and responsible for a relatively low
particles are more likely to be disseminated through the ore zone and responsible for a relatively low
background grade of between 0.5 g/t Au and 5 g/t Au. The coarse particles will be more dispersed
background grade of between 0.5 g/t Au and 5 g/t Au. The coarse particles will be more dispersed
and/or clustered, and are more likely to be related to higher grade zones or shoots.
and/or clustered, and are more likely to be related to higher grade zones or shoots.
Identification of the background grade and associated gold particle size characteristics is
Identification of the background grade and associated gold particle size characteristics is
important, since they may represent the only part of the mineralisation that can be sampled relatively
important, since they may represent the only part of the mineralisation that can be sampled relatively
easily [15]. For example, if a deposit with a mean grade of 10 g/t Au has a background grade of 2 g/t
easily [15]. For example, if a deposit with a mean grade of 10 g/t Au has a background grade of 2 g/t
Au related to fine gold and the rest of the inventory (8 g/t Au) related to coarse gold, then 70% of the
Au related to fine gold and the rest of the inventory (8 g/t Au) related to coarse gold, then 70% of the
contained gold can potentially be under-sampled.
contained gold can potentially be under-sampled.
There is evidence to show that relationships exist between increased gold grade and larger
There is evidence to show that relationships exist between increased gold grade and larger
potentially clustered gold particles [16]. In some cases, 50% of the gold is greater than 100 µm in size, with
potentially clustered gold particles [16]. In some cases, 50% of the gold is greater than 100 µm in size,
extreme cases of 90% above 100 µm and including 60% about 1000 µm. Coarse particles may be more
clustered and related to high-grade zones. In rare cases, fine gold particles may cluster to form pseudo-
coarse gold masses of up to 500 µm in size. Clustering can have a significant impact on sampling protocols,
particularly during the early stages of sampling broken rock, whether it is bulk sampling of development
rounds, grab sampling of stockpiles, field sampling of RC chippings, or laboratory sampling of crushed
Minerals 2018, 8, 232 4 of 45

with extreme cases of 90% above 100 µm and including 60% about 1000 µm. Coarse particles may
be more clustered and related to high-grade zones. In rare cases, fine gold particles may cluster to
form pseudo-coarse gold masses of up to 500 µm in size. Clustering can have a significant impact on
sampling protocols, particularly during the early stages of sampling broken rock, whether it is bulk
sampling of development rounds, grab sampling of stockpiles, field sampling of RC chippings, or
laboratory sampling of crushed drill core. Once the sample has been pulverised, the additional in-situ
heterogeneity related to clustering is removed, though it may still be relatively high if coarse gold
is present.

1.5. Sampling: Decisions, Financial Loss, and Compliance

Once a sampling error is introduced, it propagates through all subsequent processes contributing
to the uncertainty of the end result and any decisions made thereon [3,4,8,17]. Across the mine value
chain, these errors generate financial and intangible losses [3,4,8]. The financial consequences of
sub-optimal grade control sampling can be significant and relate to: incorrect operational decisions,
poor reserve estimates, reduced mine life, and poor reconciliation.
These financial losses arise due to misunderstanding of the principal factors that affect the size of
sampling errors, such as the mass of the sample, the effects of splitting a sample to reduce the mass,
and the influence of the nominal particle size. The handling and processing of sample material without
due regard to the sampling characteristics of the material being sampled is a major source of sampling
errors (e.g., coarse versus fine gold dominated mineralisation). This significant problem of ore/waste
misclassification is a consequence of the support and information effects. The support effect relates to
the difference in size at which samples are collected and mining blocks are extracted. The information
effect is a function of the availability of data at the time that decisions are made about the destinations
of ore and waste. Ore/waste misclassification will be minimised through samples of an optimal mass
and spacing.
In the Australian context, The JORC Code 2012 requires the Competent Person(s) to provide
commentary on the entire sampling process. The company undertaking sampling is required to
have quality assurance policies, sampling procedures, and protocols in place. There is sometimes a
casual approach to grade control sampling and associated activities, where protocols, QAQC, training,
and periodic review are not undertaken. This will nearly always result in operational performance
problems that require time and money to remedy.
Dominy [4] presents a number of underground case studies where annual financial losses due to
poor sampling range between US $2 M to US $12 M before corrective actions were taken. Key issues
related to incorrect and poorly implemented sampling methods, inappropriate sample preparation
and assay methods, and poor or no QAQC.

1.6. Current Practice and Focus of this Contribution

Many large operations have removed the need for development sampling, relying on the use of
drilling. Small mines (<0.25 Mt per annum) continue to use development samples (e.g., linear, sludge,
and grab) both for resource evaluation and grade control activities. Some will have minimal drilling to
support mine planning. Where development sampling is required, there is a need for best practice,
particularly when it feeds into a resource that will be publically reported.
This paper draws on the practical experience of the authors to provide the reader with the
background required to establish a fit-for-purpose underground grade control sampling programme.
It critically assesses the sampling methods used and their optimal collection, preparation, and
assaying through the application of TOS. A number of examples are presented to validate discussions
where appropriate. Whilst the paper focusses on gold-based operations, the conclusions are applicable
to other metalliferous mines. In addition, its coverage is relevant to surface sampling using the
same methods.
Minerals 2018, 8, 232 5 of 45

2. Theory of Sampling

2.1. TOS Overview

Theory of Sampling was developed in the 1950s by Dr Pierre Gy to deal with sampling challenges
in the mining industry, though it has far wider applications [1,10]. It provides important guidelines
for reducing sampling errors (Table 2), though there is still much to do in the context of research,
communication, and education [18].

Table 2. Definition of Theory of Sampling (TOS) sampling errors.

Sampling Effect on
Acronym Error Type Source of Error Error Definition
Error Sampling
Results from grade heterogeneity of the
broken lot. Of all sampling errors, the FSE
does not cancel out and remains even after
Fundamental FSE Correct Random Characteristics a sampling operation is perfect. Experience
Sampling Error Errors—Precision of the ore type. shows that the total nugget effect can be
(CSE) Generator Relates to artificially high because sample masses are
constitution not optimal.
and
Relates to the error due to the combination
distribution
of grouping and segregation of rock
heterogeneity
Grouping and fragments in the lot. Once rock is broken,
GSE
Segregation there will be segregation of particles at any
scale (e.g., surface stockpile or
laboratory pulp).
Results from an incorrect shape of the
Delimitation DE Sampling
Incorrect volume delimiting a sample.
Systematic equipment and
Sampling Error
Errors—Bias materials Results from the incorrect extraction of a
(ISE)
Generator handling sample. Extraction is only correct when all
Extraction EE
fragments within the delimited volume are
taken into the sample.
Samples should represent a consistent mass
Weighting WE
per unit (e.g., kg/m).
Refers to issues during sample transport
and storage (e.g., mix-up, damage, loss and
alteration), preparation (contamination
Preparation PE and/or losses), and intentional (sabotage
and salting), and unintentional (careless
actions and non-adherence of
protocols) actions.
Relates to all errors during the assay and
analytical process, including issues related
Analytical
Analytical AE - to rock matrix effects, careless actions, and
process
analytical machine maintenance
and calibration.

Its application to underground grade control sampling has been variable, with leading use in the
South African gold sector [19–21]. Some of the resistance to the application of TOS relates to the use of
the Fundamental Sampling Error (FSE) equation and confusion as to its calibration and application.
Theory of Sampling has a wider usage than simply the FSE equation and includes a number of errors
that must be considered to achieve representative samples (Table 2).
Theory of Sampling provides an insight into the causes and magnitude of errors that may occur
during the sampling of particulate materials (e.g., broken rock). Theory of Sampling strictly does not
include sampling of in-situ material such as face samples, as the sample does not have an equiprobable
chance of being collected. However, its application is relevant and helps to understand some of the
challenges of field sampling through analysis of the so-called incorrect sampling errors during the
sample collection process [4,21]. The application of TOS includes the mandatory use of an appropriate
QAQC system, including the use of duplicates. Table 3 shows the cause and effect of the common
sampling errors and provides some solutions.
Minerals 2018, 8, 232 6 of 45

Table 3. Cause, effect, and solutions to TOS errors.

Nugget Effect
Sample Type Error Type Cause Error Effect Solution
Component
In-situ (e.g., linear In-situ Local Poor Larger samples
GNE
samples and core) heterogeneity representativity precision More samples
Optimised protocols
Constitution Poor
FSE SNE Larger samples
Random Heterogeneity precision
More samples
Broken rock (from
rock chips to blasted Constitution Optimised protocols
Poor
material and and Distribution GSE SNE More sample
precision
laboratory pulps) Heterogeneity increments
Poor quality Optimised protocols
DE, EE, WE,
Systematic sampling High bias SNE Efficient training
PE, and AE
protocols Strict QAQC
GNE: geological nugget effect; SNE: sampling nugget effect.

2.2. Correct Sampling Errors

The correct sampling errors (CSE) are considered “correct” because they cannot be removed by
perfect sampling procedures (Table 2). They are random errors and relate to the inherent heterogeneity
of the material being sampled and control precision [1]. In the context of grade control, low precision
generates ore/waste misclassification and poor reconciliation.
The FSE relates to the constitution heterogeneity of the material in question. The FSE
can be estimated via the FSE equation [10,22]; it does not cancel out and remains even after a
sampling operation is perfect. The FSE is controlled via the optimisation of sample mass and size
reduction process.
The second CSE is the grouping and segregation error (GSE). From a practical perspective, GSE
cannot be measured, but it is likely to have a material effect when not controlled [23]. It is controlled
by accumulating many small increments to form a composite sample. Segregation can theoretically be
reduced by homogenisation, though in the presence of liberated gold and/or sulphide particles it is a
futile exercise that often promotes further segregation [4,24].

2.3. Incorrect Sampling Errors

The incorrect sampling errors arise as a consequence of the physical interaction between
the material being sampled and the technology employed to extract the sample (Table 2).
Incorrect sampling errors are systematic errors which result in bias. They can be reduced by the
correct application of sampling methods, equipment, and procedures, and are the errors most likely to
affect grade control sampling [1].

2.4. FSE Equation and Application to Gold Ores

2.4.1. Background
The FSE can be modelled before material is sampled, provided certain characteristics are
determined [1,10,22]. The FSE is applicable to grade control samples once they are collected (e.g., a face
sample is broken rock in a sample bag). For any process where the FSE escalates, there is an associated
loss due to uncertainty, which in terms of grade control relates to ore/waste misclassification. The FSE
equation (“François-Bongarçon modified” version) is given as [22]:

FSE(rel var) = f·g·c (d95Au )b ·dn α (1/MS − 1/ML ) (1)

where f = shape factor; g = granulometric factor; dn = nominal material size; c = mineralogical factor;
d95Au = liberation diameter; b = (3 − α), where α is determined experimentally from Duplicate Series
Analysis (DSA) tests [25], or a default value of α = 1.5 is generally applied [22]; MS = sample mass;
and ML = lot mass. The sampling constant (K) is a product of: f·g·c (d95Au )b . Using this formula, it is
Minerals 2018, 8, 232 7 of 45

possible to: (a) calculate the error for a given sample size split from the original, or (b) calculate what
sub-sample size should be used to obtain a specified variance at a given reliability.

2.4.2. Calibration of FSE Equation Inputs

The key parameter to determination is the gold liberation diameter—a particle size
parameter [1,22]. For gold mineralisation, it can be defined as the screen size that allows 95% of
gold given a theoretical liberated lot to pass (d95Au ) [13,22]. If gold particles cluster, then d95Au should
be defined as the cluster diameter—dAuclus [16].
Approaches to FSE equation calibration range from guesswork to the implementation of
Heterogeneity Tests (HTs) or DSA. The HT allows the direct determination of K, whereas the DSA
approach (and variants thereof) determines both K and α [1,25–27].
The HT is the most commonly applied calibration method in the mining industry. The DSA
approach is both complex and time consuming to apply, so it is relatively rare compared to the HT.
The HT is prone to severe precision problems, particularly when coarse gold is present [27]. A further
problem is that it provides values for K only at the fragment size at which the calibration exercise is
carried out (often 10 mm). It has been shown that in such cases, the values for d95Au are far too low.
In the presence of coarse gold, the HT approach may require samples of hundreds of kilogrammes in
size. The method may not be a standard industry approach, since it is likely only to provide a correct
value when mineralisation is disseminated [27].
As a result of these issues, the authors recommend a direct and holistic approach to
characterisation with key outputs being [7,15,16,27,28]:

• Realisation of gold deportment, in particular the partitioning of gold as free gold, gold in sulphides,
and refractory gold.
• Gold particle size curve(s), including effects of clustering and relationship between gold particle
size and grade (e.g., high grade versus background grade).
• Definition of key FSE equation inputs (d95Au versus dAuclus ) and the sampling constant K.
• Recommendations as to optimum in-situ sample mass requirements.

The direct approach to d95Au may include a combination of methods such as core logging,
underground face observation, optical and automated microscopy, X-ray tomography, and staged
liberation and concentration [7,28–30]. Characterisation studies allow the practitioner to set the ore
sampling expectations across a number of d95Au -grade scenarios. In the production environment,
optimisation at the breakeven cut-off grade is usually the critical point. Where this requires specialist
and potentially costly protocols, then a cost-benefit analysis is required to understand the theoretical
need versus practical reality.

2.4.3. Applicability to Gold Ores

François-Bongarçon [31] and François-Bongarçon and Gy [22] addressed some key issues
in the application of the FSE equation and proposed a modified equation (Equation (1)).
The modified equation has been applied successfully by practitioners to optimise sampling
campaigns [7,17,23,29,32,33]. The use of the FSE equation represents a model that may or may not be
attained in practice, but provides a starting point from where protocols can be optimised. Results of
QAQC programmes will provide evidence for representativity through the application of field and
laboratory duplicates [34,35].

2.5. TOS and Geostatistics

Theory of Sampling and geostatistics are inextricably linked through the nugget effect [36].
In geostatistics, the nugget effect is the random component of variability that is superimposed on the
regionalised variable. In sampling, the magnitude of the nugget effect relates to: in-situ heterogeneity
of the mineralisation; sample support; sample density; and sample and assay procedures. The nugget
Minerals 2018, 8, 232 8 of 45

effect has two principal components [13,14,36]: the geological nugget effect (GNE) and the sampling
nugget effect (SNE).
The GNE relates to the microscopic differences in composition throughout the
mineralised zone/domain. It principally refers to the distribution of single grains or clusters
of gold particles distributed through the ore zone [16]. Variability is greatest where there are
small-scale, low continuity structures such as high-grade gold carriers within the main structure or
veinlets within wall rocks [14].
The SNE is directly related to the sampling process: that is, to the size of the field samples
taken; the effectiveness of collection; sample preparation; the size of sub-samples after splitting; and
analysis [13,14]. Throughout the mine value chain, optimised sampling protocols aim to reduce
the total sampling variance thereby also reducing the total nugget variance, skewness of the data
distribution, and number of extreme data values.

2.6. Sample Representativity

2.6.1. Representative Sampling

A sample is representative when the analytical results are within acceptable levels of bias and
precision [37]. Whilst precision (i.e., reproducibility) can be determined from duplicate samples, bias is
difficult to estimate without generally impractical and costly experimental efforts.
The Danish Horizontal Standard [38] uses the relative sampling variability (RSV) metric to
measure total sampling variability—the percentage coefficient of variability for repeat sample values.
The higher the RSV, the poorer the precision indicating that the sampling procedure requires
improvement. Relative sampling variability measures the total empirical sampling variance influenced
by the heterogeneity of the lot sampled under the current sampling procedure [38]. The RSV comprises
all stages of the sampling protocol, including all errors incurred by mass reduction, as well as
analytical error. The accepted value of RSV is up to the practitioner and based upon the nature
of the mineralisation in question, the data quality objectives, and what is cost-effective and practical.
Experience shows that the RSVs of underground linear samples generally lies between 50% and 400%,
but can reach 750% depending on the mineralisation. The RSV for a given sample type and support can
be reduced through TOS and QAQC application [7,39,40]. The ultimate test of sample representativity
is reconciliation between predicted grade and actual performance to within acceptable limits.

2.6.2. Data Quality Objectives

The data quality objective (DQO) is the level of total error that a given sampling protocol is
designed to achieve. It is quoted as a precision value reported at a given confidence limit, usually 68%
or 90%. The confidence limits can be defined as the reliability, where, for example, 90% means that
the assay results will lie within the given precision in 90 out of 100 times [2]. Pitard [37] states that
the total precision (error) for resource and grade control sampling should not be more than ±32%,
with the FSE component not more than half of the total error (±21%). Fundamental Sampling Error is
generally reported at the 68% reliability. It is up to the practitioner as to which reliability to accept, but
for grade control samples to achieve an FSE of ±21% at the 68% reliability is reasonable.
The key error components across the sampling value chain are: sample collection (Table 4;
sampling value chain (SVC) stages 2–3), preparation (Table 4; SVC stage 4), and analysis (Table 4;
SVC stage 5). These can be estimated via an empirical approach using duplicate samples [34,39].
The sampling component error contains both the GNE (e.g., the in-situ variability) and the
sample collection errors (e.g., delimitation (DE), extraction (EE), etc.). All TOS errors (e.g., FSE,
GSE, DE, etc.; Table 2) contribute to the SNE. The typical ranges of component errors are given
based on experience [6,7,34,37,39]. The optimal total target error is within ±20%, though for
high-nugget/coarse-gold mineralisation a value of ±40% may be more appropriate. Values for
the FSE and other TOS errors are based on general guidelines for acceptable uncertainty (Table 4) [37].
Minerals 2018, 8, 232 9 of 45

Table 4. Distribution of errors across stages of a gold grade control sampling programme. Refer to
Table 1 for the sampling value chain.

1 Duplicate 3Other
Error SVC Stage Sample Type/Activity TOS Errors Component 2 FSE TOS
Error Range Errors
In-situ sampling (e.g., core GNE
GNE, DE, EE, WE, PE
Sampling 2&3 and linear samples) ±20–70%
±16%
Broken rock sampling
(e.g., grab, sludge, and RC FSE, GSE, DE, EE, WE, PE
samples) ±23%

Drying PE
Preparation 4 ±5–20% ±11%
Crushing/grinding PE
Splitting FSE, GSE, DE, EE, PE
Splitting FSE, GSE, DE, EE, PE
Analytical 5 ±1–15% ±8%
Analysis AE
Total error ±20–70% ±21% ±23%
(±20%)
Target error (fine-gold)/coarse-gold ±32%
±40%
1 Potential component error range as determined from duplicate sample (pair) analysis [34,41]; 2 Maximum
recommended FSE distribution across the sampling stages [37]; 3 Maximum recommended other TOS error
distribution across the sampling stages [37]; SVC: sampling value chain; RC: reverse circulation.

2.6.3. Fit-for-Purpose Samples

A sampling and assaying programme must produce data that are fit-for-purpose for the required
need [4,17]. In this context, fit-for-purpose refers to sampling and assaying that contributes to a
resource/reserve that will be reported in accordance with the 2012 JORC Code (or other codes), and/or
a grade control programme that minimises ore/waste misclassification.
Development of sampling protocols in the context of TOS must be based on the specific
mineralisation [7,28]. A critical input is that of QAQC to maintain data quality through documented
procedures, sample security, monitoring of precision, accuracy, and contamination. If a batch of
samples can be deemed to be representative and associated assaying complies with QA documentation
and QC metrics, then fit-for-purpose is indicated. In addition, the fit-for-purpose test must consider
the spatial distribution and number of samples. For example, if all samples comply with QAQC
requirements, but do not sample the entire mineralisation (e.g., a certain vein set cannot be accessed
and sampled in large profile drives), then they are not fit-for-purpose to evaluate the mineralisation
in question.

2.7. Sample Spatial Distribution

2.7.1. Primary Sample Mass

Definition of gold particle size distribution within geo-domains supports the estimation of a
primary (in-situ) theoretical sample mass [15]. With discrete gold particles, use of a Poisson-based
method provides an indication of the sample mass that may be required to yield a given
precision [23,42,43]. For gold deposits with discrete gold, there is a high probability of drawing
non-gold particles. Hence, if a sample is too small, it will contain no gold. The Poisson model is a
limiting case of the Binomial model where the proportion (p) of the rare component (e.g., gold) is
very small, while the proportion of the material surrounding the gold (q) is practically unity (where
q = 1 − p). It can be applied to both liberated and non-liberated (i.e., in-situ) cases, since on the sample
scale, gold particles are rare events and can be considered as being discrete [23].
The approach assumes the equant particle (or grain) model where: (1) particles are free with no
composite particles; (2) particles are either nuggets or gangue minerals; (3) nuggets have the same size,
shape, and composition; (4) the gold occurs only in the nuggets; (5) a small number of nuggets are
present in samples; and (6) a large number of gangue particles are present in the samples. The critical
Minerals 2018, 8, 232 10 of 45

assumptions are (3) and (5) above. Assumption (3) is unlikely to be true since the population will have
variable particle sizes and shapes. For assumption (5), the larger nuggets will be rare, but the fine
background gold is likely to be more abundant and disseminated. The Poisson approach defines a
maximum theoretical sample mass based on the largest gold nuggets present.
For example, gold mineralisation at a potential underground breakeven cut-off grade of 3.5 g/t
Au and a d95Au of 1 mm, requires a 250 kg theoretical sample mass to yield ±20% precision at 90%
reliability. If each face is represented by a composite channel sample of 4 kg, then 63 samples are
required to evaluate the ore zone. Such a figure could be used to guide the sampling around a
stope panel. A 2800 t stope panel could be blocked out by raises and drives totalling 120 m. If each
advance was 1.8 m, then this is 70 samples around the panel. Seventy 4 kg samples yields a total mass
of 280 kg, which is above the theoretical optimum.
Given that theoretical sample mass, face separation, and stope panel size are relatively fixed,
the variables that can be changed are individual sample mass and number of samples per face.
An individual field sample may not be representative, particularly in coarse-gold mineralisation.
A theoretical field sample mass of many tonnes could be required to achieve an acceptable precision.
As noted previously, a reasonable strategy is to collect multiple samples across a given domain (e.g.,
stope panel). Each sample may be locally unrepresentative, though by taking numerous spatially
distributed samples the mean grade may be ultimately approached and variability reduced.

2.7.2. Sampling Domains

The optimum sample mass should be considered in the light of potential sampling domains,
which are spatial zones of similar sampling characteristics based principally on grade and d95Au .
The definition of sampling domains is an iterative process, where characterisation studies will allow
sampling expectations to be set across a number of d95Au -grade scenarios, thus allowing the worst
case scenario to be applied via an achievable protocol across the mineable width. In the simplest case,
a vein and wall rocks may have different sample mass requirements (Table 5).

Table 5. Example from a South American project of sampling characteristics for vein-stockwork wall
rock mineralisation domains.

Parameter/Domain Wall Rocks Central Vein

Mineralisation Characteristics
Quartz vein stockwork hosted in black
Geology Composite laminated quartz vein
shale. Individual veins to 10 mm width
Gold deportment Free gold in quartz and pyrite Free gold in quartz
Domain width (m) 5 1
Grade (g/t Au) 5 16
Nugget effect (%) 25 60
Sampling Characteristics
>100 µm (%) >2 50
d95Au (µm) 75 650
Sampling constant (K: g/cm1.5 ) 630 2500
Theoretical sample mass (kg) 5 20
Theoretical mass per unit length (kg/m) 1 20
1 Duplicate Channel Sample Analysis
Field component error (%) 14 33
Preparation component error (%) 8 14
Analytical component error (%) 8 11
Total error (%) 18 37
No. pairs (field, preparation, and pulp) 50 50
1 Sample protocol comprises 5 kg field sample crushed to P90 −1.5 mm and rotary split to 2 kg, pulverised to P90
−75 µm, and the entire pulp subjected to screen fire assay. Protocol FSE for domains ±7% (wall rocks) and ±20%
(central vein), respectively, at 68% reliability. Errors estimated via the method presented in [34,41].
Minerals 2018, 8, 232 11 of 45

In the example presented, a central high-grade narrow vein is surrounded by lower-grade

wall rocks, with stockwork vein sets (Table 5). Characterisation of both domains indicates that
the high-grade central zone contains coarse gold. Consequently, the theoretical sample mass for both
zones varies from 5 kg (wall rocks) to 20 kg (central), respectively, to achieve ±20% precision at 90%
reliability. Mining focusses on the central vein, with the wall rocks making up diluting material in
3 m by 3 m development drives and a minimum stoping width of 1.5 m. Face chip-channel samples
of 5 kg/m were collected across the wall rocks and vein. These are of a similar support to the NQ
diamond drilling used. A duplicate sample study gave total errors of 18% and 37%, respectively.
Both values indicate acceptable error for the fine-gold wall rock and coarse-gold vein scenarios.

2.7.3. Sample Density

Linear and drill sample spacing can be optimised using variography. However, this requires the
existence of substantive data to produce a realistic variogram model. The spacing of development
samples is controlled by the separation of faces, and thus ranges between 1.8 m and 4 m. Extension
variance theory can be used to assess the geometric distribution of samples around and within a block;
however, it requires input of variogram parameters. Initial sampling programmes, particularly for
drilling, are likely to be driven by experience-based spacing until variography can be completed.

3. Grade Control Sampling Methods

3.1. Type of Samples

Many gold deposits are vein- or shear-hosted structures that have a thickness dimension that
is considerably less than their strike and dip dimensions. Various sample types have traditionally
been used for grade control (Table 6). The most common are in-situ samples collected by the geologist
or geotechnician from development walls or faces. Broken rock sample collection is conducted by
grabbing from rock piles (e.g., stockpiles, dumps, trucks, etc.). Core and percussion drilling methods
are also applied (Table 6). Samples are collected at all stages of the mine development chain from
pre-production, to stoping, and post-production (Table 7).

Table 6. Underground grade control sampling types and methods. Note, strictly drilling is a linear
sampling method, though it is classified separately here as it is different from the main grade
control methods.

Sample Type Sampling Method Definition Photograph

Individual punctual chip specimens
Chip collected in a discontinuous fashion Figure 2A
aggregated to form a sample
Series of continuous chips aggregated to
Linear Chip-channel Figures 2B, 3 and 4
form a sample
Consistent volume cut across sample zone,
Channel
can be hand or diamond saw cut Figure 5A,B
Large and/or multiple chip or
Panel
chip-channel samples
Broken rock Grab Hand “grab” of sample from a rock pile Figures 6 and 7
Diamond core Solid core of sample over a given length -
Chippings over a given length recovered
Reverse circulation -
though the centre of the drill string
Drill hole
Chippings over a given length recovered up
Blast hole/sludge Figure 8
the outside of the drill string
Minerals 2018, 8, 232 12 of 45

Table 7. Sample type for different stoping methods during the mining production chain.

Stoping Type Development and Production Post-Production (Mucking

Stoping Method
[Selectivity] Pre-Production Stage (Stoping) Stage and Transport) Stage
Shrinkage
Linear samples
Cut and fill
Entry [Selective] Linear samples Grab samples
Room and pillar Grab samples Blast hole/sludge
Drift and fill Diamond core Grab samples
Reverse circulation
Longhole (open and Blast hole/sludge
sub-level variants) Non-entry
Grab samples
[Non-selective]
Sub-level caving
Block caving

3.2. Linear Sampling

3.2.1. Chip Sampling

This technique involves taking a series of chips over a continuous or semi-continuous band
using a hammer and chisel (Figure 2). When a series of discontinuous chips are collected,
a series of punctual samples result which are effectively “specimens”. Depending on geology, the
chip-band can be cut horizontally, vertically, or perpendicular to the dip of the ore body (Figure 3).
The separation of the individual chips varies from touching (a chip-channel) to centimetres or
more apart. Chip sampling is susceptible to extreme errors (DE and EE) through operator preferences
and rock hardness (Tables 8 and 9).
Minerals 2018, 8, x FOR PEER REVIEW 12 of 43

(A) (B)

Figure 2. (A) Chip-channel sampling. Samples collected from sidewall of shanty drive. Each sample
Figure 2. (A) Chip-channel sampling. Samples collected from sidewall of shanty drive. Each sample
comprises near continuous chips collected across the vein; (B) Hand channel sampling of side walls.
comprises near continuous chips collected across the vein; (B) Hand channel sampling of side walls.

Table 8. Likely sampling errors during the collection of chip or chip-channel samples.

Sampling Error 1 Error Magnitude Error Effect

DE Moderate-high Often high as the sample is generally poorly demarcated.
Expected to be high, as the sample is chipped out and
generally leaves an irregular band in 3D. May be
EE Moderate-high extreme if very different rock properties are encountered
across a sample. Prone to operator bias for
high-grade/shiny sections and/or soft material.
Dependent upon EE and specifically rock hardness. Can
WE Moderate
be variable.
1 Indicative error magnitude: Red: high (> ±15%); Orange: moderate (±5–15%).

Figure 3. Orientation of linear samples (e.g., chip-channel) to match geological requirements. Sub-vertical
quartz reef in foreground with area of flat veining on the footwall (blue line marks footwall contact).
Sub-vertical reef sampled horizontally (samples 1 and 7), whereas flat footwall veins sampled
vertically (samples 2–6).

Table 8. Likely sampling errors during the collection of chip or chip-channel samples.

Sampling 1 Error
Error Effect
Error Magnitude
 (A) (B)
Minerals 2018, 8, 232 13 of 45
Figure 2. (A) Chip-channel sampling. Samples collected from sidewall of shanty drive. Each sample
comprises near continuous chips collected across the vein; (B) Hand channel sampling of side walls.

Figure 3. Orientation of linear samples (e.g., chip-channel) to match geological requirements. Sub-vertical
Figure 3. Orientation of linear samples (e.g., chip-channel) to match geological requirements.
quartz reef in foreground with area of flat veining on the footwall (blue line marks footwall contact).
Sub-vertical quartzreef
Sub-vertical reefsampled
in foreground with
horizontally area of
(samples flat 7),
1 and veining onflat
whereas thefootwall
footwall (blue
veins line marks
sampled
footwallvertically
contact).(samples
Sub-vertical
2–6). reef sampled horizontally (samples 1 and 7), whereas flat footwall veins
sampled vertically (samples 2–6).
Table 8. Likely sampling errors during the collection of chip or chip-channel samples.

Sampling Table 9. Chip sampling attributes for

1 Error
grade
Error Effectcontrol.
Error Magnitude
DE Moderate-high Often high as the sample is generally poorly demarcated.
Attribute Expected to be high, as the sample is chipped Comment
out and generally leaves an irregular
EE Moderate-high band in 3D.Easy:
May becan
extreme
be collected withrock
if very different properties
minimum are encountered
preparation, across acan
though
Logistics and planning sample. Prone to operator bias for high-grade/shiny sections and/or soft material.
be time consuming if rock is hard.
WE Moderate Dependent upon EE and specifically rock hardness. Can be variable.
Moderate–high:
1 Indicative error magnitude: any
Red: high (> ±15%); reasonable
Orange: number
moderate of samples can be
(±5–15%).
Frequency of collection
collected, but will take time.
Table 9. Chip Low–high:
sampling attributes
Depends for upon
grade sampling
control. location. Local safety
Hazard exposure
Attribute protocols must beComment
applied. Requires sampler to be at rock face.
Logistics and planning Easy: can be collected with minimum preparation, collection.
Risk of fly rock during though can be time consuming if rock is hard.
Frequency of collection
Moderate–high: any reasonable number of samples can be collected, but will take time.
Flexibility of method High: can be undertaken at relatively short notice.
Low–high: Depends upon sampling location. Local safety protocols must be applied. Requires
Hazard exposure
Cost sampler to be at rock Low–moderate:
face. Risk of fly rock labour intensive though relatively fast.
during collection.
FlexibilitySample
of method High: can be undertaken
quality Low:at relatively
generally short
lownotice.
quality sample.
Cost Low–moderate: labour intensive though relatively fast.
Sample quality Moderate–low:
Low: generally low quality sample. whilst relatively quick in the production
Value proposition
Moderate–low: whilst setting, thequick
relatively poorinquality sample provides
the production setting, theapoor
low quality
value that may
sample
Value proposition
provides a low value lead to ore/waste
that may misclassification.
lead to ore/waste misclassification.

3.2.2. Channel Sampling

This method involves cutting a channel across the rock face. Depending on geology, the channel
can be cut horizontally, vertically, or perpendicular to the dip of the ore body (Figure 3). The channel
should be kept at a uniform width and depth to ensure a low DE and EE. Channels should generally
be around 5 cm to 10 cm wide and 2.5 cm deep to yield a sample of between 3–6 kg/m. In practice,
hand-cut channel samples are relatively difficult to keep uniform, and can be more like a “chip-channel”
sample. Cutting of channels rarely permits equal depth representation over the entire cross-section
(e.g., DE) due to different rock hardness and human effects. Sample quality may decrease as the
channel progresses across the face due to poor sampler motivation and fatigue.
Some operations use a diamond saw to cut channels in an attempt to improve sample quality
(Figure 4A–C) [7,32,44,45]. Two parallel saw cuts to about 2.5 cm depth are made 5 cm to 10 cm apart
across the required width (Figure 4B). A hammer and chisel (or compressed air pick) is then used to
extract the intervening block of sample (Figure 4D).
 sample. Cutting of channels rarely permits equal depth representation over the entire cross-section
(e.g., DE) due to different rock hardness and human effects. Sample quality may decrease as the
channel progresses across the face due to poor sampler motivation and fatigue.
Some operations use a diamond saw to cut channels in an attempt to improve sample quality
(Figure 4A–C) [7,32,44,45]. Two parallel saw cuts to about 2.5 cm depth are made 5 cm to 10 cm apart
Minerals 8, 232
2018,the
across required width (Figure 4B). A hammer and chisel (or compressed air pick) is then used to14 of 45
extract the intervening block of sample (Figure 4D).

(A) (B)

(C) (D)

Figure 4. Collection of saw-cut channel sample (A) and resulting channel (B); Cutting a sidewall
Figure 4. Collection
channel sample (C)of saw-cut
and channel
using an air pick sample (A)
to remove theand resulting
sample (D). channel (B); Cutting a sidewall
channel sample (C) and using an air pick to remove the sample (D).
A typical 1 m length saw-cut channel cut should yield a sample mass of 4 kg (2.5 cm by 5 cm
profile).
A typical It is1 unlikely
m length that perfect sample
saw-cut channelextraction will yield
cut should be achieved due mass
a sample to irregularities at the
of 4 kg (2.5 cmbase
by 5 cm
profile). It is unlikely that perfect sample extraction will be achieved due to irregularities at the base of
the cut (e.g., where the sample block breaks free, EE). Where a large sample is required, channel width
and/or depth can be increased or multiple channels cut. The principal barrier to low DE in saw-cut
channels is rock face roughness. Where the face is very uneven, then even a saw cut is difficult to place
and may result in high DE and EE across the sample zone.
Single- or double-blade saw units are generally powered by compressed air and require water
for dust suppression. Single-blade units are the most appropriate, from both a safety and efficiency
perspective. Double-blade units have the advantage that two parallel cuts are made at the same time
2 cm apart, though double-blade units are generally heavier and pose more of a manual handling and
safety issue. Multiple cuts, one below the other, can be made using the upper or lower original cut as
a guide.
Saw-cut channels are generally of higher quality, though bias may occur when fines are washed
away during cutting, particularly in soft sections (EE). With rough/uneven rock surfaces, sample
quality can be compromised since the extracted sample has a variable support along the channel.
A larger saw blade may help to ameliorate a rough surface. The cost and time for taking saw-cut channel
samples depends on rock hardness and the availability of services to run the saw, but experience
Minerals 2018, 8, 232 15 of 45

shows that a well-trained two-person team can cut and collect a metre of channel sample in 10–15 min.
Proper training and the use of personal protective equipment will reduce safety concerns with the use
of saws.
A reasonable quality channel sample can be extracted using an air pick chipping device
(Figure 4D), but it is prone to operator induced DE bias, either purposeful (e.g., a deep cut around a
high grade section) or accidental (e.g., a deeper cut around soft material). Air picks have the tendency
to produce fly-rock, which is both an EE and safety issue. Channel sample attributes and errors are
given in Tables 10 and 11.

Table 10. Likely sampling errors during the collection of channel samples.

1 Error Magnitude
Sampling Error Error Effect
Hand Cut Saw Cut
Generally good as the sample is similar to a
DE Moderate Moderate
drill hole.
High if different rock properties are
encountered across a sample. For hand
collected samples, hard rock leads to loss of
sample in hard rock as fly particles. In
EE High Moderate–low
soft/friable material will lead to bias due to
oversampling or loss. Saw-cut channels
generally better, though cooling water may
lead to loss in soft/friable areas.
Dependent upon EE and specifically rock
WE Moderate Low hardness. Can be variable. Reduced if saw-cut
channel.
1 Indicative error magnitude: Red: high (>±15%); Orange: moderate (±5–15%); Green: low/acceptable (<±5%).

Table 11. Channel sampling attributes.

Attribute Comment
Moderate–difficult: requires time and/or compressed air saw
Logistics and planning
to collect.
Moderate: any reasonable number of samples can be collected, but
Frequency of collection
will take more time than chip
Low–high: Depends upon sampling location. Local safety protocols
must be applied. Requires sampler to be at rock face. Risk of fly rock
Hazard exposure
if collected by hand or air pick. Care required if a diamond saw is
used due to rotating parts.
High: can be undertaken at short notice. Slow to cut by hand, 30 min
Flexibility of method or more for 1 m sample. Can be quite quick to collect a saw-cut
sample once the saw is set up.
Cost High: labour intensive and can require compressed air saw.
Sample quality High: potentially a high quality sample.
Moderate–high: whilst relatively time consuming and costly, the
Value proposition
higher quality sample provides increased value.

3.2.3. Panel Sampling

Panel samples comprise multiple chips spaced across a panel (i.e., <50% rock sampled, chip-panel
sample) or contiguous chips over the entire panel area (panel-channel sample; >50% rock sampled)
(Figure 5).
Dispersed clusters of coarse gold in a vein system at the Gwynfynydd Mine (UK) created a
negative bias in grade control samples, leading to the replacement of standard chip-channel samples
by panel-channel samples [12]. Faces were divided into horizontal quarters and close-chipped to give a
25 kg sample from across each of two panels. Processing of the combined 50 kg micro-bulk sample was
Minerals 2018, 8, x FOR PEER REVIEW 15 of 43

Dispersed clusters of coarse gold in a vein system at the Gwynfynydd Mine (UK) created a
negative bias in grade control samples, leading to the replacement of standard chip-channel samples
by panel-channel
Minerals 2018, 8, 232 samples [12]. Faces were divided into horizontal quarters and close-chipped to 16give
of 45
a 25 kg sample from across each of two panels. Processing of the combined 50 kg micro-bulk sample
was through a gravity separation unit located on surface. The resulting data were used to map grade
throughalong
trends a gravity separationand
development unitto
located on surface.
provide The resultinggrade.
a semi-quantitative data were
Theused to map grade
potentially larger trends
mass
along development and to provide a semi-quantitative grade. The potentially larger mass
(>100 kg) of panel samples compared to chip or channel samples, means they are typically used for (>100 kg) of
panel samplessample
metallurgical compared to chip (Figure
collection or channel
5). samples, means they
Panel sampling areare
errors typically
given used for metallurgical
in Table 12 and key
sample collection
attributes in Table 13.(Figure 5). Panel sampling errors are given in Table 12 and key attributes in Table 13.

(A) (B)

(C) (D)

Figure 5. (A) Panel sample preparation. Large (100 kg) samples were required for metallurgical testing
Figure 5. (A) Panel sample preparation. Large (100 kg) samples were required for metallurgical testing
and grade verification; (B) Collecting panel samples with hammer and chisel. Samples were chipped
and grade verification; (B) Collecting panel samples with hammer and chisel. Samples were chipped
from the drive sidewalls onto a plastic floor sheet; (C) Part of panel sample. Chip fragment size is
from the drive sidewalls onto a plastic floor sheet; (C) Part of panel sample. Chip fragment size is
reasonably well controlled, with a maximum fragment size of about 8 cm, and most below 5 cm. With
reasonably well controlled, with a maximum fragment size of about 8 cm, and most below 5 cm. With
samples
samples collected
collected from
from brittle
brittle material
material such
such as
as quartz
quartz there
there are
are aa substantial
substantial proportion
proportion of
of sub-1
sub-1 mm
mm
fragments present; (D) Alternate sample capture method using rubber
fragments present; (D) Alternate sample capture method using rubber mats. mats.

Table 12. Likely sampling errors during the collection of panel samples.
Table 12. Likely sampling errors during the collection of panel samples.
Sampling Error 1 Error Magnitude Error Effect
Sampling Error Low–moderate1 Error Magnitude Error on
Effect
DE Generally good as the sample is delineated the face.
DE Low–moderate Generally
Can be acceptable, but maygood as thehigh
be locally sample is delineated
as rock on the face.
chips are collected. May be
EE Moderate–high Can be acceptable,
very high if different but may
rock properties arebe encountered
locally high as rock chips
across a panel.
are collected. May be very high if different rock
Oversampling of “shiny” and/or soft/friable material.
EE Moderate–high
properties are encountered across a panel. Oversampling
WE Low–moderate Dependent upon EE and specifically rock hardness. Can be variable.
of “shiny” and/or soft/friable material.
1 Indicative error magnitude: Orange: moderate (±5–15%);upon
Dependent Green:
EE low/acceptable
and specifically(<±5%).
rock hardness. Can
WE Low–moderate
be variable.
1 Indicative error magnitude: Orange: moderate (±5–15%); Green: low/acceptable (<±5%).
Minerals 2018, 8, 232 17 of 45

Table 13. Panel sample attributes.

Attribute Comment
Logistics and planning Moderate–difficult: requires time to collect by hand.
Low–moderate: any number of samples can be collected, but will take
Frequency of collection
time and depend upon sample size.
Low–high: Depends upon sampling location. Local safety protocols
Hazard exposure must be applied. Requires sampler to be at rock face. Risk of fly rock
during collection.
High: can be undertaken at short notice, but may take at least 30 min
Flexibility of method
per sample to collect.
Cost Moderate–high: potentially cost intensive for large samples.
Moderate–high: potentially a high quality sample with large area of
Sample quality
face coverage.
Moderate–high: whilst relatively time consuming and costly, will
Value proposition
provide a large sample mass.

3.3. Discussion: Practical Issues of Linear Sampling Methods

3.3.1. Integration with Geological Mapping

Geological mapping is a critical activity during grade control and is particularly applicable where
geology and geometry are variable and mine openings need to respond accordingly [12]. The primary
objective is to identify and locate the ore zone or at least the section likely to contain the ore with
respect to the minimum stoping width. The secondary objective is to determine the attitude and
position of relevant structures that control the 3D form of ore shoots or gold carriers (Figure 3). Thus,
mapping has an intimate relationship with sampling [12,46]. Sample locations should be picked up by
the mine surveyors to ensure that they can be recalled in 3D space to support resource models and
mine design.

3.3.2. Sample Collection and Methodology

The geologist must ensure that samples are delineated and extracted efficiently to minimise DE
and EE. This means an even width and depth across the sample zone must be cut and that all liberated
rock fragments are captured in the sample. Extraction of saw-cut channels is relatively easy, but more
challenging for hammer cut channels. Chisels should be of high quality with a sharp point, as blunt
points eject fine rock fragments (EE).
Sample capture is important (minimisation of EE), where one of the most effective methods is to
use of a bucket or purpose-designed receptacle that can be held by an assistant close to the rock face.
It is important to ensure that the receptacle opening is wide enough to allow sample material to enter.
A tarpaulin can be placed on the drive or stope floor immediately below the delineated sample area
(Figure 5B,C). An alternate sample capture method is to use rubber mats (e.g., lengths of conveyor
belt; Figure 5D), which whilst relatively heavy, can be cleaned more effectively. Sample contamination
(PE) needs to be minimised during collection. Key areas for consideration include: cleaning of faces or
walls prior to sampling (avoid use of dirty water); ensuring that all sample collection equipment is
clean; avoiding the re-use of sample bags (particularly calico bags where fines can become entrained in
the weave); and minimising of fly-rock contamination during sampling.
Chip and channel sampling are prone to EE because as much as 20% (typically 5–10%) of the
extracted (chipped) sample mass is lost as fly rock rather than being captured in the sample. Extraction
error can be reduced by careful chipping and ensuring that any sample receptacle is large enough.
A two-person sampling team reduces both DE and EE, where one person extracts and the other catches
the sample material. Where channel sampling is used, a piece of wood or aluminium cut to the channel
dimensions can be used to gauge delimitation.
Minerals 2018, 8, 232 18 of 45

Where multiple channels or chips are being collected across a face or sidewall, they should be
collected from the bottom sample upwards. This approach avoids contamination of lower samples
from those above from falling fragments. Where the sample band/channel is inclined to cut the ore
zone at right angles, sampling should commence at the lower end and work upwards.
The greatest EE during face sampling arises from differential, hard versus soft rock properties, as
a greater quantity of softer material usually enters the sampling bag (Fourie and Minnitt, 2016). If this
is high-grade, then the sample will be positively biased, and if low-grade, vice versa. In very hard
sections, there is a likelihood of sample loss, as rock chips from each hammer blow will fly away from
the sampling bag or a lack of sample collected.
In a study based on Witwatersrand (RSA) gold mines, Fourie and Minnitt [21] identified an
“in-stope sample discard process”, where samplers were prone to make a selection of what they believe
was better looking material. Such discard of waste and preferential selection of mineralised material
leads to a positive “waste discard bias” and hence higher sample assays [21].
In well-jointed and/or bedded rocks, collection by hammering presents difficulties in that adjacent
non-sample material may fall into the sample lot (e.g., contamination; PE). This alien material is
unwanted and needs to be either removed or avoided by screening or barring down the loose zone
prior to sampling.
The quality of chip and channel samples is highly dependent upon the sampling crew and the
underground environment. The best-trained crews are unlikely to collect a high-quality sample.
Quality will vary from crew to crew and from shift to shift. The mine environment will also effect
quality through local temperature/humidity (e.g., fatigue effects), sample location (e.g., drive vs. stope
vs. raise), and time of shift (e.g., fatigue/motivation effects). Additional bias (DE) also comes from
the human eye, where the geologist will be drawn to collecting a larger sample from visible gold or
high-grade areas (e.g., sulphide-rich). Collector bias can be reduced by ensuring that all samples are
collected along a marked grade line and staff training (Figure 1).
In many mines, face samples are taken at a height above the drive floor, often 1 m ± 0.5 m, though
this depends on the type of ore body. For example, in the shallow-dipping Witwatersrand reefs, face
samples are taken across the reef irrespective of how high or low it is from the floor of the stope.
In some operations, the face is divided into two horizontal halves, with a single channel taken along
the half line or one sample taken along the centre of each half (i.e., two face samples). For larger drive
profiles, 3 m by 3 m and greater, it is difficult to sample safely above 2 m. In this case, a service vehicle
or rugged step-ladder is required to provide a safe working platform. In a production environment
this may be difficult, and in some cases a sample may not be collected. If relevant, the geologist will
have to use their experience to evaluate the effect of not sampling a particular geological domain.
Rowland and Sichel [47] compared chip and channel sampling within selected Witwatersrand
mines, concluding that: cutting chip or channel samples by hand was difficult, sampling supervision
and training are critical, and when using compressed air saws or picks it was possible to collect better
quality samples quicker. Storrar [11] emphasises the need to minimise DE and EE during sample
collection, and the need for the sampler to be motivated, trained, and encouraged to maintain a high
standard in sampling practice.

3.3.3. Sampling under Production Conditions

Time spent at a working face is critical, therefore the geologist must develop skills to collect the
best information in the shortest time. There is rarely an hour to spend at a face and more likely it is
15–30 min. Operational pressure tends to be higher in large mechanised mines, whereas in smaller
mines there is usually a higher focus on grade rather than tonnes. Whilst not wanting to unduly hold up
development, time to do the job properly and collect samples in a geological context is imperative. The
authors maintain that once the mining crew has made a face safe, it is not unreasonable for a geologist
to be allowed 30 min at the face per shift. Actual time required will depend on ore body complexity
and drive size. In reality, the actual delay cost by allowing geologists access to a face is minimal, though
Minerals 2018, 8, 232 19 of 45

few mines will neither realise or admit this. However, it can be very difficult to get the cooperation
of the drilling crews if their bonus depends on meters of advance or number of holes drilled. Grade
control activity should be factored into the mining contract through geological-related quality checks,
for example, stope and drive profile measurement and sampling.
Where shanty drives are cut in narrow low-angle ore zones, the exposure of mineralisation in
sidewalls allows sampling to be potentially undertaken behind the development face if immediate
ore/waste decisions are not required (e.g., Nalunaq [32]; Figure 2A).
There can be an inverse relationship between the size of a mine and sample mass collected at
face, generally where smaller mines collect larger samples. Similarly, there can also be a relationship
between mine size and sample quality, where smaller mines collect higher quality samples. However,
some large mines, notably those on the Witwatersrand, have put much time and effort into improving
sample quality [19–21]. Mass and quality issues relate to: the rate of face advance, time pressures, and
availability of the geologist. Some mines require 24/7 coverage by the geological team to ensure all
development faces are sampled.

3.3.4. Sampling and the Geological Context

A major focus is to ensure that samples are collected in a geological context. Beyond the digital
camera, some mines use rugged computers to record face data. Automated 3D digital face mapping
systems are now also available [48]. The importance of geological input into grade control cannot be
overstated and has implications for selective mining and stope design [12].
Sample lengths should generally be less than 1.2 m and greater than 0.3 m. Where the vein is
very narrow (<0.5 m) or a key geological domain within a larger structure is very narrow, care must
be taken. In some instances, any chip sample collected will be small (<1 kg). In one example, the
author has seen geologists collect a purposely large sample (e.g., 4 kg) in a narrow (25 cm) high-grade
(>30 g/t Au) vein domain within a 2.5 m host reef. This yielded a support of 16 kg/m compared to
other samples in the same face, which had a support of 6 kg/m. A strong grade bias resulted in the
face grades. In high-grade narrow structures, the probability of sample collection bias, planned or not,
is high when visible gold is present.
Geologically-controlled high-quality samples are particularly important where the ore body
boundary is based on an assay contact. There may be no convenient geological indicators, with reliance
being placed on the final assay grade to determine ore and waste. In this case, samples down to
0.1 m may be collected. During the sampling of faces, it is important for the geologist to take note
of the contacts between mineralised and unmineralized material. In some high-grade deposits it is
common for visible gold to occur at the boundary between two rock types (e.g., quartz versus shale or
massive versus laminated quartz). When cutting samples, it is likely that the high-grade material can
be sampled twice for the different rock domains. In this case, it is important for the true high-grade
lithology sample to be extended a few cm into the lower grade zone to avoid contamination of the
lower grade zone.
Fourie and Minnitt [21] identified “soft-reef” bias within the carboniferous reefs of the
Witwatersrand. So-called soft-reef bias contributes to the generation of significant sampling error and
bias is the result of the over-extraction of soft, carbon-rich material compared to the harder host rocks.
This bias results in an overestimation of the grade of the in-situ sample [47,49,50]. Cawood [50] suggests
that the soft-reef bias results in the overestimation of grade by up to 26%, due to the over-extraction
(e.g., EE) of the high-grade carbon-rich material.
Wall rocks may contain disseminated gold and/or stringer veins, which will need to be sampled
as separate domains from the main gold carrier (e.g., vein, shear, dyke, etc.; see Table 5). At an early
stage, characterisation is required to investigate which geological domains bear gold and in what form.
Minerals 2018, 8, 232 20 of 45

3.3.5. Recording Sampling Quality

The key qualifiers of sample quality are DE and EE. A simple observational index that can be
recorded on the sample log/face sheet to report quality, which can then be integrated into the database
(Table 14). Rating is based on the visual review of DE and EE of the sample. It is then possible for
the geologist to review the quality of samples being used to inform a given stope/resource block(s).
Should the majority of samples have an “LQ” (i.e., low quality) rating for example, then any resource
estimate can take this into account by using a lower reporting category if appropriate.
A study undertaken by Freese, Flitton, and Pillay [19] on chip sampling at the Kopanang mine
(Witwatersrand) led to the introduction of a quality index (QI). The QI measured a number of
parameters based on before and after photographs of each sample site including: area and depth of the
sample, width of sample, sample mass, and time taken to collect the sample. They found that closer
supervision of sampling activities and application of the QI led to increased sample masses being
recovered by samplers. Fourie and Minnitt [20] used a laser scanner to profile a face before and after
sampling to produce a 3D image of the sampled zone.

Table 14. Sample quality rating as applied at the Gwynfynydd mine. Rating applied to both
chip-channel 1 and panel 2 samples.

Sample Quality Code Definition

Well-defined and extracted/recovered sample. Clear cut channel or
High HQ
paneled area. Sample mass >90% of target mass
Moderate to well-defined and moderately extracted/recovered sample.
Medium MQ
Sample mass 70–90% of target mass
Moderately defined and poorly extracted/recovered sample. Sample
Low LQ
mass <70% of target mass
1 Target channel sample mass generally 5 kg/m; 2 Target panel sample mass 50 kg across 2–3 m by 0.5 m panel area.

An alternate method is to record the mass of each sample at the laboratory after drying.
This measures the quantity of recovered rock over the expected mass. A target acceptance of ±5%
is nominally taken around the expected. Above or below this range, the samples are either over- or
under-sampled. This method is not perfect, for example an under-extracted sample could yield a close
to normal mass based on contamination by fallen material out of the delimited area.
Example of a mass-recovery QI for linear samples is shown in Table 15, where the rating is based
on percentage of samples on target. The data was collected from the same mine by the same crew
from 2.5 m by 2.5 m faces, which comprised 1.0–1.3 m of quartz vein hosted in weak–strongly altered
dolerite wall rocks. The highest quality sample was the saw-cut channel, which achieved 75% of
samples on target. The lowest quality was the chip samples, achieving 0% with all under-sampled.
The air-pick cut channel yielded a low value, which was due to excessive loss (EE) by fly rock.

Table 15. Example of mass recovery (quality index (QI)) for linear samples across quartz-dominated
vein mineralisation from a North American mine [42].

No. 1 Expected 2 Data Range 3 Under-Sampled 4 Target (%) 5 Over-Sampled 6 Quality

Sample Type
Samples (kg/m) (kg/m) (%) [within ±5%] (%) Rating (%)
Chip 105 2.0 0.4–1.8 100 0 0 0
Chip-channel 135 2.0 1.0–3.2 50 24 26 24
Panel (hand cut) 45 6.0 4.7–7.1 50 41 9 41
Channel (air-pick cut) 135 2.5 1.3–2.8 43 42 5 42
Channel (hand cut) 120 3.0 1.8–3.9 36 53 11 53
Channel (saw cut) 125 3.0 2.2–3.5 20 75 5 75
1 Mass of sample based on chip or channel target volume; 2 actual range of mass collected: 3 proportion of samples
below 95% of target mass; 4 proportion of samples within target ±5% mass; 5 proportion of samples above 105% of
target mass; 6 and proportion of samples within target ±5% mass.
Minerals 2018, 8, 232 21 of 45

3.3.6. Sample Support

Sample support (e.g., volume) and quality (e.g., minimised ISE) exert controls on the sampling
nugget effect, where larger samples yield a lower nugget effect [13,14,36]. Magri and McKenna [45]
report a comparison study between chip and channel (saw cut) samples in two Witwatersrand mines.
In one mine (Randfontein), the chip samples were taken under close supervision and were then
duplicated with minimal supervision. Variography of the two sets revealed very similar structure
(e.g., model type and range), but different total nugget effects with chip sample nugget effect
being higher. In the second mine (Western Areas), the initial chip samples were duplicated with
saw cut channel samples. Variography showed that the total nugget effect of the channel samples was
half that of the original chip samples. The authors also concluded that a spacing of 5 m for the channel
samples would be equivalent to a spacing of about 2.5 m for the chip samples. Similar results were
also noted in other Witwatersrand studies [44,49].
Dominy [42] reports on a study undertaken in a North American gold operation. Three 50 m
long drives up-dip from one another were re-sampled under close supervision using air-pick cut
channel samples (support of 3 kg/m). The new channels were cut as close as possible to the site on the
backs of the previous face chip-samples (support of 3 kg/m). Sample preparation and assay protocols
were defined for a moderately coarse gold-bearing ore for the new sample set, including 1 kg screen
fire assays. The new data (98 samples) displayed a 20% upgrade in comparison to the original data.
The nugget effect of the new data was 61% compared to 80% for the original data. Dominy et al. [7]
report on the evaluation of a sheeted vein gold deposit, where 60 saw-cut channel (8 kg/m) and chip
samples (3 kg/m) were collected as field duplicates. The channels yielded a nugget effect of 63% (RSV
145%) and the chips a nugget effect of 87% (RSV 296%).

4. Grab Sampling

4.1. Introduction
There are a number of mines using grab samples (or muck or broken rock samples) as a method of
grade control [51,52]. The goal of grab sampling is to predict the grade of ore from a particular source
(e.g., stope, stockpile, or truck) and/or reconcile with the predicted grade. It is often used because of
access issues (e.g., non-entry stopes—sub-level caving operations), safety (e.g., to avoid unsupported
backs), or lack of other sample data. On occasion, samples may also be taken from drive or stope floors
to monitor gold loss. Surface grab sampling is often used to monitor the grade of stockpiles prior to
blending or feeding to the plant or for reconciliation purposes. A fuller discussion and case studies are
given in Dominy [51], and Carswell et al. [39] present a case study based on the high-grade Gosowong
mine (Indonesia).

4.2. Collecting Grab Samples

Grab sampling involves collecting a large sample or series of smaller samples (that are later
combined) from the muckpile at a face or drawpoint, or from tram cars or trucks transporting the ore
from these points. Samples are generally collected by hand or shovel by a geologist, field technician,
or miner (Figure 6). Non-scientific protocols usually define sample mass and number of samples,
where an individual mass may range between 1 kg and 6 kg. The mass is limited in practical terms
by health and safety issues, as well as the capacity of the laboratory to process samples within a
reasonable timeframe. The number of samples collected is generally arbitrary and often constrained
by laboratory capacity.
from these points. Samples are generally collected by hand or shovel by a geologist, field technician,
or miner (Figure 6). Non-scientific protocols usually define sample mass and number of samples,
where an individual mass may range between 1 kg and 6 kg. The mass is limited in practical terms
by health and safety issues, as well as the capacity of the laboratory to process samples within a
reasonable
Minerals 2018, timeframe.
8, 232 The number of samples collected is generally arbitrary and often constrained
22 of 45
by laboratory capacity.

Figure 6. Grab sampling from a surface stockpile showing the challenges of collecting a representative
Figure 6. Grab sampling from a surface stockpile showing the challenges of collecting a representative
sample and oversampling fine material.
sample and oversampling fine material.

4.3. Grab Sampling in Block Cave Mines

4.3. Grab Sampling in Block Cave Mines
Block cave operations have been traditionally reliant upon drawpoint grab sampling for grade
Block cave operations have been traditionally reliant upon drawpoint grab sampling for grade
control [52]. Similar to other large mines, drawpoint sampling is used to monitor grade over a period of
control [52]. Similar to other large mines, drawpoint sampling is used to monitor grade over a
Minerals 2018, 8, x FOR PEER REVIEW 21 of 43
period of time. When the plant is reconciling to the predicted feed grade, then the samples are not
important.
time. WhenHowever,
the plant when reconciliation
is reconciling to the becomes
predictednegative,
feed grade, they canthe
then be used
samplesto track grade
are not across
important.
active
However,drawpoints.
when reconciliation becomes negative, they can be used to track grade across active drawpoints.
The
The question
question of of representativity
representativity is is aa continual
continual point
point ofof disagreement
disagreement between
between practitioners.
practitioners.
Drawpoint
Drawpoint grab grab sampling
sampling focuses
focuses onon the
the fraction
fraction ofof material
material that
that is
is small
small enough
enough to to be
be picked
picked upup by
by
hand and put into a sample bag. This introduces a bias (e.g., DE and EE) due to the
hand and put into a sample bag. This introduces a bias (e.g., DE and EE) due to the fact that no samplefact that no sample
is
is taken
taken from
from thethe coarse
coarse size
size fraction
fraction (>15
(>15 cm),
cm), which
which isis relevant
relevant when
when ores
ores have
have aa demonstrated
demonstrated
characteristic
characteristic for increasing grade in the finer fractions. A drawpoint may produce 200 ore
for increasing grade in the finer fractions. A drawpoint may produce 200 t of t of per
oreday
per
and be sampled every 10 days, which equates to 10 kg of sample for every 2000
day and be sampled every 10 days, which equates to 10 kg of sample for every 2000 t [52]. A grid t [52]. A grid cancan
be
drawn onto the rill surface indicating where samples are to be taken in an attempt
be drawn onto the rill surface indicating where samples are to be taken in an attempt to spread sample to spread sample
coverage
coverage (Figure
(Figure7). 7).

Figure 7. Drawpoint with sampling grid (Left) and sample collection (Right).
Figure 7. Drawpoint with sampling grid (Left) and sample collection (Right).

The use of grab sampling in block cave mines, whilst problematic, provides management with
The use of grab sampling in block cave mines, whilst problematic, provides management with
information which allows them to make decisions where the grade appears to be reducing. Future
information which allows them to make decisions where the grade appears to be reducing. Future very
very large-scale caving operations will require a new approach to production sampling to meet safety
large-scale caving operations will require a new approach to production sampling to meet safety and
and health expectations. The numbers of draw points (potentially up to 1000) combined with the use
of more automation mean that a faster and smarter sampling will be required to meet the demands
of bigger mines [52].

4.4. Grab Sampling and Errors

Minerals 2018, 8, 232 23 of 45

health expectations. The numbers of draw points (potentially up to 1000) combined with the use of
more automation mean that a faster and smarter sampling will be required to meet the demands of
bigger mines [52].

4.4. Grab Sampling and Errors

4.4.1. Error Generation

The efficacy of grab sampling has frequently been questioned due to the presence of large size
distribution biases in the method and hence high FSE and GSE [51]. Bias can be exacerbated due to
the natural tendency of the sampler to be drawn to richer fragments or to the fact that fines are often
enriched in gold and/or sulphides. Another challenge relates to the fact that muckpiles are rarely
mixed; therefore, the collection of samples from the pile surface is never going to be representative
of the block of ground from which it was drawn. This problem is exacerbated by the fact that in
development headings, detonator time delays eject material sequentially from the face, initially from
the centre, and finally from the edges. Thus the muckpile will be zoned with the last material blasted on
the top. Samples collected from trucks, tram cars, or stockpiles/dumps may be more “homogenised”.
In long-hole stoping, the last material to be blasted will generally be from the hangingwall and
footwall of the stope, and may be entirely dilution (i.e., low grade or barren). Diluting blocks are likely
to be relatively large, but will be under-sampled. In narrow ore bodies extracted by the long-hole
method, minimum stoping widths may require a significant amount of edge dilution, which will
generally end up on top of the pile. In cut-and-fill operations there is a risk of cement fill entering the
ore stream. If this is the case, then it should be included in any grab samples since it is part of the mill
feed and its non-collection is an EE.
High or low-grade material may preferentially segregate in the pile during mucking.
Additional issues result from the use of water underground to suppress dust. Sprinkler systems
or water jets spraying onto a muckpile exacerbates segregation of fines to the pile base. In some cases,
spraying can lead to gold loss into drive or stope floors. The same is true for surface stockpiles where
water is used to reduce dust or where there is high rainfall.
In order to obtain a more reliable sample, a large quantity of broken rock (up to 100 kg or
more) will usually have to be collected on a day-to-day basis, which is generally impractical in a
mining situation. Sampling needs to be throughout the pile volume, which is again impossible, and
hence tram car sampling is a preferred, but still not optimal option. Some correlation generally exists,
whereby large fragments are enriched or depleted with respect to gold. The sample mass that should
be ideally collected is dependent on the size of the largest fragments in the pile and on the nature of
the contained mineralisation (e.g., coarse versus fine gold content). The coarser the rock fragments
and the coarser the gold, the larger the sample mass required.

4.4.2. Comparison Between Contrasting Ore Types

A comparison between two contrasting ore types is shown in Table 16, illustrating that in some
cases, grab sampling can work relatively well. The two deposits have similar head grades, but contrast
as low-nugget no-coarse gold (Mine I) and high-nugget dominant-coarse gold (Mine II). Notably the
field sampling error component for the two deposits is very different at 27% and 68%, respectively.
As a result, the grab sample-derived grade estimate for Mine I is good (107%), whereas for Mine II is
very poor (227%). This difference principally relates to the granulometry and gold particle distribution
of the ore types. In Mine I, gold is relatively evenly distributed through the muckpile and does not
differ substantially from one fragment size range to another. However, in Mine II, there is a dominance
of gold in the sub-2.5 cm fragment fraction, and thus the grab samples are biased since the +2.5 cm
fraction is not represented properly.
Caution is required with the interpretation of duplicate grab sample analysis. In one example,
duplicate analysis indicated low variability of the sampling component (e.g., ~25%), but this masked
Minerals 2018, 8, 232 24 of 45

the fact that the samples were representing the <3 cm fraction well, but not the coarser >3 cm fractions.
In this case there was low heterogeneity in the high-grade fine fraction related to the consistent
liberation of gold-bearing sulphides. However, this fraction contained around 50–65% of the
total gold. Full characterisation, including a granulometric study is required to understand grab
sample representativity.

Table 16. Examples of grab sample performance from two contrasting mineralisation styles from South
America [42].

Parameter Mine I Mine II

Characteristics of Mineralisation
Mineralisation style Epithermal vein Mesothermal vein
Gold deportment Free gold disseminated through quartz Free gold within quartz and pyrite
Percentage > 100 µm 0 45
Liberation diameter (µm) 50 850
Sampling constant (K) 90 7400
Nugget effect (from drilling) (%) 17 72
Granulometric Characteristics 1
Screened tonnage (t) 51 54
−3 mm grade (g/t Au) 10.1 49.4
+3–6 mm grade (g/t Au) 9.6 36.9
+6–12 mm grade (g/t Au) 9.2 28.6
+12–25 mm grade (g/t Au) 9.0 19.7
+25–50 mm grade (g/t Au) 8.8 4.6
+50 mm grade (g/t Au) 8.4 2.2
Weighted lot mean grade 9.2 12.2
Grab Sample Duplicate Analysis 2
Field component error (%) 27 68
Preparation component error (%) 10 26
Analytical component error (%) 7 8
Total error (%) 30 73
No. of pairs (each of field, preparation and
50 50
pulp duplicates)
Test Stope Reconciliation 3
Grab sample mass (kg) 3–4 4–6
Sample location Underground re-muck pile Surface stockpile
No. grab samples collected 285 325
Stope tonnage (t) 8000 9400
Plant reconciled head grade (g/t Au) 9.0 10.9
Grab sample mean grade (g/t Au) 9.6 24.7
Grab/mill head grade (%) 107 227
Grab sample RSV (%) 75 335
1 Granulometric tests undertaken on an individual development round. Each lot was screened and then crushed in
its entirety to P90 −5 mm and 100 kg split off by rotary sample divider to yield an FSE of ±20% at 68% reliability.
Each sample was submitted to a laboratory where it was ground to P90 −2 mm and 25 kg split off by rotary sample
divider to yield an FSE of ±20% at 68% reliability. The entire 25 kg was cyanide leached. 2 Analysis was undertaken
on a number of duplicates taken from the test stope programme.3 Granulometric and duplicate studies based on
stope ore.

4.5. Summary
The key issue with grab sampling is the size of primary sample required to be representative.
The few kilogrammes that are usually collected are generally inadequate and leads to a large FSE,
potentially to ±500% or more. This error may be reduced by collecting multiple small samples over a
pile, for example fifty 5 kg samples giving a composite of 250 kg. Pitard [23] suggests that a minimum
of 32 individual samples should be taken from a pile to reduce the GSE. In the most challenging
of cases (e.g., some coarse gold systems), it is possible that hundreds of kilogrammes or tonnes are
required for each sample composite. Where large samples are collected, the best option is sample
size and fragment reduction via a sampling tower. The FSE equation can be used to optimise sample
reduction, where Equation (1) can be rearranged to solve for MS .
The method is prone to high sampling errors (Table 17); key attributes are given in Table 18.
Grab sampling may be effective where grades vary little between size fractions (e.g., fine disseminated
Minerals 2018, 8, 232 25 of 45

gold that is locked in sulphides; Table 16) and where both the ore and waste break into pieces of
approximately equal size. In many gold mines, despite best efforts, the geologist should expect
precision to be poor and ore/waste misclassification high.
There are situations when grab sampling is the only way to evaluate grade, for example where
there is no drilling and/or development cannot be safely accessed. Grab sampling is known to
work sometimes in low-nugget effect mineralisation styles (e.g., disseminated gold deposits), but in
heterogeneous coarse gold systems a positive bias is likely given that the fine fraction is generally
enriched in gold (Table 18) [39]. If grab sampling is the only option, then gold particle characterisation,
granulometric, and duplicate programmes must be undertaken to determine representativity.

Table 17. Sampling errors and their impact on grab sampling.

Sampling Error 1 Error Magnitude Error Effect During Grab Sampling

Relates to collection of a sample that is generally too
FSE High
small for the lot and is biased to the finer fractions.
The muckpile bears a natural segregation particularly
GSE High
of fines that settle to the pile base.
Routine grab samples only sample the surface of the
DE High muckpile. Delimited zone is not representative of the
pile. Large tracts of the pile go effectively unsampled.
The poorly delimited sample does not extract all the
EE High rock fragments within the sample zone due to
differential fragment sizes.
Consistent mass usually collected, though can be
WE Moderate
variable from sample to sample.
1 Indicative error magnitude: Red: high (>±15%); Orange: moderate (±5–15%).

Table 18. Grab sampling attributes.

Attribute Comment
Logistics and planning Easy: relatively easy to undertake.
Frequency of collection High: numerous small (<6 kg) samples can be collected.
Low–high: Depends upon sampling location; risk with multiple
small samples as composite high mass results in manual handling
Hazard exposure issues. Sampling of surface or underground stockpiles may pose
traffic hazards. May require sampler to be at the rock face if
collecting from a muck pile.
Flexibility of method High: can be undertaken quickly and at relatively short notice.
Low: relatively easy and quick to collect, cost principally sample
Cost
preparation and assaying.
Sample quality Poor: prone numerous sampling errors.
Value proposition Low: strong likelihood of bias, and thus ore/waste misclassification.

5. Drilling Methods

5.1. Diamond Core Drilling

Diamond core samples are generally the highest quality samples that can be produced when core
recovery is above 85% (e.g., low DE and EE). Core can be oriented spatially in 3D for geological and
rock mass characterisation. Coring in broken ground may require more expensive large-diameter core
and triple-tube drilling techniques to achieve acceptable recoveries. Similarly, core drilling in swelling
ground may require special mud to achieve higher recoveries. Core has a small relative volume (e.g.,
ranging from BQ 3 kg/m to PQ 15 kg/m) of material and may result in locally unrepresentative samples.
The method is relatively slow and expensive, which may pose some restrictions on close-spaced drilling.
Minerals 2018, 8, 232 26 of 45

Diamond core drilling is used in a number of mines for grade control and short- to medium-term
planning purposes [7,40,53–55].

5.2. Reverse Circulation Drilling

Reverse circulation (RC) drilling technology is now available to underground mines [56].
It provides a larger volume (e.g., 35–45 kg/m) of sample than drill core in a relatively shorter time
(7–10 m/h) and at lower cost. Holes can be angled orthogonally to the mineralisation for better
geometric representation. Chip sizes are relatively small and consistent. However, sample recovery in
poor ground conditions may be inadequate (high EE). Reverse circulation cuttings can provide useful
information, but lack the detailed structural component which is available with core. Care needs to
be taken to ensure correct sampling equipment whilst splitting a sub-sample at the rig (minimisation
of DE and EE). Despite producing a relatively good quality primary sample, a negative of RC is that
a large FSE can result where a small sub-sample is taken from the primary sample. Thus, a 5 kg
sub-sample from an original 1-m 40 kg composite from a medium-grade (3–4 g/t Au) coarse gold ore
could yield an FSE of greater than ±40%. Duplicate sample analysis provides an appropriate way to
empirically evaluate any RC protocol [56].

5.3. Sludge Drilling

Sludge or blasthole drill sampling involves using a jumbo, long-hole, or air-leg drills to create
open holes and collect the cuttings in catchers or strategically placed containers [57]. Sample mass can
vary from 2 kg/m for air-leg drills up to 12 kg/m for long-hole rigs.
The traditional sludge sample paradigm is the collection of sample material per rod length (e.g.,
1.2 m to 1.8 m) per hole (Figure 8). However, a small number of operations collect cuttings from an
entire blast ring that may yield around 2–3 t of sample. Collection of the primary sample is a challenge,
but can be achieved via a hole collar device feeding to a hopper or simply placing the hopper under
the hole whilst drilling. The primary sample mass then needs to be reduced, which can be undertaken
via a splitter unit located underground or on surface. For a 2–3 t primary sample, a 200–500 kg split
is potentially required to achieve an FSE of ±20%, though this depends on the nature of the gold
and grade. The sub-sample can be processed via a laboratory-scale gravity plant or split down further
for multiple bulk cyanidation. Such an approach has been applied to high nugget mineralisation
Minerals 2018, 8, x FOR PEER REVIEW 25 of 43
extracted by long-hole stoping [42].

(A) (B)

Figure 8. Types of sludge sampling. (A) Collection of stope blasthole sludge via plastic boxes placed
Figure 8. Types of sludge sampling. (A) Collection of stope blasthole sludge via plastic boxes placed
below
below the
the hole
hole leads
leads to
to very
very high
high DE
DEand
andEE;
EE;(B)
(B)Collection
Collectionof
offace
faceblasthole
blastholesludge
sludgevia
viascooping
scoopingfrom
from
the drive floor leads to high FSE, GSE, DE, and EE.
the drive floor leads to high FSE, GSE, DE, and EE.

A small number of operations have practiced the collection of sludge samples from development
faces by scooping cuttings from the floor of the level (Figure 8). This method is prone to very high
sampling errors and best described as a gambling process. In all sludge sampling exercises, there are
usually problems with loss of fines and the sample being biased by contamination from the sides of
the hole. Care needs to be taken whilst splitting a sub-sample to minimise DE, EE, and PE. Like RC,
sludge drilling produces a large primary sample, though generally of low quality. A similarly large
Minerals 2018, 8, 232 27 of 45

A small number of operations have practiced the collection of sludge samples from development
faces by scooping cuttings from the floor of the level (Figure 8). This method is prone to very high
sampling errors and best described as a gambling process. In all sludge sampling exercises, there are
usually problems with loss of fines and the sample being biased by contamination from the sides of
the hole. Care needs to be taken whilst splitting a sub-sample to minimise DE, EE, and PE. Like RC,
sludge drilling produces a large primary sample, though generally of low quality. A similarly large
FSE can result, where a small sub-sample is taken from the primary sample.
Ovalle et al. [58] propose the “Sampdrill” method which utilises a hand-held drill to place multiple
small diameter holes into the drive face or walls. The cuttings from each hole are then taken as an
individual sample or composited to a larger sample. This methodology was applied to porphyry
copper mineralisation, but is less likely to be relevant to gold unless it is fine-grained and disseminated.
Sludge sampling is often used to check where the ore zone is greater than the drive width. In other
cases, it is used to check that blastholes are in ore via assay and/or visual inspection of channels.
Sludge-based assays are generally not used to support resource estimates.

5.4. Comment
A reduction in the need for linear samples results from the use of more pre-development drilling
at a spacing to allow local estimation [53,55]. McBeath and Whitworth [55] present a case study
where drilling was reduced to 5 m to improve ore/waste definition in a geometrically complex and
high-nugget effect ore body. Both diamond core and RC drilling can be used underground, with RC
drilling providing a relatively quick and cheap alternative.

6. Other Sampling Methods

6.1. Bulk Sampling

In the underground context, bulk samples are used to assess grades in complex deposits and
also provide geological, geotechnical, and geometallurgical information. The term bulk sample is not
rigorously defined, but can be taken as the collection of a series of large (often >1 t) samples across a
mineralised zone for test work. They are typically undertaken during pre-feasibility and feasibility
studies and are seen as an insurance policy by verifying drill- or linear sample-based grade estimates
and/or to support metallurgical pilot test work [7,29,59]. Bulk sampling is generally not relevant to
routine grade control activities. However, some high-nugget coarse-gold operations have collected
micro-bulk samples (100–250 kg) via face panel sampling (Figure 5) for grade control [12,60,61].

6.2. Conveyor Belt Sampling

So-called belt sampling is not generally relevant underground, though may be locally important
for grade control activities as a measure of plant feed grade. Where undertaken, belt sampling must
apply all TOS principals and in particular, a correctly designed automatic sample cutter should
be employed [5]. The manual scooping or grabbing of samples from a belt is inappropriate and
is dangerous. The FSE equation (Equation (1)) can be used to optimise the mass of sample required for
the given feed granulometry. Cutters must collect a complete cross-section of the stream, with both the
leading and trailing edges of the cutter clearing the stream at the end of each traverse. In addition, the
cutter must be non-restrictive and self-clearing, fully discharging each increment without any reflux,
overflow, or hang-up [5]. Falling stream cutters are preferable to direct belt cutters. Poorly designed
and operated cutters are prone to high DE and EE, and the sample mass and number of increments
should be optimised to the feed rate to ensure a low FSE.

7. Sample Preparation and Assay

In some cases, a mine will have its own laboratory which speeds up the turnaround time for assays.
Where samples inform a resource estimate, it is preferable for the laboratory to be operated by an
Minerals 2018, 8, 232 28 of 45

independent and accredited third party. If the laboratory is run by the mining company, then it is
critical that it be independently audited on a regular basis and that umpire assays are undertaken as
part of its QAQC programme.
Sample mass and fragment size reduction must be undertaken within the framework of the TOS;
the FSE equation can be applied (Equation (1)). Particular care is required in the presence of coarse
gold, with possibilities for gold loss and/or contamination [24,30,32,54,59,62]. Broken rock samples
from the mine (e.g., a face channel sample) must be crushed prior to splitting and pulverisation to
avoid high DE, EE, and GSE. Large pulverisers such as the LM5 (maximum capacity 3 kg) are useful;
however, removal of the pulp needs to be maximised (EE) and contamination controlled (PE). It is
recommended that laboratories include a barren flush after visible gold-bearing samples. However,
if the ore bears coarse gold, then flushing after each sample is optimal. Assaying of the pulveriser
wash (e.g., sand) is recommended at a rate of 1 in 10–20 for coarse gold ores. DE and EE are potentially
high during sample mass reduction, and a Jones riffle splitter or rotary sample divider should be used.
Mat-rolling and grab sampling from the pulp must be avoided (DE, EE, and GSE) [1,4,24].
The traditional paradigm of crushing and pulverising the entire sample (e.g., chip or channel
sample, or half drill core) and taking a 30–60 g sub-sample for fire assay is potentially flawed [4,7,17].
The approach is prone to high CSE and ISE, particularly when the assay charge is scooped from
the pulp. The propensity of gold not to pulverise efficiently (exacerbated in the presence of coarse
gold), promotes FSE and GSE, and potentially loss or contamination (e.g., PE) during sub-sampling for
the fire assay charge.
The use of an entire sample for total assay effectively yields FSE and GSE values
of zero [7,12,54,59,63]. With good laboratory practice, the PE and AE can be minimised.
Arguments against whole core sampling in particular, revolve around no reference core remaining for
re-logging, check sampling, etc. However, the authors contend that with modern digital photography,
detailed logging and internal and/or external peer review that this should not be an issue. A reference
sample can be split by rotary sample divider after crushing.
There are a number of techniques employed to assay samples for gold (Table 19); these include
traditional fire assay, and large charge methods such as screen fire assay, LeachWELL, pulverise and
leach, and whole sample by gravity. The LeachWELL method is particularly suited to gold assaying,
as samples of up to 5 kg can be leached [64]. LeachWELL is a catalyst that increases the dissolution
rate of gold, where samples can potentially be leached in less than 24 h. The gold laden solution is
then analysed via an instrumental method such as atomic adsorption spectrometry. LeachWELL is not
a total assay, therefore evaluation of its performance on a given ore type is required. In most cases, the
leach tails are subjected to a duplicate fire assay to provide a total gold grade.
Sample protocols should not be left solely to the laboratory to decide, but established by the project
team based on characterisation. In the absence of duplicate samples, the FSE equation (Equation (1))
can be used to evaluate protocol precision across different ore types. Examples are given in Table 20
for a series of protocols based on a fine- and coarse-gold ore types. It must be remembered that the
FSE equation only evaluates the FSE; an acceptable FSE value can be swamped by high GSE, DE, EE,
etc., thus making it inappropriate for use.

Table 19. Summary of gold assay methods.

1 Assay Type Assay Charge Nature of Method Application Outcome

Small charge mass. Poor
The standard method of fire
Fire assay (FA) 30–50 g reproducibility in the presence Grade
assaying for gold.
of coarse gold.
A variant of FA, the SFA reduces
the coarse-gold problem by An effective method for dealing
Screen fire Grade
0.5–2 kg sieving out the coarse fraction with coarse gold. Can be used
assay (SFA) Per cent coarse gold
(100–150 µm screens) and on samples up to 10 kg.
assaying it separately.
Minerals 2018, 8, 232 29 of 45

Table 19. Cont.

1 Assay Type Assay Charge Nature of Method Application Outcome

Large charge mass. Effective
Method overcomes the problem
method for dealing with coarse
of coarse particles by assaying Grade
LeachWELL gold. The method needs to be
0.5–5 kg the entire sample. Leach recovery (if tails
(LW) controlled by assaying the
24 h leach time effective for assayed)
undissolved residues to check
most coarse gold deposits.
for partly dissolved gold.
Smaller charge mass. Potentially
effective method for dealing
Essentially same as LeachWELL. with coarse gold. Some issues
Crushed (approx. P80 −10 mm) relate to contamination of Grade
Pulverise and
1 kg sample is leached and pulverise/leach pots. Method Leach recovery (if tails
leach (PAL)
pulverised at the same time. needs to be controlled by assayed)
1–1.5 h leach time. assaying the undissolved
residues to check for partly
dissolved gold.
Novel method based on
Gamma bombarding the sample with Moderate charge mass. Sample
activation high-energy X-rays to produce preparation requires just coarse
0.5 kg Grade
(Chrysos gamma rays, which are crush. Reported 1–2 min
Photon Assay) measured across a given energy analysis time.
for gold.
Very large charge mass.
Effective method for dealing
with coarse gold. Method needs Grade
Takes a large sample and
Whole sample to be managed to ensure GRG (single-stage)
processes entirely via gravity
gravity maximisation of gold recovery. Leach recovery (if tails
>5–500 kg (e.g., Knelson and/or Wilfley
processing Not useful for fine gold, unless leached)
table) to produce a gold
(GRAV) gravity used for gold-rich Gold particle size profile (if
concentrate and tails for assay.
sulphide concentration and/or GRG screened)
clustered fine gold is present,
hence larger sample warranted.
1 Assay type (e.g., LW) with charge size (e.g., 2000 g) given as LW2000. GRG: gravity recoverable gold.

Table 20. Sensitivity of the FSE and sampling errors based on ore type (grade and liberation diameter)
and sample protocol.

2 FSE (%) 3 Total Error

1 Preparation-Assay Protocol Comments
Fine Gold Coarse Gold Fine Gold Coarse Gold
Coarse gold FSE range reflects
(1) 2.5 kg crush to P80 potential efficiency of the
−3 mm pulverising process, where
(A) (2) Pulverise to P90 ±5 ±8–30 coarse gold (>100 µm) may Low High
−75 µm, scoop off 30 g remain after pulverising
(3) Take for FA30 Scooping of 30 g from pulp will
promote DE, EE and GSE
High FSE in both cases
(1) 2.5 kg crush to P80 indicates splitting at P80
−3 mm, riffle split off 500 g −3 mm
(B) ±17 ±165 Mod High
(2) Pulverise to P90 −75 µm As entire 0.5 g sample is
(3) Take for LW500 pulverised and LW500, then
minimal errors at this stage
Scooping of 0.5 kg from 2.5 kg
(1) 2.5 kg crush to P80 of pulp will promote DE, EE
−3 mm and GSE
(2) Pulverise to P90 Splitting of 0.5 kg from 2.5 kg
(C) ±1 ±2–10 Low High
−75 µm, riffle split off must be undertaken using a
0.5 kg rotary splitter, particularly in
(3) Take for LW500 the coarse gold case to
minimise DE, EE, and GSE
Approach minimises all
(1) 2.5 kg crush to P80 sampling errors including FSE
−3 mm Splitting of 2 kg from 2.5 kg
(D) (2) Pulverise to P90 ±1 ±1–5 should be undertaken using a Low Low
−75 µm, riffle split off 2 kg rotary splitter, particularly in
(3) Take for LW2000 the coarse gold case to
minimise DE, EE, and GSE
1 Based on gold grade of 6 g/t Au; fine gold liberation diameter 50 µm (K = 115 g/cm1.5 ); coarse gold liberation
diameter 1000 µm (K = 10,200 g/cm1.5 ); Francois-Bongarçon modified FSE equation (Equation (1)) applied using
default alpha at 1.5 and 68% reliability; sampling errors as defined in Table 2; assay methods as defined in Table 19.
Protocols expedited with good practice and QAQC (Table 21). FA30: 30 g fire assay; LW500-2000: LeachWELL
500 g to 2000 g. 2 FSE error definition: Red: high (>±30%); Orange: moderate (±20–30%); Green: low (<±20%).
3 Indicative total error: Red: high (>±50%); Orange: moderate (±30–50%); Green: low (<±30%).
Minerals 2018, 8, 232 30 of 45

8. Quality Assurance/Quality Control and Error Quantification

8.1. Quality Assurance and Quality Control

Quality assurance and QC are the key components of a quality management system [35]. Quality
assurance is the collation of all actions necessary to provide adequate confidence that a process (e.g.,
sampling, test work, and assaying) will satisfy quality requirements. While QA deals with prevention
of problems, QC aims to detect them. Quality control procedures monitor precision and accuracy of
data, as well as possible sample contamination during preparation and assaying. Throughout any
grade control programme, QAQC is a key activity to determine fit-for-purpose samples [17]. Some
mines do not apply QAQC to grade control samples as they are not included in resource estimates;
however, this attitude is generally flawed, given that ore/waste decisions are based on the samples.
Where grade control samples inform a resource estimate, then QAQC is an absolute.
Protocols should be set up to cover field collection, laboratory preparation, and analysis (Table 21).
During grade control, QAQC should include field duplicates and certified reference material (CRM)
submission (e.g., minimum three CRMs at a range of grades including blanks). Laboratory QAQC
will include internal CRMs, pulp duplicates, umpire sample submission, pulp screen tests, and
contamination tests. In particular, duplicate field samples provide a measure of variability of the entire
sampling process.
Documentation of sample collection and laboratory activities is an important part of QAQC,
and staff training and monitoring. It is the opinion of the authors that quality samples flow from
well-trained and experienced personnel. Operations should ensure that all grade control staff are
trained in sampling and during their first few months have adequate mentoring. This will be additional
to all standard operational and safety training. Proper training will be facilitated by well-written
and illustrated documentation. Training and mentoring should be linked to continuous quality
improvement, where protocols are internally and externally audited at least annually. On-going
supervision and periodic re-training are recommended, where activities must be practical-based at the
rock face and not simply in the classroom.

Table 21. Best practice QAQC for a grade control sampling programme that informs a resource estimate.

11 Key Performance Indicators

QAQC Action Rate Instigator
Fine Gold/[Coarse Gold]

1 90% ±10–25% HARD 10 90% ±13–35% RSV

Field duplicates 1 in 20 Operation
[90% ±25–50% HARD] [90% ±35–70% RSV]

2 Depends upon sample type; for saw-cut channels >80%; for

Sample quality index All Operation
diamond drill core >85% recovery
3 Coarse reject 90% ±10–20% HARD 90% ±13–28% RSV
1 in 20 Laboratory
duplicates [90% ±20–50% HARD] [90% ±26–70% RSV]

4 90% ±10% HARD 90% ±13% RSV

Pulp duplicates 1 in 20 Laboratory
[90% ±10–20% HARD] [90% ±13–28% RSV]
<2δ (“safe zone”) no action required
5 Certified Reference Operation and 2δ–3δ (“warning”) investigate (re-assay 25% of batch if
1 in 20
Materials laboratory required)
>3δ (“action”) re-assay 100% of batch

6 Operation and
Blanks 1 in 20 Less than 0.05 g/t Au
laboratory
7 Pulp quality 1 in 20 Laboratory 95% to be P95 −75 µm
8 Barren flush 1 in 20–50 Laboratory <<0.5% gold loss
9 Umpire 1 in 20 Operation 90% ±10% HARD 90% ±13% RSV
Laboratory audit Quarterly Operation Full adherence to agreed practices and performance levels
Minerals 2018, 8, 232 31 of 45

Table 21. Cont.

11 Key Performance Indicators

QAQC Action Rate Instigator
Fine Gold/[Coarse Gold]
Operation and
QAQC review Monthly Compliance across all metrics
laboratory
1 Applies to any sample type collected. 2 Applied to linear and drill samples; key performance indicators (KPIs)
based on sample type and expected mass. 3 Laboratory crusher or RC rig rejects. 4 Dependent upon nature of
ore and assay method. For samples assayed via screen fire assay (SFA), a high precision would be expected for
undersize fraction. 5 Recommendation to have a minimum of three CRMs at grades ranging from cut-off, ROM, and
high-grade. For any batch of (say) 20–30 samples, the three key CRMs should be added. Note that by their very need
to be homogeneous, CRMs do not bear coarse gold, but they can be matrix matched by being quartz-dominated,
sulphide-bearing, etc. The laboratory will also insert its own CRMs. CRMs used for SFA process will just be fire
assayed. Action is required if 3δ breached, usually re-assay of the entire batch if possible. 6 Blanks provide a
measure of contamination. They should be inserted after expected high-grade and/or visible gold-bearing samples.
If substantial visible gold is present, two separate blanks should be placed after the sample. One blank should
be added together with the three CRMs per batch. Laboratory will also place blanks into the sample stream. 7
Test involves screening (or use of an autosizer) of the pulp to ensure 95% passing. All samples should pass or
the entire batch should be reground. 8 Barren flush may be inserted after each and every sample for coarse gold
samples. Assaying of the barren flush; for fine gold ores, a rate of 1 in 50 is appropriate increasing to 1 in 20 for
coarse gold ores. Careful management of coarse gold ores is required. It is suggested that laboratories include
a “wash” after visibly high-grade (e.g., visible gold-bearing) samples. However, if the ore bears notable coarse
gold, then cleaning is best after each sample given that even low grade samples can bear coarse gold particles. 9
Monthly submission of samples (typically pulps), including standards and duplicates is sufficient to provide a
check of primary laboratory results. This is especially important where an on-site laboratory is being used as it
provides independent confirmation of the results. Where SFA, LW, or PAL is used, there may be no pulp residues
to submit. In this case, coarse rejects can be used. Umpire samples (e.g., pulps or coarse duplicates) should be
returned to the mine and submitted by mine staff to the umpire laboratory. In some cases, the laboratory (mine
or off-site) may submit umpire samples as part of their internal QAQC. 10 HARD is half the absolute difference
of the pair divided
√ by the pair mean; HARD value for fine versus [coarse gold]; HARD can be expressed as RSV,
where HARD = 2/2.RSV, e.g., ±10% HARD is ±13% RSV. 11 It is important to ensure that enough QA data is
collected, particularly during a small sampling programme. The rate of insertion of CRMs, blanks, etc. may need to
be increased beyond the nominal 1 in 20 to achieve a minimum of 10 results.

8.2. Chain of Custody

Chain of custody (COC) procedures are particularly relevant if samples inform a resource estimate.
Chain of custody must provide a documented, legally defensible record of the custody of samples
from collection through analysis [35]. A sample is considered to be in custody if it meets at least one of
the following conditions, where the sample is: in someone’s physical possession or view, secured to
prevent tampering, or secured in an area restricted to authorised personnel.
Chain of custody procedures in the underground environment are more difficult than for
surface-based activities. The greatest risk of sample tampering is during collection, and prior to
the sealing of sample bag(s) or core trays. At the mine face, samples should be bagged and sealed with
a wire tie. Drill core trays can be covered with a metal lid and strapped. Samples and trays should be
returned to surface as quickly as possible via the most efficient means.
A sample or core record sheet can be used as a COC document to track sample progress from
face/rig, to core shed and laboratory. Face, grab, or core samples will need to be securely transported
to the laboratory, which may be either onsite or offsite. In either case, the COC document can be used
to track progress and record handovers. Where samples have a journey to an offsite laboratory, then
transport should be secure.

8.3. Errors along the Sampling Value Chain

The results of duplicate sampling programmes document the magnitude of errors across the
sampling value chain (Tables 1–4). These generally show that a large component of the total error
is introduced during sample collection. As a result, undertaking excessive efforts to reduce errors
during preparation and analysis will not always result in a substantive error reduction. In contrast, the
collection of larger high-quality field samples will result in error reduction providing that protocols
are optimised appropriately.
Minerals 2018, 8, 232 32 of 45

Test work from a narrow vein deposit exemplifies the impact of sampling error through the
comparison of chip and channel samples (Table 22). Seventy-five sample triplicates (chip, hand-cut
channel, and saw-cut channel) were collected from around a 40 m by 20 m stope block. The block
was sampled from faces located every 1.5 m along its upper and lower drives and two raises. After
cleaning, a reference line was drawn across each face centre and the samples collected from the bottom
up; chip sample (7 cm wide by 1.5 cm deep), hand-cut and then saw-cut channels (4.5 cm wide by
2.5 cm deep). The sample delimitation dimensions were designed to achieve a theoretical sample
support of 3 kg/m. The mineralisation was known to have a moderate variability, containing visible
gold to 1.5 mm in size. In the test area, the vein was near vertical and its width relatively consistent,
ranging from 0.8 m to 1.2 m. Only the vein was sampled. All samples were prepared and assayed in
the same way, via a total sample preparation and screen fire assay route. The FSE for the protocol was
effectively zero. A QAQC programme was applied, with all CRMs and blanks within expectation.
The results show a marked reduction in RSV and nugget effect between the three sample sets
(Table 22). The rigorous laboratory protocol and QAQC indicate that errors within the laboratory were
at a minimum. Therefore, the remaining variability relates to the geological nugget effect and sample
collection. The dominant error for the channel samples relates to the geological nugget effect, given
that sampling error was minimal. The dominant difference between the chip and saw-cut channel
samples relates to sampling error. The results corroborate previous findings, showing that saw-cut
channel samples provide the best sample quality.

Table 22. Example from an Australian mine showing comparison between chip and channel sample
replicates. Total number of samples in each type was 75. Mean grades cut and diluted to stope width.
Reconciled stope head grade 13.7 g/t Au.

Mean Target 1 Sample

Mean Percent of Nugget
Support Support (%) 2
Type Grade Mined RSV (%) Effect Collection Preparation and Assay Errors
(kg/m) [within
(g/t Au) Grade (%) (%) Errors
[Range] ±5%]
High:
2.1
Chip-channel 22 21.3 +55 198 68 DE, EE,
[1.4–3.4] Low: entire sample crushed and
and WE
pulverised prior to total sample
Low-moderate:
Channel 2.6 screen fire assay with triplicate
63 16.7 +22 135 54 DE, EE,
(hand cut) [2.1–3.3] fine-fraction fire assay. All
and WE
equipment cleaned between samples.
Low: DE,
Channel 2.8
77 16.3 +19 81 41 EE, and
(saw-cut) [2.3–3.1]
WE
1 Indicative sample collection error definition: Red: high (>±50%); Orange: moderate (±20–50%); Green: low

(<±20%). 2 Indicative preparation and assay error: Red: high (>±20%); Orange: moderate (±10–20%); Green: low
(<±10%).

Duplicate sample analysis across the sampling and laboratory process allows errors to be
apportioned to each phase of activity [34,41]. Dominy [42] reports test work from a moderate coarse
gold-bearing (d95Au : 750 µm) vein system. A series of fifty non-contiguous underground faces were
sampled via chip, hand-cut channel, and saw-cut channel samples. A field duplicate was taken for
each, and at the laboratory coarse and pulp duplicates were taken (Table 23). The results demonstrate
the dominance of sample collection error and higher chip sample error compared to the channel
samples (e.g., reduced DE and EE). The high preparation error most likely represents liberated coarse
gold present in the crushed material. The analytical error is at the high end of acceptable, reflecting
some remnant heterogeneity from the coarse gold. The total error could be reduced to around 23% by
putting the entire saw-cut sample through a LeachWELL assay.
There a need to move towards the quantification of errors, where a first step is the application of
the RSV as defined in DS3077 [38]. Resolution of individual relative errors across sampling, preparation,
and analysis can be gained from duplicate sample pairs (Tables 23 and 24). From a spatial perspective,
variography provides a method to determine the relative proportion of in-situ/geological versus
sampling nugget effect based on duplicate field pairs.
Minerals 2018, 8, 232 33 of 45

Table 23. Example from a North American mine of RSV values for duplicates taken across each mine
face [42]. See also Table 24, Mine B example.

Sampling (%) Preparation (%) Analytical (%) Total (%)

Sample Type/Error Crushed to <1.5 mm; Pulverised to 75 µm; 50 1 m 3 kg Sample
(Preparation Route) 1 m 3 kg Field
Riffle Split into 1.5 kg g Riffle Split from to Duplicate 50 g
Sample
Sub-Samples 1.5 kg for Fire Assay Fire Assay
Chip 67 31 17 76
Channel (hand) 48 29 14 58
Channel (saw) 21 27 15 37
Estimated stage FSE - 23 18 29

Table 24. Examples from South (Mine A) and North (Mine B) America showing error comparison
between two differing styles of gold mineralisation [42].

Parameter Mine A Mine B

Characteristics of Mineralisation
Mineralisation style Epithermal vein Mesothermal vein
Mining method Underground (long-hole open stoping) Underground (shrinkage stoping)
GRG component (%) 20 55
Grade (g/t Au) 10 15
Grade RSV (%) 67 117
Nugget effect (%) 20 65
Liberation diameter (µm) 70 750
Sampling Constant (K) 115 2600
Routine sample type used Chip Channel (hand cut)
Pairwise RSV for Each Error
No. of duplicate pairs used 50 50
Chip (sampling error) 31% 67%
Channel [hand] (sampling error) 27% 48%
Channel [saw-cut] (sampling error) 25% 21%
Range of preparation errors 10–14% 27–31%
Range of analytical errors 7–11% 14–17%

9. Integration into the Mining Process

9.1. Mining Method and Grade Control

Grade control sampling strategy is related to mining method, which broadly relates to ore body
width and dip, and rock mass properties (Tables 6 and 7). Mining methods are described as being
either entry or non-entry, where entry methods allow access to the stope and non-entry methods
do not.
Entry methods such as shrinkage and cut-and-fill stoping offer opportunities for on-going
sampling during production. However, such methods are inherently more risky from a safety
perspective, and require careful risk assessment and planning before undertaking in-stope grade
control activities. These are well-suited to complex narrow (<3 m) mineralised systems that require
strong geological control to achieve some selectivity and flexibility. On-going sampling and mapping
allows monitoring of grade, placement of pillars, and reaction to changes in ore zone orientation.
Where strong mining selectivity is required, the effects of high sampling error can be particularly
severe leading to ore/waste misclassification.
Non-entry methods such as long-hole stoping and sub-level caving can only be sampled in
development drives and sub-levels, and potentially slot raises. They are therefore relatively safe
from a grade control perspective, as humans do not enter stopes. The methods are suited to wide
structures where geological control may be less critical and bulk extraction appropriate. In mass
mining operations, such as those using sub-level caving, grade control is reliant on grab sampling.
The best approach for optimal mine design is to use diamond drilling at an appropriate spacing
in advance of development. This will tend to be the case in larger operations employing non-entry
Minerals 2018, 8, 232 34 of 45

methods. Smaller, often narrow vein operations with a small resource base will be more reliant on
definition of stope blocks by linear sampling methods. For these mines, the decision as to what will be
mined has effectively already been made. A retrospective decision to process a development round
can only be made if it has been stockpiled separately. Small mines generally require selective mining,
therefore optimal sampling and integration with mapping becomes critical. A low-grade stockpile
is recommended to ensure that misclassified ore can be recovered, even if at the end of the mine
life [13,14,65].
An asset of the geologist is the ability to work and communicate with mining crew(s).
Communication needs to provide a rationale for good samples, why poor samples are problematic
and provide an explanation for the methods used. There are often excuses the geologist does not
have enough time at the face because the miners do not want them there. There is some truth in
this, though a good working relationship with crews will pay dividends. In particular, if a crew
understands why a geologist is mapping and sampling a face, then their attitude is more likely to be a
positive one. A progressive operation will include grade control familiarisation in its induction and
training programme.

9.2. Safety Aspects of Grade Control

Of overwhelming importance is the consideration of safety in the sampling of stopes, drives,
crosscuts, and raises. A primary consideration is rock mass conditions and the need for support
systems such as mesh, bolts, and/or shotcrete. Under no circumstances should samplers work under
unsupported ground unless a clear case for no support is in place. Before any rock face is sampled, it
should be reviewed by a competent person and scaled if necessary.
Some mines have already removed face sampling from the routine and replaced it by either grab
sampling or more drilling. Grab sampling is generally not an optimal alternative, though depending
upon mineralisation type could be appropriate. Drilling is a costlier option (e.g., diamond core, RC,
or sludge), though should be considered in the light of better local grade estimation and short-term
planning. In all cases, geological information can be collected back from faces via visual (mapping and
photography) or automated methods [48].

9.3. Grade Estimation

A grade control programme may produce a number of different sample types, typically linear
samples (e.g., channel) and diamond core samples. These have inherent differences in sample quality,
spatial distribution (e.g., linear samples close-spaced along drives versus a regular wide-spaced
drill grid), sample support (e.g., 2 kg/m chip samples versus 8.7 kg/m HQ core), and potentially
assay methods. Bias between different sample types can be investigated via statistical (e.g., graphically
and numerically) and spatial (variogram) analysis. If it is identified, there are options available to the
estimator, including removal of the less reliable data or factoring one of the datasets if a consistent
relationship can be defined and explained.
Grade control samples generally result in grade populations that have large numbers of low-grade
assays, long tails, and high skewness. Linear, particularly chip or chip-channel samples may be
imprecise and biased, and display variability when compared to drill samples. Their usage must be
driven by analysis and understanding of limitations. Underground mines generally collect samples
in a “block out” approach, essentially along 2, 3, or 4 sides of a stope panel. Block models should
encompass as much sample data as possible to inform a kriged estimate. Local estimates will generally
show better reconciliation with mining when based on quality drill samples at an appropriate spacing.
Recognising that linear samples may input into the short- to medium-term mine plan, it is
important that the practitioner consider all aspects of the programme. Key issues are the sampling
method(s) used (e.g., linear and/or drilling), sub-sampling techniques and sample preparation, assay
method, QAQC application and compliance, data spacing and distribution, and orientation of data in
relation to geological controls. Based on these and potentially other considerations, the practitioner
Minerals 2018, 8, 232 35 of 45

can apply an appropriate resource classification to provide a measure of this risk in the estimate
(e.g., Inferred, Indicated and Measured Mineral Resources) [9]. For example, in a high-nugget
deposit where multiple high-quality samples have been collected around a series of stope blocks,
the classification may only be Inferred, whereas exactly the same scenario in a low-nugget deposit
may yield a Measured resource. As the nugget effect increases, local estimation becomes more
difficult, and thus reconciliation becomes potentially poorer on a short-term basis [14]. Many high-
to extreme-nugget effect ore bodies will operate solely on Inferred Mineral Resources, though this is
dependent upon the data spacing available [14,54]. Moderate- to low-nugget effect ore bodies will
generally define at least Indicated Mineral Resources with Probable Ore Reserves [66].

9.4. Reconciliation
The ultimate challenge of any mine is that of achieving acceptable reconciliation between the
long-term reserves and short-term grade control model with actual plant performance. In many gold
mines, the reserve estimate overstates the achieved grade. Whilst reconciliation problems can be
related to estimation approach and/or mining, there are often underlying reasons related to sampling
error particularly bias (e.g., in-stope sample discard or soft reef bias: refer Section 3.3) and incompatible
datasets (e.g., mixing drillhole and channel sample results).
In higher nugget mineralisation, reconciliation over short periods of time (e.g., month by month)
will generally be difficult as variability may be ±50% or more [14]. Thus, longer periods of time
are likely to show more acceptable results (e.g., ±20% over three months). Operators should avoid
knee-jerk reactions such as altering cut-off grades or blending strategies based on one or two months of
poor reconciliation. Viewing a resource/reserve estimate as an “estimate” is important and operators
should understand the likely tolerances of grade and tonnage figures. In particular, they should
understand the sampling process and its effect on estimates and decision making.

9.5. Sampling Historical Mine Workings

Grade control sampling methods are also applied in the re-evaluation of historical mine workings
and may be used to support a resource estimate. Samples can be collected from the backs, faces, and
sidewalls of drives, crosscuts, raises, and stope pillars. In some cases, samples will also be collected
from drive and/or stope floors where broken material needs to be checked for potential gold content.
Care is required when accessing historical mine workings, particularly in terms of rock stability, air
quality, and open voids.
Methods are no different to operating mines, though there is often a focus on backs or sidewall
sampling given that faces are already extracted. This imposes some working at height issues depending
on drive size. Backs sampling is usually dominated by chip-channel samples given that it is difficult to
collect a sample from an uneven surface above head level (Figure 9). This leads to a potentially large
DE and EE. A tarpaulin placed on the floor immediately below the sample point generally provides a
practical solution to catching as much chipped/cut material as possible.
In some cases, legacy grade control data may be available. This will require upcycling via
validation (e.g., re-sampling/assaying), integration with modern data (e.g., new mapping and/or
other geochemical data), and interpretation/modelling. New samples may be used to validate legacy
data and support a resource estimate, in which case QAQC and COC protocols will need to be in place.
Minerals 2018, 8, 232 36 of 45
Minerals 2018, 8, x FOR PEER REVIEW 34 of 43

Figure 9. Backs channel sampling in historical workings. A chip-channel is produced, where sample
Figure 9. Backs channel sampling in historical workings. A chip-channel is produced, where sample
fragments are supposedly being collected in a plastic bowl. In this case the EE is high, as many extracted
fragments are supposedly being collected in a plastic bowl. In this case the EE is high, as many extracted
fragments are not being collected. In addition, DE on the shoulders of the drive is particularly high since
fragments are not being collected. In addition, DE on the shoulders of the drive is particularly high
the delimited samples are biased within the delimited zone on the curved drive shoulder.
since the delimited samples are biased within the delimited zone on the curved drive shoulder.

10. Future of Grade Control

10. Future of Grade Control
10.1.
10.1. Making
Making Efficiency
Efficiency Gains
Gains
The
The ultimate
ultimate grade
grade control
control programme
programme will will have
have all work at
all work at the
the face
face automated,
automated, thus
thus eliminating
eliminating
the
the need for humans to be in contact with the rock face. Automated mapping capability already exits
need for humans to be in contact with the rock face. Automated mapping capability already exits
through such systems as Sirovision
through such systems as Sirovision [48]. [48].
Sensor-based
Sensor-based technology hosted on
technology hosted on mining
mining equipment
equipment and/or
and/or conveyor
conveyor belts
belts has
has the
the potential
potential
to
to assist in this matter [67,68]. Kurth [68] reports the application of the Geoscan system which
assist in this matter [67,68]. Kurth [68] reports the application of the Geoscan system which uses
uses
prompt gramma neutron activation analysis (PGNAA) to determine the
prompt gramma neutron activation analysis (PGNAA) to determine the grade of crushed material grade of crushed material
on
on aa conveyor
conveyor belt.
belt. Analysis
Analysis can
can bebe undertaken
undertaken over
over time periods potentially
time periods potentially representing between
representing between
10 t and 40 t of feed which can then be diverted into high- and low-grade stockpiles (e.g., a bulk
10 t and 40 t of feed which can then be diverted into high- and low-grade stockpiles (e.g., a bulk ore
ore
sorting process). At the current time, gold grade cannot be directly determined
sorting process). At the current time, gold grade cannot be directly determined via PGNAA though itsvia PGNAA though
its correlation
correlation withwith other
other elements
elements (e.g.,
(e.g., CuCu and/orS)S)can
and/or canenable
enableaagrade
gradetotobe
beestimated.
estimated. Such
Such technology
technology
has the potential to improve the grade control process and also to feed into geometallurgical-based
has the potential to improve the grade control process and also to feed into geometallurgical-based ore
control.
ore control.
Hand-held X-ray fluorescence has already been used in some base metal mines for the determination
Hand-held X-ray fluorescence has already been used in some base metal mines for the
of grade at the face; however, it has yet to be applied successfully in precious metal mines. Laser-
determination of grade at the face; however, it has yet to be applied successfully in precious metal
induced breakdown spectroscopy (LIBS) has been slated as a potential contributor, though its
mines. Laser-induced breakdown spectroscopy (LIBS) has been slated as a potential contributor,
application within the mining industry is still at an early stage [69].
though its application within the mining industry is still at an early stage [69].
An automated and potentially robotic approach to sample collection that results in high-quality
An automated and potentially robotic approach to sample collection that results in high-quality
face samples under geological control would be a key gain. This would ensure that geologists do not
face samples under geological control would be a key gain. This would ensure that geologists do not
have to access the face directly and ensuring sample collection is consistent and safe. The authors are
have to access the face directly and ensuring sample collection is consistent and safe. The authors are
aware of various attempts to mechanise face sampling, but none that have been commercialised to
aware of various attempts to mechanise face sampling, but none that have been commercialised to date.
date. One observed by the primary author (S.D.) in the late 1990s, involved a grinding head mounted
One observed by the primary author (S.D.) in the late 1990s, involved a grinding head mounted on a
on a hydraulic boom that was pulled across the face to excavate a relatively consistent channel. A
hydraulic boom that was pulled across the face to excavate a relatively consistent channel. A rubber
rubber cuttings catcher device was mounted below the cutter head. At the time, the grinding head
cuttings catcher device was mounted below the cutter head. At the time, the grinding head appeared
appeared to work relatively well, though further work was required to improve sample collection. It
to work relatively well, though further work was required to improve sample collection. It is unknown
is unknown how the trials progressed.
how the trials progressed.
Recent advances in gamma activation analysis (PhotonAssay) provides for faster laboratory
Recent advances in gamma activation analysis (PhotonAssay) provides for faster laboratory assay
assay requiring less preparation and an assay mass of around 0.5 kg. Tickner et al. [70] report a sample
requiring less preparation and an assay mass of around 0.5 kg. Tickner et al. [70] report a sample
throughput of 20–80 per hour and samples presented as either crushed rock or pulps.
throughput of 20–80 per hour and samples presented as either crushed rock or pulps.
The digital mine is high on the agenda of many companies [67]. From a grade control perspective,
assay data can feed into the digital realm and can be interfaced with digital photographs, outputs
from automated mapping, and surveys. Once into the digital platform, artificial intelligence may then
allow direct feed into development control and design, and grade models. Data analytics will play a
Minerals 2018, 8, 232 37 of 45

The digital mine is high on the agenda of many companies [67]. From a grade control perspective,
assay data can feed into the digital realm and can be interfaced with digital photographs, outputs from
automated mapping, and surveys. Once into the digital platform, artificial intelligence may then allow
direct feed into development control and design, and grade models. Data analytics will play a role in
analysing grade and other data to produce more effective and agile short-term mine plans, potentially
real-time information, and lead to automated reconciliation. Sensors in drill holes, at the mine face and
in the plant, will be critical for this development.

10.2. Ore Control versus Grade Control

Grade control has a traditional focus on measuring and controlling grade outputs from the
mine. Ore control has the same aim, though provides a wider scope across grade, metallurgical, and
geoenvironmental parameters. It is a key activity of operational (or tactical) geometallurgy, which
is essentially a mine optimisation process. Geometallurgy seeks to resolve grade, metallurgical, and
mining variability based on information such as geochemistry, mineralogy, grade, and lithology
obtained from spatially distributed samples or measurement points. Key parameters that can
be obtained from ore (grade) control activities include metallurgical recovery, comminution, and
geochemistry, which are fed into the short-term block model. Grade control samples can potentially be
utilised to collect geometallurgical data during mining (Table 25). Ore control links with the digital
mine through a data-rich environment that produces real-time information, agile short-term mine
plans, and automated reconciliation.

Table 25. Potential contribution of grade control samples to geometallurgical ore control. Assumes
best practice sample collection for all samples. Green: applicable; Orange: potentially; Red: no.

In-Situ Rock Broken Rock

Sample Type/Geomet Parameter Drill Linear
Core RC Sludge Chip Channel 1 Grab
Geochemistry/mineralogy
Rock mass
Comminution
Metallurgical recovery
Geoenvironmental
1 Includes hand-cut and saw-cut channels and panel samples.

11. Public Reporting of Grade Control Sampling

The public reporting of grade control results is as important as resource drilling results when
they contribute to a resource estimate. The JORC Code 2012 has global application for the reporting
of exploration results, resources, and reserves [9]. It requires the Competent Person(s) to provide
commentary on the sampling and assaying process. When it comes to sample representativity, it is
not unusual for a public release or report to simply state: “the sampling and assaying programme
has been extensive, where representative samples were collected” or “the samples were deemed to be
representative” with little or no justification.
The JORC Code 2012 (Table 1 Section 1 (Sampling techniques and data)) [9] is aimed at the
geologist and subsequent resource model (if appropriate), even if it is based on grade control samples.
It covers disclosure on key matters such as: sampling techniques; sub-sampling techniques and sample
preparation; quality of assay and laboratory tests; verification sampling and assaying; location of data
points; data spacing and distribution; sample security; and audits or reviews.
Sampling and assaying protocols can be written in the framework of DS3077 [38] to provide an
auditable trail. Protocols can be summarised in the general format of Table 26 to provide an assessment
programme quality. Even if a resource estimate based on grade control samples is not publicly reported,
it is best practice to report at least internally using The JORC Code 2012 Table 1 framework [9].
Minerals 2018, 8, 232 38 of 45

Table 26. Example of risk review for a grade control sampling programme used to support a resource
estimate [6].

Key Parameter Comment Material Error 1 Total Error Rating

Samples collected at approx. 2 m intervals
Spatial distribution and along drives. Raises not sampled
1 GNE Moderate
number of samples Each stope block informed by around 20–25
samples
Each face composite sample approx. 2–4 kg;
total sample mass collected approx. 65–70 kg
2 Sample mass GNE High
Indicated optimum mass around 450 kg to
achieve 90% ±15%
Wall rock and central shear sampled
3 Degree of domaining - Low
separately
Samples collected by chip-channel
Uneven weighting across the face, where the
soft narrow central-shear is oversampled
4 Collection and handling EE, WE High
(approx. 5–6 kg/m) compared to the wall
rocks (approx. 2–3 kg/m)
All samples placed into calico bags and tied
No chain of custody recorded between mine
5 Transport and security PE Moderate
and off-site laboratory
Entire sample lot crushed and pulverised,
FSE, GSE, DE, EE,
6 Preparation and 30 g riffle split for fire assay High
PE
Equipment cleaning variable
Fire assay process undertaken correctly, but
7 Assay together with 6 not appropriate for coarse - Moderate
gold ore
One CRM used at high grade (70% within 3δ;
30% above 3δ)
8 QAQC Few blanks used PE, AE High
No written protocols for the
sampling-assaying process
9 Validation Plant recovers 35% less gold than estimate - High
Sample representativity (1)–(4) Poor
Assay quality (5)–(8) Poor
Fit-for-purpose rating (1)–(9) Poor
1 Indicative overall error definition: FSE: Red: high (>±30%); Orange: moderate (±20–30%); Green: low (<±20%).

12. Planning and Implementing a Grade Control Programme

Acceptance of programme value by management will be critical and requires the business case and
deliverables to be promoted early on. Recommendations to plan and implement a new or improved
underground grade control sampling programme are provided in Table 27.

Table 27. Summary of steps to achieve optimal grade control sampling programmes.

Stage Key Activity

• Present business case to support introduction or improvement

1: Scope • Aims—define what decisions grade control samples will be used (define data quality objectives)
• Understand deposit geology and grade distribution
• Undertake a characterisation programme to determine gold particle size distribution metrics

• Select a sample strategy based on geology, mining method/selectivity and safety requirements
• Consider sampling method(s)
• Ensure that sample collection, preparation, and assaying protocols are designed with reference to the ore type
2: Develop • Ensure TOS and QAQC is applied
• Document within the framework of DS3077 [38]
• Review cost-benefit of more grade control drilling to provide better grade control/local estimates and thus
negating the need for face sampling

• Ensure that staff are properly equipped and trained, with a focus on reducing sampling errors and increasing
3: Execute sample quality
• Ensure regular staff supervision and introduce a sampling quality index

• Ensure regular review of sampling strategy

4: Validate • Data and QAQC analysis
• Reconciliation with production

5: Refine • Update where required and/or return to stages 2 and 3 for further work
Minerals 2018, 8, 232 39 of 45

A five 2018,
Minerals stage8, xapproach is recommended starting with (1) scoping; moving to (2) development
FOR PEER REVIEW 37 of 43
through to (3) execution; (4) validation; and (5) refinement. Beyond establishing the business case
and aimsAoffive thestage
gradeapproach
control is recommended starting with (1) scoping; moving to (2) development
programme, a major precursor is the gold characterisation study that
through to (3) execution; (4) validation; and (5) refinement. Beyond establishing the business case and
focusses on the evaluation of gold particle size and distribution. It is important to understand the
aims of the grade control programme, a major precursor is the gold characterisation study that
mineralisation and controls to gold distribution through evaluation of geology, ore body shape and
focusses on the evaluation of gold particle size and distribution. It is important to understand the
architecture, and grade
mineralisation distribution.
and controls to goldSome mineralisation
distribution styles areofmore
through evaluation forgiving
geology, than
ore body others
shape andgiven
their architecture,
sampling characteristics (Figure 10).
and grade distribution. Some mineralisation styles are more forgiving than others given
Characterisation leads to development
their sampling characteristics (Figure 10). of the sampling strategy and selection of sampling
methods.Characterisation
Each method has leadsrelative
to development
positivesofandthe negatives,
sampling strategy and selection
which should of sampling
be considered during
methods.
scoping (TableEach method
28). Early inhas
the relative
planningpositives
processand negatives,
(Stage which
2), there mayshould be considered
be minimal during data
actual sample
scoping
available, (Table 28). duplicates.
particularly Early in the planning process
In this case, (Stage 2),by
supported there may be minimal
characterisation actualthe
results, sample data
model-based
available, particularly duplicates. In this case, supported by characterisation results, the model-based
FSE equation (Equation (1)) can be applied to evaluate sampling protocols. As the programme develops,
FSE equation (Equation (1)) can be applied to evaluate sampling protocols. As the programme
then duplicate samples permit an empirical approach to the estimation of error.
develops, then duplicate samples permit an empirical approach to the estimation of error.

d
o
C 75-100%
m
o
i
a
n
r
a
s
t
e 50-75%
e
d

d
o
25-50%
m
F
i
i
n
n
a
e
t
0-25%
e
d

[1] [2] Channel hand- Channel saw- DC & RC

Gold particle size Nugget effect Grab Chip Chip-channel Panel Sludge Bulk
cut cut drilling

Applicability to resource estimation No No [3]

Potentially Yes Yes [4]
Potentially No Yes Yes

Figure 10. Generalised applicability of sampling methods based on mineralisation type (dominance
Figure 10. Generalised applicability of sampling methods based on mineralisation type (dominance of
of coarse vs. fine gold and nugget effect). Red: avoid method; Orange: avoid method where possible;
coarse Green:fine
vs. gold applicable,
method and nugget effect). good
assuming Red: avoid method;
practices, Orange:
optimised avoidand
protocols, method
QAQC. where possible;
[1] Coarse

Green: method applicable, assuming good practices, optimised protocols, and

nominally defined as gold >100 µm in size; Nugget effect sub-divided into four ranges: low (0–25%),
[2] QAQC. [1] Coarse
nominally
medium (25–50%), high (50–75%) and extreme (75–100%); Chip-channel samples potentially appropriate low
defined as gold >100 µm in size; [2] Nugget[3] effect sub-divided into four ranges:
(0–25%), medium
if chips (25–50%),orhigh
are contiguous (50–75%)
at least and extreme
dominantly (75–100%);
semi-contiguous; and[3] Chip-channel
[4] Panel-chip samples
samples potentially
potentially
appropriate
appropriate if chips
if chips are contiguous
are contiguous or at or at dominantly
least least dominantly semi-contiguous.
semi-contiguous; and DC: diamond core
[4] Panel-chip samples
drilling;appropriate
potentially RC: reverse circulation
if chips aredrilling.
contiguous or at least dominantly semi-contiguous. DC: diamond
core drilling; RC: reverse circulation drilling.
The execution stage focusses on ensuring that staff are properly equipped and trained, with the
emphasis on reducing sampling errors and increasing sample quality. Validation and refinement
The execution
follows, led by QAQCstage focusses
analysis. on ensuring that staff are properly equipped and trained, with the
emphasisThe on investment
reducing sampling
required to errors
designand anincreasing
underground sample
gradequality. Validation
control sampling and refinement
programme is
follows, led by
variable andQAQC analysis.
dependent upon the type of mineralisation and size of operation. Indicative costs for a
characterisation
The investmentstudy are USto
required $50–150
design K per domain, thoughgrade
an underground much of the data
control can oftenprogramme
sampling be gained is
from and/or shared with metallurgical programmes [6,7]. Characterisation
variable and dependent upon the type of mineralisation and size of operation. Indicative is required during project costs
for a evaluation or during
characterisation production
study are USwhen new ore
$50–150 bodies
K per domain,and/orthough
ore typesmuch are encountered.
of the dataOn-going
can often be
grade control sampling costs relate to mineralisation type and mining method. For a small operation
gained from and/or shared with metallurgical programmes [6,7]. Characterisation is required during
(e.g., 100,000 t per annum) an annual cost may be in the region of US $0.3–0.5 M, whereas for a much
project evaluation or during production when new ore bodies and/or ore types are encountered.
larger operation (e.g., 1 Mt per annum) more likely US $1.5–2.0 M or more per annum. These
On-going grade
indicative control
figures sampling
include the costcosts relatecollection,
of sample to mineralisation type andand
sample preparation mining method.
assaying, Forand
staffing a small
operation (e.g., 100,000 t per annum) an
consumables, but do not include drilling costs. annual cost may be in the region of US $0.3–0.5 M, whereas for
a much larger operation (e.g., 1 Mt per annum) more likely US $1.5–2.0 M or more per annum. These
indicative figures include the cost of sample collection, sample preparation and assaying, staffing and
consumables, but do not include drilling costs.
Minerals 2018, 8, 232 40 of 45

Table 28. Summary of sampling errors and risk rating for grade control sampling. For linear/broken
rock sampling, it assumes a fully equipped two-person team to achieve best practice sample collection.
Errors are best-case scenarios based on careful sample collection/operation.

1 Sampling Errors
Sample Type Summary 2 Total Error Rating
DE EE WE
Linear
Easy to collect, relatively fast
Chip High number can be collected High
Cheap
Relatively easy to collect and fast
Chip-channel Moderately high number can be collected High
Moderately cheap
Less easy to collect, requires effort
Channel
Moderately high number can be collected Mod
(hand-cut)
Higher cost
Requires effort and specialist equipment
Channel
Moderate number can be collected Low
(saw-cut)
Higher cost
Medium-large sample size (>25 kg)
Large sample size
Panel Mod
Less easy to collect
Moderate number can be collected
Broken Rock
Relatively easy to collect
Grab Moderate number can be collected High
Cheap
Drilling
Well established method
Diamond core Good geological information Low
Relatively costly
FSE potentially mod-high when sub-sample split from
Reverse
original, though modern technology much improved Low
circulation
Cheaper than core drilling
FSE potentially mod-high when sub-sample split from
original
Sludge
Sample collection/splitting at the rig problematic, though Mod
(blasthole)
some mines have devised a reasonable catcher system
Cheap
Mini-bulk (<1 t)/Bulk (>1 t)
Comprise (i) panel or (ii) blasted/excavated blocks
Careful planning required
Bulk Low
Good geological and geotechnical information
Relatively costly depending upon mass
1 Indicative sampling error definition: FSE: Red: high (>±30%); Orange: moderate (±20–30%); Green: low (<±20%).
2 Indicative total error: Red: high (>±50%); Orange: moderate (±30–50%); Green: low (<±30%).

13. Conclusions
(1) Practitioners can be justifiably critical of linear and grab sampling methods, given their potential
for poor precision and high bias. However, they have a role to play in underground grade control,
depending on their ultimate use and effort put into making them representative. Maintaining
representative samples is advantageous; however, there will always have to be a balance between
what is theoretical, practical, and safe. It is important to understand what decisions the sample
data will be used for and the inherent risks involved both in collection and decision-making.
Critical end-member applications are (a) to support a reserve estimate for short-term mine
planning, or (b) as ”indicators” to map grade trends and/or deportment along development and
in specific geological domains.
(2) A range of sampling methods are available, all of which require evaluation before routine
acceptance. The greatest error is generally introduced during sample collection (e.g., DE and EE).
A cost-effective way to reduce sample collection error involves staff training in TOS and correct
collection methods. Whilst correctly collected channel samples are of a higher quality than chip
samples, chip samples will continue to have importance given their relative ease of collection.
A reduction in the need for linear samples will only come from innovation or the use of more
pre-development drilling at a spacing to allow local estimation.
Minerals 2018, 8, 232 41 of 45

(3) Where grade control samples support a reserve estimate, a QAQC programme must be introduced
to ensure on-going quality of sampling and assaying. Written protocols and procedures, periodic
auditing of protocols and people, and training are required. DS3077 provides a framework on
which to compile transparent protocols [38].
(4) Sample protocol optimisation is best undertaken at the breakeven cut-off grade given that it is
the ore/waste decision point. An optimised sampling programme will never be perfect and as a
result, ore/waste misclassification will occur particularly where the nugget effect is high; thus a
low-grade stockpile is recommended. Many small mines do not have drilling, therefore the most
practical and representative linear samples are required.
(5) Samples feed into reserve and grade control models that are reconciled with plant performance.
All errors along the sampling value chain are additive and impart data variability making
estimation less reliable. Errors should be evaluated by empirical means to quantify their impact
on the estimation and decision making process. Where sets of mixed data such as channel and
drill core samples are used for estimation, it is important to ensure that any bias between the
samples is identified and acted upon if appropriate.
(6) The human impact on sampling is great, given that the operating environment is generally
unforgiving and hazardous. The sampling process is a physical one and often carried out by
relatively junior and inexperienced staff. Team training is critical to improve sample quality
and representativity.
(7) Geologists must appreciate the rigors of the TOS and also understand the needs of production.
The “ideal” is advantageous, but not always practical. However, where the ideal becomes
impractical, the potential effects of the non-ideal on the decision making process must be
understood. Management and technical staff should be educated to ‘sample better’ throughout
the mine value chain.
(8) In some cases, the application of grade control samples is an exercise in futility. This usually
reflects extreme-nugget effect coarse-gold mineralisation. Challenges are particularly exacerbated
when grades are low (<5 g/t Au) and/or where the dominant gold particle size is >1 mm or
clusters. In these cases, the best option is likely to be bulk (or mini-bulk) sampling and strong
geological control during mining.
(9) In context of other mineral deposits, particularly those of base metals, the application of TOS
and its error evaluation is relevant. Additionally, use of the FSE equation is also relevant,
though the original FSE equation is usually applied in this case. The challenges of applying the
various sample types are the same. QAQC is required, particularly when samples feed into a
resource estimate.
(10) Company management in their pursuit to reduce costs, often reduce grade control expenditure
with no consideration of the impact on the quality of samples and decisions made thereon.
Persuading managers to spend the money to rectify the problems, even when they are clearly
identified, can be challenging.

Author Contributions: S.C.D. conceived the paper. S.C.D., H.J.G., and L.O. analysed the numerical data presented.
C.K.L. and S.P. reviewed publically available reports to assess industry practice. R.C.A.M. contributed his expertise
from the South African gold mines. All authors contributed to the writing and editing of the paper.
Acknowledgments: The authors acknowledge a number of current and former mining companies across Australia,
Africa, East Asia, Europe, and the Americas for the opportunity to input into grade control programmes. S.C.D.
acknowledges discussions with the late Allan (Bon) Royle and Alwyn Annels on many aspects of mine sampling.
R.C.A.M. acknowledges extensive support from the South African gold mining industry. Thanks are due to the
Minerals reviewers for helpful comments on the manuscript.
Conflicts of Interest: The authors declare no conflict of interest.
Minerals 2018, 8, 232 42 of 45

References
1. Pitard, F.F. Pierre Gy’s Sampling Theory and Sampling Practice; CRC Press: Boca Raton, FL, USA, 1993; p. 488.
2. Glass, H.J.; Zegers, T.W. Statistical aspects of sampling particulate matter. In Proceedings of the Surface
Mining Conference, Johannesburg, South Africa, 30 September–4 October 1996; Southern African Institute of
Mining and Metallurgy: Johannesburg, South Africa, 1996; pp. 157–164.
3. Minnitt, R.C.A. Sampling: The impact on costs and decision making. J. South. Afr. Inst. Min. Metall. 2007,
107, 451–462.
4. Dominy, S.C. Importance of good sampling practice throughout the gold mine value chain. Min. Technol.
2016, 125, 129–141. [CrossRef]
5. Holmes, R.J. Sampling procedures. In Advances in Gold Ore Processing; Elsevier: Amsterdam, The Netherlands,
2016; pp. 45–56.
6. Dominy, S.C.; O’Connor, L.; Glass, H.J.; Purevgerel, S.; Xie, Y. Towards representative metallurgical sampling
and recovery testwork programmes. Minerals 2018, 8, 193. [CrossRef]
7. Dominy, S.C.; O’Connor, L.; Glass, H.J.; Xie, Y. Geometallurgical study of a gravity recoverable gold orebody.
Minerals 2018, 8, 186. [CrossRef]
8. Carrasco, P.C.; Carrasco, P.; Jara, E. The economic impact of incorrect sampling and analysis practices in the
copper mining industry. Chemom. Intell. Lab. Syst. 2004, 74, 209–214. [CrossRef]
9. Joint Ore Reserves Committee (JORC). Australasian Code for Reporting of Exploration Results, Mineral Resources
and Ore Reserves—The JORC Code; Joint Ore Reserves Committee of the Australasian Institute of Mining and
Metallurgy, Australian Institute of Geoscientists and Minerals Council of Australia: Melbourne, Australia,
2012; p. 44.
10. Gy, P.M. Sampling of Particulate Materials: Theory and Practice; Elsevier: Amsterdam, The Netherlands, 1982; p. 431.
11. Storrar, C.D. Underground sampling with special reference to gold reefs. In South African Mine Valuation;
Southern African Institute of Mining and Metallurgy: Johannesburg, South Africa, 1981; pp. 36–63.
12. Dominy, S.C.; Platten, I.M.; Fraser, R.M.; Dahl, O.; Collier, J.B. Grade control in underground gold vein
operations: The role of geological mapping and sampling. In Proceedings of the International Mining
Geology Conference, Perth, Australia, 17–19 August 2009; Australasian Institute of Mining and Metallurgy:
Melbourne, Australia, 2009; pp. 291–307.
13. Pitard, F.F. The in-situ nugget effect: A major component of the random term of a variogram. In Proceedings
of the World Conference on Sampling and Blending, Porto Alegre, Brazil, 23–25 October 2007; Fundacao
Luiz Englert: Porto Alegre, Brazil, 2007; pp. 91–110.
14. Dominy, S.C. Predicting the unpredictable—Evaluating high-nugget effect gold deposits. In Mineral Resource
and Ore Reserve Estimation; Australasian Institute of Mining and Metallurgy: Melbourne, Australia, 2014;
pp. 659–678.
15. Dominy, S.C.; Xie, Y.; Platten, I.M. Gold particle characteristics in narrow vein deposits: Implications
for evaluation and metallurgy. In Proceedings of the Narrow Vein Mining Conference, Ballarat, Australia,
14–15 October 2008; Australasian Institute of Mining and Metallurgy: Melbourne, Australia, 2008; pp. 91–104.
16. Dominy, S.C.; Platten, I.M. Clustering of gold particles and implications for sampling. Appl. Earth Sci. 2007,
116, 130–142. [CrossRef]
17. Lyman, G.J.; Robertson, I.G.; Day, T. The simple facts about sampling gold ores. In Proceedings of the Mill
Operators Conference, Perth, Australia, 10–12 October 2016; Australasian Institute of Mining and Metallurgy:
Melbourne, Australia, 2016; pp. 159–172.
18. François-Bongarçon, D.M. A brief history of Pierre Gy’s sampling theory and current trends in its acceptance
in the mining world. In Proceedings of the Sampling Conference, Perth, Australia, 27–29 May 2008;
Australasian Institute of Mining and Metallurgy: Melbourne, Australia, 2008; pp. 13–14.
19. Freese, B.; Flitton, T.; Pillay, A. Quality control and assurance of underground chip sampling at Kopanang
Mine, South Africa. In Proceedings of the Sampling and Analysis Conference, Muldersdrift, South Africa,
4–6 June 2013; Southern African Institute of Mining and Metallurgy: Johannesburg, South Africa, 2013;
pp. 277–278.
20. Fourie, D.; Minnitt, R.C.A. The simulated chip-sample model as a method for qualifying error and bias in
sampling thin carboniferous reef types. J. South. Afr. Inst. Min. Metall. 2014, 114, 19–30.
Minerals 2018, 8, 232 43 of 45

21. Fourie, D.; Minnitt, R.C.A. Review of gold reef sampling and its impact on the mine call factor. J. South. Afr.
Inst. Min. Metall. 2016, 116, 1001–1009. [CrossRef]
22. François-Bongarçon, D.M.; Gy, P.M. The most common error in applying Gy’s formula in the theory of
mineral sampling and the history of the Liberation factor. J. South. Afr. Inst. Min. Metall. 2002, 102, 475–479.
23. Pitard, F.F. Pierre Gy’s Theory of Sampling and Ingamells Poisson Process Approach: Pathways to Representative
Sampling and Appropriate Industry Standards; Aalborg University: Esbjerg, Denmark, 2009; p. 309.
24. Royle, A.G. Splitting gold assay pulps containing coarse gold. J. Leeds Univ. Min. Assoc. 1989, 89, 63–68.
25. Minnitt, R.C.A.; Assibey-Bonsu, W. A comparison between the duplicate series method and the heterogeneity
test as methods for calculating the sampling constants. J. South. Afr. Inst. Min. Metall. 2010, 110, 251–268.
26. Minnitt, R.C.A. Calibrating K and Alpha in Gy’s formula: A new approach. Math. Geosci. 2016, 48, 211–232.
[CrossRef]
27. Dominy, S.C.; Xie, Y. Optimising sampling protocols via the heterogeneity test: Challenges in coarse gold
mineralisation. Min. Technol. 2016, 125, 103–113. [CrossRef]
28. Dominy, S.C.; Platten, I.M.; Xie, Y.; Minnitt, R.C.A. Underground grade control protocol design: Case study
from the Liphichi gold project, Larecaja, Bolivia. Appl. Earth Sci. 2011, 119, 205–219. [CrossRef]
29. Cintra, E.C.; Scabora, J.A.; Viegas, E.P.; Barata, R.; Maia, G.F. Coarse gold sampling at Sao Francisco
mine, Brazil. In Proceedings of the World Conference on Sampling and Blending, Porto Alegre, Brazil,
23–25 October 2007; Fundacao Luiz Englert: Porto Alegre, Brazil, 2007; pp. 187–197.
30. Gonzalez, P.; Cossio, S. A review of sampling protocol for a gold ore based on liberation study. In Proceedings
of the World Conference on Sampling and Blending, Porto Alegre, Brazil, 23–25 October 2007; Fundacao
Luiz Englert: Porto Alegre, Brazil, 2007; pp. 163–174.
31. François-Bongarçon, D.M. Extensions to the demonstration of Gy’s formula. Explor. Min. Geol. 1998, 7,
149–154.
32. Dominy, S.C.; Petersen, J.S. Sampling coarse gold-bearing mineralisation-developing effective protocols
and a case study from the Nalunaq mine, Southern Greenland. In Proceedings of the World Conference on
Sampling and Blending, Sunshine Coast, Australia, 10–12 May 2005; Australasian Institute of Mining and
Metallurgy: Melbourne, Australia, 2005; pp. 151–165.
33. Fallon, M.; Rowley, M.; Du Plessis, E.; Golenya, F.; Robertson, I.; Swanson, R. Confirming the sampling,
sample preparation and assaying protocols at two Yilgarn gold deposits. In Proceedings of the Sampling
Conference, Perth, Australia, 21–22 August 2012; Australasian Institute of Mining and Metallurgy:
Melbourne, Australia, 2012; pp. 47–62.
34. Stanley, C.R.; Smee, B.W. Strategies for reducing sampling errors in exploration and resource definition
drilling programmes for gold deposits. Geochem. Explor. Environ. Anal. 2007, 7, 329–340. [CrossRef]
35. Simon, A.; Gosson, G. Considerations on quality assurance/quality control and sample security.
In Proceedings of the Sampling Conference, Perth, Australia, 27–29 May 2008; Australasian Institute of
Mining and Metallurgy: Melbourne, Australia, 2008; pp. 135–140.
36. François-Bongarçon, D.M. Theory of sampling and geostatistics: An intimate link. Chemom. Intell. Lab. Syst.
2004, 74, 143–148. [CrossRef]
37. Pitard, F.F. Guidelines for acceptable allotted sampling uncertainty. In Proceedings of the World Conference
on Sampling and Blending, Lima, Peru, 19–22 November 2013; Gecamin: Santiago, Chile, 2013; pp. 89–98.
38. DS3077. Representative Sampling–Horizontal Standard; Danish Standards Foundation: Copenhagen, Denmark,
2013; p. 41.
39. Carswell, J.T.; Yulia, K.; Lesmana, D.; Steamy, K. Grade control sampling quality assurance/quality
control in a high-grade gold mine—Gosowong, Indonesia. In Proceedings of the International Mining
Geology Conference, Perth, Australia, 17–19 August 2009; Australasian Institute of Mining and Metallurgy:
Melbourne, Australia, 2009; pp. 285–290.
40. Sims, D.A.; Benton, R. Life on the Rollercoaster—Mining and milling the Kencana K1 orebody, Indonesia.
In Proceedings of the International Mining Geology Conference, Perth, Australia, 17–19 August 2009;
Australasian Institute of Mining and Metallurgy: Melbourne, Australia, 2009; pp. 319–327.
41. Stanley, C.R.; Lawrie, D. Average relative error in geochemical determinations: clarification, calculation and
a plea for consistency. Explor. Min. Geol. 2007, 16, 267–275. [CrossRef]
42. Dominy, S.C. Sampling and Evaluation of Vein Gold Deposits; Final Technical Report for DevR-MIDAS Project;
University of Greenwich: Chatham, UK; Welsh Gold PLC: Dolgellau, UK, September 1998; p. 230.
Minerals 2018, 8, 232 44 of 45

43. François-Bongarçon, D.M. Fishy samples: How big a sample to avoid the infamous Poisson effect.
In Proceedings of the World Conference on Sampling and Blending, Cape Town, South Africa, 21–23
October 2009; Southern African Institute of Mining and Metallurgy: Johannesburg, South Africa, 2009;
pp. 43–46.
44. Gouws, J. The use of diamond saws in underground sampling at Randfontein Estates Gold Mining Company,
Witwatersrand. J. Inst. Mine Surv. S. Afr. 1986, 23, 85–97.
45. Magri, E.J.; McKenna, A.P. A geostatistical study of diamond saw sampling versus chip sampling. J. South.
Afr. Inst. Min. Metall. 1986, 86, 335–347.
46. Potter, D.; Sheriff, C.; Collins, P. Significance of geological control on assay data in the narrow, high-grade
gold quartz veins at the Sand Queen mine, Comet Vale, Western Australia. In Proceedings of the Narrow
Vein Mining Conference, Ballarat, Australia, 14–15 October 2008; Australasian Institute of Mining and
Metallurgy: Melbourne, Australia, 2008; pp. 71–78.
47. Rowland, R.S.; Sichel, H.S. Statistical quality control of routine underground sampling. J. South. Afr. Inst.
Min. Metall. 1960, 60, 251–284.
48. Van Der Merwe, H.C. Sirovision: A proposed solution for the implementation of a digital geological mapping
system at AngloGold Ashanti’s Moab Khotsong mine. In Proceedings of the World Gold Conference, Misty
Hills, South Africa, 26–30 October 2009; Southern African Institute of Mining and Metallurgy: Johannesburg,
South Africa, 2009; pp. 183–191.
49. Lerm, C. Evaluation of chip sampling vs the gold analyser. J. Inst. Mine. Surv. S. Afr. 1994, 27, 77–87.
50. Cawood, F.T. Underground face sampling on narrow reefs—Sampling practice and its impact on the mine
call factor on Witwatersrand gold mines. J. Inst. Mine Surv. S. Afr. 2003, 31, 202–224.
51. Dominy, S.C. Grab sampling for underground gold mine grade control. J. South. Afr. Inst. Min. Metall. 2010,
110, 277–287.
52. Ross, I.T. Sampling in block cave mines. In Proceedings of the Sampling Conference, Perth, Australia, 21–22
August 2012; Australasian Institute of Mining and Metallurgy: Melbourne, Australia, 2012; pp. 81–85.
53. Arslan, N.; Ozkayhan, M. Diamond drilling at Efemcukuru gold mine, Turkey—A case study on the
benefits of timely and cost-effective drilling to short-term planning. In Proceedings of the International
Mining Geology Conference, Hobart, Australia, 20–22 September 2017; Australasian Institute of Mining and
Metallurgy: Melbourne, Australia, 2017; pp. 285–290.
54. Dominy, S.C. Sampling coarse gold-bearing mineralisation-developing effective protocols and a case study
from the Ballarat mine, Australia. In Proceedings of the World Conference on Sampling and Blending, Perth,
Australia, 9–11 May 2017; Australasian Institute of Mining and Metallurgy: Melbourne, Australia, 2017;
pp. 3–19.
55. McBeath, R.; Whitworth, D. What drilling density is optimal—A case study from the Golden Pig
mine. In Proceedings of the Underground Operators Conference, Townsville, Australia, 29–31 July 2002;
Australasian Institute of Mining and Metallurgy: Melbourne, Australia, 2002; pp. 33–37.
56. Clark, F.; Carswell, J.T. Using reverse circulation drilling to improve sampling in a complex underground
gold operation, Sunrise Dam mine, Western Australia. In Proceedings of the International Mining Geology
Conference, Adelaide, Australia, 18–20 August 2014; Australasian Institute of Mining and Metallurgy:
Melbourne, Australia, 2014; pp. 143–152.
57. Prouhet, J.-P. A quick method of sampling development headings. In Proceedings of the Ore Reserve
Estimation and Grade Control Conference, Ville d’Esterel, QC, Canada, 25–27 September 1967; CIM: Montreal,
QC, Canada, 1968; pp. 97–102.
58. Ovalle, A.; Magri, E.; Zuniga, P.; Pitard, F.F. Sampdrill. Quant. Geol. Geostat. 1997, 8, 767–779.
59. Johansen, G.F.; Dominy, S.C. Development of sampling and assaying protocols at the new Bendigo gold
project, Victoria, Australia. In Proceedings of the World Conference on Sampling and Blending, Sunshine
Coast, Australia, 10–12 May 2005; Australasian Institute of Mining and Metallurgy: Melbourne, Australia,
2005; pp. 175–183.
60. Roberts, L.S.; Dominy, S.C.; Nugus, M.J. Problems of sampling and assaying in lode-gold deposits—Case
studies from Australia and north America. In Proceedings of the International Mining Geology Conference,
Bendigo, Australia, 17–19 November 2003; Australasian Institute of Mining and Metallurgy: Melbourne,
Australia, 2004; pp. 387–400.
Minerals 2018, 8, 232 45 of 45

61. Dominy, S.C. Effects of sample mass on gravity recoverable gold test results in low grade ores. Appl. Earth Sci.
2014, 123, 234–242. [CrossRef]
62. Dominy, S.C.; Annels, A.E.; Johansen, G.F.; Cuffley, B.W. General considerations of sampling and assaying in
a coarse gold environment. Appl. Earth Sci. 2000, 109, 145–167. [CrossRef]
63. Hutchison, R. Geology and definition of the Wattle Dam coarse gold deposit. In Proceedings of the Gravity
Gold Conference, Ballarat, Australia, 21–22 September 2010; Australasian Institute of Mining and Metallurgy:
Melbourne, Australia, 2010; pp. 109–118.
64. Elder, D.J. Leachwell versus fire assay: Comments from assay quality assurance at Tarmoola gold mine.
In Proceedings of the International Mining Geology Conference, Coolum, Australia, 14–17 May 2000;
Australasian Institute of Mining and Metallurgy: Melbourne, Australia, 2009; pp. 43–52.
65. Pitard, F.F. Sampling correctness: A comprehensive guide. In Proceedings of the World Conference on
Sampling and Blending, Sunshine Coast, Australia, 10–12 May 2005; Australasian Institute of Mining and
Metallurgy: Melbourne, Australia, 2005; pp. 55–66.
66. Sims, D.A. Sampling practices at Vera Nancy gold mine. In Proceedings of the International Mining
Geology Conference, Coolum, Australia, 14–17 May 2000; Australasian Institute of Mining and Metallurgy:
Melbourne, Australia, 2009; pp. 35–42.
67. Benndorf, J.; Buxton, M.W.N. Sensor-based real-time resource model reconciliation for improved mine
production control—A conceptual framework. Min. Technol. 2016, 125, 54–64. [CrossRef]
68. Kurth, H. Measuring ore quality using real-time on-belt elemental analysis—More than grade control.
In Proceedings of the International Mining Geology Conference, Hobart, Australia, 20–22 September 2017;
Australasian Institute of Mining and Metallurgy: Melbourne, Australia, 2017; pp. 213–220.
69. Rifai, K.; Laflamme, M.; Constantin, M.; Vidal, F.; Sabsabi, M.; Blouin, A.; Bouchard, P.; Fytas, K.; Castello, M.;
Nguegang Kamwa, B. Analysis of gold in rock samples using laser-induced breakdown spectroscopy: Matrix
and heterogeneity effects. Spectrochim. Acta B 2017, 134, 33–41. [CrossRef]
70. Tickner, J.; Ganly, B.; Lovric, B.; O’Dwyer, J. Improving the sensitivity and accuracy of gamma activation
analysis for the rapid determination of gold in mineral ore. Appl. Radiat. Isot. 2017, 122, 28–36. [CrossRef]
[PubMed]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).