Drill hole spacing analysis for classification and cost optimisation –

a critical review of techniques

I Glacken1, O Rondon2 and J Levett3

1. FAusIMM(CP), Executive Consultant, Snowden Optiro, Perth WA 6000.

Email: [email protected]
2. Principal Geostatistician, Snowden Optiro, Perth WA 6000.
Email: [email protected]
3. Principal Consultant, Snowden Optiro, Perth WA 6000. Email: [email protected]

ABSTRACT
Reporting Codes are not prescriptive on methodologies to report or classify Mineral Resource
Estimation results, but the assessment of risk and uncertainty is required, and is likely to be
increasingly important in future Code updates. The 2012 JORC Code indicates that relative
confidence in tonnage-grade estimations should be considered for classification, and that there
should be a discussion of relative accuracy and confidence levels in the Mineral Resource Estimate.
These considerations are also relevant to the drill hole spacing required for various classification
categories, along with other geological and data quality criteria.
This paper summarises the theory, principles and current practice of various drill hole spacing
analysis techniques for both achieving a desired level of resource categorisation and for optimising
grade control or resource drill-out costs. The authors provide practitioners with a guide in their use
to assess uncertainty and relative accuracy, as illustrated by a number of case studies. While all drill
hole spacing studies rely on the assumption of consistency of the continuity model from current to
future drilling, the range of methods available provide much more objective outcomes than previously
adopted ‘rules-of-thumb’.

INTRODUCTION

The motivation for drill hole spacing analysis (DHSA)

Prior to the capital expenditure associated with building a mine and a processing plant, one of the
largest costs associated with the mine value chain relates to drilling. There is an ever-increasing
emphasis on the consideration of Mineral Resource classification categories, whether via the JORC
Code (Australia) the CIM Guidelines (Canada), the SAMREC Code (South Africa), S-K 1300 (USA),
the PERC Code (Europe) or other CRIRSCO-based schemes. As a consequence of this, explorers
and miners wishing to achieve production recognise the criticality of attaining an Indicated level of
classification, since this is the lowest confidence level which can be converted to reserves and thus
contribute to a forecast cash flow via the pre- and feasibility process. While all of the aforementioned
Codes emphasise the prerogative of the individual Competent or Qualified Person to allocate the
various resource categories (and especially the Indicated Resource category), for various reasons
there is an increased search for objectivity in Mineral Resource classification. These reasons may
include:
• the need to assess projects competing for scarce capital.
• the desire, despite the individual CP/QP decision, to achieve an objective standard for
classification across what may be a significant number of projects in a large mining company.
• in the case of external funding, the need to achieve a certain minimum level of higher-
confidence resources and reserves in the early years of production.
As a result, increased emphasis is now placed on drill hole spacing analysis (DHSA) to achieve pre-
determined classification levels. However, there is also a strong motivation to optimise the costs of
drilling throughout the exploration-development-production cycle.
This paper presents a critical assessment of the relatively wide range of approaches to DHSA,
whether to achieve a certain classification level (usually Indicated) or whether to optimise the costs
of grade control drilling. Many of the techniques are common across both motivations. A review of

Mineral Resource Estimation Conference 2023 | Perth, Australia | 24–25 May 2023 179
the relatively simple and straightforward group of methods based upon estimation variance and its
derivatives is presented, along with reminders of the key assumptions inherent in the application of
these techniques. The connection between a Kriging variance (KV) or confidence interval (CI) -based
approach and the very common ‘90:15’ standard for resource classification (Verly, Postolski and
Parker, 2014; Parker and Dohm, 2014; Owusu and Dagdelen, 2019) is emphasised. A case study
using these approaches to quantify drill spacing in Mineral Resource development is presented for
a nickel-cobalt laterite project.
The increasing use of conditional simulation in DHSA is detailed, both for classification and for
general cost optimisation. The benefits of using conditional simulation and the drawbacks (increased
workload, quantification of uncertainty) are emphasised. The combination of simulation and cost
optimisation for DHSA is presented, with an overview of the benefits of considering the cost versus
revenue dimension for simulated drill holes. Finally, a case study of grade control drilling optimisation
for a nickel-copper project, using multivariate simulation and value optimisation, is presented.

Overview of techniques and previous work

The description and documentation of DHSA techniques, whether for achieving a desired resource
classification level, or for general grid optimisation, has become a relatively popular topic in the past
decade. DHSA studies reinforce the themes of compliance with increasingly prescriptive
classification Codes, the capture and quantification of uncertainty through simulation, and the use of
risk profiles to optimise the risk-reward balance associated with various levels of drilling.
Rivoirard et al (2016) proposed a two-step approach for classification. First, a measure of spatial
sampling density of the deposit is derived. Second, a resource classification is obtained by
thresholding a coefficient of variation derived from the measure of spatial sampling density. This
approach has the advantage of comparing objectively different drill hole spacings across the volume
of interest to derive a classification category.
Verly, Postolski and Parker (2014) describe a production-linked DHSA approach designed to achieve
a certain classification level. The principles underlying this approach are simple:
• define the set of drill hole spacings to evaluate.
• compute the relative error of an attribute within a large production block corresponding to a
period of interest using the specified drill hole spacings.
• assume that the relative error has a Gaussian distribution and deduce relative confidence
intervals accordingly.
• assign a classification based on whether the confidence interval is within a defined error
tolerance.
• select the widest drill hole spacing which achieves less than the desired error.
2
The relative error is determined as the ratio of the Kriging variance 𝜎𝜎𝑂𝑂𝑂𝑂 and the square of the mean
value 𝑚𝑚 of the composited drill hole data. Therefore, under the Gaussian assumption, the 𝛼𝛼%
confidence interval is:
1 + 𝛼𝛼 𝜎𝜎𝑂𝑂𝑂𝑂
𝛼𝛼% 𝐶𝐶𝐶𝐶 = 𝐹𝐹 −1 � �
2 𝑚𝑚
where 𝐹𝐹 −1 is the inverse of the standard Gaussian distribution function. This function can easily be
evaluated using Excel.
Industry practice (Parker and Dohm, 2014; Verly, Postolski and Parker, 2014) is to consider a
90 per cent confidence interval, a 15 per cent error tolerance and classification as follows:
• Measured resources – material which has a 90 per cent confidence of being within 15 per cent
of the true value for parcels reflecting quarterly production.
• Indicated resources – material which has a 90 per cent confidence of being within 15 per cent
of the true value for annual production parcels.

Mineral Resource Estimation Conference 2023 | Perth, Australia | 24–25 May 2023 180
The 90 per cent confidence means that 𝐹𝐹 −1 (1 + 𝛼𝛼 ⁄2) = 𝐹𝐹 −1 (0.95) = 1.645. Therefore, it is only
2
required to compute the Kriging variance 𝜎𝜎𝑂𝑂𝑂𝑂 for the different drill spacings and check
if 1.645 ∗ 𝜎𝜎𝑂𝑂𝑂𝑂 ⁄𝑚𝑚 is less than the error tolerance of 15 per cent. Hereafter, this approach will be
referred as to the 90:15 method.
The authors have recently observed that a wider error margin, such as 30 per cent, is being used to
separate Inferred areas from unclassified areas and endorse this refinement. Verly’s co-author, the
late H M Parker, presents a historical perspective of the development of this technique in the 2014
paper and also in a presentation with C E Dohm (Parker and Dohm, 2014). The 90:15 method was
also first addressed by Parker and others in Murphy et al (2004) and by Wawruch and Betzhold
(2004).
Considerable effort has been devoted to the development of DHSA techniques which align with this
standard. Nowak and Leuangthong (2019) describe a number of approaches, based upon KV and a
90 per cent CI but also including simulation and the idea of resampling a reference realisation, a
method covered by the authors in a case study below.
Usero, Misk and Saldanha (2019) also detail the use of conditional simulation of grades, which are
then resampled and scaled up to the various production units required by the 90:15 method; the
optimal drill spacing is a presented as a function of the accuracy, as measured at a 90 per cent CI
with a 15 per cent simulation risk (error). In this work the different aspects of grade risk and thickness
risk (in a nickel laterite case study) are compared, and it is important to acknowledge that grade
uncertainty (as provided by conditional simulation) is often only one aspect of the total risk.
Neves et al (2023) also examine the role that simulation plays in optimising a drilling grid, again with
a case study on a nickel laterite deposit. In this case, the simulation examines three dimensions of
uncertainty; nickel grade, thickness, and ore type (as simulated via a categorical indicator approach).
Using the 90:15 approach for quarterly panels, the drilling grid required to achieve a maximum
15 per cent error is defined. This study does consider geological uncertainty (via the categorical ore
type simulation) along with thickness and grade, which is not available in most cases due to the
relative complexity of the geological controls on mineralisation.
Alfonseca and Silva (2022) present two alternative DHSA workflows, both based upon simulation.
One approach involves the sampling of a dense simulation to generate pseudo-drill holes (as in the
case study below) followed by estimation and comparison with the original simulation using a number
of misclassification measures. The second documented workflow involves re-simulation of pseudo-
drill holes from a prior simulation based upon the existing drilling data, followed by calculation of the
error of the group of realisations. In each case the chosen drill spacing is a function of the desired
level of error.

VARIANCE-BASED METHODS

Overview
Kriging variance (KV) based methods of DHSA are simple and relatively straightforward to
undertake. These DHSA approaches allow for the basic assessment of estimation uncertainty, where
risk assessment requirements are not complex, and the time available to complete a simulation is
lacking.
KV-based methods calculate the Slope of Regression (SR) and Kriging Efficiency (KE) for a range
of conceptual drill grid spacings, where it is assumed that the overall grade continuity model remains
constant with increased (conceptual) information. These methods use project-specific variograms
and estimation block parameters, although it is important to note that the quality of the variogram will
impact the result of the study, and that the variogram models may change with increased physical
information. The theoretical estimation quality can be compared for each grid, and the results can
provide a general indication of risk and uncertainty with each potential grid spacing. The KV based
methods can also be used to optimise expenditure, given some simple cost assumptions.
Variance-based methods of DHSA do not consider potential risks, including variations in geology,
deleterious elements, or mining considerations. The methods, as described in Verly, Postolski and
Parker (2014), link to the 90:15 assumptions as discussed above, and provide a reasonably objective

Mineral Resource Estimation Conference 2023 | Perth, Australia | 24–25 May 2023 181
framework for the definition of drilling for Measured or Indicated Resources. The 90:15 approach is
best applied to large production situations; there is an attempt to quantify error via a confidence
interval, but there is no consideration of grade or geological risk.

Case study – Wilconi nickel laterite

This case study describes the use of a KV-based approach to define a drill spacing intended to give
Measured Resources. A-Cap Energy’s majority-owned Wilconi Project is situated 700 km from Perth
within the Norseman-Wiluna Greenstone Belt. The nickel-cobalt mineralisation has developed
through lateritisation of the Perseverance ultramafic sequence, extends for around 20 km along
strike and is up to 1500 m wide. Nickel and cobalt mineralisation have been formed by intense
weathering and have concentrated in a saprolite clay layer that overlies the ultramafic rock unit. The
depth to significant (>0.5 per cent Ni) mineralisation ranges from 2 m to 60 m. The nickel
mineralisation zone can be up to 30 m thick, averaging around 4 m in thickness (Figure 1).

FIG 1 – Location and geology of the Wilconi nickel laterite project.

Two variance-based analyses were undertaken to recommend appropriate drill spacing for
Measured Resources. A very basic DHSA, using the project variograms and drilling data and the
Mineral Resource Estimate parent cell size, was undertaken. The mineralisation domain selected to
undertake the analysis was chosen as it had the ‘worst’ variography – that is, the highest nugget and
shortest range, and it also wholly encompassed another higher-grade mineralisation domain
(Figure 2, where Y is north and X is east). The logic is that any drill spacing recommended for this
‘worst’ domain would be more than adequate for zones of better continuity. A range of conceptual
drill grids with vertical holes was tested down to a 10 m × 10 m spacing, and KE versus drill costs
were graphed. The optimal drill grid spacing was selected where the improvement in Kriging quality
(KE and SR) plateaus, and the drill costs start to increase significantly (Figure 3).

Mineral Resource Estimation Conference 2023 | Perth, Australia | 24–25 May 2023 182
FIG 2 – Variogram model (left) and search ellipse with existing drilling data (right) for mineralisation
domain.

FIG 3 – Outputs of the DHSA at the Mineral Resource block size.

The same process was undertaken to encapsulate the 90:15 approach; however, in this case, a
quarterly production volume was applied to the block size (250 m × 250 m × 5 m) rather than the
parent cell volume (Figure 4). The same mineralisation domain variography and conceptual drill grids
were used, although a large number of samples were used to offset the potential impact positively-
skewed populations may have on confidence interval calculations. Post processing for the 90:15
requirements, using the method of Verly, Postolski and Parker (2014) suggests that a drill grid of
50 m × 50 m is appropriate (Figure 5). This shows the 15 per cent error (orange line) and the scaled
90 per cent CI based upon the Kriging variance as the blue line. As a comparison, a nearby
advanced nickel laterite project defines Measured Resources based upon a 50 m × 50 m drilling
grid.

Mineral Resource Estimation Conference 2023 | Perth, Australia | 24–25 May 2023 183
 FIG 4 – Production block scale located within the mineralisation domain.

FIG 5 – Output of the 90:15 approach to variance-based drill hole spacing analysis.

It is important to reiterate that a key assumption of the KV methods is that the current variogram,
based upon the existing drilling, will be preserved with infill drilling. A refinement of the KV-based
approaches is to investigate the impact of a ‘worse’ and a ‘better’ variogram than the current
evidence-based model, but these are not based upon real-world outcomes, just predictions.

Mineral Resource Estimation Conference 2023 | Perth, Australia | 24–25 May 2023 184
SIMULATED REALITY APPROACHES

Overview
As mentioned above, the use of conditional simulation provides the DHSA practitioner with the
potential to define a dense grid of data, using the parameters of the existing real-world data set,
which can then be post-processed in a variety of ways, either to define a confidence interval as called
for in the 90:15 pseudo-standard, or to look at the drilling required to generate a maximum error, or
more sophisticated methods. The approach described here, and the associated case study, looks at
the cost of misclassification and the drilling grid required to minimise the cost of misclassification,
based upon some relevant cost and revenue assumptions. The key objective is to define an optimal
grade control grid, but the same approach can be used in the 90:15 context to achieve a certain
classification confidence level.
A typical workflow for this variant of the simulated approach would be as follows:
1. Using the real-world data, simulate on a dense grid over the area of interest.
2. From the suite of realisations, choose one which reflects the key parameters (histogram,
variogram, multivariate statistics) of the existing drill holes.
3. Using the chosen realisation as the ‘truth’, resample to generate pseudo drill holes on the
range of potential drilling grids.
4. Generate a grade estimate for the key variables for each potential drill grid.
5. Using a model of the misclassification and associated revenue and losses, calculate the total
net revenue (or loss) associated with each drilling grid. The optimal drilling grid is that which
gives the best cost-benefit.
This approach is similar to one of the workflows presented by Alfonseca and Silva (2022), but the
emphasis is on defining the misclassification costs and benefits. As with the KV/CI methods, the
outcome depends upon the assumption that the quality of the continuity model does not diminish as
infill drilling takes place.

Cost optimisation
The theoretical relationship between estimated blocks and true blocks in a mineral deposit has been
documented since the introduction of modern geostatistical techniques and first appeared in Journel
and Huijbregts (1978). A version is presented in Figure 6.

Z 1 2

Ore loss Ore as ore

True Grades

z
Cut-off
grade

Waste as waste Dilution

4 3
Cut-off
z
grade
Estimated Grades
Z*
FIG 6 – Relationship between estimated and true grades above and below an applied cut-off.

In its simplest form, the relationship between estimated block grades and true block grades is
portrayed with respect to an applied mining cut-off. This divides the blocks into four quadrants:
1. Blocks predicted to be waste which are in reality ore.
2. Blocks predicted to be ore which are truly ore.

Mineral Resource Estimation Conference 2023 | Perth, Australia | 24–25 May 2023 185
 3. Blocks predicted to be ore which are in reality waste.
4. Blocks which are predicted to be waste, and which actually are waste.
This rather simplistic formulation generates two quadrants, 2 and 4, where there is no
misclassification, and two quadrants, 1 and 3, where potential misclassification exists. Figure 6
shows an ellipse reflecting the scatter of true versus estimated grades and a 1:1 regression line. Due
to conditional bias the diagram typically exhibits a regression line having a slope less than unity; in
other words, high-grade estimated blocks underestimate the true grade and low-grade estimated
blocks overestimate the true grade.
The key application is that the ‘true’ blocks on the Y axis can be represented by a simulation which
reflects the key parameters—histogram, variogram and inter-analyte correlation (if necessary)—of
the sample drill holes. The estimated blocks on the X axis reflect ordinary kriged (or similar) models
built from pseudo-drill holes selected from the ‘reality’ at various grid spacings. Thus any one block
in an estimate will reside in one of the four quadrants.
Cost profiles can be assigned to the four quadrants. In the simplest possible formulation, this may
be:
• quadrant 1 – the negative value of the material not processed (‘ore loss’)
• quadrant 2 – the revenue from the material sent for processing
• quadrant 3 – the cost of mining and processing waste material which has no value (‘dilution’)
• quadrant 4 – the cost of mining waste.
A simplified model (as in the case study below) is to assume an open pit scenario where all material
(ie quadrants 1–4) has to be removed, whether to the waste dump or to the processing plant – thus
the mining cost is not a factor in the cost as it is the same for each block in whatever quadrant. There
are many refinements which may be made to this simple model, including a variable mining cost for
ore and waste material.
The key point is to choose the drilling configuration which either minimises the loss or (hopefully)
maximises the profit. For each proposed drilling grid, however, there is the cost of drilling; thus a
very tight drilling grid should maximise the profit, but the cost of drilling may outweigh the revenue
from the reduced misclassification.

Case study – nickel-copper project

The objective of this case study was to define the optimum drilling grid for grade control at a nickel-
copper project scheduled to be mined by an open pit. The misclassification approach was adopted,
and four potential drilling grids, with different spacings, were drawn from a ‘truth’ simulation which
replicated the statistics of the input (and relatively wide-spaced) drilling on a fine grid. The revenue
from mining ore blocks, or the loss of revenue of dumping ore blocks, was defined as a net smelter
return (NSR) value, which built in the revenue from the nickel and the copper, adjusted for the relative
metallurgical recoveries which are a function of grade, and also consider treatment and refining
charges. Figure 7 shows the orebody in real space (left) and in flattened space (right). There are
some artefacts of the flattening process which are not material to the outcome. The exploration
drilling is on an approximate 25 m × 25 m to 50 m × 50 m grid. The mineralisation is around 750 m
in the north–south dimension and around 950 m in the east–west dimension.
The reference simulations for the DHSA use an additive log ratio compositional transformation (ALR)
method, followed by Gaussianisation to preserve the histogram, covariogram and correlation
between the input nickel and copper grades. A detailed description of the multivariate simulation is
beyond the scope of this paper, but the theory and practice are summarised in Cook et al (2023).
Figure 8 shows the input composite nickel and copper data and the simulated values from one
realisation, intended to represent reality. The selected realisation preserved the histogram of the
nickel and copper values, the covariogram and the correlation between the samples.

Mineral Resource Estimation Conference 2023 | Perth, Australia | 24–25 May 2023 186
 FIG 7 – 3D view of the mineralisation and drill holes, coloured on nickel grade in real-world space
(left) and in flattened space (right).

Once a realisation of simulated nickel and copper grades was selected, this was used to generate
pseudo drilling grids which were derived from the simulated samples. Four potential candidate grids
were chosen; 10 m × 10 m, 15 m × 15 m, 20 m × 20 m and 25 m × 25 m, each with 2 m downhole
samples from vertical holes. In each case holes were generated on an offset ‘dice 5’ pattern. Figure 9
shows two pseudo drilling grids selected from the realisation, 10 m × 10 m and 25 m × 25 m.
Once the four drilling grids had been extracted, these were used to generate an ordinary kriged block
model at a 10 m × 10 m × 5 m SMU size. All four models used the same variograms, which were
generated from the 10 m × 10 m pseudo-drilling, not from the original simulation. It would have also
been appropriate to generate different variograms in each drill grid instance, but the differences in
continuity were not considered material, given the generally good variograms of the nickel and
copper mineralisation.

Mineral Resource Estimation Conference 2023 | Perth, Australia | 24–25 May 2023 187
 FIG 8 – Histogram of the sample nickel and copper composites (in ppm, top) and the simulated
values from one realisation (bottom).

The OK models were validated against their own set of input pseudo-drill holes and then each were
compared back to ‘truth’ simulation. Drilling and assaying costs were supplied and were based upon
current contract prices, assuming that all samples were to be assayed for both nickel and copper.
Figure 10 shows the scatter between estimated and ‘true’ grades for the 10 m × 10 m and 25 m ×
25 m drilling cases, applying a cut-off of 0.15 per cent copper. The quadrants as per Figure 6 are
identified.

Mineral Resource Estimation Conference 2023 | Perth, Australia | 24–25 May 2023 188
 FIG 9 – Two instances of simulated drilling grids showing nickel grades, 10 m × 10 m (top) and
25 m × 25 m (bottom).

4 3

FIG 10 – Comparison of the estimated model versus the truth for copper grades in the 10 × 10 m
drilling (left) and 25 × 25 m drilling (right) cases.

Figure 10 shows that quadrant 3 (dilution) is a greater contribution to misclassification than

quadrant 1 (ore loss). It can be predicted in advance of the estimation-simulation comparison that a
tighter drill spacing will result in less misclassification errors; however, this approach allows the
quantification of the economic impact of the misclassification compared to the overall drilling cost.
For the misclassification algorithm, a simplified model was used, assuming that all tonnes had to be
mined in an open pit and had the same mining cost; therefore, the following revenue/cost streams
were allocated to the quadrants of Figure 6:

Mineral Resource Estimation Conference 2023 | Perth, Australia | 24–25 May 2023 189
 • quadrant 1 – the revenue from the lost ore sent to the waste dump
• quadrant 2 – the revenue minus the processing cost
• quadrant 3 – the processing cost of the waste material
• quadrant 4 – no cost or revenue.
The net revenue or cost per block is the sum of these four (three) items, accumulated over the entire
deposit, minus the drilling cost for each scenario. Once the consolidated net revenue is available,
several metrics can be generated to optimise the drilling grid; one is shown in Figure 11, which plots
the net revenue relative to the 10 m × 10 m drilling case against the drilling grid.

Net revenue, relative to 10 x 10 pattern

100.5%

100.0%

99.5%
Net revenue

99.0%

Net revenue is maximised for 15 x 15 drilling

98.5%

98.0%
10 x 10 15 x 15 20 x 20 25 x 25
Drill spacing

FIG 11 – Chart of net revenue relative to the 10 × 10 m drill pattern.

It is noted that the differences in revenue are small, but that the revenue is maximised for the 15 m
× 15 m drilling case. Considering the cost of drilling alone, the 10 m × 10 m drilling grid costs
50 per cent more than the 15 m × 15 m grid.
The misclassification model used to generate the net revenue is robust but simplistic, and could be
refined to reflect more realistic cost models, for instance:
• the relative and different costs of mining ore and waste can be used
• an increasing mining cost with depth could be applied
• the opportunity cost of displacing ore from the mill by processing waste can be invoked.
The sophistication of the misclassification revenue algorithm depends in part on the predicted mine-
mill balance; if mine production outstrips mill capacity then quadrant 3 (dilution) incurs a higher
penalty, while if the processing capability is greater than mine production, the cost of dumping ore is
more than the cost of accepting dilution at the plant. This invokes the theories of risk profiles and
loss functions (Glacken, 1996) which can be used to define a more realistic outcome.

CONCLUSIONS – THE FUTURE OF DHSA

The practice of drill hole spacing analysis is mature and the literature is replete with detailed case
studies, especially over the past five years. DHSA is used either to define the spacing for a particular
Mineral Resource classification, usually Indicated, based upon the industry 90:15 pseudo-standard,
or to optimise the return from drilling by quantifying and minimising misclassification. There are a
range of techniques available to the practitioner, which can be based upon relatively simple and
straightforward treatments of Kriging variance for various production-related units, or upon more
complex (and time-consuming) approaches which consider the risk inherent in grade (and
sometimes thickness and material type) through conditional simulation. Because the simulation
approaches have the ability to generate a dense grid of data which reflects the input sample data,
multiple pseudo-drilling grids can be used, either for re-simulation (generating a confidence interval

Mineral Resource Estimation Conference 2023 | Perth, Australia | 24–25 May 2023 190
and associated mean error) or to compare estimates at the proposed scale of mining with the ‘truth’,
allowing for some relatively sophisticated cost and revenue analysis.
Further developments will see the routine inclusion of these tools in either generalised or specialised
mining software packages and the analysis of optimal drill spacing as part of estimation and
classification workflows. The increased ability to quickly generate simulated data on a dense grid
and then resample either for re-simulation or estimation will lead to more accurate DHSA studies
which should result in cost and time optimisation. The introduction of risk profiles will provide a more
realistic outcome which considers differential costing.
All techniques, however, whether simple or complex, do rely on the assumption of consistency of
the spatial continuity model from the existing, sparse data to the future infill drilling, whether for
resource classification or grade control purposes. Any scenario testing of alternative variograms can
never be based upon reality but may allow the consideration of upside or downside scenarios.
Furthermore, while the DHSA methods described are complementary to the Mineral Resource
Classification, the consideration of other classification criteria, such as data quality, geological
uncertainty, and other factors must form part of the classification allocated by the Competent or
Qualified Persons, and thus the DHSA results can only be a starting point, albeit a firm foundation
for the ultimate outcome.

ACKNOWLEDGEMENTS
The permission of A-Cap Energy to publish results from the Wilconi Project is gratefully
acknowledged. Permission from another minerals company to discuss the multivariate DHSA case
study and findings is also appreciated. The paper has benefited from a review by Paul Blackney.

REFERENCES
Alfonseca, B V and Silva, V, 2022. Defining optimal drill-hole spacing; a novel integrated analysis from exploration to ore
control, Journal of the Southern African Institute of Mining and Metallurgy, 122(6):1–12.
Cook, A, Rondon, O, Graindorge, J and Booth G, 2023. Iterative Gaussianisation for multivariate transformation, in
Geostatistics Toronto 2021 – Quantitative Geology and Geostatistics (eds: S A Avalos Sotomayor, J M Ortiz and R
Mohan Srivastava), pp 21–36 (Springer Verlag).
Glacken, I M, 1996. Change of support by direct conditional block simulation, Master’s thesis, Stanford University, USA.
Journel, A G and Huijbregts, C J, 1978. Mining Geostatistics, 600 p (Academic Press: London).
Murphy, M, Parker, H M, Ross, A and Audet, M-A, 2004. Ore-thickness and grade resource confidence at the Koniambo
Nickel Laterite (a conditional simulation voyage of discovery), Geostatistics Banff 2004 (eds: O Leuangthong and
C V Deutsch), pp 469–478 (Springer Verlag).
Neves, C M S, Costa, J F, Souza, L, Guimaraes, F and Dias, G, 2023. Methodology for defining the optimal drilling grid in
a laterite nickel deposit based on a conditional simulation, Geostatistics Toronto 2021 – Quantitative Geology and
Geostatistics (eds: S A Avalos Sotomayor, J M Ortiz and R Mohan Srivastava), pp 151–162 (Springer Verlag).
Nowak, M and Leuangthong, O, 2019. Optimal drill hole spacing for resource classification, Mining Goes Digital (ed:
U Mueller), pp 115–124 (Taylor and Francis Group: London).
Owusu, S K A and Dagdelen, K, 2019. Critical review of mineral resource classification techniques in the gold mining
industry, Mining Goes Digital (ed: U Mueller), pp 201–209 (Taylor and Francis Group: London).
Parker, H M and Dohm, C E, 2014. Evolution of Mineral Resource classification from 1980 to 2014 and current best
practice, Finex 2014 Julius Wernher Lecture, 72 p.
Rivoirard, J, Renard, D, Celhay, F, Benado, D, Queiroz, C, Oliveira, L and Ribeiro, D, 2016. From the Spatial Sampling of
a Deposit to Mineral Resources Classification, Geostatistics Valencia 2016 (eds: J J Gómez-Hernández, J Rodrigo-
Ilarri, M Elena Rodrigo-Clavero, E Cassiraga and J A Vargas-Guzmán), pp 329–344 (Springer-Verlag).
Usero, G, Misk, S and Saldanha, A, 2019. An approach for drilling pattern simulation, Mining Goes Digital (ed: U Mueller),
pp 59–66 (Taylor and Francis Group: London).
Verly, G, Postolski, T and Parker, HM, 2014. Assessing uncertainty with drill hole spacing studies – applications to Mineral
Resources, in Orebody Modelling and Strategic Mine Planning Symposium 2014, pp 109–118, AusIMM, Melbourne.
Wawruch, T M and Betzhold, J F, 2004. Mineral Resource classification through conditional simulation, Geostatistics Banff
2004 (eds: O Leuangthong and C V Deutsch), pp 479–489 (Springer Verlag).

Mineral Resource Estimation Conference 2023 | Perth, Australia | 24–25 May 2023 191