Glacken, I M and Snowden, D V, 2001.

Mineral Resource Estimation, in Mineral Resource and Ore

Reserve Estimation – The AusIMM Guide to Good Practice (Ed: A C Edwards), pp189-198 (The
Australasian Institute of Mining and Metallurgy: Melbourne).

Mineral Resource Estimation

By I M Glacken1 and D V Snowden2

INTRODUCTION constructed about the mid-point of intersections, with the grade

of each triangle being the arithmetic mean of the grades at each
The process of estimating a Mineral Resource can only take of its vertices. Other approaches include averaging values at
place after the estimator is convinced of the soundness of the the corners of other regular shapes or within grade contours.
fundamentals underlying the estimation process. Thus the
database of sampling, density, and other quality data for both The next step up in computational complexity is to apply
estimation and geological interpretation must have integrity and some weighting function to grades surrounding a point or block
robustness (Chapter 2); the geological data must be suffi- to be estimated. This is attractive where a series of regular
ciently complete for the definition of a geological model; the points or blocks need to be estimated in two or three dimen-
geological model itself must have internal consistency, should sions for subsequent planning or manipulation. The simplest
explain the observed arrangement of lithological and mineral- weighting function in common usage is based upon the inverse
ogical domains, and should represent the estimator’s best of the distance of the sample from the point to be estimated,
knowledge of the genesis of the mineral deposit; and the usually raised to the second power, although higher or lower
geological model should support the distribution of mineralisa- powers may also be used. Such inverse distance techniques
tion seen in the sampling (Chapter 3). It is only at this stage that introduce issues such as sample search and declustering
a resource model may be generated. The resource estimation decisions, and cater for the estimation of blocks of a defined
process involves the definition of mineralisation constraints or size, in addition to point estimates.
geological domains, the statistical and/or geostatistical analysis Further complexity in calculation is engendered by the use
of the sample data, and the application of a suitable grade inter- of geostatistical techniques, all based upon the theory of
polation technique. The final stage of the estimation process is regionalised variables developed by the French mathematician
to classify the resource according to the JORC Code (JORC, Matheron (1962, 1963). These methods all seek to utilise the
1999 and republished in this volume, see Appendix 1). spatial relationship between samples, as quantified by the semi-
variogram, to provide weights for the estimation of the
Synopsis of resource estimation techniques unknown point or block. The standard technique of
geostatistics, named kriging by Matheron in honour of the
Resource estimation techniques range in complexity, roughly
South African mining engineer Danie Krige, has many variet-
proportional to the amount of computation involved in deriving
ies, but those most commonly used are the variants of ordinary
the estimate. The simplest techniques involve assigning to an
kriging, the so-called linear kriging techniques.
orebody intersection its own clearly defined area of influence,
defined in relation to the other intersections. Such techniques The evolution of geostatistics in the last decade and a half
are known as polygonal, area-of influence, or has seen the development of a range of non-linear kriging
nearest-neighbour approaches. Polygons may be developed in techniques, based upon non-linear transformations of grades.
the plane of mineralisation, or may be projected onto a suitable These include the commonly used methods of indicator kriging
horizontal or vertical plane with the corresponding geometric and the various flavours of uniform conditioning and disjunct-
transformation. One very common version of this polygonal tive kriging. Non-linear kriging approaches seek to estimate a
approach is the sectional area-of influence method, with distribution of grades into each point or block, thus providing
polygons defined on sections (perpendicular to mineralisation) some measure of local uncertainty.
extended orthogonally to the section plane halfway to the next
section. The most recent development, and the most complex and
computationally intensive method used in resource estimation,
Limited averaging of grades occurs in some techniques, not is that of conditional simulation. This builds upon kriging and
common nowadays, where, for instance, triangulations are the use of stochastic (random) sampling approaches to provide,
in theory at least, a full measure of uncertainty. Conditional
simulation, while honouring data values locally, overcomes
many of the shortcomings of kriging methods. The negative
1. FAusIMM, CPGeo, Manager – Resource Evaluation Division, Snowden aspects of complex approaches such as this are the time of
Mining Industry Consultants, PO Box 77, West Perth WA 6872. E-mail: computational power required for its implementation and the
[email protected] lack of simplicity, which hampers understanding and accep-
tance.
2. FAusIMM, CPGeo, Director, Snowden Mining Industry Consultants, PO
Box 77, West Perth WA 6872. E-mail: [email protected]

189
I M GLACKEN AND D V SNOWDEN

All methods described above have a number of common The mining software packages continued to develop in
attributes; all must be based upon a carefully defined geologi- power and sophistication during the 1980s, and great advances
cal model, and all require extensive validation by the practitio- were made in the visualisation and modelling of complex
ner. A good overview of the various techniques is given by geological domains. It was during this time that the first
Carras (1998 and this volume). resource estimates sensu stricto were generated. These were
then modified by the mining design and other factors to yield
It should be noted that the resource estimation process,
Ore Reserves.
while often driven by the geoscientist, is a team endeavour, and
should include contributions from mining, metallurgical and Following its initial poor reputation, kriging techniques
often the commercial disciplines. continued to be used increasingly, especially the non-linear
approaches such as indicator kriging (Journel, 1983). However,
Another essential aspect of the process is quality control,
a large number of resources were still estimated using
including management of the estimation process and most
non-geostatistical techniques. This polarisation continues to the
importantly, management of the data sourced from the various
present day, although some form of block model is now
disciplines.
accepted as the norm in the majority of estimations. As
computers continue to grow in power and sophistication, a
History of resource estimation number of very computationally-intensive techniques such as
Since the start of the modern mining era there has been an conditional simulation are starting to become common.
increasing use of sample data to estimate the amount of material There have been many studies of the history of resource
mined or to be mined. Early mining operations had little estimation and the evolution of methods over time. Reeve and
forward planning, and consequently little use was made of Glacken (1998) describe the history of reserve and resource
exploration or mining data to estimate resources or reserves, estimation for one large Australian mining company, WMC
although there was extensive use of data for grade control and Resources Limited. Arvidson (1998 and this volume) describes
for monitoring of the mining operations. The early resource the process of bringing together over 100 years of data to gene-
estimations – which were in fact reserves – arose out of this rate a resource estimate for the gold deposits of the Golden
grade control function, where, for example, underground face Mile in Kalgoorlie. Murphy et al (1998) provide an overview
and raise samples were used to delineate a mining block. It was of the resource estimation process at the Bronzewing and
only in the middle part of the 20th century that exploration drill Jundee gold operations, Yandal province, WA.
holes started to be used to define a form of mineral inventory.
The earliest methods used were all variants of the polygonal Lutherborrow (1999 and this volume) describes the history
estimator, and typically, dilution and recovery were built into of estimation on the broken Hill field and the recent develop-
the estimation process to define an Ore Reserve without first ment of a three-dimensional methodology.
progressing through the resource stage.
As the importance of planning and scheduling of an opera-
tion’s mineral endowment became apparent, the use of all types
THE RESOURCE ESTIMATION PROCESS
of sample data to define Ore Reserves increased. However,
there was little regulation of the reporting standards and these Assignment of domains
tended to vary widely. The methods used to estimate reserves Once the geological model is as complete as the available data
also developed, and in the 1950s and 1960s there were a and knowledge of the setting and genesis of the mineralisation
number of attempts to fit simple mathematical models (such as allow, the data must be coded according to its domain. A
multiple linear regression) to the sample data and thus derive domain in this context is defined in the loosest sense, and repr-
simple statistical parameters such as the mean and variance of esents an area or volume within which the characteristics of the
the underlying population. In South Africa some success was mineralisation are more similar than outside the domain.
achieved by fitting lognormal distributions to mine data (Krige,
1978), but this failed to translate well to Australian orebodies. Hopefully the geological modelling will have highlighted a
number of domains, which should conform in some way with
With the advent of increasingly fast and reliable comput- the geology wherever possible. In many cases, the geological
ers, mine planning packages, which incorporated resource and units are the same as the mineralisation domains, such as in
reserve estimation, were introduced from the late-1970’s. many iron ore, sedimentary base metals or base metal sulphide
These provided for the first time the ability to generate a deposits. In this case, the grade modelling is constrained
model of the orebody comprising a large number of orthogonal, entirely by the geological modelling, and the resource grade
similarly-sized blocks. Accompanying this block modelling model will be a true reflection of the geology. In many other
ability the packages generally offered a variety of resource cases the mineralisation of interest does not entirely correspond
estimation techniques, including polygons (more strictly a to a geological unit, or transgresses geological units. This is a
nearest-neighbour interpolator), inverse distance estimation, typical feature of structurally-controlled mineralisation, such as
and simple or ordinary kriging. Although much work was shear-hosted gold deposits (eg Porgera, PNG). A number of
carried out in developing mining geostatistics in the 1960s and deposits, such as those in the Callie and Granites ore systems of
1970s in France and in North America, kriging was not the Northern Territory, combine elements of structural and
routinely used as a resource estimation technique until the end lithological control. A further class sees disseminated minerali-
of the 1970s. The method gained a somewhat tarnished reputa- sation scattered widely throughout a limited range of rock
tion due to its poor application by inexperienced practitioners, types, with no clear geological points to the distribution of
and inverse distance estimation, along with the various variet- grades. Typical examples of this style of mineralisation are the
ies of polygonal estimation, established themselves as the large porphyry-hosted copper and gold deposits, such as those
methods of choice. at Ok Tedi in PNG and Grasberg in Indonesia.

190
 CHAPTER 4 – Mineral Resource Evaluation

The resource estimation domains should honour the Once a series or several series of coherent domains have
geology wherever possible, but where this is not achievable, been defined, the numerical characteristics of the mineralisa-
some other form of domain boundary needs to be imposed. tion in each of these areas should be described. This will not
Typically this is a grade boundary defined by a cut-off grade only assist with the choice of a grade interpolation method, but
which should bear some relation to the economics of the deposit will also highlight any special data treatments (such as grade
to be evaluated, however preliminary the assessment is which cutting) which need to be applied. Where there are several
needs to be made. Domains may be defined by a combi-nation minerals or variables of interest, statistical analysis will reveal
of statistical and geostatistical means, in addition to or instead any patterns or correlations between them which need to be
of by a cut-off grade. Where grade alone is used to define the taken into account. Statistical analysis should take place within
domain boundaries, then it is risky to use a cut-off grade too the domains defined either by the geological model or by other
close to the overall economic cut-off of the deposit. If this is the approaches.
case, the result is often the overestimation of grades within the
One prerequisite of data analysis is that the samples all
domain, and the underestimation of grades outside the domain.
represent an equal volume. This is called the support of the
Some deposits show a rapid change from ore to non-ore, so
sample. The accepted way of ensuring equal support of all
selecting a natural cut-off is relatively safe (eg Osborne copper-
samples within a domain is to composite the samples (usually
gold deposit, NW Queensland, and most of the Archaean lode
drill- or interval-related data) into equal lengths. There are
gold deposits of the Yilgarn craton in Western Australia).
various algorithms for compositing, but the practitioner should
These boundaries may be termed hard, and greatly facilitate
always check that the results are as expected.
resource estimation. The other extreme is the gradual or soft
boundary, requiring much more careful treatment when Another problem which needs to be tackled before data
estimating resources. Structurally complex deposits such as analysis can take place, is preferential clustering of data, which
those at the Macraes gold mine, New Zealand, display a not only affects statistical analysis but can also bias the
combination of hard and soft boundaries. The well-defined variography. Since there is a natural tendency to drill more
hangingwall of the shear zone is a hard boundary, but the holes or take more samples of the higher grade portions of the
footwall is gradational and areas of associated stockwork mineralisation, clustering or irregular representation of samples
mineralisation also have soft boundaries. is a real possibility. Domaining might be necessary to correct
this. Alternatively, various declustering approaches are
It is possible, in defining domains for resource estimation,
available to ensure that each sample represents an equal
to impose several types of boundary conditions. Soft domain
volume for statistical analysis. Failure to take heed of preferen-
boundaries allow grades from either side of the boundary to be
tial clustering of data may lead to biased results. Note that it
used in estimating both domains, to varying degrees. Hard
may be desirable to treat different drilling types (eg diamond
domain boundaries do not permit interpolation of grades across
and reverse circulation) differently within the same domain.
domains. One-way soft domains are often used in estimation,
data within a high-grade domain is not used to estimate within If sampling is done on fixed lengths, then compositing
an adjacent low-grade domain, but estimation of the high-grade should be on multiples of the original sampling interval to
domain will use data from within the low-grade domain. Such minimise unnecessary smoothing. Zonal compositing (that is,
one-way boundaries introduce conservative bias and avoid compositing within domains) prevents this in many commer-
artificially sharp grade boundaries. More useful are partly soft cial packages, as composites commence after a change in zone
domain boundaries, where only limited crossing of domain at an odd interval, not at the start of an original sample,
boundaries is allowed. especially where zones are not defined on sample intervals.
The method of compositing adopted should take into account
Stegman (1999 and this volume) discusses the effects of
the style of mineralisation and boundary requirements. For
domaining on resource estimation in the Cobar region of New
example in a narrow vein, hard boundary environment, it is
South Wales.
essential to composite by zone to avoid over- or under-dilution.
Finally, it is worth bearing in mind that several different This is of less consequence where the boundaries are
types of domain may be used in the same region. A typical gradational.
example is the use of oxide, transition, and primary domains to
There are some schools of thought that believe that
allocate specific gravity in an oxidised mineral deposit.
composites should always be equal to the height of the mining
Similarly, metallurgical domains may be used in addition to
bench. There are advantages in employing bench compositing,
geological domains to define areas of differing metal recovery.
in that the input data is fully diluted to the bench height.
Often one set of domains is used for estimation and another for
However, where drill holes are angled at various orientations
mining/metallurgical purposes. As many different types of
with respect to the mining bench, this could result in different
domain as are needed to define the resource and to provide
support lengths for different composites, particularly when
information for reserve estimation should be used.
drilling is orientated close to horizontal. Downhole
compositing ensures that each sample represents the same
support. Assay lengths should not be split into smaller compos-
ite lengths, as this results in an artificially low variance for the
modified support as adjacent composites could be identical in
Data analysis value. Issues of bench compositing are starting to have less
importance as true three-dimensional methods of resource
Statistical analysis may help decide the nature of the domain estimation predominate, and in this paradigm each composite
boundaries. The analysis should include studies of how grades represents a point in three-dimensional space, irrespective of
change at domain boundaries. orientation.

191
I M GLACKEN AND D V SNOWDEN

Once compositing and declustering have been applied to the Geostatistical analysis will also reveal any anisotropy in
data, statistical analysis can take place. There are no recipes the domains of the mineral deposit and seek to quantify the
for this, but a good summary of statistical analysis techniques is magnitude of that anisotropy. The anisotropy may be repre-
given in Isaaks and Srivastava (1989). Desired outcomes from sented by the same total variation but at different ranges in the
statistical analysis of the data would be: various directions (geometric anisotropy), or by different
magnitudes of the variation in different directions (zonal aniso-
• depiction of any trends evident within the domains;
tropy). The analysis should also seek to verify the decision to
• depiction of data distributions in both histogram and use hard or soft domain boundaries, and should confirm the
cumulative probability form; amount of random variation, whether due to sampling
problems or due to inherent randomness in the data, in each
• a decision as to whether a distribution-based interpolation direction.
technique may be used;
Although variography should not be used as a substitute for
• definition of any data subsets within each domain; geological interpretation, it can indicate whether the geological
• comparison of different sample types; model is appropriate.

• establishment of the basis for any top cuts to be applied; Indicator variography may demonstrate that the amount
and direction of anisotropy varies with grade, eg high-grade
• choice of thresholds for indicator kriging, and veins may have a different orientation compared to the bulk of
• establishment of any relationships between multiple the mineralisation.
variables. Finally, the definition and modelling of semi-variograms
Statistical analysis may indicate that some domains have will help with the definition of the basic block size to be used in
very mixed populations, eg excessively high coefficients of any block modelling techniques, and will provide information
variation or multi-population probability plots. This may signal for the aggregation of grades into larger block sizes if required.
the need for more or different domains or, if it is not possible to Some Australian operators use variography to determine
separate out the populations, indicator techniques may be the range of influence and hence filter size for inverse distance
necessary for variography and kriging. interpolations.
Where there is a positive correlation between the bulk Geostatistical techniques for data analysis are presented by
density of the ore and the grade of the minerals of interest, Coombes (1997), Longley-Sinitsyna and Snowden (1997) and
typically in massive sulphide deposits, the bulk density should Vann and Sans (1995).
be involved in the actual estimation process, in other words,
sample intersections used for grade estimation should be
weighted by density as well as by length. Another option which
is often considered is the direct interpolation of bulk density The volume model
along with the grade variable(s). This method of density
weighting is used at many base metal deposits (eg Kambalda, Prior to grade estimation, it is necessary to convert the geologi-
(Glacken et al, 1998 and this volume)). cal model and/or the domain model into a physical, usually
three-dimensional, representation of the volume of mineralisa-
tion to be estimated. Common practice is to define a block
model, comprising a series of orthogonal cuboid blocks either
Geostatistical analysis of the same size or of subsets of a nominated ‘parent’ block
size. This is usually a semi-automatic process, but generally
As a precursor to any of the various kriging or conditional requires a confining shape in which to generate the blocks.
simulation techniques, spatial (geostatistical) analysis of the Typically this is given either by a three-dimensional enclosed
domained data – that is, the calculation and modelling of solid or by a series of surfaces. These are generated by
semi-variograms – is an obvious and necessary step. However, wireframing strings of points on section or plan, or by triangu-
the analysis of the continuity of data values in three dimensions lation of a series of points and strings into a digital terrain
is also a very useful precursor to almost any form of estimation, model, more commonly termed a surface. Surfaces may also be
as it defines, at the very least, the classical ‘range of influence’ interpolated from the raw data by a number of surface-fitting
of the data. Knowledge of this can and should have a bearing on techniques.
the choice of a suitable grade interpolation technique. For
instance, the direction and magnitude of the ranges may be Modern mining software has advanced to such a degree
used to define grade search parameters or the maximum size of that almost all of the major packages provide moderate to
polygons of influence. excellent tools for defining both the three-dimensional shapes
and for filling them with blocks. Key decisions for the practi-
The generation of semi-variograms and their subsequent tioner include:
modelling should reveal the structure of spatial continuity of
the data, and should confirm geostatistically any geological • How large should the blocks be compared to the data?
trends previously modelled or noted (see Chapter 3 for details • What should the relative shapes of the blocks be in the three
of geological interpretation techniques). The practitioner
dimensions?
should always seek a geological explanation for the principal
directions revealed by semi-variogram analysis. Quite often • How complex should the wireframes be?
such a step will reveal some subtle controls on mineralisation
• Should subcelling be introduced, or will the extra resolution
not immediately evident in the geology.
produce too many blocks?

192
 CHAPTER 4 – Mineral Resource Evaluation

While the advantages of a block modelling approach are tages, including the lack of applicability of the method to thick,
that it provides the framework for a good local estimation, that non-tabular bodies, and the assumption of an unrealistic model
it provides a model which lends itself readily to reserve estima- for grade variation. The major objections by advocates of
tion, and that it allows the modelling of mining selectivity, not geostatistical techniques include the issue of ignoring sample
all resource estimation is carried out to this detail. In some support (every sample has a different support, equivalent to the
cases a simple cross-sectional area of influence model is appro- size of the polygon or the area of influence) and possible condi-
priate, or a nearest-neighbour (polygonal) approach. tional bias (high-grade areas are overestimated and low-grade
areas are underestimated). This is discussed by David (1977,
p309).

Unsmoothed grade estimation

As mentioned above, despite the speed and power of modern
computers and the sophistication in estimation algorithms
Smoothed grade interpolation
which this allows, many practitioners carry out grade estima-
tion using unsmoothed or polygonal techniques. The essential Most resource estimation techniques, and most in common
aspect of all of these algorithms is that each grade, or series of practice today, use some form of grade smoothing to interpo-
grades in a defined intersection, is allocated unaltered to a late values into a block based upon surrounding samples.
specific area of influence. The simplest means of allocating
These fall neatly into two categories – the non-geostatistical
grades is via polygons of influence, which are generated by
methods and the geostatistical methods.
constructing the perpendicular bisectors between adjacent
samples or intersections. These are generally constructed in The methods of grade interpolation use some relationship
two dimensions on composited intersections, and rarely in between the distance of a sample from the block centre and the
three dimensions. In the block modelling context the polygonal weighting which it is given. The most commonly-used
approach is represented by the nearest neighbour approach, in approach weights each sample by some power of the inverse of
which each cell or block to be estimated assumes the grade of its distance from the block to be estimated, usually the second
its closest sample within the defined domain. This in effect or third power. The power chosen is somewhat arbitrary,
generates three-dimensional polygons. Although conceptually although it is well-known that the lower the power, the greater
very simple, many resource and reserve systems based upon the smoothing of grades. Higher powers of inverse distance
the polygonal approach are very complex in their treatment of tend to approximate a nearest neighbour approach, with distal
individual samples and their geometrical relationship to the ore samples receiving almost no weight.
surface. Many systems involve projection onto either a
horizontal or vertical plane and the subsequent geometric The geostatistical approaches to grade interpolation all rely
manipulation of areas and volumes. on some form of kriging, whereby the weights given to each
sample are derived from the semi-variogram model, which
A very common adaptation of the polygonal approach is defines the continuity of grades in two or three dimensions.
the cross-sectional resource estimation method. In this These geostatistical methods may in turn be subdivided into
approach, sectional interpretations are constructed, generally three classes – linear kriging, non-linear kriging, and simula-
orthogonal to the strike of mineralisation. Each separate ore tion. All geostatistical methods rely to a lesser or greater extent
intersection on each drill hole is allocated its own volume of on the assumption of stationarity, which is seen as the decision
influence, which usually extends halfway to the next drill hole to pool data within a given area or domain, and not as a hypoth-
up and down dip, and halfway to the next section in each strike esis which can be proven or disproven.
direction.
Linear kriging techniques are the simplest to apply, and
An adaptation of the polygonal approach in narrow centre on simple or ordinary kriging and their variants. These
orebodies is the use of accumulations (sometimes termed techniques are generally based on classical parametric statis-
service variables). Metal accumulations are the product of tics, which are affected by the distribution of the grade popula-
grade and thickness. In this approach the accumulation is tion underlying the data. Ordinary kriging is more resistant to
estimated, and the thickness is estimated independently. The departures from the assumption of stationarity than simple
final grade is obtained at each point by dividing the accumula- kriging, and is optimal for normal or Gaussian distributions of
tion by the thickness. The accumulation approach only works data, although still effective in other circumstances.
well where there is no correlation between grade and thickness,
and also requires the spatial orientation of the samples to be Non-linear techniques have gained in popularity in the last
taken into consideration as true (normal to dip) ore thicknesses ten years, and address some of the deficiencies of the linear
are required. It is worth noting that the accumulation and thick- techniques. All non-linear kriging techniques are based upon
ness variables can be estimated using any technique, including non-linear transformations of the sample data such as the
geostatistical methods. natural logarithm, the Gaussian (normal scores) transform, or
the indicator transform. The most widely-used non-linear
The advantages of the polygonal, nearest neighbour, and techniques are the various flavours of indicator kriging,
sectional methods are their simplicity and theoretical ease of uniform conditioning and disjunctive kriging. A comprehen-
application (although, as noted, the geometrical manipulations sive review of most non-linear kriging approaches is given by
may be extremely complex). Another bonus is the speed of Vann and Guibal (1998 and this volume). The perceived
obtaining a result. The polygonal estimator also has the added advantage of non-linear kriging techniques is firstly that they
advantage of being the perfect declustering technique for irreg- are able to cope with highly-skewed or mixed distributions of
ular data. However, there are a number of distinct disadvan- data (in other words, data most commonly seen in practice) and

193
I M GLACKEN AND D V SNOWDEN

secondly, that they are able to derive local distributions of

uncertainty which lead to a practical estimate of resources Volume-variance, sample search and block size
above a range of cut-off grades. Such estimates are known as issues
‘recoverable resources’, and while representing the correct It is well-known that there is a volume-variance relationship,
support for mining still need to be subjected to the reserve whereby grades based on a sample support are more variable
process. It is fair to say that indicator kriging is the most easily than grades based on a block support, and moreover that the
understood and the most commonly applied of all the larger the block, the less variable, or smoother, the distribution
non-linear kriging techniques; overviews of indicator kriging of the grades. It is thus important to model the appropriate
are given by Glacken and Blackney (1998) and Khosrowshahi block size to achieve a representative distribution on which to
et al (1998). Practical implementations of non-linear kriging predict the grade/tonnage relationship for a given mining
techniques are presented by Elliott et al (1997 and this volume), scenario.
Collings et al (1997 and this volume), and Matthews et al
(1999). Whatever the estimation approach adopted, all techniques
which seek to interpolate grades into blocks (except the nearest
Indicator kriging is a reasonably robust technique in that neighbour method) depend on the sample search procedure.
considerable changes in variogram parameters can be The methods by which samples are selected for subsequent
withstood before greatly affecting estimated grades. Some weighting are critical to the process, and in some cases (eg
other geostatistical estimation methods may be much more Carras, 1998 and this volume) are of more consequence than
dependent on the accuracy of the variogram models. As gold the estimation algorithm itself. A good sample search plan
typically has difficult variography (because of mixed minerali- should have some or all of the following features:
sation styles and often noisy sampling and assaying) robustness
is desirable. • declustering of data via octant or quadrant selection;
• restrictions on the number of samples from one drill hole;
• minimum and maximum numbers of samples specified for
search;
CONDITIONAL SIMULATION
• preferred search directions (ie anisotropic search); and
Conditional simulation, which typically uses a kriging
technique combined with a Monte Carlo sampling approach, • restriction of the influence of high-grade samples.
has the potential to enhance most of the geostatistical In common with the sample search plan are issues of block
approaches currently used. Conditional simulation produces a size for interpolation into block models. The block size to be
number of equally-likely pictures of the grade estimation, each used should ideally bear some relation to the mining equipment
of which honours the distribution and spatial continuity inher- planned or used, often referred to as the selectivity. The
ent in the input data. Conditional simulation also has the ability concept of the selective mining unit is the smallest parcel of
to reproduce the level of variability in the samples, in contrast ground on which mining decisions, such as the allocation to ore
to other smoothing methods which reduce the variability. or waste, may be made. The block dimensions should also be
Simulations allow the practitioner to choose a result which is considered in relation to the sampling grid; blocks that are too
commensurate with the objective of the investigation, and have small will result in oversmoothing of the sample data and
the ability to incorporate the risk and cost profiles of the opera- subsequent very low precision results.
tion under investigation. An overview of conditional simula-
tion for resource estimation is given by Khosrowshahi and Oversmoothing results in conditional bias, whereby
Shaw (1997 and this volume). Practical aspects are presented high-grade blocks are underestimated and low-grade blocks are
by Rossi and Alvarado (1998), Dowd (1996), and Glacken overestimated. A number of tools are available to decide on a
(1996). Schofield (1993 and this volume) describes a particular suitable block size – these are discussed in some detail by Krige
use of conditional simulation, that of identifying the optimum (1996).
drilling density. Undersmoothing results in a form of conditional bias
Conditional simulation may be most useful where data are whereby high-grade blocks are overestimated and low-grade
closely spaced (ie ample conditioning data) and may be seen as blocks are underestimated. Elliott et al (1997 and this volume)
supplementing other estimation methods (ie by providing describe the change in the grade/tonnage relationship depend-
sensitivity analysis) rather than replacing them. ing on whether a model is based on kriged block estimates
(smoothed) or raw assay data (unsmoothed). Reconciliation of
The key aspect of simulation which differentiates it from the estimates with the actual head grades achieved during
estimation is the consideration of probability or risk. It is possi- mining, plus the eventual processing of low-grade stockpiles,
ble to derive the probability of exceeding any given cut-off illustrates the sensitivities of the volume-variance effect in this
grade, or to define true confidence intervals of a specified case study.
degree of certainty around any single figure. The risk or confi-
dence is a function of the production unit; the chance of achiev- Conditional simulation provides some useful tools to check
ing target grades or blends is greater for large parcels of ore or determine optimal block sizes and the degree of smoothing
(such as six months production) than daily production units. In inherent in a resource. Small-scale simulations on representa-
any case, however, simulation allows the quantification of that tive areas or domains allow the change of support to be deter-
risk. mined more accurately and locally than using one of the global
techniques. The use of a simulation approach also allows the

194
 CHAPTER 4 – Mineral Resource Evaluation

determination of the selectivity and inherent smoothing in an Another technique expands search volumes during
estimated model. A case study using this application of simula- multi-pass estimations, with those blocks being estimated by
tion (Glacken et al, 1998 and this volume) details how a kriged the most distal samples having the lowest-confidence resource
model can be calibrated to achieve the desired selectivity. category.
One of the more useful criteria for resource classification is
the kriging variance or error arising from estimation. While
Checking the resource useless as an actual grade confidence measure (unless the data
is strictly normally distributed), the kriging variance depends on
Validation of the resource estimate can be undertaken using
the arrangement and continuity of samples around the block,
various tools including:
and thus is a good indicator of overall sample spacing which
• graphical plots of sections/plans showing input data and takes anisotropy and sample clustering into account. Other
block grade estimates; numerical approaches include the regression coefficient and
kriging efficiency measures proposed by Krige (1996).
• histograms of input and output grade distributions within
domains; There is a move in some sectors of the industry towards the
use of conditional simulation as a resource classification aid
• input and output mean grades within domains; and (Guibal, 1997; Snowden, 1996 and this volume). Since simula-
• generation of resource estimates using alternative tion quantifies the grade confidence, for a given block size, in
algorithms or approaches. addition to considering data position and continuity, it will
provide more information to assist resource classification than
The graphical validation plots of the resource need to make numerical indicators such as the kriging variance alone. Any
geological sense and show accord with the geological model. If simulation-based resource categories must be tempered by
not, further work must be carried out to resolve any anomalies. detailed consideration of non-grade factors, as with any
Once again, the importance of clustering on data statistics classification.
should be taken into account as this can bias the mean of the In most cases the resource classification will be based upon
input data. It is also important to note that the mean grade a combination of criteria, numerical and geological, with
obtained can also be biased if the population is skewed in distri- overall manual override by the Competent Person (Snowden,
bution, and the input mean grade needs to be corrected accord- 1996 and this volume).
ingly. This is typically done by applying a top cut. If the data is
perfectly lognormally distributed (ie the logs of the data show a
normal or Gaussian distribution), the Sichel mean is a better
estimator of the unbiased population mean.
Overview papers
A number of papers included in this volume provide an excel-
The best validation tool for any resource estimate is recon-
lent overview of the entire resource estimation and classifica-
ciliation with production (see Chapter 8). Comparisons
tion process. Collings et al (1997 and this volume) describe the
between the grade/tonnage curves for the resource model and
geological modelling, resource estimation and classification
the grade control model within a given identical volume and
aspects of a sedimentary-hosted iron ore deposit, while
with the production records for the corresponding mining
Lutherborrow (1999 and this volume) charts the evolution of a
period will give an operation the greatest confidence in the
new resource estimation system at the Broken Hill deposit. A
merit of the resource model.
similar historical perspective is presented by Glacken et al
(1998 and this volume) for the Kambalda nickel mines.

Resource classification Moorhead et al (1999 and this volume) present the resource
estimation procedures at the Cadia Hill mine in New South
The final task in resource estimation is to classify the resource – Wales. Pocock (1999 and this volume) describes why resources
this not only imparts levels of confidence in the results, but also at the Peak gold deposit, Cobar, NSW, have always been
dictates which classes of Ore Reserve may be generated by underestimated. The evaluation and classification of mineral
subsequent modification of the resource figures. The task of sands resources is comprehensively described in Lee (this
resource classification under the JORC Code (JORC, 1999 and volume).
also republished in this volume, see Appendix 1) is the duty of
the Competent Person, and therefore is ultimately a subjective
duty, but notwithstanding this, the resource classification OUTLOOK
should ideally be based upon as many objective factors as
possible. Clearly those factors related to the data and its config- The two developments that will most affect the future of
uration are of most importance, but these may be tempered by resource estimation are technology and the introduction of
overriding geological, mining or data-related issues. Chapter 9 more sophisticated estimation/simulation algorithms. The
discusses reporting and regulatory requirements. power of computers still continues to double roughly every 18
months, both in sheer numerical computational power and in
At the simplest level, resources may be classified on the graphical processing ability. This has greatly affected, and will
basis of the overall drill spacing within the various domains or continue to affect, the resource estimation procedure.
regions of the deposit. This is easily understood, but may not Real-time three-dimensional rotation of orebody and develop-
take into account any anisotropy which exists. Another ment models and pit and underground ‘fly-throughs’ are now
approach is to consider the average distance from the block commonplace, and this virtual reality interaction with models
centre of those samples used to estimate that block, or simply will continue and increase both speed and the depth of model-
the number of samples defined inside the search volume. ling features.

195
I M GLACKEN AND D V SNOWDEN

Developments in computing power in the last five years Coombes, J, 1997. Handy Hints for Variography, in Proceedings
have made such techniques as conditional simulation routinely National Conference on Ironmaking Resources and Reserves
Estimation (Eds: V N Misra and J S Dunlop), pp127-130 (The
possible, and further speed increases will expand the breadth Australasian Institute of Mining and Metallurgy: Melbourne).
and depth of such techniques which are available to the practi-
tioner. Conditional simulation for sensitivity and risk analysis David, M, 1997. Geostatistical Ore Reserve Estimation, 364 pp
(Elsevier: Amsterdam).
will become a routine tool in most resource estimation. The
increase in graphical processing power will also make possible Dowd, P A, 1996. Structural Controls in the Geostatistical Simulation
the simultaneous real-time visualisation of multiple simulated of Mineral Deposits, in Geostatistics Woloongong ’96 (Eds: E Y
models. Baafi and N A Schofield), pp647-657 (Dluwer: The Netherlands,
1997).
Although the current generation of mining software Elliott, S M, Snowden, D V, Bywater, A, Standing, C A and Ryba, A,
packages is highly sophisticated and will continue to increase 1997. Reconciliation of the McKinnons Gold Deposit, Cobar, New
in complexity, the changes will be evolutionary rather than South Wales, in Proceedings Third International Mining Geology
revolutionary. A new class of mining software will start to Conference, pp113-122 (The Australasian Institute of Mining and
emerge, combining powerful database facilities, extensive Metallurgy: Melbourne). Paper also revised and republished in this
volume as: Elliott, S M, Snowden, D V, Bywater, A, Standing, C A
modelling, visualisation and virtual reality features, and incor- and Ryba, A, 2001. Reconciliation of the McKinnons Gold
porating sophisticated software modules such as finite element Deposit, Cobar, New South Wales, in Mineral Resource and Ore
analysis, open pit and underground optimisation, interactive Reserve Estimation – The AusIMM Guide to Good Practice (Ed: A
data analysis, detailed cost and financial modelling, mining and C Edwards), pp257-268 (The Australasian Institute of Mining and
process simulation, and risk analysis, all centred around a three- metallurgy: Melbourne).
dimensional model or series of models of the orebody. Glacken, I M, 1996. Change of Support and Use of economic
Parameters IN Block Selection, in Geostatistics Wollongong ’96
Outside of the computer, the development of rapid and (Eds: E Y Baafi and N A Schofield), pp811-821 (Kluwer, The
highly-accurate spatial positioning systems such as differential Netherlands, 1997).
GPS are already affecting grade control and will start to impact Glacken, I M and Blackney, P C J, 1998. A Practitioners
on resource estimation as more data becomes available more Implementation of Indicator Kriging, in Proceedings of a one day
quickly. Depletion of resource models by excavation volumes symposium: Beyond Ordinary Kriging. (The Geostatistical
will become quicker and easier, leading to monthly or more Association of Australasia: Perth).
frequent resource updates and the commensurate flow-on Glacken, I M, Sommerville, B L and Arnold, C G, 1998. Reserve
effects on scheduling and forecasting. Remote assaying Estimation at Kambalda Nickel Operations – from 1970 to 2000, in
technology will provide instantaneous turnaround for sample Proceedings Gold and Nickel Ore Reserve Estimation Practice
data. The resource models of the future will incorporate many Seminar (held 31 October 1997), pp35-38 (The Australasian
Institute of Mining and Metallurgy, Kalgoorlie Branch:
diverse data types, collected and processed in real time.
Kalgoorlie). Revised and republished in this volume as: Glacken, I
M, Sommerville, B L and Arnold, C G, 2001. Reserve Estimation
at Kambalda Nickel Operations from 1970 to 2000, in Mineral
Resource and Ore Reserve Estimation – The AusIMM Guide to
Good Practice (Ed: A C Edwards), pp237-248 (The Australasian
REFERENCES Institute of Mining and Metallurgy: Melbourne).
Guibal, D, 1997. Techniques of Resource Estimation and Conversion
to reserves – Conditional Simulation, in Proceedings Resource to
Arvidson, H, 1998. KCGM – Fimiston Ore Resource Estimation
Reserve Inputs Seminar, pp23-25 (The Australasian Institute of
Practice, in Proceedings Gold and Nickel Ore Reserve Estimation
Mining and Metallurgy: Melbourne).
Practice Seminar (held 31 October 1997), pp59-69 (The
Australasian Institute of Mining and Metallurgy, Kalgoorlie Isaaks, E H and Srivastava, R M, 1989. An Introduction to Applied
Branch: Kalgoorlie). Revised and republished in this volume as: Geostatistics, 561p (Oxford University Press: New York).
Arvidson, H, 2001. KCGM – Fimiston Ore Resource Estimation
Practice, in Mineral Resource and Ore Reserve Estimation – The JORC, 1999. Australasian Code for Reporting of Identified Mineral
AusIMM Guide to Good Practice (Ed: A C Edwards), pp207-214 Resources and Ore Reserves (The JORC Code), The Joint Ore
(The Australasian Institute of Mining and Metallurgy: Melbourne). Reserves Committee of the Australasian Institute of Mining and
Metallurgy, Australian Institute of Geoscientists, and Minerals
Carras, S N, 1998. Let the Orebody Speak for Itself, in Proceedings Council of Australia (republished in this volume, see Appendix 1,
Gold and Nickel Ore Reserve Estimation Practice Seminar (held 31 pp683-698).
October 1997), pp3-18 (The Australasian Institute of Mining and
Metallurgy, Kalgoorlie Branch: Kalgoorlie). Revised and Journel, A G, 1983. Non-Parametric Estimation of Spatial
republished in this volume as: Carras, S N, 2001. Let the orebody Distributions, Math Geol, 12(4):285-303.
Speak, in Mineral Resource and Ore Reserve estimation – The
Khosrowshahi, S, Gaze, R and Shaw, W J, 1998. A Proposed Approach
AusIMM Guide to Good Practice (Ed: A C Edwards), pp199-206
to Change of Support Correction for Multiple Indicator Kriging
(The Australasian Institute of Mining and Metallurgy: Melbourne).
Based on p-field Simulation, in Proceedings of a one day
Collings, P S, Khosrowshahi, S and Ness, P K, 1997. Geological Symposium: Beyond Ordinary Kriging (The Geostatistical
Modelling and Geostatistical Resource Estimation of the Hope Association of Australasia: Perth).
North Deposit, in Proceedings National Conference on Ironmaking
Khosrowshahi, S and Shaw, W J, 1997. Conditional Simulation for
Resources and Reserves Estimation (Eds: V N Misra and J S
Resource Characterisation and Grade Control – Principle sand
Dunlop), pp105-116 (The Australasian Institute of Mining and
Practice, in Proceedings World Gold ’97 Conference, pp275-282
Metallurgy: Melbourne). Revised and republished in this volume
(The Australasian Institute of Mining and Metallurgy: Melbourne).
as: Collings, P S, Khosrowshahi, S and Ness, P K, 2001.
Revised and republished in this volume as: Khosrowshahi, S and
Geological Modelling and Geostatistical Resource Estimation of
Shaw, W J, 2001. Conditional Simulation for Resource
the Hope North Deposit, in Mineral Resource and Ore Reserve
Characterisation and grade Control – Principles and Practice, in
Estimation – The AusIMM Guide to Good Practice (Ed: A C
Mineral Resource and Ore Reserve Estimation – The AusIMM Guide
Edwards), pp269-284 (The Australasian Institute of Mining and
to Good Practice (Ed: A C Edwards), pp295-292 (The Australasian
Metallurgy: Melbourne).
Institute of Mining and Metallurgy: Melbourne).

196
 CHAPTER 4 – Mineral Resource Evaluation

Krige, D G, 1978. Lognormal – De Wijsian Geostatistics for Ore Reeve, J S, and Glacken, I M, 1998. The Evolution of Resource and
Evaluation, S Afr Inst Min Metall Geostatistics Monograph I, Reserve Practices within WMC – Past, Present, and Future, in
Johannesburg. Proceedings Gold and Nickel Ore Reserve Estimation Practice
Seminar (held 31 October 1997), pp19-34 (The Australasian
Krige, D G, 1996. A Practical Analysis of the Effects of Spatial Institute of Mining and Metallurgy, Kalgoorlie Branch:
Structure and of data Available and Accessed, on Conditional Kalgoorlie).
Biases in Ordinary Kriging, in Geostatistics Wollongong ’96 (Eds:
E Y Baafi and N A Schofield), pp799-810 (Kluwer, The Rossi, M E, and Alvarado, S B, 1998. Conditional Simulations Applied
Netherlands, 1997). to Recoverable Reserves, in Proceedings of APCOM ’98, 27th
International Symposium on Computer Applications in the
Lee, G, 2001. Mineral Sands – Some Aspects of Evaluation, Resource Minerals Industries, pp187-199 (The Institution of Mining and
Estimation and Reporting, in Mineral Resource and Ore Reserve Metallurgy: London).
Estimation – The AusIMM Guide to good Practice (Ed: A C
Edwards), pp315-321 (The Australasian Institute of Mining and Schofield, N A, 1993. Determining Optimal Drilling Densities for Near
Metallurgy: Melbourne). Mine Resources, in Application of Computers in the Mining
industry, Wollongong, 1993, pp34-41. Revised and republished in
Longley-Sinitsyna, D and Snowden, D V, 1997. Using Geostatistics as this volume as: Schofield, N A, 2001. Determining Optimal
a Tool in Fine Tuning of Iron Ore Resource Modelling, in Drilling Densities for Near Mine Resources, in Mineral Resource
Proceedings National Conference on Ironmaking Resources and and Ore Reserve Estimation – The AusIMM Guide to Good
Reserves Estimation (Eds: V N Misra and D S Dunlop), pp95-97 Practice (Ed: A C Edwards), pp293-298 (The Australasian Institute
(The Australasian Institute of Mining and Metallurgy: Melbourne). of Mining and Metallurgy: Melbourne).
Lutherborrow C, 1999. Evolution of Resource and Reserve Estimation Snowden, D V, 1996. A Practical Interpretation of Resource
Methods at Pasminco Broken Hill Mine Southern Underground Classification Guidelines, in Proceedings 1996 AusIMM Annual
Operations, in Proceedings Resource/Reserve Estimation Practice Conference, pp305-308. (The Australasian Institute of Mining and
in the Central West New South Wales Mining Industry (Ed: C L Metallurgy: Melbourne). Revised and republished in this volume
Stegman) (held 23 July 1998), pp3-10 (The Australasian Institute as: Snowden, D V, 2001. Practical Interpretation of Mineral
of Mining and Metallurgy: Melbourne). Revised and republished Resource and Ore Reserve Classification Guidelines, in Mineral
in this volume as: Lutherborrow, C, 2001. Evolution of Resource Resource and Ore Reserve Estimation – The AusIMM Guide to
and Reserve Estimation Methods at Pasminco Broken Hill Mine Good Practice (Ed: A C Edwards), pp643-652 (The Australasian
Southern Underground Operations, in Mineral Resource and Ore Institute of Mining and Metallurgy: Melbourne).
Reserve Estimation – The AusIMM Guide to Good Practice (Ed: A
C Edwards), pp215-220 (The Australasian Institute of Ming and Stegman, C, 1999. How Geological Envelopes Impact on the Resource
Metallurgy: Melbourne). Estimate – A Case Study of Advanced Projects in the Cobar Gold
Field, in Proceedings Resource/Reserve Estimation Practice in the
Matheron, G, 1962. Traité de Geostastique Appliqúe, (Technip: Paris). central West New South Wales Mining Industry (Ed: C L Stegman)
Matheron, G, 1963. Principles of Geostatistics, Econ Geol, 58:1246- (held 23 July 1998), pp97-120 (The Australasian Institute of
1266. Mining and Metallurgy: Melbourne). Revised and republished in
this volume as: Stegman, C, 2001. How Domain Envelopes Impact
Mathews, S, van Luyt, P, and Smart, G, 1999. Grade Domains Versus on the Resource Estimate – Case Studies from the Cobar Gold
Geological Domains at the Browns Creek Gold Mine, NSW, in Field, NSW, Australia, in Mineral Resource and Ore Reserve
Proceedings Resource/Reserve Estimation Practice in the central Estimation – The AusIMM Guide to Good Practice (Ed: A C
West New South Wales Mining Industry (Ed: C L Stegman) (held Edwards), pp221-236 (The Australasian Institute of Mining and
23 July 1998), pp89-94 (The Australasian Institute of Mining and Metallurgy: Melbourne).
Metallurgy: Melbourne).
Vann, J and Guibal, D, 1998. Beyond Ordinary Kriging – An Overview
Moorhead, C F, Dunham, P B, Eastwood, G J and Leckie, J F, 1999. of Non-Linear Estimation, in Proceedings of a one day symposium:
Cadia Hill – From Discovery to a Mine – A Case History, in Beyond Ordinary Kriging, (The Geostatistical Association of
Proceedings Resource/Reserve Estimation Practice in the Central Australasia, Perth). Revised and republished in this volume as:
West New South Wales Mining Industry (Ed: C L Stegman) (held Vann, J and Guibal, D, 2001. Beyond Ordinary Kriging – An
23 July 1998), pp67-78 (The Australasian Institute of Mining and Overview of Non-Linear Estimation, in Mineral Resource and Ore
Metallurgy: Melbourne). revised and republished in this volume as: Reserve Estimation – The AusIMM Guide to Good Practice (Ed: A
Moorhead, C, Dunham, P, Eastwood, G and Leckie, F, 2001. Cadia C Edwards), pp249-256 (The Australasian Institute of Mining and
Hill: From Discovery to a Measured Resource – A Case History, in Metallurgy: Melbourne).
Mineral Resource and Ore Reserve Estimation – The AusIMM
Guide to Good Practice (Ed: A C Edwards), pp97-108 (The Vann, J and Sans, H, 1995. Global Resource Estimation and change of
Australasian Institute of Mining and Metallurgy: Melbourne). Support at the Enterprise Gold Mine, Pine Creek, Northern
Territory – Application of the Geostatistical Discrete Gaussian
Murphy, R, Landmark, J, Warner, J, Riddington, D and Ekers, B, 1998. Model, in Proceedings of APCOM XXV, pp171-179 (The
Towards a Better Resource Estimate – The GCML Experience, in Australasian Institute of Mining and Metallurgy: Melbourne).
Proceedings Gold and Nickel Ore Reserve Estimation Practice
Seminar (held 31 October 1997), pp101-120 (The Australasian
Institute of Mining and Metallurgy, Kalgoorlie Branch:
Kalgoorlie).
Pocock, J, 1999. Why Feasibility Resource Estimates Under-Valued
the Peak Orebody, in Proceedings Resource/Reserve Estimation
Practice in the Central West New South Wales Mining Industry
(Ed: C L Stegman) (held 23 July 1998), pp13-28 (The Australasian
Institute of Mining and Metallurgy: Melbourne). Revised and
republished in this volume as: Pocock, J, 2001. Why Feasibility
Resource Estimates Under-Valued the Peak Orebody, in Mineral
Resource and Ore Reserve estimation – The AusIMM Guide to
Good Practice (Ed: A C Edwards), pp299-314 (The Australasian
Institute of Mining and Metallurgy: Melbourne).

197