Professional Documents
Culture Documents
Jounel - Huijbregts 1978 - Mining Geostatistics PDF
Jounel - Huijbregts 1978 - Mining Geostatistics PDF
Ch. J. HUIJBREGTS
ACADEMIC PRESS @
Mining Geostatistics
A. G. JOURNEL
and
CH. J. HUIJBREGTS
Centre de Geostatisrique
Fontainebleau, France
ACADEMIC PRESS
Harcourt Brace Jovanovich, Publishers
London . San Diego . New York Berkeley
Boston . Sydney . Tokyo . Toronto
WWW.minebook.bIogfLCom
ACADEMIC PRESS LIMITED
24-28 Oval Road
London NWl 7DX
Copyright 1978 by
ACADEMIC PRESS LIMITED
Reprinted with corrections 1981
Third Printing 1984
Fourth Printing with corrections 1989
WWW.minebook.blogfa.com
Foreword
by G. MA THERON
W'NW.minebook.blogfLCom
VI FOREWORD
treatise banal and of little interest. I feel that they would be mistaken. At this
point, a comparison with geophysics could be instructive (besides, the very
term "geostatistics" was formed after the model of "geophysics" or of
"geochemistry' ').
If a theoretical physicist were to read a treatise on geophysics, he would
risk getting the disappointing impression that it is but a collection of
particular cases of the theories of potential and of wave propagation: a bit of
gravitation and magnetism, quite a lot of electromagnetism, acoustics, etc.,
nothing very original, really. And yet, without special training, that very
physicist, be he a Nobel prize winner, would admit to be unable to interpret
the results of a seismic prospection campaign.
In the same way, a specialist in mathematical statistics, having paged
through the present book, would not find much more than variances,
covariances, optimal linear estimators, etc., and would find no justification
in this for creating a separate field. However, let us submit to this specialist
the following problem, which is quite a standard one for the mining engineer
(in geostatistical terminology, it is the problem of the estimation of recover-
able reserves).
"We have, at a certain depth, an interesting copper deposit. The geolo-
gists who have studied the deposit have arrived at certain conclusions about
its genesis and structure, which you will find in this report. The available
numerical data consist of a square grid of drill -holes 200 m apart. The core
sections were analysed metre by metre, and, thus, we have a total of some
5000 chemical analyses. We are considering open-pit selective mining. This
means that, during the mining of our approximately 200 Mm 3 deposit, which
has been mentally divided into 50 000 blocks 20 m x 20 m x 10m, we have
the choice of sending each newly excavated block either to waste , or to the
processing plant. Taking account of an economic pre-feasibility study, we
have decided that the blocks with a mean ore grade greater than x = 0.4%
Cu should be sent to the plant, and that the rest should be set aside with the
waste . This cut-oft grade x is subject to revision, since the market price of
copper as well as the mining and processing costs can vary. x should be
considered as a parameter, and the evaluations should be made under
various hypotheses, i.e., with x =0.3%,0.4%,0.5% , for example.
"Moreover, we must consider the fact that our block by block decision is
based, not on the unknown true block grade but on an estimator, this
estimator being based on information which will be much richer than that
now available (core drillings on a 200 ill X 200 m grid): at the time of
selection, we shall have the results of the chemical analyses of blast holes on
a 5 m x 5 m grid. Based upon this Anal information, we shall build an
estimator z 1 for each of our blocks, and we shall send this block to the plant
only if its estimated grade z i is greater or eq ual to the cut-off grade .r.
WWW.minebook.bIogfLCom
FOREWORD VII
W'NW.minebook.blogfLCom
viii FOREWORD
WWW.minebook.blogfiLCom
Contents
Foreword by G. Matheron v
Contents xi
INTRODUCTION 1
VtIWW.minebook.btogfiLCom
x CONTENTS
Bibliography 581
Index A, Geostatistical concepts 591
Index B. Quoted deposits 595
Index C. FORTRAN programs 596
Index D. Notations 597
WWW.minebook.blogfa.com
Introduction
WWW.minebook.btogfa.com
2 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
INTRODUCTION 3
words, the mining engineer must be involved. Outside the mining field,
geostatistics applied to forestry requires the ability to interpret spatial
correlation between the densities of growth of different species of trees;
geostatistics applied to bathymetry must take into account all the problems
related to the localization of the hydrographic survey ship at sea. Thus, a
book on geostatistics cannot be dissociated from its practical context, in
this case mining.
This instrumental nature of geostatistics both opposes and completes the
explanatory aspects of the standard, deterministic, geological approach .
For example, even if a metallogenic study is sometimes successful in
explaining the overall genesis of a given deposit , such a deterministic
approach wil1 be clearly insufficient on a smaller scale, such as that of metal
concentrations. In such a case, it is said that the grades are "randomly"
distributed within the defined genetic outline.
But nothing is more precise than this "random" distribution. It will differ
from a gold placer deposit to a porphyry-copper deposit, and the random
distribution of the grades within the deposit will condition its survey and its
selective mining, which represent considerable investments. On the smaller
scale of mining, the probabilistic approach can relay the deterministic and
genetic approaches by which the deposit was delineated on a larger scale.
Our lumberman is not a botanist; similarly, the mining geostatistician
accepts the reality imposed by nature, and is content to study the
consequences of this reality on the mining project he must prepare. A
probabilistic approach can, of course, be guided by genetic considerations,
and the best mining geostatistician is the one who is able to combine his
geological knowledge and his technical mining skill with a good use of the
probabilistic language.
WWW.minebook.blogfiLCom
4 MINING GEOSTATISTICS
VtIWW.minebook.btogfiLCom
INTRODUCTION 5
First notations
For the uninitiated, one of the main difficulties of geostatrsncs is the
notation. These notations are precisely defined as they enter into the work,
and are summarized in Index D. However, it may prove helpful at this
stage to introduce the following basic notations.
A deposit is always a regionalized phenomenon in the three-dimen-
sional space. To simplify notation, the coordinates (xu, xv, x.,) of a point are
usually denoted x. Thus, the integral (extended over the three-dimensional
f-co I(x) dx =
r
space) of the function [(Xu., xv, x.,) is written as
+OO
-co dx,
f+>
-:x;l dx,
r.
-co [(x"' Xli' x.,) dx w
WWW.minebook.blogf&com
WWW.minebook.btogfa.com
I
Geostatistics and Mining
Applications
SUMMARY
This chapter is intended to be both an introduction to, and an overview of,
geostatistics as applied to the mining industry. Each successive problem encoun-
tered from the exploration phase to the routine operating phase is presented along
with its geostatistical approach. The problems of characterizing variability struc-
tures of variables of interest within the deposit are solved with the "variogram",
Optimal local estimates with corresponding confidence intervals are found by
"kriging". The choice of various working techniques (machinery, need for stock-
piling or blending) can be facilitated by simulating them on a geostatistically
determined numerical model of the deposit.
The main steps in proving a deposit are outlined schematically in section LA
along with some typical evaluation problems arising from such steps.
Section LB introduces the geostatistical terminology from a practical and intui-
tive point of view. Such key notions as regionalized variables, variogram, estima-
tion and dispersion variances, are presented.
The geostatistical approach to several of the problems encountered in section LA
is given in section I.e, which also outlines the sequence of steps in a geostatistical
study.
WWW.minebook.btogfa.com
8 MINING GEOSTATISTICS
Geological surveys
A preliminary survey of a province or favourable metallogenic environ-
ment, by such various means as geological mapping, geochemistry or
geophysics, will usually result in the location of a certain number of
prospects. If one of these prospects locates a mineralized zone, the next
step is to determine its value and extent by a more detailed survey in-
volving, for example, drilling, trenches or even, sometimes, drives. The
locations of these samples are generally not regular, but are based on
geological considerations such as the need to verify the existence of the
mineralization and to provide a qualitative description of it (e.g.,
mineralogy, tectonics, metallogenic type). The results of such sampling
may be-a proper basis on which to plan further reconnaissance, but they are
not entirely suited for quantitative estimations of tonnage and yardage.
The first estimations will tend to be more qualitative than quantitative , as
will be the- geologist's image of the deposit.
WWW.minebook.blogfiLCom
1. GEOSTATISTICS AND MINING APPLICAnONS 9
resources, depend on the particular deposit and the type of are considered.
And, finally, he knows that for a given ore the variability of assays of core
samples, cuttings or channel samples can be markedly different.
In this chapter, it will be shown how geostatistics takes these various
concepts into account by means of the variogram and structural analysis.
WWW.minebook.btogfa.com
MINING GEOSTATISTICS
(i) The definition of the most precise local estimator and its confidence
interval, taking account of the peculiarities of the deposit and the
ore concerned. It is obvious that a gold nugget deposit cannot be
evaluated in the same way as a sedimentary phosphate deposit.
(ii] The distinction between in situ resources and mineable reserves. It
should be possible to model rigorously the effect of the mining
method on the recovery of the in situ resources.
(iii) The definition of the amount of data (drill cores, exploration drives)
necessary for a reliable estimation of these recoverable reserves.
Mine planning
Once the estimations of resources and recoverable reserves have been
made, the next step is to plan the mining project itself. This involves
choosing mining appliances, type of haulage, requirements of stockpiling
and blending, mill planning, etc. Some of these choices are conditional, in
all or part, upon the structures of the variabilities of the are characteristics,
either in situ or at any of the various stages of mining, haulage, stockpiling
and mill feeding. It will be shown later how geostatistics can be used to
provide representations of these variabilities by means of numerical models
of the, deposit. The effects of various extraction and blending processes can
then"be tested by simulating them on these numerical models.
WWW.minebook.blogfLCom
r. GEOSTATISTICS AND MINING APPLICATIONS 11
WWW,minebook,blogf&com
12 MINING GEOSTATISTICS
two experimental measures [z (X;), Z(Xi + h)] at any two points separated by
the vector h ; i.e.,
(1.2)
where N(h) is the number of experimental pairs [z (Xi), Z(X; + h)] of data
separated by the vector h.
Note that the intrinsic hypothesis is simply the hypothesis of second-
order stationarity of the differences [2 (x) - 2 (x + h)]. In physical terms,
this means that, within the zone D, the structure of the variability between
two grades z (x ) and z (x + h) is constant and, thus, independent of x; this
would be true, for instance, if the mineralization within D were homo-
geneous.
The intrinsic hypothesis is not as strong as the hypothesis of stationarity
of the random function 2 (x) itself. In practice, the intrinsic hypothesis can
be reduced, for example, by limiting it to a given locality; in such a case, the
function y(x, h) can be expressed as two terms: )I(x, h) = I(x) . )lo(h),
where yo(h) is an intrinsic variability constant over the zone D and [(x)
characterizes an intensity of variability which depends on the locality x, d.
the concept of proportional effect, treated in section IILB.6.
WWW.minebook.blogfa.com
I. GEOSTATISTICS AND MINING APPLICATIONS 13
Continuity
In the definition of the variograrn 2y(h), h represents a vector of modulus Ih I
and direction a. Consider a particular direction a. Beginning at the origin,
y(O) = 0, the variogram increases in general with the modulus Ih I. This is
simply an expression of the fact that, on average, the difference between
two grades taken at two different points increases as the distance Ih I
between them increases. The manner in which this variogram increases for
small values of Ih I characterizes the degree of spatial continuity of the
variable studied. Figure 11.10 shows profiles of four piezometer readings,
giving the height of a water table, measured at four different points
(denoted by 3, 4,33 and 18), as a function of time. The four corresponding
semi-variograms have markedly different shapes, from which a geostatisti-
cian would be able to deduce the degree of continuity of the four
piezometer profiles.
- - - - Horizontal
C7
Ihl
NNI.minebook.blogfa.com
14 MINING GEOSTATISTICS
1
HV, v)=NV ,IN f y(x;-x)dx.
,~l v(y)
The general formula (1.3), shows that the estimation variance, i.e., the
quality of the estimation, depends on all four of the following.
(i) The relative distances between the block V to be estimated and the
information v used to estimate it. This is embodied in the term
y(V, v).
(ii) The size and geometry of the block V to be estimated. This is
embodied in the term y(V, V).
(iii) The quantity and spatial arrangement of the information v, which is
embodied in the term -rev, v).
(iv) The degree of continuity of the phenomenon under study, which is
conveyed by its characteristic serni-variogram y(h).
Thus, the theoretical formula (1.3) expresses the various concepts that
we intuitively know should determine the quality of an estimation.
Confidence interval
The estimation variance itself is not enough to establish a confidence
interval for the proposed estimate, the distribution of the errors must also
WWW.minebook.blogfa.com
I. GEOSTATISTICS AND MINING APPLICATIONS 15
WWW.minebook.blogfa.com
16 MINING GEOSTATISTICS
each datum is defined increases: the mean grades of mining blocks are less
dispersed than the mean grades of core samples.
These two phenomena are expressed in the geostatistical concept of
dispersion variance. Let V be a domain consisting of N units with the same
support v. If the N grades of these units are known, their variance can be
calculated. The dispersion variance of the grades of units v within V,
2
written D (v/ V), is simply the probable value of this experimental vari-
ance and is calculated by means of the elementary variogram 2y(h)
through the formula
D 2(v/V)= i'(V, V)-y(v, v). (IA)
1.8.5. Coregionalizations
The concepts of variogram and various variances, which so far have been
defined for the regionalization of one variable, can be generalized to spatial
coregionalizations of several variables. In a lead-zinc deposit, for example,
the regionalizations of the grades in lead, ZI (x), and in zinc Zz(x), are
characterized by their respective variograms 2YI(h) and 2Y2(h). However,
these two variabilities are not independent of each other and we can define
a cross-variograrn for lead and zinc:
"( 1 "(IZ],
[
"(12 "(2
VNIW.minebook.blogfa.com
I. GEOSTATISTICS AND MINING APPLICATIONS 17
100
50
Rain - go uge
FIG. 1.2. Locations of the rain gauges.
WWW.minebook.blogfa.com
18 MINING GEOSTATISTICS
variant I
50
Variant 2
Voria nt 3
WVNI.minebook.blogfa.com
I. GEOSTATISTICS AND MINING APPLICATIONS 19
NNI.minebook.blogfa.com
20 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
I. GEOSTA TISTICS AND MINING APPLICAnONS 21
'#NNI.minebook.blogfa.com
22 MINING GEOSTATISTICS
(0 )
FIG 1.5. Recoveries estimated from cut-off grades applied to histograms. (a)
Influence of the support; (b) influence of the standard of information.
and that the same cut-off grade Zo applied to both histograms will result in
very different recoveries. These different dispersions are characterized by
2(V/D)
the two dispersion variances D\v/D) and D with D\vID
D 2( VI D). By assuming some hypotheses related to the conservation of the
type of distribution law, it is possible to deduce these two histograms from
the experimental histogram defined on the support C of core sample
lengths. For any given selection unit size, the estimated histogram can then
be used to determine the recoveries corresponding to different cut-off
grades zoo These recoveries are then used to build the so-called "grade-
tonnage" curves.
WWW.minebook.blogfa.com
L GEOSTATISTICS AND MINING APPLICATIONS 23
the time of selection. Selection is made on the basis of the estimated grades
of these units, which will inevitably result in a recovery different from the
optimum recovery based on the true grades.
Figure 1.5(b) shows the histogram of the true grades of the units V
superimposed on the histogram of their estimated values. As actual selec-
tion will be made on the basis of these estimated values, the proportion of
recoverable units V must be read from the histogram of the estimated
grades (dotted area on Fig. 1.5(b)) and not from the histogram of the true
grades (hatched area on Fig. 1.5(a)).
If the estimated grades are obtained by kriging, then the dispersion
variance D 2 ( V* I D) of the histogram of the estimated grades can be cal-
culated from a relation known as the "smoothing effect" due to kriging:
(I. 7)
NNI.minebook.blogfa.com
24 MINING GEOSTA TISTICS
VNIW.minebook.blogfa.com
L GEOSTATISTICS AND MINING APPLICATIONS 2S
WWW.minebook.blogfa.com
II
The Theory of Regionalized
Variables
SUMMARY
This fundamental theoretical chapter gives the conceptual basis of geostatistics,
then introduces, successively, the three main geostatistical tools: the variogram, the
estimation variance and the dispersion variance.
Section II.A gives the interpretation of a regionalized variable z (x ) as a realiza-
tion of a random function Z (x ), The spatial correlations of this random function,
i.e., the variability in space of the variable being studied, are characterized by the
covariance function C(h) or by its equivalent, the semi-variogram function defined
by relation (JIA):
y(h) = 1E{[Z(x + h)- Z(X)]2}.
Section II.B introduces the distribution function of the error of estimation. If a
normal distribution is assumed, this error is then completely characterized by its
expectation (or estimation bias) and its variance (or estimation variance). For the
important class of linear estimators, the estimation variance of any domain V by
any other domain v is given by a combination of mean values of the semi-variogram
y(h), d. formula (11.27):
(T~(v, V)= 2-y(V, v)- )I(V, V)- Y(v, v).
Section II.C introduces a second variance, the dispersion variance D 2 (v/ V),
which characterizes the dispersion of the grades of mining units v within a stope V,
for example. This dispersion variance is written as a combination of mean values of
the semi-variogram 'Y(h), cf. formula (11.36):
D 2 (v/ V) =)i( V, V) - )I(v, u),
Section II.D introduces the important notion of regularization (taking the
average value in a certain support v). If Z (x) is the point grade with the point
serni-variogram vin). the mean value Z,,(x) in the volume v is then characterized
by a regularized semi-variogram y,,(h), which is written as a combination of mean
values of the point serni-variogram, d. formula (11.41):
'Yv(h}=y(v, Vh)'--Y(V, v}.
26
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 27
Two sets of information defined on two different supports v and v' and charac-
terized by the two regularized semi-variograms Y~ and Yv' can then be compared.
As all the previous geostatistical operations make continuous use of mean values
y of the point serni-variogram y(h), section II.E is devoted to the practical
calculation procedures of those mean values y. Either a direct numerical calculation
or the use of appropriate auxiliary functions and charts is possible.
WWW.minebook.blogfa.com
28 MINING GEOSTATISTICS
f.--------~400 m --------+-1
....---- 20 m --..........,~
10
s:
ii.
Q)
p
20
30
Poor
zone
x
FIG. Il.l. Random and structured aspect of a ReV.
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 29
NNI.minebook.blogfa.com
30 MINING GEOSTATISTICS
Statistical inference
The probabilistic interpretation of a ReV z (x) as a particular realization of
a certain RF Z (x) has an operative sense only when it is possible to infer all
or part of the probability law which defines this RF in its entirety.
Obviously, it is not rigorously possible to infer the probability law of a
RF Z (x) from a single realization z (x) which is, in addition, limited to a
finite number of sample points Xi- It is impossible, for instance, to deter-
mine the law of the RV "the result of casting the die" from the single
numerical result 5 resulting from a single cast of the die, and, in particular,
it is not possible to determine whether or not the die is biased. Many casts
of the die are necessary to answer such a question. Similarly, many realiza-
tions z 1(x), Z2(X), ... , zdx) of the RF Z (x) are required in order to infer
the probability law of Z (x). Since, in practice we shall be limited to a single
realization {z (Xj)} of the RF at the positions Xj, we appear to have come to a
dead end and, in order to solve it, certain assumptions are necessary. These
assumptions involve various degrees of spatial homogeneity and are intro-
duced under the general heading of the hypothesis of stationarity.
In practice, even if only over a certain region, the phenomenon under
study can very often be considered as homogeneous, the ReV repeating
itself in space. This homogeneity or repetition provides the equivalent of
many realizations of the same RF Z (x) and permits a certain amount of
statistical inference. Two experimental values z (xo) and z (xo + h) at two
different points Xo and Xo + h can, thus, be considered as two different
realizations of the same RV Z (xo). This type of approach is not peculiar to
geostatistics, it is used to infer the distribution law of the RV Z (x) from a
histogram of data values {z(Xj)}, or more simply, to infer the mathematical
expectation E {Z (x)} from the arithmetic mean of the data.
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 31
The set of all these distribution functions, for all positive integers k and for
every possible choice of support points in R", constitutes the "spatial law"
of the RF Z(x).
In mining applications, the entire spatial law is never required, mainly
because the first two moments of the law are sufficient to provide an
acceptable approximate solution to most of the problems encountered.
Moreover, the amount of data generally available is insufficient to infer the
entire spatial law. In geostatistics (more precisely, linear geostatistics), only
the first two moments of the RF are used; in other words, no distinction is
made between two RF's Z1(X) and Z2(X) which have the same first- and
second-order moments and both functions are considered as comprising
the same model.
Second-order moments
The three second-order moments considered in geostatistics are as follows.
(i) The variance, or more precisely the "a priori" variance of Z(x).
When this variance exists, it is defined as the second-order moment
about the expectation m(x) of the RV Z(x), i.e.,
Var {Z (x)} = E{[Z(x) - m (x )]2}.
As with the expectation m (x), the variance is generally a function of
x.
(ii) The covariance. It can be shown that if the two RV's Z(XI) and
Z (X2) have variances at the points XI and X2, then they also have a
covariance which is a function of the two locations x I and Xl and is
written as
NNI.minebook.blogfa.com
32 MINING GEOSTATISTICS
(ii) for each pair of RV's {Z (x ), Z (x + h)} the covariance exists and
depends on the separation distance h,
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 33
WWW.minebook.blogfa.com
34 MINING GEOSTATISTICS
(for distances Ih I less than the limit b). This limitation of the hypothesis of
second-order stationarity (or the intrinsic hypothesis if only the variograrn
is assumed) to only those distances Ih I~ b corresponds to an hypothesis of
quasi-stationarity (or to a quasi-intrinsic hypothesis).
In practical terms, we can define sliding neighbourhoods inside of which
the expectation and the covariance can be considered as stationary and the
data are sufficient to make statistical inference possible.
The hypothesis of quasi-stationarity is really a compromise between the
scale of homogeneity of the phenomenon and the amount of available data.
In fact, it is always possible to produce stationarity by considerably reduc-
ing the dimension b of the zones of quasi-stationarity, but then most of
these zones would not have any data in them, and, thus, it would be
impossible to infer the quasi-stationary moments in these zones.
From a data grid of 1 to 3 m, a 200-m-Iong hillside would appear as a
quasi-stationary model, but would require a non-stationary model
(because of the increasing altitude) for a grid of 20 to 30 m, cf. Fig. 11.3. If
the data grid is of the order of 200 m, then a certain stationarity again
becomes evident: at the scale of the peneplanar basin the entire hillside
appears as a local variability of the stationary RF "altitude". This example
shows, incidentally, that it is not possible to devise a theoretical test for the
a priori verification of the hypothesis of stationarity using a single realiza-
tion of the RF. It is always possible to find a particular realization of a
stationary RF which would display a systematic and continuous change in
value (apparent non-stationarity) over a limited distance, so that a
theoretical test could never refute the hypothesis of stationarity.
- I b=IO m I
x
~~
x
I I'
I I <,
x <,
<,
z(xo} <,
x x
200 m
..
x
VNNI.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 35
The covariance function must be such that it ensures that the previous
variance will always be positive or zero. By definition then, the function
C(h) is said to be "positive definite".
Thus, not any function g(h) can be considered as the covariance of a
stationary RF: it is imperative that it be positive definite.
Taking account of relation (11.4), expression (11.8) can be written in
terms of the serni-variogram:
In the case where the variance C(O) does not exist, only the intrinsic
hypothesis is assumed, and the variance of Y is defined on the condition
that
The variogram function must be such that Var {Y} is positive or zero, with
the condition that It Aj = O. By definition, -l'(h) is said to be a "conditional
positive definite" function.
WWW.minebook.blogfa.com
36 MINING GEOSTATISTICS
Absence of correlation
Very often, in practice, the correlation between two variables Z (x) and
Z (x + h) disappears when the distance h becomes too large:
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 37
r(m)=c{O)
C(O) - - - - - - - - - - - --~--_----:._----
r (II)
C(II)
C{co)::O
o o
NNI.minebook.blogfa.com
38 MINING GEOSTATISTICS
.- - Vertical
1!vertlcal
COre .I
rcnqe
- - - Horizontal
o 01 02 lnl
FIG. II.S. The variogram as a structural tool.
t In fact, the right and left derivative, /,'(0+) and )"(O~) exist but have different signs.
+A RF is said to be mean-square continuous if
2
lim E{{Z(x +h)-Z(x)J } =0, when h ..... 0.
'INNI.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 39
o h a h o h
FIG. II.6. Behaviour near the origin of the variograrn. (a) Parabolic behaviour;
(b) linear behaviour; (c) nugget effect; (d) pure nugget effect.
(c) Discontinuity at the origin: y(h) does not tend towards zero when h
tends towards zero, although by definition 1'(0) = O. The RF is no
longer even mean-square continuous; the variability between two
values z(x) and z(x +h) taken at two very close points may be quite
high and increases as the size of the discontinuity at the origin of
y(h) increases. This local variability can be compared to the random
phenomenon of white noise known to physicists. As the distance h
increases, the variability often becomes more continuous and this is
reflected in the continuity of y(h) for h > O.
The discontinuity of the variogram at the origin is called a "nugget
effect" and is due both to measurement errors and to micro-vari-
abilities of the mineralization; since the structure of these micro-
variabilities is not accessible at the scale at which the data are
available, they appear in the form of a white noise. A whole section
(lILA) will be devoted to the study of this nugget effect.
(d) Pure nugget effect. This is the limit case when y(h) appears solely as a
discontinuity at the origin:
y(O)= 0 and y(h)= Co when h > 8.
WWW.minebook.blogfa.com
40 MINING GEOSTA TISTICS
11.A.4. Coregionalization
A regionalized phenomenon can be represented by several inter-correlated
variables and, in certain cases, it may be useful to study them simul-
taneously.
In a lead-zinc deposit, for example, the two mineralizations are closely
related and it may be of interest to know if, on average, zones that are rich
in Pb are equally rich in Zn, In other polymetallic deposits, it would be
useful to be able to use the information provided by one variable to make
up for missing information on another variable.
The probabilistic approach to a coregionalization is similar to that of the
regionalization of a single variable.
The simultaneous regionalizations (or the coregionalization) of K ReV's
{Zl(X), . . . , ZK(X)} are interpreted as a particular realization of the set of K
inter-correlated RF's {Z 1(x), ... , ZK (x )}.
Under the hypothesis of second-order stationarity we define:
(i) for each RF Zk (x), the mathematical expectation
E{Zk(X)} = m; = constant, Vx;
(ii) for each pair of RF's Zdx), Zdx), the cross-covariance
E{Zdx + h). Zk(X)} - mi-mi. = Cek(h), Vx; (I1.l5)
(iii) and the cross-variogram
E{[Zdx + h)- Zk'(X)][Zk(X + h)- Zk(X)]} = 2Yn(h), Vx. (11.16)
Properties of cross-moments
1. When k' = k, the formulae (11.15) and (11.16) yield the definitions of the
covariance and variograrn, Ci; (h) and 2 'Ykk (h).
VNNI.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 41
WWW.minebook.blogfa.com
42 MINING GEOSTATISTICS
If the coregionalization {Zl (x), ... , ZK (x)} verifies the intrinsic hy-
pothesis, then the RF F(x, Ak ) is also intrinsic and has the following
stationary serni-variogram:
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 43
where v = 00075 fe
is the area of cross-section of the analysed core
samples, and V is the horizontal surface area of the field. The a priori
variance of the logarithms of the gold grades should be the limit of (T2( V)
as the field size V tends towards infinity. Experimentally, such a limit does
not exist and, as a consequence, a second-order stationary model of RF
cannot be used. It then becomes necessary to use a model that is only
intrinsic, having a variogram but no covariance.
06
0-5
0-4
4>
u
<:
<:::
... 03 0
c OJ QJ
> 0 u
02
to
,2 ::
".
toO
U 0 ~
<lJ o,
o
0-1 '" .2:
'l-
"0
C
0
0
0'" 0:
-I 2 3 4 5 6 7 8 9
Log area
WWW.minebook.blogfa.com
44 MINING GEOSTATISTICS
3 r'--~------~------------'
,/
,/
/'
/'
'" I
W'INI.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 45
Surface 1 Rain
1
1
1
Water table
.. 500 m
D~
of discharge
At piezometer no. 3 The water table is very close to the surface and the
water level reacts each time it rains (the amount of water retained by the
ground is proportional to the thickness of the ground through which the
water passes before reaching the water table). The profile from piezo no. 3
is very erratic and is directly related to the daily rainfall. The variogram
along this profile is linear with a nugget effect (Brownian motion for small
lags t).
At piezometer no. 4 The water table is deeper and the absorption effect of
the ground begins to take effect. The corresponding variogram has a
smaller nugget effect and a transition phenomenon is becoming evident
(appearance of a sill).
At piezometer no. 33 The water table is deeper still and the regularizing
effect of the ground absorption is almost total. There are, however, a
certain number of discontinuities evident on the profile and these are due
to abrupt refilling of the table. The corresponding variogram has a small
WWW.minebook.blogfa.com
46 MINING GEOSTATISTICS
At piezometer no. 18 The water table is even deeper and the effects of the
discontinuous refilling are no longer evident. The nugget effect has
completely disappeared and the variogram presents an almost perfect
parabolic behaviour near the origin, characteristic of the differentiability of
the profile z(t)at any time t.
3
E
4
~
..c
.;: 5
l' 6
0
3:
7
'0
s: 8
n.., Piezo no. 33
a 9
10
II
12
13 Piezo no. 18
14
July August September October November December
I
Olf-
l f-
,-! I
....
....
....
I
l .'
1
i
/
.'
N
E 3 4 33 18
./ I
-;:'005
.'
- I -
.,fI
.1
.,........
l-
.'
I
." .'
1-.
~ .~t
I I I ...... I I I
0 5 10 15 0 5 10 15 0 5 10 15 0 5 10 15
Time (days)
FIG. 11.10. Variography of the piezo heights at Korhogo.
VNNI.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 47
4. Coregionalizations at Exotica
The experimental direct and cross-semi-variograms for Fe, soluble Cu and
total Cu, shown on Fig. I1.II, were calculated from the grades defined on
13-m vertical bench heights in the Exotica copper deposit (Chile) and
represent the horizontal coregionalization of Fe, Cu S, Cu T grades.
If-
Cu T-Cu T
Cu S - Cu T
CuS~CuS
"}' 05-
.~.~. Fe-Fe
01-
o f----------;---
h (m)
WWW.minebook.blogfa.com
48 MINING GEOSTATISTICS
Vi
)(i
x
a mE b
Block ,ll'( xl
Borehole r ( xl I... 95 01.0 pro b 0 hI'
I Ity .1
FIG. Il.12. Distribution of the error.
Suppose now that the entire deposit is divided into blocks of equal size
V, and that each block is intersected by a vertical bore-hole passing
through its centre. If the deposit is an homogeneous mineralization, i.e., if
the RF Z (x) can be considered as stationary, then the error R (x ) is also
stationary and any two errors r(xi) and r(xj) can be considered as two
different realizations of the same stationary RF R(x)=Zv(x)-Z(x).
If the histogram of experimental errors is available in a control zone,
then it will be possible, because of the stationary assumption, to infer the
complete distribution function of R (x). Even if such a histogram is not
available, it will still be possible to calculate the stationary expectation
mE = E{R (x)} and variance (T~ = Var {R (x)} of the distribution function of
the error.
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 49
The particular error r(x;) = zv(x;) - z (x;) involved when estimating the
block V (x;) remains unknown, but the mean and variance of the errors (or
the complete distribution function if it is known) will provide a measure of
the quality of the estimation.
If the RF R (x) is stationary, then this function does not depend on the
location x. Given the function F(u), it is possible to calculate the prob-
ability that the error will fall within any interval [a, bJ;
E{R(x)}=rnE=Constant, 'Ix;
The expectation mE characterizes the mean error, while the variance O"~ is
a dispersion index of the error, cf. Fig. 11.12. Thus, a good estimation
procedure must be such that it ensures
0) a mean error close to zero; this property of the estimator is known as
unbiasedness;
(ii) a dispersion of errors very concentrated around this mean zero
value, this second criterion being expressed by a small value of the
., 2
estimatron vanance O"E.
At the estimation stage, the type of the distribution function of the errors
is unknown in most cases. But, since the two most important characteristics
of this function - its expectation and variance - can be calculated, we shall
refer to a standard two-parameter (mE and t:T~) function which will provide
an order of magnitude of the required confidence interval.
WWW.minebook.blogfa.com
50 MINING GEOSTATISTICS
,-h-
I I \
i...
FIG. 11.13. Normal standard for the error distribution. , Gaussian standard;
- - - -, experimental distribution of errors.
YNNI.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 51
z, E [ Z *i 2 (O'E) *J
z* Z i (11.21)
This formula will not ensure that the lower bound is always positive,
since it becomes negative when the relative standard deviation is greater
than 50%, O'el z* > 0'5, but it is sufficient for practical applications. The
case where erE! f* > O 5 corresponds to poor estimations and, in such cases,
it would be better not to use confidence limits at all.
Formula (H.2l) corresponds to a non-stationary standard deviation of
estimation, since it depends on each estimator z 1. It can be shown (d. G.
Matheron, 1974, and section II1.B.6, Remark 3 and 51
that this formula
corresponds to a conditioning of the RF Z (r ) to the estimator i*, the
distri bution of Z (r ) being lognormal.
NNI.minebook.blogfa.com
52 MINING GEOSTATISTICS
Estimation variance
Let Z (x) be a second-order stationary RF with expectation m, covariance
C(h) and semi-variograrn y(h), ct. definition formulae (11.1) to (11.4).
Consider first the simple case of the estimation of the arithmetic mean ZK
of the K unknown values {z (xd, k = 1 to K}:
1 K
ZK = K k~l Z(Xk)'
ZK* =1;'
- L. )
Z(Xi .
ni=l
1
E{Zk} = - L E{Z(Xi)} = m
n i
with
2 1
E{ZK} = -2 E {L L Z(Xk)Z(Xe}}.
Kkk'
VNIW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 53
Similarly,
:I< 1 2
E{ZKZK} =-I I [C(Xk -x;)+m ].
Kn k i
the constant term m 2 is then eliminated from the expression of (T~, leaving
O"i:: = E{[ZK - Z'J( ]2}
1 1 2
= K2 L L C(Xk - Xd+2 L LC(Xi - X j ) - - I I C(Xk -Xi)' (II.23)
k k" n ; j Kn k i
These mean values zv(x) and z,,(x ') are interpreted as particular real-
izations of two RV's Zy(x) and Z,,(x'):
Zv(X)= VI f
y(x)
Z(y)dy and Zv(X')=~f
v ,,(x')
Z(y)dy.
NNI.minebook.blogfa.com
54 MINING GEOSTATlSTICS
For the estimation variance, the formulae (II.23) and (II.24) for the
discrete case give, for the continuous case of estimating the mean grade of
V by that of v,
uk = E{[Zv(x)- z..,(X')]2\
j
If the domains V (x) and v (x") are three-dimensional, each of the integral
signs of formula (II.25) represents a triple integral and the variance o'~ is a
sum of sextuple integrals. To simplify notations, we shall denote by
C (V, v) the mean value of the covariance C (h) when one extremity of the
distance vector h describes the domain V(x) and the other extremity
independently describes the domain vex'). It is then written
C(h)= C(O)-y(h).
Formulae (II.23) to (I1.26) can then be rewritten in terms of semi-
variograms. Since the constant term C(O) disappears, it is enough to
replace C or C by -1' or -if and the estimation variance becomes
where y( V, v), for instance, represents the mean value of y(h) when one
extremity of the vector h describes the domain V(x) and the other ex-
tremity independently describes the domain v (z '),
It can be shown that this formula (11.27) remains valid even when the
covariance C(h) does not exist, provided of course that the semi-variogram
is defined (the RF Z (x) is then only intrinsic).
Remark 2 The two formulae (II.26) and (rI.2?) are completely general
whatever the domains v and V. In particular, the domains need not
VNNI.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 55
Remark 4 Formulae (11.26) and (11.27) express four essential and intui-
tive facts which condition all estimations. The quality of the estimation of
V by v is a function of the following.
0) The geometry of the domain V to be estimated: term -Y(V, V)
appearing in formula (11.27). As the serni-variogram y(h) is an
increasing function of h, YC V, V) increases with the size of V.
Hence, if v and the interdistance (v, V) remain fixed, it is easier to
estimate the mean grade of a large block V than the grade of an
unknown point (V = x). Finally, if the size V is fixed, the term
YC V, V), and, consequently, the estimation variance cr~, depend
also on the geometry of V.
(ii) The distances between the domain V to be estimated and the
support v of the estimator: term +y(V, v) appearing in formula
(11.27). As the distance between V and v increases, so does the
value of HV, v) and the value of the estimation variance a~.
(iii) The geometry of the estimating domain v: term -y(v, v). As the
dimension of the domain v increases, the value of y(v, v) also
increases and, consequently, a~ will decrease. But, for fixed
volumes V, v and fixed interdistance (v, V), the estimation variance
will also depend on the configuration of the information v. The
estimation of the same block V on Fig. 11.14 by the two distinct
samples v, and V2 will be better than that provided by the two
closely spaced samples v ~ and v; ; indeed, y( v /, + v~, V /1 + v 2) will be
less than ii(VI + V2, VI + V2). This intuitive notion of the importance
WWW.minebook.blogfa.com
56 MINING GEOSTATISTICS
----- ------ -
0. ~\lhI __
--
VI
1111
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 57
L Ai = 1.
j= 1
O'~ = E{[Zv - z t. f}
=2LA i -Y (Xj, V)-y(V, V)-IIAiAiY(Xi-Xj), (II.28)
i i i
where Y(Xi, V) denotes the mean value of y(h) when one extremity of the
distance vector h is fixed at point Xi and the other extremity independently
describes volume V.
WWW.minebook.blogfa.com
58 MINING GEOSTATISTICS
blast-hole samples are taken from it, an average of six being taken from
each block. Experimental observation has shown that the arithmetic mean
of these six holes, Z v, can be taken as the true grade of block V.
--~-~----------------~~~~-~
c
B
% Cu
FIG. II.l? Histogram of kriging errors. - - , Histogram, m = 0-01, (J"~ = 0-273;
---, Gaussian distribution, m =0, (J"~ =0273.
VNIW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 59
The experimental variance of the 397 errors was 0273 (% CU)2 while
the predicted theoretical estimation variance calculated from formula
(11.28) was 0261 (% CU)2. Therefore, the estimation variance was predict-
ed with the very acceptable relative precision of 4%: (0' 273 -
0261)/0-273 = 004.
A normal distribution with zero mean and a variance equal to that of the
experimentally observed errors (u~ = 0,273) is also shown on Fig. 11.17. It
fits reasonably well the experimental distribution of errors with the follow-
ing remarks.
(i) The normal distribution underestimates the proportion of both small
errors (between O'5% Cu) and large errors (greater than
15% eu).
(ii) The standard normal interval (2UE) correctly estimates the
experimental 95% confidence interval: 96% of all the observed
errors fall within 20'E = 1 04 % Cu. If the predicted theoretical
variance, O'~ = 0261 had been used, the result would have made
little difference: 2UE = 102 % Cu.
The nickel silicate deposit of Mea (New Caledonia) was sampled by vertical
bore-holes on a pseudo-regular square grid of 80 m, d. Fig. 11.18. One of
the problems in the exploration of this deposit was to map the hanging wall
and foot wall of the mineralized bed. Let Z (x) be the height of the hanging
wall at the horizontal coordinate point x. At each point Xi where a drill
intersected the bed, the height Z(Xi) of the hanging wall was measured. In
order to map the hanging wall, the unknown value z (xo) of the height at
Xj
? Z(Xi) 0
,,/
Xo
X x X
Xi
0 '0 ~
x x
Foot
wall
FIG.
(J
bore-holes; x, Xo
FJ
node of
the estimated grid.
WWW.minebook.blogfa.com
60 MINING GEOSTATISTICS
each node of a regular square grid of 40-m side was estimated. The
estimator used was the BLUE Z*(xo) provided by kriging.
Isobath contours of the hanging wall were plotted, using the estimated
40-m grid pattern. Similarly, the foot-wall depth and the mineralized
thickness (hanging-wall height minus foot-wall depth) were plotted.
The kriging estimator z*(xo) used was a linear combination of the nine
bore-hole data closest to X():
9
z*(xo)= I Ajz(xj).
j = 1
Since the configuration of the available data was not strictly regular, the
estimation variance depends on the position of each point x(} to be esti-
mated.
In order to verify the procedure, each one of the known values z(x;) was
estimated by kriging, i.e., by the optimal linear combination of the nine
closest data values {Xj, j = 1 to 9, and j oj:; n.Using the theoretical estima-
tion variance (T~ (x,) and the known experimental error z (x;) - z*(x;), the
deviation ratio e (x.) = [z (Xi)- z*(x; )1/ O'E(X;) was calculated for each data
point Xj. For the theoretical estimation variance (T~ (Xi) to be an effective
estimator of the true estimation variance, the variance of the n values of
e(x;) should be approximately equal to 1.
Figure 11.19 shows the histogram of 167 e(x;) values along with a normal
distribution with the same variance and a zero mean. The arithmetic mean
of the 167 experimental errors [z (Xi) - z *(x;)] is 015 m and, given that the
mean hanging-wall height is 840 ill and the mean bed thickness is 20 m, the
non-bias can be accepted as having been verified.
The variance of the deviation ratio is O'863, which is reasonably close to
1, and, thus, the theoretical estimation variance a~ (x ) provided by formula
(11.15) is an effective predictor of the experimental variance of the errors.
As in the preceding Chuquicarnata case, the normal distribution under-
estimates both the proportion of small errors (between O' 5 m) and the
proportion of large errors (greater than 2 m). Again, the usual normal
interval of 20'E correctly estimates the 95% experimental confidence
interval: 92% of the deviation ratios fall within this interval (2aE =
186m).
VNNI.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 61
0,1
\
\
\
\
\
I \
I \
005 I \
\
\
\
-3 -2 -I o 2 3
WWW.minebook.blogfa.com
62 MINING GEOSTATISTICS
Let Z(y) be the ReV grade at pointy. The mean grade of each uni t v (Xi)
centred on point Xi is
1
Zy(x)=-
f z(y)dy
IN =- I zv(x;).
V V(x) N ;~1
,...-
<I> ,...-
u
.,
0::
-
~S2(X}
~
'"
u
u
o
Vi - -
1--,...-
I In
FIG. 11.20. Dispersion of units v grades within stope V.
The problem is then to estimate, if not the entire histogram, at least its
two main characteristics, mean zv(x) and variance S2(X).
Probabilistic interpretation
When the point ReV z (y) is interpreted as a particular realization of a RF
Z (y), the unit grades z v (x;), as well as the stope grade z v (x ), appear as
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 63
realizations of RV's Zv(x;) and Zv(x); the preceding mean square devia-
tion, s2(x), also appears as a particular realization of a RV S2(X) defined at
point x, i.e., corresponding to the stope V centred on point x:
2 1 2
S (X)=-L[Zv(x)-Zv(Xi)] . (11.30)
N;
Variance of dispersion
Under the hypothesis of stationarity of the point RF Z (y), the stationary
expectation of this RV S\x) is, by definition, the dispersion variance of the
units v within V. The following notation is used:
Remark It must be stressed that the RV S\x) and its particular realiza-
tion S2(X) depend on the particular position x of stope Vex), while, under
the hypothesis of stationarity, D 2 (v/ V) no longer depends on the position
x but only on the geometries of v and V and the covariance C(h), as will
be demonstrated.
For each one of K stopes, identical in shape and size, {V(xd, k = 1 to
K}, there is an experimental mean square deviation S2(Xk) calculated from
(II.29). As K ~ co, the arithmetic mean of these K experimental deviations
tends towards the stationary expectation of S2(X), i.e., the variance of
dispersion D\v/ V).
If we are only interested in one particular stope V(x), then D\v/ V) will
be an estimator of the experimental mean square deviation S2(X), just as
the expectation of a RV is an estimator of a particular realization of this
RV.
Example Suppose that a stope V(xd is divided into four equal units
{vex,), i = 1 to 4} and that grades Zv(Xi) are assigned to each unit according
to the result of casting an unbiased six-sided die.
The distribution resulting from casting the die is uniform with a mean
equal to (1 + 2 + 3 +4 + 5 + 6)/6 = 35 and a variance (J"2 = 2917.
Consider three stopes {V(Xk), k = 1 to 3} having the same mean grade
Zv = 35 and consisting of the following three series of four units:
2
k = 1, Z",= 6, 3, 2, 3, Zv = 35, s = 225;
k =2, z, = 6, 5,1,2, zv = 35, S2 = 4.25',
2
k = 3, z.; = 6, 2, 2, 4, zv = 35, s = 275.
NNI.minebook.blogfa.com
64 MINING GEOSTATISTICS
which is equivalent to
4
2 1 2 1 22
D (v/V)=- [E{{Zv(x)-Zv(Xi)] }=-[(T =(T =2,917.
4 i~l 4 i
1
S 2 (x)=-
V
f V(x)
[Zv(X)-Z1J(Y)] 2 dy.
v V. (II.32)
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 65
It can be shown that the order of the integral sign and the expectation
symbol can be reversed to give
D2 (v/ V ) = 2- f E{[Zv(x)-Z,,(y)]2}dy
V Vex)
(II.33)
1
= V f O"E'
Vex)
(V(x), v(y)) dy, u V.
The variance of dispersion D 2 (vl V) thus appears as the mean value over
V of the estimation variance of grade Zv(x) by grade Z,,(y) of a unit v
located inside V.
(In practical applications, care must always be taken to distinguish
2
between the dispersion variance, denoted by D , and the estimation vari-
ance, denoted by aI.)
Remark For v = 0, the unit v becomes a point and is said to have a point
support. In this case, the restriction v V is verified and formula (11.33)
gives the variance of dispersion of a point variable within V, which will be
denoted by
1
D 2 (O/V)=-
V
f v(x)
E{[Zv(x)-Z(y)] 2}dy (I1.34 )
with
Zv(x)= VI f v{x)
Z(y)dy.
with
If n is not too small in relation to N and if the n known values z,,(xd are
distributed approximately uniformly over the field V, then s~ can be
WWW.minebook.blogfa.com
66 MINING GEOSTATISTICS
1
D 2 (vjV)= V J 2
CTdV(x), v(ydy, v V.
Vex)
~
V
f
V(x)
C(V(x), v(ydy = C(V(x), V(x= C(V, V), if v V.
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 67
I 2
D (v/ V) = ii( V, V)- ii(v, v). I (11.36)
It can be shown that this last expression remains valid even if the covari-
ance C(h) does not exist, provided that the serni-variogram y(h) is defined
(the RF Z (x) is then only in trinsic).
The dispersion of the unit v within the deposit G is equal to the sum of
the dispersion of v in the stope V and the dispersion of these stopes V
within the deposit G.
Note that D2(vIG)~D2(V/G)if v < V.
The dispersion variance decreases as the support increases; it is well
known that there is much more variability among the grades of core
samples (v of several kilograms) than there is among the mean grades of
blocks (V of several thousand tons).
D 2(O/oo)= lim
V-a:>
.If
V v
E{[Z(y)-m]2}dy.
NNI.minebook.blogfa.com
68 MINING GEOSTATISTICS
(ct. relation (11.3)), and, thus, the preceding limit exists and is written
2(O/oo)= 2(0/V)=Var{Z(y)}=
D lim D
v--,>c<)
CeO). (11.38)
When the RF Z (y) has a stationary and finite a priori variance, this
variance is the limit of the dispersion variance of the point variable Z(y) in
a very large field.
Under the hypothesis that this a priori variance exists and, thus, the
covariance C(h) exists, expressions (11.35) and (11.36) give
2(0/00)=
D C(O)- C(oo, (0)= CeO),
D\O/oo) = y(oo, (0) - 1'(0) = Hco, co) = C(O),
since 1'(0) = O. Thus, for the finite a priori variance RF, we have
C(oo, (0) = C(oo) = 0 (11.39)
and
ii(oo, (0) = '}'(oo) = C(O) = Var {Z (y )}.
However, there are RF's which have an unlimited capacity of dispersion,
at least over an entire deposit or a metallogenic province. In such cases,
neither the a priori variance nor the covariance are defined and the dis-
2(0/
persion variance D V) increases indefinitely with V, d. Fig. II. 7.
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 69
WWW.minebook.blogfa.com
70 MINING GEOSTATlSTICS
\
\
\
~
~
0-1 \
\
I \
I \
I
I
I
J
J
I
'I
001
o 02 2 3 4
--
FIG, II.23. Dispersion of V-block grades in the bench. Block V,20mx20rn x
13 rn: rn =2'16, S2 = 0346% Cu:t.
VNNI.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 71
\
\
\
\
~
a; \
O~
>,
\
oc; \
<1> \
:>
\
...
<:T
\
~ \
\
\
\
-,
'\
-,
001 "-
I --------.-.-L_ _---'------="'-_ _- - '
2 3
NNI.minebook.blogfa.com
I (a)
I I
II
II
)
II
~ I
~ 0 I I
>.
u
...
<::
::l
<:r
Ol.l
~
0-01
65 70
(b)
03
o 2
% sic,
FIG. 11.25 (a). Dispersion of Fe grades (hematite ore). m = 666% Fe, S2 = 2 3
(% Fef. (b) Dispersion of Si0 2 grades (hematite ore). m =0.58% Fe, s2=020
(% Si0 2 ?
VNNI.minebook.blogfa.com
~
s;
o-,
u
c 0-1
'"0-
:::J
'"
~
50 70
%. Fe
FIG. II.26. Dispersion of Fe grades (global ore). m ::;: 588% Fe, s~ = 84.9% 2.
~
~
>- 0-1
..
1,I
c
u,
....
:::J
0- ~
I
I
I
I
~
I
I
I
30 50 70
"/0 Fe
FIG.IL27. Dispersion of Fe grades (deep poor ores). m:= 505% Fe, S2::;: 50-1%2.
WWW.minebook.blogfa.com
74 MINING GEOSTA TISTICS
200
-r:>
..,.,.L.x~ xL
>:
/
/x
100 x
/
10 20 30 90 99 995
Curnulc tive f req uancy (,,; x )
VNIW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 75
(i) When the dimensions of support v are small compared with those of
the field of dispersion V (e.g., v representing a core sample and Va
mining unit or a stope), then (a) the histogram of the grades of
elements of weak concentration (Au, Ag, precious metals, U, Cu,
Ni, impurities in weak concentration such as Si0 2 , Ah03, Fe, ... )
are asymmetric with a tail towards the higher grades (direct
lognormal type); (b) the histogram of the grades of macro-elements
of strong concentration (Fe in deposit, Mn in deposit, phosphate
P 20s, ... ) are asymmetric with a tail towards the lower grades
(inverse lognormal type).
(ii) As the support v increases in size (e.g., v representing a mining unit)
a blending effect is produced, which tends to slowly tone down the
previous asymmetries. Lognormality is not preserved when the
support v becomes too large.
(iii) A cleary multimodal histogram reveals, in general, t the existence of
heterogeneous mineralizations (the presence of several types of ore
or the influence of poorly mineralized areas on the fringes of the
deposit). The interpretation of such a histogram must be made with
due regard to statistics (e.g., fluctuation due to an insufficient
number of samples), geometry (e.g., position of the field of dis-
persion in the deposit) and geology (e.g., mineralogy, faults, direc-
tions of enrichment of the are).
(iv) The preceding remarks concern the grades of elements. For
geometric-type regionalized variables, such as mineralized or over-
burden thickness, or mining type ReV such as percentages of ore
recovery or granulometric ratios, the histograms are of a more
symmetrical type, as shown on Fig. 1I.29(a) and (b).
Figure 1I.29(a) shows the histogram of the mineralized Ni-Iaterites
thickness of the Prony are body (New Caledonia). The histogram was
constructed from 176 values measured on the 176 vertical bore-holes of a
regular square grid of lOO-m side.
Figure I1.29(b) shows the histogram of the percentage of are (mesh
>6 mm) recovered on mining units at Moanda (Gabon) manganese
deposit. This histogram was constructed from the results of mining 1400
unit blocks corresponding to an area of about 42 ha.
t But the presence of several modes on a histogram does not imply that each of these modes
necessarily corresponds to one type of mineral that can be distinguished at the mining stage.
For example, a banded hematite quartzite ore surveyed by drill cores (of a metre or so in
length) would show a multimodal histogram of the type shown on Fig. II.27 and still be
considered as an homogeneous are at the scale of mining (on blocks of 10 or 20 m in
dimension).
NNI.minebook.blogfa.com
76 MINING GEOSTATISTICS
7 f--- ( 0) (b)
6 l-
f- 20 -
i:? 4 f-
<:
."
g. 3 f---
."
~ 10 f-
2 f-
o
f-
h
124
mfm 10 20 30
~ rhh,n
40
4f-
2 f-n
51-5
rrf
57-2 64-7
Thickness (rn) Mn (%)
~/
20
o
Va
o o o
'"
u
<:
<l>
& 10
o o '"
LI..
o 0 o 5
~
Negative
drillholes
o I 2 5
Thickness {rn)
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 77
11.0. REGULARIZATION
z,,(x)=!
v
f,,(x)
z(y)dy.
Z.,(x)=! f Z(y)dy.
v v(xl
E{Z.,ex)}=! f
v "
E{Z(y)} dy =! f
v u~)
m. dy =m, Vx;
WWW.minebook.blogfa.com
78 MINING GEOSTATISTICS
Since the point serni-variogram "1(h) is stationary, the last two terms of
the previous expression are equal and, thus,
Remark 1 For distances h which are very large in comparison with the
dimension of support u, the mean value y(u, Uh) is approximately equal to
the value y(h) of the point semi-variograrn and we obtain the very useful
practical approximation
y,,(h) ~ y(h)- Yev, v) for h v. (11.42)
At large distances, h v, the regularized serni-variogram is simply derived
from the point serni-variograrn by subtracting a constant term Hv, u)
related to the dimensions and geometry of the support v of the regulariza-
tion, ct. Fig. II.3l.
C(O)
p,,", ,," ---- ----- --~-------I---=
~g\Jlon~_~.~__S.~_I_~_ !y(v,v) /v(h) I
...-
/// -:
/
/
,-- ...-
L --
/
o
/
.....- .....-
,/
-
...-/
- - --- ----------
h
WWW.minebook.blogfa.com
II. THE TI-IEORY OF REGIONALIZED VARIABLES 79
x ~----+--.
x+h --t-.
FIG. 11.32. Core samples aligned along a bore-hole.
Zv(x)=!
v
f
v(x)
Z(y)dy,
where the sign Iv(x) dy represents, in fact, a triple integral over the volume
v. When the diameter of the core is smallt compared to length I, the
regularization effect of the cross-sectional area s of the core sample can be
t This approximation will not be valid for samples taken from sampling pits, for example,
because the diameter of the pits will be of the order of the distance between the samples (0' 5
to 1 m).
NNI.minebook.blogfa.com
80 MINING GEOSTATISTICS
neglected. The mean value over the length I of the core sample can then be
written in terms of a single integral in the direction of the length of the core
as
Zv(X)=Z/(X)=_,l iI(x)
Z(y)dy. (11.46)
Regularized semi-uariogram
When the cross-sectional area s of the core sample is negligible, two core
samples can be considered as two aligned segments I and I" of the same
length I and separated by a distance h, ct. Fig. II.32.
According to formula (H.4l), the regularized semi-variogram can then
be written
(11.47)
If u denotes the coordinate along the length of the bore-hole, the first term
y(l, lh) of the expression (11.47) for the regularized semi-variogram is
written
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 81
r I r+1 I r r+1
------
... u I
a o I
I
I
t t, t t
u u u u'
r~1
0) Ihl~/:
2
1 y (l, lh)= r du f+I(U'-U)dU+ fdU[i
U
(u-u')du'
+ f+1 (ul-U)du}
These integrals can be evaluated quite simply to give
1 ,2 I
y(l,lh)=31 2(31-r)+3 forr=lhl~1
r -1/3, Vr ~ I,
d. Fig. 11.34.
For distances Ih I~ I, the behaviour at the orrgin of the regularized
semi-variogram is parabolic. The cores of length I overlap each other and
the regularizing effect is strong enough to change the behaviour of the model
at the origin from linear to parabolic.
For distances Ihl;?= I, which, in practice, are the only distances observable,
the linear behaviour of the point model is preserved and the regularized
model differs from the point model by a constant value
'Yl(h) = y(h) - y(l, I) = r -l/3, r = Ih I?: l.
(Note that for the linear point model the approximation of formula (II.42)
becomes a strict equality.)
If the regularized curve for Ih I ~ I is projected towards h = 0, the
ordinate axis is intercepted at -1/3 and this negative value is called a
"pseudo-negative nugget effect" due to regularization.
2l f - - - - - - - r - - - - - -
3
'Y
/
/
o
.:
L. _
,
. i //
3~ .....
r
In practice, this exact formula is rather difficult to use and the approxima-
tion of formula (11.42) is preferred:
r 3
ydh)=y(h)-y(l, l)=log /+2' Vr ";3/, (II.50)
with
y(l, I)=log l-l
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 83
Given the conditions, Ih I~ I and a ';3 31, which cover almost all practical
cases, this approximation is excellent.
WWW.minebook.blogfa.com
84 MINING GEOSTATISTICS
Given the conditions, Ih 1;:;.1 and a 3a ;:;. 3/, which cover almost all prac-
1=
Regularized range For a transition point model, i.e., for a point model
with a sill and a (practical or actual) range a, two point variables Z (x) and
Z(x + h) are uncorrelated when the distance h separating them is greater
than the range (lh I> a.)
As a consequence, the two variables regularized by cores of length I,
Z, (x) and Z, (x + h) are uncorrelated when the distance between each point
of l(x) and each point of lex + h) is greater than the range a, cf. Fig. II.35;
for which it is necessary and sufficient that lh I> a + 1.
The regularized range is, thus, (a + 1) and this is the distance at which the
regularized model )'f(h) reaches (actually or practically) its sill y/(oo). Note
that the previous practical approximation, 'Y1(h) = y(h) - y(l, l) amounts to
assigning to the regularized model the same range as that of the point
x x+1I
I 0+1 1
FIG. 11.35. Regularized range.
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 85
model. In fact, for the two transition models in current use (the spherical
and the exponential), when IhI> a and a ~ 31, ydh) can be approximated
with excellent precision by its sill value y/{co).
A
---f--- f
B
BenCh I
__J _
OeD
u (x+h)
Horizontal plane
Zo(X u , xv)=i
111 Z(X
0 u, xv, xw)dx w (II.54 )
WWW.minebook.blogfa.com
86 MINING G EOSTA TISTICS
Remark Even though the support of the regularization (I) may be the
same, care should always be taken to distinguish between grading over a
constant thickness l and regularization by core samples of length t along
the length of a bore-hole, d. Figs 1I.36 and 11.32. In the case of grading,
the two variables Zo(x u, xv) and ZG(x u + h.; Xu +~ are defined on two
segments I and lh which are parallel and separate(f by a distance vector
h = (h", h u ) orthogonal to the direction of the length l, while, in the case of
regularization by cores, the two segments I and lh are aligned and
separated by a distance vector h parallel to the direction of length I.
Graded semi-uariogram
The serni-variogram of the graded RF over the constant thickness I is
written in the standard form of formula (II .41):
(11.55)
where x = (xu, xv) and h = (h", h v) denote the coordinates in the plane
orthogonal to the direction of the grading thickness, d. Fig. 11.36.
In calculating the mean value y(l, Ih), three directions in space are
involved and the calculation is simple only when the point model
vtb: h.; h w ) is isotropic, i.e., if 'Y(h)= 'Y(r) with
r=(hl=J(h;+h~+h:)
t The regularization effect of the cross-sectional area s of the core can be neglected in relation
to the thickness f of the formation when integrating over the support (v = s x f) of the
intersection of the bore-hole with the formation.
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 87
r
Mk::----------~
r
M'
This part is not essential and the reader may go directly to the next
section, II.DA.
Remark For distances h ";3 I, the model rdh) graded over the length I is
practically indistinguishable from the model Yl (h) regularized by cores of
length I, d. formulae (II.58) and (lIAB). On the other hand, for distances
Ihi;;;:; I, the effect of regularization by cores of length l is greater than that of
WWW.minebook.blogfa.com
88 MINING GEOSTATISTICS
y(h)=logr withr=lhl.
(Recall that this logarithmic model should only be used when regularized
over non-point supports.)
The exact isotropic expression of the semi-variograrn graded over I is
found by expanding formula (I1.56):
28 log cos 8 .
'YG(h)=--+ 2 -log Sin 8 (II.59)
tan () tan ()
r r2 r 3,2 1 r4
for' ~ I, l'G(h)=7f/+[ilog,-zP-12 [4+"';
(II.60)
3 r 1 [2 1 14
for, ~ I, YG(h) = 2+ log [+12 r2 ~ 60 ,4 + ....
(II.61 )
3,4. The spherical and exponential models The analytical expressions for
these graded semi-variograms are long and awkward to handle and for this
reason the following charts are used:
(i) chart no. 1, which gives the graded serni-variogram of the point
1/
spherical model (with sill 1 and range a) as a function of Ih a and
for different values of 1/ a;
(ii) chart no. 11, which gives the graded serni-variogram of the point
exponential model (with sill 1 and parameter a and, thus, practical
range a'=3a) as a function of Ihl/a and for different values of ll a.
VNNI.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 89
The graded sill is defined as follows. If the sill of the point model is 1,
then, according to formula (HA2), the sill of the model graded over [ is
1'0(00) = 1- y(l, f),
the mean values .y(l, I) being given by formulae (11.51) and (II.53).
The model graded over l and the model regularized by cores of length [
thus have the same sill.
The graded range may also be defined. When the distance Ihl becomes
greater than the (actual or practical) range, every point of [ is independent
of every point of L; and, as a consequence, the range of the graded model is
equal to the range of the point model, whatever the grading thickness [
may be.
Recall that the range of the model regularized by cores of length I is,
strictly, equal to (a + I), where a is the range of the point model.
In practice, for both the spherical and the exponential models we have
the following.
(i) For distances lh] ;3[, we will use the standard approximation of
(IIA2):
yo(h) = 1'(h) - y(l, I). (11.62)
The relative error involved in this approximation never exceeds 10%
in the most unfavourable cases (e.g., Ihl = I and If a = 0,1) and is less
than 1 % in the majority of cases encountered in practice (!h! > 3I
whatever the ratio If a may be).
(ii) For distances !h] < l, we can either (a) use the exact formulae of the
graded models given in J. Serra, (1967 a), Vol. 2, pp. 446 and 461) or
(b) if a I (in practice if the actual or practical range is greater than
3[), we can assimilate the point models to their tangents at the origin
(linear model of slope w); the graded models can then be deduced
from formulae (II.57) or (11.58) corresponding to the linear model.
WWW.minebook.blogfa.com
90 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 91
(ii) The regularized theoretical expression 'Y... (h) (or its approximation
given by formula (11.63 of this point model y(h) is then calculated
and compared with the experimental curve y~ (h). The various
parameters (ranges, sills, ratios of anisotropies, etc.) of the point
model are then adjusted in such a way as to bring 'Yv (h) in line with
y~ (h).
(iii) Once the point model y(h) is determined, the theoretical expression
'Y.,{h) for the regularization over a possible second support VI can be
deduced and checked with the corresponding experimental curve
'Y~' (h).
The calculation of the regularized theoretical expressions 'Yo (h)
and 'Yv{h) is simple only when the point model consists of the
established models which are in current use, and for which formulae
and graphs are available, e.g., if the point semi-variogram y(h) is a
spherical model.
with 1 =;: 3 m, a = 40 m and C, =;: 133. Thus, the sill of the point spherical
model is C= 138 (% Fel.
NNI.minebook.blogfa.com
92 MINING GEOSTATISTICS
In Fig. 11.38, the theoretical point spherical model y(h) with range
a = 40 m and sill C = 138 is shown as a dashed line. The theoretical
regularized model YI deduced from the previous point model by the prac-
tical approximation (II.63),
ydh) = y(h) - 1'(1, I), Yh ~ I, a> 3/ = 12 m,
is shown as a first solid line. The regularized model 'Y41 deduced by the
same approximation is shown as a second solid line.
200 f-
150 f-
a 3 12 24 36 0 48 60 72
h (m)
41 (41)3
C 4 1 = 1'41(00)= C [ 1 - - +--3 = 117'5,
2a 20a J
and correctly fits the experimentally observed sill.
Thus, the practical rule (II.63) for the regularization by cores of length 1
provides models 'Yl and Y41 which are very good fits to the corresponding
experimental semi-variograms.
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 93
WWW.minebook.blogfa.com
94 MINING GEOSTATISTICS
But the experimental curve 'Y~ (h) is clearly above the theoretical
straight line 'YG(h) derived from the hypothesis of isotropy, and we can,
thus, conclude that the phenomenon is anisotropic: the horizontal vari-
ability is greater than the vertical variability.
(Note that this study dealt only with small distances, h being less than
10 m in Fig. II.39.)
1 1-~
Vertical
2/ (l Ho riz ont o l
10
05
2 3 4 5 6 7 8 9 10 20
2h (m)
FrG. II.39. Core-regularized and graded semi-variograms.
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 95
j(v, Vi) = ~
vv"
f f,,'
dx y(x - x') dx '. (II.65)
y(v, Vi) denotes the mean value of the function y(h) when one extremity of
the vector h describes the domain v and the other extremity independently
describes the domain Vi.
It is recalled that if the domain v is three-dimensional, then the notation
Iv dx in fact represents the triple integral Ht d.r, dx, dx w , (xu, xv, x w ) being
the three coordinates of the point x.
Thus, in the three-dimensional space, the mean value f(v, Vi) between
the two volumes v and ' involves a sextuple integral. The advantage of
avoiding a direct analytical computation of these sextuple integrals is
obvious.
Two solutions are possible. The first one is to calculate the mean values j
numerically using a computer. The second one is to break the analytical
resolution of the multiple integrals into successive stages, some of them
being calculated in advance once for all, These stages or intermediate
integrations correspond to the definition of auxiliary functions which can
be presented either as charts or in their exact analytical form.
NNI.minebook.blogfa.com
96 MINING GEOSTATISTICS
r: (Xi -j(;)
x;
,
v
Remark 2 The zero effect. Whatever the adopted structural model, yeO) =
o by definition.
As a consequence, the two centred regular grids should be chosen so that
none of the discrete approximation points (Xj, i = 1 to n) and (x;, j = 1 to
n') coincide. Each time two points coincide (x, = xi) in the numerical
estimation of y, a greater importance is attached to the zero value yeO) = o.
This zero effect will cause the mean value of the semi-variogram to be
systematically underestimated and, correlatively, the mean value C of the
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 97
WWW.minebook.blogfa.com
98 MINING GEOST ATISTICS
greater than the largest range of the model y(h), then y(v, Vi) = sill value,
Again, if the dimensions of the two domains v and Vi, separated by a
distance h, are very small with respect to the range of the transition model
y(h), then y(v, v')= y(h), Now, if only the dimensions of v' are small with
respect to this range, then v' can be approximated by its centre of gravity
(n ' = 1) and v can be approximated by n discrete points in the usual
manner. Still again, if the domain v is a parallelepiped very much flattened
in the vertical direction, and if the model y(h) is isotropic, then v can be
approximated by its median horizontal plane - which is then approximated
by discrete points over two dimensions (n ~ 6 x 6 points); on the other
hand, if there is a strong vertical variability represented by an anisotropic
model y(h), it is advisable to approximate the vertical dimension of the
parallelepiped v also.
Y(V,V)=2
v
1 f. dx
l)
f y(x-x')dx'=21 1+
l) V -00
k(x)dx 1+
-00
y(x-x')k(x')dx'
with
k(x)= {0
I if x E v
if and
1
~
1+ 00
k(x)dx=1.
I not v -00
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 99
with
k(u +x')k(x')dx'.
y(v, v,,) = 21
v
f. dx f
v Vh
y(x -x') dx' = 21
V
f
-co
+00
y(h + u)K(u) duo
(11.69)
The Cauchy algorithm for reducing the order of the integral is of interest
only if the function K(u) is known a priori. Now, for rectangular domains v
with sides parallel to the system of axes, the corresponding function K (u),
also called the "geometric co variagram of v", can be readily determined.
(a) In one dimension, v is a segment of length l and its geometric
covariograrn is a function of one variable u:
K (u ) = { I -I u I ~f Iu I'!% t,
o If not.
Consider now the symmetric function of y(h):
f(h + u) = Hy(h + u)+ y(h - u)].
Then, for aligned segments of identical length l translated one from
the other by the distance h, d. Fig. HAl, the Cauchy algorithm
(H.69) leads to the following expression for the mean value y(l, l,,):
1
,
-----
I I
h
...
FIG. 11.41. Two aligned segments
WWW.minebook.blogfa.com
100 MINING GEOSTATISTICS
y(S, Sh)=4 t1 i I
'I
L2 (
---------------
~--+-------~-----------,-
5
,
I
I
I~"~-~--~---~- ~I
FIG. 11.42. Two coplanar translated rectangles.
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 101
y(v'Vh)=8 fa
,
11, 1
(1-ud(1- U z)(1 - u 3)
(11.72)
x f(h I + flU " h + Izuz, h + 13u3) dUl duz dU3.
The mean value nv, v) is obtained by setting the vector h to zero in
formula (11.72).
11
~
V
I, (
/1 2
hi
h3
WWW.minebook.blogfa.com
102 MINING GEOSTATISTICS
The four abscissae Xi, Xi, Xk of the discrete approximation and the cor-
responding weights Ai, Ai' Ak have been calculated by A. Marechal (1976,
internal report) and are given in Table 11.1.
1 2 3 4
For a spherical or exponential model y(h) and for practical values of the
range, the relative precision is always better than 1 %.
FUNCTION GBAR(---)
v and Vi can be any domain, provided that they are of rectangular form.
The dimensions of the two domains do not have to be the same, e.g., v can
be a three-dimensional parallelepiped (a mining block) while Vi is a seg-
ment (core samples along a bore-hole), cf. Fig. 11.44.
Each of the two domains is approximated by a centred regular grid of n
points, n taking the values given in (11.67).
When the two domains v and Vi have a non-empty intersection, and
particularly when one contains the other, the index INTER = 1 allows
w
101MI:3
( v)
\
-.h._-----CCAA 1
01
IDIM 2: I
CA2
WWW.minebook.blogfa.com
104 MINING GEOSTATISTICS
avoiding the zero effect: each point of the discrete approximation of one
domain is shifted randomly from its corresponding node on the regular
centred grid.
The three-dimensional point semi-variogram y(h) is read as a sub-
routine FUNCTION GAM(HX, HY, HZ). The three coordinates
(HX, HY, HZ) of the vector h are relative to the reference system of axes.
(In section IILB.5, an example of FUNCTION GAM corresponding to a
very general structural model y(h) is given.)
Relative to the reference system of axes, each domain v or v' is defined
by one apex and one, two or three side vectors according to whether the
domain is one-, two- or three-dimensional. Thus, in the example of Fig.
11.44,
(i) the rectangular parallelepiped v is defined by the coordinates of one
of its apex 01 and the coordinates of the three side vectors CAl,
CBl and cci,
(ii) the segment v'is defined by the coordinates of one of its extremities
02 and the coordinates of the vector CA2.
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 105
'#NNI.minebook.blogfa.com
106 MINING GEOSTATISTICS
FUNCTION F(---)
This subroutine calculates the mean value y( v, v) over a rectangular
domain of any dimension (segment, rectangle or parallelepiped). The
subroutine considers the Cauchy-Gauss discrete approximation method
and the corresponding formulae (II.73) to (11.75).
w
cc
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 107
c
C a ~ ~ _ ~ ~
WWW.minebook.blogfa.com
108 MINING GEOSTATISTICS
21 G=G+G~M(HX,Hy,Hlj
F=F+wII)*WIJ)*G/4.
20 CONTINUE
~ETURf\
c
C PAEUllElEPIPEC
C
:3 00 30 1=1,4
CO 30 J=l,it
CO 30 K=l,4
G-=O.
Jl=-1
J2:1:-1
J3=-1
CO 31 IN=I,2
Jl=- J 1
00 3 1 I ~:= 1 , 2
J2=- J 2
00 31 IL:tl,2
J3=- J ~
HX=Jl.PII).CA(1)+J2*PIJ).(E(1)tJ~.f{~}.CC{11
HY;Jl*P(1)*CAIZ)+J2*PIJ}'CEI2)+J3*FCK).CC(2)
~l=Jl*P(IJ*CA(3}+J2*P(J)'(E(3)+J~*FIK)*CC(3)
31 G=G+f~~(~X,hy,HlJ
F=F+W([)*W(J)*wIK)*G/8.
30 CONT HUE
HTUR ...
END
L
A B
FIG. II.46. Segment AS of length L.
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 109
The auxiliary function X(L) is defined as the mean value of y(h) when
the extremity of the vector h is fixed at the point A and the other extremity
describes the segment AB of length L:
1
X(L)= y(A, AB)= L I, y(u) duo
r
L
(11.76)
The auxiliary function X(L) thus represents a simple integral of the semi-
variogram y(h).
The auxiliary function P(L) is defined as the mean value of y(h) when
the two extremities of the vector h describe, independently of each other,
the segment AB of length L:
F(L)=y(AB,AB)=2
L
1 lL du lL y(u-u')du'.
0 0
(II. 77)
2
F(L)=2
L
fL (L-u)y(u)du=22
0 L
iL
0
uX(u)du. (11.78)
L
A--'~B
/ u
1\ h
~--
v
--L._---'
C D
I
u
WWW.minebook.blogfa.com
110 MINING GEOSTATISTICS
It can be shown that this double integral of the isotropic function y(r) can
be reduced to the following simple integral:
2
a(L; I) = HAC, BD) = /2 f0
I
Note that the auxiliary function a (L; l) is not symmetric in (L; I):
ai]; L) = y(AB, CD) y(AC, BD).
Note that the auxiliary function X(L; I) is not symmetric in (L; I):
x(l; L) = y(AB, ABCD) oj; HAC, ABCD).
The two-variable auxiliary function F(L; I) is defined as the mean value
of y(h) when both extremities of the vector h describe, independently of
each other, the entire rectangle ABeD:
F(L; l) = y(ABCD, ABCD) (11.81)
This auxiliary function is symmetrical in (L; I). It represents a quadruple
integral of the isotropic function y(r). The Cauchy algorithm of formula
(II.70) can be applied to the auxiliary functions a(L; 1) and x(L; l) to
reduce this quadruple in tegral to the following simple integrals:
2
F(L; 1)="2
L
iL
0
(L - u)a{u; I) du = 22 fL
L 0
UX{u; I) duo (II.82)
Note the similarity between these formulae and those of (II. 78) cor-
responding to one dimension. In fact, the two-dimensional function Fil; 1)
is nothing other than a one-dimensional function peL) with the mean value
over I having been taken beforehand.
The two-variable auxiliary function H (L; I) is defined as the mean value
of y{h) when one extremity of the vector h is fixed at anyone of the
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 111
L
A- __ B
I I
C D
FIG. lIAS. Rectangle ABeD compressed (l = 0).
The auxiliary function F(L; l) corresponds to the highest order (4) multiple
integral of y(h), all other auxiliary functions can be deduced from F(L; I)
by differentiation or by setting a parameter to zero:
1 i
a(L'l)=- - L 2 a
F (L ' 1)=-Lx(L' I)'
2
, 2 aL 'aL "
1 a 2P(L;
x(L; l) = 2L aL L I) and X(L) = x(L; 0);
(11.87)
2
1 a 2 2
H(L; 1)= 41L al aL I L F(L; I);
WWW.minebook.blogfa.com
112 MINING GEOSTATISTICS
L
A B
I( C I
D
E - - - - - - - - - -L.F-..
~:""""G-----_..::..o H
......
-
FIG. 11.49. Parallelepiped with a square base.
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 113
-y(h)=r withr=lhl.
In one dimension,
1 2 2 2 L2 2 2 L
2
l-r J(L 2 + [2)
a(L; l)=-J(L +[ )+- ~/2 (L-J(L +1 +-log ;
3 3 [L
3 2 2 2
1L
X( L , [ ) = ~6 - + J(L 2 +[ 2 ) (1 - -1 -L-) +--log-----'-----.:....
1L I+J(L + f )
12 4 6 f 3 l L
1 z2 L + J(L 2 + [2)
+-12 L log [ ;
2 3 (II.91)
F(L'/)=J(L 2+[2)(---
IlL 1 -/2) +-
-2 - - 1 (L
- +z3-)
, 5151 15L 2 15 z2 L 2
2 2 2
1L I+J(L + / ) 1 l2 L+../(L 2 + Z2 )
+ 6-/ log L + (; L log Z ;
1 2 [2 L+J(L 2+ f ) L 2 1+J(L 2 + z2 )
H(L; l)=3 J(L + [2)+ 6L log I + 6l log L .
WWW.minebook.blogfa.com
114 MINING GEOSTATISTICS
+ (7T_0)
23'
tan (J. (11.93)
. 25 tan ' ()
F(L; I) = log I-log sm (J - 12 +-6-log sin ()
1T L 1 L 2 L 11 L 2 1 L 4
3
x(L; 1)= log 1-"2+2/+3 [2 log i-18 [2- 60 14+' .. ; (II.94)
2 2 4
3 1T L 1L L 25 L 1 L
F(L', i) = log 1--+-
2 3 -+-
I 6 -log
[2
- -- -
[72 P- 180
- - {4
+ ....
.. ;
7T I 1 [2 1 14
X( L ', I) = log L -1 + -6 - -- -+ - - +
L 12 L 2 180 L 4 . .
. (II. 95)
3 1T 1 1 [2 I 25 [2 1 [4
F(L; 1)=IOgL-"2+'"j L +"6 L z log L -72 L 2-180 L 4 + ....
Remark For values r = Ihl < 1 the function log r has negative values.
Consequently. for small values of L and I, the auxiliary functions a, x, F
can take negative values. Every estimation or dispersion variance defined
on non-point supports and calculated with these auxiliary functions is.
nevertheless, always positive. since every regularization of the logarithmic
model is a conditional positive definite model.
VNNI.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 115
Linear equivalents
From formula (11.92), the one-dimensional auxiliary function F(l) IS
written as
F(l) = log I ~ 3/2.
The auxiliary function F(a; b) for a rectangle with sides (a, b) and a > b,
is given by the first three terms of the limited expansion (II.95):
3 7Th 3 b
F(a; b)=log a --+- -=log a --+-
2 3 a 2 a
3
F(a;b)=log(a+b)-2=log [a( 1+-;;- b)] -2=loga-2+~+""
3 3 b
Thus, the two-dimensional auxiliary function F(a; b), for the rectangle
with sides a and b, is approximately eq ual to the one-dimensional auxiliary
function F{l) for the segment I = a + b :
F( a; b) = F(l) = log l- 3/2 with [ = a + b. (11.96)
WWW.minebook.blogfa.com
116 MINING GEOSTATISTICS
(II.99)
WWW.minebook.blogfa.com
II. THE THEORY OF REG IONALIZED VARIABLES 117
In two dimensions, the exact expressions for the auxiliary functions can be
found in J. Serra (1967) and D. Guibal (l973a). In practice, the following
charts, graduated in terms of L/ a and II a, are used.
Chart no. 11 gives the function YG(L)=a(L; l)-a(O; l)=a(L; l)-F(l).
Chart no. 12 gives the function X(L; I).
Chart no. 13 gives the function H(L; I).
Chart no. 14 gives the function F(L; I).
Chart no. 15 gives the function F(L; [2).
Chart no. 16 gives the difference a(L; P)-a(O; p)= a(L; l2)-F(l; l).
5. The r B model
8
y(h)=r withr=lhland OE]O,2f.
The function rO is conditional positive definite only when the parameter 0
is contained in the interval ]0,2[, the limits a and 2 being precluded.
Although this model is not currently used in practice, it has some
theoretical importance and, therefore, the following one-dimensional
auxiliary function can be of interest:
X(L)= [/(8+1), }
(II. roo)
peL) = 2[8/[(0 + 1 )(0 + 2)].
WWW.minebook.blogfa.com
118 MINING GEOSTA TISTICS
with
ii(O, AB) = ii(O, OA) = y(O, OB) = x(l/2);
because of symmetry,
HAB, AB) = F(l) and ii(O, 0) = 1'(0) = 0
and the extension variance becomes
O"t = 2xUI2)-F(l)
(ii) The estimation variance of a segment AB of length I by two samples
located at its extremities A and B. Letting if;' = {A + B} designate the
set of the two sample points, this variance is expressed as
ut = 2y(g', AB)- y(AB, AB)- Hg', it)
with
'Y(g', AB) = yeA, AB) = y(B, AB) = xU),
because of symmetry;
HAB, AB) = F(l),
Hit, g') = y(A, g') = t[y(A, A)+ y(A, B)] = ~y(l)
and, thus,
uL = 2x(l) ~ F(l) - y(/)/2,
A I B
I
, --------t J
o
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 119
because of symmetry;
y(ABCD, ABCD) = F(l; l) and HO, 0) = 1'(0) =
and, thus,
O't : : ; 2H(lj2; tns-r, I).
(iv) The estimation variance of the square ABCD of side l by the set 'iC
of its four corners, d. Fig. 11.50, is written as
O't = 2y('jg', ABCD)- y(ABeD, ABCD)- y(g, fg)
with
-y(g', ABCD)= yeA, ABeD)= H(l; l),
because of symmetry;
y(g, 'iC) = yeA, 'iC) = ~[y(A, A)+ yeA, B)+ yeA, C)+ yeA, D)]
= !-yU) + ~'Y(lv'2)
and, thus,
crt = 2H(l; l)-F(l; l)-~y(l)-ky(l.J2).
Charts no. 8 and 18 give the estimation variance of a rectangle ABCD of
sides I and L by its median IJ of length l, ct. Fig. II.51:
cr~5 = 2y(IJ, ABCD)-y(ABCD, ABCD)-y(IJ, 11)
= 2x(Lj2; I) - F(L; 1)- F(l).
A I B
C~D
L
FIG. 11.51. Estimation of a rectangle by its median.
As the functions F(L; 12 ) and F(l; I) are given elsewhere (Charts no. 4, 5
and 14, 15), these charts no. 9 and 19 can be used to calculate the
2
three-dimensional auxiliary functions X (L; 1 ) .
WWW.minebook.blogfa.com
120 MINING GEOSTATISTICS
L
FIG. 11.52. Estimation of a parallelepiped by its median square.
As the functions Fil; [2) and P(L) are given elsewhere (Charts no. 5 and
15 and formulae (11.98) and (11.99)), these Charts no. 10 and 20 can be
used to calculate the three-dimensional auxiliary function H(L; [2).
I
I
-\
\
J \
\
\
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 121
ax(b; a)+bx(a; b)
= a+b
y(ABCD, ABeD) = i'(AC+ AB, ABeD)
ay(AC, ABCD)+ bii(AB, ABeD)
=
a+b
'#NNI.minebook.blogfa.com
122 MINING GEOSTA TISTICS
Since
-(AC ABeD)= a'HAC, AC)+Zby(AC, AB)+ay(AC, BD)
'Y , 2(a+b)
aF(a)+2bH(a; b)+aa(b; a)
=-....:..-..;~-----=-....;,.........:-....._....;,.......=--...:.
2(a + b)
we finally obtain
UE
2
I
= Zax(b;a)+2bx{a;h)
a+b
( b)
-F a"
'
a 2P(a)+4ahH{a; b)+b 2P(b)+a 2a(b; a)+b 2a(a; b)
2(a +b)2
Considering formula (11.91)of the linear model and a square panel (a = b),
we get the following values for the auxiliary functions:
F(a)=a/3, a(a;a)=1'077a, xCa;a)=O'652a,
F(a; a)= 0'521a and H(a; a)=0765a.
Thus, (]'~l = 0'0479a for a total sampled length 4a.
=~[F(b)+a{a; b)]
and, thus, we have
u~z = 2x(a; b)- F(a; b)-!{F(b)+a(a; b)].
The numerical application gives lT~2 = 00778a, for a total sampled length
2a.
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 123
Considering the values x(aI2; a)= O'454a and H(aI2; aI2)=O383a, the
numerical application gives ui 3 = Q0288a, for a total sampled
length 2a.
U~3 = Q0288a < U~l = O'0479a < (T~4 = O'056a < U~2 = Q'0778a,
II.E.4. Charts
All the charts given hereafter correspond to the two following spherical
and exponential isotropic models with a unit sill value. All the distances
considered by the charts are relative distances ria (divided by the
parameter a of the model).
Spherical model:
3 r 1 r3
VrE [0, a],
y(r)= 2-;-2 a 3 '
{
1 = sill, V r ;3 a (actual range).
Exponential model:
NNI.minebook.blogfa.com
124 MINING GEOSTATISTICS
Chart no.
Spherical Exponential
Object of the chart model model
Grading
'YG(L)=a(L; 1)-a(O; I) 1 11
Auxiliary functions
X(L; I) 2 12
H(L; I) 3 13
F(L; I) 4 14
F(L; (2) 5 15
u.. f)-a(O; 12 ) 6 16
7 17
8 18
9 19
10 20
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 125
090 -----;..----i--
,
I
,-, ,-'--------t-- ~- j--i---+-
____ ~_+---L
060
I -----j---+----l
i
------->---i-I------:.-
N
-
-
- ,
,
r---,-!----'---+! ,,-
!
0_60
...J
0,50-
'"
>...
_ ..._+-.-, ..
V. h.~
OAO
0,30
020
WWW.minebook.blogfa.com
126 MINING GEOSTATISTICS
10
--------t ---::---~ " t~s~~:."~F'" "1\."" """ "" "~0!J o.!J!J~
9
J""'.....
"I\. '\
8 'j.,.
r-, <, " " .:\. ~
7
6
----- _p. . . -I 1"--,' . !"\.. '" """
-...., .>:i" :""'" ""'" ",,\" <, '\
0.$l~ .~~
1'.. '" I 0'!J !J...
'PO'
2
. r--- - 'I'. _ 0.,
~ - ~ 1'-" 1\ 1\
0,
1\
l\l\ [\1\"
r, 1\ 1'\ l'~~
-.... I
O,g -, ~ :~~I-"':'>'"
-......... ,............ ,,0"'0" \
\ \ \ 1\ \ \ i\ \
\ _
\
1\ 1\
0'8
0-7 ---r- r-, r-, '" I" ~" \ \ \ \ \ \ 1\ \ "
- -......,,, q G'"" ~\ \ \ \ \ "\ \
06
05
r=====::::: <, <, -, "~o""" 1\ \ r\ \ 1\ \ 1\ \
~ i\ \ \ \ 1\
\ \\\ \ \\ \ \1\
0-4 . _, I o \" \ ' 1\ \ \
~" 0'
0 0 \
_" \ \
a I '1,"'0\ I
0-1 02 0-3 04 OG 08 I 2 3 45678910
i/o
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 127
10 r-------..,.--==-----.------,-.o::::-r--..:--r""'r"O<l'""""ll-.....--....-.........,..,...................,...,......................~.......__ .............._______,
9 I----=o::;;i:::::;:=--
8 +--~~-....-:-____+_:::......I,._----=~P-Irt~~~~----"!.~~>J....."~~~~
7 ;----k=---_+__
61---"""""4:::.::::
5 ~==_.jk-~""_<d-.p"..J-Nx.:~~-~~~~~~
4 j----_~.:::- -~-.,-----f"--+~""-<
2!---=:t::::------''''''''''--.;;;;::-
0::.
....
<,
I
0'9
oa
07
06
0-5
0-4
03
02
02 03 04 06 08 2 3 4 5 6 7 8 910
L/o
WWW.minebook.blogfa.com
128 MINING GEOSTATISTICS
~
..... I
0-9
08
07
0-6
05
04
0-3
0-2
06 08 2 3 4 5 6 7 8 910
Llo
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 129
I:l
-.... ,
0'9
06
0-7
0-6
05
0-4
0-3
02
NNI.minebook.blogfa.com
I
~.
090 i
i
s- oso
,
o f
I
I- +- ;-
J 070
II
- ... _-f
!
n
S -
060 -;
--
'"
r5 0-50
1:1
I
--t j---_._-
N
- 1:1
,
(;
030
020
__ -v} -
-t--,.
'I/o ~
,'10,160
140
\'
~
~- ~/O: 180
1/0'2,00
010" "
'"' ., I/o' 2 50 '1'-"""'----' ---- .,,-
V I/o
['300!
= 4'00_-
L'IU-OO
_,n
I
_1..___
o
1- o 01 02 03 04 05
Llo
06 07 OIS ' 09
1:
CHART NO.6. Spherical model. Grading over f: (L: f)-a(O; t).
II. THE THEORY OF REGIONALIZED VARIABLES 131
2
a: E~
-, 18 -- 10
.............
t
- ~,
t
!
-I
t
03
02
. b
01
009
008
007
006
005
/
004
003
002
I/o
WWW.minebook.blogfa.com
132 MINING GEOSTATISTICS
Liz LIZ
10 r-------...,...~-rr__r-..,_____,_,~..,..,,,.....,..._r____,_______..-------,...,...-7T-_,___,________,~.....,_____,__,
9
8
7
6
5
:3
~
<, I
09
0'8
0-7
06
05
04
0'3
02
Uo
CHART NO.8. Spherical model. Extension variance crL.
VNNI.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 133
r:====r==r=~o=r==l+===:L~~r==C~*:r:l
~ ---~~-=::~f~f-~~- c__ -+;
10
---+------;.++----+--+-------r- ----:io-""-
6~-----:---+
51-----
3 1--------.;----._..--
21-----+--+--f'------+------+-
'"
<,
I
0-9
0-8
0-7
0-6
05
0-4
03
0-2
06 0'8 2 3 4 5 6 7 8 910
L!a
CHART NO, 9. Spherical model. Extension variance CT~".
WWW.minebook.blogfa.com
134 MINING GEOSTATISTICS
Ia r - - - -........-------,,..-----rT"-..,..-...-----,----y,-----r--,-__.,~r____.,_-___._-_r___,______,_____,......,..___r_l
9
8
7
6 I------------cf-
5
4
~
<,
I
-..j
0-9
0-8
07
0-6
0-5
0-4
0-3
0-2
2 3 4 5 6 7 B 910
(/a
WWW.minebook.blogfa.com
....-
I
~,
090
1. .
+.
S-
o 080 I
I-
J
';n 070 f-- _~r~M
; 060 -
-
---'
::.'" 0 50
040-
,(;
030
020 h_ _
110 700
/101000
~.
~
o 110 02000 -,
'" 10
v - ,~
0
1-
[
0 0-5 I 15 2 2-5 3 35 4 45 5
L/a
1: CHART NO, 11. Exponential model. Grading over I: ydL).
136 MINING GEOSTATISTICS
I~ c===::r:::::=::F3;;;J==r::r:S:r=:S=3:S::::I~Qr=D-:-1
-- ,------i'
8 r=====;;===1===r:="
-----
~ ~=~=~r===
5 ~===-~+:-~~-.--J
4r-----....=-4:::::::.:=-=~
...----.--...-t.. -....
3 r------'---__
2 - _
1
0:,
....
<,
I
09
08
07
06
0-5
004 -+- ..
03 ---
02
VNNI.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 137
10 r-----.--~~====:'
9
8 F"---'=c;;;..;.:=:.....f--"-"::=;;,;.e:.;....-.~~..:.-
~ ~=--:----:-=:::-~~-~
5 F='--"-~=-=l~-.:--.--:..~~
3--
II-----~
09r---~~
08t-- _
07r--_~
06 --
r---_
05
04
o 3 ~. . .
o
o I 01oQ~---'----'~~-'----'--;~~:-'::"------;~.J.J...7-::"-'7::-,-:"-------'_.L..~LL..Il......l...l-~-.-L....L.L-~-.L....L~L-.....J
2 "3 4 6 7 8 9 10
NNI.minebook.blogfa.com
138 MINING GEOSTATISTICS
I~ C:==:I:=::======Eb:====:=:;;]C~=s[~h:5IS:CS:O;~
:~--_._._---i...-._-------T---_._----':----_ ....j,~".,_.
8 ---------------
~ r::==r=+=::=;=-=:"-+=4::~
....----.i.__~.~~m ~ ;
5F==:~~~~~
2 ---
~
I
09
08
07
06
0-5
04
03
0-2
0-1
01 0-4 06 0-8 ( 2 3 4 5 6 7 8 9 Ie
Ua
CHART NO. 14. Exponential model. Function F(L; I).
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 139
8
7
6
5
I
0-2 03 0-4 06 08 I 2 3 4 5 6 7 8910
Lla
CHART NO. 15. Exponential model. Function F(L; [Z).
WWW.minebook.blogfa.com
I _, ~i . -----t,,------,
i'
---+-------- - ----
---+-----; -----t- "-~
l--J-~
~-=- ~
~
,-----1---+---'-
I 090 !
f---= ",
M
_,
__
__'_" _"
_ .- -.,.0,10
l J0
,
e
I- 080
~,-
M' -
1 II 0 ~--+----------+- ._----~-+-- I
~ ~I
-~_
'
070 1 -. t L
"b--+
'
n
-
S
~ ..
~
060 t--"-- r -. .
--_.... -.+--_._-
o
'" 050
,
'----i------
'"
'-
-:::J 040
,
l:l
030
(;
020 ----
\'
~
'"'
~
1- 05 10
L
15
-----'~_
2-0
i
25 30 35 40 45 50
1: Lta
CHART NO. 16. Exponential model. Grading over /2: a(L; 12)~a(O; n.
II. THE THEORY OF REGIONALIZED VARIABLES 141
I 0
--
1--'
{
,--------r--:-...,--,.----r--r--r--,-,-r-----,----r---,----r~,..._I"1___rl
0-9 ..
~ .._ .. ._.._ .....1...-. ---i i- I ; -1 - --- -------~-"'~""" -------,--,,,,,,,--.
-'I, "'''-ri ----------,'
i
08
-J, ~~~ ~t ;_ :~_:;' ~ _,. ~.~
t
0,7 --------------i-------+----+----
t
"+ , +
',j;;
- __ -:--1;; . . .... ... __._
0'6
I ------_.._---+- ...__ ._+._._+_._._~ ---~
05 -----1---- --i-- .
I
"It
0-4 -
0'" 2 ' ,
03 ......... "'-r _-t , E~ ~t"'- ~.
-...--1-........
, ! 1"-
----
-- : , :
I
01
009
008 -
007
a 06 f------------ ------ ------------ --i-------'/--- ---.-- -f-'! +
005
a 04 ~-- ,L.. -- ~i!----+---- -----.- --i-----~--_i_-~---~-"---------------------------------- ---- --;-------- t------! --- -+,-,- --"It--"
/
/
0- 03 J._
/
i i
a2 03 04 06 08 I 2 3 4 5 6 7 8 9 10
I/o
WWW.minebook.blogfa.com
142 MINING GEOSTATISTICS
10 ,-----------;--~-_.,___.~~__,__._______"_____,______,....__.____r_____,_______._______r_______r..,.....__,..,.__....,.......,,,......____.__r.>""l
9 . . --------
S
7
6
5
4
~
<,
I
09
08 .
07
0-6
05
004
03
0-2
02 0-3 004 2 3 4 5 6 7 8 9 II
WWW.minebook.blogfa.com
II. THE THEORY OF REGIONALIZED VARIABLES 143
~ f----~~_~---+_--+-+----+----!--++-!------7''---,,L-_____7'''II______~~__!r_+_7'q._+___:J,.~
61-------f---
51-------+---+
4 r------+-----t--_+___
31-----+---+--+
21-------f--------1f--
0-8 f-----.-"f--------Jl'-------+-+-+---++-+-+-+
0-6 r----+---+--,'--t--------.+-++---+-:fi---If--i---if--r
o 5 f----------t'--------+-+---I--- .f--f------A-----1I'------hI--+-----tf-t-.f--+--t
o 2 1-----+-------11-1'---......<-,
WWW.minebook.blogfa.com
144 MINING GEOSTATISTICS
8
7
6
5 ----
2 ---------
1-
08
0-7
O6 1-------,1-----
05
004
02
06 08 I 2 3 4 5 6 7 8 9 10
l/a
CHART NO. 20. Exponential model. Extension variance a-~.,.
WWW.minebook.blogfa.com
I
~. I
00
I i .: I ' __ : I ! " ',' . I
s-
o 090
I-
J 080
; 060 ;.
o 50 f------------+- -- -..... . ~
040
. i
....-----;------.-.-t- "... MM"'''
030
,
(;
020
---I
I
.+_~_+c_.+_
t-
~.
~
~ 0/1
1- CHART NO. 21. Spherical model. Regularized semi-variogram. First point, y/(l);
sill )'1 (00).
1:
I
~.
s- 07 J 1 I ,
,
o
oil: 1- 3
I-
J
n
06 I
0-511-----
m __ ---+-- j
~ 1 ~-:-~ 0-9
S
~
~
04r I'
___
5=1 R I :: 05
031 I
___ I I 0-4
0-21 I ';/",/:;>/ J I 03
J- I I I 02
'"' 05 10 15 2-0 25
V hit
1:
- -
I
~.
09
/'
-: 7 / / v / /
"/ / /
....--
v /
./ ./
"/
//
./ ./
.//
~
,/
-"'"
V
.---
s- 08
/ / / 1/ / -:
~/ I / / / /"
/ / 0/1 =2-0
o
I- J.7 / / I / / /
/
/
18
H3 ./"
./ ..-/
J 0-7
, A / I I I / /v / 1-4/ /"
';n ,31//// / / / V V
;
--c::
~
0-6
0-5
"oil'" r!}///IJ j/ -: /
v
/'
V
~
0-4
If,IV/li' / V V
0-3
II//;W /20/V 15/
\ 0-2
W//J'// /
~.
~
~~V
01
'" V
V
1 2 3 4 5 6 7
h/t
8 9 10 II 12 13
SUMMARY
The structural analysis of a regionalized phenomenon consists in constructing a
model of a variogram which characterizes, in an operational way, the main features
of the regionalization. This modelling requires good physical knowledge of the
phenomenon under study as well as good "craft" in the practice of fitting geo-
statistical models.
Section lILA introduces the concept of nested structures to model the succession
of various scales of variabilities of a natural phenomenon; e.g., from the petro-
graphic scale to the macro-scale of the distribution of deposits within a metalli-
ferous province. The micro-variabilities, i.e., the variabilities of which scale is
overpassed by the available grid of information, are represented on the variogram
by an apparent discontinuity at the origin called the "nugget effect". '
Section III.B gives the various models of regionalization and coregionalization in
current use and shows how they should be combined to provide an operational
representation of structural features, such as nested structures, anisotropies, quasi-
stationarity. Then, the study of the statistical inference of a moment of order two
gives an order of magnitude of the reliability of an experimental variograrn or
covariance.
Section III.C shows how a structural analysis should be guided in practice. A
critical review of the available data, along with elementary statistics of these data,
should be made before proceeding to the effective geostatistical analysis. Three
main steps can be distinguished in this analysis, successively: the construction of
experimental variograms, their interpretation and the construction of a synthetic
structural model.
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 149
WWW.minebook.blogfa.com
150 MINING GEOSTATISTICS
--- - -- - --
1--- ---- - - Yo+Jj +Y2
F-~ -- 70+ 71
--/ 71
Co
I
/ I
Yo 72
V
-=='------
-- -- --
-----~-~
1 2 h
(10 m) (200 m)
VNNI.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 151
The operator 'l acts identically on each of the components Yi of the nested
model y.
The same holds true for the covariance; the operator 'itt operates iden-
tically on each of the components C, of the nested model C:
g(C) = I gee) if C(h) = L C(h). (111.3)
j
WWW.minebook.blogfa.com
152 MINING GEOSTATISTICS
at the origin of amplitude Co, which is called the" nugget constant", d. Fig.
liLt.
Definition
The term "nugget effect" can be generalized from its specific meaning in
gold-bearing deposits. It is used to characterize the residual influence of all
variabilities which have ranges (ao) much smaller than the available dis-
tances of observation (h ao). A nugget effect will appear on the vario-
gram (or covariance) as an apparent discontinuity at the origin.
y ,----------------,
y(h)=Co=C(Ol
c(Ql - ---------
C( h)= 0
o .. - -------------'
h
FIG. lIL2. Pure nugget effect.
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 153
In the course of this book, a pure nugget effect will often be represented
by the notation with square brackets: yo(h)=[Co(O)-Co(h)], so as to
distinguish it from a structure 'Y1 (h) with a greater range.
WWW.minebook.blogfa.com
154 MINING GEOSTATISHCS
It was shown previously that the main geostatistical operators are linear
functions of the covariance C. If 'l: is one of these operators, then, from
(III.3), we have
k
g'(C)= ~(Co)+ I ~(C;).
i= 1
- A A
Co(v, V) = V and )lo(v, V) = Co(O)- V' (III.6b)
n
(iii) v V = 0, v and V are disjoint and the distance between them is
much greater than the range ao:
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS IS5
and, thus,
-
CoCv, 1
V) =-
v
f A
-dx
v V
=A
-.
V
If not, i.e., if there exist points XoE v such that the sphere Bao(xo), centred
on Xo and with radius ao, is not entirely contained within V, then, for these
points xo, fv Co(x - y) dy ~ A. But, as regards the second integration, Iv dx,
the influence of these points Xo is negligible (since ao is very small
compared with the dimensions of v). Thus, the previous expression holds
true: Co(v, V) = AI v.
WWW.minebook.blogfa.com
156 MINING GEOSTATISTICS
o if h =0,
YOlJ(h)=
A
v -
[1- Meas v
v
n VhJ If. Ihl < v, t.e..,
. if v n Vh ,t 0 ,
(I1I.7)
A
if Ihl > v,
v
with all the dimensions of observation much greater than the range ao, i.e.,
v, Meas v n V, and: h ao
Generally, the supports v and Vh do not overlap, so that the only
discontinuity at the origin is of amplitude A/v, and the a priori variance of
the regularized RF Z, (x) is given by expression (11.45) as
Vv ao.
In practice, point data are seldom available and the nugget effect will be
revealed by the regularized nugget constant Cov = A/ 0, when 0 is the
smallest support of information (e.g., piece of core). For all other supports
V, the regularized nugget constant Cov is given by the inverse pro-
portionality relation:
v
COy = Cov V' with v, V ao (III. 8)
if V contains v,
VNNI.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 157
VV no, u ao,
(III.ll)
t Strictly speaking, the standard formula a.2 J n of the statistics of independent variables
corresponds to the estimation variance of the expectation E{Z(x)} of the RF Z(x}.
However, under the hypothesis of stationarity and ergodicity, the mean grade Zv~'
( I J v>S vZ (x) dx of a fixed domain V, with V (10' tends towards the expecta lion E {Z(x) } .
WWW.minebook.blogfa.com
158 MINING GEOSTA TISTICS
..
.----------~,".~- -----------,
h) r, (
60
>-..
10 50 60
h
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 159
These experimental variances are greater than the above nugget constants,
the difference corresponding to the additional variability and nesting effect
of the macro-structure observed at large distances (h > 50 m).
On the scale of observation, hE [5,50 m], which is also the scale on
which production is carried out, this pure nugget effect indicates that
selective mining is not possible in this deposit, unless the size of the
selection unit is of the order of the vein (5 to 40 em). In fact, the absence of
spatial correlation makes impossible any local differentiation by estima-
tion: at any location in the deposit, the best local estimator of the cumula-
tive quartz thickness is the global mean, rn = 5 em m -1 in the direction
studied (approximately vertical).
This example is one of the rare examples of deposits in which the main
variable exhibits no spatial auto-correlation at the scale on which the study
is made. There is, however, a macro-structure which appears at greater
distances.
WWW.minebook.blogfa.com
160 MINING GEOSTATISTICS
100 m
15 to 20 m
(a )
( b) (c)
-N- S
/
,,-
,,- ./
2 I N-S io /
N "/
E
I
I _
" I I
05
/
/.
I I ./
/ /
1/
II Co
Co
a 100 500 0 10 50 100
h (m)
FIG. lIlA. (a) Sampling a bauxite thickness. (b) Nested structures at Mazaugues On
a hectometric scale. (c) Nested structures at Mazaugues on a decarnetric scale.
Thus, what is seen on the hectometric scale as a nugget effect is, in fact,
the micro-structure of the jagged outline of the karstic foot wall. The total
variability of the bauxite thickness, up to h = 500 m in the N-S direction,
can, thus, be represented as a nested model which is the sum of two
transition models (spherical, for example) with ranges an = 20 m and a I =
250m.
This nested succession of variabilities at different scales is a very
frequent phenomenon in mining geostatistics. The best example of it is the
nested structures of the ferriferous basin of Lorraine (France) studied by J.
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 161
WWW.minebook.blogfa.com
162 MINING GEOSTATISTICS
Two properties
By means of the two following properties of positive definite functions, a
vast family of theoretical covariance or covariogram models can be defined
in terms of basic positive definite schemes.
(i) Every linear combination of covariances with positive coefficients is
a covariance. The same holds true for variograms:
C{h)= L A7C(h),
i:
n1 2 1V the weights Ai. (111.12)
y(h)= L: Ail'i(h),
i= 1
Z (x ) = Il" Y 1(x )
1=1
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 163
Remark For the moment, only isotropic models will be considered, i.e.,
RF's which have the same spatial variability in all directions of space. Thus,
in the three-dimensional space, the isotropic notation y(h)= y(r) denotes
y(.J(h~+h;+h~)),
where (h u , h", hw ) are the three coordinates of the vector h.
WWW.minebook.blogfa.com
164 MINING GEOSTATISTICS
b) Exponential model:
y(r)= 1-exp (-ria). (III. 16)
Note that the spherical model effectively reaches its sill for a finite distance
r = a = range, while the exponential model reaches its sill only asymp-
totically, d. Fig. IlLS. However, because of experimental fluctuations of
the variogram, no distinction will be made in practice between an effective
and an asymptotic sill. For the exponential model, the practical range a I
I u0-F========*?~====3?~=::::::::~;;=:::::::::=======4
095-
I Ex ponenucl
I
I
I
I
I
I
I
o 2/3 IT 2 3 r/o
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 165
Parabolic behaviour at the origin This very regular behaviour at the origin
is seldom found in mining practice. In the absence of nugget effect, it
corresponds to a RF Z (x), which is mean-square differentiable, and, thus,
to a ReV z(x), which is very continuous in its spatial variability. Thickness
variograms of continuous sedimentary beds sometimes display such a
parabolic behaviour at the origin, but are usually accompanied by a slight
nugget effect due to errors of measure.
(c) Gaussian model:
(Ill. 17)
The sill is reached asymptotically and a practical range can be considered
with a' = a.J3, for which yea') = 095 = 1, d. Fig. IlLS.
A parabolic drift effect (d. section ILA.3, "Behaviour at infinity" and
section II.A.5, Fig. 11.8) does not stabilize around a sill for large distances
and, thus, cannot be confused with the parabolic behaviour at the origin of
a Gaussian model.
However, for small distances (r < 2aj3), it makes little difference
whether a very regular local variation is interpreted as a drift effect or as a
stationary Gaussian structure.
Remark 1 For small distances (h ~ 0), the linear model can be fitted to
any model that has a linear behaviour at the origin (e.g., spherical and
exponential models).
WWW.minebook.blogfa.com
166 MINING GEOSTATISTICS
o r
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 167
WWW.minebook.blogfa.com
168 MINING GEOSTATISTICS
Within the interval r E [I, 13i] the relative error involved in fitting the
nested model "If (r) to the regularized model 1'1 is less than 3 %, 11'1-
"If 1/1'1 < O 03. In practice, this can be considered as a very satisfactory fit.
/
2
x
I I
3 5 7 9 II 13
rll
3. Hole-effect models
A semi-variograrn y(h) is said to display a hole effect when its growth is not
monotonic, The hole effect can appear on models with or without sills, d.
Fig. III.8(a) and (b).
The following hole-effect model, positive definite in the three-dimen-
sional space, is given:
C(r)=sinrlr and y(r)=l-sinr/r. (III.21 )
r being expressed in radians
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 169
This model has a sill and parabolic behaviour at the origin, cf. Fig. III.8(a):
y(r)=r 2j6 when r~a.
Amplitude of the hole effect Let a be the relative amplitude of the hole
effect, i.e., the minimum value of the covariance divided by the sill value
C(O):
, . . - - - - - - - - - - - - - - - - - - - - - - - , r(r) r--------.....,
(o) (b)
.:
C(r]
4". r
r
FIG. III.S. Hole-effect model. (a) With sill; (b) without sill.
WWW.minebook.blogfa.com
170 MINING GEOSTATISTICS
Note that the model (111.22) is periodic without any dampening of the
oscillations. In practice, however, such a hole-effect model is always asso-
ciated, in a nested manner, with other models with or without sills and the
considerable oscillations of the cosine will, thus, be dampened, d. Case
Study 6 in section IV.B.
z{ x)
t ( b) I z( x) +
I
;
1
Rich "'tI'''i+f-I'tH-l/lI"ttf\t-+-tIf-'''i
40{.,\>~44>-,,: I __
. - - - - _ . ._----
x x
Tin' R,c" ~rI.-h-)----~-----------'
{a )
/ Waste I
ol~. h h
FIG. II 1.9. Verti cal hole effect. (a) Vertical dire ction ; (b) hori zontal direction.
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 171
WWW.minebook.blogfa.com
172 MINING GEOSTATISTICS
(III. 24 )
with y~ and y~' having the same direct covariance K;(h); y~ and being yr
orthogonal except when i = j and 1=[' simultaneously. The cross-covari-
ance of Zdx) with Zdx) is then written as
n lIj n
Ckk'(h)= I [La~ja~.;]K,(h)= I b~k,Ki(h),
,=11=] ,~l
with
Vi = 1 to n.
This last expression gives the n matrices of coefficients [b~k'] and is nothing
more than the general expression of a positive semidefinite matrix.
The linear model for coregionalization is then a model where
(i) all the direct and cross-covariances arc derived from linear
combinations of n basic direct covariances {Ki (h), i = 1 to n}, i.e.,
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 173
WWW.minebook.blogfa.com
174 MINING GEOSTATISTICS
This last cross-covariance Ckk,(h) is no longer symmetric in (h, -h) as, Vi,
Ki(h -hk)r=. Kj(-h -h k) unless the lag h k is zero.
Intrinsic coregionalization
This particular example of the linear model corresponds to the case in
which all the basic RF's Yj(x) of formulae (III.23) and (111.24) have the
same direct covariance Ko(h) (or the same direct semi-variogram 'Yo(h )):
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 175
yo(h). This model thus corresponds to variables Zk(X), Zk'(X), the vari-
abilities of which are proportionally related to a unique principal cause
characterized by the model yo(h). This occurs in mining practice in the
study of the two proportional variables, thickness t(x) and accumulation
a (x) = t(x) . z (x); z (x) being the mean grade of the drill core of length t(x).
Isotropy When the function y(lhl, a, q; depends only on the modulus Ihl
of the vector h, the phenomenon is said to be "isotropic". The variability of
the three-dimensional RF Z (x"' xv, x w ) , characterized by the serni-vario-
gram y(ih I), is identical in all directions of the space and, thus,
(i) in spherical coordinates,
Va and tp ;
WWW.minebook.blogfa.com
176 MINING GEOSTATISTICS
Representation
All the theoretical models of regionalization presented in section 111.8.2
are isotropic, i.e., y(h) depends only on the modulus r = Ih I of the vector h.
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 177
1. Geometric anisotropy
A serni-variogram y(h u , h.; h w ) or a covariance C(h", h.; h w ) has a
geometric anisotropy when the anisotropy can be reduced to isotropy by a
mere linear transformation of the coordinates:
with
h~ = as-h; +a12hv +a13hw,
I - - - ;;>"T""------=~-----1
o Ihj
WWW.minebook.blogfa.com
178 MINING GEOSTATISTICS
represented by two spherical models with the same sill 1 and different
ranges aOl l and a 0l 2 Thus, since 1''''1 (h)"= y"'2(h), the phenomenon represent-
ed is anisotropic: the variability depends on the directions Cl'j and a2.
In order to make the two curves Ycq and 1''''2 in Fig. 111.10 coincide, it is
sufficient to multiply h, by the ratio of affinity QOI7./a"" or, conversely, to
multiply the distances h by the ratio a",) a",7.' Indeed, consider the two
spherical models
and
i.e.,
a
h',1 -h
- I ~.
a",
for all horizontal directions ai, and the phenomenon can thus be
considered as isotropic and characterized by a spherical model of
range a.
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 179
"'I
I
{i) lsotropy (Ii) GeomelrlC orus otropy
If, instead of transition structures of range a"" as in Fig. III. 11, the direc-
tional semi-variograms are of the linear type, 'Y",; (hi) = W",,h'i, then the
directional graphs of the inverses of the slopes at the origin WO: will be
j
(1) The first step is to rotate the coordinate axes by the angle q; so that
they become parallel to the main axes of the ellipse. The new
WWW.minebook.blogfa.com
180 MINING GEOSTATISTICS
Y1
FIG. 111.12. Correction of geometric anisotropy.
(ii) The second step is to transform the ellipse into a circle with a radius
equal to the major range of the ellipse. This is achieved by multiply-
ing the coordinate Yz by the ratio of anisotropy A > 1. The new
coordinates (y~, yD are then deduced from the coordinates (Yl'
Y2)
by the matrix expression
(iii) This last step is not necessary for a mere correction of geometric
anisotrophy.
The initial orientation of the coordinate system can be restored by
rotation through the angle -cp, and the final transformed coor-
dinates (x~, x~) are given by
The final transformed coordinates (x~, x~) can then be derived from
the initial coordinates (xu, Xl!) by the transformation matrix [AI
which is the product of the three matrices [R-..,][AJ[R~], i.e.,
(III.32)
with
[Al=[; ~J
WWW.minebook.blogfa.com~U4iCtJJo.-..~-.aU
III. STRUCTURAL ANALYSIS 181
and
2 \.")
a = cos '1'+/\ s1O-'P,
. 2 2
b =sm 'I' +A cos 'P,
c = (l - A) sin ({! cos tp,
[hh;J= fAJ[hr.;]
L hI;
and then we substitute these coordinates in the isotropic model
y'(lhll), which has a range equal to the major range of the directional
ellipse, i.e., according to formula (III .31),
anisotropic isotropic
Remark 2 Optimum sampling grid. The minor axis of the ellipse (OY2 on
Fig. III.12) is the direction in which the phenomenon varies fastest, and the
major axis the direction in which variability is slowest. Thus, the optimum
sampling grid for this phenomenon is a rectangular one, aligned in the
directions of these two main axes (OYl and OY2) with sides in the ratio A,
the smallest side corresponding to the direction 0Y2 of fastest variability.
2. Zonal anisotropy
The model of zonal anisotropy is the one most currently used in practice,
since any observed anisotropy which cannot be reduced by a simple linear
transformation of coordinates will call for this model.
Let y(h) be a nested model characterizing a variability in the three-
dimensional space y(h) = 'i.m(h), where h is a vector with coordinates (hr.;,
WWW.minebook.blogfa.com
182 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 183
Remark. The model of zonal anisotropy, built from the nest of structures
having their own anisotropy, is flexible enough to be used as a model for
almost any type of experimental anisotropy.
However, the flexibility of this model must not be misused. When
a specific direction is differentiated (e.g., the vertical direction h.; of
the previous example), this direction must correspond to a preferential
direction of the mineralization. The known geology should always be
used as a guide when fitting a mathematical model y(h) to a natural vari-
ability.
N
y(h) = I 1'i(lh i l) (111.33)
i=1
(111.34)
NNI.minebook.blogfa.com
184 MINING GEOSTATISTIC$
h3 w = a. h., + a. hv + a. h; = O.
The last component structure, Y3(h w ) , depends only on the vertical
direction (h,,,). This structure, Y3(h w ) , can be considered as isotropic with
regard to the coordinates (h3 , u , h3 ,,, , h3 , w) of the transformed vector [h 3 ] =
fA 3 ] [h]:
Computer subroutine
The subroutine FUNCTION GAM(HX, HY, HZ) gives the values of a
three-dimensional model y(h) of the general type (111.33). The component
structures 'Yi are transition models, either spherical or exponential. A
nugget effect is represented by a transition model with a zero range.
CAX(4,3)=
I a 0 ponent Yo
all al2 0 -;. for the first coordinate of the third com-
O 0 a ponent Y2
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 185
1 1 1 1 1 1
0 1 0 0 0 1
CAY(4, 3)= and CAZ(4, 3)=
a2 1 a 22 0 0 0 0
0 0 0 a a 1
FUNCTION GA~(rX,rY,HZJ
c
C
C SE~I-VARICGRA~ FU~CTIC~
C DEFINED AS TrE SUM OF EXPJ~E~TIAL CR SFrfRICAL
C ~E~TED STRUCTURES TrES= SIRLCILRES ARE ISCTRCPIC
C AFTER L!NfAP TOA~SFOPMATIO~ ( f INITIAL CC(RDINATES
C hX.Hy,Hl ~E~ CCORDINATES ARE [X,DY,Ol
C
C CCMMCf\S IGAfJ"Aj
C t\ST NU~~ER JF ELE~ENIA~Y STRU(TL~ES
( (NSfl SILLS
C A.NST) RA"IGES <0 .. EXPCP\E!\ liAl
C =0. NLGGEI EFFECT
C >0. SPHERlcn
C CAX{I\iST,~} LINEAR TRA~SF(~~ATI[~
C C/. Y ( ~J ST , ~ ) tJATRICES CF
C CAI(NST,~) :3 COGROIf\ATU
C ......................................................
C
CIMEt\SION /l(ll,C(ll,CA)l(!),CAY(ll ,Ct.l(l)
COMMC~/GA~~A/~Sr,c.A,cAX,C~Y,CAI
GAM'" o.
H=SQRT(HX*rX+rY*~Y+rlZ*rZ)
IF(H.LT.l.E-03IRETUR~
CU 1 IS=l,NST
I JS= IS
IJSl=IJS+\JST
[JS2=IJSl+NST
CX=HX*CAX( IJS)+HY*CAX( IJSll+Hl*CA}l( IJS2J
*
ov =H X*C II V( I J S I +H V*C A'(( 1 J S 1 ) +HZ C A V ( I J S 2 )
Cl=HX*CAZ( IJSI+HV*CAZ( IJS 1 )+Hl*CAZ (1 .. 52)
H=$QRT (C)(*Cx'+-CY*CY+DZ*CZ)
IF!A{lS) 110.12,11
10 Gt\M=G/lt-l+(IS~*(I.-ExP( .. JtO{I5HI
GC Tel
11 IF(H.GE.A(ISllGO r c 12
G /J ~= Gll ~ + c( 1 S ) * I 1 '; "" H I A ( IS)-
1 C. 5* r *H'1' 1--/ ( A( IS) * A( IS) *A ( IS J) I
GO TO 1
12 G" M.::; G A,.. + C( I S J
1 Cf'NT I f'.UE
RETLJRI\
END
WWW.minebook.blogfa.com
186 MINING GEOSTATISTICS
(ii) the semi-variogram y(x, x') or the covariance C(x, x') depends on
the two locations x and x' of the R V's Z (x) and Z (x '), i.e., on the six
coordinates (x"' x"' x.,) and (x;" x ;" x~),
~E{[Z(x) - Z(x')f} = y(x, x').
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 187
Proportional' effec t
In practice of mining applications, it has been observed that experimental
serni-variograms y(h, xo), -y(h, xo), ... , defined on different neighbour-
hoods V{xo), V(xo), . . . , can be made to coincide by dividing each one of
them by a function of the corresponding experimental mean m *(xo),
m*(xo), of the data available in each neighbourhood, d. Fig. 111.15:
This amounts to assume that there exists a stationary model yo(h) indepen-
dent of the neighbourhood location Xu and such that:
The two quasi-stationary models y(h, xo) and y(h, xb) are said to differ from
each other by a proportional effect.
___- - - - - - - - - - y(h,xo)
---------~___- - - - - - - - - - y ( h , x O )
WWW.minebook.blogfa.com
188 MINING GEOSTATISTICS
0 z a m
r r
r(h,xo) r(h,xol
m*(xol>m*(x~l m*(xom*(xo)
r(h,xo) r(h,xo)
(c) h
FIG. 111.16. (a) Direct proportional effect. (b) Inverse proportional effect.
(c) Proportional effect on the nugget constant only.
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 189
and:
g[ y(h, xu)] = f[m*(xo)] . g[ yo{h)]. (III.36)
It is enough to work on the stationary model yo(h) and multiply the result
\E[ ')Io(h)] by the corresponding proportional effect factor f[ m * (x 0)].
With a direct effect in m*2, the relative semi-variograrn, i.e., the local
serni-variogram divided by the square of the local mean, appears as a
stationary function independent of the localization Xo:
WWW.minebook.blogfa.com
190 MINING GEOSTATISTICS
Note the similarity of this expression with the one quoted in (11.21), m *(xo)
being replaced by the estimate zf itself.
the structure )'1 (h) being stationary (independent of the location x).
A good practical way to reveal a proportional effect is to draw the
relative serni-variogram (when a direct effect is suspected) on trans-
parencies: the visual comparison thus shows whether or not the pro-
portional effect affects the micro-structures (e.g., nugget effect) and the
macro-structures equally.
In fact, the local serni-variogram y(h, xo) calculated over the neigh-
bourhood V (xo) is relative to the RF Z (x) conditioned by the set of
available experimental data on V(xo) and not to the RF itself. This local
conditioning effect on V(xo) can be represented by a new RF
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 191
This does not hold true for any other definition of the conditioning
variable m *(xu).
(ii) When the RF Z(x) has a lognormal spatial law (i.e., Y(x)= log Z(x)
has a Gaussian spatial law) and if it has been conditioned by the
kriging BLUE estimator m;:"(xo) of the mean grade Yv(xo) of Y(x)
on V(xo), then the conditioning is expressed on the local semi-
variogram 1"z (h, xo) of the conditioned RF {Z (x)/ m 't-(xo)} as a pro-
portional effect in m t (xo):
2
WWW.minebook.blogfa.com
192 MINING GEOSTATISTICS
(111.39)
where V' represents the intersection of the domain V and its translation by
n
the vector h ; in other words, x E V' = V V h is equivalent to x, x + h E V.
The experimental variograrn 2y*(h) and the local variogram 2y(VJCh)
are both relative to the particular realization z (x) of the RF Z (x); these
two variograrns thus appear as random variables, the expectation of which
is precisely the theoretical variogram of the RF Z (x):
2y(h) = E{[Z (x + h) - Z {x )]2} = E {2y*(h)} = E{2Y{v)(h)}.
(111.41)
In practice, only the experimental variogram 2y*(h) is available; this
gives rise to the first deviation [Y(vlh) - y*(h)], which is the difference
between the experimental value and the local value, due to the fact that
only a finite number N ' of data pairs, [z(xi+h),z(x;)] separated by a
distance h, is available. This deviation is characterized by the variance
E {[Y( VJ(h) - y* (h) ]2} called the" variance of esti rnalion" of the local semi-
variogram.
However, even if the domain V, and, thus, the local semi-variogram
"Y(v)(h) were perfectly known, there would still be a second deviation
between this local expression and the theoretical serni-variogram y(h).
This deviation is called a "fluctuation" and is characterized by the variance
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 193
E {[ y(h)- y(v)(h )]2} called the "fluctuation variance" of the local semi-
variograrn.
In fact, since the ReV z (x) is the only available reality (the RF Z (x)
being a mere probabilistic model), only the local semi-variograrn y(v)(h)
and its estimator y*(h) have any physical meaning; consequently, only the
estimation variance of the local semi-variogram is of practical interest. The
fluctuation variance merely gives an order of magnitude of the deviation
which can be tolerated between the local semi-variogram y(v)(h) and the
fitted model y(h) which is chosen to be as simple as possible.
L
for h <-
3
L
prohibitive value for h >-
2
with A = 4/3 for the linear model y(h) = Ih I.
WWW.minebook.blogfa.com
194 MINING GEOSTATISTICS
For the linear model and for h > L/2, the relative fluctuation variance
E{[y(h)-Y(Llh)]2}/fy(h)]2 is about 2 (i.e., a prohibitive value of 200%).
Note that even if the estimation variance vanishes (very large number N'
of pairs), the fluctuation variance allows the choice of any possible
behaviour of the theoretical model y(h) for distance Ihl > L/2 greater than
the distance of reliability which is half the dimension of the field L con-
sidered.
Note again that this considerable value of the fluctuation variance pro-
hibits the construction of goodness-of-fit tests for the fitting of a theoretical
model y(h) to an experimental semi-variograrn y*(h): such tests would
almost never invalidate the fit.
Practical rule
Although there is no reason for the RF Z (x) representing the ReV z (x) to
follow a Gaussian spatial law , the preceding expressions for the estimation
and fluctuation variances provide the following orders of magnitude:
2 1 N'
C*(h)+m = N' j~l [Z(Xi +h). z(x;)].
On the other hand, to get an estimator of the centred covariance C (h), the
expectation m must also be estimated from the same set of data {z (x;)}, and
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 195
and, thus,
1
= C(h)- N2~ ~ C(Xi - X j').
J J
The estimator C*(h) of formula (111.43) then shows a bias which is nothing
more than the mean value of the theoretical covariance C(h) when the two
extremities of the vector h independently describe the set of the N avail-
able data:
1 N N
E{C(h)-C*(h)}= N 2 j~l j"~1 C(Xj-x/). (III.44)
This bias can be assimilated to the mean value C( V, V) when the N data
are uniformly distributed on the field V. When the covariance model C(h)
is of transition type, i.e., C(h) = 0 when the distance Ih I is greater than a
range a, then the bias C(V, V) vanishes if the dimensions of the field V are
much greater than the assumed range a of the covariance model.
This bias is linked to the simultaneous estimation, from the same set of
data, of the expectation m and the non-centred covariance E{Z(x +
h) . Z (x)}. This difficulty does not appear in the estimation of the uariogram,
since the expression
E{[Z(x + h) - Z(x )]2}
does not depend on the unknown expectation m.
WWW.minebook.blogfa.com
196 MINING GEOSTATISTICS
knowledge gained from the studied phenomena, and also involves a certain
amount of "craft" in the choice and use of the various geostatistical tools.
While skill can only be gained by experience, structural analysis has certain
general attributes. In particular, it should:
(i) be adapted to the proposed aim of the study. Structural analysis is
most often a prelude to an intended estimation and, as such, should
be limited to the purpose of the estimation. Thus, it is no use to
specify structures on a petrographic scale if it is intended to estimate
blocks on a hectometric scale.
(ii) take critical account of all qualitative information. The experimental
knowledge of a geologist may lead him to give all importance to a
data file; conversely, and more rarely, an erroneous geological
hypothesis may bias the use of such a file.
Before beginning the geostatistical treatment of a data set, it is advisable
to become familiar with both the physical nature of the phenomenon under
study (geology, mining technology) and the available data. This pre-
liminary phase is essential in any structural analysis, and also for a correct
formulation of the study; very often, it even defines the framework within
which geostatistical methods will be applied.
The next step is the "variography" phase, i.e., the construction, analysis
and fitting of the variogram models. The purpose of the basic geostatistical
tool- the variogram - is to condense the main structural featurest of the
regionalized phenomenon into an operational form. Variography makes as
much use of the previously introduced geostatistical concepts as it does of
the gained (qualitative or not) knowledge of the physical nature of the
phenomenon (metallogeny, structural geology, tectonics, mineralogy,
mining and treatment of are, etc.).
t Most readers have seen feasibility studies in which excellent studies of structural geology or
geomorphology are presented, only to be completely ignored in the estimation phase when
methods such as polygons of influence are used. If estimation is divorced from geology, one
might rightly wonder about the purpose of the geological study. Obviously, geological
information must playa part in the estimation and the variograrn is used in geostatistics to
interpret the structural information to be used in the estimation procedures.
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 197
WWW.minebook.blogfa.com
198 MINING GEOSTATISTICS
Through time An original sampling campaign may have been carried out
by different companies with different objectives and different equipment.
In space The same are body may have been sampled in different places
by very different methods, dragline samples on the surface, drill cores in
depth, samples taken along a drive, etc. Topography, rock type and
equipment used may vary from place to place and can lead to different
recovery rates and to biases (loss of fines in blasting holes for example).
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 199
1. Additivity of the variables The ReV must be such that all linear
combinations of its values retain the same meaning. If f(x) is a grade value,
:LA!(x;) and especially the arithmetic mean (l/nYLf(x;) must have the
same meaning as a grade.
Thus, a volumetric grade (in kg m", for example) defined on a constant
support is an additive regionalized variable. The same volumetric grade
defined on variable supports is no longer an additive ReV and, in fact, the
mean grade of two different supports is not the arithmetic mean of their
grades. In such a case, the additive ReV to be considered should be the
accumulation (or quantity of metal in kg), i.e., the product of the volu-
metric grade (kg m:') and the support volume (m'). However, when study-
ing the spatial variability of this accumulation, it would not be possible to
NNI.minebook.blogfa.com
200 MINING GEOSTATISTICS
distinguish between the variability due to the grade and that due to the
variable supports.
If the preceding volumetric grade z (x) is associated with a granulometric
fraction r(x) and both analyses are defined on variable supports v (x)
centred on point x, then the additive ReV accumulation or quantity of
metal is defined as
a (x) = r(x) . z (x) . v (x).
Other examples of non-additive variables are permeability, direction of
a vector a (x) with origin at point x, indices of all sorts (hardness, colour),
ratios of grades, certain metallurgical recoveries, etc. The particular case of
the non-additive variable Y (x) = log Z (x), logarithm of a grade, for
example, is discussed later.
2. Adequateness of the variables chosen for the purpose of the study The
chosen ReV's should be such as to encompass the main characteristics of
the problem at hand. They should also lead to operational solutions
without any superfluous precision or calculation.
Consider the example of a thin subhorizontal sedimentary bed with
obvious foot and hanging walls. Rich and poor zones will be selected by
mining on a horizontal plane, but no vertical selection is to be considered
(the entire bed thickness will be recovered). Since there is no vertical
selection, it is of no use to study the vertical regionalization of the grades,
at least for the strict purpose of ore reserve estimations. The chosen
additive ReV's for this estimation are:
(i) the thickness l(x) of the bed measured along the bore- hole located at
the point x of the horizontal two-dimensional space;
(ii) the vertical accumulation a(x)= t(x)xz(x), defined as the product
of the thickness and the corresponding mean grade z (x) sampled on
the mineralized section (of length t(x of the borehole.
Actually, the thickness variable is used in the estimation of the volumes
of recoverable ore, while the accumulation variable is used in the estima-
tion of the quantity of recoverable metaL The recovered mean grade within
a zone is given by the ratio of the quantity of metal to the recovered
tonnage in this zone.
In this example, it can be seen that the following are true.
(i) It is of no use to consider the three-dimensional regionalization of
the grades.
(ii) It is of no use to cut the cores into small lengths to represent the
various facies met along a vertical. Moreover, to do so would risk
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 201
then, the required estimation variance E{[Z(xo)-e v*(x o )j2} is not readily
derived from the variance
E{Y(xo)- Y*(XO)]2}
calculated from the structural function of the transformed variable Y(x).
WWW.minebook.blogfa.com
202 MINING GEOST ATISTICS
1. The actual core samples are not of constant length but have all been
analysed, c]. Fig. II!.17
C1 C2 c~
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 203
(III.45)
WWW.minebook.blogfa.com
204 MINING GEOSTATISTICS
2. The actual sample lengths are not constant and are not all analysed
This case is more often associated with a preferential analysis (e.g., of rich
facies only) rather than with a random one (e.g., one sample on two
analysed).
Random analysis is more favourable, and the non-analysed samples can
be considered as missing data. The missing datum is sometimes in rich facies
and sometimes in poor facies and, thus, does not cause any bias in the
structural analysis.
Preferential analysis is much more critical, especially when the non-
analysed facies (generally the poor ones) may actually be recovered by
mining. To reconstruct an homogeneous and unbiased data set, it is possi-
ble to act in one Or more of the following ways.
(i) To simulate the missing analyses. If the missing analyses correspond
to intersections with pure waste, a zero grade is given to the inter-
section; if the waste is, in fact, poor are which will be recovered (all
or in part) during mining, it is advisable to simulate the missing
analyses from the knowledge of the characteristics of this poor are,
deduced, for example, from systematically analysed drill cores.
(ii) To limit the calculation of the variograms to within the rich analysed
facies, thus characterizing the variability of these rich facies only.
This supposes that, during the mining stage, it will be possible to
select perfectly those facies with a negligible amount of contamina-
tion from other facies.
(iii) To calculate an inter-facies variogram by assigning a global grade
+ 1 or 0 to the various facies, according to whether they are rich or
poor. The drill core is then represented by a series of values of the
"all or nothing" type with constant support and the sequences of
facies may be characterized by the ranges of the variogram.
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 205
Case study
The mineralization of the precambrian iron deposits of West Africa
(Mauritania, Liberia, Guinea) is made up of itabirites concentrated in iron
oxides by departure of the silica and is essentially composed of two types of
ore:
(i) type cr, rich shallow hematite ore corresponding to a maximum
departure of the silica (67% Fe, and Si0 2 < 1%);
(ii) type {3, mixed poor ore (55 % Fe and SiD 2 > 5%) located at the
fringes of the deposits or deep at the contact with the clean itabirites.
Of course, when mining within each of these two main types of are,
several subdivisions will be considered according to the granulometry and
the degree of alteration (i.e., presence of goethite).
These two types Ct' and {3 can be clearly distinguished by
(i) their statistical. distributions (histograms and correlation diagrams of
the three main components, Fe, Si0 2 and AhO;> of the ore);
(ii) the structure of their spatial variability (variograms);
(iii) their spatial localization within each deposit.
WWW.minebook.blogfa.com
206 MINING GEOSTATISTICS
Figure 111.18 shows the Fe-Si0 2 correlation diagram derived from the
core samples (of constant support) available on one of the Nimba range
deposits (Guinea). This diagram clearly reveals two sets corresponding to
I I
I
t I
40 J I
, I "
2 I
30 3 ,
I I ,
II I '" I
I :2: 2 I
I I I
3 I ,
I I
, I
20 , I
j I
r I ,
1 I 2 j
I
I
I J"
10
Fe ("Ie,)
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 207
(0 ) Fe { b)
30
20
'"!'...-
:::e
>-., Fe
10
OI~(O)
005
....---.-J1 ..L.-
1 _
10 50
FlO 111.19. Experimental semi-variograrns (Nimba range). (a) a -type ore;
(b) 13 -type ore.
WWW.minebook.blogfa.com
(c)
20
18
16
14
,.('
0
12
0 '"
VI 10
Fe (0/0)
7
(b)
6
5
><'
0
4
0'"
3
'"
"! 2
I
56 58 62 66
Fe (%)
(c)
16
14
12
><' 10
~
0'"
Vl 8
2 :3 4 5 6
Al 20 3 (%)
FIG. W.20. (a) Fe-SiO z correlation diagram (F'derick). (b) Fe-Ah03 correlation
diagram (F'derick). (c) Ah03-Si02 (F'derick),
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 209
b
n 0 0 0 0011
{ (
\ n Q ? 9 0 9 n
1 1
JI (k{,a) I 1
1
I
1 1
1 I
1
b 1 I 1
'1 1
I 1
I
JrI1t I
(0 )
FIG. III.2L Data aligned and regularly spaced. (a) Rectilinear drill core;
(b) channel samples along a gallery.
WWW.minebook.blogfa.com
210 MINING GEOSTA TISTICS
The significance of the effect of this smoothing over the interval [r e(r)]
decreases as the tolerance e (r) becomes smaller with respect to the range
of the theoretical model y(r) to be estimated.
y (r) , . - - - - - - - - - - - - - - - - - - - - - - ,
I I
I ---+---r---
I
r Range
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 211
increase to the sill value) has been calculated using at least three or
four classes; thus, the tolerance e (r) should not be too large.
(iii) Ensure that each distance class [r s(r)] contains enough pairs so
that the corresponding estimator of the variogram is reliable; thus,
the tolerance s(r) should not be too small.
(iv) Detect any risk of bias due to preferential location of data.
3. Non-aligned data
This category can be reduced to one of the former two.
(a) By defining approximately rectilinear pathways passing through the
available data locations. Each of these approximate alignments is
then treated separately with a possible grouping into distance
classes, cf. Fig. 1I1.23(a). The disadvantage of this method is that all
the available data are not used, and it is difficult to program.
(b) By grouping the data into angle classes followed by distance classes,
d. Fig. III.23(b). To construct the variogram in the direction a, each
data value z (xo) is associated with every other value located within
the arc defined by [a da ]. Within this angle class, the data can be
grouped into distance classes [r 8 (r).
(0 )
(b)
FIG. 111.23. Non-aligned data. (a) pathways; (b) grouping into angle classes.
WWW.minebook.blogfa.com
212 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 213
1't and 1'~ are calculated from two different zones A and B, or from
core samples for one and channel samples with the same support for the
other, or, again, 1'X and 1'~ correspond to two different directions Q'A and
as
Suppose that, after examining the two experimental curves, it is
concluded that they are not significantly different, and it is required to
group them together to form a single mean experimental variagram
21':t+s (r). This mean variogram is calculated from all pairs of data values
separated by a distance r, whether they come from A or B, which amounts
to weighting each of the elementary variograms 21't (r) and 21'~ (r) with
the respective number of pairs used in their construction:
INk(r)yk(r)
k
1'*(r)=----- (I1I.48 )
I N~(r)
k
WWW.minebook.blogfa.com
214 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 215
WWW.minebook.blogfa.com
216 MINING GEOSTATISTICS
al
'f'3-5==-35~33- 3 3 - 3 4 - 31-35 ~37-41-41
I
35-35-35-
I I I-
I I
---35-33~
I I -
I I -41
a21 I
37-35-37---35-
1 I I I
~37-37-39-39-41
I I I I
I
37-40-42-
I I I I I
-34-36-41-
I I -
I I -34
I
37-
I -41-
I I I
-
I-33-
I -
I -
I I ---
I -
I --
I -35-
I I I -- ---
I I
-42-33~
I I
I -
I -
I -
I I I
~- --- -
I --39-
I I I -31
I I -
I --
I I I
-- --30-
I --
I -
I I -
0'1 24 41 20 84 18 121
0'2 22 425 18 82 15 109
0'3 19 5 16 119 10 173
0'4 18 65 14 113 8 154
h
a a,J2 2a 2a,J2 3a 3a,J2
N' 46 37 38 30 33 18
y(h) 41 57 83 116 116 163
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 217
Computer subroutines
/
/
/'
/
15 /
/'
/
/
/
/
'*~ 10
/
/
/
/
/
/
o 2 3
r/a
FIG. Il1.2S. Fitting a linear model to the mean isotropic serni-variograrn.
WWW.minebook.blogfacom
218 MINING GEOSTATISTICS
SUBRGLTI~E GA~Al(VRt~D.K~A)tPAS,NC,GtltV,N,ISJ
C
c ..................................
c SEMJ-VARICGR~~ AlC~G ~ ll~E
c REGULAR GRIC OF POI~TS M~~ t~~E ~ISSl~( CAlA
c
c ..... PARAt" ETERS
c VP(~C) CtT~ ARRAY
c NO NLMBER CF PCINTS CF THE LI~E
c K~AX ~~Xl~U~ Nu~aER cF CC~PUT~TI(~ LAGS
c p~S LENGT~ OF LAG
c NC(KMAXI NU~8ER OF COUPlESIl1G
C G(KMAX) SE~I-V~RIC{RA~ VALLES/LAG
C U A~ERAGE I
C V V~RIANCE I OF DAT~ )lEST
C N NUMBER I
C
C CPH C ~s
C IS.NE.l RESULTS ARE PP(~TEC 8Y SLBRC~lI~E
C
C CCp.l.MC ~S
C IGUT LINE P~INTER UNIT ~lMBE~
C TESI INFERreR BCUNCAR~ ( f EXISTI~G CAIA
C IF V~.LE.IEST MISSING OR ElI~INATEC CATLJ'"
C
C
OIME~SIO~ VR(lJ,~C(l),G(l)
CC~MC~ lCUT,TEST
c
C INITIALIZE
C
CO 1 tK~l,KMA)t
t-C (110=0
1 G e J K ) =0.
IFeVRCNC)-TEST)12,12,1:
12 "'=0
tJ=o.
\1=0.
GC rc 11
13 U=VR no
V:VR o..r.*VR(NC)
"'=1
11 ccnr H.UE
c
C COMP~TE SEMj-VARIOGRA~ A,.,C STA11~lI(S
C
,., C1=='" C-1
co 2 1=I.NOl
'vRl= \lJH l )
IFIVRl.LE.TEST)GC TO 2
NaN+l
U=U+VRI
~~,,+vF11JR.1
I1~I + 1
J~~MI~O(I+KMA~tNCI
CO 21 J~Il,JM
I(:oJ-t
tF(VPtJI.lE.TESTJGO TO 21
t\C.K).1:~C(K)+l
'w~R'Rl-~R(JJ
G(K)=GtK)+0.5*VRR*VRR
21 CONT H,UE
2 CONT INUS
c
c RESLLTS
c
IF(~.eQ.O)GO TC 3
yotV-l*U/N)/N
U=U/N
co 30 IK-l,KM.AX
30 GlIKj-G(IKl/M.AXotl,NC(IKII
3 CONT HUE
IFtlS.EQ.1IGO TO 5
c
C PRINT RESULTS If IS.NE.l
C
~~lle(ICUT,20CO)
RPITE(OUT,2001JU.V.N
DC 41 1<.=1,K/o(,6X
C-K*P,aS
41 ~RITE(IOLT,2002)K,O''''C,G
4 caNTI~UE
WWW.minebook.blogfa.com
220 MINING GEOSTATISTICS
C
2000 FORMAT(1~l,53Xt'SE~1-VA~I((R~M'/Ir ,
1~3Xt'**** ****.* '.lC>,
21(REGLl~RGRIC 1 CI~EN~I(~)')
2001 FORMAT(l~" AVERAGE = ttFIC.5,cX,'v~RIA~CE = t
I.cll.5,' NUM8ER OF DATA = 'tl5/1H ,'LAG CIST'
3,'ANCE I NC VAQIOGR~~ I')
2002 FOR~ATtl""" tlXd2,2XtF8.?,~)I,' I ',I4,2),Ell.5,
12X,' I ')
c
5 RETURI\
ff\ 0
2 - - - - - - NCO columns
0 0 0 0 0
2 0 0 0 0
I
I 0 0
I
I 0 0 0
I
I 0 0 0 0
Nli
lines
The data for subroutine GAMA2 (NLI lines and NCO columns) is input in
the form of a column vector VR(NLI*NCO). The subroutine calculates the
four experimental serni-variograms corresponding to the four main direc-
tions (the two directions of the grid and the two diagonal directions). The
subroutine also supplies the experimental mean and dispersion variance
corresponding to all the data available on the plane.
SLBRClTI~E GA~A2(VR,l\lI,~C(,K~AX,I\(,(,L,V,~,IS)
C
C 6
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 221
C
CIMEt--SION VRll),NC(I),(I"lO(4),JC(~J
COMMC~ leUT,TEST
CATA IAII 'I
c
c INITIALIZE
c
I(~M"'K"AX*4
u=o.
11=0.
Nc::O
CC 1 IK=l,KM~
"'((IK)=O
1 G(lKJ"O.
cr 10 1""1,-4
IOU 1=1
10 JO(J '=1
10(21=C
10(4)=-1
JC(!J"'O
c
C CCMPUTE SfMI-IJARIC(qAM,NE~ PCINT
C
NNI.minebook.blogfa.com
222 MINING GEOSTATISTICS
CO 2 l=l,NlI
CO 2 J=l,NCO
[J=IHLI*(J-l)
c
C ST~l1STICS
C
vlH=\lR( IJ)
IrIVRl.lF.TESrtGO TO 2
N=N+l
V=V+'-'Rl*'vRl
l;=U+\lRl
3 U1NTlf\UE
c
C NEk C(RECT!CN
C
CO 30 KC"lI:l,"
J3:JCCKDJ
13'-'= I C{ K C)
I 1.. I
J 1 ~J
(
C /liE, LAG
C
CO 31 j(:l,KMAX
11'-11+13
IF(Il.LT.l.OR.Il.GT.NLOG( TO ;C
Jl=JliJ3
IF(Jl.LT.l.CP.Jl.GT.NCCIGC TO 3C
JJ1= 11+MI*CJl-l)
IK="..KMAX*CI(C-IJ
IF(VRUJl).LE.TESTIGC rc 11
f\C(IK)=NC( {KIf!
Yf.lR=VR(lJl)-YRl
GCIK)=GC IK)+O.S*VPR*vRR
31 CONTtr...UE
30 CONTIf\UE
2 CONT II\UE
e
C flESUI.TS
C
lfIN.EC.O.GO TO ~
Y-IV-U*U/NI/N
U-U/M
DO 40 II(-l.KMfIl
.0 GIIKJ-GIIKJ/.AXOC1,NCCJKIJ
,. CONTINUE
c
e PRJNT If IS.HE.1
e
IFCIS.EC.l.GO TO ,
WIUT E CI OUT. 2000)
WRITE.(OUT.2001)U.Vt~
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 223
WRJTEfIOUT,2002JIIA,IO,ICt,4)
IKMaK'AX*3
WRITECIOUT,2003)IIA,IO1,~)
DO ~5 K-t, KMAX
IK2-K+IK"
~S WRITf(IOUT,2004JK,INCIIK)"tIK),IK=K,IK2,K~A})
C
2000 FORMATIIH1,53X,'SEMI-VARI(GRAM '/lH ,
153X, ',10)
2'(REGUlAR GRID 2 DIMENSIO.~)I)
2001 FORMATI1~ , I JVERAGE ',F1C.5,6X,'\JRIA~CE : I
I,Ell.5, NUMBER OF DATA ',15)
2002 FORMAT'l~ ,6X,'I',6(Al,2J,'Dl~ECTICN ',{Z,3X,'I'))
2003 FORMATCIH " LAG I ',6(Al,'NC 1/2 V~~10GRAMI')
2 004 FORM AT U H ,IX, 13, 2X, I I E( 14 t 2 X.sr i, ~ "1 X, I
c
.5 flETURN
END
Oistonce
tolerance---
-
- --
y ....
)l.
NNI.minebook.blogfa.com
224 MINING GEOSTATISTICS
SUBROUTINE GAMA3(VR,X,~,~C,K~AX,PAS,CP,NOI,
lAlP,O_,NC,G,O.U,V,N, IS)
c
C
C SEMI-V_RIOGR'M IN lwe [IME~SIO~S
C IRREGULAR GRID.TMERE MAY BE ~1~SlNG DA1A
C CALCULATION BY CLASS OF ANGLE 'hD DIST.~(E
C
C PARA ..ETEIliS
C VRCHOI O'TA ARRAY
C XCNDI,YCNOJ X ANO Y (VORDIN'TES CF FClhTS
C NO NUMBER OF POINTS
C KMAX H'XIHU~ HUMBER CF CCfIIPUTATICN LAGS
C PAS LENGTH Of BASIC LAG
C Op WIDTH OF DISTANCE CLASS.IF CP~O.
C OP=PAS/2. IS TAKEN
C HDI NUMBER OF DIRECTIO~5
C ALPeNOl} ANGLES DEfINING Dl~ECTIONS ehlTH
C RESPECT TO X AXIS IN OEGREE~1
C OA WIDTH OF ANGLE CLASS.IF CA=C.
C OA=45. DEGREES IS TAKEN
C NC(~HAX*NDl) NUHfER Of CO~PlES/LAG/OIPE(TION
C GCK~AX*NDl) VARIOGRAH VALUES/lAG/CI~EC110N
C OCKMAX*NDI) AVERAGE DISTANCE/lAG/DIRECTION
C U AVERAGE I
C V VARIANCE I OF DATA )TEST
C N NUMBER
C
C OPTIONS
C IS.NE.I RESULTS ARE PRINTEO:O~E P~GE CF PR[~TCUT
C PER EAC~ 5 DtRECTICN~
C
C COMHO~S
C lOUT LINE PRINTER LNIT NlMBER
C TeST INFERleR 8CUNnA~V OF EXISTI~G DATA
C IF VR.LE.TEST MISSING OR ELIMINATED CATUM
C
C CAP4ClTY:
C 10 DIRECTIONS:SANCIO),CANCICI
C
C
OIMENSION VRCll,XClJ,Y(1)
DIMENSION AlP(ll.CANtlC),S'NClOJ
DIMENSION Gtl),Otll,NCClJ
COHMON I(Ul,1EST
DATA tAl' 'I
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 225
c
c INITIALIZE
c
PI=3.1"1~92t5
Tel: CF
IFITOL.LE.O)TOl=PAS/2.
DALPH_=D,6
IFCOA.lE.O)0,6LPHA~5.
c
CO 1 KO=l,NOl
_LPHA:PIAlPC~O)/180.
CANIKCJ=COSIAlPHA)
1 SANIKC)=SINIALPHA)
THETA=PJ*OAtPHA/180.
COAzCtlSITHETA)
KMM#KflAX
C
DO 10 IKzl,Kf!4f11
NCClK)=O
OI'K I :z0.
10 G( IK }aO ..
IFIV~INCI-TEST.12,12,1~
12 N-O
u~o.
v-=o.
GO TO 11
13 H.-I
U-VRCfl,CJ
V=VAChO}.VR(NCJ
11 CONT INUE
c
c COMPUTE seM'-VARtCGP_~,NEW P~tNl
c
N01-NO-l
00 2 lal.NCl
VIUaYAl1 J
IFeVR1.LE.TESTIGO TO 2
NaN.}
lJ::aU+ VR 1
Y-V+VRl*VRl
nal +1
c
C NEW LAG
C
00 21 J- n,NC
IFIVRIJ).LE.TESTJGO TO 21
DX-X J'-JCC J)
Oy,.y (J ,-YI I)
H=SQATCOX*OX+OY*OY)
IFIH.LT.I.E-Q3JGO TO 25
K=INTIH/PAS+O.5)+1
c
WWW.minebook.blogfa.com
226 MINING GEOSTATISTICS
"lAeSCH-(~-l)PAS)
IFtK.GT.KMAX.CR.Hl.GT.TCL1EO TO 21
c
c NEW OIRECTICN
c
CO 22 KO~l,NO I
COSDzCOX*CANCKO)+OY*SANCKCII/H
IFfABSfCOSOJ.GE.COAIGO TC ;3
22 CONT INUE
GO TC 21
23 IK.K.K~AX*fKC-l)
NCU";)zNC( IK)+l
OfIK)=OlIK)+ti
VRR=VFI CJ I-VR 1
GfIKI-GfIKI+O.S*VFlP*VRP
GO TO 21
25 WRITECIOUT,20001 I,xt U .... Ol,J,XCJ},'t(JI
21 CONT I~UE
2 CONTINUE
c
c AESUlTS
c
IFlN.EQ.01GO TO 3
V.. (V-l*U/NI/N
U=U/N
CO 30 IK-=1.KJ4M
O(IK)~O(tK)/MAXOCl,NC(tKII
30 GC IK )=Gl II<. I/fUXOC I,NC( itO)
3 CONTJ~UE
c
C PRlhT RESULTS IF IS.NE.!
C
IF(IS.EQ.l)GO TO 5
IMP= It<-CI-IHS+l
IO~~FlOATlNOII/FlOAT(I~P)+C.9999
00 42 1f~al, IMP
WRITE(IOUT,20011IM
WRITECIOUT,Z002)U.V,N
101=1+10 .... ( 114-1)
I02;MINOINC].IOM.IM)
WRITElIOUT,200310AlPHA,P.S,TOl,
l( lA, ] C, I Os J 0 1, 10 2 )
WRITECIOUT,2004)(AlPIID),J[I01,tD~1
WRITECIOUT,2005)(IA,IO=I[l,ID2)
IKO: KP'AX*( ]01-1)
lKM=]KO+KMAX*CIC2-101)
00 43 K: 1 , K/'til A)(
IK1=K+IKO
IK2= K+ IKf'
43 WR]TEIIOUT,2006)K,CNCCJKJ,C(IK).GCIK),
11K=IKl,IK2,K"~X)
tt2 CONTlf\.Ue
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 227
41 CONT II\UE
C
2000 FORMAT(lH ,' OOUBLY OEFJ~EC P(JNT [AIU~,
X=',F9.4,' ~=',F9.4,' DATlJftI',I4,' X=',
114,'
2F9.4,' Y=',F9.4}
2001 FORHAT(1~1,53X,tSEHI-VARICGRAH ',3EX,' PAGE:'
1,I2/54X,' ',lCX,(IRREGlLAR GRIO',
2' 2DIMENSION5}'/1
ZOOl FORMAT(l~ " ~~ERAGE = ',FIC.5,6X,'\~PIANCE = t
i,Ell.5, NUMBER OF DATA = ',15
2003 FORMAT(l~ ,'OIRECTION TOLErANCE = ,F4.0,
l' DEGREES ,LAG = '.Fl.2,' ,OISTANCE lCLERANCE'
2,' = ',F7.2/1H ,5x,'1',SCAl,5X,' CIlHCTlGN',IZ
3,6X,'I'j
2004 FORM.T(lH ,5x.'I',5()X,F1.1,'DEGREES',6X,"')
2005 FORMATIIH " LAG 1',5(Al, 'NC DISTANCE "
1'1/2 ~ARIO I '1)
2006 FORMATI1H ,lX 3,' 1'.5(I?,tX,F1.2,1),
lEl1.5,'I'))
C
5 RETURN
END
WWW.minebook.blogfa.com
228 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 229
experimental variograms calculated from the two sets of data, rather than
to compare a single vertical drill core with a single horizontal channel
sampling profile.
Second, if the two experimental variograms are defined on different
supports (channel sample and core sample in the preceding example) it is
advisable to determine whether the difference in support is sufficient to
explain the observed difference between the variograrns.
Third, if the means of the data used to calculate the variograms are
different, then it is advisable to look for a possible proportional effect; for
example, by plotting and comparing the relative semi-variograms y* / m *2
if a direct proportional effect is suspected.
Fourth, the preliminary review of the data may immediately reveal the
causes of an observed difference between two variograms: a single aberrant
data value, when squared, is enough to completely change the behaviour of
the variogram; the two sets of data may not have been collected with the
same equipment, etc.
Variography does not create data, it simply presents it in the synthetic
and quantified form of the variogram, A significant heterogeneity dis-
played by a variogram usually has an obvious physical explanation; when
this is not the case, an artefact of measurement or calculation should be
suspected. If only a slight difference is observed between two variograms
and there is no obvious physical explanation, then it is best to ignore it - by
assuming isotropy, for instance - rather than developing a complicated
model for anisotropy that may turn out to be illusory. The rule in such a
case is to adopt the simplest model.
WWW.minebook.blogfa.com
230 MINING GEOSTATISTICS
(a) (b)
o h 0 h
FIG. 111.28. Nugget-like variograms, (a) Actual pure nugget effect; (b) apparent
nugget effect.
Operational representation
The representation of a natural variability by a point or regularized vario-
gram can never be unique. In practice, it is enough that this representation
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 231
reflects the various experimental features which have been found (experi-
mental variograms, anisotropies, a priori geological knowledge, etc.). This
representation must also be operational, i.e., simple and adapted to the
purpose of the study. In particular, certain minor features revealed by
structural analysis may be omitted from the adopted representation.
Similarly, when two geological or mineralogical causes contribute in the
same way to the observed variability at some scale, it is useless to compli-
cate the model by distinguishing between these two causes.
can sometimes be used in fitting the sill, provided that it is in fact a good
2(vjoo),
estimator of this sill (i.e., of the a priori variance D when v is the
support on which the serni-variogram is defined), ct. section Il.Cd and
section III.C.4 (the note about the experimental dispersion variance S2
WWW.minebook.blogfa.com
232 MINING GEOSTATISTICS
y* (II) .------------~-
xl
x
x x X
x
x
/'
/'
o II
Nested models
It will be recalled from III.A.l that the nested structure y(h) = Lty,.(h) is a
model particularly well suited to applications because of the linearity of
geostatistical operators. Thus, it is not strictly necessary that the different
components 1", have a distinct physical interpretation. However, two
components Yi and l'i with markedly different sills and ranges will, in
practice, have a clear physical meaning.
Automatic fitting
As far as mining geostatistics is concerned in this book, any attempt at
blind automatic fitting of parameters to experimental variograms - such as
least squares methods - should be avoided.
It should be recalled that each estimator point y* (h) of an experimental
serni-variogram is subject to an error of estimation and fluctuation which is
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 233
particular to that point; this error varies according to the separation dis-
tance h, the zone or data set from which the varicgram is calculated, etc.
The best method for weighting each estimated point y*(h) with its
"coefficient of reliability" comes from a critical review of the data and from
practical experience.
Among other things, physical knowledge of the phenomenon (geology,
for example), common sense involved in approximations and adapting
available information to the ultimate purpose of the study are all as
essential to a structural analysis as the calculated values of the experi-
mental variograms.
'#NNI.minebook.blogfa.com
234 MINING GEOSTATISTICS
For these two models, exponential y(r) and its fit y*(r), the values of the
following geostatistical operations in the one, two and three-dimensional
spaces are given in Table Il1.3:
(i) The dispersion variance of a point within a segment, square or cube
of side 1= 3a = 3, d. formulae (11.98) and (11.99) and Charts no. 4, 14 and
5,15;
(ii) The estimation variance of this segment and this square by a central
point, and of the cube by an axial core of length I, d. Charts no. 7, 17 and
10,20.
1-0 I--------------:==::;;:;;:;;;:;;,;;~~
...
0-5
7
'/
;t
'/
'/
If
0-1
o 2 3
rio
FIG. 111.30. Exponential model (~-) approximated by two nested spherical
models (- - - -).
TABLE 111.3. Robustness of the geostatistical variance with respect to the type of
model chosen
Dispersion Estimation
vanance variance
R1 R2 R3 R1 R2 R3
Exponential y 0544 073 082 042 057 018
Fit ,.*
Relative deviation
0535
0017
072
0014
081
0012
0419
0002
0571
0002
0179
0006
WWW.minebook.blogfa.com
III. STRUCTURAL ANALYSIS 235
Note that the relative deviation between two results derived from the
two models is of the order of 2% for the dispersion variances and of several
per thousand for the estimation variances. The excellent concordance of
the results is to be expected since the fitting of y*(r) to y(r)on Fig. III.30 is
remarkable, at least within the working interval r E [0, 3].
WWW.minebook.blogfa.com
IV
Case Studies of Structural
Analysis
SUMMARY
The 15 case studies given in this chapter were selected to cover the main structural
models of variograms in current use as well as the most critical problems which may
be encountered in the course of an actual structural analysis.
The first five examples of section IY.A show how to fit the various variogram
models currently used and introduced in section III.B.2.
Section IV.B gives a typical example of model fitting to a regionalization which
shows a hole effect and a proportional effect.
Section IV.C gives two case studies of coregionalization and their fit to the linear
model introduced in section I1LB.3.
The two examples in section IV.D present the two models of anisotropy intro-
duced in section 1I1.B.4, i.e., the geometric and the zonal anisotropies.
While the presentation of each of the previous ten case studies is focused on a
limited number of theoretical concepts, the three case studies of section IV.E give
three examples of structural analyses in three dimensions, where most of the
previous concepts are involved simultaneously.
The two examples of section IV.F are particularly important in mining practice,
as they show how the two geostatistical tools - estimation and dispersion variances
- can be used in a simple but operational way to control grades during production.
This study concerns a section of the Athabasca tar sands (Canada) that
were proven by 40 vertical diamond drills, the cores of which were cut into
approximately constant lengths, IE (1S to 22 tr]. The drill-holes themselves
236
WWW.minebook.blogfa.com
IV, CASE STUDIES OF STRUCTURAL ANALYSIS 237
/
(783)
10 50 100 150
hOI)
(ii) The arithmetic mean of the 7000 grade values used is m" = 88%
saturated weight.
(iii) The mean experimental dispersion variance of a core sample of
length I within the total core length Lis, s 2(lj L) = 173 % 2. This
WWW.minebook.blogfa.com
238 MINING GEOSTATISTICS
2 A A A( I)
Do(l/L)=---=- 1 ~- = Co 1 - - = 596 ( 2)
1 L I L 350
and, from formula (11.36), the variance contributed by the" structure 'Yt is
given by
where 'Y1 (h) is the underlying point model of the model 'Yl(h) regularized
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 239
by core samples of length 1. The range a ~ of the point model is thus equal
to a ~ = al -I = 34 ft, and the point sill value is given by formula (I1.51) as
C
1 = C l 1- 2a~ - 20a~j
I [ / [3] = O 97C I
1,
giving C t = 1339. Approximating the point model 1'1 (h) to a new spheri-
cal model with range a~ = 34 ft and sill value C; = 1339%2 and using
formulae (II.51) and (II.98), the following mean values y~ are obtained:
fiU, 1)= Ci -C1 =039,
I
'h(L , L)= C 1 1-
,[ 3 ai
T 1 a ~2]
+s ""2 ,
=O947C 1 = 1267.
4
Thus, the variance contributed by the structure Y1 is
Dt(l/L) = 12-67-0'39= 1228
Remark 1 As the support 1=2 ft of the data is small with respect to the
experimental range al = 36 ft, it could have been approximated by a point.
The fitted model, y(h)= Co+'Yl(h), would then be considered as a point
model and the experimental dispersion variance could be accepted as an
estimator of the theoretical dispersion variance D 2(O/ L) of a point in L.
This variance D 2(O / L) can then be calculated directly from the model
y(h), i.e.,
D 2(O/L)=y(L,L)=CO+C l [ 1- 3 T
al
+:S1 L: =18,02. at]
4
This value of D 2(OI L) is very close to the value 1824 calculated assuming
a support of length 1= 2 ft. Note that the quasipoint support approximation
greatly simplifies the calculations (see the remark on support and
regularization in section III.C.6).
NNI.minebook.blogfa.com
240 MINING GEOSTATISTICS
I
I
Co
01
o 4 20 a/ 40 60
r( m)
(0 ) (b)
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 241
WWW.minebook.blogfa.com
242 MINING GEOSTATISTICS
..a 2..-
E-W
c
'"E
I I
o 2000 4000
Ii (km)
The N-S drift can be seen directly on the data themselves; the N-S
profile of the mean values of the geopotential within zones of 5 of latitude
is shown on Fig. IVA. By contrast, the E-W profile of the mean values of
the geopotential within vertical bands of 10, shown on Fig. IV.S, appears
stationary.
(km)
50 52 5-4 56 58 6,0
Pole 90 I J I N
0 '"------- .....:...--J S
EQualor
VNNI.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 243
-
E
..
~6.....
~ 5 -4 f-- ,_~,
-
I '.-.-.
\ .-.' - - .-.- \..
\ /--., /.- -e"
/
/-
, .......\ ,.
- \..
\ _/ .-
\/
.
."
52f---
l.onqltude Eosl
WWW.minebook.blogfa.com
244 MINING GEOSTATISTICS
/'
/'
/
/
/
/
/'
10
>< '/'"
N /'
E ,/
.,.-
/
,/
100 200
r (m)
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 245
-
50m
1=725
(al
FIG. IV.7. Survey of Mehengui deposit. (a) Vertical section; (b) horizontal section.
NNI.minebook.blogfa.com
246 MINING GEOSTATISTICS
is not possible to determine visually which is the best fit, at least within the
interval of reliability of the experimental semi-variogram (r < 500 m).
The theoretical calculations using both models lead to completely
similar results, d. the following evaluation of the theoretical dispersion
variance.
(a)
10
(b)
This case study shows that the important thing in fitting variograms is not
the type of models, but rather-the agreement of the model with the various
experimental curves available, d. section III.C.6 (the paragraph on
robustness). Note the importance of the graph scale in plotting the vario-
grams.
WWW.minebook.blogfa.com
TV. CASE STUDIES OF STRUCTURAL ANALYSIS 247
WWW.minebook.blogfa.com
248 MINING GEOSTATISTICS
/~
I "
I
(
5 I
(
f
f
7
I
4
tFor proprietary reasons, all the mean grades quoted in this example have been multiplied by
a factor.
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 249
S'4 L '22m,286%" I
10
-
N
::J
0
::<:
-
0
)...
SI6 L'IT4m,I16%"
I~
[l-__-----'--- L ---'-----_ _---.-.l.._ _............J
o 2 3 4
h (rn )
7 ,-----~~~-----~~----~-~-~--__.
I
/
5 /
I
/
I
2 3 4
h (m)
WWW.minebook.blogfa.com
250 MINING GEOSTATISTIC5
can also be seen on the vertical profile of the grades of drill-hole S4 on Fig.
IV.12.
Fitting a proportional effect
The preceding nested spherical model is only representative of the mean
vertical structure over the zone sampled by the 25 drill-holes with mean
grade m * = 22%0 U. In order to represent local values on a zone with a
mean grade mt
- 22, a proportional effect should be added to the model.
%"U
o 5 10
I I I
10
~------------- - - -- - ~
------
g
~
~
15
E
L4'11-6m
-
s:
a.
<II
o
~
20
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 251
variance s; is calculated with a larger number of samples (five per metre) and
can be considered as an estimator of the a priori variance D 2 (I/ oo ) of the
grades of core samples of support 1= 20 ern. A scatter diagram of cxperi-
mental variance s1 is shown on Fig. IV.13. A parabola has been fitted to the
scatter diagram:
'im*E [13 to 3% U]
15
o
~ 10
t\l
V,
,
.:
. r..
I
WWW.minebook.blogfa.com
252 MINING GEOSTATISTICS
6'8-48
a = Imin {C (h )} I/ C (O) = =0417,
48
The maximum value of the experimental serni-variogram on Fig. IV.9 is
68 and, thus, (6'8 - s~) is the minimum absolute value of the correspond-
ing covariance. The experimental variance s~ = 48 is an estimator of the a
priori variance C(O).
As the value of 0417 is larger than the maximum amplitude of an
isotropic hole effect in three dimensions, CX ma x = 0217; the hole effect
observed on the semi-variograrn S4 will not be reproduced in the non-
vertical directions of the deposit. And, indeed, in such a subhorizontal
sedimentary deposit, the pseudo-periodicity of the alternance of rich
mineralized seams and waste has no reason to be isotropic,
As this hole effect of amplitude a = 0417 is limited to the vertical
direction, it can be modelled by positive-definite-type models in only one
dimension, in particular the cosine model, y(h) = 1 - cos h, for which the
hole-effect amplitude is a = 1, d. formula (III. 22), The following nested
model has been fitted and is shown on Fig. IV.9:
Y2(h)={Bh, VhE[O,a],
Ba, \:fh??:- a,
with
B = 9(%'" U)2 m-- I , a =0'28m,
The model ')It(h) is constructed from formula (III. 13 ) for the covariance
function exp(- uh] A). cos J.Lh, which is the product of an exponential
covariance and the cosine covariance.
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 253
IV.C. COREGIONALIZATIONS
o 50m
I I
Hanging tuff
B:SM
'4":
.' Sandstone
NNI.minebook.blogfa.com
254 MINING GEOSTA TlSTICS
E-W
N-S
Vertica 1
o 5 10 15 20
h(m)
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 255
E-W
Verti col
2 N-S
o 5 10 15 20
II (rn)
---
2
.-.- .>: .-
.-'
-- --- -- ---
,/
//
/
,/
v
'/
o 5 10 15 20
II (m )
WWW.minebook.blogfa.com
256 MINING GEOSTATISTICS
Thus, the three experimental curves of Fig. IV.I? are not consistent. One
possible explanation is that the experimental cross-serni-variogram Y;-R
has not been calculated from the same data as the direct semi-variograms
(in practice it is difficult to do otherwise, unless we consider only the few
points where the two measurements - grade and radiometry - exist simul-
taneously). Inconsistent and excessively small values of the means m; and
m ~ can then explain the excessive variability of the experimental relative
cross-serni-variogram Y:-RI m: .m ~ as observed on Fig. IV.I?
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 257
Mining is carried out using a bore and pillar method in an E-W direction;
pillars are 19 ft wide and stopes are 33 ft wide, cf. Fig. IV.19. On each
mining level, both faces of each pillar were sampled continuously by 10-it
chip samples, providing a set of data on a regular grid measuring 10 it in
the E-W direction and alternately 33 it and 19 it in the N-S direction.
Each chip sample is analysed for Pb, Zn and Ag.
100lj
1
30 m
(0 ) (b)
FIG. IV.18. Cross-sections showing the three types of ore. (a) Vertical;
(b) horizontal.
For the E-W and N-S directions, the experimental direct semi-vario-
grams for lead, zinc and silver grades are shown on Fig. IV.20(a), (b), (c), as
well as the experimental means and dispersion variances of the data used in
the calculation. The experimental variances calculated from formula
(III.49) are estimators of the dispersion variances D 2 (11L) of the grades of
a chip sample of length l within the mean length L of the drives ([' = 400 it
in the E-W direction).
WWW.minebook.blogfa.com
258 MINING GEOSTATISTICS
100
(a)
m:::;13-2Ofc, E-W
5 2 *' = 4 9 .5
N
:0- ~- - ---
n,
:::"':"_...,..:==--~::=oo:::::'._. _. _
~
a;
lo...
10
40
(b)
N
30 ,...
... - - - - - - - - - - - - - - - - -
---
E-W
N-S
N
C
~
"" ~.
~ 20 - ~'..-
..-
">... m 10.3 %
10 ..../
5 2"-:::; 209
I I
5
(c)
m=372 oz/ton
4 2
N 5 *= 2 -6 2 "" ....
~
___ .........
----~ N-S
<:
0
---- - ---.-=-=-~ _. ~-W. _ . _
"'
.....
0
lo...
3
a 10 100 200
lid (ft.)
(ii) There are obvious sills which are well approximated by the experi-
mental dispersion variances S2 and which, in all cases, are reached at
distances of about [hi = al = 60 ft.
(iii) There are obvious nugget effects.
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 259
(iv) The variabilities begin to increase above the sills for distances lhl
greater than 150 ft. Although located in a zone close to the limit of
reliability of the experimental semi-variograrns (L/2 = 200 ft), this
increase may be due to the non-stationarity at large distances
caused by the presence of three different gangues: for distances Ihl
greater than 150 ft, when a point x is in one gangue, the point x + h
will probably be in a different gangue causing the square deviation
[z(x + h) - z (x)] to become particularly large.
The experimental cross-semi-variograrns for lead-zinc, silver-lead and
silver-zinc are shown on Fig. IV.21(a), (b) and (c). According to the
preceding hypothesis of isotropy, only the E-W direction has been con-
sidered. There are no nugget effects on these cross-semi-variograms, and
there is a transition structure with a range approximating that found for the
direct serni-variograms (al = 60 ft).
Fitting a model
The simultaneous presence of the same transition structure with range
al =60 ft on all direct and cross-serni-variograms suggests that the linear
model of coregionalization (111.27) would provide a good fit. This model is
given by
Vh horizontal E [0, 150 tt].
The indices (k, k' = 1 to 3) represent, respectively, the grade in lead, zinc
and silver. The index (i = a and 1) represents the two basic structures; a
transition model yo(h) with a very small range or nugget effect, and a
transition model Yl (h) with range a I = 60 ft, respectively.
As the nugget effect is not present on the cross-semi-variograms, the
coefficients b~k" Vk - k' are all zero.
A spherical model with a unit sill and a range al = 60 ft has already been
fitted to the basic structure Yl (h), and Table IV.1 gives the two matrices of
coefficients [b~k'] for i = 0 (nugget effect) and for i = 1 (Yl structure). The
corresponding fits obtained using this model are shown by the dotted lines
on Figs IV.20 and IV.21.
[ i~ ~ ~n
Ag 0 0
gg]
1.1
39
14
[ 5_5
14.5
15
3.8
5 ]
3.8
1.8
i=O i=l
WWW.minebook.blogfa.com
260 MINING GEOSTATISTICS
Remark 1 The coefficients b~k' verify the constraints of the linear model,
i.e., both matrices of Table IV.l are positive definite, d. conditions
(III. 26).
30 r---~~----~~---------.
(a)
20
----~_:-::_=-=:_~_::""':-- - - - - --
10
(b)
I I
I
5 (c)
I
/
o 10 100 200
r(f!)
Remark 2 Note the quality of the fit achieved for small distances (h <
range al = 60 ft). This coregionalization model is thus ideally suited to the
geostatistical procedure of co-estimation of one grade (e.g., Ag) by the
three data sets (Pb, Zn and Ag), d. section V.A.5, Case Study 2.
This co-estimation will be carried out within an horizontal neighbourhood
of approximately 60 ft around the unit to be estimated.
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 261
Vk = 1 to 3.
The RF's YOk (x) represent the micro-variabilities of the grades for
distances (h < 10 ft) which are not accessible from the available data; these
micro-variabilities vary from one metal to the other and appear as different
nugget constants on the direct serni-variograms. The unique RF Y1(x)
represents a second variability common to all three metals. When the
hypothesis of a sedimentary origin of the deposit is accepted, this common
variability can be explained in terms of horizontal beds of deposition
micro-basins, the mean horizontal dimension of which is around 60 ft, ct.
Fig. IV.22.
60 It
(i) Nugget constants strictly equal to those found for the horizontal
structures: the micro-variabilities Yodx) would then be isotropic in
the three-dimensional space.
(ii) A common transition-type structure with a range a ~ = 30 ft
representing the mean vertical dimension of the micro-basins of
deposition, d. Fig. IV.22. A proportional effect is added to this
geometric anisotropy of ratio a 1/ a I = O 5, due to the fact that the
three types of gangue are not encountered in the same proportions in
each level.
WWW.minebook.blogfa.com
262 MINING GEOSTATISTICS
IV.D. ANISOTROPIES
10..---------------------------.........,
/-
I
I
-- -_./ ./
E-w
/'
-
'" 5
500 1000
h(m)
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 263
/
,... -- --"E-W
N-S
I "-
I "- .... .......... ......
-, ..... / ................ >
I
I
I
5 I
01' I
100 500 1000
h(m I
10.-----------------------------
This right shift of the hole effect along the distance axis Ih I suggests a
model of geometric anisotropy between the E-W and N-S serni-vario-
grams. When the E-W distances are multiplied by the anisotropy ratio
A = 700/200 = 3,5, the E-W and N-S maximum values of the hole effect
are made to coincide, cf. section IILBA, Fig. III. 10.
Genetic interpretation
It is generally accepted that the origin of the Lorraine ferriferous basin is
sedimentary with marine deltaic deposition of the upper toarcian, d.
WWW.minebook.blogfa.com
264 MINING GEOSTATISTICS
Marine sedimentation
Sea N-S
700 m
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 265
(a)
(h)
s2
10
NE-SW
\-
N
( c)
s
FIG. IV.27. Horizontal regionalization in the grey seam (Sancy). (a) N-S and E-W
directions; (b) NW-SE and NE-SW directions; (c) ellipse of geometric anisotropy.
WWW.minebook.blogfa.com
266 MINING GEOSTATISTICS
d'Avril, and, thus, the experimental semi-variograrn from Sancy will not
reveal the hole effect quite so clearly. However; the beginning of the hole
effect is quite obvious on the NW-SE semi-variogram which shows the
most rapid variability.
The first step in establishing a model of the geometric anisotropy at
Sancy was to approximate the sill of the various serni-variograms by the
experimental dispersion variance, 52 = 115 (%Fe)2 of all the data con-
sidered in the calculations. Next, a line representing the average slope of
the first few experimental points of each semi-variogram was extrapolated
to this sill, providing an abscissa which was transferred to the directional
graph on Fig. IV.27(c) (e.g., 260 m for the NW-SE direction). An ellipse of
geometric anisotropy can be fitted to this directional graph and shows that,
at Sancy and in the grey seam, the sedimentary units seem to be elongated
in the NE-SW direction and appreciably longer than those of the other
two mines, all facts which are corroborated by various lithological
observations.
The Salafossa zinc deposit is located in the dolomitic Alps (Italy) and
consists of an ellipsoidal mass with a main axis in the N-S direction,
dipping between 0 and 30 W. The mineralization (blend and galena)
consists of dolomitic breccia cement. Three types of ore can be clearly
distinguished during mining, each one being related to the dimension of the
elements of the host breccia, d. Fig. IV.28(a) and (b):
(i) type lore in the heart of the deposit, associated with gravelly
breccia (elements with dimensions of several centimetres);
(ii) type 2 are in the upper part of the deposit, associated with fractured
dolomitic rock (elements with dimensions of several decimetres);
(iii) type 3 ore at the base of the deposit in karstic cavities consisting of
dolomitic sand rich in sulphurous mineralized cement.
Thus, the two upper ores (types 1 and 2) are associated with altered
dolomite, within which stratification can still be distinguished, cf. Fig.
IV .28(b). The direction D perpendicular to this stratification is vertical in
the upper part of the deposit (type 2 ore) and then tends progressively
towards a slope of 30 West in the central part of the deposit (type lore).
No stratification can be distinguished in the type 3 are. The following
structural analysis by variography will identify these preferential structural
directions.
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 267
Data available
The data come from fans of drills set in an E-W vertical plane approxi-
mately perpendicular to the N-S direction of advance of the main drives,
ct. Fig. IV.28(b). The cuttings from the drills were analysed in 15-rn and
3-m lengths according to the zone. Because of this configuration, no data is
available in the N-S direction.
NJ\I.1215
Niv,1I81
-------------
200 ~O
(a )
400
FIG. IV.28. The three types of ore at Salafossa mine. (a) Vertical N-S cross-section;
(b) Vertical E-W cross-section and location of the N-S drives.
VtIWW.minebook.blogfa.com
268 MINING GEOSTATISTICS
Direction 1
10
N
)--_ _~3
30 45
h(m)
FIG. IV.29. Regionalization in the type 2 are. - - - , Experimental;
model; support, l::::: 3 m.
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 269
30
":-
<::
N
2 4 ~
0
~
>-
20
3 3
r" ...., - -
/' ..... /
/ -, ..... ' ...... -. -"..--' /
/'". Direction 2 +3
10 _I
4 2
I
/
15 30
Ihl (rn )
FIG. IV.30. Regionalization in the type lore. Support, (=: 15 m.
WWW.minebook.blogfa.com
270 MINING GEOSTATISTICS
20
....
\
\ Di rection 4
15
Direction 2+4+1
'"C
N Direction 2
~ 10
e;
>-..
4 2
o J5 30
Inl (rn)
~
- 30 0 W
'~3 m JC O ~ I (%Zn)2 , " I 5 m ; Co = 47 (% Zn) 2
01 = 21 m i C 1 ~ 4(%2n)2 01" 12 m i C 1 '" 12('%Zn)2
02 = 36 m ; C2" 8{'%Zn)2 02= 24 m i C 2 " 20[%Zn) 2
Zone E Zone W
Type 2 Type 1
~
L = 1-5m; Co =3-3(%Zo)2
~
01= 9 m ; C 1 = 15(%Zn)2
C2 = 0 (%2n)2
Zone W
Type '3
WWW.minebook.blogfa.com
IV, CASE STUDIES OF STRUCTURAL ANALYSIS 271
and
C 2 = C 2 1/ [I - Y2(l,I)]
= e 2 1 x 104 = 8.
Finally, the regularized model 1'1(h D ) in the direction D of zonality is
written as
v; 3 m,
Direction 1
10
/.
/
/
/
/
/
/
/ Direc lion 3 t 4
/
/
/
/
/
/
/'
o 5 15 30 45
h (m)
WWW.minebook.blogfa.com
272 MINING GEOST ATISTICS
C 21Y21(h D ) being a spherical model with sill Cu and range a21. This
regularized model 'YI (h D ) is shown on Fig. IV.29 and is a good fit to the
experimental semi-variogram for direction 1.
Available data
Two sampling campaigns are available:
(i) systematic horizontal channel sampling along the drives and cross-
cuts. The channel samples have been cut into 10-ft lengths and
analysed.
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 273
(ii) drills are collared in each drive and developed in fans. The directions
are variable but the cores have been analysed in constant 10-ft
lengths.
The channel sample and core sample supports can be mixed with good
approximation and statistical studies have shown that there is no significant
bias between the two data sets.
Granodiorite
Andesite E
WWW.minebook.blogfa.com
274 MINING GEOSTATISTICS
increases with the mean m" of the data from which the serni-variogram was
calculated.
All the data of the channel samples from six levels were combined to
provide the two mean serni-variograms for the N-S and E-W directions
05
--- [-w
shown on Fig. IV.36. A considerable number of data were used: 3360 for
the E-W direction and 2730 for the N-5 direction. Again, the radial drift
effect on the E-W semi-variograrn and the proportional effect on the
--
04
[- w
N-S
3 9 15 30 60 90
Ih!(m)
FIG. IV.36_ Channel-sample rnean-variograms (at six levels). E-W: N = 3360,
m = 153, ct = 0-200. N-S: 2730, m = I-51, ct = 0-128.
~_~_~'#!)NW.f!linebook.blogf
. .com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 275
WWW.minebook.blogfa.com
276 MINING GEOSTATlSTICS
0-2,.--------------
(0 )
01
_--_
- - - - . -.. ..__ ._----.-----~---------'
01 (b) ~ _... ~
... .-
-----------
/'
- --~------~--
~
- I
02 ------",.,-,., ..---
(c)
- --~
......................... --
01
./
N
OJ
U
e~ 0'2 ( d)
>-..
---
01
,.-----------------'~""--------
(e)
03
02
01
--
I I
a 3 9 15 30 60 90
Inl (rn)
WWW.minebook.blogfa.com
I
~.
s-
o
I-
J TABLE IV.2. Proportional effect observed on the nugget constant
';n
; Drills
Direction
Channel samples
~ A --., ,,'- -- --,
Horizontal - 4)
-0
+45 0
Horizontal T4 T4 All levels
Vertical N-S E-W E-W E-W N-S E-W N-S E-W
,
Fitted parabola C0 0-037 0004 0-035 0-064 0-026 0184 0-242 0118 0-19
(;
~.
~
'"
V
1-
1:
278 MINING GEOSTATISTICS
These experimental points (CS, m*) are shown graphically on Fig. IV.38;
the parabola Co = 018 [m - 0'5]2 is a remarkably good fit.
To take account of the radial drift in the E-W directions, it is necessary
to add an additional anisotropic term to the preceding isotropic model
(formula (IV.2)). The deviations from the isotropic model of the E-W
020
0-10
./a
.
001 ~
.::'-~ ----l ---"-- ---'
o 10 1-5
m(%Cu)
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 279
C3 = 2 . 2 2 2 [ 2 . 2 2
au SIn 'P cos cp +a u SIn cp SIn a + COS ffJ J
196x 10-8
01 C)2 ~'1
= . 2 2
Sin 'P cos a +
400 [sm
. 2 'P Sin
. 2
a + COS2cp]
( 10 U m ,
0'15
-
--
~ 0-10
x x_--.. )(
x ~.
005
x x~-
DOl
o 3 9 15
----------
.~-=''''';~'''------'-------_ _----L
30
....L..
60 90
L -_ _--------J
hu
FIG. IV.39. E-W horizontal deviations from the isotropic model. ., Core samples;
x, channel samples.
N s
v
u
FIG. IVAO. Ellipsoidal model for the radial E~W drift.
NNI.minebook.blogfa.com
280 MINING GEOSTATISTICS
Available data
The results from three types of sampling campaigns are available.
(i) The entire deposit was sampled by vertical diamond drilling (DDH)
on a pseudo-regular 100 m x 100 ill horizontal grid. The diamond
drill-holes were cut into constant lengths 1= 13 m and analysed for
copper. The 13-m core sample length corresponds to the height of
mining benches.
(ii) The primary production zones were sampled by rotary drilling
(RDH) on a regular 50 m x 50 m horizontal grid. These holes were
also cut into 13-m lengths.
(iii) The production benches were sampled by the same rotary drills used
for blast-holes. The drill cuts coming from these blast-holes (BH)
were approximately 14 m long. The blast-holes are located on an
irregular horizontal grid with constant density and a mean 8-m
distance between two BH. At the time this study was carried out, the
are production had just begun and the information from the BH was
very limited.
All samples were analysed for total copper (Cu T) and soluble copper
(Cu S).
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 281
The three types of samples (DDH, RDH and BH) were studied
separately. The vertical regionalization can only be studied through the
DDH, which are the only sampled long enough. On the horizontal plane
(by grading over the constant 13-m thickness of a horizontal bench), these
DDH provide data over large horizontal distances (h > 100 m.) The RDB
provide a comparable horizontal information, i.e., graded over 13 m, for
horizontal distances (h ~ 50 m). The BH provide horizontal information
graded over 14 m (14 = 13), but at smaller distances (h ~ 8 m).
-......::..
10 /
/-~-~-~-..
cu T
/
/
/
/
/
--..:::.
Cu S
WWW.minebook.blogfa.com
282 MINING GEOSTATISTICS
The two other groups of samples - RDH and BH - were used to evaluate
these small-range structures. Care was taken to choose benches and zones
for which the Cu T and Cu S grades were similar to the corresponding
DDH grades (1,1 % Cu T and 08% Cu S). This precaution cancelled the
influence of the proportional effect.
10
/"
/"
-- ~:-..;:--~--
/"/"~- Cu T
-
/"
, /"
Cu S
,, -~-----
/ -,
/ -,
, /
0-5
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 283
{Co(O)- Co(r)} represents the nugget effect which appears on the grading
over the I = 13~m support as the nugget constant COl, ct. formula (lII.7).
Cj)'1 (r) and C 2 Y2(r ) are two spherical models with respective sills C 1 and
C 2 and ranges a 1 = 45 m a2 = 200 m.
The theoretical model of the semi-variograrn graded over the constant
13-m thickness is obtained from the point model by applying the practical
rules of section II.D.4, and is given by
yo(h)= C O/+CllYI(r)+C21l'2( r ), V horizontal vector with modulus rE
[13,400 m], (IVA)
where the graded sills are written as
Cll = C 1[l- Yl(l, l)] and C 2 1 = C 2 [1 - Y2(/, I)].
The graded parameters Co/, C II, C 2 1 are fitted to the experimental
graded semi-variograms, By way of example, the fitted parameters for total
copper are
COl = 0,22, Cll = 0,10, C2 l = 064 (% ce T)2
and, thus, the parameters for the point model (IV.3) are
C j = Cllxl16=O12 and C 2=C2 I Xl03=066(%CuTl.
The graded model (IVA) thus considered is shown on Fig. IVA2 and is a
good fit to the experimental curves.
NNI.minebook.blogfa.com
284 MINING GEOSTATISTICS
1-5,-----------------------------------,
E x per rme nt cl
semi- vor ioqrorn .....
,/ ">--- The 0ret i ca I
/ <.
./
.> .:
<,
vertical model
c><> ......
-....-.
10
".
..... .....
/
./
/ "
I /
I /
I /
I I
1/
;1
;1
;'
05 I
~ Theoretical
horizontal model
o 13 26 39 52 65 78 91
s; (m)
For a vertical vector h; (Stich that h; = h., = 0), the following point
model is obtained:
(IV.6)
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 285
a3 = a3/- j = 65 -13 = 52 m,
The regularized model (IV.6) thus considered is shown on Fig. IV.43 and is
a good fit to the experimental vertical serni-variogram.
The three-dimensional point model adopted for the Cu S grades is
completely similar, with the same vertical anisotropy, the same values for
the nested ranges at, a2, a3 and the same nugget constant COl, but with
different values for the sills Cj, e 2 , C 3
Case Study 13. Structural analysis of the Mea nickel deposit, after J.
Deraisme (1976)
The nickel silicate deposit of Mea (New Caledonia) consists of altered basic
rocks. The following three components can be schematically distinguished
on the vertical, cf. Fig. IV.44.
(i) A seam of yellow laterites with low, uneconomic nickel grades.
(ii) A seam of silicate ore consisting of various facies of altered basic
rocks (dunites, harzburgites). These facies are more or less nickel
rich. Unaltered inclusions of variable sizes are found within this
silicate ore and consist of hard blocks caned pure waste, that can be
distinguished both when drilling and mining (selective crushing and
selection at the face by the miner himself).
(iii) A karstic bed-rock consisting of peridotites.
WWW.minebook.blogfa.com
286 MINING GEOSTATISTICS
t.
,80m
I
. I
I
I
____ Laterites
" '-waste
The drill cores were cut into constant l-m lengths within the silicate
seam. The pure waste was removed from the softer, altered material with a
geologist's pick and the lengths of the two separated materials were
measured and analysed for density and grades (of dry are) d. Fig. IV.45.
L en g t hl I-Ii
...r~------"7'Altered Density. Of
Grodes,Nf , Co
{
Co and impurities
1m
Pure L engt h, Ii
:J-------~waste
{ Density, a,
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 287
Core If
sample _
{ver~coi )-1--
' - - - - r G ( ! l l _ _---J
(a) (b)
FIG. IV.46. Available structural information. (a) Regularization over I; (b) grading
over 31.
Vertical regionalization
The I-m and 3-m regularized vertical serni-variograms, 'Y/(h) and 'Y31(h)
for each of the above three variables, were calculated from the 167 vertical
drill-holes and are shown on Fig. IV.47(a), (b), (c).
'#NNI.minebook.blogfa.com
288 MINING GEOSTATISTICS
( o) Support i
0-5
Support 31
I
/ --------
/"
/"
/
01
Support l
---- -----~-----
Support 3'
=
Support l
(cJ
2
Support 3t
---
/
o 1 3 10 20
r (m)
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 289
Horizontal regionalization
The three accumulation variables graded over a constant thickness of
31 = 3 m were defined by the intersections of the vertical drill-holes with
each of the 3 m thick horizontal mining benches, cf. Fig. IV .46(b). The
graded variables were defined by weighting the analyses of the I-m core
lengths by the intersection of these lengths with the bench, according to
formula (III.45).
Using these variables graded over 3/, the graded experimental semi-
variograms were calculated, bench by bench, and for various horizontal
directions. The bench variograms are not significant, due to the small
number of data pairs used in the calculations, and they were combined into
mean directional variograms for all benches. These mean directional
variograms show no significant anisotropy and, thus, were combined into a
single mean isotropic, horizontal variogram valid for all benches. These
three mean isotropic semi-variograrns, graded over 31, are shown on Fig.
IV.48(a), (b), (c).
On each of these experimental curves, there is a horizontal macro-
structure with a range a3 = 180 m and a very prominent nugget effect with
an amplitude approximately equal to the sills of variability of the vertical 31
regularized experimental serni-variograms 'Y3/(h), d. Fig. IV.47.
Thus, a three-dimensional model can be established by adding a macro-
structure with a range a, = 180 m to the structure revealed by the vertical
regionalization,
WWW.minebook.blogfa.com
290 MINING GEOSTATISTICS
0-5,-------------------------------.,
l a)
-~-~----------~---- -----
Suppor t 3/
02
01 0.-- - - - ---------_.~,,_._-----------------'
02r~lb)-~-,
--------------~
(c )
2
Support 3t
Of I I I I
10 100 200 300
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 291
To sum up, nested models of the type (IV. 7) have been adopted for the
three-dimensional point support regionalizations of the three accumulation
variables. The values of the sills, regularized on I and 3[, are given on Table
IV.3. The corresponding theoretical models, regularized by core samples
of lengths [ and 3/ and graded over 31, agree remarkably well with all the
available experimental curves.
(b) Support 3l
WWW.minebook.blogfa.com
292 MINING GEOSTATISTICS
r
1'31(h) = [CO,3i + C(l' 5-log 3)] + C log I' 'r/r';331;
thus
1'1(h)- 1'31(h) = CO,I - C O,3i + Clog 3
= 2eO,3 1 + Clog 3,
Consequently, if the logarithmic model is adopted, the experimental
serni-variograms for metal accumulation plotted on logarithmic paper
would be approximated by straight lines with slope C. The two lines
corresponding to the two supports I and 31 should differ by a translation of
modulus 2CO ,3i + Clog 3.
On Fig. IV.49, the two parameters fitted to the experimental curve
regularized on 31 are C O3i = 021 and C = 036.
Over small distances (r:s;: 25 m), there is a very good agreement between
this logarithmic model and the experimental curves. On the other hand, the
logarithmic model slightly overestimates the variabilities at large distances
- -- - Support I
-- -
2
Su pporl 31
F------,~,--~~--+-----
-
10
rlt
FIG. IV,49 Fitting a logarithmic model to the regionalization of metal accumula-
tion.
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 293
The two following case studies show how the characterization of a spatial
variability through the variograrn model allows the prediction and, thus,
the control of the production grades. These two case studies can also be
considered as examples of practical applications of the two geostatistical
variances: the estimation and the dispersion variances.
~t
To to I tannoge,~-
Day J True grade, T;
{
Esti mo ted grade, Ii*
'#NNI.minebook.blogfa.com
294 MINING GEOST ATISTICS
tonnages Pu are calculated so that the predicted mean grade T] of the total
production P, of the day j satisfies the sales or milling constraints. Thus, for
a day i. the following relations hold:
N
total tonnage produced, P, = L = 1 Pij;
CTt = L pt E{[tjj -
i
eil 2 } / Pi.
The elementary error can be written as
ti j - et = (tif - OJ)+ (8 i ---> en
and, thus,
The fluctuation (tij - Oi) of the elementary grades Iii within the block B I can
be assumed to be independent of the estimation error (01 - On of this
block, and, thus,
E{(ll; - Oi)(ei - (1)} = o.
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 295
1 r 2 2
V JE{(fij-Oi) }dx=D (PdB i ) ,
8,
As for the term E{(Oi - Or )2}, it is, by definition, the estimation variance
a;, of the mean grade of block B; If this block has been estimated by
kriging, then 0";,
is simply the kriging variance of the block given by the
kriging system of equations.
Finally, the estimation variance of the mean grade TI of the total pro-
duction PI of day j is written
2
<rEi = Py1:!- 2[ 2 / ) 2]
;::-1 Pu D (Pi B, +a c , . (IV.8)
This variance thus depends not only on the dispersion of the grades of the
production units Pi in each block B, but also on the quality of the estima-
tors of the mean grades of these blocks B;
2 1;'
aE i =~ L,
2
PiiE{(tii- 1*
ij )
2}
(IV.9)
Pi ;=1
This case corresponds, for example, to an open-pit deposit for which the
mining program is based on blast-hole analyses which provide the estima-
tor t~ of each production unit.
WWW.minebook.blogfa.com
296 MINING GEOSTATISTICS
(IV.I0)
If the daily production units Pu extracted from each block B, are more or
less equal, then
and, thus,
(IV. 11)
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 297
WWW.minebook.blogfa.com
298 MINING GEOSTATISTICS
82.,2'''10 Ni
x 1 x 1--------1 1 1 x
60
'%
V//
"/.
%% ;5 I'
8 3,1 5"/" Ni
1 1 x x ------ x x x 1
///
///
x 1 x 1 - -- ------x x x x
WWW.minebook.blogfa.com
IV_ CASE STUDIES OF STRUCTURAL ANALYSIS 299
between the two auxiliary functions P(Pi) and F(B;) defined on the two
rectangles Pi and B; cf. section I1.E.2, formula (11.82), e.g.,
D
2
(p d 8 1 ) = "1(8[, f3d ~ y(P" PI) = F(60; 60)~ F(20; 5).
2
(For the calculations of the values D (pJ B i ) , note that as the supports Pi
and B, are considerably larger than the support of the 5 m-Iorig core
samples, the influence of the nugget effect disappears and the relative
spherical model with range a = 300 m and sill elm 2 = 0-075 is sufficient.)
Immediate calculation and chart reading provide the following values:
Thus, from formula (IV.S), the estimation variance of the daily grade
production is
Thus, adopting the Gaussian standard, the mean grade T, actually re-
covered each day will, in 95% of the cases, fluctuate in the interval
[2 0-6] % Ni around the predicted value = 2 % Ni. r:
The mediocre result of this grade control is due to the rather hasty
estimation of the panel grade 0; by the grade of the central drill core
(polygon of influence method). Using the same data (drills on a regular
60 m x 60 m grid), simple kriging would, without any additional cost,
reduce the estimation variance a;;
by half, and daily grades could then be
controlled with a variance of at
= 0-0511 (% Ni)2, i.e., a standard devia-
tion of (TE, = O' 23% Ni, which represents a noted improvement.
The lower limit of the variance (T~'i corresponding to the perfect know-
ledge of the mean grades ()i of the working sites is given by formula (IV.IO):
WWW.minebook.blogfa.com
300 MINING GEOSTATISTICS
t There is no need to give the value of the nugget constant, as the nugget effect vanishes when
regularized on the previous supports sand S of the weekly and annual productions, d.
formula (IlLS).
WWW.minebook.blogfa.com
IV. CASE STUDIES OF STRUCTURAL ANALYSIS 301
720 m
,I
I 5:---
I I
5mL~L.......---_1_125m
I. S S .1
2x 180 ~ 360 m
500 rn
'I
5
180 rn
,
o I~~-----+-----
FIG. IY.52. Two methods of shovel advancement. (a) Two shovels in line;
(b) two shovels in parallel.
WWW.minebook.blogfa.com
302 MINING GEOSTATISTICS
d. the definition (II.79) of the auxiliary function atL; l). On average, over
the 50 weeks of production, the term 1'(51,52) can be written as
1
Yes!> 52) = - I a [5(2i -1); 180] = ex 084 = 0252
50 i
Remark The knowledge of the dispersion over one year may not be
sufficient to control the variablity of grades. For example, it is important to
ensure that weekly mill-head grades will be sufficiently regular over a short
period of time (e. g.. 1 month). For this purpose, the dispersion variance
2
D (5/ M) of the weekly grades within a month's production can be cal-
culated. Krige's additivity relationship (11.37) gives
2
D (sl S) = D 2 (s/ M) + D 2(M/ S).
Thus, the distribution of the total annual variability D 2 (s/ S) between the
short-term variability D\s/M) and the long-term variability D\M/ S) can
be evaluated.
If it is desired to further increase the control of variabilities, the deposit
can be simulated, i.e., the spatial regionalization of the grades can be
simulated, ct. Chapter VII. A simulation of the mining method can then be
applied to this simulated deposit to give an idea of the fluctuations of
grades from one period to another, e.g., day per day, month to month, etc.,
d. the case study given in section VII.B.3.
WWW.minebook.blogfa.com
v
Kriging and the Estimation
of in situ Resources
SUMMARY
The in situ resources of a deposit are distinct from the recoverable reserves which
can be mined within some given technical and economical context. The object of
this chapter is the estimation of the in situ resources prior to the application of any
cut-off; the estimation of recoverable reserves will be treated in Chapter VI.
Section V.A introduces the technique of kriging, which provides a minimum-
variance unbiased linear estimator of the mean characteristic of a block of any
given geometry. In the case of non-stationarity, several non-bias conditions are
required and this leads to the technique known as "universal kriging", or unbiased
kriging of order k. When a group of variables are correlated, the cokriging tech-
nique can be used to estimate any variable (e.g., silver grade) from the data
available on all the correlated variables (e.g., lead and zinc grades in addition to the
silver grades).
Section V.B deals with applying kriging techniques in practice and demonstrates
that it is possible to krige thousands of blocks with reasonable speed and cost. Some
typical subroutines are given along with numerous examples of kriging plans which
can be adapted to the estimation of a large range of deposits of variable
morphology.
Section V.C deals with the global estimation of a deposit or a large zone. The
global estimator is obtained by combining the various local kriged estimators of the
blocks or units which make up the zone. When calculating the global estimation
variance, a distinction is made between the geometrical error involved in evaluating
the limits of the mineralized surface or volume and the quality error due to the
estimation of the mean characteristics (e.g., grade) within a zone of fixed limits.
These two errors are then combined to provide the estimation variance attached to
global estimators, such as quantities of ore and metal and the corresponding mean
grades.
303
'#NNI.minebook.blogfa.com
304 MINING GEOSTATISTICS
Zv(xo)=2- f Z(x)dx
V V(xo)
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 305
E{Z~} = m LA a = m = E{Zv},
E{[Zv - Z~ f}
2
= E{Zt} - 2E{ZvZk}+ E{Zk },
with
E{ZvZ~d=LAa-
.
1
'" VV a
J dx f
V Va
E{Z(x)Z(x')}dx'=LAaC(V,va)+m,
'"
2 -
E{Z~2}=LLA"AB_l_,
(); (3 V",f-{3
f
u"
dXf E{Z{x)Z(x')}dx'
U(3
2
The terms m are eliminated and, thus, we have
NNI.minebook.blogfa.com
306 MINING GEOSTATISTICS
(The standard notation C(V, Va) designates the mean value of the covari-
ance function C(h) when the two extremities of the vector h independently
describe the domains V and Va:, respectively.)
The estimation variance can, thus, be expressed as a quadratic form in
Aa , As, and is to be minimized subject to the non-bias condition L" AD: = 1.
The optimal weights are obtained from standard Lagrangian techniques by
setting each of the n partial derivatives a[E{[Zv - Z~ ]2}_ 2M If] ,1/31/ (lA",
to zero. This procedure provides a system of (n + 1) linear equations in
(n + 1) unknowns (the n weights Aa and the Lagrange parameter J.t) which
is called the" kriging system".
tl
V" ~ 1 to n, }
(V.3)
L A13 = 1
13
and
t In practice, even when the covariance is not defined, the system (V.l) written in terms of
covariance is preferred for reasons of programming efficiency. For this purpose, it is enough
to define the "pseudo-covariance" C(h) such that YCh)=A-C(h), the constant A being
any positive value greater than the greatest mean value y used in the kriging system (V.3).
The non-bias condition I" A" = 1 allows the elimination of this constant A from the system
(V.3), which then becomes a (V. I)-type system written in terms of the pseudo-covariance
C(h).
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 307
Matrix form
Both kriging systems can be expressed in matrix form as
[K] . [A I = [M2] (VA)
from which, for system (V.I),
[A] = [Kr 1 . [M2],
and, [A r being the transposed matrix of rA],
O"k = C( V, V) - [A It. [M2],
where the matrix of unknowns fA] and the second member [M2] can be
written as two column matrices:
Al C(VI,V)
A2
An C(v", V)
-/.J- I
The first member [K], or "kriging matrix" is written as:
[K] =
~
Main diagonal
Note that the kriging matrix is symmetric, i.e.,
Va, (3.
WWW.minebook.blogfa.com
308 MINING GEOSTATISTICS
[C (Va, VI3)] is strictly positive definite and, thus, necessarily has a strictly
positive determinant. For this purpose, it is enough that the point covari-
ancet model C(h) used is positive definite and that no data support
v'" coincides completely with another one. In fact, Va E Va' entails
that ct; v(3)= ci; v(3)' '1(3, and the determinant IC(v"" Vp)1 is, thus,
zero.
This condition for the existence and the uniqueness of the solution of the
kriging system thus entails that the kriging variance (V.2) is non-negative,
d. section II.A.3, relation (II.8).
Remark 3 The expressions of the systems and the kriging variances using
the notations C and l' are completely general,
(i) whatever the supports Vet> V{3 of the data and the support V to be
estimated, some data supports may partially overlap, Vet n V{3 ~ 0,
but for {}' ::P {3 it is imperative that V a ?!= Vj3 - some data supports may
be included in the volume V to be estimated, Vet C V;
(ii) whatever the underlying structure characterized by the model C(h) or
y(h), the structure may be isotropic or anisotropic, nested or not.
Remark 4 The kriging system and the kriging variance depend only on
the structural model C(h) or y(h) and on the relative geometries of the
various supports Va, V(3, V, but not on the particular values of the data Za.
Consequently, once the data configuration is known and before drilling is
undertaken, the kriging system can be solved and the corresponding
minimum estimation variance can be forecast. Thus, the kriging variance
can be used to balance the cost of a new drilling campaign with its forecast
utility (new data would decrease the estimation variance, thus providing
narrower confidence intervals) cf. O. Bernuy and A. Journel (1977).
+If the serni-variogram is used, then -y(h) must be conditional positive definite.
VNNI.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 309
Remark 6 The kriging system and the kriging variance take into account
the four essential and intuitive facts, previously mentioned in section
II.B.2, remark 4, which condition every estimation. These are as follows.
(i) The geometry of the domain V to be estimated, expressed in the
term y( V, V) in the expression of the kriging variance Q"~.
(ii) The distances between V and the supports Va of the information,
expressed by the terms Hv w V) of the matrix tM2].
(iii) The geometry of the data configuration as expressed by the terms
y(vn, V(3) of the kriging matrix [K]. The accuracy of an estimation
depends not only on the number of data but also on their configura-
tion in relation with the main features of the regionalization as
characterized by the structural function y(h) in the various terms
Yev{O v(:J).
(iv) The main structural features of the variability of the phenomenon
under study as characterized by the semi-variogram model y(h).
Consider the estimation in two dimensions of the panel V by the sym-
metric configuration of the four data A, B, C, D as shown on Fig. Y .1. The
underlying mineralization shows a preferential direction u of continuity
WWW.minebook.blogfa.com
310 MINING GEOSTATISTICS
A
x
Dx
0 x
xB
~c
Direction of continuity
The mean values C(v"" VtI) appearing in the kriging system are written,
according to formulae (111.6), as
if Vf3 c v"',
if v'" and vl3 are disjoint,
A", [AI v", + C\(v", va)] + 2: Af3 C\ (va , VfJ)-1-t = C\(v"" V),
{3r''''
va'l (V.S)
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 311
If all supports v", V/3, V are not disjoint, then each of the terms C of the
kriging system must be calculated.
A case often found in practice is that of a configuration of data disjoint
among themselves, some of which (v c, ) being located within the support V
to be estimated:
A -
= V + Cl(V a ' , V), v,
(V.6)
I A~ = 1
a
and
In mining applications, the support V is very often much larger than the
data supports Vo;, V,,(, and the term A/ V is negligible when compared with
the terms Aio; or A/Va" When all the data supports Va, V(3 are disjoint,
system (V.S) is obtained.
If the semi-variogram l'(h) = {Co(O) - Co(h)} + 1'1 (h) is to be used instead
of the covariance C(h), it is enough to substitute (-Yt) for C t in the
preceding systems (V.S) and (V.6).
Note in system (V.S) that the influence of a nugget effect with nugget
constant COli = A/ v affects only the elements of the main diagonal of the
kriging matrix as an additional term A/ Va'
WWW.minebook.blogfa.com
312 MINING GEOSTATISTICS
I AJ3 = 1
{3
and, thus,
CT~ = A [ I 1
V{3 +V
1 J.
{3
The weights for the n available data values Zo. are proportional to the
volumes va of their respective supports.
Moreover, if the n supports Vc< are equal (e.g., n core samples of
constant length and volume v), then the system (V.?) becomes
Ao.=l/n, Va = 1 to n,
if V v. }
The n weights Ac< are all equal and the kriging variance (T~ = COl)/ n is
simply the variance of estimation given by the statistics of spatially uncor-
related variables. The nugget constant COt; is the a priori variance of the
support v data, according to formula (IlL!!):
2
COt; = A/ v = D (v/ OO ) = Var {Zv(x )}.
Thus, for a pure nugget effect, we have the following.
(i) All data having the same support v (disjoint from V) have the same
weight whatever their distances from the volume V to be estimated.
The close data no longer screen the influence of the more distant
data and the nugget effect is said "to remove the screening effect".
(ii) Within a given neighbourhood of quasi-stationarity, it is no longer
possible to make different local estimations: all estimators Z~ of all
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 313
Pone I to be
estimated
"-
Available
drill-holes
NNI.minebook.blogfa.com
314 MINING GEOSTATISTICS
m( x)
D(xO)
... ..
xo x
FIG. V.3. Drift and residual fluctuations.
The form of the drift m (x) is supposed known. For example, m (x) may
be a linear combination of any known functions {II (x), I = 1 to k}, the
coefficients a/ of which are unknown, so that the drift m(x) remains
unknown:
k
m(x)= I ad/(x). (V.9)
/= 1
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 315
WWW.minebook.blogfa.com
316 MINING GEOSTATISTICS
to estimate y(x, y) directly from the initial experimental data z (x",). This
would entail the simultaneous estimation of the drift m (x) and the semi-
variogram y(x, y) from a single realization z(x) of the RF Z(x), which is
not rigorously possible.t
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 317
Z~ = l:" AaZ".
"=1
E{Zv} = V1 f
V(xol
E{Z(x)} dx
.
= L arr:1
I V
f . fi(x) dx
V(xo)
= L alb Iv,
I
E{Z~} = L AaE{Z,,} = I
I
a, L Au ~
a V"
f~
!E(x) dx = L a,
I
L Aab~",
a
denoting the mean value on the support v of the known function fl(x) by
l .
b v, I.e.,
b Iv = -1 f fT(x)dx.
v v
This estimation variance then appears as a quadratic form in A", A/3, just
as it did in the case of kriging under a stationary hypothesis. However, in
this case the estimation variance must be minimized subject to the k
non-bias constraints, instead of the single constraint for stationary kriging.
Using the standard Lagrangian technique, a system of (n + k) linear equa-
tions in (n + k) unknowns is obtained (the n weights A", and the k Lagrange
WWW.minebook.blogfa.com
318 MINING GEOSTATISTICS
parameters J.1-1)' This system is called the "unbiased kriging system of order
k" or the "universal kriging system".
11 k
I ABC(vm Vi~ L /-Llb~" = ('(v"" V), Va = 1 to n,
13 ~ 1 1= I
(V.12)
11
I At3b~a=blv, VI = 1 to k.
B= 1
A(.c~(v(>, V).
I
ak=E{[Zv-Zkf}=c(v, V)+ L J-tlb V - I (V.U)
minimized 1=1 <>: ~ 1
I
t3
Ai>Y(v,., V(3)+LJ.1-lb~a =Y(v,., V),
1
\fa ~ 1 to n, }
(V.14 )
I Af3b~13 = b 'v, Vi = 1 to k,
(3
and
O'~ = I A",')i(Va , V) + L /-LIb ~ - ii( V, V).
a 1
Remark 1 The kriging system (V. 1) in the stationary case can be seen as
an unbiased kriging of order k = 1, the stationary expectation being
reduced to an unknown constant aj:
Note that the mean value of the constant function I, (x ) = lover any
volume v is equal to 1: b~ = 1, Vv.
Remark 2 Existence and uniqueness of the solution. As was the case under
the stationary hypothesis, cf. section V.A.I, remark. 1, the covariance
matrix [C(v a , VB)] is assumed to be strictly positive definite. It can then be
shown that the system (V.12) of unbiased kriging of order k has a unique
solution if and only if the k functions fl(x) are linearly independent on the
set of the n data, i.e., if the n following linear relations
k
L clb~" = 0, Va = 1 to n, (V.IS)
1= 1
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 319
imply that ci = 0, V1= 1 to k. For example, consider a data set in which all
the data are aligned perpendicularly to the direction u, as on Fig. V.4. It is
obviously impossible to take a drift in the direction u into account, using
just this set of data, even if this drift is only linear: m (u) = a 1+ a2U.
Remark 3 Matrix form. The first member matrix [K"J of the system
(V.12) of unbiased kriging of order k can be obtained by adding (k -1)
rows and (k - 1) columns to the first member matrix [K] of the system (V. 1)
for kriging under the stationary hypothesis, i.e.,
b~" 0
::
~o~ .. 0: (k -1)
" . rows
b~, b~n 0 ~
0 .. . 0
(n + 1) columns (k -1) columns Main diagonal
Note that the matrix [Ks ] is still symmetric.
WWW.minebook.blogfa.com
320 MINING GEOSTATISTICS
The column matrices of the unknown [.A 1 and of the second members
[M2 k ] are obtained likewise:
C(VIl,V)
[A] = [M2d = 1
bt.
-p.,k
The system (V. 12) for unbiased kriging of order k can then be expressed
in matrix notation as
[Kd [A 1= [M2d (V.16)
[A] = [K k r .[M2 k ],
and, thus,
(T~ = C(V, V)-[.Ar. [M2 k ],
[A r being the row-matrix transpose of the column matrix [A].
V.A.3. Combining kriged estimates
Very often, in practice, we will be required to combine the kriged estimates
of local units to form a "global" estimator of a zone or even of the entire
deposit. Assuming an hypothesis of stationarity for the entire zone, the
theorem for combining kriged estimates shows that the direct kriging of the
required global mean yields the same results as combining the local kriged
estimates. In practice, the second procedure is preferred for its greater
flexibility, and because it does not require the preceding stationarity over
the entire zone.
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 321
thus linear in V.
More precisely, let A,,(x) be the solution of the kriging of the point value
Z(x) at point x from the data configuration {Z",; a = 1 to n}. The mean
t Remark The global mean ZD = (l/D)JDZ(x}dx, which has a precise physical meaning,
should not be confused with the stationary mathematical expectation m =E{Z{x)} of the
probabilistic model. ZD is an estimator of the expectation m only when the dimensions of
the deposit D are very large with respect to the actual range of the stationary variogram or
covariance representing the model (ergodic property).
WWW.minebook.blogfa.com
322 MINING GEOSTATISTICS
Va = 1 to n, (V.I?)
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 323
inside the polygonal contour which best approximates the form of the zone
to be estimated, d. Fig. V.5.
and, thus,
(V.19)
: :6:
~===~~~
FIG. V.5. Combining point-kriged estimates.
Each time two elementary krigings zt, and Z't (j,,= i) have considered
j
[Zvj-ztJ are correlated. Thus, most of the N(N -1) error covariances
E{[ ZV; - zt, ][Zv; - Z'tJ} are non-zero, and their sum cannot be neglected
with regard to the first term (11N 2 ) Li O"~ v, of the expression of the global
variance (T~D. Theoretically, the calculation of these error co variances is
WWW.minebook.blogfa.com
324 MINING GEOSTATISTICS
V.A.4. Cokriging
Sometimes, one variable may not have been sampled sufficiently to provide
estimates of acceptable precision. The precision of this estimation may
then be improved by considering the spatial correlations between this
variable and other better-sampled variables. For example, the silver grade
of a panel can be estimated by lead and zinc grades in addition to those of
silver.
From a theoretical point of view, cokriging is no different from kriging;
difficulties arise essentially from notations, since an extra index k is
required to differentiate between the variables.
Consider a regionalization characterized by a set of K spatially intercor-
related random functions {Z, (x), k = 1 to K}. The first- and second-order
moments of these RF, assuming stationarity, are
E{Zdx)}=mk, Vx;
E{Zk'(X + h). Zk (x)} - me- . me = Ck'k (h), cross-covarrance ;
E{[Zdx + h) - Zdx )][Zk (x + h) - Zi; (x)J) = 2 Yk'k (h), cross-variogram ;
VNNI.minebook.blogfa.com
V THE ESTIMAnON OF IN SITU RESOURCES 325
The estimator k
o
zt
is a linear combination of all the available data values
of all the K variables in coregionalization:
K tl
k
Note that the condition LakO A"k0 = 1 cannot hold true if nk() = 0. In other
words, cokriging requires at least one experimental value of the main
variable Zk w
The minimization of the variance (T~kn = E {[ Z V k ,) ~ ztkO]2} under the K
non-bias constraints results in a system of (Lk n; + K) linear equations (the
unknown being the Lk ni; weights A"k and the K Lagrange parameters iLk)
called the "cokriging system";
K
c.;
"k'
t It may happen that one of the secondary variables Z1<" with k' #; k o, has the same meaning
and the same expectation (mk' ~ mko) as ZkO, e.g., ZkO and Zk' are two lead grades, but one is
obtained from core analyses and the other one from cutting analyses. The two variables Zk;o
and Zk' may have the same expectation, but different structures of spatial variability
(generally, their nugget effect would be different due to different measurement errors). The
K previous non-bias conditions are then reduced to
'r:/k #- k o, k'.
NNI.minebook.blogfa.com
326 MINING GEOSTATISTICS
Remark 1 Unlike the kriging system, the cokriging system can be written
simply in terms of the cross-serni-variogram 'Yk,dh) only if C'<'kCh)=
Ckk'(h), which entails that Yn (h) = C,,'dO) - Ce(h), i.e., if the cross-
covariances are symmetric in (h, -h), cf. section n.AA. The cokriging
system written in terms of cross-serni-variograrns is then simply derived
from system (V.21) by replacing Ck'k by (-)in).
Vk, k',
If the K data configurations of each variable Zi; are identical, then it can
be shown that the cokriging of a particular variable Zk o is identical to the
kriging of this variable from its corresponding data values {Zak o , ako = 1 to
nk o}'
In the presence of an intrinsic coregionaiization, the complication of
co kriging can be justified only if one of the variables (k o) is under-sampled
when compared with the others (k oj;. k o). This practical remark can be
extended to all other coregionalization models.
In mining applications, a coregionalization study and subsequent co-
kriging are carried out only if one of the variables to be estimated is
under-sampled with respect to the other variables with which it is spatially
correlated, d. Case Studies 2 and 3 of the following section V.A.5.
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 327
v
0t! /
/ Limit of
I~ the deposit
A I B
I
I
S3 SI I II
E F I
I
I
c J D
I
/
04 I
/
/
3
zt = L A",Z(S",) with Z(S2)=HZ(04)+Z(05)].
a< = I
WWW.minebook.blogfa.com
328 MINING GEOSTATISTICS
and
WWW.minebook.blogfa.com
Y THE ESTIMAnON OF IN SITU RESOURCES 329
Nugget
effect Kriging Poly 10 ID2
v
A2 / 2
A3 u
AI
A,z/2
FIG. Y.? Weights used for the estimation of the square panel.
VtIWW.minebook.blogfa.com
330 MINING GEOSTATISTICS
Remark 2 As the nugget effect decreases and, thus, the spatial cor-
relation increases, the influence of the central information becomes pre-
ponderant (A 1 increases from 025 to O 63). Note that for 'UJ -:;f:; 0 the weight
of the datum S2 (of support 2v) is always greater than twice the weight of
53 (of support v); this is due to the fact that 52 is nearer to the right-hand
side of panel V, which is less sampled.
Other estimation procedures The standard weighting methods (Poly, ID,
ID2) do not take into account any of the characteristics peculiar to the
regionalization under study and, consequently, provide the same estimator
whatever the structural function l' (h) may be, cf. Table V.I.
The polygon of influence method gives all the weight to the central
datum: A 1= 1 and A 2 = A 3 = O.
Both the inverse and inverse-squared distance weighting methods (ID
and ID2) give the same weight to the three peripheral data: A3 = 11. 2 / 2 . The
central datum Sl is given greater weight by ID2 (A 1= 0,727) than by ID
(A I = O'484). The mean distance from S 1 to the panel has been taken as a
quarter of the diagonal, i.e., lJ2/4.
In the stationary case, these standard methods ensure the unbiased ness
of the estimation, as they satisfy the unique non-bias condition L Aa = 1,
but they do not, by themselves, provide the estimation variances (T~. For
this purpose, it is necessary to characterize in some form the spatial
variability of the phenomenon under study. The geostatistical method is to
use the structural function y(h) with which the estimation variance of any
unbiased linear estimator can be calculated from formula (II.28).
The following remarks can be made on the results of Table V.I.
(i) In all cases, kriging provides the best unbiased linear estimator
(BLUE).
(ii) Depending on the nugget effect (i.e., the degree of spatial cor-
relation), one or the other of the three standard estimators is close
enough to the kriging optimum, but only a structural analysis, i.e., a
geostatistical approach, can tell which one is the closest.
(iii) Generally, the greater the nugget effect (thus, the greater the screen
removing effect), the better are procedures (such as ID) which do
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 331
not give too much weight to data preferentially placed with respect
to the panel to be estimated. On the other hand, the smaller the
nugget effect, the better are procedures (such as ID2) which favour
well-located data (by screening the influence of furthest data).
(iv) In all cases, an unbiased weighted average-type estimator such as ID
or ID2 fares much better than the POLY estimator. Had the
configuration of data been more clustered, e.g. if data 0 4 was located
nearby data O, the advantage of kriging over ID and ID2 would
have been much more decisive.
Panel to be
estimated
VtIWW.,,"-"ebook.blogfa.com
332 MINING GEOSTATISTICS
can thus be grouped into the set S2 = {04 U Os} of support 2v. And, as
previously, a linear estimator with three weights can be defined:
3
zt = L AaZ(S,,).
Cf= I
bt=-
1 f 1
h(u)du=-
1 udu=-,
1 f C+ 1/2
udu=c,
V v 1 EF (-1/2
L AI'Y (S""
.B 1
=
S(3) + f.L 1 + f.L2Ua = y(S", V), Va = 1 to 3,
(V.24)
and
VNNI.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 333
mt = ~ L m*(u) du = m*(ud=~[Z(S,)+Z(04)+Z(05)]'
If that least-square estimator, mt, is used to estimate the mean value Zv
over the panel V, then the weights are A. 1 = 1/3, A.z = 2/3, ft.3 = 0, and, thus,
the non-bias condition is satisfied. The estimation variance, (J"~ =
E{[Zv - m't]Z}, is calculated from formula (11.28), using the semi-vario-
gram 'Y(h)=~Var{Z(x+h)-Z(x)}-theknowledge of which requires a
geostatistical approach.
It is stressed that the estimation variance (J"~ of the LS estimator mt has
nothing to do with the minimized variance Q of the experimental residues.
In this example, 0 is equal to the experimental variance of the three data
2(5,),2(04 ) , 2 ( 0 5 ) :
Q = 1[[Z(5,)- m*(ut)]z + [2 (0 4 ) - m*(ud]2 + [Z( 0 5) - m*(uj)]2] -:;l; (T~.
WWW.minebook.blogfa.com
334 MINING GEOSTATISTICS
Nugget
effect Kriging LS Poly ID 102
A1 = 066
A2 = 034 idem idem
Absent "-3 = 0 idem idem
<T~ = 025 (J"t~ = 0,45 0.468
v
-'*2/ 2
A3 a
AI
Az/2
WWW.minebook.blogfa.com
V THE' ESTIMATION OF IN SITU RESOURCES 335
Remark 3 In the presence of the directional drift m (u), the ID and ID2
estimators are biased. This bias. E{Zv - zt}, depends on the unknown
drift m (u) and is itself unknown. The existence of this unknown bias thus
prohibits the definition and calculation of the estimation variance (T~ =
Var {Zv - zt}.
Addendum. The tensorial inuariance property of the kriging systems
(This subsection may be omitted in a first reading.)
Obviously, neither the kriging system (V.24) nor the corresponding kriging vari-
ance uk should depend on the arbitrary choice of the origin of the coordinates, i.e.,
of the parameter c. In particular, this system must be equivalent to the following
system (V.25) corresponding to the choice c =:: 0:
Va = 1 to 3,
(V.25)
In fact, with the change of variables J.t; = fJ.l + 11-2C, J..L; = J..L2, the two systems
(Y.24) and (V.25) are equivalent. Thus, A~ =A s, V/3, since the solution of the
kriging system is unique.
This invariance property of the kriging system with respect to changing the origin
of coordinates is related to the polynomial character of the component functions
Mx) of the drift m(x) and is the reason why only functions of this type are used, in
practice, to characterize the form of the drift.
The equation corresponding to a = 3 can be eliminated from system (Y.24) as it
only serves to define the parameter 1-1-2, which has no purpose when c = O. The
system (V.24) then reduces to (V.26) which, in the stationary case, is the same as
kriging panel V by the two data 51 and 52:
and
WWW.minebook.blogfa.com
336 MINING GEOSTATISTICS
data consist of the central point value Zl(5 0 ) of the main variable and the
five point values Z2(5 0 ) , Z2(A), 2 2(B), Z2(C), Z2(D) of the secondary
variable located at the centre and the four corners of the square. This
corresponds to the case of a variable Zl, undersampled with respect to
another variable (22 ) , and the problem is to evaluate the gain in precision
provided by cokriging (Z1 v is estimated from the two data sets on Z1 and
2 2 ) over kriging (ZI v is estimated from the single data set on Zl)'
A 8
x
( )(------:x
x
o
So
o
FIG. V.10. Cokriging on the square ABCD. 0, Data set Z); x, data set Z2.
Kriging amounts to estimating the panel V from its central value 2 1(50 ) .
The kriging variance is, thus, given by the general formula (II.2?):
(T~ =E{[ZIV-ZI(SO)]2}=2Yll(SO, V)-YI1(V, V)-Yll(SO, So)
(V.2?)
= 2H l1(lj2, Ij2)-F ll(l, I),
where Hi, and Fi, are the auxiliary functions corresponding to the point
model 'Yl1(r) which characterizes the regionalization of ZI(X),
For cokriging, reasons of symmetry (the coregionalization matrix being
isotropic) allow the four peripheral samples in 2 2 to be grouped together
into one single set 5 R with a grade of
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 337
The two non-bias conditions have already been taken into account by the
form of the estimator (V.28) itself, and the cokriging system (V.21) reduces
to the following three linear equations in three unknowns (A and the two
Lagrange parameters, f.Ll and f.L2):
!
~ l l (S() ' 50)- A~12(SO' 5 0 ) + A~12(SO' 5 R )+ J.Ll = ~11(50' V)
1'12(SO, 5 0) - A1'22(SO, So)+ A1'22(SO, SR)+ J.L2 = 1'12(50, V)
Y12(SR, 5 0 ) - AY2Z(SR, So)+ AY22(SR, SR)+ J.L2 = Y12(SR, V)
and
(J"~K =E{[ZIV -Zivf}
minimized
Numerical application 1 Consider first the case in which all the semi-
variograms consist of a linear isotropic model with nugget effect:
Vr> I,'; (very small range)
WWW.minebook.blogfa.com
338 MINING GEOSTATISTICS
formula (V.27).
Similarly, Y22(r) will be fixed so that the variance D is small, which means
2
that the gain, g = N / D, will be large. For example, the model I'22(r) =
~hll(r) will be taken with the parameters Ci = 01 and "U1221 = 041.
According to formula (V.29), D thus becomes
D = 21'22(lJ2/2)-h'22(l)-h'22(lJ2)
= ~C22 + O' 5607"U1221 = 0.355.
With the two direct fixed models ')'11 and ')'22, the cross-model I'12(r) will
be made to vary, taking care that the coregionalization matrix [ - Ykk.(r)}
remains conditionally positive. For this purpose, it is enough that the two
parameter matrices [C kk ,] and [trrkk'] be positive definite, d. section III.B.3,
conditions (111.26). It is thus enough that the two following determinants
be positive or zero:
ll
lee 12
el21 >0
C 22
::> C12<J(C I l . C n ) = O2236 .
12
trrll UT 1 >0 ::> UJ121 < J ('fJJ1l1. UTnl) = O9168.
I 'fJJ12 trr22
Table V.3 gives nine pairs (Cn , "UT 12 1) which satisfy these conditions, and
gives for each one of them the value A of the weight assigned to the
secondary sample Z2(SR), as well as the relative gain gr provided by
cokriging.
Remark 1 When the increments [ZI(X + h)- Zl(X)] and [Z2(X + r)-
Z2(X)} of the two variables are independent, the cross-semi-variograrn
')'12(r) is zero, and the secondary data on Z2(X) adds nothing to the
estimation of the main variable ZI(X),
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 339
e 12 'UJ12 1 A gr
WWW.minebook.blogfa.com
340 MINING GEOSTATISTICS
51 and 52 being the two sets of core data and cuttings data, respectively.
The supports of the core lengths and cuttings are negligible with respect to
the support V of the block.
As the matrix of coregionalization of the two variables ZI and Z2 is
['Ykk'(h), k, k' = 1,2], the cokriging system (V.2l) is written in terms of
se mi-variograms:
L Ai>I'Y1 1(X", I ' Xi>J+ L A/32Y12(X",p X.aJ+J.Ll
BlESI fhES2
(V.3D)
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 341
and
(TZ'K = L AU 1 I'll (x,,!, V)+ L Au 2 1'12{X" 2' V) +,a1 - iill (V, V).
"lES1 a2 ES2
It should be recalled, from section V.AA, remark 2, that the set 5 I of the
unbiased data ZI(X a .) must not be empty. This restriction is quite natural,
since the unknown bias of the data set 52 cannot be corrected unless there
are actually some unbiased data in SI. Under the condition 51 - 0, the
influence of the unknown bias of the data set S2 is made nil by the relation
L "'/31 + L "'/32 = 1.
I3,ES, 132ES2
But there is still a need for cokriging, since the two spatial structures )'11
Numerical application The three RF's ZI(X), Zz(x), s{x) have the
following characteristics.
First, they are all stationary with unknown expectations:
E{ZI(x)} = m1; E{Z2(X)} = m i:
E{c (x)} = E{Z2(X)- Zl(X)} = mi - ml,
WWW.minebook.blogfa.com
342 MINING GEOSTATISTICS
the quantity of fines is inversely proportional to the iron grade Zl). Thus,
Y12(h) = tE{[Zl(X + h)- Zl(X )][Z2(X + h)- Z2(X)]} = '}'ll(h)
as
Z2(X + h)- Z2(X) = [ZI(X + h)- Zl(X)] + [s (x + h) -sex)],
and the increments of sex) are independent of the increments of Zl.
Similarly,
Y22(h) = !E{[Z2(X + h) - Z2(X )]2} = '}'II(h)+ Ye (h)
= Yl1(h)+ C".
Thus, the following regionalization model is obtained:
Indeed, in practice, the two direct serni-variograms /'11 and '}'22 are
calculated from the corresponding experimental data and they usually
differ only by the nugget effect:
Y22(h)-Yl,(h)= C e,
The cokriging configuration of Case Study 2, above, cf. Fig. V.1 0, will be
taken as an example, i.e., the estimation of the mean grade Z, v over a
square block V of side 1= 20 m by the two data sets 2,(50 ) with weight 1
for the central core analysis, 2 2(So) with weight -A and Z2(SR) with weight
+A for the cutting analyses at the centre So and the corners SR of the block.
The quasi-point serni-variogram YI1 (h) is the model fitted to the
experimental semi-variogram of the iron grades of core length at Tazadit
(Mauretania), d. section II.D.S, Case Study 1, i.e., a spherical model
without nugget effect, range a = 40 m and sill C = 133 (% Fe)2.
For the direct estimation of 2 1 V by the only central core grade Z, (So),
the variance O"i, given by (V.27), is
(]"~ = Q185C = 24'6(% Fe)2.
Cokriging, i.e., taking the cutting data into account, gives the following
results from (V.29):
N = 0243C = 323 and D = 1.25 Ce +0451 C.
By taking the variance of the measurement errors C; equal to the first
experimental value YlI(3), i.e., C, = 15, D becomes 787, and the weight
assigned to the corner cuttings data Z2(SR) is A = N / D = 041. The gain of
cokriging over kriging is thus equal to
g = (]"~ -(]"~K = AN = 133 (% Fe)2,
i.e., a fairly good relative gain: gr = g/a~K = 054.
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 343
In spite of their bias, taking the cuttings data into account provides a
relative gain of more than 50% in terms of estimation variance.
WWW.minebook.blogfa.com
344 MINING GEOSTATISTICS
and
1L:/3
A/3 = 1
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 345
The various steps in the practical application of the kriging of each panel
V are as follows.
(i) Select the data Z", to be used in the estimation of Z v
(ii) Calculate the mean covariances C(v"" VJ3) between all the selected
data (these are the elements of the first member matrix [K] of the
kriging system).
(iii) Calculate the mean covariances C(vO" V) between each of the
selected data and the panel V to be estimated (these are the ele-
ments of the second member matrix [M2]).
(iv] Choose an algorithm for the solution of the linear system.
To carry out a "crude" kriging without any consideration for computer
cost would consist of taking all available data without distinction within a
neighbourhood of quasi-stationarity centred on each panel V, calculating
exactly the various mean values C and then applying a precise algorithm
which would give an exact solution of the linear system (such as S[MQ~SSP
IBM). Such an application would yield very precise results, but would also be
very costly in terms of computer time.
To establish a practical alternative, each one of the preceding steps must
be critically examined to provide an approximate procedure which will
minimize computer time while yielding results that are within acceptable
limits of approximation.
The five essential points to be used in reducing the kriging cost are:
1. the reduction of the dimension of the kriging system;
2. the reduction of the number of kriging systems;
3. the rapid calculation of the various mean values C (or ii)~
4. the preparation of a data file adapted to the kriging plan;
5. the choice of a time-competitive algorithm for the resolution of the
kriging system.
'#NNI.minebook.blogfa.com
346 MINING GEOSTATISTICS
for distances Ih 1< b. Thus, to estimate a panel V(xo) centred on xo, only the
data inside a neighbourhood b(xo) centred on Xu are considered, d. Fig.
V.II. This neighbourhood is not necessarily isotropic.
.,..---.,.
/" '
I
/ "' "-
I \
I I
I J
\
\ /
\
" __
.....A(xo).... /"
/
It may happen that the data within b(xo) are either too few, in which case
more distant data may be included in considering a non-stationary kriging
system, or too many, in which case a reduced neighbourhood A(xo) may be
considered. In fact, this procedure amounts to neglecting the weights of
data exterior to A(xo), d. Fig. V.II. This approximation is justified when
the data within A (xo) can be considered as sufficiently screening the
influence of the data exterior to A(xo). This screen effect increases as the
nugget constant of the underlying structural function becomes smaller (d.
section V.A.I. remark 7, "The influence of the nugget effect") and also as
the data within A(xo) surrounds more completely the panel V(xa) in all
directions, without gaps.
A prime objective in defining a data neighbourhood A (xa) should be to
ensure that it avoids all risks of bias in the estimation. Thus, in two
dimensions, for example, when the data configuration is such that one
direction f3 is undersampled, although the structural continuity in this
direction is significant, it is advisable to extend the estimation neighhour-
hood A(xo) in this direction so as to include more data 2(3, which will take
into account the .a-direction continuity in the estimation, d. Fig. V.12.
Examples of such a procedure are the automatic mapping programs that
partition the space around Xo into equal parts (e.g., four quadrants in R 2 ,
or eight octants in R 3), and only the two or three data closest to Xo are
considered within each of these parts.
It often happens when estimating panels on the border of a deposit that
most of the data lie on one side of the panel, i.e., they are preferentially
located in a rich zone, ct. Fig. V.13. In such a case, it is advisable to do one
of the following:
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 347
(i) limit the extent of the neighbourhood A (xn) to the favoured side of
the panel, so as not to give too much influence to the preferentially
rich data;
(ii) introduce, with due care, a supplementary non-bias condition cor-
responding, for example, to a linear drift m (x a ) = an + a lX a in the
direction a of impoverishment;
(iii) not carry out the estimation without any additional data - this would
be the proper course if preferential data positioning as well as
heterogeneity of the mineralization were suspected.
xx- - - - - - - - -~~Z/l
)( I
x
x
Xo
x
,
J
I
)( A(xo) /
---------_.--
a
...-..-- x
x
X
/'
\
X
/// x\ x X
I x
W
,/
/ x x t
( x x
x J
\ I
"- ...... / x
" Border
WWW.minebook.blogfa.com
348 MINING GEOSTATISTICS
k =1 2 3
x x x x
x
x
4 5 6
x )(
V x
x x x
7 8 9
x x
x )(
Z9 = (Z;9 + Zi.,J/2,
y(Sg, Sk) = [Y(Xi9 ' Sd+ y(xh' Sk)]/2, vs;
This grouping of the available data into nine sets allows the initial
dimension (17) of the kriging system to be reduced to 10. The corollary of
this reduction in the dimension of the system is a loss in precision: the
kriging variance resulting from the nine mean grades Z; will be greater than
that resulting from the 16 elementary grades. If the data are distributed
more or less uniformly, as shown on Fig. V.14, then this loss in precision will,
in general, be negligible. To evaluate this loss in precision, the two kriging
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 349
Structural symmetry All data that would receive strictly equal weights
from kriging can be grouped together. This grouping reduces the dimen-
sion of the kriging system but does not alter the kriged value or the kriging
variance.
Such data must be "structurally" symmetric with respect to the support
V to be estimated. This symmetry will be identical to geometrical sym-
metry only if the structural function y(h) is isotropic. For example,
consider the estimation of the square panel V on Fig. V.I5 by the
geometrically symmetric configuration of the five quasi-point grades
Z (x d, ... ,Z (x 5)' If the structural function y(h"" h(3) is isotropic, then the
four exterior data will have exactly the same weights and they can, thus, be
grouped into one single set called the "first aureole" with a mean grade
ZA = [Z(X2)+ Z(X3)+ Z(X4)+ Z (xs)]/4. The two-weight linear estimator is
then zT = A j Z ( X l ) + A 2ZA'
x,. W X"
a
FIG. V.IS. Structural and geometrical symmetries.
NNI.minebook.blogfa.com
350 MINING GEOSTATISTICS
A B
x x x x
x )( x x
x x x x
x x x x
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 351
x x x
Kriging of gaps in the data grid It often happens that certain points on a
data grid have, for various reasons, not been sampled (difficulty of access,
accidental interruption of drilling, not enough time, etc.). Before consider-
ing these gaps as missing data, it should be verified (d. section III.C.l,
"Review of the data") that, in fact, they are not zero values nor pref-
erentially located in some way. Thus, consider a regular grid with non-
preferentially located gaps. There are as many different kriging systems as
there are different locations of the gaps, cf. Fig. V.IB. The presence of gaps
in a regular data configuration thus increases the number of kriging systems
to be solved.
In order to return to a unique kriging system corresponding to a
complete data configuration, each gap can be replaced by its kriged esti-
mate. It has been shown, d. A Journel (1977, French version), that the
x x x x x
x x x x 0
XI X2
Gaps
X X X X 0
'#NNI.minebook.blogfa.com
352 MINING GEOSTATISTICS
x x x...1- Z(.Xi)
x
x jJ,.
v x v' x
x x
x
x
x A
x x )( Z(Xj)
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 353
with
'Vk~k',
and, finally,
(V.3!)
YNNI.~ebook.blogf
. .com
354 MINING GEOSTATISTICS
The diagonal terms thus depend on the number and supports of the data
Z (Xjk) in each domain {3;, and the random kriging system is not identical for
each panel V to be estimated.
.---"""0'101 ".~~
/1,. <,
i
I
:
I
:
I
:
I /
[]
I I I I
~--- Cross-section
( a) ( b)
rDrif'l
x
x x )(
(c ) ( d )
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 355
(iii) In one dimension, (3; may be one of the sides of the panel V to be
estimated; this side may correspond to a drift sampled by channel
samples, d. Fig. V.20(c).
(iv) In two dimensions, (3i may be the discrete set of the eight nodes of
the first aureole of a regular grid. Gaps may occur at random in any
of these eight locations, ct. Fig. V.20(d).
NNI.minebook.blogfa.com
356 MINING GEOSTATISTICS
In order to build a data file adapted to this kriging plan, the space is divided
into a certain number of grid squares, each square being indexed by a pair
of integer coordinates (iUk> jVk)' For each domain V k the following
information is recorded in the data file.
(i) The number n of data (with support supposed constant and quasi-
point) within Vk .
(ii) The coordinates (Ui,., Vi,.) of the location Xi,. of each of these data
(ik = 1 to nk). This information allows the calculation of the mean
values y of the first member kriging matrix [KJ:
VNNI.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 357
Va,f3.
The symmetry property of the first member matrix [K] suggests the use
of the square roots method (the matrix [K] is expressed in the form of a
product of two triangular matrices, one the transpose of the other: [K] =
[Tf [T]), or the Cholesky method ([K] = [R] . [D] . [R]\ where [D] is a
diagonal matrix). Both these methods can be found in B. Demidovitch and
I. Maron (1973, pp. 287-294). These two methods were tried and rejected
at the Centre de Geostatistique of Fontainebleau, because of the machine
t If the structural analysis has been carried out in terms of the serni-variogram y(h), the
kriging system can be written in terms of the covariance C(h) = C(O) - y(h) = 1'(0) - y(h),
provided the covariance exists. If the covariance does not exist (e.g., a serni-variogram
model without a sill, i.e., I' (co) = co), a pseudocovariance C'(h);; A - y(h) can be consid-
ered, where A is a constant with a value greater than the greatest value of Y(v", v/l) that
can be encountered in the various kriging systems. Because of the non-bias condition
L.a A(3 = J, A is eliminated from the kriging equations, which are then written as
and provided there is no data grouping, nor hole effect in the model:
'rIa,{3.
WWW.minebook.blogfa.com
358 MINING GEOSTATISTICS
time involved. Two other methods are used at Fontainebleau: the standard
pivotal method and the method of successive approximations when the
number of unknowns is greater than 20 (which is not often the case in
mining applications).
VNNI.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 359
or
,..
Ilall=max L la;jl<l
i i= I
or
,.. " 2
liaW = ;=1
L L
i~1
laill < l.
(V.32)
(V.33)
or
v = =- - - - - - - -
L [A ~k -I) - Aik -2)] . (A i
k
)- A~k -I 11 L (LU ;k -1)] . [~A ~k 1]
i
2
2
6,(1" E k =
v" [ (k)]2 ;
1_ v '7 k ii LiA i (V.34)
VtIWW.minebook.blogfa.com
360 MINING GEOSTATISTICS
Using this last relation, the iterations can be stopped as soon as the
imprecision ~(Tlk is less than a given limit.
The listing of program REITER is given in section V.B.3.
As it is REITER works only for simple kriging systems, d. section
VIlLA.!, i.e. for kriging systems without any unbiased ness condition.
REITER should therefore be modified before being used in KRI3D.
(i) GELG (standard IBM subroutine) pivot method with search for
maximum elements.
(ii) RECHOL, Cholesky method.
(iii) RELMS, pivot method without search for maximum elements.
(iv) REITER, method of successive approximations with Lusternik's
6
correction. The precision on the variance calculation is dO"t < 1 0-
for a kriging variance close to 1. The number of iterations is 2 for
n ~ 10 and 10 for n = 40.
Dimension,
n GELG RECHOL RELMS REITER
5 002 0-02 0-01 0-01
10 009 0-07 003 003
15 024 018 008 006
20 050 0-36 0-17 010
30 157 1-20 0-48 020
40 366 272 105 036
The times given in this table are only relative (in IRIS-SO seconds), the
same FORTRAN programs translated by another computer would result
in different absolute times.
Note that, for the most common dimensions (n :s; 15), RELMS is three
times faster than the standard GELG, and, for large dimensions, (n ~40)
REITER is more than 10 times faster than GELG. Moreover, the per-
formance of RELMS and REITER can be improved if they are rewritten
taking account of the particularities of the machine language of the
computer used.
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 361
Program CLAS
This program is designed for the classification of drill-hole data within a
regular network of parallelipedic blocks. The drill-holes are of any length
and direction in the three-dimensional space. For each block of the
network and, according to the index, when INC = 1, the number of distinct
samples having their centre of gravity within the block is calculated. The
characteristics of each of these samples (coordinates, length of the sample,
number of the drill-hole from which it originates, mean grade, etc.) are
then put into memory. Thus, for the block in Fig. V.21 which is intersected
by two drill-holes, two distinct samples are considered on drill-hole no. I,
and four distinct samples on drill-hole no. 2.
When INC = 0, within each block the samples belonging to the same
drill-hole are grouped, thus constituting pieces of core originating from
/ Four samples
Drill-hole 2
Drill-hole I
tThe Centre de Geostatistique of Fontainebleau gives out (under certain conditions) its own
Mining Geostatistics package, d. Ch. Huijbregts (197 5b).
VtIWW.minebook.blogfa.com
362 MINING GEOSTATISTICS
distinct drill-holes (e.g., two pieces of core for the block of Fig. V.21). The
characteristics of these pieces of core are put into memory.
If required, the program CLAS provides a listing of this classification
once completed.
IZC = I
BlOCk no NY
IYC e NY
f--+-+----------(
IZC" NZ
Origin of the block
r network XYZ (3)
Organization of the data file Only the informed blocks (which contain at
least one sample or piece of core) have a memory reservation on the file.
The size of this memory reservation is given by the parameter NIMB -
maximum number of data per block; each datum being characterized by
nine values (the 2 x 3 coordinates of head and end of sample, the length of
sample, the number of the drill-hole and the mean grade); there are
9 x NIMB words reserved per block.
The location within the file of this memory reservation is read from the
integer table NUM (NX, NY, NZ). The number of data contained in each
informed block is given by the integer table NPB (NX, NY, NZ), with
1 < NPB >s; NIMB. If NPB;2< NIMB, only the NIMB first data are stocked in
the file.
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 363
The CLAS version given here corresponds to a direct word coding on file
(ell French computer). This type of coding can easily be adapted to the
particular coding required.
SUBROUTINE CL~SINS,XYZ,N)~ltD8,INC,Il,IFl,IF2
l,NIMB.NUM,NPB,IS'
C
C
C SORTING OLl CF ORIlL-~OlE DATA
C WITHIN A REGULAR SYSTEM OF BLGCKS (CEFI~EC BY
C ORIGIN ANr GRIO):SCRTING C~N EE SA~FlE B~ SA~PlE
C DR BY DRILL-HOLE INTERSECTIO~ RESULTS A~E WRITTEN
C ON A DIRECT ~CCESS FILE
C
C PARAIitETEPS
C N$ NUMBER OF CRIlL-~CLES Te EE SORTEe eLl
C XVZ(3) CCOROINATES Of CRIGl~ OF 8lCC~ SYSTE~
C fLOCKS ARE RANKED It-. INCREASI~G )C
C It\ CECRE.6SHG 'y
C F~C~ lOP TC M~ll IN l
C ~XYI(3J GRID OF BLOCKS AL(~G OX.OY,C2
C ORIGtN CF BLOCKS 15 CENTRAL lC BLOCK:
C LINE NY,COLUMN 1,lE~EL 1
C LINE 1 CORRESPONDS 1( ~AXt~u~ ~
C OB(3) OI~E~SION OF BLOCK AlCt\G AXIS eX,CV.Cl
C Il UNIT NU"8EA OF CRIll-HOLE DATA FILE
C IFI(20) FORMAT Of DRIll-HClE COll~R I~FCR~AT[C~
C :DRILl-HOLE NUMeER,~L~BER CF SA~PlES,
C COCRDINATES OF PCINl ~HERE S~~FlING
C ST~RTS
C IF2C20J FOR~AT CF SAMPLES:EN( COORDl~AlE,~ALUE
C NIMB MAXIMUM NU~8ER OF INFC~~AT[CNS (O~SIDEREO
C IN A BLOCK:OEFINES RE(C~O LENGT~
C NUM(NX*N~.Nl) ACCRESS ON FILE IE CF eLCC~.IF
C NUM=Q ai oc s (O",HIN$ ~o Ir-.FCR"'ATIC~
C ~PBCNX.~Y*Nl) INCICATES ~u~eER OF DATA OF BLOCKS
C
C CPTIOt\S
C INC.EQ.O All DISTINCT CRlll-relE INTE~5ECTIO~S
C ARE CO~PUIED AND ~RITTEN
C INC.EO.J SCRTlhG AND FllIN( SAMPLE 8~ SA~PlE
C IS.NE.l SORTING RESULTS ARE PRIN1EC B~ SVBRCU1INE
C
C ..... COMMOf\S
C lOUT LINE PRINTER UNIT ~lMeER
C TEST INFERICR BCUNCARY CF EXISTI~G DATA
C IF T.lE.TEST MISSING OR El(~I~ATEC DA1A
C IWRIT/1E,Ifl,IE2 UNIT NU~BER~ OF CA FILES:
C SORTED CA14,ARRAYS ~u~ AND NPB
C 4
WWW.minebook.blogfa.com
364 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 365
I AD=NU~ PliOBl)
IFIIA[.GT.01GC TO 11
ICL=ICL-tl
NUM(NCBl)=ICL
11 IFlINC.EC.01GC TC 2
c
C SORTI~G AND W~ITING SA~PLE/SAMPlE
C
NPBCNCBL1=NPB(NOBl)+1
IF(NP8lNCBL).GT.NIMB)GC l( 13
12 NCUT=NCUT.l
C
C W~ITE ON DIRECT ACCESS FilE IE AT ~C(RESS
C IDA IN WORDS
C
IDA=IRE*(NUM(NOBl)-1)+9.(~Fe(N08l)-lJ
WRITE DISC IE,IDA,NUS,XO,~C.IO.XF,'F,ZF,HC,T
C
GC Te .:
13 ~RJTECIO~Tt2002)hCBl~I~C,J)C,IIC,NFB(NCBl)
GO TO 3
C
C COMPUTATION CF DRllL-HCLE INTERSE(TICh
C
2 IFCNOEl.EQ.OIGC TO 23
20 IF(NOEl.NE.Noel1GO Te 22
If(NPECNC8L).GT.NIMB)GC TC 3
I\CT=NCT"l
TV=TV+f-iC*T
t1l=Hl+t;C
XTF=XF
YTF=YF
ITF=Zf
IF(IEC~.lT.NECH)GO TO ~
c
C END OF DPIlL-hCLE
C
21 JND= 1
IFCHl.lE.O.JGC TO 4
TV""T\I' IHL
NCUT~'CU1+NCT
c
IDA=IRE*(NU~("OBl)-1)+9.(~Fe(NOBL)-1)
WRITE DISC JE.IDA,NUS.JTtt'TO.ZID.)lf,~TF,
lZTF,Hl,T'V
c
GO TC 4
c
C .RITE ORILl-HClE INTERSEClION
C
22 IFlNPECNOBl'.GT.NIMB)GC lC 23
XTF=)(C
WWW.minebook.blogfa.com
366 MINING GEOSTATISTICS
YTF=YC
ITF= Z r
TV=lV n"L
NCUT .. to.CUT+NCT
c
C WRITE ON DIRECT ACCESS FILE IE AT ~OCRE~S
C IDA I~ WCROS
C
IOA=IRE*CNUM(NOBl)-ll+9*C~feCNOBl)-lJ
WRITE DISC IE,IOA,NUS,X1C,'TO,Z10,)lF,YTF,ZTF,HL,TV
c
23 NOB1=t..OBl
t..P8(NCBl)=NPB(NOBL)+1
IFCNPBCNOBl).GT.NIMBJGC Ie 13
C
C INITIALize NEW INTERSECTICt..
C
NCT= 1
NTUT=NTUT+l
JND= 1
IF(IEC~.EQ.NEC~)GC TO 12
JNO=O
XTD=XC
YTD=YC
ZTD=Z r
XTF=Xf
YTF=Yf
ZTF=: If
TV=HC*T
HL=HC
3 CONTlt..UE
XO=XF
YD=YF
ID=Zf
31 caNTI~ue
IF(INC.EC.O.A~C.JNO.EQ.OJG( TO 21
4 CONT I to.UE
DO 41 I J l =1 , NB
41 NPB(IJL)=MINOCNPe(IJL),Nl~E)
IAO=O
WRITE DISC IEl,IAO,(~U~(I),I=l,NB)
WRITE DISC IE2,lAD,(NP8(IJ.I=1,NBJ
IFCINC.EQ.OlGO TO 42
WRITE(IOUT,2003)ICl,NECT.~CUT
GO TC 43
42 W~ITECIOUT,2004)ICl,NECT,NCUT,NTUT
43 IFCIS.EO.I1GO TO 5
c
c PRl~l RESULTS IF IS.to.E.l
c
wRITEIICUr,2010)
co 41 L=l,NI
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 367
11=IOE-CE(3)*(L-l)
~RITE(IOUT,2C05)L,ZZ,TESl
DO 47 I=l,N'I'
00 46 J=l,NX
IJL=I+NY*(J-l)+NX*NY*{l-])
If(NP8IIJL).EC.OlGO TO 4t
VY=YOe+08(2J*(NY-I)
WRITE(IOUT,2006II,YY
r AD= NlM ( (Jl J
XX=XOE+OS(I)*(J-l)
WRITECIOUT,20C7IJ,XX,NFBCJJl),IAC
NO=N P E ( I Jl )
IDA= I RE* (IAD-l)
DO 45 1(=I,NG
READ else lE,IDA,NUs,xo,~r,ID,)F,VF,lF,Hl,T
WRITE(IOUT,2008'NUS,XO,VC,ZO,XF,YF,IF,Hl,T
IDA= ICA+C;
45 CONT II\UE
46 CONTItl:UE
47 CONTI~UE
C
2000 FORMAT(l~l,'SCRTING eN FJlE OF CRIll-hOLE "
l'INFO~MATIONS 'IIH ,5X,'BlCCK GRID: CRIGIN "
2'XO= ',FIO.3,' VO= ',FI0.~,' ZO= '.FIO.3/lH ,
317X, , NUfiIlBER IOX='.I4,' IC'-=',14,' /(1=',14
4/lH ,11X,' SIZE OBX=',FIC.3,' OB~=',FI0.3,
5' OBZ=',FIO.))
2001 FORMAT(l~ ,'NUMBER OF DRIll-HOLES: ',I4,lOX,
l'MAXI~UM NUMBER Of INFOR~~110NS TA~E~ IN A '.
2'BlOCK =',12)
2002 FORMAT(l~ ,' BlOCK ',IS,' IYC;',13,' IZC=',
113,' IIC='tI3,4X,'NU~8ER (F DATA 'tI3,'A80~E "
2'AllOwEO MAXIMUM')
2003 FORHAT(I~O,' NUMBER OF BLOCKS COhT~I~IhG DATA'
1,' .',15/lh ,'TOTAL NUMBER OF SA~PLE5 =',16,
2' NUMBER OF SAMPLES TAKEN ~',I6)
2004 fORHAT(l~O,' NUMBER OF BLeCKS COhT~I~ING DATA'
1,' ;',I5JIH ,'TOTAL NUMBER OF SA~PLE5 :',16,
2' NUMBER OF S,tMPLES TAKEr. =',16,' I" THE "
316,' INTERSECT IONS WRllTe .. f
2005 FORMATflllHO,' LEVEL ',15,' ElE~ATJOh,
l'lz ',FlO.3,' LIST OF \ALUES PER BLOCK*.'
2/1H ,Ell.5,'sMISSING ~Al~E5'.
2006 FORMAT(lHO,'.lINE ',15,' '=',FIC.3)
2007 FORMATCIH ,'COLUMN ',15,' X=',F1C.3,' hU~BERI
1,' OF DATA .',12,5X,' FilE AOORESS~ ',16)
2008 FORMATCIH ,'DRILL-HOLE ',18,' ST,tRT AND END'
I,'COQRDINATES '.6(2X,F8~3),SX,'LENGT~ =',FIO.3
21tH ,'VALUE ',FIO.3.
2010 FORHAT(llllHO.
C
5 RETURfIi
END
NNI.minebook.blogfa.com
368 MINING GEOSTATISTICS
Program KRI3D
This program is designed for the kriging estimation of parallelepipedic
units from three-dimensional drill-hole data, previously sorted by the
program CLAS.
The unit Uk to be estimated is a submultiple of the block B of CLAS
network. There are NSB units to be estimated in each block B. The
location of these units Uk are deduced from the location of the block B by a
system of integer coordinates similar to that used in CLAS (the unit no. 1
within B is located at the superior left corner of the upper level of B, cf.
Fig. V.22).
The same estimation neighbourhood is considered for all the units Uk
comprising block B. The form of this neighbourhood is parallelepipedic,
centred on block B, and consisting of an integer number of blocks B;. The
data within the estimation neighbourhood are grouped into NL sets. Each
of these sets consists of the data which are within a certain number NP
(L = 1 to NL) of blocks B, comprising the neighbourhood. A particular
kriging weight will correspond to each of these NL sets of data. All the
units Uk of the same block B are estimated from the same data configura-
tion; their kriging systems differ only by their second member matrices.
The mean value C(LI, L 2 ) calculations, between two sets L 1 and L 2 , are
carried out taking into account the precise locations of each datum (sample
or piece of core) constituting L 1 and L z . For the calculation of the mean
values C(L;, Uk), the unit Uk is approximated by a regular set of discrete
points.
The algorithm used for solving the system of linear equations is RELMS,
cf. following section.
The total volume to be estimated is contained in the network of blocks B
defined by the program CLAS, and is constituted by an integer number of
blocks B; (those indicated by the index IEXP = 1). All the units Uk, com-
prising a block B to be estimated, are estimated.
WWW.minebook.blogfa.com
V THE ESTIMAnON OF IN SITU RESOURCES 369
for the definition of the nugget effect, the structure of this subroutine is
similar to that of FUNCTION GAM given in section III.B.5.
t
C
C KRI-30 KRIGING PROG~~~
C
C KRIGING O~ A T~REE-DIMEN~lCN_l SYSTEM Cf B~OC~S
C DEFINED BY ORIGIN ANt GRID F~C~ ORILl-~(lES WITH
C ANY CRIENTATION .. CATA HAve BEEP! INllIAlLl' SORTED
C IN BLOCK SYSTEM B~ SUBROUTINE ClAS EACH elOC~
C MIGHT BE SUBDIVIDED IN A SVSTE~ OF SUB-SLeCKS wITH
C SAME KRIGING NEIGhBOURHOOD
C
C PARA"ETE~S
C :SORTING OF CATA lIN BLOCK SYSTEM)
C XOB,Y08,ZOB ORIGIN OF BLOCK !'STEM : CE~TRE OF
C BLOCK LOCATED DOWN lEFT ON TOP lE~EL
C tax,DeY,CBI SIZE OF BLOCKS ALONG AXI~ (X,OY,Ol
C NX,NV,NZ BLOCK SYSTEM 15 AA~'NGED AS I~ <ClAS>
C HIMB SEe SUBROUTINE ClAS
C TEST,JS SEE COM~ON AND PR!P!l (PTICN
C ,.S8 ,NSY ,HSZ SUB-BLOCt< SYS 1EII :lJ,J TH IN E'CH BteCK
C :KRIGING PLAt-
C NL NUIlIBER OF DATA GROUPINGS (KRIGING I.E IGIHSJ
C IJV(3) NUMBER Of lINES,COL~~~S,lE~El~ (f
C PARAllELEPIPEOIC K~IGING NEIGHBOURHOOO
C "PCf~l) NUMBER OF BLOCKS COP~ESPONDI~G TO EACH
C CAT' GROUPING
C NoeIMAX.NP) LOCATION OF THE!f BLOCKS .ITHIN
C NEIGHBOURHOOO.BteCKS ARE NlMBEREC
C ACCORDING TO LINE ,COlU~~,lE~El
C POSITION IN Nel'~BOURHOOC
C :STRUCTURAL P~~AMETERS
C NST.cO NUMBER OF STRUCTUAES,~UGGET EFFECT
C C(NST) SILLS
C A.eNST) RANGES:<O. EXPCNEN~I'L,>O. SPHERICAL
C C'XtNST,3J,CAYtNST.3J,CAIIN!1,3) SEE <C(~A>
C NBX,NBV.NBl SUBCIVISION Of SUB-8LeCKS fOR
C COVARIANCE CALeltATICN
C :tIRECT-ACCE~! fILES U~IT ~U~BERS
C IE: SORTED DATACPROVIDED BY <elAS
C lEX: AR~AY IEXP LOCATION Gf KRIGEC BLeCKS
C I EXPCNX*NV*NIJ =1 KAJ~EO BLOCK =C NO KRIGING
C IN SA"E OAeER AS BLCCK SYSTE~
C lEI. AR~AY NUM FROVIOEO ev LAS>
C IE2: ARRAY NPB PROVIDED fy LAS>
C IE2: ARRAY NPB PROVIDEO BV eLAS
C 10: OUTPUT FilE OF RESULTS.ONlY BlCCJS SUCH
C THAT IEXP=l ARE WRIITE~ IN SEQUENCE.wITHIN
C EACH BLOCK SUB-BLOCKS APE NU~8E~EC ACCORDING
C TO lI~e,CCLUMN.lEVEl
C
WWW.minebook.blogfa.com
370 MINING GEOSTATISTICS
C MEMOR~ peCUIRE~EN1S
C CB,CBl,CB2 ~S8=NSX.NSY.~5Z
C XOB.YCB,ZOB ~SB.NBX*N8Y.~EZ
C IP 3.N8T=3.(~P'1)+NPC2)+ +NPC~L))
C NIH ~l.l
C XA,YA.ZA ~8T.NIMB
C TA ~l
C TK,SK NX*NY*NS8
C NUM,NPB NX*NY*IJV(3)
C R,RR,X (NL+l'*NSe
C A (HL+l)*CNl+c)J2
C
C (0I4MC,..5
C lOUT LINE PRINTER UNIT ~lMBER
C TEST lNFERICR BOUNDA~Y CF EXI5TI~G DATA
C IN CASE Cf fLOCK WITH NO INFCPMATICN
C VALUES A~E THEN =1E51.
C l'JOtSII PARA~&lERS FeR ~(IS
C ISTRUtl PARAMETERS FOR C(~A
C
C CPTIOfl,S
C IS.EQ.l STATISTICS Of ~RJGE( BLOCKS _RE PRINTED
C IS.NE.l ALSO RE5UlTS BLOCK/BLOCK ARE P~I~TEO
C
C SUBROUTI~ES CALLED
C MRIGMS+RElMS KRIGING SYS1E~+~OlUTION
C VOIS NEIGHBOUPHOCO OF ESTIMATEC flOCK
C tBARS AVERAGE COVARJA"CE
C COVA COVARIANCE FU~ClICN
C
C COMME"TS:
C BLOCKS ARE NUMBERED AS I~ SteFoUTINE CLAS:BLOCK
C ~UMBER '+NY.(J-l)+NV*NX*(L-I) IS eN LINE I,
C COLUMN J,LEVEl l lINE 1 CC~RESPONCS lL~"S Te
C ~AXIMUM ~ COORO[NATE,COlU~~ I to CPIGIN,lEvEL 1
C TC TOP lEVEL KRIGING PLA~:~E[GHBCU~H(C[ IS
C DEFINED AS ~ PARAlLELEPIPECIC ENVELOPfE MADE UP
C CF BlOCKS INFOR~ATIONS BeL(~(ING TC ELeCKS OF
C TtE ESTI~ATION NEJGHeOU~HCCC IPE GPCUFE[ IN NL
C SUB-SETS EACH OF WHICH Will t'VE A PA~TICULAR
C KRIGING WEIGHT (HOICE OF T~E5E SLe-SET~ SHCULD
C BE ~AOE ACCCRDING re lOCATICh,SVM~ET~\,EIC TO
C DEFINE A PAPTICUlAR SUB-SET L THE NUMEE~S OF THE
C BLOCKS DELINEATING IT ARE GJ~EN AS A lI5T OF ~P(L)
C ~AlUES (ARRAY NCe) THESE BLeCKS ARE ~U~8eRED
C ACCORDING TO THEIR RELATIVE LINE (1 TC IJVCIJ)
C POSITION 11 (LINE 1 CORRESP(~[S ALWAY5 IG Y
C ~AXIMUM),COlUMN (1 TO [JV~2) AND LE~EL (1 TO
Co IJV(3)) POSIT10NS 12 AND 13 B' 11+JJVCll.(12-1'+
c JJV(2).IJV(1)*CI3-1) SUC~ ~U~BERS ARE CC~VERTEO
C INTO ~ElATIVE POSITIONS USEC EY SU8ROlTl~E VOIS
C CSEE AR fl:.A YIP J
.",,,.,....... ..
.. ~,"""ebook.blogf
..com
V THE ESTIMATION OF IN SITU RESOURCES 371
c
c ........
C
OIME~SlCN NP[lO),NOB(75)
DIMENSION C(5),A~t5),CA~C!,3),CA~[~,~),CAlI5,3J
DIMENSION CB(S),CBICSJ,CB2IBJ
DIMENSION XOf(64),YOB(641.ICBC64J
DIMENSION IP(225),NINIIIJ.1A(lC)
OIMENSIO~ XA(150),YAI150),ZAt150)
DIMENSION TKIEOOJ,SKISOOJ,JEXPIIOQ)
OlMENS10N NUM(300),NPBI3CC)
c
DIMENSION R(SBJ,RR(SB),XIEJ,A(33C)
c
DIMENSION IJ~e3J,N~Yl(3J,)'1(3),OBe3J,JlI3)
c
CO~MCN ICUT,lEST .
COMMO~/VC(SI/~IMBtXB,Ye,lE,IJVL
COMMON/STRUC/NST.CO,C,~A,('~,CAY,C_l
C
C READ INPUT PARAMETERS AKC FRINT TIlLE
C
INP~ 1
IOUT-=3
READ(INP,lOOO)XOe,YOB,ICe,[eX,CeY,CBZ,NX,N~,
INl,TEST,NIM8,IS
~EAD(INP,lOOl)NSX,NS,NSl
cSX=CeX/NSX
CSY""C8YJfliSY
CSl:CEI/NSZ
"SB=N5X.~SY.NSl
WRITE(tOUT,2000)OBX,OB~,Cel,XOe,YCE,lCe,~y,~X,
INI,OSX.CSY,CS1,NSe
c
C CEFINE NEIG~B(URHOOO A~O CJ1A G~GLFI~G
C
READ(INP,l002)~L,IIJV(I),I:l,3)
~EAO(JNP,l003.(NP(l),l:1,~l)
~RITEIIOUT,20Cl)Nl
Nl=O
NLJ=[JV(l)
NlC=IJV(1)*IJ~(2)
IJVL=JJV(3)IZ-tl
c
C READ BLOCK NU~8ERS DEFINI~G GROUPING L
C
CO 100 l=l,NL
N=NP ( l )
REAOe INP,l003J(NOB( IJhIJ:I,N)
W~ITE(lOUT,2002)L,(NOe(IJ.,IJ=1,N)
DO 100 1=1, N
Nl=Nl+1
NNI.minebook.blogfa.com
372 MINING GEOSTATISTICS
N2=3(~1-1)
lB=(NCB(I)-lJ/NlC+1
JB=CNCB(I)-Nl(.llB-l)-l)/~ll+l
IB=NOe(I)-NLI*(JB-l)-NlC*(lB-lJ
IPtN2+1'=IB-IJV(lJ/2-1
IP(N2+2J=JB-(JV(2)/2-1
IPIN2.3J=lB-IJV(3J/2-1
100 CONTIr-UE
c
C READ AND PRINT ST~UCTUP~L FARA~EIEFS
C
~EAD(INP,1004JNST,CO
~RITE(IOUT,20C3)NST,CO
DO 103 ~=l,NST
READ(INP,l005JCIN),AA(NJ,(CAX(~,IJ,C~~(N,IJ,
lCAZtN, 1),1=1,3)
103 WRITE(IOUT,2004JC(NI,A~(~J,(CA)l(N,IJ,I=I,3Jt
I (C AY (N, I J, J = 1.3) (CAL ( ~ I) ,1= 1,3)
c
C INPUT OUTPUT CIRECT ACCE5~ FILES
C
READ(INP.l006)IE,IEl,(E2,IE~,IC
NXYZ t l':::NX
NXVI12J=NY
NXYZ(?I=NI
XVZ(l}=XCa
XZ(2J=YCB
XY l ( 3 ).:;.ICB
CB(lJ=CBX
[;8(2)=08'1
CeC3J=CBI
c
C SUBCIVIDE SUB-RLeCKS IN FCI~TS
C
PEAD(INP,lOOl)~8x,NBYt~el
NCB=NeX*~BY.NBZ
11=0
00 14 lS=l,NSZ
Iso=ceI12.-0SI/2.-0SZ*tlS-lJ
DO 13 IS=l,NSY
YSO=C8V/2.-D5Y/2.-0SY*(IS-I)
Co 12 J$=l,NSX
XSO=-D8X/2.+0SX/2.+0SX*(J~-1)
Jl=-l
J2=-1
XO=XSC-DSX/2.+0SX/(4.*Ne~J
YO=YSO-O$Y/2 0SY/(4.*hB'I)
ZO=ZSO+CSZ/2.-0SZ/(Z.*hBZ)
L=O
DO 11 ll=l,NBI
J 1=- J 1
XO*XC+Jl*OSX/CZ.*NBXl
00 11 l2:::1.2
L=l+ 1
VNNI.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 373
IFCL.~T.~BZ)GC Te 12
J2=-J2
YO=YO+J2*OSY/C2.*NBV)
00 10 J=l~NB)(
00 10 l=l,NB'1'
r 1= 11 +l
X08(Il)=XO.(J-1)*OSX/NBX
YOBCll)=YO.(!-l)*OSY/NBY
10 loBt j 1)= 20
lO=lO-OSl/NBl
11 CONT If\UE
12 CONTI~UE
13 CONTI~UE
14 CONTIhUe
C
C COMPUTE VARI.hCE Of A FOINT ~ITHIN SlB-BlOCK
C
NINt 1)=1
NIN(2)=NC8+1
CALL CBARS(XCE.YCe.IOB.l~,NIN,XCe,YCBtIDB,
IhOB,NSB,CBJ
c
C KRIGI~G BLeCK PER BLeCK
C
"Bl= a
UK=O.
IIK=O.
~K=O
IAD=O
"O=NX*~Y
IFtIS.NE.IJWRITECIOUT,20CtJ
00 1 lB=l,NI
le=IOE-C~I*(Le-1J
IF(IS.~E.l)W~ITE(IOUT,20Ci)L8~Ze
NBK=O
lO=MAXO( 1, lE- IJV (~J/2 J
ll=MI~O(~I,LB.IJV(3J/2J
C PEAC IEXF,NuM,~pe
C ON RESPECTIVE UNJ1~ IEX,IEl,IE2
C ADCRESSES START 'T a
IADl=~X*~Y*(le-l)
READ else IEl,IAOl,( IE)cP(IJJ,IJ=l,l\CJ
IAD2="X*~Y*(LO-l)
~1=NX*NY*LI-IAC2.1
READ CIS C IE 2 , I A0 2, NU,.. ( J J J, I J'" 1 , ~ 11
READ rise IE3,JAC2,(NP8(lJJfIJ=1,~l)
C COMPUTE LINE BY Lll\E
CO 2 I8=l,NY
CO 3 JE!"'l,NX
IJB= I EHa* (JB- U
IfCIEXPCIJB).EQ.O)GO Te 2
f\ e1(= N E! K + 1
~BL=NEl+l
XB=XOf+ceX*CJB-IJ
WWW.minebook.blogfa.com
374 MINING GEOSTA TISTICS
YB=YOf+OBY(N~-Ie)
Jl ( l ) = I 8
JLC21=J8
JLC3J=L8
c
C SORT CUT INFOR~ATICN WJT~Jh NEIGHe(U~hCOO
C
CALL VOIS(Nl,NP,IP,Jl,NX~Z,NPB,hU~,IE.)A,~A,
lZA,NI~.TA.NAJ
If(NA.EQ.OJGO TO 37
If(NA.GT.IJGO TO 31
C (NlY C~E JNFCRM~TJ(N fCLNC
CAll CBARS(XA,YA.ZAtl,1,~I~.XDB.YCf,2CB.NCB,
INse, CEll
CALL CeAPS(XA.YA,ZA,l,C.NI~.xoe,~CE,lCB,NDe,
INSB,Ce2)
DO 30 /C.=l,NSB
J=I(.N S8. CNBK-l)
TK(J )=TAC 1)
$KlJI=C8CIJ-2*C82(K).CBICIJ
30 CONTI~UE
GO TO 34
C SOLVE KRIGIk SYSTE~
31 CALL KRIGMS(NA,XA.YA.ZA,NI~.)Oe,YCe,lOB.Noa,
INSB,R,RR,A.X,IER,CBll
IFlIER.NE.OIGC TO 36
C SCLUTION
DO 33 K=l,NSB
J=K.NSB*tNBK-l)
1TK=0.
lK={N,A+l)*K
SSK=cetl,-x( IK)
DO 32 l=l,NA
lK=I.(~A+l).CK-l)
TTK=TTK+X(IK)+TA(I)
32SSK=SSK-)(IKj*RRCIKI
TKIJJ=TTK
33 SK(JJ "5SI<
34 00 35 K=l,NSB
J=K"'Nse* (NBK-l)
UK=UK+TK(JJ
VK=VK+1KCJI*TKlJJ
35 NK=NK+l
[F(IS.EQ.l)GO TO 3
C PRINT OUT If IS.NE.!
WRITEtIOUT.200~)18,JB,)B,~f,NA
CO 40 K-l,NSB
JzK+f\I sa* lNBK- L
40 WRITEtIOUT,2010)K,TKlJ),SKCJJ
GO rc 3
c ERRO~ED S~SlE~
36 WRITElIOUT.200SJIB,JB
GO TC 38
C ~o INFORMAT]C~
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 375
37 WRITE(IOUT,2012tJ8,JB
38 00 39 K-" NSB
,J-K+NSS* 'N8K-l)
TK(JI-1EST
39 SKCJ.-TESt
:3 CONTINUE
2 CONT INUE
N2-H8K*NSB
c
C W~ITE RESULTS ON FILE
C
WRITE DISC IO,IAD,(T~(N),SK(N),~~l.~il
IAC=I.6C+2*NZ
1 CONT I t>.UE
C
C P~INT STATISTICS CF KRJGEC BLOCKS
C
WRITe(IOUT,2013JCSX,DS~,CS2,NBL
IFtNK.EQ.O)GC TO 4
VK~(V~-UK.UK/"'K)/NK
UK=UK/f\K
4 WRITE(IO~T,2014JUK,VK,NK
C
1000 FCRMAT(6F8.3,313,F8.3, 11)
1001 FORMAT(3IZ)
100Z FORHATl4I2}
1003 FORMAT(4012}
1004 FQ~MAT(IZ,F10.4J
laOS FORMAT(fl0.4,F8.3,9F4.2J
1006 FOAMATI5IZ)
c
2000 FORMAT( UH, 4t I ),4EX, '****Kf<IGI hG Of fLOCKS* '
111.*O~IlL-HOlE DATA SO~TING GRID (I'" BLOCK S',
2'VSTE~).'II. sIze : 0)& ',Fe.3,' tv ,Fa.3
3,' oz. I,FB.3/tH " ORIGIN OF BLOCKS: XO-',
4FII.3,' YO- ',FII.3,' lea ,fll.!/lH ,
5' GRID OF ',13,' LINES, ',13,' CQlL~NS AND "
613,' lEVELS 'I' .~RIGee 8leCKS. SIZE : OSX- "
7F8.3,' OSy. I,Fe.3,' OSZ& I,Fa.3,' NUMBER',
8 'OF SUB-BLOCKS: ',(3)
2001 FOA"AT(l~ ,'KRIGING PLAN: ',12,' .eIGNTS'/IH ,
2'WEIGHT:CORRESPONCING BLaCKS WITHJfI~fJGt180UAt,
3'HOOD (NUMBERS)')
2002 FOAHAT(l~ ,2X,12,2X,':.~~,40tl),I2J)
2003 FORMAT(lHO,' STRUCTURAL PAPAMETERS OF THE ' t
l'ESTJMATEO VARIABlE'/Ui ,'''UfltBER OF !TRUCTURES'
Z,t - ',II,' NUGGET EfFECl ,F12.~/' SIll I ,
3' RANGE I tAX I CA'I .,
It' I CAl' )
2001t FORHAT(lH .Fl0.4,lX,FB.3,1),9('I',F4.Z))
2006 FORHAT(lHO,4(!',48X,'.*KRIGING RES~lTS BLOCK "
1'PEA ELOCK**'II)
20~1 FORMAT(l~O,/tlH,' .*lEVEl ',12,' Z-',
IF8.311)
VtIWW.minebook.blogfa.com
376 MINING GEOSTATISTICS
SUBRCUTINE KRIGMSINA,X~,~J,lA,NIN,)CE,YCe,lCB
1,NDB,~SB,R,RR,A,X,IER,CB)
C
C
C CO~PUTE KRIGING MATRIX At-C SClVE S~STEM
(
C PA~A"'ETE~S
c NA NUMBER Of EFFECTIVE DATA G~OUPI~GS
c XA,VA,IA COCROINATES OF O.TA ~ITHI~ NEI(~BOUR~OOO
c hlN LOCATION OF tAlA GROUPING IN XA,'A,ZA
c NSB NU~BER OF SUB-BLOCKS
c R(CNA+ll*NSBl RIGHT-~ANO SITE MATRIX
c 'CCNA.ll*(N'+21/2) COEFFICIE~l MATRIX
c XNA+ll*NSBl SOLUTION
c KTlll ERROR INDICATOR IF KIILT.NE.O
c
c SUBROUTINES C'llED
c (BARS COVARI~NCE COMPUTAl](N
c PElMS SYSTEM SOLUTION
C
C
C
OIME~SION xAtl),YA(l),IAll),XOBllJ,~CB(lJt
IIDIHlI,NIN(l)
o1ME NSI 0 ~ C8 0
OIME""SIO~ RClJ,RRIU,AU ),)0'
NEQ=N ... 1
hN=(NEC+!.*NEC/2
c
C COMPUTE COEFFICIENT ~ATP]~
C
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 377
IHuO
CO 1 J=l.NA
00 2 l:l.J
IN:J N+l
CAll CaAPS(XA.YA.ZAtl,J,~Ih,X08,YCe,lD8,NOB.
INSB,CeJ
"C INJ=CaiU
2 CONT I'~UE
C.t.ll C8ARSCXA,YA,lAtJtC.NI~tXDB,YCetlCB,NCB.
INSB.CE)
00 10 l""l,NSB
f(=-J+NEQ+(l-lJ
R (K)-CBCL)
10 RR(K)-R(K)
1 CONTINUE
00 3 J=l.,NA
INaIN+l
3 AlIN)a}.
IN:I:IlJN+t
UJNI=O.
CO 4 L=l.NSe
K-NEQfoNE,. fl-H
RU:)-l
.. RR(K):ltl.
NSHaNSe
M...NEO
c
C CALL SYSTE~ SCLUTICN RCU11~E
C
CALL REl~SIX,"'.R.~.NSM.IEPI
RETURN
END
SUBROUTINE REl~SIX,A,8'~t~,KTllTJ
c
C
C OBTAIN SOlUllCN OF S~!TE~
C OF SIMULTANEOUS LINEAR EQUATI(~S ~ITH ~~~METRIC
C COeFFICIE~T ~ATR[X UPPER TRIA~GUlA~ PA~T BEING
C ASSUMED tc BE STORED COLUMNWI5f
C
C PARAMETERS
C X(M*NJ FINAL SOLUTION
C
C
A "'*
CM+ 1 J /2 J UPPER TRIAN~UL_F PART OF
COEFFICIENT MA1PIX DESTRO~EC BY
C TRIANGULATION
C BCM*N) RIGHT-HAND SIDE HATRI) DESTRO~EC
C ~ NUMBEP OF EQUATIONS
C N NUMBEP OF RIGhT-~AND SIOE~ECTCRS
C KTILT RESULTING ERROR PARA~E1ER
C xr n, T:O NO ERROR
VtIWW.minebook.blogfa.com
378 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 379
R-At ll)*PIV
lP=l P.. l
IJ" I [-KM 1
CO ~ J"J,Ml
IJ-I J"J
ll""ll+J
it ~CIJ)-~(IJ)-R*~CLl)
Lle-K,NM,flI
DO 5
IN"'LLB+lP
5 BtIN)-81IN)-R*BellBJ
6 CONT I~UE
7 CONTl~UE
~:cA(IJI
IFCR-TOL19,8,t;
8 KTIlT-M
lleTU~N
C
C END TRIA~GUl.TION
C
C START BACK SOLUTION
C
9 PI V-l./R
DO 10 llB.M,N~,M
10 XCllBJ-BtlLB)+PIV
taM
KK-IJ
DO 13 11*1 , M1
IlK-K.,-(
PIV-I.I... U(IO
1-1-1
DO 12 LLB-I,NM,M
IN-LLB
R-Be ltd
IJ-KIC
DO 11 J-J.Ml
IJ-IJ+J
IN-INti
11 R-R-A IIJ
12 Xlll8)-R*PJV
'.JI( IN)
1'3 CONTINUE
C
C END SOLUTION
C
RETURN
C
C ERROR RETURN
C
14 KTJLT--!
Q:ETURN
END
WWW.minebook.blogfa.com
380 MINING GEOSTATISTICS
3~BRCLTINE Ce~RSI~A,YA.l~.l,J,~IN,)Cf,~Ce,lCB
l,NOB,~SB,C8)
C
C
C CCMPUTE AVERAGE CC\lRIA~(f eEl~EE~
C SETS OF CAT~ I A~C J OR Bla(~ IF J=~
(
C P/IRAMETEPS
C CBeNSB) AVERAGE COVARI~NCE
C CT~ER PftRA~ETERS SEE KRIG~S
(
C SUBRCUTI~ES CftLlEC
C COVA COVARIANCE FU~CTIrN
C .................................................
C
o I ME'" 5 ION X /J. ( 1 ) , YIi 1 ) , 1A 1 It N IN ( 1 ) ,C E 1 J
DI~E~SION XDBC11,~08(1),Z[e(1)
NC=O
CC 100 I=l,Nse
100 (811)=0.
IFe I .H:.CIRETliRN
Nl=NHdIJ
N2=NHdt+U-l
IFCJ.EQ.O)GO TO 3
c
C AVER~~E COVARIANCE BETMEEh SETS I /INC J
C
fIIl=NHdJJ
lQ=NUdJ+U-l
00 1 I2;:;Nl,~2
DO 2 J2=p.ll,~2
CB ( l ) =CB ( l ) + COVA, ( xAI 12) , '1 A ( 12 J ,I A 12 ) XA J 2 J ,
lYA(J2J.ZAI J2J)
NC=NC+l
2 CONTIJ\UE
1 CONT INUE
CBIl)=CB{lJ/~AXO(l,NC)
flETURt\
c
C AVERAGE COVARIA~CE BET~EEN ~ET I _NO StB-BLOC~S
C
3 DO 4 Kl*l,NSB
Nt*a
00 5 J2*hl,~2
00 5 Jl*l,Noe
Jl*J l+ttOS* IK2-1)
C8(K2)-tBIKl'+COVAIXA(12),'1AfJ2),Z.(12',
lX08(J2),YOBCJ2),lOBIJ2))
Nt-Ne.l
5' CONT INUE
CB(K2)ceIK2J/~AXO(1,Nt)
4 tONT It<iUE
END
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 381
SUBROYTI~E VCISINL.NP,IP.4l.NXYI.~PB,~LM,IE,XA
1,YA,1~.N(N,TA,NA'
C
C
C SORTS OUT I~F(RMATIC~~
C BELONGING TO NEIGHBOURHOOD OF ESTf~~TEC ELCCK
C AND GROUPS T~ESE BY SETS WITh SAME ~PI'I~G WEIGHT
C ACCQRDING TO KRIGING PLAN CA1' ARE READ ON FILE
C PROVIDED ev <CLAS>
C
C PARA MEIE~ S
C IPI3*N8TI LOCATION OF 8l0CK~ ~ITHIN NEI~HBCURHCOO
C XAINIMAI,YAfNIMAJ,lAINIMA) C(CRDI~ATES CF DATA
C NIMA=Nel*~I"B
C NINe NL+l1 LOCATION OF EAC~ GROUPING I~ ARRAYS
C XA,VA,lA
C TAINt) AVERAGE VALUE OF EAC~ GROUPI~G
C COMMONS
C IVOISll SEE ~AIN PRCGPA~
C
C
CIMENSION NPIIJ,JLlIJ,JP(lJ
DIMENSION NXYZ(l),NPB(l'.~l~IIJ
DIMENSION XAIIJ,YA(lJ.IA(J),NIN(lJ,T~ll)
c
COMMON/VOISI/NIMB,XB,VB,Ze,IJVl
c
"X=NX'tZC 1)
N't=NX'tlC2J
"'Z=NXYlI3J
NA"'O
IA=O
J::zQ
NINCl)=l
N=O
c
C NEW G~OUPING CONSIOEREC
C
00 1 Jlal.Nl
NB=NPHLI
IA=NA.l
t"A=O.
TACI.&'=O.
c NEW BLOCK CF l~(S GROLPI~G
DC 2 JII':l,NB
N~N+ 1
Nl=3.1I\-U
IO=JLCll+IPINl+!)
JO=JL(2).IPINlt2J
lO=JL(3J+IPtNl+3J
l8=IPC~1+3)+IJVl
IF(IO*(NY+I-IOJ.lE.OlGC lC 2
IFCJO*(NX+I-JCI.lE.OtGC lC 2
,_lJJi ...-'..J
t...... ' J ,'~, ~ ~
'#NNtI.minebook.blogfa.com .- . . . - ..,............
382 MINING GEOSTATISTICS
IF(LO*CNZ+I-lO).LE.O)GC lC 2
IJL*IO+NV*CJO-l)+NX*NY*(LB-lJ
I A0" NFe ( I J U
IF(IAC.EC.OJGC TO 2
IDA=9*CNUMCIJLJ-ll*NIMB
CO 3 I"l,IAC
c
C READ t~TA ON FILE IE AT ~CCRESS IC~ lh WCROS
C
READ CIse IEt]CAINUS.XC,'C.ZO,~f.YFtZf.HLtl
c
IF(T.lE.TEST)GO TO 3
J:J+l
XA(J)=~D+(XF-XC)/2.-XB
YAIJ)*YO+IYf-YCl/2.-VB
ZAtJ)=IC+(lf-lC)/2.-Z8
HA""'HA ..... L
TA(IA)=TA{lA)+T*~l
IOA-I[A+C;
3 CONT I"UE
2 CONTINue
IFCHA.EO.OlGO TO 1
C G~OUPING kITH tATA
NINIIA+l)-aJ+l
TAU ,I TAIIAI/HA
NA-NA+l
1 CONTltoiUE
FlETURN
END
C
c
c COVARIANCE F~~CTION
c DEFINED AS ThE su~ OF EXPO~EhTIAL CR SPHERICAL
c STRUCTURES W~IC~ ARE ISOTRCPIC AFTER LINEAR
c TRANSFORMATION Of THE INITIAL COOROI~A1E~ CF
c T~E VECTOR (XI-X2.YI-Y2,Zl-Z~I
c
c COMMONS ISTRuel
c NST NU~8ER OF ELEMENTARY SlFllCTURES
c co NUGGET EFFECT
c C(NST) SILLS
c AACNST RANGES <~. eXPONENTI_l
c >0. SPHERICAL
c C~X(NST,3)tCA~(NS1.3 CAZ{N~1,3) llNeA~
c TRANSFOR~ATICN OF ]~ITIAl CCCJCJ~ATE$
c
c
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 383
t As it is REITER works only for simple kriging systems (without the condition La A.. == 1).
NNI.minebook.blogfa.com
384 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 385
c
C TEST OF ~CCURACY AND G(N\EFGENCE
C
L=O
fIoI1ER=N ITER. IT
CNMU =0.
ODMU =0.
CSIG=C.
DO 6 I:l,M
L=L+ [
Y::XC I)fA IL J-I( I J
CNfllU=CNMUt-V*Y
COMU=CCMU.~.OT(IJ
6 CSIG=(SIG"~*V.A(lJ
ANU= O(~U IOM~ll
C CIVERGI~G $Y~lEM
IF(ABSCAhUJ.lT.I.OIJGO IC ~
DS:DSIG/CANU*(ANU-IJEPSJ
C GIVEN ACClR~(' ACHIEvEC
IFeOS.LT.l.JGD TO 7
IfINITER.GE.20JGO TO 9
n-2
GO TO 2
c
C FINAL SOLUTICN
C
1 ANU=l./IJNU-l.}
l=O
CO 8 l-l,M
L=<L+I
V""'XIIJ/AIl)
8 X'I)~V.(~-l(I))*ANU
~ETUflN
9 JER=-l
~ETUP~
END
WWW.minebook.blogfa.com
386 MINING GEOSTATISTICS
1. Stratiform deposits
In geostatistical applications, the techniques of stratiform deposit are rele-
vant when the structure of the mineralization and the layout of the avail-
able data, in the three-dimensional space, favour one dimension with
respect to the other two. The most characteristic example is a sub-
horizon tal sedimentary deposit sampled by vertical drilling: the vertical
direction is the best known, and it is often the direction of preferential
variability (due to surface leaching by water for example).
The study of these deposits can very often be reduced to a two-dimen-
sional problem, by considering, for instance, the horizontal regionaliza-
tions of variables accumulated over the total vertical thickness.
100 m
51 5, 1% !Ie (% Ni)
Service
varia bles
(Q) ( b)
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 387
The kriging plan The zone that was studied covered 176 squares of the
100 m x 100 m grid, I.e., 17600 small 10 m x 10 m units. It is, of course,
out of the question to solve 17600 kriging systems for each of the 12
studied service variables (three variables for each of the four cut-off
grades). First, note that provided the regular data grid is complete, the data
configuration is the same for all 1OO-m grid squares. This limits the number
of kriging systems to be solved to 100 for each service variable; there are
100 unit panels within each lOO-m grid square, to which correspond 100
kriging systems with a unique first member matrix [K] and 100 second
member matrices [M2] (v), v = 1 to 100. Furthermore, because of the two
symmetries with respect to the medians and diagonals of the 100-m grid
square, the 100 second member matrices [M2](v) can be reduced to 21, cf.
Fig. V.24. Thus, for each service variable, the problem of solving 17600
kriging systems is reduced to that of solving a single system with 21
different second member matrices. In such a case, kriging proves no more
costly than any of the more traditional estimation methods.
There are several non-preferential gaps in the 100-m data grid due to
nine accidental breaks in the drilling. These gaps were filled in, before
kriging was undertaken, by simply taking the arithmetic mean of the eight
data values surrounding each of them on the 100-m grid, cf. kriging of gaps
'#NNI.minebook.blogfa.com
388 MINING GEOSTATISTICS
in section V.B.2. All 100 unit panels within the 100-m grid square were
estimated considering the 16 data {Za, a = 1 to 16} of the first and second
aureola, d. Fig, V.24. The kriging estimator is thus a linear combination of
these 16 data,
16
Z*(V)= I Aa(v).ZO'
0'=1
for each of the 100 unit panels (v = 1 to 100). The dimension of the
stationary kriging system is, thus, 17 (16 weights A", plus the Lagrange
parameter J.L).
e @ <XI @
4 8 12 16
7 II 15
@
.. III .. ...
'"
r&i~- ~.-.7"'~6"
5 .. "43 111 ...
4!I /34
/ .
~i ./.:2.~ II .
2i/I~ ..
/'4
I~ ill @ ~
6 " 21 31 41 5' 61 71 8' 9' 10
~-+-Iom
Avoilo ble data
1
@
100 m
.. I
<Xl <XI
9
/~ lSI
13
1 5
VNNI.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 389
fraction and for the tricalcic phosphate grade z (x) of this fraction. The two
studied regionalized variables are the percentage p(x) and the accumula-
tion a (x) = p (x). z (x). These variables are defined on the constant su p-
port, O 5 m x horizontal section of the pit, and are regionalized in the
three-dimensional space. In addition to the 50-m regular square grid,
data were also available on two diagonal lO-m in-fill grids to detect
micro-structures.
The three-dimensional variograms that were fitted consist of three nest-
ed structures:
y(h". h.. h w ) = Co + 'YI(.J(h~ + h~), h w )+ 'Y2(.Jh~ + h~))
where Co is the nugget effect.
"1 is a structure characterized by a marked geometric anisotropy: the
range in the vertical direction h.; of sedimentation is 1 5 m, while the
isotropic range is 25 m in the horizontal directions (h", h,,).
"Y2 is an isotropic horizontal macro-structure with a 30G-rn range cor-
responding to the horizontal silting of beds. The problem consisted of
kriging a 25 m x 25 m x 05 m panel slice P, d. Fig. V.25. The object of this
kriging was to produce a vertical mean grade curve for each mining unit of
25 ill x 25 m horizontal dimensions. Cut-off grades would then be applied
to this curve so as to define optimal roof-wall profiles, taking into account
various constraints such as ore quality control and accessibility to mining
equipment.
The kriging plan was essentially determined by the structure of the
three-dimensional model l' adopted.
Only the structure 1'1 differentiates the vertical distance h ; and, there-
fore, the DS-m sections in any given pit. The horizontal range of this
structure 'Y1 is 25 m, which is not greater than half the distance between the
pits. Thus, when kriging the slice P which is wedged against the pit 55, it
serves no purpose to differentiate between any of the vertical pit section
other than those of 55' Furthermore, since the vertical range of "II is 15 m,
it is only necessary to distinguish those data in 55 which are less than 1 5 m
from P; in the example shown on Fig. V.25, only the first seven sections of
pit 55 should be distinguished.
As the horizontal structure "2is isotropic, the data from 52 and 56, 51
and 59, 54 and 58 can be grouped together because of their symmetrical
locations in relation to P. It was shown that, in regard to the structure "2,
the first aureole (pits 5 10 52, 53, 54, 56, 57, 58, 59) forms an almost total
screen again the influence of more distant data.
Eight different weights were distinguished in the kriging plan, as shown
on Fig. V.26: the five weights {3t, {32, {33, {34, /35 for the eight pits in the first
aureole (grouped into five sets, 53, 5 2+56 , 51 +59,54+58,57 ) ; and three
WWW.minebook.blogfa.com
390 MINING GEOSTATISTICS
weights, al, a2, a3, for the vertical data of pit 55 which runs down on one
side of the panel slice P.
Of course, if any of the data corresponding to the weights f3i or Cj are
missing, the dimension of the kriging system will be reduced accordingly.
The eight weights resulting from kriging the accumulation variable for
the particular configuration considered are also given on Fig. V.26. Note
that the value a 1 = 011 is significantly different from the value 1 which
$, 52 53
r-
, ~
A $4 I 56 8
~-- ---+-
I
55 I
-- _..J
57 56 $9
..
5Cm
~
8
A
,
2
3
4
t
5
6
7
8
9
50 em
would have been obtained if the value of each vertical section had been
assigned to the corresponding slice P (polygons of influence method). Note
also the significant influence of the eight pits of the first aureole, L f3 =
067; kriging of slices P results in little vertical differentiation. Although
kriging causes a smoothing effect (d. Chapter VI, relation (VI. 1, it is
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 391
al'II%,
a2' 12a 0/0
113'9-3%
A 54
_.--...c--------5r:-6 B
2. Massive deposits
In geostatistical applications, a massive deposit is one in which the
geometry of the mineralized mass has no preferential direction of
'#NNI.minebook.blogfa.com
392 MINING GEOSTATISTICS
'Vr E [2,60 m]
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 393
Figure V.27 shows the weights of each core as a function of the distance
to the median horizontal plane of the block to be estimated. Note the
following:
(i) the rapid decrease of the curve 51;
(ii) the screen effect of cores 52, 54 on 53, which means that enlarging
the estimation neighbourhood would serve no purpose.
(iii) the relative importance of the weight assigned to the global mean
grade, Am =0143, which is due to the presence of the relative
nugget effect Csl C, = 028.
12 r-------~-------__.
10
o I---i--+-'--~----~--~---"' 12 54
10
-2
8
-4
6
2 I'" r-,
I
- __ .1_
-
f\
~
I
-2
-, I
I
-4 K "
-6
Weight (%)
FIG. V.27. Donoso's kriging plan. Global mean grade weight = 0143.
'#NNI.minebook.blogfa.com
394 MINING GEOSTATISTICS
(i) Drilling in all directions. Drill cores are cut and analysed on l'5-m
lengths. These samples are used essentially for long term estimation.
(ii) Vertical blast-holes analysed over the thickness of l3-m benches.
These samples are used for short-term estimation of mining units.
The regularized variograrns calculated from the lS-m core lengths were
compared to the horizontal variograms graded over the constant 13-m
bench thickness calculated from the blast-hole data. The three-dimen-
sional regionalization of the copper grades, with the l'5-m support assim-
ilated to a point, is characterized by a model of anisotropic, nested
variograms, specifying two horizontal directions and the vertical direction.
The presence of a proportional effect means that neighbourhood of quasi-
stationarity must be considered.
For kriging purposes, the blast-hole data are grouped together inside
each of the nine neighbouring blocks already mined, d. Fig. V.28. Since
the number and exact arrangement of the blast-holes within each of these
blocks are variable, it is assumed when calculating the mean values 'Y of the
kriging system that each block contains, on average, six blast-holes on a
regular 6 m x 10 m grid, the grid being positioned in a uniform random
manner within the block, cf. Fig. V.29(a).
~~-
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 395
Thus the terms Yea, f3) between two blocks a and (3 for which data is
available, each with six blast-holes, are approximated by the mean value
1 6 6
-yea, (3)= 36 i~l i~l Y(Xi -Xj).
x x x x x X
I to 30
x x x k ={ or
x x x I to 6
Block a Block JJ
; : I to 6 j : I to 6
(a ) ( b)
FIG. V.29. Discrete approximation of the blocks (a) blocks for which data is
known; (b) block V to be estimated.
If the object is to estimate the mean grade of the entire block V, then the
terms Ha, V) are calculated using a discrete approximation of V consist-
ing of 30 blast-holes on a regular centred grid within V, d. Fig. V.29(b),
On the other hand, if the object is to estimate the arithmetic mean of the
six blast-holes within V, then the terms yea, V) are calculated using a
discrete approximation of V consisting of six blast-holes on a regular
6 m x 10 m grid positioned at random within V:
1 6
Yea, V) ="6 j~l ii(x;, V) with y(x;, V) =
Using the blast-hole data file from the benches already mined, the
experimental arithmetic mean zt
of the blast-holes within V can be
calculated and compared with the kriging estimator Z ~ corresponding to
the discrete approximation (k = 1 to 6). Note that the same experimental
value zt. can be compared to two kriging estimators Z:<'w and Z~E
depending on whether the block V is considered as coming from the
Western front (the data a are then located East of the block) or from the
Eastern front (the data a are located West of the block), The comparisons
NNI.minebook.blogfa.com
396 MINING GEOSTATISTICS
(Z~w, zt) and (Z~E, zt) are given by the histograms of errors (zt-
Z~w ), and (zt - Z~E) on Fig. V.30.
If the results corresponding to the East and West data configurations are
grouped together, the following remarkable verifications can be made:
(1) the non-bias of kriging, the mean error being 000% Cu;
(ii) the representativity of the kriging variance, (T~ = 0117, which IS
very close to the experimental variance of 0118.
W front E front
01
>.
<J
c
'"
::I
c;
~ 005
001
-05 0 05
% ell
FIG_ V.30. Histograms of errors for the stationary case in which the kriging system
is of dimension 10.
For the Western front, mean error = ~011 % Cu; experimental estimation
variance = 0-125 COlo Cu)"; number of values=410. For the Eastern front, mean
error = +011 o Cu; experimental estimation variance = 0111 (% Cu?; number of
values = 400.
Grouping the two fronts, mean error = O 00 % Cu; number of values = 810;
experimental estimation variance = 0-118 (% Cul; kriging variance O'~ = 0-117
(% Cu?; mean grade of the considered zone = 2% Cu.
On the other hand, if a distinction is made between the East and West
configurations, there is a significant bias of modulus 011 % Cu, the sign of
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 397
w E
'Moil1
fault
The solution consists in taking this local E-W drift into account in the
short-term kriging system, which means introducing more non-bias condi-
tions than the single condition,
9
L Aa = 1,
a=1
presen t for the stationary case. The variogram model 'Y remains
unchanged, and a linear drift is considered which, in the three-dimensional
space, amounts to introducing three additional non-bias conditions:
9
L Aau a = Uv,
a=l
where (u"", Va, wa ) and (Uy, vv, Wy) are the three-dimensional coordinates
'#NNI.minebook.blogfa.com
398 MINING GEOSTATISTICS
of the centre of gravity of the block a for which data is known and the
block V to be estimated, d. relation (V.lO) in section V.A.2.
Using the same data file coming from the same already-mined benches
considered above, the comparison of the kriging estimates Z~w, Z~E and
the experimental values zt
yields (cf. Fig. V.32):
(i) an almost null bias, whether the East or West data configuration is
used;
(ii) an excellent agreement between the kriging variance ai = 0261
and the experimental estimation variance. Note that the value 0261
is greater than the value 0117 previously obtained for the stationary
W front E front
0-1
>.
u
c:
'"
J
cr
Lt'" 0-05
0-01
-05 0 0-5
% Cu
FIG. V,32. Histograms of errors for the non-stationary case in which the kriging
system is of dimension 13,
For the Western front, mean error = -0,02% Cu, experimental estimation
variance = 0247 (% CU)2, number of values = 397. For the Eastern front, mean
error = +001 % Cu; experimental estimation variance = 0273 (% CU)2; number of
values = 397.
Grouping the two fronts, mean error = 0,00% Cu; number of values = 794;
experimental estimation variance = 0261 (% Cuf, kriging variance. ui = 0261
(% CU)2; mean grade of the considered zone = 2% Cu.
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 399
60 m
'#NNI.minebook.blogfa.com
400 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 401
within this square); a random kriging plan was adopted for this estimation
of 100 m x 100 m x 15 m blocks. The estimation neighbourhood consisted
of 27 blocks: the block B to be estimated and the 26 blocks of the first
aureole surrounding it; cf. Fig. V.34. These 27 blocks were divided into six
domains {Dcl'; a = 1 to 6} within which it was assumed that the data were
randomly located within each 100 m x 100 m square.
D} is the domain consisting of the block B to be estimated; D 2 and D 3
are the blocks located above and below B, respectively; D 4 is made up of
the eight blocks of the first aureole surrounding B on the same level; Ds
and D 6 consist of the eight blocks located, respectively, above and below
the eight blocks of D 4
The data inside each domain D a consist of no; drill intersections, all
vertical, 15 rn long and of mean grade Zo;o In the general case, the six
domains Do; will all contain some data (no; > 0, Va), and the kriging
estimator will be the linear combination of the six mean grades Zo;:
6
Z~ = I
0; = 1
x.z:
The stationary kriging system is then written as
6
I A{{y(Du , D(3) + f.L = y(D at , B), Va = 1 to 6,
,B=!
I A,a = 1.
{3
4 4 4
t 4 4
100m
Lf------...---"~4__~
4 4 4
~I-p~-
5 5
4 4
6 6
15m
FIG. V.34. Six domains with random positioning of the data.
'#NNI.minebook.blogfa.com
402 MINING GEOSTATISTICS
As the three intersections S1, S2~ 53 come from the same vertical drill
intersecting the block B, cf. Fig. V.35, then y(D lo D z) = y(Slo Sz), and
similarly for the other two rectangular terms y(Dl~ D 3 ) and y(D z, D 3 ) .
This relation holds, strictly, only when there is a single vertical drill inter-
secting each block B. For this reason, only one intersection per block was
retained in the classifying program of the data (nl; nz. n3 ~ 1); in any case,
only rarely did more than one drill intersect a block.
$2
$1
$'3
For the calculation of the other rectangular terms y(Da , D(3), a :-13~ the
hypothesis of random locations was used. Thus, for example y(Dl~ D 4 ) is
the arithmetic mean of the eight mean values 1(B, B k ) , k = 1 to 8, between
the block B = D 1 and the eight blocks Bk of the first aureole that make up
D4
The diagonal terms are
Similarly,
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 403
the second member matrix. Thus, for example, -y(D j , B)= y(8, B) and
1 8
y(D 4 , B) ="8 "~1 y(Bk' B).
'dr=lhl>A
VtIWW.minebook.blogfa.com
I
~.
300m
I-
,
e
I- / /
/_---~reOle
----- ---
J /
/
n /
S 0-"- /
/
100m
0"'"
")0
,"- /
'i;>-~/
Block B
/
// a<$' Zx= 172
/ ,0
o-}=O-006%2
/
/
/
I A,o:; 33, z. 2-IBo/o, ,I' 21
I A,05-01, Z ol-B3 % , ,, =19
\ A~"605, Z 02-I?o/0,,,=19
A7 6-93, Z, 2 20 %
0 , ,, ' 17
AgO 7 83, Zo 1-47 % , no 15 /
\\' 2
All' 10-31,,2'01'46%, n> 15
AI~'7-77, Z.j-23 %
v'/
":>.,'//
1i,o/
/
'"' 4 '1-54%,n 08 /
V E
0
Q
5
G
~jBo~;:'/7 ~~
A
1/
/
? ___ ..... : I
1- 8
9
~~~e
I
1: 10
J
FIG. V.36. El Salvador's kriging plan. n = number of core samples within each
slice, Z = mean grade of each slice, A = weight in %.
V THE ESTIMATION OF IN SITU RESOURCES 405
(iii) the set of channel samples taken along the drifts which run along the
base of block B.
This provides a total of 21 weights for the 21 data sets {Om a = 1 to 21},
and a stationary kriging system of dimension 22. As the results of this
kriging were to be used for production planning, it was essential that they
be as accurate as possible. For this reason, the number and exact location
of each in tersection S, constituting the cluster G, were taken into accoun t
when calculating the terms ii( Go:., G/3) rather than assuming random loca-
tions of the clusters Go: in their respective domains:
Similarly,
WWW.minebook.blogfa.com
406 MINING GEOSTATISTICS
_ - -=:-- -.J:.!91l!-
of the vein
-I I
I _ _ _ ____
~l:-
I I
I I
I" "I
x x+h
Data for this study consist of vein width and copper assay values taken at
approximately 10ft intervals along three main horizontal development
drifts located in the plane of the vein. The two regionalized variables used
in the study were the vein width and the corresponding copper accumula-
tion.
Nested structures, y(h)=Yl(h)+Y2(h), were fitted to the variograms of
these two regionalizations. The two structures are spherical with respective
ranges for the width of a 1 = 20 ft and a2 = 100 ft. These two ranges can be
interpreted as pseudoperiods of the successive bulges in the vein, as shown
by the thickness profile obtained on one of the development levels d. Fig.
V.38.
WWW.minebook.blogfa.com
V THE ESTIMA nON OF IN SITU RESOURCES 407
In the plane of the vein, the three levels are about 600 ft apart, which is a
much greater distance than the largest range az = 100 ft. Thus, local esti-
mations of ore resources can be made only within 100 m of workings for
which samples are available, i.e., the zone of influence of the levels. This
Smoll st ructure
, Lorge s truer ure Small structure
.
------
---._- ----- --- ------- .. --
----
... ---
.
1200 ft
Le~el
6930
~~-l-4-+-1-4-44-l--I-4-1-~~-+-+-+-+-++-+-+-+-t-1~=
Le~e~
6360
The three vertical bands of influence are split into panels of vertical
dimension 100 ft x 200 ft. For the estimation by kriging of the mean
horizontal thickness and the corresponding mean copper accumulation of
WWW.minebook.blogfa.com
408 MINING GEOSTATISTICS
each panel, the information on the central level is grouped into three data
sets, S1' 52, 53, d. Fig. VAO. 51 is the data set within the panel to be
estimated; 52 and S3 are the next data sets taken along the central level.
Existing
drift
100 ft
FIG. VAO. Data configuration for the estimation of a 100 ftx 200 ft panel.
As the mining company planned to carry out additional sampling for the
local estimation of all the mining panels, several kriging plans were
established to predict the improvement to be gained by different additional
sampling plans, d. Fig. VAL It was shown that, with an additional level
between each of the existing ones, the relative standard deviation 0"/ m, of
the global estimation of the ore-tonnage, would be reduced from 48% to
23%, the relative standard deviation (TKI m, of the local kriging of the
ore-tonnage of 100 ft x 200 ft unit panels, would be reduced from 17% to
15%. This latter decrease is not especially significant; however, all the
units in the deposit could be differentiated by kriging, while, with the three
existing levels, only 40% of the panels were differentiated (had different
G1
--If-----------+--
200
II
(a) ( b)
FIG. VAL Kriging plans with additional data. (a) Two additional drifts; (b) one
additional vertical raise.
VNNI.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 409
kriging estimates). Similar results were obtained for the estimation of the
quantity of metal and the mean copper grade.
Drift d
dZ dl dZ
~~~
Raise Block r
[50m r
I I I 1
I
I Block I
I
t
20 m
I.~---~I
40m
d d
50 m
~
WWW.minebook.blogfa.com
410 MINING GEOSTATISTICS
has a physical meaning, and the local estimation variance E {[Zv; - zti f}
of a block V was then defined. In global estimation, the deposit D is
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 411
unique as well as its mean grade ZD. The probabilizationt of the global
error (zv - zt) is obtained, no longer by considering various particular
deposits DJ, but by considering various locations of the data configuration
as a whole over the unique and fixed deposit D: a particular global
estimate z~ corresponds to each particular position i of the origin of this
data configuration, cf. Fig. V.54, the example of two surface estimates
S*(xo) and S*(xo) linked to two particular locations Xo and xb of the data
grid origin. Thus, the global estimation error (z D - Z 0) has been prob-
abilized, and the global estimation variance is defined as E{[ZD - zt, ]2}.
However, the distribution of the error (ZD - z}b) will always remain
inaccessible since, in practical applications, only one particular location io
of the data configuration over D is available. Hence, all confidence inter-
vals (e.g., the 20" Gaussian standard) attached to the global estimation
variance will have but a mere conventional meaning.
Choice of the structural model y(h) or C(h), used in the calculation of the
global estimation variance
Consider the "local variogram" on D, i.e., the variogram 2 Y{D)(h) that
would be available if the domain or deposit D were perfectly known, ct.
section III.B. 7, formula (111.40):
where N' is the number of pairs of data (distant of vector h) available over
the whole domain D.
As this experimental variograrn 2y'b(h), and, thus, its fitted model
2YD(h) are defined over the whole deposit D, they may differ from the
quasi-stationary models 'Yi(h) used in the local estimation of each partic-
ular zone or block Vi of the deposit D. Often, in mining practice, these
t Note The global estimation error probabilization as well as the conceptual basis of the
global estimation variance calculation is related to an approach different from the approach
of regionalized variables considered in this book. This approach is called "transitive
theory", cf. G. Matheron (1965, p. 17) and (1971, p. 9), and results in calculation algorithms
and solutions remarkably similar to those provided by the regionalized variables theory. In
the present text, this transitive theory approach is briefly outlined in section V.CA, which is
devoted to the surface or volume estimation problem.
NNI.minebook.blogfa.com
412 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 413
Regular grid Consider the regular square grid of side I shown on Fig.
VA3. The surface S of the deposit is divided into N elementary grid
surfaces s; each of which has a positive central drill-hole (i.e., with a
.
non-zero mineralized thickness Z (Xi The mean thickness ZS; of each of
these elementary units is estimated by kriging from the thicknesses
measured 011 the central drill-hole and, for example, the eight holes of the
first aureole surrounding s; If the deposit is assimilated to the surface S of
known geometry, then the estimator of the mean thickness Zs of the
deposit is the average of the N local kriging estimators, i.e., = z;
(lIN)IiZ~,
Any two units s, and s, which are separated by a distance less than 3! will
have common drill-holes in their estimation neighbourhood and, as a
consequence, the two estimation errors [ZSI - Z:'l and [ZSj - Z~] cannot
WWW.minebook.blogfa.com
414 MINING GEOSTATISTICS
E {[ Zs-Zs]
* 2} 1
=N 2 ,2:N 1 2
E{fZsi -Z(x,)] 2 }=-<TEs (V.3?)
,~1 N
..
with 01s = 2H(lj2; 1/2)- F(l; l), according to the formula for cor- ut
responding to Fig. II.50 in Section II.E.3.
Note that this sum of elementary independent errors, Zs, - Z (Xi)' is
related to the approximation
Zs* =-IZs;
1 * =N,-Z(Xi)'
1"
N i j
X
W W
Xi
X
x
X
Xj
X
x
X
L-_~~~ _______
t The N elementary errors [Z.. - Z (Xi)J are independent but only as a first approximation,
although they do not involve any common data. This is because the spatial regionalization
entails a certain degree of correlation between the values ZS!' ZSj and their estimators Z(Xi),
Z(Xi)' However, formula (V.37) provides a good approximation for the estimator variance. as
will be shown later. in the addendum, on Figure V.47.
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 415
and E{[Zs - Z s]
* 2
=
1 2
N 2 ~ O"Es;'
In this case, each of the elementary errors [ZSi - Z (x;)] has its own
estimation variance O"~s, which depends on the particular position Xi of the
drill-hole in the grid square 5j. Over a large number N of such grid squares,
x will take all possible locations inside the grid square s and the elementary
estimation variances at will have the foIlowing mean value:
1 N
N i~l
2
O"Es j
If
= --; E{[Zs - Z (x)]
S
2
} dx = D (01 s).
2
Indeed, from formula (11.33), the mean value over s of the variance of
extension E{[Zs -Z(x)f} of a point datum Z(x) to the mean value Z; is,
2
by definition, the dispersion variance D (Ol s ) of this point datum in s. The
global estimation variance is then written as
*2 1 2 1 2
E{[Zs-Zs]}= N2~(jES' = N D (Ols). (V.38)
x
x x x
sI
X
X X
xI X
X X
X
X
NNI.minebook.blogfa.com
416 MINING GEOSTATISTICS
(V.39)
Irregular grid Consider the same sedimentary deposit, this time with the
N positive data on an irregular grid, d. Fig. V.4S (the N surfaces of
influence s; are unequal). Note that under no circumstances can a pref-
erential grid, with all its associated risks of bias, be assimilated to an
irregular grid, cf. section III.C.l.
The global estimator Z~ is the weighted sum of a certain number of
local estimators obtained, for instance, by local kriging; note that these
local estimators may not be defined on the surface of influence s, of each
positive hole. By expressing the global estimation error [Zs - Z~ J as the
sum of the N independent elementary errors [ZSi - Z (Xi)], of extending the
thicknesses Z (x;) to their respective surface of influence Si, the following
formula is obtained:
(VAG)
Zs* = S
1 ~ SiZs;* = S1 ~ SiZ(Xi),
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 417
x x
(VAl)
with
ukl = E{[ZI - Z (x )}2} = 2X(112) - F(l).
The same line L = NI may be sampled by (N + 1) points in a closed
arrangement, each segment l, having two extreme data Z(Xi) and Z(Xi+l)'
as shown on Fig. V.46(b). Two contiguous segments I, and l, have one
datum in common and, as a consequence, the two corresponding local
WWW.minebook.blogfa.com
418 MINING GEOSTATISTICS
( a )
\
Z(x)
ii Ii
~ ...-----------.......
II II II' K ( b)
I ,
Z(Xi) Z(Xi+ l)
FIG. V.46. Direct combining of elementary errors on a line. (a) Centred arrange-
ment; (b) closed arrangement.
(V.42)
where L = L. l, and
ukl, = E{[Zjj - Z (Xi)]2} = 2X(//2) - F(li)'
Addendum
On this simple example of centred arrangement in one dimension, it is advisable to
test the robustness of the approximation underlying formula (V.41), i.e., the
influence of the hypothesis of independence of the N elementary errors of extend-
ing the central data to their segments f of influence. For this, a spherical scheme
with range a and unit sill and without nugget effect, was considered, d. formula
(IIL1S). The estimation variance of a line L =NI surveyed by N point data
regularly spaced on a central arrangement is then calculated from the exact formula
(II.27) and compared with the approximation given by formula (VAl). These
compared results are shown on Fig. V.47 (by courtesy of Michel David) for various
values of N and the quotient lJ a. The approximation is excellent for all small values
of the spacing (lla < 05, which is a common case in practice); for greater values of
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 419
the spacing (if a> 0-5, the number N must be great (N > 30) to provide a good
enough approximation.
007
/
006 /
I
"- /
~" I
005 /
/
/
,'\,
004 ~'
NlI)
~
003
002
0-01
--
01 0-2 0-3 0-4 05 0-6 0-7 08 0-9 ro
110
I I
1 I
~----------------~
I SI Li I
I I
~ I 0
I I
~ - - - - - - - - - - - - - - - __ I
I I
I !
1 J
FrG. Y.48. Direct combination of elementary errors from a configuration of
median drifts.
NNI.minebook.blogfa.com
420 MINING GEOSTATISTICS
where
-,+--5,.
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 421
(V.44)
Note In the last two examples, corresponding to Fig. VAS and V.49, the
elementary extensions of the mean grade of a median drift or a median
level to its rectangle or parallelepiped of influence were considered. But
both the median drift and the median level have been estimated with a
certain estimation error. Thus, a term corresponding to this estimation
error must be added to the estimation variances given by formulae (V.43)
and (V.44); this results in combining the various terms of the global
estimation error.
L;
- Level 2
-lev~1 3 Drift
"
Chennel
scmples
WWW.minebook.blogfa.com
422 MINING GEOSTA TISTICS
(i) The estimation of the drifts by the channel samples. The estimation
variance corresponding to this step is called the "line term", T/.
(ii) The estimation of the levels by the drifts which are supposed to be
perfectly known. The estimation variance corresponding to this step
is called the "section term", T s .
(iii) The estimation of the slices of the mass by their median levels,
which are supposed to be perfectly known. The estimation variance
corresponding to this last step is called the "slice term" , Tv.
As a first approximation, the three estimation errors corresponding to
these three steps can be considered as independent, and the global estima-
tion variance of the mass then consists of the sum of the three preceding
terms:
(VAS)
The following example of the calculation of these three terms is based on
Fig. V.50. Another example is given in the study of Ity Mount Flotouo gold
deposit in section V.C.S.
2 12
(TEL, = N- (TEl>
r
2 1 2
T 1 = (TEL = N (TEl
Section term The three levels are assimilated to the sum of n rectangles
s, = hi X L; each estimated by a median drift L, supposed to be perfectly
known. The estimation variance of these three sections (of a total surface
S = I~=l 5,), known as the section term, is then obtained from the direct
combining formula (V.43):
2 1 2 2
T, = (rES = 52 ~ S i (TE,-"
where (Ttl is the variance of extension of the median L, to its rectangle of
influence 51 = hi XLi.
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 423
Slice term The final term is the error of estimation of the three slices of
the mass by their median levels supposed to be perfectly known. Each slice
may, for example, be assimilated to a rectangular parallelepiped estimated
by its median level Si, and the slice term is then obtained from the direct
combining formula (V.44):
where
Section
A B
\v
Vertical
drills
WWW.minebook.blogfa.com
424 MINING GEOSTA TISTICS
z, = l/S
I
Ssa(x)dx.
f ( estimated by
* A~
Zs =~.
1 S stx)dx Ts
Similarly, if the mineralized surface S is estimated by S*, the estimator
of the global volume of ore is in the form of the product of the two global
estimators S* and T~:
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 425
P" -<Prr e;
NNI.minebook.blogfa.com
426 MINING GEOSTATISTICS
Putting PAB = (E{[A - A *][B - B*]}/O"A . 0"8) for the correlation coefficient
of the two estimation errors of the two factors, the formula can be written
as
(V.46)
Q *= ( Q+-
A)
B
B e 1 E 1
. [ 1--+-..,--+
B 3 ...
B- B
J (VA?)
and
(0*) = o[ 1+ cr~ -
B
E{EAEB} + ...
AB
J.
There is, thus, a relative bias, ((T~I B2)-(E{fAC8}! AB)+ . . " .which persists
even when the two estimations are independent, i.e., even when E{cAB} =
E{A} . E {fa} = O. In mining applications, this relative bias is negligible
when the relative estimation variance 0"1/ B 2 is less than 0.01, which is
often the case in global estimation. Using (V.4?), the relative estimation
variance of the quotient Q can be expanded up to the second order as
crb _ EUG - Q*]2} _ (T~ + 0'1_ 2 EHA - A *][8 - B*]} +
Q2 - Q2 - A 2 B2 AB " .. ,
i.e.,
(V.48)
Note first that if the two estimations of the two factors A and Bare
independent, the covariance EHA - A *][B - B*]} of the two errors is null,
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 427
and the approximate formulae (V.46) and (VAS) reduce to the standard
formula of addition of relative variances:
2 2 2 2
(TQ crp (T A (TB
Q2 or p2 = A 2 + B2 + .... (V.49)
If the two variables a(x) and b(x) are intrinsically coregionalized, i.e., if
the two structural functions Ya(h) and Yb (h) are proportional, d. section
lILB.3, formula (IIL30), and if the two configurations for the estimation of
their mean values A and B are identical (similar in practice), then it can be
shown that the coefficient PAB is equal to the standard correlation
coefficient pal> between the two variables a (x) and b (x).
If the two variables a (x) and b (x) are not intrinsically coregionalized,
then the covariance of the two errors E{[A - A *][B - B*]} must be cal-
culated, i.e., the coefficient PAB is calculated from the cross-structural
function
Yab(h) = !E{[a(x + h)- a(x )][b(x + h)- b(x)]}.
When the two configurations for the estimation of the two factors A and
B are identical (or very similar), a "reduced difference" variable can be
defined, d (x) = a (x)1 A - b (x)1 B. Its structural function can then be
determined:
= 2Ya(h)
2
2Yb(h)
+--..-".-.-
2
4Yab(h)
A B AB'
If the two configurations for the estimations of the two factors A and B
are supposed to be identical, then the two estimation variances (T~ and a1
are obtained by applying the same linear operator g to the two structural
models Ya and Yb, respectively, cf. formula (111.2):
(T~ =E{[A-A*]2}= g'(Ya) and a1 =E{[B-B*]2}= %'(Yb);
similarly,
E{[A - A*][B - B*]} = gCYab).
The relative estimation variance of the quotient AI B can then be written
as
ub 1 1 2
Q2= A 2 g'(Ya) + 2 ?E'(Yb)- ABg(Yab)+ ...
B (V.50)
= g'(Yd) + ....
WWW.minebook.blogfa.com
428 MINING GEOSTATISTICS
a~ 1 1 2
p2 = A 2 g (Ya) + B 2 ~n Yb) + AB g (Yab) + ...
(V.51)
= g(y,,)+ ....
of the volumes of these blocks. In doing so, an error (V - V*) was incurred
and must be evaluated, for example, through a variance of the geometric
error. This geometric error (V - V*) not only affects the volume and the
are tonnage, but also the estimations of metal tonnage and mean grade of
the deposit D. Thus, it is equally important to combine the influence of the
geometric error with that of the different quality errors, e.g., the quality
error due to the regionalization of the grade variables in the total volume
V.
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 429
The transitive approach Let k(x) be the indicator variable of the mineral-
ized volume V to be estimated, This indicator is an all-or-nothing variable
defined as
{~
if the point x is in V,
k(x) = (V.52)
if not, xe V.
The mineralized volume V is then the triple integral, III the three-
dimensional space, of this indicator variable:
f -e-co f
+00 +00
LXl
V = k(x) dx = f
-co dr, -co dx, k(x", xv, x w ) dx w . (V.53)
WWW.minebook.blogfa.com
430 MINING GEOSTATISTICS
Pu. and P being two integers defining the various nodes of the regular grid.
If n(xo) is the number of positive drills, the estimator S*(xo) is then written
as S*(xo) = a laz . n (xo). Note that this estimator depends on the choice of
the origin within the elementary grid surface alaZ, ct. Fig. V.53.
0 0 o o o o o
0 0 x x o
0 0 x x x o
,--
I
0
I x x )( o o
L_
I
0 I x x x I o o
_____________ I
a2(~ 0
o o o o o
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 431
For each new position xoof the grid, there is a corresponding estimator
S*(x&) and, thus, a new realization [S - S*(x&)] of the estimation error of
the surface S, d. Fig. V.54. One possible way of defining the quality of the
estimator S* is to take the mean value of the quadratic error [5 ~ S*(XO)]2
when the origin Xo of the grid takes all possible locations inside the
elementary grid surface alQ2. This mean quadratic error is expressed as
Note the similarity of this expression for the mean quadratic error to that
of the estimation variance E {fS - S* ]2}. Using this similarity, even though
no probabilistic interpretation is involved in this "transitive" definition, the
notation O'~ for the estimation variance can be used, i.e., in one-dimen-
sional notation:
2 1
O'E(a)=-
Q
J
a
' ] 2 dxo.
[S-S>i'(xo) (V.55)
This mean quadratic error, called the "variance of the geometric error", is
a function of the available data grid; here, a = (a 1, a2).
If S*(Xo) is replaced in the integral of (V.55) by its one-dimensional
expression
S*(xo)=a L k(xo+Pa),
p=-oo
then, after some calculations, the following formula is obtained:
+00 J+co
(T~(a)=a p='2:. K(pa)~ -00 K(h)dh. (V.56)
oo
The variance of the geometric error linked to the grid spacing a can be
expressed uniquely with the geometric covariogram K (h) of the surface 5
(or volume V) to be estimated. However, this covariogram is just as
unknown as the geometry and surface area of S (or V). It is at this point
that an important property of geometric covariograms is introduced, which
allows both the formula (V.56) and the transitive approach to geometric
errors to be of practical use.
Whatever the geometry and the extent of the domain to be estimated (S or
V) are, their geometric couariogram K (h) exhibits a linear behaviour at the
origin:
K(h)=K(O)-w"lhl+ ... , when Ihl~O.
The direction subscript a indicates that this linear behaviour may not be
isotropic.
WWW.minebook.blogfa.com
432 MINING G EOSTATISTI CS
Using this property, the first term in the limited expansion of the
expression (T~ (a) can be evaluated when a tends towards zero. This first
term is sufficient to give the order of magnitude of the geometric error
variance.
For the two-dimensional problem of the estimation of a surface from a
regular, rectangular grid (all az), it can be shown (d. G. Matheron (1965,
p. 108)), that the relative estimation variance can be expanded up to the
first order as
u~ 1 [1
2=-2 6 N Z+ 0 0 6 -
(N 1)2
- + ... , (V.57)
S n . J
where n ~ lOis the number of positive drills (the estimator is then S* =
n . a laZ), and Nz:<:S; N 1 ; 2N 1 and 2N z being the number of elements parallel
to the two sides of the grid (a l , az) which make up the perimeter of the
reunion S* of the n positive zones of influence. The perimeter of any
possible gaps in the mineralization must be taken into account in the
calculation of the values 2N l and 2Nz .
Taking the example of Fig. V.53, with n = 14 positive drills, the values
for the perimeter of the estimated surface S* are:
2N 1 = ten elements of length a 1 parallel to the side a 1 of
the grid;
2N z = eight elements of length az parallel to the side a : of
the grid.
Thus, the order of magnitude of the relative estimation variance of the
surface S is
(T.~1 [4 2S ]
52=1426+0-064 =0005,
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 433
5, the more jagged (i.e., the longer) the perimeter, the more difficult is the
surface estimation.
When a portion of the perimeter is perfectly known (e.g., the limit of a
lease), this portion must not be considered in the calculation of the values
2
2N j and 2N2 , and this will reduce the relative variance (T;,/5 by approxi-
mately the ratio of known perimeter to total perimeter.
---'-.j--+----+-
, ; - 11
-- ---t-+;-
""1'---1'--+-
......--+----'t-""'t---i
'r-
(a ) ( b)
FIG. V.54. Estimation of the same surface from two distinct origins of the data
grid.
(a) n = 37, 2N I = 26, 2N 2 = 22, s* = 37ala2' crl s = 45%, s = (3733)a 1a2'
(b) n = 39, 2N j = 32. 2N 2 = 28, s* = 39aj Q 2 , (T/ S =4 7",{" s = (39 37)alQ2'
WWW.minebook.blogfa.com
434 MINING GEOSTATISTICS
where u, = 1;. S;
I'
--~---~--~-~~-
___________ BJ ~_~ J
5,.
t,
a
FIG. Y.55. Estimations of a volume by combining sections.
L=- I t;
m ;=1
at S*
V = L
2 *2
1 [
n 2 0, 06A
1T 1]
+ 90 A z +, .. , (V.59)
WWW.minebook.blogfa.com
V THE ESTIMAnON OF IN SITU RESOURCES 435
Positive
hole
0 0 0 0 0
.-
1
C12
a
0
0
0
0
D 0
0
0
0
S1f~ 170102
ii
+ + + +
+ + + + +
} /(x)
+ + + + +
oLd + + + + +
+ + +
5*
NNI.minebook.blogfa.com
436 MINING GEOSTATISTICS
Ts~=~l
S s~
t(x)dx,
estimated by T~* ;
(ii) the mean metal accumulation
estimated by A~*;
(iii) the mean grade Zs* = As*jTs*, estimated by zt* = A~*ITt*. (The
relative bias due to the quotient of the factors A~* and T~* is
neglected, d. section V.C.3, formula (V.47).
By means of the study of the two regionalizations, the thickness l(X) and
the accumulation a (x) = l(X) . z (x), defined as the product of the thickness
t(x) by the corresponding grade z (x), the corresponding estimation vari-
ances can be calculated, i.e.,
As =~ I a (x) dx.
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 437
(V.61)
+When the limits of the mineralization are economic limits linked to the quality of the ore
(e.g., cut-off grade), the geometric error and the quality error cannot be considered as being
independent, particularly at the borders of the deposit. But if the extent of these border
zones is negligible with respect to the extent of the deposit (of surface S), an hypothesis of
internal independence can be made as a first approximation.
NNI.minebook.blogfa.com
438 MINING GEOSTATISTICS
(V.63)
where the density d of the are is supposed to be constant and known. If this
density, d(x), is regionalized and sampled along the drill-holes, it should
then be studied by means of a new accumulation variable b (x) =
a(x). d(x).
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 439
r ~
b z " 30 m
x
x
x
) b l ~ 20 m
x x x x x
x
x
CJ
x
x
x
x
x
x
x
x
x
X x x x x x
x x x x x x
x x x x x x
)( x X x x
x x x x x
x x
x x
x )(
)0 2 3 4 5 6
WWW.minebook.blogfa.com
440 MINING GEOSTATISTICS
22 J
2
u1 12 [10
-==-- -+O06~ = 00014
52 58 6 10 '
i.e., a relative standard deviation of 37 %, us/ s = O'037.
VNNI.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 441
with
For a spherical model of range a and sill 1, the auxiliary functions Hand F
are given on Charts no. 3 and 4 in section ILEA:
(T~blb2 = Co + C x 011 = 038 and (T~A = 038/58 = 00066.
Combining terms. In this case, the global estimation variance is cal-
culated as the sum of a line term and section term. The line term (cor-
responding to the N-S lines which have the greatest data density, b, < b 2 )
measures the extension error of the 58 central accumulations to their
segments tb , = 20 m) of influence. This line term is expressed by formula
(VAl) as
The value of the variance of the elementary extension a~bl is read from
Chart no. ? and gives (j"~bl = C o+ C x 0077 = 0.36. (Note the prepon-
derance of the nugget effect Co = 0,30.) The line term is then T1 =
036/58 = 00062.
The section term measures the extension of the mean accumulations of
the lines L;, j = 1 to 6, to their sections of influence (rectangle L, X b 2 ) . The
elementary extension variance of a median line L, to its rectangle of
influence L, x b: is calculated using Chart no. 8, and, for the first line
(L I = 160 m), this gives
a~Llb:l = C x 001 = 00075.
In this case the nugget effect has no influence, d. formula (111.9); this is due
to the fact that the nugget constant COLI' regularized over the length L 1 ,
can be considered as zero because L 1 is much greater than the diameter of
the sampling pit, which is the support on which the constant Co = O 3 was
defined.
NNI.minebook.blogfa.com
442 MINING GEOSTATISTICS
Table V.5 gives the variances (T~Lh as well as the variances weighted by
7,
the square L of their length of influence, for each of the six lines L; The
section term is expressed by formula (VA3):
1 6 2 2
t: = 5*2 j~l u.; b 2 ) . (jEL;b2'
7"L~ 2
I (TEL;b2 1534
Ts = 2 = (1160)2 = 00011.
(~ L j)
J Lj (T~L; X b: Lf(T~Li X b 2
E{(dsAf} = ~~ x D~(O/S),
where (T~/ 52 = 00014 is the relative estimation variance of the surface,
WWW.minebook.blogfa.com
V THE ESTIMATION OF IN SITU RESOURCES 443
t Recall that, since the regionalized variable is the relative accumulation, the calculated
variances are also relative.
WWW.minebook.blogfa.com
VI
Selection and Estimation of
Recoverable Reserves
SUMMARY
This chapter is intended as a study of the influence of the regionalization of the
selection variable on the recovery of the resources of a deposit.
Section VI.A is concerned with the respective influences of the support of the
selection unit (mining selection is made on blocks and not on core samples) and of
the level of available information (in general, selection is made on estimates while
real grades are recovered and treated in the mill).
Section VI.B is a study of hypotheses of permanence of distribution laws, by
means of which one histogram - e.g., the histogram of block grades used for mining
selection - can be deduced from another with a different variance - e.g., the
experimental histogram of core-sample grades.
-~
WWW.minebook.blogfa.com
.
VI SELECTION AND ESTIMATION OF RESERVES 445
WWW.minebook.blogfa.com
446 MINING GEOSTATISTICS
practice, amounts to saying that the sampling of G by the set of drill cores
is unbiased ~ no particular zone of G is preferentially sampled.
WWW.minebottblogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 447
NNI.minebook.blogfa.com
448 MINING GEOSTATISTICS
I
I
I
I
I I
L ~
L __
I..
If the polygonal influence method is used, then the mean grade of the
central drill core S, is attributed to each polygon Vi. Now, if the cut-off
grade Zo is applied to the estimates obtained by this method, the estimated
recovered tonnage will correspond exactly to the hatched area of Fig. VI. 1,
with all the associated risks of bias. The results of the case study given in
section VLA.4 are particularly good illustrations of this point.
Another typical example concerns the method of estimation by inverse
squared distances and, more generally, all estimation methods which do
not take account of the particular geometry of the panel to be estimated.
For a given data configuration, such a method would give the same esti-
mated value regardless of the estimation support v. Accordingly, if selec-
tion is made on these estimated values, the result will not depend on the
size v of the selection unit. Now, it is obvious that mining a deposit with a
small hammer (very selective mining with a very small support c) will yield
results very different from block-caving (the mining unit v being very
large), see P. Switzer and H. Parker (1975).
Within any homogeneously mineralized zone G (at the scale of mining),
geostatistics can be used to calculate the variance of dispersion D 2 (v/ G) of
~~~~~' """"'"
VNNI.minebook.blogfa.com
.
VI SELECTION AND ESTIMATION OF RESERVES 449
the grades Zv of the mining units of support v (d. section ILC.2), either
formally by the general formula (I1.36), D 2(v/G)= y(O, G)-f(v, v), or
experimentally by means of Krige's additivity relation (11.37), D 2 (v/ G) =
D 2(cIG)-D 2(c/v). The two dispersion variances of the core grades in G
and v can be obtained experimentally.
The mean over G of the grades of the units v can be estimated, for
example, by the mean m * of the grades of available core samples. But the
first two moments, mean m and dispersion variance D\v/ G), are not
sufficient to determine the distribution of the grades Zv. It is also necessary
to know the type of function.
(i) In some cases, this distribution may be known a priori; for example,
as a result of mining, or from blast-holes on benches, or from zones
that have already been mined, cr. section II.e.3, the case study of the
block grade distribution at Chuquicamata.
(ii) In other cases, an hypothesis of permanence of law can be postu-
lated, and the shape of the histogram of experimental data Z; can
be adopted for the distribution of Zv. Thus, if a lognormal dis-
tribution can be fitted to the histogram of the Zc, the Z; are
supposed to be lognormally distributed with the variance D 2 (v/ G).
The problems involved in determining one distribution (Zv) from
another (Zc) defined on a different support by means of
permanence-of-law hypotheses will be treated later, in section VLB.
WWW.minebook.blogfa.com
450 MINING GEOSTA TISTICS
Thus, on each level, the pit design defines the accessible surfaces; one then
proceeds to a second stage of selection which consists in distinguishing
waste from ore on the basis of the mining unit.
In this chapter, we limit ourselves to a one-stage selection only, made on
the basis of a mean characteristic (grade, for example) of the unit v,
independent of its location in the deposit or in the zone G. The difficult
problems of estimation of the recoverable reserves after a two-stage selec-
tion are still in the domain of geostatistical research. One approach to these
problems may consist of simulating the deposit and applying a simulation
of the mining process to this numerical model: the influence of tech-
nological conditions on the recovery of reserves can then be observed, d. J.
Deraisme (1977) and section VILB.3.
Smoothing effect
If the estimator Z~ used in the actual selection is a kriging estimator, then
the variances D 2(vjG) and Dk*(vjG) of the true and estimated values can
WWW.minebook.blogfa.com
----------------------------------
VI SELECTION AND ESTIMATION OF RESERVES 451
is the mean value of the kriging variance of a unit v. There are N such units
v in the zone G. If the grid of available data over G is not regular, then the
various local kriging variances O"~Oi will be different. Ii = (11 N) L f-I-; is the
mean value of the Lagrange parameter used in the kriging system of each
unit v, d. system (V.l).
(T~G = E{[ZQ ~ Z'b J2} is the kriging variance of the mean grade ZG over
the zone G. In the case of a complete kriging system, it can be shown (d.
section V.A.3, "Theorem of combining kriged estimates") that the global
kriged estimator Z~G is the average of the N local kriged estimators Z~Vi'
i.e.,
*
ZKG = N1 ~ Z *KV;,
7
The proofs of these two smoothing relations VLI) and (VI.2 are
given in the addendum at the end of this section, VI.A.2.
The smoothing relation (VI.2) implies a smoothing effect: D~* (vi G)~
Indeed, the global kr~ing variance is always less than the mean
2(v/G).
D
local kriging variance: O"~G ~ (TKv' On the other hand, in the local kriging
WWW.minebook.blogfa.com
452 MINING GEOSTATISTICS
At the time of selection, the true values z v are generally unknown, and the
actual selection is made on the estimators z ~ (preferably on the kriging
estimator z L). It is, thus, imperative to evaluate the recoverable tonnage
from the histogram of the estimators zt, i.e., by the dotted area (z~ ;:: zo)
on Fig. VI.3.
1. If, at the time of the evaluation of the reserves, the ultimate estima-
tors zt, upon which selection will be based, are known, then the histogram
of the z~ is experimentally available and the recoverable tonnage (z~ :;:;: zo)
can be estimated without any problem. In the next section, VLA.3, it will
be shown that the kriging of the selected units (those for which z~v;;::: zo)
provides a fair estimator of the mean grade of this recoverable tonnage.
From the histogram of the kriging estimators z~", it is possible to obtain
an estimator of the histogram of the true values z, by the following means.
(i) Use of a permanence-of-law hypothesis to extend the type of law
observed on the distribution of the z~" to the distribution of the z".
(ii) Use of a correction of variance: the variance of the distribution of
the z, is n 2 (vl G). This variance can be deduced from Dic* (vi G) by
one of the smoothing relations (VI.1) or (VI.2), or calculated from
the general formula (II.36).
t Formulae for deriving recovered tonnage and average grade from a known distribution are
given in the addendum VI.B.3.
WWW.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 453
In practice, at the time of the study, both the histogram of the core
grades z; and that of the kriged grades zt, are available, these two
histograms having respective variances D 2(c/ G) and D~* (vI G). The
problem is to obtain an estimator of the histogram of the ultimate estima-
tors z ~* (to be used for selection) from these two experimentally available
histograms. To do this, it is necessary to act as follows.
WWW.minebook.blogfa.com
454 MINING GEOSTATISTICS
(i) To evaluate the level of ultimate information which will be used for
future selection and to calculate the corresponding mean kriging
variance ~~*. Remember that a kriging variance can be calculated
in advance, once the corresponding data configuration is known. d.
section V.A.!, remark 4. However, the prior calculation of (T~~* is
often approximate, since it must take into account future informa-
tion which may be difficult to quantify, e.g., the experience of the
shovel operator or information about the stope face.
(ii) Using one of the smoothing relations (VI. 1) or (VI.2), to deduce the
dispersion variance D 2**(vIG) of the ultimate estimators Z~*.
Thus, for example, if the following approximate smoothing relation
is used:
2
the variance D ** (vi G) is written as
(iii) To deduce an estimator of the histogram of the Z~* from one of the
two experimentally available histograms by a variance correction
and assuming permanence of law.
WWW.minebook.b1c)!3fa.com
VI SELECTION AND ESTIMATION OF RESERVES 455
The dispersion variance of the true grades of the units v in V can be written
according to formulae (11.31) and (Il.35):
Consider now the N unbiased estimators Z~ (Xi) of the grades of the units v. The
global mean grade estimated over V is then
Z '"v (x ) = -I '\'
L. Z*
._" ( Xi ) .
N,
In a similar way. we can define the dispersion variance of the estimated grades of
the units v in V:
D
2
(v/ V) = E{ ~~ [Z~ (x,)-ztex )1 2
}
The case of kriged estimators If the estimators used above are kriged estimators,
then Z~ (Xi) = Z~v (Xi) is a linear combination of the n, data {Z,,(Xi)' a '" 1 to nil in
the neighbourhood of estimation, i.e., zt" (Xi) = L" A",Z", and the kriging ex, ),
system (V. I) is written as
{ IA s = 1
13
and
Considering that C(h) is a centred covariance, and that the stationary expec-
tation is m = E{Zku (x)} = E{Z~v(x)}, the following relation can be expressed:
(VI.3)
WWW.minebook.blogfa.com
456 MINING GEOSTATISTICS
Now, by putting
and noting that C(Vj, Vi)= C(v, v), 'Vi (all units Vi have the same support v), the
following relation is obtained for the dispersion variance of the kriged grades:
Di* (vi V) = C(v, v)+ 2,iL - a~u - E{[Z~v(x)- m f}.
By substituting D\vl V)+ C(V, V) for C(v, v), this relation becomes
D 2 (vl V) = D~*(vIV) +aL, - 2.u - [C(V, V) - E{[Z~v(x)- mf}]. (VIA)
The term between the brackets is of the order of magnitude of the global
estimation variance over the field V, and can generally be disregarded when
compared to mean local estimation variance o-L. The first approximate smoothing
relation (VI. I) is then obtained:
D (v/ V)= D~* (vi V)+uL- 2ji.
2
The case of complete kriging The term "complete kriging" is used when the N
krigings Z~,,(Xi) have been carried out using the same data configuration compris-
ing alit the available data in V. From section V.A.3, "Theorem of combining
kriged estimates", it can then be shown that the sum zt = (1/ N) L Z~" (x.) cor-
responds exactly to the direct kriging of Zv(x) from the previous data configura-
tion. For E{[ Zt. v (x) - mY} we then obtain an expression similar to (VI.3):
t In practice, the information used for the kriging of a unit v is limited to a neighbourhood
surrounding v. However. and provided there is no strong nugget effect (since the nugget
effect removes the screen, cf. section V. A.l, re mark 7), this nearby informatio n screens the
influence of more distant data and, thus, the kriging of v can be assumed to be complete and
taking all the available information over V into account.
WWW.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 457
--1f-----t- m
==-_z~
FIG. VI.S. Bivariate distribution of the true and estimated grades, Z" and Z~.
WWW.minebook.blogfa.com
458 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 459
Now, the kriging estimator za,is precisely the linear estimator which
minimizes the estimation variance, i.e., the sum of the accuracy criterion
and the dispersion criterion. By minimizing the sum of these two criteria,
kriging represents a compromise between satisfying one or the other, and,
because of this, it is preferred to any other linear estimator for selection
problems.
WWW.minebook.blogfa.com
460 MINING GEOSTATISTICS
The deposit
This simulation can be considered either as a zone of sedimentary deposit
of variable vertical thickness (such as the lateritic or silicated nickel
deposits of New Caledonia, d. section IV.E, Case Study 13), or as a
horizontal mining bench of constant vertical thickness in a massive deposit
mined by open pit (such as a porphyry - copper deposit).
The simulated zone is a rectangle of dimensions SOu x lOu, made up of
500 square blocks Vi (of dimensions u xu), cf. Fig. VI.6. A total of 60 500
samples were simulated on a regular u] 11 x u] 11 grid over this rectangle,
giving a total of 11 xlI = 121 samples per block. These 60 500 samples are
considered as the real deposit G to be studied.
IOu
I' "1
u
I' 'I
:w; ~ x :w; x x :w: x x x x
x
.
x
r----t----t:,..--.;;- - - - x II x II drill- holes
vI
u :--------
: S,. Centro I drill- note 5,.
,
x
Block v,.
1----+--+ - - - - -
Deposit G
WWW.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 461
(i) A given distribution - with mean 957% and variance (T~ = 22,4%2
- which is, in fact, the experimental histogram of the nickel accumu-
lations (thickness x grade) of the Mea deposit (New Caledonia).
Obviously, if z(x) is interpreted as a grade, the value m = 957% is
closer to a lead grade, while it is closer to a copper grade when
divided by 10.
(ii) A given model of spatial variability; in this case, a spherical model
with a nugget effect and isotropic in the horizontal plane:
!E{[Z(x +h)-Z(X)]2} = y(h)= Co+C,y,(r), 'rtr = Ihl> u/ll,
where Cs> 6%2, C'Yl(r) is a spherical model with sill C 1 = 18%2
and range a, = lOu.
x~x
This kriging configuration is complete only for the 384 blocks Vj inside
the zone 0' of dimensions Su x 48u, defined as the erosion of 0 by a unit
distance u. The following study will be confined to these 384 "blocks for
which the kriging system is identical.
The results AI, AZ, A3 , ai
of the above kriging system are given in Table
VI.1. Since the true grades of these 384 blocks are available, the kriging
variance u~ = 1 32 % 2 can be compared to the experimental estimation
WWW.minebook.blogfa.com
462 MINING GEOSTATISTICS
Krig Poly
Al 0204 1
..1. 2 0-458 a
..1. 3 0338 o
O"~, theoretical 1323 7307
0'1, experimental 1348 6919
VNNI.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 463
U
True
selection, 203 1297 0 0 2633
Zv 67 1619 0 0 1085
Kriging, 365 1013 1005 008 1-36 3698 3668
Z*
"
Poly,
U- 4
205
65
337
12-82
16-04
10-51
12-66
1591
016
0-13
1-36
156
10-68 -0-17 6-94
2628
1043
3542
2593
1034
3599
Z*
{1~ 199
67
12-52
14-93
13-52 -1-00 708
16-95 -2'02 781
2491
1000
2690
1136
Note that the selection made on kriged estimates results in a better re-
covery from the deposit, in terms of the global quantity of metal actually
recovered, Also note that the standard polygons of influence estimator
always overestimates the effectively recoverable quantity of metal, which is
particularly dangerous during a feasibility study. Selection based on the
VtIWW.minebook.blogfa.com
464 MINING GEOSTATISTICS
(0 )
20
15
10
>,
<.>
c:
9
<u
:I
C'
e
lL.
(b)
/
:>?
o /
/
/
/
20 /
/
/
/
/
/
/
15
,,-
'"
'"
.... ....
...
,/
.... ....
.....
..... "..-
"..-
IOc...__--
""
9
15 20
FIG. VL8. Conditional bias of the estimators. (a) Kriging. ~--, E{Zv/ zt, ~
zo}; ----, E{Z:'v/Zkv ~zo}. (b) Polygons of influence. , E{Zv/Z~ ~zo};
----, E{Z~/Z~ ~zo}.
If, instead of a regular grid, the 384 blocks were estimated from data on
a pseudo-regular grid with clusters of data and gaps (which is often the case
in mining applications) the results obtained from kriging would be still even
better than those obtained from ID, ID2, and Poly.
WWW.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 465
20
(a)
,--
-- ( b)
>-
<.>
"
'"0-
~ - - '--;.--
,--
-
' - ,--r--
10 f- I--
--
u:'" -
~
r..
0
o
I
18 9 14-4
rh I--
9 144
20 20
20
(c) -
-
-
----
-
-
- I--
o
r
1-8 9 144
rh
20
FIG. VI.9. Histograms of true grades and of their estimators. (a) Kriging: m* =
9,72, a 2*=1471. (b) Polygons of influence: m*=9'70, a 2*=2313. (c) True
grades: m* = 9,82, (T2 = 1614.
WWW.minebook.blogfa.com
466 MINING GEOSTATISTICS
Zv,---------------------- -----..,
(a)
I
"'1
,.,
~ .... 1
, 1';'.':')'
l
"
"
I
I 2
1
2" I
I I
I
l r- l I
I ,-
:3. /1 I ~
r .2 <
r-z ~ I
I 2
I 2: 2'2
2'
III
I .....
E Z
l
2
I I
I Z l<t'
* I dose to
I 3 .... 2" I l I ~ ....
'"
I
I
M
(b)
,I I I
t I I
, I
, I
I I
I
I I
I I I
,
,, I
. ,
I
M Z"
v
FIG. VI. 10. Scatter diagrams. true grade/estimator. (a) Kriging; (b) polygons of
influence.
WWW.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 467
It was seen in the preceding section that the estimation of the recoverable
tonnage requires the inference of the histogram of the estimators zt*
which are ultimately used in the future selection. For example, this
inference is made from the histogram of the grades Z; of the available core
samples, or from the histogram of the estimators Z~ available at the time
of the study.
More generally, we have the experimental histogram of a stationary
variable Z, from which it is required to deduce the distribution of a variable
Y that has the same expectation, but a different, known variance: E{ Y} =
E{Z} = m and Var{ Y};:= Var{Z}. This problem as posed has no solution,
since the type of the distribution of Y is undetermined. However, the
WWW.minebook.blogfa.com
468 MINING GEOSTATISTICS
1. Permanence of normality
Let Z be a Gaussian stationary variable (in practice, a variable the histo-
gram of which can be approximated by a normal distribution), with expec-
tation m and variance <T~. Let U be a standard normal variable with zero
expectation and unit variance. These two variables are related by the
standard formula Z = m + Uz. U.
The hypothesis] of permanence of normality means that Y is considered
to follow a normal distribution with expectation m = E{ Y} = E{Z} and a
variance <T~ #- (T~. Y can be expressed in terms of the standard normal law
as
Y= m+Uy. U. (VI.7)
The distribution of Y can then be readily determined from tables of
standard normal distribution.
2. Permanence of lognormality
Let Z be a lognormal variable with expectation m and variance a~. The
Gaussian transform function 'Pz which transforms a standard normal vari-
able U into this lognormal variable Z is
Z = cpz( U) = em'+a~u (VI.8)
or, conversely,
U = cP 2/ (Z) = (log Z - m')j <T~ ,
where m' and cr'l are the expectation and variance of the normal variable
log Z. These logarithmic parameters are deduced from the known arith-
metic parameters m and (T~ by the standard relations
m = E{Z} = em'+a~/2, }
{ u~ = Var {Z} = m 2(e(T] -1). (VI.9)
t If Y can be expressed as a linear combination Y = Let AaZa of normal variables, and if the
distribution of the RF Z(x) is multivariate Gaussian, then its distribution is normal, and the
permanence of normality is a property of Y and not an hypothesis. However, normal-type
symmetrical distributions are not very frequent in mining applications; the histograms of
grades in particular are usually asymmetric of the lognormal type.
~-~~
, VNNI.minebook.blogfa.com
.
VI SELECTION AND ESTIMATION OF RESERYES 469
t Contrary to the normal case, a linear combination Y = L:,,, A",Z", of lognormal variables Z",
does not follow a lognormal distribution. Thus, if the grades Z; of core samples follow a
lognormal distribution, then, to be theoretically rigorous, the grades Z; of blocks cannot
follow a lognormal distribution. In mining applications, however, the histogram of the block
grades z" very often has the same lognormal type of asymmetry as the histogram of core
sample grades Zc' For this reason, a- working hypothesis of permanence of lognormality is
adopted. A case study involving permanence of lognormality is presented in section II.C.3,
"Grade dispersions at Chuquicarnata mine", ct. Figs. 11.22 to 11.24.
t The lognormal recovery functions formulae are given in the addendum to section VLB.3.
WWW.minebook.blogfa.com
470 MINING GEOSTATISTICS
the distribution of the variable Y with the same expectation but different
variances (T~;t a~, and not in the quality of the fit of the histogram of Z.
If the two modes of the Z histogram are to be exactly reproduced on the
estimated histogram of Y, then, under certain conditions, t the following
procedure can be used.
0/ , . . . - - - - - - - - - - - - - - - - - ,
0
(a) (b)
Experimental Two lognormal distributions
histogram
(m,cr})
-', )IC
\.
" Log normal fi I
:; I (m,cr})
"' -, \.
-,
"-
"
o z
(i) Using the simple two-parameter distribution which fits best the
experimental histogram of Z, make the variance correction a~ ---;. (Tt,
cf. the two lognormal distributions on Fig. VI. 11(b).
(ii) From these two two-parameter distributions, read the recoveries
corresponding to the cut-off grade zoo The quotient of these two
recoveries gives an estimate p of the correction coefficient, to be
applied to the recoveries deduced from the experimental histogram
of Z to obtain the recoveries corresponding to the histogram of Y.
Thus, if the recovery read on Fig. VI. 11(b) from the lognormal dis-
tribution (m, a~) is 20% (hatched area) and if the recovery read from the
lognormal distribution (m, a~) is 25% (dotted area), the correction
coefficient is 25/20= 125. The hatched area on Fig. VI. 11(a) gives the
recovery read from the experimental histogram of Z, say, for example,
18%. The estimator of the recovery of Y for this cut-off grade Zo is, then,
Prob* {Y;2< zo} = 18 x p = 225%.
t This indirect correction procedure retains all the secondary modes observed on the histo-
gram of Z As a consequence, and in mining applications, this procedure will be used if these
secondary modes are significant and therefore should be reproduced on the histogram of Y.
For this, the variance correction (T~ ~ 17~ must not be significant; in practice, the relative
differences l(Ti-l7~I/u~ must be less than 30%. In fact, if this variance correction
represents a significant smoothing (l7~ (T~), this strong smoothing should mask the
secondary modes, which then should not appear on the Y-histogram.
WWW.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 471
This procedure has the advantage of being simple and of reproducing the
major characteristics of the experimental histogram of Z on the estimated
histogram of Y. However, the histogram of Y estimated in this way will not
have the exact required values tm, O"~) of the mean and variance. In
practice, the deviations from the required values (m, at) can usually be
corrected by a slight modification of the few extreme values of the histo-
gram of Y. In any case, no permanence-of-law procedure is robust enough
to be used for these extreme values (y greater than m + 3ay for instance).
WWW.minebook.blogfa.com
472 MINING GEOSTATISTICS
where the H;(u) are the Hermite polynomials. The definition and main
properties of these polynomials will be presented in the following section,
VLB.3.
In practice, the Hermite expansion is carried out up to an order n, i.e.,
n l/J;
(f)z(U) = .L --:-;H";(u). (VI. 15)
'=0 I.
VI SELECTION AND ESTIMATION OF RESERVES 473
The order n is such that the distribution of the variable fPz( U) correctly fits
the experimental histogram of the data z. The method of fitting the (n + 1)
coefficients l/Ji of the expansion of (fJz from the experimental Z -histogram
is given in the following section, VI.B.3. It can be shown that
~ IjI~ 2
1/10 = m, L, -'f
j~l I.
= (Tz, (VI. 16)
where m and O'~ are the mean and variance of the experimental Z-
histogram.
Applying the hypothesis of permanence to this Hermite expansion
consists of considering that the Hermite expansion of the transform
function (fJy of the variable Y can be written as
n !/J.e.
(,Oy(u)= I, ~ ~(u) (VI. 17)
i=O I.
where
n .1,2 2
" o/iCi _ 2
0/0 . Co = rn, i.e., Co = 1, c... ~l--(Ty
.=1 l.
Thus, (fJz and cpy have the same Hermite expansion, except for the vari-
ance correction coefficients c; The variance correction is written c, = a' (a
to the power i) which entails
2 n I/ITa 2 i
O'y=.2: -.,-=!(a). (VI.18)
,= 1 1.
The justification for adopting the form c, = a' for the variance correction
is given later, in remark 1.
Since the n parameters {I/Ii, i = 1 to n} are known, the function f( a) can
be plotted as on Fig. VI.12. The known value (T~ then determines the
parameter a of the variance correction. Formula (VI.17) gives the expres-
sion of the Hermite expansion of ([iy, i.e.,
n IjIja i
fPy(u)=.2: -.-, H;(u). (VI. 19)
.=0 1.
WWW.minebook.blogfa.com
474 MINING GEOSTATISTICS
f( a)
2
0-
Y
o a
where u'i. is the variance of the normal variable log" Z. It can be shown that
the Hermite expansion of this function 'Pz can be written as
~ 0/; ~ (-u~Y )
'Pz(u)= L.. ~H;(u)=m L. 'r H;(u;
;~O 1. j=() I.
thus,
0/; = rn(~u~Y, i.e., % = m = E{Z}
and
I =() 1.
i.e., a = a'yJ cr~, the ratio of the standardldeviations of the variables loge Y
and loge Z.
WWW.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 475
WWW.minebook.blogfa.com
476 MINING GEOSTATISTICS
G(u)=Prob{U<u}= f"
r, g(x)dx= /
v 2 7T Lco
r
e- x 2j 2dx. (VI.20)
where
I if i = i.
i {
b; = 0 if i ~ j.
WWW.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 477
This condition is satisfied for any variable Z with finite variance, which is
always the case in practice since this variance is deduced from the experi-
mental histogram of Z. The following property then results: any function
(fl(x), such that
co co 0/.
cp(x) = .L k;7Ji(X) = ,=0
,=0
2: ~Hi(X).
I.
(VI.22)
From this, it follows that the expansion (VI.22) is the best approxima-
tion, up to any order n, of If'(x) by a polynomial of degree n, in the
weighted (by g(x least squares sense. Indeed, with
the equations allao/i = 0, Vi, providing the minimum of I, give the relations
(VI.23) which define the coefficients !/Jj.
Now, let <pz be the Gaussian transform function of Z ~ the moments of Z
are written as
E{Z} = E{lf'z(U)} =
rco
Loo 'Pz(u)g(u)du = !fro,
eo eo r
1/1.1/1.
E{Z2} = E{cp~(U)} = ;~o j~O i;jt teo +00
Hi(u). ~(u). g(u)du
00 !/JT
=L~.
i=O I.
WWW.minebook.blogfa.com
478 MINING GEOSTATISTICS
There are, thus, two relations relating the coefficients of the Hermite
expansion of ((Jz to the mean and variance of Z:
E{Z} = m = l/Jo,
co ,1,2 co ,1,2
1 (VI.24)
2 'fJi 2 -, wt
Var{Z} = <Tz = i~O i!-I/Jo= i~l iT'
In practice, the Hermite expansion of ((Jz is stopped when L;= 1 o/~ / i! is
close enough to the required variance u~. The contributions !/J?/ i! of the
terms of higher degree (i > n) tend rapidly toward zero.
((Jz being expanded up to the order n, it must be verified graphically that
the distribution of the variable
Graphical transform
The transform function cpz can be determined graphically from the two
cumulative distribution curves, G(u) = Prob {U < u} corresponding to the
standard normal variable U, and F~(z)= Prob {Z < z} corresponding to Z,
d. Fig. VL13. A class (Uk' Uk+l) of the theoretical Gaussian curve G(u)
corresponds graphically to each class (Zb Zk+l) of the experimental cumu-
lative Z-histogram. An interpolation (linear, for example) can be used
Cumulative frequency
,/
,/
,/
,/
,/ -.
--
j
,
Experi menIal
J histogram
1
j
I
I 1
I
u,z
WWW.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 479
within each class interval (Zk' Zk+l). This way, a one-to-one correspon-
dence Z ~u can be defined graphically, i.e., not only the transform z =
If'z(u), but also its inverse u = If' Zl (z).
If there are N data values zs, one solution consists in sorting these N
values by ascending order: z 1";: Z2";: . ~ ZN. The experimental cumulative
frequencies are approximated by: F'i: (Zk) = k/(N + 1). The graphical trans-
form, d. Fig. VL13, consists in identifying the two cumulative distributions,
i.e. G(ud = F'i:(Zk). Hence the ordered series of gaussian values u is
given by:
Vk = 1 toN. (VI.25)
with:
u~ = tu; + Uk-l)/2.
N -1 is the number of classes representing the gaussian distribution, the
Uk'S are the class bounds and the u ~ 's are the central class values. If N is
taken as the number of original data values Zb the Uk'S are the corresponding
gaussian values obtained by graphical transform through formula (VI.25),
and: cp(u~) = (z, +Zk-I)/2.
NNI.minebook.blogfa.com
480 MINING GEOSTATISTICS
Remarks
Often in practice the density function fz (z ) is given through the discrete
form of a histogram.. The previous integrals should then be replaced by
discrete summations.
Note that for a zero cut-off:
T(O) = To, 0(0) = Tom, m(O) = m,
(z-m)
Fz(z) = G --;;- ,
g(u) and G(u) are the standard normal density and distribution functions,
d. Abramovitz, M. and Stegun, I. (1965), p. 931.
The preceding recovery functions are then written:
WWW.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 481
(VI.28)
WWW.minebook.blogfa.com
482 MINING GEOSTATISTICS
o I 10 ,j,
m
z
FIG. VI.14. Thickness histogram (core section support). Experimental values for
m = 1209, D 2(Oj G) = 4898 are shown by the histogram. , Hermite fit,
n = 15; - - - -, Hermite fit, n = 30.
Fig. VI.15.
Note that the secondary modes for the large thicknesses are no longer
present, but the secondary mode around 8 m remains. The following
models were fitted to this experimental histogram, cf. Fig. V1.15.
VNNI.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 483
30
FIG. VL15. Kriged thicknesses histogram (100 m x 100m support, 200-m sam-
pling grid). Experimental values for m = 11 83 and Di* == 241 are shown by the
histogram. , Hermitian permanence, n = 15; - - - -, Gaussian fit.
(i) A normal distribution with the same mean and variance. There is a
slight probability that this normal distribution will result in negative
thicknesses.
(ii) An Hermite expansion, deduced by variance correction D1.(OIG)~
Di* (vi G) from the n = 15 order Hermite expansion fitted to the
histogram of thicknesses measured along the drill cores (d. Fig.
VI. 14). The correction factor a used in this variance correction is
given by formula (VI. 18) and a = 07064. Note that this histogram,
deduced by Hermitian permanence, tends at the limit towards the
normal distribution with the same mean 1183 and variance 241.
Indeed, the decrease in the variance is particularly significant here,
the difference between D 2(OjG)=4898 and D~*(vjG)=24'1
being more than 50%. Note that if an Hermite model were fitted
directly to the experimental histogram on Fig. VI. 15, the secondary
mode around 8 m would show up more clearly, d. Fig. VI.20.
Selection study
The 100 m x 100 m panel v is considered as the selection unit on which
future production will be carried out. The selection variable is the mean
mineralized thickness.
If the only information available at the time of study is the data from the
200-m sampling grid, the cut-off will be made on the preceding kriged
WWW.minebook.blogfa.com
484 MINING GEOSTATISTICS
values Zkv(X), i.e., on the experimental histogram on Fig. VI.15, the results
of which are shown in column 1 of Table VI.3. From this table, for
example, a cut-off thickness Zo = 5 m provides a recovery of 94 5% of the
panels, i.e., 945% of the kriged values z~v(x) arc greater than the cut-off
zo=5 m.
In fact, much more data are available at the time actual selection is
made; this future information is assimilated to a regular, square 25-m grid,
i.e., 16 samples per paneL The actual future selection is thus assumed to be
made on the ultimate estimators z~t (x) deduced by kriging from the data
on the 25-m grid. Therefore, the problem is to deduce an estimator of the
histogram of these 2214 ultimate selection values z~~ from the histogram
available at the time of study, i.e., the histogram of the kriged values z~v.
To do this, the procedure outlined in section VI.A.2 is followed.
(i) and (ii). The kriging variance (T~~* corresponding to the ultimate
25-m grid is calculated, and then the dispersion variance of the ultimate
kriged values z~t is calculated by means of the smoothing relation (VI. 1).
This results in a value D~** (vI G) = 328 m".
(iii) Using an hypothesis of permanence of law, an estimator of the
future histogram of the z~~ is deduced, either from the sample thick-
nessses histogram by decreasing the variance D\OI G)....::,. D~** (vi G),
or from the histogram of the now available kriged values zkv by
increasing the variance D~* (vI G)....::,. D~** (vi G). Applying the cut-off
thickness Zo to this estimated histogram of the z~t results in the required
prediction of actually recoverable reserves.
Two procedures based on two different permanence-of-law hypotheses
were considered, namely:
(i) the affine variance correction, d. section VI.B.l, point no. 4, the
correction by increasing the variance D~* . . : ,. D~** is made directly
on the ztv -histogram, according to formula (VI. 14);
(ii) the correction by decreasing the variance D 2 (OI G )---')o D~** is made
on the Hermite expansion fitted to the sample thicknesses histogram.
WWW.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 485
/-,
I \
\
I '-
I "-\
I \
/ \
\
\
~
\
\
I \
J \
\
/
J
J -,
...... _- -- ~- ...........
o I 10 t 20 30
m
r
FIG. VI.16. Estimated histograms of the future kriged thicknesses (100 m x 100 m
support, 25-m sampling grid). , Hermitian permanence; - - - -, affine cor-
rection. m = 1183, D~** = 328.
ZL Z**
Kv
A
WWW.minebook.blogfa.com
486 MINING GEOSTATISTICS
% V(zo) = f
20
+<:0 zf(z) dz/ f
()
--cc zf(z) dz,
(iii) the mean thickness of the recovered panels, d. columns 3,6 and 9 is:
expressed in meters.
VNNI.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 487
WWW.minebook.blogfa.com
488 MINING GEOSTATISTICS
100r---~
'">
B
'"
E
50 ~
'\
'"
u \
~ \
0
\
,,
"-
"-
~O
"
o I 10
30
/'
.;
/
/
/
/
/
/
a I 20 30
WWW.minebook.blogfa.com
VI SELECTION AND ESTIMATION OF RESERVES 489
II"
II
1\
I 1
I I
15 I 1
I I
I I
I I
I I
I I
I I
I I
I :
10 I 1
I j
(\ I
I \
I I
I I
! \
I \
I
5 I
I
I
I
I
f \
I
- ---
01 10 t 20 30
m
VtIWW.minebook.blogfa.com
490 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
VII
Simulation of Deposits
SUMMARY
A regionalized variable z(x) is interpreted as one realization of a certain random
function Z(x). This RF - more precisely this class of RF's - is characterized by a
distribution function and a covariance or variogram model. The idea of simulations
consists in drawing other realizations Zg(x) from this class of RF's, and to retain
only those realizations zsc(x) which meet the experimental values at the data
locations x"', i.e., zsC<x.,,} = z(x a ) , 'r/x", E data set. The realizations zsc(x) are called
"conditional simulations" of the regionalized phenomenon z(x).
The theory of conditional simulations is outlined in section VILA together with
the original "turning bands" method, which provides three-dimensional simulated
realizations from one-dimensional realizations simulated on lines rotating in space.
The practical application of conditional simulations is given in section VII.B,
through two-case studies. A basic FORTRAN program is given, which should
allow each user to write his own simulation package. Then several typical uses of
simulated deposits in the field of mine project and planning are presented.
NNI.minebook.blogfa.com
492 MINING GEOSTATISTICS
Conditioning
There are an infinite number of possible realizations {zs(x), s = 1 to co} of a
RF Z(x). From among this infinity of realizations, the simulations {zsc(x)}
that are chosen are those that meet the experimental data values at the
actual data locations x"', i.e., those simulations for which
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 493
Simulation or estimation?
Simulated deposits have the same values at the experimental data loca-
tions, and have the same dispersion characteristics (at least up to order 2)
as the real deposit. In what way then do such conditional simulations differ
from an estimation? Their difference lies in their objectives.
(i) The object of an estimation is to provide, at each point x, an estimator
z*(x) which is as close as possible to the true unknown grade zo(x). The
criteria for measuring the quality of an estimation are unbiasedness and
minimal mean quadratic error, or estimation variance E ([Z (x) - Z* (x)] 2}.
There is no reason, however, for these estimators to reproduce the spatial
variability of the true grades {zo(x)}. In the case of kriging, for instance, the
minimization of the estimation variance involves a smoothing of the true
dispersions, cf. the smoothing relations (VI. 1) and (VI.2). Similarly, the
polygonal influence method of estimation would consider the grade as
constant all over the polygon of influence of a sample, and, thus, under-
estimate the local variability of true grades. The estimated deposit {z*(x)},
is, thus, a biased base on which to study the dispersions of the true grades
{zo(x)}.
(ii) On the other hand, the simulation {zs(x)}, or better the conditional
simulation {zsc(x)}, has the same first two experimentally found moments
(mean and covariance or variogram, as well as the histogram) as the real
grades {zo(x)}, i.e., it identifies the main dispersion characteristics of these
true grades. On the contrary, at each point x, the simulated value zs(x) or
zsc(x) is not the best possible estimator of zo(x). In particular, it will be
shown that the variance of estimation of zo(x) by the conditionally simu-
lated value zsc(x) is exactly twice the kriging variance, ct. relation (VII.6).
In general, the objectives of simulation and estimation are not compati-
ble. It can be seen on Fig. VII.1 that, even though the estimation curve
WWW.minebook.blogfa.com
494 MINING GEOSTATISTICS
z*(x) is, on average, closer to the real curve zo(x), the simulation curve
z:c(x) is a better reproduction of the fluctuations of the real curve. The
estimation curve is preferable to locate and estimate reserves, while the
simulation curve is preferred for studying the dispersion of the charac-
teristics of these reserves, remembering that in practice the real curve is
known only at the experimental data points Xu'
zr----------------------------------,
x
FIG. VII. 1. Real, simulated and estimated profiles. , Reality;
conditional simulation; - - - -, Kriging; 0, conditioning data.
The right-hand side of the profiles on Fig. VII.l illustrates the fact that a
simulation cannot be used to replace sampling, which is always necessary
for a good local estimation of the deposit. In any case, the better the
deposit is known, the better the structure of variability can be modelled
and the denser the grid of conditional data will be, all of which will result in
a simulation closer to, and more representative of, reality.
WWW.minebook.blogfa.com
vn SIMULATION OF DEPOSITS 495
experimental data points the simulated values and the experimental data
values must be the same:
zsc(x,,) = zo(x,,), VXIJI E data set 1.
Consider the true value zo(x) and its kriged value z tK (x) deduced from
the available data {zo(xa ) , Xa E I}. These two values differ by an unknown
error:
zo(x) = Z6K (x)+ [zoex)- Z~K (x)J (VII. 1)
or, in terms of random functions,
Zo(x) = Z6K (x)+ [Zo(x)- Z6K (x)]. (VII.2)
Now; a characteristic property of kriging is that the kriging error
[Zoex)- ZSK (x)] is orthogonal to the kriged values, i.e.,
E{Z6K(Y) [Z(J(X)-Z~K(Xm=O, 'T/x, y.
(The proof of this result is given in the addendum at the end of this
section).
To obtain the desired conditional simulation, it is thus enough to replace
the unknown kriging error [Zo(X)-Z6K(X)] in formula (VII.2) by an
isomorphic and independent kriging error [Z, (x) - Z~K (x)]. More pre-
cisely, we consider a RF Zs(x) which is isomorphic to, but independent of
Zo(x). Given a realization zs(x) of this RF Zs(x), the kriging procedure,
when applied to the same data configuration {zs(x a ) , X'" E I} will result in a
kriging error [zsCx) - Z ~K (x)] isomorphic to the true error [zo(x) - ztK (x)]
and independent of Z6K(X). The required conditional simulation is then
written as
(VII. 3)
(ii) The RF Zsc(x) is isomorphic to Zo(x). This follows from the fact that
the simulated kriging error [Zs(x) - Z~K (x)] is isomorphic to the
true error [Zo(x) - Z~K (x)] and independent of Z~K (x). In the
addendum, it will be shown that this isomorphism applies rigorously
to the increments of the RF's Zsc(x) and Zo(x) only. In other words,
the variograms of these two RF's are identical, but not necessarily
their covariances. This does not pose any problem since, in geo-
statistics, the variogram structural function is always preferred to
the covariance.
(iii) The simulated realization Z se( x) is conditional to the experimental
data. This follows from the fact that, at the experimental data
points, the kriged values are equal to the real values, i.e.,
'rIxo E I,
Remark 1 The kriging weights {.Ao , Va E I} are the same for Z~K (x) and
Z;K (x), since the two RF's Zo(x) and Z, (x) are isomorphic and the two
kriging configurations are identical.
Remark 2 The conditioning principle provides both values, kriged ztK: (x)
and conditionally simulated Z:e(X), at each point x, cf. formula (VII.3). By
taking into account the independence of the two kriging errors [Zo(x) ~
Z~K (x)] and [Zs{x) - Z:K (x )], the variance of estimation of the real value
zo(x) by the conditional simulation zsc(x) can be written as
E{[Zo(x)- Zsc(x)f} = E{[Zo(x)- ZtK (X)]2}+ E{[ZAx)- Z:K (x)]2}
= 2E{[Zo(x)- ZtK (X)]2} = 2(T~.
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 497
'fIXaE J,
The covariance between the kriging error and the kriged values then becomes
and the kriging error R (x) is thus orthogonal to the kriged values ztdy).
In the usual case of a kriging system with a non-bias condition, the ordinary kriging
system (V.1) is written as
This covariance becomes zero when Ia v: = 0, i.e., for all linear combinations
La v", Zo(xa) known as "authorized" linear combinations, ct. section II.A.3. In this
case, the kriging error R(x) is orthogonal only to authorized linear combinations of
the data Zo(xa ) .
Now, the difference between any two kriged values is such an authorized linear
combination; indeed,
with
and, thus,
E{R(x). [Z~K (y)-Z6K (y')]) = 0, 'fix, y, y'.
The variogram of the RF Zo(x) can then be written as
2y(x - y)= E{[Zo(x)- ZO(y)]2} = E{[Z6K (X)-Z6K (y)]2}+ E{[R(x)- R(y)]l}.
WWW.minebook.blogfa.com
498 MINING GEOSTATISTICS
From expression (VIL3) for conditional simulation, this variogram is exactly that
of the RF ZscCx), since the error [ZAx)-Z;dx)} is isomorphic to R(x) and
independent of ZtK (x). Thus, the conditional simulation Zsc(x) ensures the
reproduction of the variograrn of the RF Zo(x), but not necessarily its covariance.
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 499
w
,X
Bond \
\
~~-- h
\ o
\
\
I
x+h
A final value is then assigned to each point x by taking the sum of the N
contributions from the N lines:
1 N,
z.,(x)=i L z;(x).
'IN i~l
(As the N lines can be deduced from each other by rotation, so can the
slices (or bands) which they define in the three-dimensional space, hence
the name "turning bands" method.)
The resulting realization zs(x) is a realization of a three-dimensional RF
Zs(x) = Z,(u, v, w) which is second-order stationary, has a zero expectation
and a covariance
WWW.minebook.blogfa.com
500 MINING GEOSTATISTICS
where <h, k> is the projection of the vector h onto the axis k.
'T = Ih) =! J (h ~ + h ~ + h~) is the modulus of the vector h.
The preceding integral can be written in terms of spherical coordinates,
d. Fig. VII.2:
1 f217" f7T/2 1 IF
C(r) =- dB C(l)(lrcos cpl)sio cp dcp =- C(l\s)ds.
21T () 0 r 0
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 501
where r = Ihl; K o, Ki, Kz, K 3 are positive constants; and Co, C h C 2 l C 3 are
isotropic covariances defined in three, one, two and three dimensions,
respectively. These covariances are, for example, spherical models with
ranges ao, at> aZ, a3 and unit sills.
In this case, the RF Z(x) can be interpreted as the sum of four inde-
pendent RF's:
3
Z(x)= L ~(x),
;~O
VtIWW.minebook.blogfa.com
502 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 503
(i) select from the simulation file the realization, spherical or exponen-
tial, that has a range closest to the specified value;
(ii) multiply all values of this realization by the constant J K.
WWW.minebook.blogfa.com
504 MINING G EOSTA TISTICS
which are the preceding 15 lines that form a regular partition of the
three-dimensional space
-(k+ 1)
k
1 k
-1
k
+1
l (VILS)
[ ~ ~ ~J.
001
c(l)(s)= I
reo f(u) f(u+s)du.
: J= L; (VIL9)
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 505
Y(u) = t: cc
feu +r)T(r) dr
with
+ s)} = E{f L: cc
feu + r)f(u + r' + s). T(dr). T(dr')}
which, except for a multiplicative factor which is the density of variance (T2,
WWW.minebook.blogfa.com
506 MINING GEOSTATISTICS
This formula is practical only when the function f(kb) decreases rapidly
towards zero as )kl increases, so that the discrete sum does not have to be
taken over the range k = - co to + oo,
I ,
o
weigh'~rtgl lu nction
flu)
Ii t, + I li+k D
1<>:::J I I I I ~
b It
This discrete sum C~l) (s) is simply the discrete approximation of the
integral in (VII.9), which defines the specified one-dimensional covariance
C(1)(s). This discrete approximation C~1) (s) can be calculated for each
weighting function feu) and for each value b of the discrete interval. The
bias C(l)(S)- C~l)(S) caused by the discrete approximation can then be
corrected. In practice, a multiplicative correction is enough, and this can be
achieved by manipulating the parameter (T2 = E{ T 2 } to ensure the equality
of the variances: C(l)(O) = C~1) (0). An example of the correction of this
bias is given in the case study of the simulation of the Prony deposit, cf.
section VII.B.2.
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 507
'tt'rE[O,a],
.r 1--+~3'
fiL 3s 2s J
3
WWW.minebook.blogfa.com
508 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 509
Zo(x) is not restricted to the Gaussian type. In particular, the real values
{zo(x), x E deposit} are often non-negative, as in the case of grades or
thicknesses, whereas the simulation {zs(x), x E deposit}, having a Gaussian
distribution, may have a significant proportion of negative values. Thus, it
is particularly important to ensure that the simulated values zs(x) have the
same histogram as the experimental histogram of the available data zo(xo:).
Of course, it must be assumed that this experimental histogram is
representative of the distribution of the variable zo(x) in the deposit, cf., in
section VLA.l, the remark on the representativeness of a histogram.
Let Zo(x) be a stationary random variable with a distribution represen-
ted by the experimental histogram of the available data {Zo(X,,)E I}. By
means of a Gaussian transform function, this variable Zo(x) can be
converted to a standard normal variable U(x), i.e.,
2 0 = cPZ:J( U) and U = cp z,~ (Zo).
Instead of simulating the grades zo(x), the transformed grades u(x) are
simulated. This simulation {us(x), x E deposit}, obtained by the turning
bands method, has a Gaussian histogram and the transform CPZu results in a
simulation {zAx) = cpZu(us(x)]} with the required histogram, i.e., a histo-
gram similar to that of the experimental data zo(x",,). The covariance
imposed on the simulation uAx) must, of course, be the covariance
deduced from the transformed experimental data, u(x",) = cp~~ [zo(xa ) ],
V X(l E I. Although not rigorously, the inverse transformed simulation
zs(x)= 'PZo[us(x)] will then, in practice, have the required covariance
C(r).
Thus, the simulation zs(x) will reproduce not only the first two moments
- expectation and covariance or variogram - but also the univariate dis-
tribution or, more precisely, the experimental histogram of the data zo(x",).
If, in addition, the conditioning of the simulation to the experimental
data zo(xa ) is desired, it is advisable first to condition the simulation us(x)
to the transformed data u(X(l)=fPZ~ [zo(x",)] through relation (VII.3), i.e.,
usc(x) = u~(x)+ [uAx)~ U~K(X)],
and then to take the inverse transform zsc(x) = cPa, [usc(x )]. The kriged
values u~(x) and U:K(X) are deduced, respectively, from the transformed
WWW.minebook.blogfa.com
510 MINING GEOSTATISTICS
data u(x",J and from the simulated values u.. (x.) located at the same points
Xc.' These krigings are carried out using the covariance Cu(r) of the trans-
formed RF Vex).
By way of example, consider an experimental histogram of real grades
{zo(x",), 'tIxa E I}, to which a lognormal distribution can be fitted. The
transformed variable which follows a normal distribution is: V(x) = log
Zo(x), and, thus, Zo(x)=exp [U(x)}. After determining the covariance
Cu(r), which characterizes the spatial regionalization of the transformed
grades {u(xa)=logzo(xa), 'tIxaEI}, the turning bands method is used to
provide a simulation u,.(x) which has a Gaussian histogram and the
imposed covariance Cu(r). This simulation is then conditioned to the
transformed data u(x by formula (VII.3). Upon taking the inverse trans-
Ol
)
A fundamental remark
The above procedure reproduces the required univariate distribution of
the RF Zo(x), but this does not mean that the 2-, 3-, ... , n-variate dis-
tributions are also reproduced, i.e., the spatial law of the RF Zo(x) is not
reproduced in its entirety. As far as geostatistical applications are con-
cerned, this is not very important since:
(i) most geostatistical techniques (kriging, estimation of recoverable
reserves and study of their characteristic fluctuations) require only
the univariate distribution of Zo(x) and its covariance or variogram,
function;
(ii) the conditioning of the simulation zsc(x) to the experimental data
zo(xa ) induces a certain robustness with respect to the features of the
reality zo(x), which are not specifically known or imposed on the
simulated model. This robustness increases with the amount of
conditioning data.
However, it should be stressed that the turning bands method produces
simulations with a Gaussian spatial law and, as such, these simulations
cannot correctly represent mineralized phenomena having spatial laws
which do not have the very continuous character of the Gaussian spatial
law. Consider, for example, a non-stationary phenomenon consisting of
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 511
two states, one highly mineralized and the other waste, d. Fig. VIlA. Each
of these two states considered separately may have a Gaussian spatial law,
but when considered together they definitely do not. A straightforward
simulation of such a process, using the turning bands method, would result
in a profile (the bold line on Fig. VIlA), which, over large zones, may have
the expectation and covariance of the combined states of the real
phenomenon, but nevertheless would be a poor representation of this
reality.
In such a real case, the two states could be differentiated by structural
analysis, and they would then be simulated independently; the transition
from one state to another being simulated separately. Alternatively, one
could consider a regularized phenomenon, taking averages within volumes
sufficiently large to include a great number of different states; this
regularization would then provide moving average profiles with a more
continuous character open to simulation by the turning bands method.
FIG. VIlA. The turning bands method is not suited for the straightforward simula-
tion of a two-state process.
WWW.minebook.blogfa.com
512 MINING GEOSTATISTICS
zsc(X) is conditional to the data zo(x",), the block simulation will also be
conditional to the same data.
The block is approximated by n interior points (Xi' i = 1 to n), and the
true grade of the block
2u (x)=1. fZ(y)dy
v v(x}
is approximated by the arithmetic mean of the n interior point grades
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 513
D. From the histogram of the simulated val ues z v,s (x), determine experi-
mentally the Gaussian transform function for a support v, i.e., z",s(x) =
'Pz~ {uv.s(x )}. Once tpz~ is known, the Gaussian simulated values can be
deduced from U",s (x) = tpz~ {zv,s(x )}. Note that, since the function 'P z~ is,
in general, non-linear, u".s (x) :;i: (1/ n ) L Us (Xi)'
E. Using the Gaussian simulated values uvAx) and Us (x",), simulate the
kriging error [uv,s (x) - u t. dx)]. The kriged value U :,sK (x) is a linear
combination of the simulated values u, (x",) at the data points X.,. E J, i.e.,
U ~.sK (x) = L", A", (x). uAx",,). In the kriging system, the first member
matrix consists of the experimental covariances E{us(x )us(y)} and the
second member matrix consists of E {us(x )u v .s (y)}.
F. Calcula te the kriged values of the blocks which are linear com binations
WWW.minebook.blogfa.com
514 MINING GEOSTATISTICS
VNNI.minebook.blogfa.com
----
VII SIMULAnON OF DEPOSITS 515
the term Z~K (x ) = 'L" A,,20(Xa ) . But the final term Z~,SK (x) requires a point
simulation z; (x"J on the actual data locations X a and this point simulation
must be coherent with that of the blocks zu,s(x), i.e., it cannot be carried out
independently.
One solution to this problem consists in conditioning the block simula-
tion to constructed data z~(xa), defined on the same support v and located
at the same points X a E I, rather than to the real point data zo(x a ) . To do
this, the dimensions of the support v must be very small with respect to the
grid (xu> x,B) of the conditioning data. In practice, the dimensions of v must
be less than one tenth the mean distance between pairs of conditioning
data (x m x,B) However, it is always possible to restrict the number of
locations X a to be considered in the construction of the v-support data
z~(xaJ These constructed data z~(xa) may be, for example, the kriged
value of the block v(xa ) centred on Xa, plus an additional term to correct
the smoothing effect of kriging:
{VII. 19)
where Z:K (xu) is the kriged value of the block v {xaJ using the neighbouring
real data {zo(x a ) , x, E I}, and e; is a realization of a random variable with
zero expectation and a variance E{E~} such that the variance of the
constructed data z~(xa) is equal to the dispersion variance D 2 (vj oo ) of the
true grades of support v, cf. the kriging smoothing relation {VI.1) or (VI.2).
This latter variance D 2 (vj oo ) is calculated from the point structural model
'ro(h), d. formula (11.35) or (11.36).
This method of constructing the e-support conditioning data is an
approximation, since the variable E v is supposed to be independent of the
kriged value Z~K (x), whereas, in fact, the kriging error [Zv(x",)- ZtK (x a ) ]
is only orthogonal to the kriged value ZtK (x). This approximation alters
the spatial variability of the true values of support v, and can be justified
only when the variance E{E:~} is small, i.e., when the kriging variance of a
unit v centred on a data location x, is itself small with respect to the
dispersion variance D\v/ ro).
Once these u-support conditioning data z~(x",) are constructed, the
conditioning formula (VII.3) can be used in exactly the same way as in the
point case:
WWW.minebook.blogfa.com
516 MINING GEOSTATISTICS
In this model, all cross- and direct covariances Ckk,(h) are expressed as
linear combinations of n basic direct covariances K;(h). (It will be recalled
here that, in the above notation, i in the superscript position represents an
index and not a power.)
Consider now n sta tionary RF's {Y; (x), i = 1 to n} with direct covari-
ances K;(h), respectively, and all independent of each other. The L sta-
tionary RF's {Zk(X), k = 1 to L}, defined by the L linear combinations
n
Zk(X) = L akiY;(x) (VII.21)
;=1
Ckk,(h) = L akiak'iKi(h).
;=1
VNNI.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 517
'tNNI.minebook.blogfa.com
518 MINING GEOSTATISTICS
Drift and cut-off grades The vertical profiles of the Ni grades measured
along the drill cores all have a similar shape, d. Fig. ILL There is a
progressive enrichment in nickel from the upper surface, followed by a
rapid impoverishment, and, finally, a further enrichment at the bedrock
contact. This is translated into geostatistical terms as a vertical drift of the
Ni grades. The drift curve (or mean tendency) can be estimated by uni-
versal kriging from the vertical profile of the grades of the core sample
lengths, and it can be shown that this drift curve is an estimator of the
vertical profile of the mean grades of small panels centred on the drill
holes, d. Ch. Huijbregts and A. Journel (1972) and Figs.I1.1 and VII.5.
100 m
o 0--'~ ~ - - -..... 0 o
Smoll panel
~
c e n t ra l dr il l hol e
o c, 0, o
I
I I i I
I I I I
I I I I
I I I I N' cut - off
I ~ I I t,
M ine o bls
ore
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 519
drill-hole on each mining unit. Thus, the cut-off grade t c. must be applied to
the previously defined drift curve estimating the vertical profile of mean Ni
grades of the mining unit, d. Fig. VII.5. For each value te , a certain number
of "service variables" are defined. These variables, regionalized in the
two-dimensional horizontal space, are: overburden thickness Po, mineral-
ized thickness PL and the corresponding accumulations ANi and A M gO ' The
simulation of the deposit thus consists of generating a realization (core-
gionalization) of the corresponding service variables on a 10 m x 10 m grid
for each value of the cut-off grade t.,
Horizontal structural analysis At the time of this study, the first pro-
duction zone was sampled on a systematic 100 m x 100 m grid, and by two
cross-shaped drilling sets spaced at 20 m to detect small structures. This
information was used to characterize the horizontal regionalizations of the
previously defined service variables. In the actual study, four cut-off grades
were considered; six service variables for each cut-off. But only the results
obtained for the mineralized thickness PI., the overburden Po, and the Ni
accumulation ANi for the two cut-off grades rc == 0 (no vertical selection) and
fc = 10% will be given here.
For proprietary reasons, all values given here have been multiplied by a
constant factor, which does not affect the relative curves (variograms
divided by the squares of the means), nor the relative dispersions.
Table VII.l gives the means m and the relative dispersion variances
D / m 2 of the various service variables within the zone of interest. These
2
experimental values were deduced from the 176 drills covering the first
production zone on a 100 m x 100 m grid. Table VII.2 gives the correlation
coefficients between these service variables for each cut-off grade.
Figure VII.6(a), (b), (c) gives the experimental relative variograms of the
variables Po for t; == 10 and PL and ANi for te = O. Figure VII.7(a), (b)
shows the scatter diagram for Po and Pl. and the corresponding cross-
variogram for tc = 10.
TABLE VILI. Experimental means and
dispersion variances
tc Po PL ANi
m 237 269
2Jm 2
o
D 008 017
WWW.minebook.blogfa.com
520 MINING GEOSTATISTICS
o 087
10 -045 098
0-4
( a)
0-3 l-
N
~
"'-
0-2 l -
./
..... /'
0-1V-
I I I I I
0 20 100 200 300 400
0-10
(b)
N
~
0-05
"'-
( c)
0-15
~.
Direct ANi
---- .
N
~
"'-
0-10
I
/ ..-: .............
____
~ .
Cross ANi-fl
-
_.
005
WWW.minebook.blogfa.com
VII SIMULAnON OF DEPOSITS 521
.' ( 0)
30
20
.. .. .
. . ..'
} "
... ,.,.
t- l1li. .!
.
10
'II
... ...
.
...
...
..
-:......... ..-:
oil
.::...
l1li l1li
..
. . I\
.. ... ,....
" ..... ,. :...
.Jo.~
.... . iII-. .
5
"
..-.. ..
":
o 10 30 40
-0-07 r-----------~-------___,
( b)
~ -0-15
"
-023
FIG. VII.7. Correlation PO-PL for tc = 10. (a) Scatter diagram; (b) relative cross-
variogram.
NNI.minebook.blogfa.com
522 MINING GEOSTATISTICS
simulated values Zsc which do not satisfy these inequalities are replaced by
the corresponding kriged value ZtK, d. the conditioning relation (VII.3).
Such an approximation will slightly modify the structural variability of the
simulation Zw but is quite acceptable provided that the number of altered
simulated values is small.
Correlation PL - ANi. Except for their nugget constants, the direct and
cross-variograms on Fig. VII.6(b and c) can be deduced one from the other
by a multiplicative factor. This suggests a linear coregionalization model
with a single basic model, d. formula (111.25), i.e.,
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 523
t; Po PL AN;
m 238 262
2 0
D 2/ m 0-08 0-175
m 117 112 16-7
2 2 10
D /m 030 046 0-51
PP" -PL
o 0-83
10 -029 095
WWW.minebook.blogfa.com
524 MINING GEO$TATISTICS
(a)
r-
.--
I-- \----
f-- - -
-- -
r--
r-- - .---
-- -
f--
I-- I- --
- --
- - -- r-r-
I- --
f--
- - f--_
- l-
- ~
-
I- r-r-
I--
I\-- I- l - I--
-- -- l -
o I 10 20
Ft=rl-r
30
Metres
(b)
40 -- ---------
30
lo..
20
10
010 100 200 300 400 500
hem)
VNNI.minebook.blogfa.com
VII S{MULATION OF DEPOSITS 525
(a)
-
5 - -
F' F
-
-
- .
I-
h LP-r-F1ft -~r-
01 5 10 20 30 40
Metres
( b) --- .-
--
20 ....
I I I I I I
o 10 100 200 300 400 500
him)
WWW.minebook.blogfa.com
526 MINING GEOSTATISTICS
C(O)~ C(h). U"'" log Z is then a normal variable with the following characteristics,
given by formulae (VI.9):
mean, E{U} = m' = log m-of 2 / 2 ;
variance, Var {U} ""- (T12 = log (1 + (T2/ m 2 ) = K(D);
covariance, K (h) = log [1 + C( h)/ m 2];
serni-variogram, 'Yu(h)= K(O)-K(h).
Suppose a spherical model with range a I, nugget constant K o and sill K] is adopted
for 'Yu(h), i.e.,
3 h 1 h.l)
yu(h) = x; + x, ( 2 ~- 2: a~' \;f h e ]0, aJ. (VlI.26)
The total sill is, thus, K (0) = K o + Kj, and the corresponding covariance is given by
3
3 h 1h )
K(h)=K(O)-'Yu(h)=K I ( 1---+- VhE ]0, ad.
2 al 2 3
aj
,
VNNI.minebook.blogfa.com
VII SIMULAnON OF DEPOSITS 527
conditioning stage when the simulation is limited to the precise polygonal contour
defined by the 176 conditioning drills on a 100 m x 100 m grid.
The turning bands method (cf. section VIl.A.2) Each of the 15 generating lines D
are considered as NGMAX::= 300 points spaced at a constant interval of u ::::: 10 m.
Next, 300 bands (slices) are centred on each of these 300 points, See Fig. VII. 10.
The number NGMAX = 300 was chosen so that each simulated point of the square
of dimensions 200u x 200u would correspond to a band on each Qf the 15 generat-
ing lines, for which it is enough that NGMAX u > 200u,h (length of the
diagonal).
For each line D, the subroutine RANDU(SSP-IBM) is used to generate 300
independent realizations of a random variable T with a uniform distribution on the
interval [0, 1]. By definition, this variable T is such that E{T} = 1/2 and Var {T} =
1/12.
r Inlerval 2R +1 --I
I
I
I
I
I
1--1.1
I
I
1
I
'
D
--+- Bond
I
WWW.minebook.blogfa.com
528 MINING GEOSTATISTICS
where
Eo=k p T n - k p / 2.
The expectations of these two nugget effects are satisfied: E{e p } "" E{e a } = 1 and, by
identifying the variances, the parameters k p and a can be determined:
Var {E p } =:: k~ Var {Tp } = k;/12 = Ch i.e., k p = 069;
Var {eo} = k; Var {To} = Co;
Var {lOa} = Var {Ep}+ a 2 Var {Eo} = CO + a 2 C O = aaCn = 0-06, i.e., a = 0707.
In addition, the nugget constant aapC) of the relative cross-variogram
"Y A-pJ rna mp can be verified:
since aap = 1.
Simulation of the regionalized variables z and z' The semi-variogram y(r) of the
RF Z consists of two nested spherical models: y(r)= C 1Y l(r)+ Cz'Yz(r), d. model
(VII.25). Thus, the required simulation can be achieved by simulating two
VNNI.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 529
independent RF's Y r and Y2 on each of the 15 lines D and then taking the sum
Y=Y I + Y 2:
Y r =: krn l = kj(TI * fr) and Y z = k 2n 2 = k z(T2 * 12)'
The simulated range (2R + l)u must approximate the real range and, thus, R is
an integer equal to a/2u. For the first structure )'1 with a range a r = 80 m, R, =:
80/20 =: 4 and the simulated range is, thus, a l = (8 + 1)10 = 90 m, which is a close
enough approximation of the real SO-m range. Similarly, for the second structure )'Z
with a range a 2 =: 800 ill, R z =: 40 and the simulated range is a; = 810 m. Note that
the two ranges a! and a 2 can be reproduced exactly by altering the spacing of the
simulated points along the lines D. In practice, the approximations (ai, a;) and
(a2, a2) are largely justified, as can be seen by comparing the variogram on Fig.
VII.1 L
As the variables T I and T 2 are regularized by an odd weighting function r(x) = x
on the interval [-R, +R], the expectations of the RF's 11 r and Hz are zero:
E{n 1} = E{fl 2 } = 0, 't:/ R,
which entails
The parameters k, and k 2 can be determined from the variances Var { Yd and
Var {Yz}. For a range a = 2R, the regularized RF n can be expressed according to
(VILIO) as
and, thus,
Now,
E{ T(l + x)T(t + y)} = m~+ (Ti. Qq,
where
I if x = y,
mr = E{T} = 05, Var{T}= 1/12 and 8..(y)= { 0 ifnot.
The variance of the RF Y = kO, can then be deduced from this latter variance:
(2R + 1)'
Var{Y}=k 2Yar{0'}=k 2
144
WWW.minebook.blogfa.com
530 MINING GEOSTATISTICS
This last relation can be used to determine k. For t; = 0 and for the structure 'Yl
with a sill C 1 =0'015,
(2 R + 1)3
C 1 = 15F1 I
144 an d kI = 1405x 10--2 .
For the second structure 'Y2 with a sill C2 = 0,036,
(2R + 1)3
C 2 = 15k~ 1~4 and k 2 = 0806 X 10- 3 .
( 0)
---
40 ---_ ........ --~
30
>-..
20
( b)
70
---~----_...--
50
30
o 10 100 200 300 400
VNNI.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 531
550 m ~ I ~u
~~,~,- ~~~~I -~"o m m ,
11m ~
----.----- ~
'<----------
Zone 0
1m II m
~
/-/ I m
t ----(lxlxO 5)
- I
- I
I I I
10 m + I'
i I
I
0-5 m;
LY Pit
Ponel
WWW.minebook.blogfa.com
532 MINING GEOSTATISTICS
of a vertical profile of grades, ct. Figs ILl and VII.l3. The second term -
R (u, v, w) - represents the residual fluctuations around this drift.
We impose the condition E{Z} = E{D}, which entails that E{R} = O. The
realizations d(u, v, w) and r(u, v, w) of the two RF's, drift and residual, are
simulated independently. In this particular study, it was assumed that the
elementary grades followed a lognormal distribution, and so a Gaussian
transform was used to ensure that the final realization z = d + r produced a
lognormal histogram.
Simulation of the drift The vertical drift along each profile of grades
represents a progressive enrichment followed by an impoverishment, and,
finally, a further enrichment toward the bottom of the deposit, as shown on
Fig. VII.l3. The following RF was chosen to characterize this drift:
D(u, v, w)=K(u, v)[A
3
w 3 B(U,v) +A2W2B(u,v)+A1WB(U,v)J, (VII.31)
where K(u, v) and Biu, v) are two stationary and independent RF's
defined in the horizontal plane (u, v), and A}, A2 , A3 are three positive
constants.
The cubic function f( w) = A3 w 3 + A2 w 2 + AI W is, thus, modulated by the
two RF's K (u, v) and B (u, v). The multiplicative factor K (u, v) defines the
maximum value f(wo) of the function f(w), d. Fig. VIL13(a) and (b). The
power Btu, v) defines the vertical coordinate Wo of this maximum, d. Fig.
VIL13(b), (c) and (d).
The horizontal variabilities of the two independent RF's K(u, v) and
B(u, v) are characterized by isotropic spherical semi-variograms without
nugget effect, with identical ranges equal to a = 100 m and sills C K and Cn.
Thus, two sampling pits less than 100 m apart on a horizontal plane will
have correlated drift curves d(u, v, w) and d'(u, v, w), while pits further
than 100 m apart will be independent. The expectations and variances (sills
Wo --- - -~-
10 -- 10 -
(a) ( b) (c) ( d)
FIG, VII. 13. Various realizations of the random drift. (a) K = 2, B = 1; (b) K = 1,
s )., (c)K=l ,. B=Og'(d)K=1
, .B=4
,.
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 533
C K and CB ) of the two RF's K(u, v) and B(u, v) were chosen so that the
random drift D(u, v, w) accounted for a fixed part of the total variability of
the grade Z = D+ R.
The positive constants ;\], ;\2, ;\3 fix the form of the cubic f( w) =
3 2
;\3 w + A2 w +;\ 1 w, i.e., the general form of the drift.
WWW.minebook.blogfa.com
534 MINING GEOSTATISTICS
Grode (%)
05 /
I
I
I ....
<'" I
I
I
.... I
I I I
I
( \ I
.... \ I
") -,
I
-, I
.... -,
,
"- 1
">
/ I
-.. I /
/
1 1 /
/
5 ..... 1
I
I
0
I
I
\
"-
',2 m
I
\
I
\
I
I ~/~} I
k"- \
;' I \
/
I
;'
;'
I
;'
-..
..- I
I
\
\
j "- ,
10
FIG. VII.14. Simulated profiles of block grades and central pit grades.
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 535
method. Kriging is then used to condition the simulation to the real data
values, and the Gaussian transform Z = cpz( U) of this conditional simula-
tion restores the required histogram. Finally ~ the simulation is verified by
means of a number of statistics, such as calculations of the direct and
cross-semi-variograrns, histograms, scattergrams, etc.
WWW.minebook.blogfa.com
536 MINING GEOSTATISTICS
Program SIMUL
Program SIMUL below handles NS independent Gaussian components
defined in spaces of the same dimension (one, two or three). Each time a
space of different dimension is required, the program should be rerun.
However, the program can be easily modified to accept components Vk(x)
that are defined in spaces of different dimension.
Each of the NS components Vk(x) has an elementary covariance model
that is either spherical or exponential. The nugget effect is represented by a
zero range covariance, and is treated separately.
with b = a/2R.
WWW.minebot)j(.blC)9f.. com
VII SIMULATION OF DEPOSITS 537
C(1)(S)=_l_(l- Ab C""2Ab) e- A S ( l - J\ s )
d 12A A
1
+--e Ab(l -e --2Ab)] e --As
-2AbAb e"2Ab[ ~ 1 +-
12A A A '
where A = 1 - e- 2 At,. The second term, which represents the additive bias,
is negligible and each value Yi can be corrected by the multiplicative factor
CK to obtain the imposed variance C(J \0) = 1. Thus,
WWW.minebook.blogfa.com
538 MINING GEOSTATISTICS
SUBROlTINE Sl~UlC~S9A,~ll,~CC9~IV,(lyCC,CN
1,NHAS 9 I UT, Y9 CX, US, CS, G)
C
c ......
C ThREE-DIME~SIONAl SI~ll~lIO~
C SLBRCUTINE OF NS INDEPENDE~T G~LSSI~N ~A~IATES
C (~OR~AL 0,1) WITr EXPONENTIAL OR SPHERICAL
C SetH!ES (ISOTROPIC IN THREE CIJilENSICNS) RESULTS
C ARE STOREC ON A CIRECT-ACCE$~ FILE
C
C PARAJI'EIEFiS
C Y ItIiORJ< ING vsc HR:
C l-CO~TAINS ThE RANOCM V~LlES Ie BE
C CILLTEO eN KO TER~S.l~KE A DlJilE~SIOh
C {RE~TER T~AN 8*NR+l
C ~R=RANGE/SUBDIV]SIOh CN BA~DS
C SUBDIVISION IS EOLAL to MIN. (F Cl,CC,Dh
C 2-CC~TAINS T~E VALUES Of t~E ~S SI~llATIQhS
C FeR A CClUMN OF NLI TERMS:~ChlJ.~S)
C ~S NUMBER Of INCEPENDENT SI~lLATICNS
C CXCNGMAX~15*NS) WORKING \lE(TCR CC~TAINING THE
C VALUES ON T~E TURNIhE EAhOS:
C ~GMAX IS EOUAL TO T~E NUMBER (F PCINTS
C SIMULATED ON A TURNI~f BAND=DI~fC~AL
C CF TOTAL PAR~LLElEPIFEC ~LI,~CC,~IV
C COU~TED IN NUMBER OF 5UBDIVISIOhS +6
C A(NS) RANGES: >0. SPHERJC~l SCHEfoIE
C =0. hUGGET EFFECT
C <0. EXPONENII~l SCHE~E
C hlI9NCO.~IV NUMBER Of LINE5,CGlUMNS,lE~ElS OF
C SlfIIUlATED GRJC
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 539
NNI.minebook.blogfa.com
540 MINING GEOSTATISTICS
IH=I"~N"'AS-t18R
lS=O
CO 200 IS=l,NS
LStlS)=Q.
CO 200 JS=l,NS
lS"'lS+l
200 CS(LS)=O.
C
UN=AMINICDL,OC,ONJ
W~ITE(IOUT,2000) UN,NlI,Cl,NCO,CC,~I\,ON
Cl=Ol/UN
DC=OC/UN
ON=DN/UN
NGMAX=SQ~TCNlJ*NllOl*Cl.~((N(O.CC[C+
INIV*N1V*ON*ON)+6
WRITE(IOUT,2aOl)hGMAX.~S
c
C SIMUlATICN ON 15 REFERENCE BANes
c
IP=O
DO 3 IS=l,NS
IF(A(IS).NE.O.JGC TO 3C
C ~UGGET EFFECl
AA=O.
I PAS =:3
NR=l
EPS= o.
PAS=UN I I PAS
GO Te 302
c
30 AA=AfSlA(lS).
NR=AA/(2*UNJ
I PAS = 1
lFfNP.GE.20)GC Te 301
NR=MAXO(l,NRJ
CO 30e IPAS=2,20
IF(N~*IPAS.GE.20)GO TO 3C1
300 CONTItolUE
301 PAS=Uh/IPAS
EPS=P"S/.6A
NR=NP*IPAS
IFIA( IS).l T.O.)GG TO 31
C SPHERICAL S(~E~E
302 CK=StPTI36./N~/(NR+IJ/12.~~+1)J
KD=2.f\R-t1
DO 30: 1<=1,1(1::
3a3 F(K)=I<-NR-l
INO=O
GO Te 33
C EXPCNENTIAl ~(HEME
31 KO=8*M~+1
DO 32 1(=I,KO
XT=EPS*(I(-lJ
VNIW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 541
32 FIKJ=fl.-XTj*EXP(-XT)
Cl=1.-EXP(-2.*EPS)
CK=SQRT(12 (1.Cl/(CI-EP~*EXP(-2 EP~))
JND=1
c
33 CO 34 10=1,15
[;0 340 K-=l,Ke
tH:IAFl*lH-tlBR
340 Y(K)=IH*OIV
C
C HEW POINT: DILUTION O~ KO R~NCO~ ~AlUES
C
DO 3~ ~ .. ltNGMAX
IP-JP+l
AO-O.
KK-MOOCCJ-1J*IPAS,KOJ
DO 350 K-l,KO
IFCK~.EO.KOJKK~O
ICKaKK+l
350 AO-AO+FCKj*YCKK)
C TRANSLATE RA~[[M ~EASl~E
DO 351 K-l,IPAS
JFCKK.EQ.KDIKK-O
kK*KK +1
IH-JAfl*lh+IBR
3'1 YIKKJ-(H*O(Y
ADaAC*CK
USCIS)aUSCIS)+AD
DXU ')"AD
DO 352 JS-l,IS
IJS- I S+N S* J 5-1 )
JP-IP+NGMAX*15*(J$-ISJ
~52 CSCIJSI-CS{IJSI+AO*OX{JPJ
35 CONTINUE
34 CONTINue
IF(IINC.EU.1 JGO TO 36
WRITE(IOUT,2002JIS,AA,PAS,EPS,IPAS,C_
GO TO 3
36 WPITElIOUT,2003)JS,AA,PA!,EFS,IPAS,C~
3 CONT lihUE
IIlR ITE HOUT, 20Q4)( IS, IS"l,h ~)
DO 31 IS-l,NS
IIS-IS+NS.C JS-1)
CS (J I S J- Cesc J I S J-U SCI S ) *l' s CI S) I U 5 *~ G.. .A X)) I
1115*HGMA)()
USCIS)aUSCJS)J(15*NG~A~J
00 3 10 J S- I. I S
JflJS.EQ.IS1GO TO 370
IJS- J S+HS* CJS-l)
JJSaJ S+NS* CJS-lJ
CSCIJS)-tCSlIJS)/CIS*NG"')J-USCIS USeJSJ)1
lSQRTCCSCllS)*CSeJJS)
370 CONTINue
WWW.minebook.blogfa.com
542 MINING GEOSTATISTICS
~~lTEII0UT.2005)JS.US(ISJ,CCS(IJS).
IIJS-l!, II S, NS
37 CONT )HUE
c
C THREE-CI~ENSIONAl SI~UlAlt(N ON NI\ LEVELS
C ~cc CClUMNS ~NC ~LI LINES
C
fl,1=CNlhU/2
~2=(N(O+11/2
fIo3 zCNIV+ll/2
"'G=NGJlI~X/2
IK=O
LS=O
CO 38 ISsI,NS
USCIS)sO.
CO 380 JS"'I,NS
L 5=l 5+1
380 CSClS)=O ..
DO 38 1=1,28
IK= (1<.+1
38 GCIKJ=O.
SQ15=SCRl( 15. J
c
I AC-= 0
NP=Ntt*NCO ... NIV
c
C NEW POINT:PRCJECT ON THE 15 BANDS .~C SU~
C UP CCNTRIBlTlr~S
(
CO 4 JlI:Il'l,NIV
X(3)=-O.5+CM-N31*ON
DO 4 J=l,NCO
X(21=-O.5+(J-~2)*OC
00 40 IsI,NLI
XCl)=-O.5+(I-NIJ*OL
DO 400 IS=l.NS
11=1+t\ll*C IS-l)
400 y{ n J :x0.
INOzO
CO 41 IR'*l,5
kO"" t IR-ll*3
CO 41 JOsI.3
XItJtJ=O.
00 42 IL a l . 3
INO=[f\C+l
42 XltJOJ=XI(JO).X(ILJ.SX(I~n)
ARG=X I IJC) +0.5
LK==A ~G
IFIARG.LT.O.JlK-LK-l
lK=U( .fIoG
KO=KO+l
CO 43 IS=I,NS
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 543
12=LK.NG~AX.(Ko-1).NGM~X.J!.(IS-l)
112:[+Hr*CIS-l)
43 YCIl):VCI1)+O)CI2)
41 CONTI~UE
00 44 IS=l,NS
II = I +fIIL 1+. 1S-1 )
V(ll)=YCll)/SC15
USCIS}=USC1S)+Y( 11)
DO 4" J S= 1, 1 S
12= 1.NL 1* ( J $-1 )
tJS= IS.NS*CJS-l)
44 CSCIJS)=CS(IJS)+Y(11).~(1~1
40 ccsr INUE
c
C ~RITE COlU~N CN DIRECT ~CCESS filE
C
00 45 IS=l,NS
11= l.~l It 15-1 )
12=Nl I* I S
IAOl=I~O.NP.(IS-l)
45 NRITE DISC IUT,IAOl,CYCIJtl=(I,I2)
lAO: IAC"f'.l I
c
C COMPUTE SE~I-VARIOGR.M
C
KV=O
CO 46 IN:el,2
GO TO (461f4~2.tIN
461 11=1
12=19
IP=l
GO TC 46:
462 11=20
12=100
IP=lO
463 00 46 K-l1,I2tIP
KL=NlJ-K
IFCKL.lE.OJGO TO "
KV=KV+l
00 464 la1,Kl
CO 464 IS==!yt\S
Jl=[+l\lI+C 15-1)
J2 z J l +K
T=yeJl)-yeJ2J
IK*[ 5+~S*(KV-ll
464 GCIKI=GCIK).T*r
46 CONTIft.UE
It CONTlhUE
c
c eDIT 5T_115TIC5 AND V.RIOGP'~ ALCNG CCLOMNS
C
WWW.minebook.blogfa.com
544 MINING GEOSTATISTICS
CO 5 IS"" 1, NS
r rs- IS+N5*( IS-II
CStI ISJ=(CS( llS)-USI IS)*\..5 ClS)J~PJ ''''P
USIIS)=UStISJ/NP
00 51 JS=l,IS
IFIIS.EQ.JSlGC TO 51
JJS:z J S+NS* (JS-l)
IJS= I Si'N S* (JS-l J
CSCIJS)=(CSIIJSJ/NP-USCI5J'USIJSJJJ
lSQRTICSIIISJ*CSIJJS})
51 CONT INUE
WRITEIIOUT,2005)IS,USIISJ,ICSIIJSJ,
1 I J S"" J S , I IS, NS J
5 CONTINUE
IMP'" INS-1)/U!+l
JSM~lOAT{NS)/FlOATIIMP).C.q99~
00 52 IM-l,lMP
151= 1+151'.( IM-l)
IS2""MINOINS,IS~*I~)
'-IRITEtIOUT,200BJ (15, IS""Hl ,152)
00 5 3 I< '" 1 , I< V
IKl""IK-l)*NS.. ISl
11<2::: IK-l)*NS+JS2
00 54 IK""IKl,IK2
NC=NIV*NCO*INlJ-K)
5~ GIIKJ""O.5.GIIK/NC
O=K.U~Ol
IfCK .GT.20JOzUN*Ol.UO.K-180)
WRITE(IOUT,2001)K,D,NC,{GIIKJ,IK:I~l,I~2)
53 CONT IHUE
52 CONT INUE
C
2000 FORMATIIH1,' CAR.CTE~JST[CS OF T~E "
I'SIMULATION t/lH ,5X,l~I'.'J,lX,' '.
210I'.'J/JIH " SUBDI~ISIC~ O~ BAhO~~',F1.3.
310X,'SIlE OF GRID: ',14,' LINES Ol =:',
4Fl.3/1H ,56X,I~,' COLU~NS DC ~t,Fl.;/lH ,
556X,I4,' lEVELS ON - ',Fl.3J
~OOI FO~"A1Cl~O,' N~MBER Of P6t~lS 51"llA1ECJ'
I,'TURNING BAND:It ',IS,' P\tI'BER. Of IN[EPENOEN1'
2,' SI'UlATIDNS ~ ',I211t
2002 fORM_Tll~ ,'.SIMUlATIO~.,J3,'
5PHEP.ICAl '.
I' SCHEME RANGE =',FS.4,' LAG ='.F(.~,
2' PRECISION "',FS.3, LAGS/SURD. a',I4,
3' FACTOR CK ""',F8.4)
2003 FORMAT(l~ ,'-SIMUlATION.,13,' EXFCNEhTIAl '.
I'SCHEME RANGE "',Fe.4, lAG =',F.?,
a' PRECISION ~'.F5.3,' lAG~/SU8D. 2',14,
3' fACTOR CK s',F8.4J
2004 FOR"AT(l~O,' STATISTICS CF RESllT~ C~ BANDS'
1.'(CORRElATION MATRIX)'/J~ ,2X,lOf'-'.,
2' -- ------- -- -----'/l~ ,'SIMUlA1ICh'/lH ,
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 545
NNI.minebook.blogfa.com
546 MINING GEOSTATISTICS
(v) A study of blending from various stopes to stabilize the tonnage and
quality of production (grades, hardness, mineralogy). A study of the
requirement for blending stockpiles, d. J. Deraisme (1977), whose
work is partially presented in the following case study.
(vi) A study of the determination of production rates and cut-off grades,
which vary in time, and their impact on recovered benefits, d. P.
Dowd (1976).
It must be understood that results obtained from simulated deposits will
apply to reality only to the extent to which the simulated deposit
reproduces the essential characteristics of the real deposit. Thus, the more
the real deposit is known, the better its model will be, and the closer the
conditional simulation will be to reality. As the quality of the conditional
simulation improves, not only will the reproduced structures of the vari-
ability become closer to those of reality, but so also will the qualitative
characteristics (geology, tectonics, morphology, etc.) that can be deter-
ministically introduced into the numerical model. Once again, it must be
stressed that simulation cannot replace a good sampling campaign of the
real deposit.
t To avoid confusion, the term "simulation" will be used here only for the simulation of the
various methods of mining and blending; the simulated deposit will be referred to as the
numerical model.
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 547
4 I-
r- -- -
...
--r-
-
I-- f - -
f--
- -: --- ,......
3 - - I--
.---- - I--
-- F=
- =- -
-f--
-
- I-
~
-
-
-~
a 05 1-0 1-5 2:0 25
0;., ClJ
FIG. VII .15. Histograms of true and kriged block grades. , True grades,
m""I%Cu, D 2(v/G)=O235%2; ----, kriged grades, m*=I%, D~*(vIG)=
0'185%2,
WWW.minebook.blogfa.com
548 MIN IN G G EOSTA TISTI CS
not significantly different from his prediction. The short-term (e.g., hourly)
fluctuations of the mill feed grade are the responsibility of the metallurgist
and the process controller.
It is assumed here that grade is the primary variable affecting mill
operation. Any other regionalized parameter (impurity grades, crushing
indexes, etc.) can also be simulated and used in an analogous manner.
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 549
450 m
A------------7
1"'-18 m
Mill
(0 )
Mill
(b)
FIG. VIL16. View of the mining fronts (blocks 10 m high). (a) Method 1;
(b) method 1'.
Simulation of the mining processes Only method 1, for which the path of
the big shovels is entirely determined, is completely simulated by
computer. For methods 1 and 2, the daily decisions such as which block to
f
mine, and whether to send it to the mill or to the stockpile, are made by an
operating mining engineer,' just as they would be in a real situation. The
decisions were made through visualization of the state of the fronts and the
stockpiles, taking the past and immediate future mill feed grades into
WWW.minebook.blogfa.com
--:.,
</)
""'o0
:0
...c
....1::>0
.l::
l
Sl
V)
'-'
o:
.-
c
0
.....I-<
00
e
';::
's<l.l
...c
....
....0
~
<l.l
:;
r-..:
.....
....
......
E >
s
-
0
u..
WWW.minebook.blogfa.com
VII SIMULATION OF DEPOSITS 551
account. The decision of the mining engineer is certainly not always opti-
mal, but the object of this study is precisely to reproduce reality (including,
in particular, human errors) and not an absolute optimum which is often
inaccessible in practice.
The engineer makes all his decisions on the basis of kriged estimates of
the mining blocks, the blocks being kriged from the central drill core
intersections. However, it is the true block grades that are delivered to the
mill, cf'. section VI.A.3.
As for the type of blending performed within the subpiles considered in
method 1', it is assumed that the true mean grade of the unit taken from a
subpile is a normally distributed random variable, with the following
characteristics.
(i) An expectation ms equal to the true mean grade of the subpile at the
time considered. This instant mean grade varies around 16% Cu for
the rich subpile, and 06% Cu for the poor subpile.
(ii) A fixed relative standard deviation as/ ms equal to O 25% 2 for the
rich subpile and O' 3 % 2 for the poor. These values were chosen so
that the confidence interval [ms2asl for the rich subpile, for
example, includes all individual true block grades sent to this partic-
ular subpile during the course of one year.
This model of the blending effect of the stockpile is very simplistic and
pessimistic, but is enough to give an order of magnitude of the influence of
the stockpile on the homogenization of the mill feed grades. In a real case,
it would be a simple matter to replace this model either by a deterministic
system or a more precise probabilistic model,
Results For the three simulated methods of mining zone G, Fig. VILl8
shows the variation in daily mean grades during the first two months (60
production days). J. Deraisme's study was carried out for a year's pro-
duction (280 days) but there is not enough space here to show all the
results.
The full-line curves show the variations of the true daily grades (i.e., the
homogenization actually achieved), while the broken-line curves represent
the variation of the estimated daily grades (i.e., the homogenization pre-
dicted on the basis of the kriged block grades).
The mill operator will obviously prefer methods 1i and 2 to method 1,
which the mining engineer may prefer because of its lower mining cost.
Thus, there is a real problem involved, which can only be solved by
assigning costs to the mill recovery losses caused by method 1, and to
balance these costs with the ones caused by stockpiling (method 1') or by
the more flexible mining operation of method 2.
NNI.minebook.blogfa.com
552 MINING GEOSTATISTICS
,r- "-"I
Melhod I
15
10 ., J- I \'. \
.'~
\ I
I \;.
\..J
\ ~
1.- \
~
I ~./
"'..
'\'.
\,
.\
" /.
I
.J
I'
05 ,.:;.' \
10
05 l_------'---_----.-J'---_-'-----_-----'-_------'--_----.-J_
I I 1 I 1 I
_---'--_-----'----_ l
........._ I
_J--_,-'-_.........J
I 5 10 15 20 25 30 35 40 45 50 55 60
Days
FIG. VILIS. Fluctuations of the daily mean grades at the mill-feed .- -....., True
grades; - - - -. estimated grades.
WWW.minebook.blogfill.com
VII SIMULATION OF DEPOSITS 553
When there arc two values, the upper one refers to the true grades while the lower refers to
estimates.
The deviation between the dispersions resulting from the different min-
ing methods increases as the time interval decreases. Thus, whatever
method is used, the amount and location of the ore extracted during a
given fortnight will be approximately constant, and the dispersion vari-
ances of a fortnight's mean grade over one year will also be approximately
the same, On the other hand, methods I' and 2 are particularly efficient in
stabilizing the daily grades within any week or fortnight.
The dispersions of estimated grades are always less than those of true
grades. This is a result of the classical smoothing property of the kriging
estimator, ct. relation (VI.l) or (VI.2): the dispersion of estimators is not a
good estimate of the real dispersion. For method 2, even though the
homogenization of the estimated grades is nearly perfect, there is still an
irreducible fluctuation of the true grades, due to the estimation variance of
the daily grades, ct. Fig. VII. 18.
The experimental estimation variances of the daily grades for each
method are:
WWW.minebook.blogfa.com
554 MINING GEOSTATISTICS
Note that for methods 1 and 2, the dispersion variance D 2 of the true
grades can be deduced from that of the estimated grades D~* by the
approximate smoothing relation (VI.2):
D 2-- D 2*
K
+ (TE
2
This relation is not applicable for method I' because daily production does
not come only from the mine but also from the stockpile.
For one year of simulated production, 17% of the production from
method l' passed through the stockpile. This represents a considerable
proportion for a stockpile with a maximum capacity of two days' pro-
duction. In fact, to obtain the homogenization shown on Fig. VII. IS,
almost 50% of production days involved departure from normal pro-
duction (two blocks per bench). One solution would be to increase the
capacity of the stockpile by, for example, adding a third subpile of average
quality are. In any case, it would appear that the capacity of the stockpile
should not be greater than three or four days' production. Method I'
produces dispersions of the same order of magnitude as those of method 2
and since its mining cost is lower (10-m benches instead of 5 m) a feasibility
study would probably result in method l ' being chosen.
Globally, there is no selection involved in any of the methods considered
in this study. However, in his more complete study, J. Deraisme (1977)
studied various selection criteria for method 2 and compared the results of
applying these criteria on the simulations to the results that could be
predicted on the basis of grade histograms alone. The differences were
significant, and illustrate the well-known effects of concentration and dilu-
tion of reserves: a rich block with a grade above the cut-off will not be
selected when it is located in an overall poor zone; conversely, a poor block
with a grade below the cut -off will be selected when located in an overall
rich zone. The prediction made on the basis of the grade histogram (cf. the
various techniques outlined in section VI.A) are unable to take account of
this spatial localization of rich and poor grades; more generally, they
cannot take the technological conditions of the mining operation into
account. Once again, a solution is afforded by the simulation of the mining
operation on a numerical model of the deposit: the simulations can be used
to evaluate the effect of these technological conditions on the actual
recovery of the reserves.
WWW.minebook.blogfa.com
VIII
Introduction to Non-linear
Geostatistics
SUMMARY
The estimators that have been considered until now are linear combinations of the
available data, and their prerequisites are limited to only the moments of order 2 of
the RF Z(x). If more is known about the RF Z(x), for example, if its 1-, 2- or
k-variate distributions can be inferred, then non-linear estimators can be built.
These estimators are more accurate than the linear kriging estimators, and allow
the estimation of unknown functions which are not simply linear combinations of
unknown values.
The word "kriging" designates the procedures of selecting the estimator with a
minimum estimation variance within a given class of possible estimators (linear or
non-linear). This estimation variance can be viewed as a squared distance between
the unknown value and its estimator; the process of minimization of this distance
can then be seen as the projection of the unknown value onto the space within
which the search for an estimator is carried out. In terms of these projections,
section VIlLA introduces the various non-linear kriging estimators such as the
condi tional expectation or "disjunctive kriging".
Disjunctive kriging is developed in section VIILB, together with its application
to the estimation of non-linear functions, such as the so-called "transfer function"
defined as the proportion of unknown values (e.g., grades of support v) greater than
a given cut-off within a given domain V.
WWW.minebook.blogfa.com
556 MINING GEOSTATISTICS
or, better still, considering a unique function of the n available data, i.e.,
z~ = l(z1> ... , z.,},
The linear estimator z~ is but a particular expression of the estimator
ZbK, which, in turn, is a particular expression of the estimator z~.
Consequently, the best estimator of the ZbK type - best in the sense of
minimum estimation variance E{[Z (xo) ~ Zf,K ]2} - is certainly better than
(or equal to) the best linear estimator of the z~ type. Similarly, the best
estimator of the z ~ type is certainly better than (or equal to) the best
estimators of the ZbK or z~ types. Thus, the larger the domain within
which the search for a best estimator is carried out, the more accurate the
corresponding estimator will be for a given amount of data. But there is no
easy way; it is shown in section VIlLA that the gain in accuracy of
non-linear estimators is balanced by a greater complexity, and, above all,
by stronger prerequisites (which are not always met in practice).
But this gain in accuracy is not the only reason for developing researches
in non-linear geostatistics. It has been shown, d. the smoothing effect
,
(formulae (VLl) and (VI. 2 that one drawback of linear estimators is that
they do not reproduce the spatial variability of the true values.
Consequently, it is not possible to derive such features as the proportion of
true values above a given cut-off from the distribution of these estimators;
cf. Fig. VI.3. Two solutions have been proposed.
The first consists in building an estimator of the true grades histogram by
modifying the variance of the experimental kriged grades, d. section
VLA.2. This solution is efficient for the estimation of a global recovery
over the whole deposit, or at least a wide field. But, if a study of local
recoveries of units v within N blocks V is desired, with N around 100 or
WWW.minebook.blogfa.com
VIII INTRODUCTION TO NON-LINEAR GEOSTATISTICS 557
F0K = L /,."(z,,,).
,\:= 1
Similarly, disjunctive kriging allows estimation of the mean grade of the
recovered ore for any cut-off zc, and, thus, provides a direct solution to the
problem of evaluating local recoveries.
It will be noted that the hypotheses underlying the construction of the
estimator F~K and the representativeness of the conditional simulation are
of the same type, i.e., the Gaussian transform Z = 'Pz( U) allows the
reproduction not only of the univariate distribution, but also of the multi-
variate (at least of the bivariate) distribution of the RF Z(x), d. section
VILA.4. This point is important from a methodological point of view, since
it shows the parallel nature of the two approaches of selection and
recoverable reserves - through disjunctive kriging and transfer function, or
through conditional simulation.
WWW.minebook.blogfa.com
558 MINING GEOSTATISTICS
The norm IIZ(x)11 of a vector Z(x) is thus defined as the positive square
root of (Z(x), Z(x, i.e.IIZ(x)11 2 = (Z(x), Z(x. The distance between two
elements Z(x) and Z(y) is defined as the norm IIZ(x)~ Z(y )'1 of the vector
Z(x)-Z(y).
Let Z(xo) = ZoE 'l: be any unknown variable and let <;g' c g be any
vector subspace (or, less restrictively, any closed linear manifold). The
Projection Theorem shows that there exists one and only one element
Z* E 'ie' which minimizes the distance 1120 - Z*II; this unique element is
then called the projection of Zo onto the subspace g', d. Fig. VIII. 1.
The kriging process is simply this projection of an unknown value onto a
particular subspace 'ie', within which the search for an estimator Z* is
carried out. The corresponding minimum squared distance, )120 ~ z*I)2 =
E{[Zo- z*f}, or minimum estimation variance, is then called "kriging
variance".
There are as many kriging processes and corresponding kriging estima-
tors Z* as there are different sets 'l;' c if within which the projection of the
unknown Z(xo) is to be made. Consider the two sets 'j;" and 'if;' on Fig.
WWW.minebook.blogfa.com
VIII INTRODUCTION TO NON-LINEAR GEOSTATISTICS 559
VIlI.2, with 'l:/I c <g' c ?t, and the corresponding kriging estimators Z*/1 and
Z*I; as '$' includes 'is/I, the projection Z*' is nearer to the unknown than
Z*/I, i.e., in terms of estimation variance
I)Z(xo)- Z*,),2~ IIZ(xo)- z*/I11 2
Hence, as the set where the search for the estimator is carried out is
wider, the estimation is better. This remark is a prelude to the classification
of the various kriging estimators, d. section VIII.A.4.
(el
z*
FIG. VIII.1. The kriging estimator defined as the projection of the unknown Z(xo)
onto the subspace g' c g.
z (xo )
FIG. VIII.2. Projection of the unknown Z(xo) onto two subspaces gil c 'C'.
11
I AaZ", +A o1
a~l
WWW.minebook.blogfa.com
560 MINING GEOSTATISTICS
Conditions of unbiasedness
Consider the expectation of any element Z* = Au + La A",Za E 'i,,+ 1:
(VIlLI)
a:
Ao = 0 and L A", = 1.
VNNI.minebook.blogfa.com
VIII INTRODUCTION TO NON-LINEAR GEOSTATISTICS 561
and is the projection of the unknown Z(xo) onto the vector space i"+l> ct.
Fig. VII!.3. This projection is unique and is characterized by the orthog-
onality of the vector Z(xo)- Zk o to each of the (n + 1) vectors generating
i,,+i:
(Z(xo)- ZL, 1) = 0,
(VIII.2)
(Z(xo)- Z~o' ZcJ = 0, Va = 1 to n. }
FIG. VIII.3. Simple kriging defined as the projection of the unknown Z(xo) onto
g'n+l'
NNI.minebook.blogfa.com
562 MINING GEOSTATISTICS
Considering the centred covariance (T~,B = u a/3 - m(xa ) . m(xJ3), the pre-
vious system (VIII.2) can then be expanded into
Ao = m(xo)- I AKoJ3m(X/3),
,B
n
,~ I !
L. AKoper ",{3 = i faro, Va = 1 to n.
13=1
Note that
m(x)=E{Z(x)} = (Z(X), 1).
r.e.,
Kriging variance The kriging variance E{[ Z (xo)- Z~ ]2} is simply the
minimum squared distance IIZ(xo)-Z~oW:
(VIIL4)
WWW.minebook.blogfa.com
VIII INTRODUCTION TO NON-LINEAR GEOSTATISTICS 563
0"
FIG. VIllA. Ordinary kriging defined as the projection of the unknown Z(xo) onto
the linear manifold C.
(VIII.5)
L AKlI'30''''13 - J.LI = (T",X;), Va = 1 to n.
~
NNI.minebook.blogfa.com
564 MINING GEOSTATTSTICS
This system is simply the "ordinary" kriging system (V. 1) most commonly
used, at least in mining practice.
(VIIL6 )
Note that the system (VIII.5), as well as the expression (Vll1.6), can be
restated by replacing the non-centred covariance by the centred covariance
(.r~y = (Txy - m". The unknown value m is eliminated from these equations,
2
The unbiasedness condition (VIII. 1) thus restricts the search for a linear
estimator to the linear manifold CL c t:" defined by the L following condi-
tions on the weights:
VI = 1 to L.
LAKI.,f3f,(xf3)=f/(xo), VI=ltoL,
~ !Ldl(x.)~ ~
(VIII.?)
1 f3
AK,pu." - LI
o:" Va 1 to n.
WWW.minebook.blogfa.com
VIII INTRODUCTION TO NON-LINEAR GEOSTATISTICS 565
Once again, note that the relations (VIII.7) and (VIII. 8) can be written in
terms of the centred covariance a~f3 Of the serni-variogram Yaf3'
Prerequisites to obtain ZL. The form of the drift m(x), i.e., the L
functions fl(x), is assumed to be known, and the statement of the kriging
system (VIII.7) requires the knowledge of the covariance function (centred
or not) or of the serni-variogram. In practice, since only one realization of
the non-stationary RF Z(x) is available, problems arise in the inference of
the non-stationary covariance E{Z(x)Z(y)}, d. section V.A.2, G.
Matheron (1971, pp. 188), P. Delfiner, (1975).
WWW.minebook.blogfa.com
566 MINING GEOSTATISTICS
FIG. VIlIS The conditional expectation E.,zo defined as the projection of the
unknown Z(xo) onto H".
WWW.minebook.blogfa.com
VIII INTRODUCTION TO NON-LINEAR GEOSTATISTICS 567
WWW.minebook.blogfa.com
568 MINING GEOSTATISTICS
estimators, one can look for an intermediate estimator, more accurate than
the linear estimators but with less severe prereq uisites than the conditional
expectation. For this, a projection space included in H" and including it"
must be considered. A good choice for this intermediate space is the vector
space D; generated by the sums of n single-variable measurable functions,
I.e.,
D; = {g\(ZI)+ g2(Z2) + ... + g,,(Z,,)}.
This space satisfies the inclusions 're" cD" c H". The "disjunctive kriging
process" is then, by definition, the process of projecting the unknown
Z(xo) on to D", d. Fig. VIII.6.
FIG. VIII.6. Disjunctive kriging defined as the projection of the unknown Z(xo)
onto D n
WWW.minebook.blogfa.com
VIII INTRODUCTION TO NON~LINEAR GEOSTATISTICS 569
WWW.minebook.blogfa.com
570 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
I
~.
s-
o TABLE VIII. I. Hierarchy of the various kriging estimators
I- Space of projection
J Vector space Vector space
D
Vector space Linear manifold
CL
';n H" => n => if"H =>
{f(Z!> . . . ,Z,,}} {Jl t. (Z,,)} {Ao + A"Z,,} A"Z" + L conditions
; on A"
Kriging estimator Conditional expectation Disjunctive kriging Simple kriging estimator Universal kriging
"Zo (best possible estimator ZbK Zk" estimator ZkL
estimator)
2 2 2 2 2
Estimation IIZ(xo)- "ZoIl2 ",; (TDK ~ (TKo ~ (TK, ",; (TK. ",;. ~ (TKL
variance
Prereq u isires (n + 1 j-variate Bivariate distributions. The (n + I) expectations. The form of the
,
distributions, e.g., In practice, The covariance drift
Z (x ) statio nary and stationarity of Z (x). function. In practice, L
Linear estimators
1-
1:
572 MINING GEOSTATISTICS
The n weights AK w r are determined from the simple linear kriging system
(VIII,3), and the variance E{[ Y(xo)-E n Yof} is simply the corresponding
kriging variance aIfo given by expression (VIII.4).
Consider now the conditional variable Z(xo)/ Z], , .. , Z; =
exp [ Y (xo)]/ Z I, . . . ,Z". This variable is lognormally distributed, and its
expectation EnZo is deduced from the parameters of the distribution of
Y(xo)/ Y i , . . . , Y n through the formulae (VIIL13), i.e.,
"Zo = E{Z(x()/ ZI, ... , Zn} = exp [ y~() + alZo/ 2]. (VIILI4)
The non-linear estimator Z~ = exp [Y~" + a-~oI2] thus appears as the
conditional expectation EnZo which is the best possible estimator of the
unknown Z(x(}) from the n available data Z(x' The corresponding estima-
tion variance is expressed as
E{[ Z(xo)- E nZ oJ2}= m 2 e rr ' 2 (1 - eo-a;?,,). (VIII. IS)
(The proof is immediate by expanding the square [Z(xo)- EnZof)
WWW.minebook.blogfa.com
VIII INTRODUCTION TO NON-LINEAR GEOSTATISTICS 573
Let us recall the results already obtained in section VIILA.3. The dis-
junctive kriging estimator of an unknown Z(xo) from n available data
tz, a = 1 to n} is the sum of n measurable functions of each datum taken
separately, i.e.,
n
Z~K = I fa(Za).
0.= I
Vf3 = 1 to n.
a
v f3 = 1 to n. (VIn.1 7)
a
WWW.minebook.blogfa.com
574 MINING GEOSTATISTICS
It can be shown that the density 0(u" Uj) can be expanded in a series of
Hermite polynomials 11k(U;):
00
I if k == k'
<TJk (Y), TJd V~ == E{ 11k (Y)TJd V)} = { 0 if k ~ k':
Consequently, from the expressions (VIII. 18) and (VIII. 19) of the Gaus-
sian bivariate distribution of Y;, Yj" it follows that
00
feu) = L ATJk(U),
k=O
/k = (f(u), Tlk(U = L: ro
f(U)71k(U)G(du), v
The conditional expectation of any measurable function feu) has the
following expansion:
00
WWW.minebook.blogfa.com
VIII INTRODUCTION TO NON-LINEAR GEOSTATISTICS 575
'Vf3 = 1 to n.
WWW.minebook.blogfa.com
576 MINING GEOSTATISTICS
Thus, the disjunctive kriging process very simply reduces to the solving
of K systems {k = 1 to K} of simple linear kriging type. The kriging matrix
of each system is deduced from the correlation matrix [Pa{3] by raising each
element Pec/3 to the power k.
Kriging variance
Taking account of the orthogonality of the two vectors Zo- Zt>K and
ZbK, i.e., E{[Zo- ZbK ]ZbK} = 0, the corresponding estimation variance
or kriging variance is wri tten as
with
zs
E { 2} =f..,
\~ 2
({'k,
k
It follows that
(VIIL22)
vu 1
(for this, it is enough to take fcr,k = 0, Va, Vk # 1) and
VI3=lton.
a
WWW.minebook.blogfa.com
VIII INTRODUCTION TO NON-LINEAR GEOSTATISTICS 577
The last n equations (for k = 1) are the simple linear kriging system of the
unknown z; d. (VIII. 3).
Thus, the disjunctive kriging estimator is written as
=~
V
f v
Prob {Y(x)~ yj Y b .. , Yn } dx (VIIL23)
with
YC=lj?-l(ZC) and Y a = ({J - 1(Z ,,).
The notation Z(:!) denotes that x is evenly distributed within V. If the
variables Z(x) are of point (or quasi-point) support, the previous prob-
ability P is also called the "point transfer function within V".
The direct calculation of this transfer function P would require the
knowledge ofthe (n + l j-variate distributions of V}, ... , Yn> Y(x), 'Ix E V.
WWW.minebook.blogfa.com
578 MINING GEOSTATISTICS
with
WWW.minebook.blogfa.com
VIII INTRODUCTION TO NON-LINEAR GEOSTATISTICS 579
vu -
Similarly, consider the Kth-order expansion of the estimator PbK :
K "
P'&K = L L
k=O a= I
fa,k TJd Y,J.
The n(K + 1) unknown parameters fa,k are determined from the dis-
junctive kriging systems (VIII.21), i.e., for each k = 0 to K,
n
L p:l3fo:,k = !kTk(V, (3), Vf3 = 1 to n,
0:=1
where
WWW.minebook.blogfa.com
580 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
Bibliography
Abbreviations
CG Centre de Geostatistique, 35 rue St Honore, 77305 Fontainebleau,
France
ENSMP Ecole Nationale Superieure des Mines de Paris, 60 Boulevard
Saint-Michel, 75006 Paris, France.
* The asterisk indicates the works that, in our opinion, are a must in a
somewhat complete geostatistical library.
"Geostat 75" NATO A.S.!. Congress "Geostat 75", Rome, Italy. The
proceedings are edited by M. Guarascio, M. David, and Ch. Huijbregts,
and published by Reidel Publishing Corporation, Dordrecht, Netherlands.
581
WWW.minebook.blogfa.com
582 MINING GEOSTATISTICS
A. MATHEMATICAL METHODS
B. GEOSTATISTICAL METHODOLOGY
WWW.minebook.blogfa.com
BIBLIOGRAPHY 583
WWW.minebook.blogfa.com
584 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
BIBLIOGRAPHY 585
WWW.minebook.blogfa.com
586 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
BIBLIOGRAPHY 587
Forrnery, Ph. (1963). Etude geostatistique du gisement d'Ity Mont Flotuo, Inter-
nal CG report.
"Francois-Bongarcon, D. and Marechal, A. (1976). A new method for open-pit
design and parametrization of the final pit contour in Proceedings of the 14th
International APCOM Symposium, Pennsylvania State University, Oct. 1976,
pp.573-583.
Guarascio, M. (1974). Valutazione dei Giacimenti Minerari: l'approccio geo-
statistico, L'lndustria Mineraria, March, 1-20.
Guarascio, M. (1975). Improving the uranium deposit estimations (the Novazza
case), in "Geostat 75", pp. 351-367.
Guarascio, M. and Raspa, G. (1974). Valuation and production optimization of a
metal mine, in Proceedings of the 12th International APCOM Symposium,
Golden, Colorado, April 1974, pp. F50-F64.
Guarascio, M. and Turchi, A. (1976). Exploration data management and evalua-
tion techniques for uranium mining projects, in Proceedings of the 14th Inter-
national APCOM Symposium. Pennsylvania State University, Oct. 1976, pp.
451-464.
Guibal, D. (1973a). Les fonctions auxiliaires a deux dimensions pour le schema
spherique, Internal, Report N-347, CO, Fontainebleau.
Guibal, D. (1973b). L'estirnation des okournes du Gabon, Internal Report N-333,
CG, Fontainebleau.
Guibal, D. (1976). Elementos de geostadistica aplicada, Apuntes del Serninario
de Minero-Peru, June 1976, Internal report, Minero-Peru, Lima.
Haas, A. and Jousselin, C. (1975). Geostatistics in the petroleum industry, in
"Geostat 75", pp. 333-347.
Haas, A. and Mollier, M. (1974). Un aspect du calcul d'erreur sur les reserves en
place d'un gisement: l'influence du nombre et de la disposition spatiale des
puits, Revue de l'1.F.?, 29, 507~527.
Haas, A. and Viallix, J. R. (1976). Krigeage applied to geophysics, Geophysical
Prospecting, 24,49-69.
Huijbregts, Ch. (I 971 a). Courbes d'isovariance en cartographic automatique,
Sciences de la Terre, 16, 291-301.
Huijbregts, Ch. (1971 b) Reconstitution du variograrnme ponctuel a partir d'un
variogramme experimental regularise , Internal Report N-244, CG, Fontaine-
bleau.
Huijbregts, Ch. (1973). Regionalized variables and application to quantitative
analysis of spatial data, in Proceedings of NATO A.S./. "Display and Analysis
of Spatial Data", Wiley and Sons, London and New York, pp. 38-53.
Huijbregts, Ch. (1975c) Estimation of a mass proved by random diamond drill-
holes, in Proceedings of the 13th APCOM Symposium, Clausthal, West
Germany, Oct. 1975, pp. A.I. 1-17.
Huijbregts, Ch. and Journel, A. (1972) Estimation of lateritic type are bodies, in
Proceedings of the 10th International APCOM Symposium, Johannesburg, April
1972, pp. 207-212.
Huijbregts, Ch. and Segovia, R. (1973). Geostatistics for the valuation of a copper
deposit, in Proceedings of the 11 th International APCOM Symposium, Uni-
versity of Arizona, Tucson, April 1973. pp. D24-D43.
Journel, A. (1969). Etude sur l'estimation d' une variable regionalisee: applica-
tion a fa cartographie sous-marine. Service Hydrographique de la Marine,
Paris.
WWW.minebook.blogfa.com
588 MINING GEOSTATISTICS
WWW.minebook.blogfa.com
SIBLIOG RAPHY 589
WWW.minebook.blogfa.com
590 MINING GEOSTATISTICS
Royle, A. G. (1977). How to use geostatistics for ore reserve classification, World
Mining, Feb. 52-56.
Royle, A. G., Newton, M. J. and Sarin, H. K. (1972). Geostatistical factors in
design of mine sampling programmes, Transactions of the Institution for Mining
and Metallurgy, 81, A81-A88.
Royle, A. G. and Hosgit, E. (1974). Local estimation of sand and gravel reserves
by geostatistical methods, Transactions of the Institution for Mining and Metal-
lurgy. 83, A53-A62.
Rutledge, R. W. (1973). Geostatistics, an introduction, in Proceedings of the
ANZAAS Meeting, August 1973, Perth, Australia.
Rutledge, R. W. (1975). The potential of geostatistics in the development of
mining, in "Geostat 75", pp. 295-311. .
Sabourin, R. (1975). Geostatistical evaluation of the sulphur contents in Lingan
coal mine, Cape Breton, in Proceedings of the 13th International APCOM
Symposium, Clausthal, West Germany, Oct. 1975, pp. I. II. 1-16.
Sandefur, R. L and Grant, D. C. (1976). Preliminary evaluation of uranium
deposits. A geostatistical study of drilling density in Wyoming solution fronts,
in Proceedings of the 1976IAEA Meeting, Vienna, Austria.
Sichel, H. S. (1972). Statistical valuation of diamondiferous deposits, in Pro-
ceedings of the 10th APCOM Symposium, Johannesburg, 1972 pp. 17-25.
Sinclair, A. J. a nd Deraisme, 1. (1974). A geosta tistical study of the Eagle coppe r
vein, N. British Columbia, CIMM Bulletin, June, 1-12.
Sinclair, A. J. and Deraisrne, J. (1975). A two-dimensional statistical study of a
skarn deposit, Yukon Territory, Canada, in "Geostat 75", pp. 369-379.
Serra, J. (1967b). Un critere nouveau de decouvertes de structures: Ie vario-
gramme, Sciences de la Terre, 12, 275-299.
Serra, J. et AI. (1974). Laws of linear homogenization in ore stockyards. Pro-
ceedings of the 11 th International mineral Processing Congress, Cagliari, Italy,
April 1975; Report N-387, CG, Fontainebleau.
Tulcanaza, E. (1972). Personal communication.
Ugarte, 1. (1972). Ejemplos de modelos de estimacion a corto y largo plaza,
Boletin de Geoestadistica, 4, 3-22.
Walton, J. (1972). Resumen de un estudio de estirnacion de bloque grande de
block-caving, Boletin de Geostadistica, 4, 43-54. Universidad de Chile,
Zwicky, R. W, (1975). Preliminary geostatistical investigations of tar-bearing
sands, in Proceedings of the 13th APCOM Symposium, Clausthal, West
Germany, Oct. 1975, pp. I. III. 1-15.
WWW.minebook.blogfa.com
Index A.
Geostatistical Concepts
A Cokriging 324,335,340
Combining terms of estimation 411,
Absence of correlation, see also Pure 441
nugget effect 36, 152 Conditional non-bias 458, 462
Accumulation (regionalized vari- Conditional expectation 457,565
able) 199,281 Conditional positive definite
Accuracy of a selection 458, 463 function 35, 161
Anisotropy 37, 175 Conditional simulation, see also Simu-
A priori variance 31, 37, 42, 78 lation of deposits 492,494
Authorized linear combination 36 Confidence interval 14, 50, 57,60
Auxiliary functions 108, 117 Convolution product 504
Coregionalization 40, 47, 171, 253,
B 256,516
Behaviour at infinity (of a Correlation coefficient 41, 256
variograrn) 39 Correlogram 33
Behaviour near the origin (of a vario- Covariance 31,34,36, 194
gram) 38,44,232 Covariogram, see Geometric covario-
Bias (of an estimation), see also Non- gram, or Transitive covariogram
bias condition 49 Cross-covariance and variogram 40,
Bivariate distribution of true and esti- 171,260
mated values 457, 465
Bivariate Gaussian distribution 573 o
Border effect (on the dispersion) 64
Border term (of the globai estimation Data file 356, 361
variance) 437,442 Deconvolution 90,231,270
Discontinuity at the origin of a vario-
c gram, see also Nugget effect 39
Discrete approximation of a block 96,
Cauchy algorithm (for integral cal- 511
culation) 99, 109 Direct variogram, see Variogram
Cauchy-Gauss method (for calculation Disjunctive kriging 567,573,575
of mean value ii) 98, 106 Dispersion of grades 68, 70, 74, 297,
Charts (for auxiliary functions) 116, 491
123 Dispersion variance 61,67,157,246,
Class of angles and distances, see Grou- 297,450
ping into- Drift 40, 271, 313, 331, 500, 530
591
WWW.minebook.blogfa.com
592 INDEX A. GEOSTATISTICAL CONCEPTS
VNIW.minebook.blogfa.com
-----
INDEX A. GEOSTIATISTICAL CONCEPTS 593
WWW.minebook.blogfa.com
594 INDEX A. GEOSTATISTICAL CONCEPTS
S - to be estimated 15,350
- of selection 21, 445
Scale effect (on the observation) 148,
152, 149
Scatter diagram 204,467 T
Screen effect (on the estimation) 310,
Transfer function (direct or
331,392
indirect) 577
Section term (of the global estimation
Transition phenomenon 37, 163
variance) 422,442
Transitive theory 411, 429
Selection (of grades and ore) 21,444,
Turning bands method (for
483
simulation) 498,508,526
Sill (of a variogram ) 37,67,78,231
Simple kriging 561
Simulation data base 502 U
-of blocks 511
Unbiasedness 49, 305,458
-of deposits 17, 491, 519
Univariate distribution 78, 468, 508
- of mining processes 17, 492, 548
Universal kriging or Unbiased kriging of
Slice term (of the global estimation
order k 313,331,393,564
variance) 422
Sliding neighbourhood 34, 186, 321
Smoothingeffect (of kriging) 313,450, Y
465, 553
Yariogram 11,31,34,162,207,227
Spatial la";" of a random function 31,
Variography, see also Structural analy-
510
sis 195,227
Sf?herical model 116,163,237,507
Vein type deposits 405
Stationarity 30, 190, 500
Stationarity of order 2 32
Statistical inference of moments 192, W
207
Weighted average, see Moving average
Stockpile (for homogenization) 548
Stratiform deposits 386
Structural analysis 12, 148, 195,236 Z
Structural anisotropy or isotropy, see
Zero effect (on the calculation of 1') 96
Anisotropy or Isotropy
Zonal anisotropy 181, 266
Support of the information 77, 148,
Zone of influence 13, 37
199,231
VNNI.minebook.blogfa.com
Index B. Quoted Deposits
595
WWW.minebook.blogfa.com
Index C.
FORTRAN Programs
596
VNIW.minebook.blogfa.com
-----
Index D. Notations
The notations appear as they do chronologically in the text.
(au, Ov, Ow) indicates the reference system of rectangular coor-
dinates.
I
x, or x indicate a point of coordinates x or x"' (x"' xu),
(x"' xu, x w ) according to whether a one-, two- or
three-dimensional space is considered.
h, or hi indicate a vector of modulus Ih 1= r, and three-
dimensional coordinates (h u , h.; h w ) .
a generally indicates a direction in space, but can also
be a data index.
indicate one-, two- or three-dimensional space,
respectively. Thus x E R 3 indicates a point x taken in
the three-dimensional space.
ReV abbreviation for regionalized variable.
RV abbreviation for random variable.
RF abbreviation for random function.
z(x), y(x), a(x), ... in lower-case letters, indicate regionalized variables
or realizations of random functions.
Z(x}, Y(x), A(x), ... in upper-case letters, indicate the corresponding
random variables or functions.
E{Z(x)} = m(x) indicates the expectation of the RF Z (x).
Var{Z(x)} indicates the a priori variance of the RF Z (x).
C(h) indicates a stationary covariance function.
y(h) indicates a stationary semi-variogram function.
p(h) indicates a stationary correlogram function.
Aj , Aj , or A", A~ indicates the weights of linear combinations.
a generally indicates the range of a variogram model.
C, or K generally indicate the sill of a variogram model.
Ckk"(h) indicates a stationary cross-covariance.
"Ykk,(h) indicates a stationary cross-serni-variogram.
indicates the correlation coefficient between two RV
z, and Zk"
V, or Vex) indicate a domain centred at point X
v, or v(x) indicate a domain (generally of smaller dimensions
than V) centred at point x. v could also indicate the
second coordinate of the reference axes system (O,
Ov, Ow).
F(u) indicates the cumulative distribution function F(u) =
Prob{Z < u}, and, more often,
597
WWW.minebook.blogfa.com
598 INDEX D. NOTATIONS
WWW.minebook.blogfa.com
INDEX D. NOTATIONS 599
WWW.minebook.blogfa.com
600 INDEX D. NOTATIONS
b~=.!rfb)dx
v J,
indicates the mean value of the function fl(x) over the
support or domain v.
Poly abbreviation for the polygon of influence weighting
method.
10,102 abbreviation for the inverse (squared) distance
weighting method.
MS abbreviation for the interpolation method by fitting
mean square polynomials.
K(h) indicates a geometric or transitive covariogram, or
simply a covariance function.
{Z/Z* > z} indicates the set of (true) values Z conditioned by the
fact that their estimators Z* are superior to a given
value z (e.g., cut-off grade).
E{Z/Z*=z} indicates the conditional expectation of Z given
2*=z.
Z = ipz(U) indicates the Gaussian transform function by means
of. which a RV 2 with any given distribution cor-
responds to a standardized Gaussian R V U.
indicates the Hermite polynomials of order k,
indicates the standardized Hermite polynomial of
order k.
indicates a non-conditional simulated realization of a
RF Z(x).
indicates the kriged value at point x deduced from
non-conditionally simulated data values zs(x,,)
indicates a conditional simulation
indicates the unidimensional covariance imposed on
the lines defined by the turning band method of
simulation.
(Z(X), Z(y) indicates the scalar product of the two elements 2 (x)
= E{Z(x )2(y)} = (T"y and Z(y), this scalar product being taken as the
non -centred covariance.
liZ (x)112 = < Z(x),2(x indicates the squared norm of the vector (element)
Z(x).
indicates a simple linear kriging estimator.
indicates an ordinary linear kriging estimator.
indicates an unbiased linear kriging estimator of
order L, or, more shortly, a universal kriging esti-
mator.
EnZo indicates the conditional expectation of the RV Z(xo)
= E{Z(xo)/ Zll Z2' ... , Zn} given the n data values Z\> Z2, ... , Zn.
E(3Zo = E{2 (x 0)/ Z(3} indicates the conditional expectation of the R V Z (xo)
given the particular data value 2(3 = 2 (x,,) at point XI3'
indicates the disjunctive kriging estimator.
Neperian logarithm of Z (base e)
WWW.minebook.blogfa.com
" Geostatistics is the application of the formalism of random func tions to the reconnaissance
and estima tion of natural phenomena. ,.
With this definitio n, G. Mathe ron introd uced the term geostatlstlcs in 1962 to describe a
scientific approach to evaluation problems in geology and mining, from ore reserve estim -
ation to grade control. This approach provides, in application, a coherent and efficient set
of techniques enabfing the geologist, mining engineer or metallurgist to use available
informatio n and obtai n more accurate esti mates.
This book , the first complete reference work on the subject of mining geostatistics, is an
atte mpt to synthesize the practical experience gained by researchers fro m the Centre de
Morphologie Mathematiqu e of Fontainebleau, France, and by mining engineers and geol-
ogists all over th e w orld w ho have cont ributed their ideas to th e question. It is designed for .
students and engineers wh o wis h to apply qeostatistlc s to practical problems occurring in
the lifetime of a mine, and for this reason, it has been built around typi cal problems
progressing from the simplest to th e most complicated: structural analysis, guiding explor-
ation, estimation of in situ resources and recoverable reserves, numerical models of deposits,
simulation of mining and homogenization processes, are grade control at production. An
exhaustive FORTRAN package is given, and the techniques developed are illustrated by a
large number of case studies. As an aid to the reader, each chapter starts with a summary of
contents, and there is a guide to the nota tion used.
The book is written by practising and experienced mining engineers for the benefit of other
practitioners, and should become the standard reference text for the field during the
1980's. It is intended for the mining industry at the exploration and project level; for
university departments of mining , geology and mineralogy; and also for the postgraduate
student in a number of related fields such as geophysics, meteorology, geochemistry ,
bathymetry and geological engineering.
' 7 he book is a practical treatis e, written by practising mmmg engineers and in tended f or
other practising engin eers. .. the best summary of geostatistical th eory as it stands at the
present time and as one of the standard reference texts for the next few years."
Mining Ma gazine
"Ibis is the book for which so many of us have been waiting: a p ractical, authoritative and
scholarly w ork on geostatistics applied to mining . It is well written, well illustrated and
usable both as a tex tboo k for advanced students of mining geology and as a reference book
for professionals . . . altogether a g ood book, the bes t available on the subject. ..
C.J. Dixon in IM M Bulletin
Academic Press
Harcourt Brace Jovanovi ch . Publishers
LONDON SAN DIE GO NEW YORK
BERKELEY BOSlON SYDNE Y
m KYO
I I - . t BI'8C'e J........ id1 UnIilH
24/2 8 Ch1II Road . london NWI l OX. &,:Iand
~i<"~
12.loO s ...h ~. s..n Diq:o. Califomta 92101-4 311. LISA
Harcou rt BI'8C'e .J<M_id1 G"""p t......... nlia) Pl y Um;lO'd
Locked & 8 16, Mlrrickville . NSW 22lM. Au.. rahl
Ha rnHl rt Brace Jon nm'i<"h JOplin . Inc.
Ic~ib . "'ho
Cenlrol Building.. 22-1. k hibancho. Chi)<>dl ku .
Tokyo 102. Japan