Capping and kriging grades with long-

tailed distributions
by M. Maleki*, N. Madani*, and X. Emery*

as outliers or erroneous measurements and to

Synopsis omit them from geostatistical analyses. Such a
procedure is questionable if the extreme data
Variogram analysis and kriging lack robustness in the presence of corresponds to true values, because it may
outliers and data with long-tailed distributions, which often arises when
lead to underestimate the resources or
estimating grades in precious metal deposits. The capping technique,
reserves. The bias can be severe when
consisting of truncating the data to some top-cut grade, is widely used
in order to mitigate the influence of the values in the upper tail of the estimating the grades of elements with heavy-
distribution. However, this procedure omits part of the grade variability tailed distributions, such as gold or silver, for
and is likely to provoke a bias in the estimates. To avoid these issues, a which removing the highest values can affect
recently proposed approach is to decompose the grade of interest into the economic appraisal of the ore deposit.
three components (the truncated grade, a weighted indicator above the To mitigate this impediment, a number of
top-cut grade, and a zero-mean residual) and jointly estimate the sophisticated robust estimation procedures
truncated grade and the indicator by cokriging. This approach is have been proposed to reduce the influence of
attractive as it provides unbiased grade estimates, allows choosing the extreme high values (Journel and Arik, 1988;
‘optimal’ top-cut value, and essentially works with truncated and
Parker, 1991; Arik, 1992; Costa, 2003;
indicator data, thus avoiding the use of outlying values for calculating
Machado et al., 2011, 2012). A simple
sample variograms and performing spatial interpolation.
This work presents an application of this approach to a disseminated alternative is to truncate the extreme high
gold deposit that has been identified through exploration drilling. The values to some threshold or top-cut value, a
effect of using an indicator covariate is assessed through leave-one-out procedure known as ‘capping’ or ‘cutting’
cross-validation, by comparing the grade estimates with the true grades (Sinclair and Blackwell, 2002; Rossi and
and with the grade estimates obtained with the conventional capping Deutsch, 2014). As an example, Costa (2003)
approach, which considers only the truncated grade as the variable of proposes to truncate the values that deviate by
interest. As a result, cokriging the truncated grade and the indicator more than their cross-validation standard
above top-cut grade outperforms the conventional capping approach, deviation error. With this method, the same
yielding significantly more accurate estimates. A few complementary
high value can result in different truncated
guidelines are provided for validating the model hypotheses and for the
values, depending on the the kriging
implementation of cokriging.
neighbourhood, and no fixed top-cut is used.
In practice, a fixed top-cut value is often
Keywords considered, chosen (with some arbitrariness)
Top-cut model, high values, outliers, cokriging, indicator.
in the last percentiles of the data distribution.
This practice is widely used in the evaluation
of precious metal deposits (David, 1988; Krige,
1999; Dagbert, 2005) and is accepted in
Introduction current international codes for reporting
Geostatistics has received a great deal of mineral resources and ore reserves (SAMREC,
attention in the mining industry over the last 2007; JORC, 2012).
few decades, as it offers varied tools and However, although less detrimental than
models to assess mineral resources and ore removing extreme high values, considering
reserves (Journel and Huijbregts, 1978; David, truncated values omits some part of the data
1988; Krige, 1999). Geostatistical analyses are information and is likely to provoke a bias in
based on exploration data and/or production the estimates, which has to be assessed and, if
data such as the grade assays in drill-hole or possible, fixed. On this subject, Rivoirard et al.
blast-hole samples. Data integrity is essential (2013) recently presented a model in which a
to obtain accurate estimates of resources and
reserves. In this regard, data with extreme
high values may be problematic, because of * Department of Mining Engineering & Advanced
the effect on the sample variograms and Mining Technology Center, University of Chile,
resource and reserve estimates (Krige and University of Chile, Santiago, Chile.
Magri, 1982; Armstrong, 1984). The modeller © The Southern African Institute of Mining and
is often tempted to consider such extreme data Metallurgy, 2014. ISSN 2225-6253.
▲

The Journal of The Southern African Institute of Mining and Metallurgy VOLUME 114 MARCH 2014 255
Capping and kriging grades with long-tailed distributions
fixed top-cut value is used and the estimates are corrected for for the non-zero residual values, should be independent of
bias. The model is attractive from several points of view, as it the low grade values (below ze) and of the geometry of the
allows choosing the ‘optimal’ top-cut value and working with set A(ze) of locations with grade values greater than ze.
truncated and indicator data, thus avoiding the use of Rivoirard et al. (2013) state that this occurs when there is no
extreme high values for calculating sample variograms and edge effects within A(ze), i.e. when the value observed at a
for performing spatial interpolation. This work presents an location of A(ze) does not depend on whether or not this
application of this top-cut model to a case study in mineral location is close to the boundary of A(ze).
resources evaluation, together with proposed guidelines for The absence of edge effects can be verified by examining
practical implementation. Before dealing with the case study, the indicator variograms. Let us denote by γze (h) the
a short review of the model is presented. Mathematical details variogram of the indicator function 1Z(x)>ze and by γze,z (h)
and proofs can be found in the paper by Rivoirard et al. the cross-variogram between the indicators 1Z(x)>ze and
(2013). 1Z(x)>z with z greater than ze (under an assumption of
second-order stationarity, these variograms are functions of
Methodology the lag separation vector h). Then, in the absence of edge
effects, one obtains (Rivoirard, 1994)
Spatial interpolation with top-cut model
Let x denote the vector of coordinates associated with a [7]
spatial location, Z the variable targeted for estimation
(typically, the grade of an element of interest) and ze a top-
cut value. The variable (original grade) can be split into two Choice of optimal top-cut value
components, a truncated grade and an excess, as follows: The previous ratio of indicator variograms is the main tool for
[1] finding the most appropriate top-cut value ze. Indeed, one can
select several thresholds (candidate top-cut values) {z1,…,
where 1Z(x)>ze denotes an indicator function: zn} and calculate the ratios between the indicator cross- and
direct variograms associated with successive thresholds. The
[2]
first threshold for which the ratio is approximately constant
(independent of the lag separation vector) corresponds to the
In turn, the excess can be written as the sum of its minimal acceptable value (zmin) for choosing the top-cut
regression onto the indicator and a residual: value ze.
[3] Also, it is convenient (although not compulsory) to
choose the top-cut value so that the residual variability above
where m+(ze) is the mean value of Z above ze, so that m+(ze) ze is pure nugget effect. In such a case, there is no advantage
– ze is the mean excess above ze, and Rze (x) is the regression in choosing a higher top-cut value, since this would amount
residual. The proposed model assumes that this residual has to incorporating poorly structured grade values in the
no spatial correlation with the indicator function or with the truncated variable. In other words, when the residual
truncated grade: variability above ze is pure nugget effect, there is little or no
loss of information entailed by working with the truncated
[4]
grade min{Z(x),ze} and the indicator 1Z(x)>ze instead of the
original grade Z(x). Following Rivoirard et al. (2013), this
Under such an assumption, the estimation of the original condition can be checked by calculating a residual indicator
grade can be made by jointly estimating the truncated grade variogram defined as a difference between normalized
and the indicator via cokriging (CK) and separately indicator variograms:
estimating the residual via kriging (K):

[5] [8]

where z and z′ (with z < z′) are thresholds greater than zmin,
If, furthermore, the residual has no spatial correlation
while T(z) and T(z′) are the proportions of grade values
structure, its estimate is equal to its mean value, i.e. zero.
above z and z′, respectively. The threshold z for which the
This results in the following estimate for the original grade:
above difference is pure nugget provides the maximum
[6] acceptable top-cut value (zmax).
In summary, the examination of indicator variograms
The model therefore amounts to cokriging the truncated allows definition of an interval [zmin, zmax] in which to
grade and the indicator function. Both of them no longer choose the ‘optimal’ top-cut value ze. In addition to these
have extreme high values, in contrast to the original grade, considerations (Rivoirard et al., 2013), the following
which lends more robustness to the calculation of sample additional checks are proposed once the top-cut value has
variograms and to the final estimates. been chosen:
1) Calculate the truncated grade, indicator, and residual at
Validation of model assumptions each data location
As assumed in Equation [4], the residual should be spatially 2) Calculate the autocorrelation function (correlogram) of
uncorrelated with the truncated grade and the indicator. In the residual, in order to determine whether or not it is a
other words, the high grade values (above ze), which account pure nugget effect. If so, the contribution of the
▲

256 MARCH 2014 VOLUME 114 The Journal of The Southern African Institute of Mining and Metallurgy
Capping and kriging grades with long-tailed distributions
residual can be omitted in the estimation of grade ordinary cokriging system (Emery, 2012). This is more
(Equation [6]). Otherwise, one may prefer the restrictive than traditional ordinary cokriging, which assumes
estimator that accounts for the residual (Equation [5]) that m1 and m2 are unknown and unrelated (a somehow
3) Calculate the cross-correlogram between the residual naive assumption, because of the strong dependence between
and the truncated grade, as well as the cross- the indicator and the truncated grade, as highlighted in Table
correlogram between the residual and the indicator, I), but more versatile than simple cokriging, which assumes
and check that these are close to zero for every lag that both mean values are known without any uncertainty.
separation vector. This corroborates that the residual For the coefficients of the relationship between m1 and m2 to
has no spatial correlation with the truncated grade and be constant (Equation [9]), it is assumed that the mean
with the indicator, as it is assumed in the model grade below top-cut value, m-(ze), is constant in space and
(Equation [4]). known, while the probability that Z(x) exceeds or falls short
The use of sample correlograms and cross-correlograms is of ze may be locally variable and unknown.
suggested because of their robustness to outliers or to data In the next section, these tools and models are applied to
with extreme high values (Isaaks and Srivastava, 1988), a case study in mineral resources estimation.
although these tools may be biased estimates of the true
underlying spatial correlations because of the centring and Case study: porphyry copper-gold deposit
normalization of the data for each lag separation vector (the
magnitude of the bias depends on the number of data and
Presentation of the data set and deposit
their spatial distribution). Unbiased estimates of the spatial
correlation could be obtained with the variogram or with the The available data consists of samples taken from 167
non-centred covariance, at the price of a loss of robustness exploration drill-holes over an area of about 0.24 km2 in a
(Rivoirard et al., 2000, Chilès and Delfiner, 2012). porphyry copper-gold deposit (Figure 1A). The mineralization
is disseminated and subvertical in orientation, with a main
Type of cokriging direction dipping about 60° with respect to the horizontal
Let us return to the estimator in Equation [5]. This estimator plane. Mineralization comprises chalcopyrite, bornite,
relies on cokriging the truncated grade and the indicator and, digenite, chalcocite, covellite, molybdenite, and pyrite. Gold
if not nugget, kriging the residual. Rivoirard et al. (2013) mineralization occurs as native gold associated with gangue
consider the use of simple and ordinary kriging/cokriging, minerals, with particle sizes up to 160 µm, and as blebs
depending on whether the mean values of the variables are within bornite and chalcocite.
deemed known or not. In the following, it is of interest to estimate the gold grade
The latter case (unknown mean values) is interesting in within the deposit. To this end, the available samples have
practice, insofar as the means can vary in space, provided been composited to a length of 5 m. The distribution and
that they remain approximately constant at the scale of the statistics of the grade data are summarized in Figure 1B and
kriging or cokriging neighbourhood (assumption of local Table II, indicating a long-tailed distribution with a median of
stationarity). When omitting the residual term and using the 0.30 g/t, a mean of 1.01 g/t, and a maximum of almost 19
estimator in Equation [6] with ordinary cokriging, one g/t. To preserve the confidentiality of the data, the original
further assumes that the mean grade above top-cut value, values have been multiplied by a constant factor.
m+(ze), is constant in space and known (but the probability
that Z(x) exceeds ze may vary locally and is unknown), so Choice of top-cut value
that the local mean value of the residual is zero everywhere A set of thresholds (from 3 g/t to 6 g/t) are considered as
(Rivoirard et al., 2013). Now, if one denotes by m-(ze) the candidate top-cut values and the associated indicator
mean grade below top-cut value, it is possible to establish a variograms are calculated. From these, it is determined that,
relationship between the mean values of the truncated grade in every case, the ratio of cross-to-direct variograms
and indicator, as shown in Table I. (Equation [7]) does not vary significantly with the lag
Accordingly, the relationship between the mean values of separation distance (Figure 2A, 2C, and 2E), denoting the
the indicator (m1) and truncated grade (m2) can be modelled absence of edge effects within the set of locations with grade
in the following fashion: greater than the thresholds under consideration. Following
[9] Rivoirard et al. (2013), any top-cut value greater than or
equal to 3 g/t is therefore eligible. On the other hand, the
Even when considering that m1 and m2 are unknown, it variograms of indicator residuals (Equation [8]) display
is reasonable to assume that the previous relationship some spatial structure at short scales (up to approximately
remains valid and to incorporate this relationship in the 20 m) (Figure 2B, 2D, and 2F). Accordingly, the maximum

Table I
Relationships between mean values of indicator and truncated grade

Case Mean of indicator (m1) Mean of truncated grade (m2)

-
True grade less than or equal to ze 0 m ( ze)
True grade greater than ze 1 ze
▲

The Journal of The Southern African Institute of Mining and Metallurgy VOLUME 114 MARCH 2014 257
Capping and kriging grades with long-tailed distributions

Table II
Summary statistics on gold grade data (g/t)
Mean Standard deviation Minimum Lower quartile Median Upper quartile Maximum

1.02 2.00 0.01 0.12 0.30 0.91 18.86

Figure 1—(A) Location of drill-hole samples (projection onto horizontal plane); (B) histogram of gold grade data

eligible top-cut value is not well-defined, since the residual The above equation gives a valid coregionalization model,
variability above the candidate top-cut values is not pure as the eigenvalues of the sill matrix are non-negative
nugget effect. (Wackernagel, 2003). The direction of main continuity (with
Based on these premises, a top-cut value of 3 g/t has a correlation range of 40 m) is dipping 60° with respect to the
been finally chosen, corresponding to the 92% percentile of horizontal plane, in agreement with the known direction of
the gold grade distribution. This choice instead of a higher mineralization, whereas the variograms are found to be
top-cut value (4, 5, or 6 g/t) is motivated by the fact that the isotropic with a correlation range of 7 m in the plane
truncated gold grade is likely to exhibit a better spatial
orthogonal to this direction. Note that the fitting relies mainly
continuity, as it is no longer ‘contaminated’ by outlying data.
on the sample variograms at lag distances greater than 5 m
The gold grade can therefore be split into three components
(composite length), since few data pairs are involved in the
(a truncated grade, an indicator associated with the chosen
top-cut, and a residual), as per Equation [3]. To validate the calculation of the very first experimental point along each
model, it is convenient to check that there is no spatial direction (only 1 pair for the direction of main continuity and
correlation between the truncated grade and the residual, as 36 pairs for the orthogonal plane, while all the other experi-
well as between the indicator and the residual. This is done mental points involve several hundreds to thousands of data
by calculating the cross-correlograms between the residual pairs).
and the truncated grade and between the residual and the It is also interesting to determine whether or not the
indicator, and verifying that these correlograms are close to residual is spatially correlated, in order to determine which
zero for every separation distance (Figure 3). estimator (Equation [5] or Equation [6]) is best suited to the
data. To this end, instead of the traditional sample variogram
Variogram modelling of the residual, we calculated its sample correlogram, which is
The direct and cross-variograms of the truncated grade and a more robust spatial continuity measure (Isaaks and
the indicator are calculated along the identified main Srivastava, 1988). This sample correlogram is then converted
anisotropy directions and are fitted with a spherical structure into a standardized variogram, showing the existence of a
(Figure 4A, 4B, and 4C): spatial correlation structure, although with a shorter range
(28 m) than the truncated grade or the indicator. The residual
variogram can be modelled by an anisotropic spherical
structure, with the same anisotropy directions as the
[10] indicator and truncated grade (Figure 4D):

[11]
▲

258 MARCH 2014 VOLUME 114 The Journal of The Southern African Institute of Mining and Metallurgy
Capping and kriging grades with long-tailed distributions

Figure 2—(A, C, E) ratios of indicator variograms (Equation [7]); (B, D, F) variograms of indicator residuals (Equation [8]) (omnidirectional calculations)

Figure 3—Cross-correlograms between (A) truncated grade and residual, and, B) indicator and residual (omnidirectional calculations). Horizontal black line
indicates zero correlation
▲

The Journal of The Southern African Institute of Mining and Metallurgy VOLUME 114 MARCH 2014 259
Capping and kriging grades with long-tailed distributions

Figure 4—Sample (dots and dashed lines) and modelled (solid lines) direct and cross-variograms for truncated gold grade and indicator (A, B, and C) and
direct variogram of residual (D). Black: direction of main continuity (dip 60°); red: orthogonal plane

Resources estimation realistic. The locations of high grade estimates are controlled
by the indicator and by the residual, which correspond to the
According to the previous models, three approaches for two corrective terms introduced in the traditional estimate
estimating gold grade are compared: (Equations [5] and [6]).
1) Ordinary kriging of the truncated gold grade, The differences between the three approaches can be
corresponding to the traditional capping approach assessed globally, by calculating the mean grade above
2) Cokriging of the truncated gold grade and indicator, as different cut-off grades (Table III). It is seen that, for all the
proposed in Equation [6]. As stated previously, the cut-offs, the traditional approach (kriging of truncated grade)
mean values of both variables are assumed unknown yields biased estimates in comparison with the other two
but linearly related (Equation [9]) approaches. The bias can be explained because the traditional
3) Cokriging of the truncated gold grade and indicator, as approach works only with the truncated grade and neglects
above, together with ordinary kriging of the residual, the excess above the top-cut grade, whereas the other two
as proposed in Equation [5]. This third estimator is approaches account for such an excess via the covariates
(indicator and residual). Although the differences between
tested because the residual exhibits a spatial structure,
the two top-cut models (with and without residual) are
as reflected in Figure 4D.
distinguishable on the maps drawn in Figure 5, their effect on
In each case, the same search neighbourhood is used,
the global statistics on the estimates is not significant, at
consisting of an ellipsoid with semi-axes 400 m along the least for low cut-offs; this is explained because, by
main anisotropy direction and 100 m along the orthogonal construction, the residual has a zero mean value.
directions. This ellipsoid is divided into octants and up to
eight data are searched for in each octant. The dimensions of
Cross-validation
the search ellipsoid as well as the large number of searched
data (up to 64 for each target block) have been chosen in In order to compare the model performances, we realized
order to obtain as accurate and conditionally unbiased leave-one-out cross-validation: each data is successively
estimates as possible, even if they inevitably yield smoothed removed and re-estimated from the surrounding data (Journel
grade models (Rivoirard, 1987; Krige, 1996, 1997b; Vann et and Huijbregts 1978). The statistics on the cross-validation
al., 2003). errors (mean error, mean absolute error, and mean squared
The traditional approach (kriging of truncated grade) error) are reported in Table IV, while the scatter diagrams
leads to an estimate that is mostly less than the chosen top- between true and estimated gold grades are shown in
cut grade (3 g/t) (Figure 5A), although the true grade is Figure 6. These results confirm the bias of the traditional
likely to exceed this top-cut, as suggested by the data approach (mean error of -0.28 g/t), for which the estimates
histogram in Figure 1B. In contrast, with the cokriging are practically limited by the top-cut grade. These estimates
approaches (Figure 5B and 5C), the grade estimates are no are also conditionally biased, insofar as the regression of the
longer limited by the top-cut grade, which looks more actual gold grade upon the estimated grade is significantly
▲

260 MARCH 2014 VOLUME 114 The Journal of The Southern African Institute of Mining and Metallurgy
Capping and kriging grades with long-tailed distributions

Figure 5—Gold grade estimates at a given elevation obtained with (A) traditional approach, (B) and (C) top-cut model without and with residual

Table III
Statistics on estimated gold grade
Cut-off (g/t) Mean grade above cut-off (g/t)
Traditional capping Top-cut model without residual Top-cut model with residual

0.0 0.465 0.874 0.869

0.5 0.831 0.879 0.876
1.0 1.267 1.478 1.556
1.5 1.765 1.914 2.063
2.0 2.270 2.470 2.617
2.5 2.666 3.176 3.238

Figure 6—Comparison of true vs. estimated gold grades with (A) traditional approach, (B) top-cut model without residual, and (C) top-cut model with
residual. Solid blue lines: linear regression of actual grades upon estimated grades. Solid black lines: main diagonal
▲

The Journal of The Southern African Institute of Mining and Metallurgy VOLUME 114 MARCH 2014 261
Capping and kriging grades with long-tailed distributions

Table IV
Statistics on cross-validation errors
Traditional approach Top-cut model (without residual) Top-cut model (with residual)

Mean error -0.2784 0.0132 0.0170

Mean absolute error 0.4345 0.3847 0.2959
Mean squared error 2.1620 1.2188 0.5291

different from the diagonal, with a slope much greater than 1 COSTA, J.F. 2003. Reducing the impact of outliers in ore reserves estimation.
Mathematical Geology, vol. 35, no. 3. pp. 323–345.
(Figure 6A). The existence of a conditional bias is a pitfall for
the resource/reserve model, as it leads to an erroneous DAGBERT, M. 2005. Geostatistics in resource/reserve estimation: a survey of the
Canadian mining industry practice. Geostatistics Banff 2004.
assessment of the mining project (Krige, 1996, 1997a). In
Leuangthong, O. and Deutsch, C.V. (eds). Springer, Dordrecht.
contrast, the top-cut model does not suffer any global or pp. 345–350.
conditional bias in both cases (the mean errors are close to
DAVID, M. 1988. Handbook of Applied Advanced Geostatistical Ore Reserve
zero and the regressions of the actual upon the estimated Estimation. Elsevier, Amsterdam.
grades are close to the diagonal line). A greater accuracy is EMERY, X. 2012. Cokriging random fields with means related by known linear
achieved when accounting for the residual, as indicated by combinations. Computers and Geosciences, vol. 38, no. 1. pp. 136–144.
the smaller mean absolute and mean squared errors, and by ISAAKS, E.H. and SRIVASTAVA, R.M. 1988. Spatial continuity measures for
the smaller dispersion of the scatter diagram around the probabilistic and deterministic geostatistics. Mathematical Geology, vol.
diagonal line. 20, no. 4. pp. 313–341.
JORC. 2012. Australasian Code for Reporting of Exploration Results, Minerals
Conclusions Resources and Ore Reserves. Prepared by the Joint Ore Reserves
Committee of the Australasian Institute of Mining and Metallurgy,
Top-cut modelling is a helpful tool to handle extreme high Australian Institute of Geoscientists and Minerals Council of Australia.
values in the estimation of variables with long-tailed distrib- 44 pp.
utions, as it improves the accuracy of the estimates and JOURNEL, A.G., ARIK, A. 1988. Dealing with outlier high grade data in precious
reduces global and conditional biases with respect to the metals deposits. Proceedings of the First Canadian Conference on
traditional capping approach. Some additional guidelines Computer Applications in the Mineral Industry. Fytas, K., Collins, J.L., and
Singhal, R.K. (eds). Balkema, Rotterdam. pp. 161–171.
have been proposed, consisting of checking the independence
JOURNEL, A.G. and HUIJBREGTS, C.J. 1978. Mining Geostatistics. Academic Press,
of the residual with the truncated grade and with the
London.
indicator above top-cut grade, through the use of cross- KRIGE, D.G. 1996. A basic perspective on the roles of classical statistics, data
correlograms, and in replacing the estimators based on simple search routines, conditional biases and information and smoothing effects
kriging and cokriging by ordinary kriging and cokriging with in ore block valuations. International Conference on Mining Geostatistics.
related mean values. Unisa Publishers, Pretoria. pp. 1–10.
KRIGE, D.G. 1997a. A practical analysis of the effects of spatial structure and of
The case study presented shows that the residual should
data available and accessed, on conditional biases in ordinary kriging.
not be discarded when it has a spatial correlation structure. Geostatistics Wollongong’ 96. Baafi, E.Y., and Schofield, N.A. (eds).
On the contrary, if the residual is pure nugget or if its Kluwer Academic Publishers, Dordrecht. pp. 799–810.
structure cannot be interpreted properly (due to large fluctu- KRIGE D,G, 1997b. Block Kriging and the fallacy of endeavouring to reduce or
eliminate smoothing. Proceedings of the 2nd Regional Symposium on the
ations in the sample variogram or sample correlogram), the
Application of Computers and Operations Research in the Minerals
estimation can be limited to truncated grade and indicator Industry. Puchkov, L.A. (ed.). Moscow State Mining University Publishing
cokriging, and the modelling can then be made in a Center, Moscow.
convenient manner as it does not involve extreme high data KRIGE, D.G. 1999. Essential basic concepts in mining geostatistics and their
links with geology and classical statistics. South African Journal of
values. In the case study, the resulting estimates turn out to
Geology, vol. 102, no. 2. pp. 147–151.
be globally and conditionally unbiased, whether or not the KRIGE, D.G. and MAGRI, E. 1982. Studies of the effects of outliers and data
residual is taken into account in the model. The top-cut transformation on variogram estimates for a base metal and a gold ore
model therefore achieves a trade-off between accuracy, body. Mathematical Geology, vol. 14, no. 6. pp. 557–567.
simplicity of use, and robustness against extreme high MACHADO, R.S., ARMONY, M., COSTA, J.F.C.L., and KOPPE, J.C. 2011. Field
parametric geostatistics - geostatistical rigorous solutions dealing with
values. extreme grades during resource estimation. Proceedings of the 35th
International Symposium on the Application of Computers and Operations
Research in the Minerals Industry. Baafi, E.Y., Porter, I., and Kininmonth,
References R.J. (eds). Australasian Institute of Mining and Metallurgy, Melbourne. pp.
ARIK, A. 1992. Outlier restricted kriging: a new kriging algorithm for handling 191–201.
of outlier high grade data in ore reserve estimation. Proceedings of the MACHADO, R.S., ARMONY, M., and COSTA, J.F.C.L. 2012. Field parametric
23rd International Symposium on the Application of Computers and geostatistics - a rigorous theory to solve problems of highly skewed
Operations Research in the Minerals Industry. Kim, Y.C. (ed.). Society of distributions. Geostatistics Oslo 2012. Abrahamsen, P., Hauge, R.,
Mining Engineers, Littleton, CO. pp. 181–187. and Kolbjørnsen, O. (eds). Springer, Berlin. pp. 383–395.
PARKER, H.M. 1991. Statistical treatment of outlier data in epithermal gold
ARMSTRONG M. 1984. Common problems seen in variograms. Mathematical deposit reserve estimation. Mathematical Geology, vol. 23, no. 2.
Geology, vol. 16, no. 3. pp. 305–313. pp. 175–199.
CHILÈS J.P. and DELFINER, P. 2012. Geostatistics: Modeling Spatial Uncertainty. RIVOIRARD, J. 1987. Two key parameters when choosing the kriging
Wiley, New York. neighborhood. Mathematical Geology, vol. 19, no. 8. pp. 851–856.
▲

262 MARCH 2014 VOLUME 114 The Journal of The Southern African Institute of Mining and Metallurgy
Capping and kriging grades with long-tailed distributions
RIVOIRARD, J. 1994. Introduction to Disjunctive Kriging and Non-linear SAMREC. 2007. The South African Code for the Reporting of Exploration
Geostatistics. Oxford University Press, Oxford. Results, Mineral Resources and Mineral Reserves. Prepared by the South
RIVOIRARD, J., DEMANGE, C., FREULON, X., LÉCUREUIL, A., and BELLOT, N. 2013. A African Mineral Resource Committee, Working Group under the joint
top-cut model for deposits with heavy-tailed grade distribution. auspices of the Southern African Institute of Mining and Metallurgy and
Mathematical Geosciences, vol. 45, no. 8. pp. 967–982. the Geological Society of South Africa, 50 pp.
RIVOIRARD, J., SIMMONDS, J., FOOTE, K.G., FERNANDES, P., and BEZ, N. 2000. VANN, J., JACKSON, S., and BERTOLI, O. 2003. Quantitative kriging neighbourhood
Geostatistics for Estimating Fish Abundance. Wiley, New York. analysis for the mining geologist - a description of the method with
ROSSI, M.E. and DEUTSCH, C.V. 2014. Mineral Resource Estimation. Springer, worked case examples. Proceedings of the 5th International Mining
Dordrecht. Geology Conference. Dominy, S. (ed.). Australasian Institute of Mining
and Metallurgy, Melbourne. pp. 215–223.
SINCLAIR, A.J., and BLACKWELL, G.H. 2002. Applied Mineral Inventory
Estimation. Cambridge University Press, Cambridge. WACKERNAGEL, H. 2003. Multivariate Geostatistics: an Introduction with
Applications. Springer, Berlin. ◆

The Journal of The Southern African Institute of Mining and Metallurgy VOLUME 114 MARCH 2014 263