Downloaded from [Link]

org/ at Pennsylvania State University on May 16, 2016

Recoverable reserve estimation by conditional simulation

PETER J. R A V E N S C R O F T
R T Z Consultants Limited, PO Box 50, Castlemead, Lower Castle Street,
Bristol BS99 7YR, UK

Abstract: Recoverable reserves are defined as the tonnage and grade that can be recovered
above a given cut-off grade in a selective mining operation. Their estimation depends on
three factors: the support, information and constraint effects. These factors are discussed,
and some traditional methods for estimation of global and local recoverable reserves are
reviewed. The technique of conditional simulation is proposed as a practical alternative to
some of the more complex methods currently used. The broad concepts of conditional
simulation are discussed, and some alternative simulation methods presented. A case study
on a set of real mining data illustrates the use of the conditional simulation approach to
recoverable reserve estimation, with the support and information effects being modelled
very simply. Results are compared with the discrete gaussian model, and with ordinary
kriging, which is shown to be unsuitable for recoverable reserve estimation.

The evaluation of a mining project rarely stops cult to reconcile ore reserve estimates without
at the estimation of a geological in-situ resource. knowing the exact implications of the terms
For the resource to be turned into an economic employed. This paper will not attempt to resolve
ore reserve, a variety of technical and economic this controversy, but the meaning of recoverable
considerations are introduced. One of the most reserve estimation within the scope of the paper
common of these is the choice of a cut-off grade, will be discussed.
below which it is not economic to process mined In strict geostatistical terminology, an esti-
ore. mate of recoverable reserve has come to be
The application of a cut-off grade implies a understood as the estimate of the final tonnage
selection procedure that divides the resource into and grade that will be mined and sent for treat-
ore and waste, and the degree of success of this ment after the application of a mining cut-off
selection has a direct impact on the economic grade. It involves all of the well known aspects
valuation of the project. Since the selection will of selectivity and mining dilution, which in geo-
never be perfect, it is important in the project statistical terms have been identified as the result
valuation stage to be able to forecast the selecti- of three factors: (i) the support effect (ii) the
vity which will be attained during the mining information effect and (iii) the constraint effect.
operation, and to adjust ore reserve estimates to
take account of the effects of imperfect selection.
This estimation of recoverable reserves is a
The support effect
complex problem which normally requires com- In most mining operations, there is normally
plex mathematical techniques for its solution. some selection of parts of the orebody which will
The method of conditional simulation provides a be processed and other parts which will either
more practicable approach to the problem, and not be mined or will be sent to a waste dump or
is becoming widely recognized as a viable tool stockpile. In a highly selective mining operation
for ore reserve estimation and mine planning. this decision may be made on the basis of the
This paper reviews the concepts of recoverable individual grade values of very small blocks,
reserve estimation and conditional simulation, with only those blocks having a grade above the
and illustrates its application in a simple economic cut-off being selected for treatment.
example. The size of these blocks, or selective mining
units (SMUs), varies depending on the oper-
ation, but may relate to the capacity of a haul
Estimation of recoverable reserves
truck in an open pit mine or to the size of a stope
The definition of ore reserve categories and such in an underground mine. In practice, the size of
concepts as geological reserves, in-situ reserves, the SMU is largely determined by physical con-
mineable reserves and recoverable reserves, is siderations such as equipment selection, blasting
the subject of ongoing discussion and contro- requirements or rock mechanics, but the SMU
versy in most mining countries. It is often diffi- size directly controls the degree of selectivity

From Annels, A. E. (ed.), 1992, Case Histories and Methods in Mineral Resource Evaluation 289
Geological Society Special Publication No. 63, pp. 289-298.
 Downloaded from [Link] at Pennsylvania State University on May 16, 2016

290 P.J. RAVENSCROFT

which can be attained, and the consequent value is not known. The selection is normally
recovered grade and tonnage. based on an estimated value, for example the
For example, in an orebody where the minera- value obtained by averaging the surrounding
lization is of a patchy nature on a very small blast-hole values in an open pit, or the closest
scale and if it were possible to mine it using face samples in an underground operation.
teaspoons and decide whether each teaspoon The quality of this estimate may not be good,
should be treated or thrown away on the basis of and in many instances incorrect selective mining
its grade, a near perfect extraction of all the decisions will be made on the basis of incorrect
economic mineral would be attained. If, how- estimates. Blocks, having an average grade that
ever, the selection can only be based on the is actually above the cut-off, may be sent to
overall grade of a wheelbarrow full of ore, most waste since their estimated value was too low,
wheelbarrows sent to the treatment plant will and unpayable blocks may be mistakenly selected
contain some uneconomic teaspoons, and simi- for processing.
larly a number of payable teaspoons will be lost This effect is illustrated by a typical plot of
in discarded wheelbarrows having an average actual SMU grades against their estimated
grade that is too low. The overall effect of the values (Fig. 2).
larger SMU is a loss of some higher grade mat-
erial to the waste category, and a dilution of Ore
recovered ore by unpayable material. discarded § §
As another example, Fig. 1 shows a schematic as waste
4" ..-"""
view of a mineralization pattern of high grade , ...'

nodules surrounded by low grade material, and 4" 4" , .."

the effect of different SMU sizes. It can be seen Actual 0

that the smaller SMU allows for much more Value 0 o
4" 4" 4",'"'+
effective ore/waste discrimination than the larger 0 0 4" . - ' " § §
blocks. 0 0 0

4"4" ; .4 '+
0 o
+o
..'F 0

o,,.82176176 Poor selection 4" '.." 4",1,4" 0

o ~ Waste treated
as ore

I_,r" of high grade

patches with
...
. ' " " 4" 4" 4"

Estimated Value
large SMU
D 4D Fig. 2. Imperfect ore/waste selection due to the infor-
mation effect.

While the dilution and downgrading effect of

lll! lq ltiiiii
Much improved
this poor selection can be minimized by better
grade control sampling and optimized esti-
ore/waste mation techniques, it can never be eliminated.
discrimination However, the impact that it has on recoverable
with small S M U reserves, known as the information effect, can be
estimated and must be included in any recover-
Itlll
~lllllll able reserve estimation study.

Fig. 1. Schematic illustration of the support effect.

The constraint effect
These examples are obviously extreme, but it Incorporation of the support and information
is clear that the size and shape of the SMU effects produces an ore reserve estimate that
relative to the variability of the mineralization assumes free selection of SMUs based purely on
has a critical impact on the selection of recover- their estimated grade, and takes no account of
able reserves, and this support effect must be the physical constraints imposed by the mining
taken into account in the estimation process. operation.
However, it may, for example, not be possible
to mine a particular SMU in an underground
The information effect room and pillar operation due to roof support
Unfortunately, at the time of the decision on considerations, or the haul road design in an
whether to treat or discard an SMU, its actual open pit may preclude mining of a certain
 Downloaded from [Link] at Pennsylvania State University on May 16, 2016

CONDITIONAL SIMULATION 291

number of payable mining blocks. The schedul- All of them depend on a good estimate of the
ing constraints required by a rapid tonnage grade frequency distribution, obtained from the
build-up in the early production period may sample histogram, and an estimate of the disper-
result in forced treatment of non-economic sion variance of SMU grades calculated from the
material, or stockpiling and materials handling variogram. However, the fundamental assump-
restrictions may impose production schedules tions differ in each technique, and the author's
which are not optimal in terms of grade selecti- preference for the discrete gaussian model is due
vity. to its robustness in situations where some of
Whilst the impact of factors such as these and these assumptions prove to be invalid.
many others is difficult to model mathematically, The information effect is taken into account
the incorporation of this constraint effect is nor- less often in practice, and this has lead to some
mally made in the ore reserve estimation by the serious estimation biases in previous mining
introduction of a number of factors. These are project valuation. The techniques above can again
usually based on previous production experi- be used with certain limitations in estimating the
ence, and may be somewhat arbitrary, but are impact of the information effect, but again the
essential in arriving at a realistic estimate of final discrete gaussian model is favoured as the most
recoverable tonnage and grade. suitable, providing a simple framework within
which to assess the complex problems of optimi-
Methods o f recoverable reserve estimation zation of grade control and selection procedures.

The support and information effects described

above can be determined mathematically once a Local recoverable reserves
model of the grade distribution and spatial con- In the local estimation of recoverable reserves,
tinuity in the orebody has been obtained by a an attempt is made to determine on a local scale
representative histogram and variogram. A good what the recoverable tonnage and grade will be
estimation technique should allow the flexibility for a given cut-off grade, including the effects of
of modelling a range of cut-off grades and a the SMU size and the imperfect selection from
variety of configurations of SMU size and shape estimated values. Although simplistic geostatisti-
and grade control sampling patterns. Many cal studies pretend that linear techniques such as
alternative techniques have been proposed and a ordinary kriging provide such results, the impli-
number are in current use in the mining industry. cations of this are well known (David 1972). It
Some of these are reviewed briefly in the dis- should rather be accepted that local estimation
cussion that follows, with a distinction made detail is not possible without local sampling
between techniques that allow for global reco- detail, and that attempts to estimate, for
verable reserve estimation, and those providing example, 5 x 5m SMUs from 50m sample
more detailed local estimates. spacing, are illusory.
It should be made clear that linear interpola- Local recoverable reserve estimation by exist-
tion techniques, such as ordinary kriging, should ing methods limits itself to the provision of grade
not be considered as tools for recoverable re- and tonnage frequency distributions within a
serve estimation: they have their uses in other local area. In the above example, it could pro-
areas but provide totally unrealistic results when vide frequency distributions of 5 m SMUs within
used to estimate recoverable tonnage and grade 50 m panels, without explicitly assigning a grade
above a cut-off for a selective mining operation. estimate to each SMU. This permits an estimate
of the recoverable tonnage and grade on a rela-
Global recoverable reserves
tively local basis, such as a mining bench or
scheduling period, but clearly does not provide a
The objective of global recoverable estimation is mining engineer with individual SMU values for
to provide an overall grade and tonnage estimate input into a detailed scheduling or planning
for an entire mining project or a large defined system. While this limitation is lamented by
portion of the project. No detail is provided of planning engineers, it does prevent the detailed
where payable tonnage is located, and the result processing of results that can be largely mean-
is only applicable over the total extraction ingless if small block values are used.
period of the area being estimated. Again, a variety of non-linear geostatistical
The support effect can be estimated by such techniques has been proposed for local recover-
techniques as the affine correction, indirect log- able reserve estimation, with the two most
normal correction, lognormal shortcut or dis- commonly used being disjunctive kriging and
crete gaussian model, all of which have been indicator kriging. Although somewhat more
applied with varying degrees of success. complex to implement, disjunctive kriging has
 Downloaded from [Link] at Pennsylvania State University on May 16, 2016

292 P . J . RAVENSCROFT

been seen in the author's experience to provide a mation, the simulated model corresponds to
more flexible result for changing cut-off grades, reality. At points falling between known sampled
support sizes and sampling configurations. Some locations, the model contains values which are
of its underlying assumptions are often criti- possible representations of reality, and repeated
cized, but newer extensions of the method which simulations will contain different values at these
allow for different mathematical transform- unknown points. Each alternative simulated
ations appear to provide greater tractability. model can be seen as one equally possible rep-
resentation of the as yet unknown reality.
Limitations of recoverable reserve estimation The essential difference between a model
obtained by conditional simulation and one
techniques
obtained by an interpolation method such as
As outlined in the preceding discussion, many kriging, is that the simulated model respects the
alternative techniques have been derived for the true spatial variability, whereas an interpolated
estimation of both global and local recoverable model provides a necessarily smoothed image of
reserves, with the direct incorporation of the reality. While conditional simulation fills in the
support and information effects. However, none gaps between known sample points with values
of them allows for the estimation of the impact which are possible realizations of what the actual
of practical mining constraints on the recover- picture may be, interpolation fills in the gaps
able tonnage and grade, and sophisticated esti- with averaged values which are in fact highly
mates are thus often subjected to coarse factor- unlikely results.
ing to include such considerations. Conditional simulation methods have evolved
In addition, the estimation methods sensibly since their initial development until there is now
limit themselves to a general distribution of a wide range of available approaches, each of
grade and tonnage, even on a local basis, rather which is suited to different applications. From
than a futile attempt to put an estimated value the early turning bands methods first proposed
in small SMUs when the limitations of the in Journel (1974), through to sequential tech-
sampling density prohibit the provision of such niques described by Isaaks (1990) and Gomez-
detail. Hernandez & Srivastava (1990), simulation
The approach of conditional simulation, how- algorithms have been devised which can now be
ever, provides a way round both of the above run on personal computers and workstations
limitations, and gives a very practical alternative and their application is becoming more wide-
method of recoverable reserve estimation. spread.
The ability to create a numerical grade model,
with values on a very dense grid, opens up a
Conditional simulation
number of possibilities of application in ore
The method of conditional simulation has been reserve estimation and mine planning, with all
in existence for nearly twenty years, but has operations in the estimation and planning pro-
had relatively limited application in the mining cess able to be tested and optimized using simu-
industry. Until recently, the need for excessive lated models. As was the case in classical stat-
computer resources limited the use of con- istics and operations research where computer
ditional simulation largely to research exercises simulation came to replace more complex math-
rather than practical application. However, with ematical procedures of estimation and optimiza-
the advent of cheap computing power and tion, so in geostatistics the complex mathe-
storage capacity, the conditional simulation matical formulations can often be replaced with
approach is becoming a much more attractive computer intensive simulation. A particular case
possibility, and is starting once again to receive of this is the use of conditional simulation to
attention from the mining industry. produce both global and local estimates of
A good general overview of conditional simu- recoverable reserves.
lation and its application can be obtained from
Journel (1979), but for the purposes of this paper
it can be described as a 2 or 3 dimensional Application of conditional simulation to
numerical model of an orebody with grade
recoverable reserve estimation
values represented on a very detailed grid. The
features of this model are that the grade values The three aspects of recoverable reserve esti-
respect all that is known about the actual miner- mation described earlier can each be approached
alization in terms of frequency distribution and by the use of conditional simulation. In the case
spatial continuity, and that at points where the of the support and information effects con-
actual situation is known from sampling infor- ditional simulation provides a direct solution,
 Downloaded from [Link] at Pennsylvania State University on May 16, 2016

CONDITIONAL SIMULATION 293

while for the constraint effect a conditional and grade can be derived from the simulated
simulation model can be used as the basis for a SMU grade.
detailed study on the mining process to quantify The method provides a direct result with no
the effect of the various practical mining con- complex assumptions or calculations, and can
siderations. easily be used to assess the effect of different
sampling patterns or even different estimation
Estimation of the support effect by methods. Repeated exercises of the same sort on
alternative simulation models can provide a
conditional simulation measure of the sensitivity of the results, and once
Estimation of the support effect by traditional again results can be used on either a global or
non-linear geostatistical techniques involves the localized basis.
modelling of the histogram of SMU grades,
normally done by estimating the variance of Estimation of the constraint effect by
this distribution as the block dispersion variance
of the SMU and applying this to some form
conditional simulation
of transform of the sample histogram. This Although conditional simulation provides no
approach necessitates a number of assumptions direct way of estimating the effect of a wide
about permanence of distribution or bivariate range of practical mining constraints on recover-
distributions of transformed values, and can be able reserve estimates, it certainly can act as the
very sensitive to departures from these assump- basis for such modelling.
tions. With a detailed spatial model of the o r e b o d y
Since a conditional simulation model, how- produced by simulation, mine design and sche-
ever, contains grade values on a dense grid of duling applications can be run directly, or more
points, these point values can easily be averaged sophisticated time/process simulation methods
to give grade values for any size of block can be applied. The simulated orebody can effec-
required. Grade/tonnage curves of resulting tively be 'mined out' and the detailed production
block values can be generated immediately, and results recorded. In this way, actual results can
the support effect of differing SMU sizes can be be compared to predicted results and a set of
determined directly. On a local basis, grade/ factors derived for application to the original ore
tonnage curves of SMUs within larger mining reserve estimates. Once again, repeated simula-
panels can be estimated in a similar way. tions can be used to test the robustness of the
estimated factors and measures of confidence in
Estimation of the information effect by the final recoverable reserve estimate can be
conditional simulation obtained.

In order to model the information effect, esti-

mates of SMU grades are required based on
General
what will be the final grade control sampling The use of the conditional simulation approach
configuration. Selection is performed on these for recoverable reserve estimation has been seen
estimated values, and a grade/tonnage curve of in the preceding discussion to provide a practical
the actual recovered grades is derived. In non- alternative for the modelling of the support and
linear techniques allowing for this calculation, in information effect and an ideal basis for the
particular the discrete gaussian model for global measurement of the constraint effect.
recoverable reserves and disjunctive kriging for A full example of the use of conditional simu-
local recoverable reserves, a complex mathe- lation in this way can be found in Isaaks (1990),
matical formation is required. Other non-linear but the limited example presented in the follow-
techniques may not even allow for the correct ing section illustrates its application on a set of
estimation of the information effect. real mining data.
By the use of a conditional simulation model,
however, a combination of simulated SMU Example of application
grades and simulated grade control sample
values can be obtained from the same model. By Data used in the study
extracting point values on a grid equivalent to The data used in this example formed the basis
the final grade control sampling configuration, of a previous study into the dangers of kriging
estimates of the SMU grades can be obtained small blocks (Ravenscroft & Armstrong 1990),
from these simulated sample values. These esti- and consist of heavily transformed values of
mated can be used as the selection criteria for actual close-spaced sampling results from one
each SMU, and the actual recovered tonnage level of a working mine.
 Downloaded from [Link] at Pennsylvania State University on May 16, 2016

294 P . J . RAVENSCROFT

Fig. 3. Actual data and extracted sampling data.

Fig. 4. Plot of first two conditional simulations.

Figure 3 shows a plan of the total data set found on the original grid. This clustered con-
together with a set of values extracted on a figuration is typical of most valuation sampling
reasonably regular grid and representing typical campaigns.
wide spaced sampling. In this example, the Both the underlying total data set and the
underlying data are on a 2 m regular grid, while sampling values, even after declustering, follow a
the extracted sampling data set is on a 24 rn grid skewed frequency distribution with a large pro-
with infill sampling around high grade values portion of low values and occasional very high
 Downloaded from [Link] at Pennsylvania State University on May 16, 2016

CONDITIONAL SIMULATION 295

values. The coefficient of variation is 1.08. The blocks is only one quarter that of the actual and
variogram of the total data set shows a nugget simulated blocks.
effect of some 25% and an effective range of In the case of 24m SMUs, the differences
about 30 m, and the sample data set shows an between kriged, simulated and actual values
experimental variogram consistent with this were not seen to be as extreme, since the larger
model. block size corresponds more closely to the
Using the extracted sampling data set as the sample spacing on which the estimation is based.
basic information, and the total data set as a The dispersion variance of the kriged block
measure of the actual grade distribution, the results was, however, still only half that of the
exercise involved using both conditional simula- actual.
tion techniques and more traditional methods to Figure 7 compares the results for 6 m SMUs in
model the support and information effect for a the form of grade/tonnage curves, comparing in
number of different SMU sizes and sampling this case the actual results with the kriged values,
configurations. The results are then compared the average of the first two simulated values, and
with the actual values. the results obtained from the discrete gaussian
model. The kriging is again seen to be an
Conditional simulation inadequate tool for modelling the support effect
Several repeated conditional simulations were with wide discrepancies in recoverable tonnage
performed to produce alternative models of estimates and a consistent under-estimation of,
point grade values on a 2 m grid, using in this recoverable grade. The conditional simulation
case the sequential gaussian simulation algor- results provide a much more acceptable model,
ithm with simple kriging. Comparisons with while the discrete gaussian model has repro-
results from the turning bands methods were duced the actual values almost perfectly.
made, but are beyond the scope of this study. Similar results were obtained for 12m and
To maintain consistency with the actual total 24 m SMU sizes, although less data were avail-
data set, values were simulated only at points for able to compare modelled results with actual
which an actual value was available. Figure 4 values.
shows the first two of the alternative simulation
models produced which, when compared with Modelling of the information effect
Fig. 3, show a good reproduction of the local
spatial variability as well as modelling the higher As an illustration of the information effect, a
and lower grade features identified by the study was made on the selection of 12 m SMUs
extracted sampling grid. Histograms and vario- based on a detailed square sampling grid of
grams calculated on the simulated results also either 6m or 12m. This represents a possible
confirm that the simulated models are appropri- practical situation where selected blast-hole
ate representations of reality. values are used for grade control and local
estimation of 12 m blocks in the mining process.
To determine the actual information effect of
Modelling of the support effect this situation, a regular grid on 6 m and 12m
The support effect was modelled for SMU sizes centres was extracted from the total original
of 6 m, 12 m and 24 m square blocks. The actual data set to represent grade control sampling
SMU values were obtained by averaging the results.
sample values within each block and retaining These grids were then used to estimate the
only those blocks with a minimum number of values of 12 m SMUs by ordinary kriging, with
included points. Simulated SMUs were modelled the estimated values representing the best esti-
in a similar way by averaging simulated point mates that would be available at the time of
values and, as a comparison with the results of selection of the mining blocks. Since the actual
ordinary kriging, the SMU values were esti- values of the 12 m blocks were also available, it
mated by kriging based on the extracted sam- was possible to model the effect of applying a
pling grid. cut-off grade to the estimated values, and calcu-
Figure 5 compares the results from the first late the actual resulting recoverable tonnage and
two simulations for 6 m SMUs, and Fig. 6 con- grade.
trasts the kriged results with the actual values. It The results of the information effect on the
can be seen that, while the simulations produce real data are shown in the grade/tonnage curves
SMU values of a very similar nature to the in Fig. 8. The curves labelled 'actual' represent
original values, the kriging shows the familiar the true distribution of 12m SMUs in the
over-smoothing of values towards the average. area under consideration, while the 'expected'
In fact, the dispersion variance of the kriged curves show the values that are expected to be
 Downloaded from [Link] at Pennsylvania State University on May 16, 2016

296 P.J. RAVENSCROFT

Fig. 5. Simulated values of 6m selective mining units.

Fig. 6. Comparison of kriged and actual 6m selective mining units.

recovered based on the estimates from the grade spacing is directly measurable in its impact on
control sampling. These estimates are, however, final recovered tonnage and grade.
subject to a certain amount of error and, Unfortunately, in a practical situation the
although the recovered tonnage will be as actual recoverable SMU values are not known,
expected, the recovered grade will be different as and the analysis discussed above is not possible.
shown by the 'recovered' curves. The effect of However, by the use of conditional simulation,
6 m rather than 12m grade control sample exactly the same exercise can be carried out on
 Downloaded from [Link] at Pennsylvania State University on May 16, 2016

CONDITIONAL SIMULATION 297

% Tonnage Grade Above Cut-Oil the simulated model, and the resulting adjust-
4 ment factors for the information effect can be
-"-- Actual applied to the actual estimated ore reserve.
grlged
z.5 Figure 9 shows a comparison of the recovered
grade/tonnage curves for the actual results pre-
80 9-D-- Average Simulated
sented above and for the same analysis on the
-'q- Discrete Gausslan z simulated model. At the same time, results
- obtained from the discrete gaussian model show
'-..~
12.5 how this method also provides a means of pre-
60 dicting the information effect with some success.
E.
The simulated results appear somewhat disap-
2 pointing, but they are based on only one simula-
tion. In a more extensive study, where averaging
40 of repeated simulations is performed, results
1.5 more consistent with the actual values would be
found.

~o.5 Discussion o f results

0 ~
0 0.5
Cut-Oil
I 1.5
t
2
I
,~
The results presented in this limited exercise
illustrate the application of conditional simula-
tion methods to model the support and infor-
mation effects for recoverable reserve esti-
mation. The study has only examined global
grade/tonnage estimation, but the extension to
Fig. 7. Support effect for 6m selective mining units. the local case is clear.
It has been shown that interpolation methods,
such as ordinary kriging, are inadequate for
providing sensible estimates of recoverable re-
% Tonnage Grade Above Cul-Ol! serves above a given cut-off grade and should
4 not be used for this purpose. The discrete
~X -- Actual gaussian model has been shown to work well
a.5 in the estimation of recoverable reserves, but
~ ~ --=-- Expected 6m
may be less applicable in other cases where
80 -+- Recovered 6m more unusual grade frequency distributions and
9~ - Expected 12m a spatial correlation patterns are encountered.
\ --[3-- Recovered 12m The conditional simulation approach, how-
2.5 ever, has been shown to provide readily accept-
60 able results in a simple way. In addition, it has
other major benefits which are a natural exten-
2 sion of the creation of a detailed orebody model.

40
1.5
Conclusion
This paper has presented the concepts of the
application of conditional simulation in recover-
able reserve estimation. The idea is not new, and
0.5 the methods have and are being used in project
valuation. However, with the ever increasing
computer power becoming available for ore re-
0 t.
0 0.5 1 1.5 serve estimation and mine planning, computa-
tionally intensive techniques such as these are
Cut-Oil
certain to have more widespread application in
the future.
Fig. 8. Observed information effect in actual data
values.
 Downloaded from [Link] at Pennsylvania State University on May 16, 2016

298 P.J. RAVENSCROFT

6m Grade Control Sampling 12m G r a d e C o n t r o l S a m p l i n g

% Tonnage Grade Above Cut-Oft % Tonnage Grade Above Cut-Oft

100 : 4
1~176
i =~ ~ , ~ Actual Actual
+ Simulated Simulated 3.5

80 lscrete Gaussian 80 Discrete Gaussian

3

2.5
60 6O

40 40
9 1.5

1
20 ~'~' 20

0.5 0.5

0 o o
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Cut-Oil Cut-Oil

Fig. 9. Estimated information effect.

References

DAVID, M. 1972. Grade tonnage curve: use and misuse JOURNEL, A. G. 1974. Geostatistics for Conditional
in ore reserve estimation. Transactions of the Simulation of Ore Bodies. Economic Geology, 69,
Institute of Mining and Metallurgy, Section A, 81, 673~87.
129-132. 1979. Geostatistical Simulation: Methods for
GOMEZ-HERNANDEZ,J. J. d~ SRIVASTAVA,R. M. 1990. Exploration and Mine Planning. Engineering and
ISIM3D: A three-dimensional multiple indicator Mining Journal, December 1979, 86-91.
conditional simulation program. Computers and RAVENSCROFT, P. J. & ARMSTRONG, M. 1990. Kriging
Geosciences, 16(4), 395-440. of Block Models--the Dangers Re-Emphasised.
ISAAKS, E. H. 1990. The Application of Monte Carlo Proceedings of 22nd APCOM Symposium, Berlin,
Methods to the Analysis of Spatially Correlated 1990.
Data. PhD thesis, Stanford University.