See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/226292931

Calculation of Uncertainty in the Variogram

Article  in  Mathematical Geology · February 2002

DOI: 10.1023/A:1014412218427

CITATIONS READS

46 215

2 authors:

Julián M. Ortiz Clayton Deutsch

Queen's University University of Alberta
95 PUBLICATIONS   519 CITATIONS    337 PUBLICATIONS   9,220 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Image analysis applied to geometallurgical characterization View project

Multiple-Point Geostatistics: Stochastic Modeling with Training Images View project

All content following this page was uploaded by Julián M. Ortiz on 17 July 2014.

The user has requested enhancement of the downloaded file.

P1: GCR/GGT/GFQ P2: GVG
Mathematical Geology [mg] PP376-matg-367314 February 15, 2002 9:37 Style file version June 30, 1999

Mathematical Geology, Vol. 34, No. 2, February 2002 (°

C 2002)

Calculation of Uncertainty in the Variogram1

Julián Ortiz C.2 and Clayton V. Deutsch2

There are often limited data available in early stages of geostatistical modeling. This leads to consid-
erable uncertainty in statistical parameters including the variogram. This article presents an approach
to calculate the uncertainty in the variogram. A methodology to transfer this uncertainty through
geostatistical simulation and decision making is also presented.
The experimental variogram value 2γ̂ (h) for a separation lag vector h is a mean of squared dif-
ferences. The variance of a mean can be calculated with a model of the correlation between the pairs
of data used in the calculation. The “data” here are squared differences; therefore, we need a mea-
sure of a 4-point correlation. A theoretical multi-Gaussian approach is presented for this uncertainty
assessment together with a number of examples. The theoretical results are validated by numerical
simulation. The simulation approach permits generalization to non-Gaussian situations.
Multiple plausible variograms may be fit knowing the uncertainty at each variogram point, 2γ (h).
Multiple geostatistical realizations may then be constructed and subjected to process assessment to
measure the impact of this uncertainty.

KEY WORDS: multi-Gaussian, multipoint statistics, decision making.

INTRODUCTION

Variogram modeling is a critical step in any geostatistical study; however, a reliable

variogram is difficult to infer in presence of sparse data. This is particularly true in
the early exploration stages of an ore deposit or petroleum reservoir. A quantitative
model of the uncertainty in the variogram would allow an assessment of uncertainty
from geostatistical simulation.
Notwithstanding robust procedures to calculate variograms and other mea-
sures of spatial correlation (Cressie, 1991; Cressie and Hawkins, 1980; Genton,
1998), there is unavoidable uncertainty in the variogram. There are many refer-
ences on the calculation and use of the variogram (including Goovaerts; 1997, Olea,
1995; Omre, 1984); however, there is little on the calculation of the unavoidable
uncertainty in the variogram.

1Received 2 October 2000; accepted 22 February 2001.

2Department of Civil and Environmental Engineering, 220 Civil/Electrical Engineering Building,
University of Alberta, Edmonton, Alberta, Canada T6G 2G7; e-mail: [email protected] or e-mail:
[email protected]

169
0882-8121/02/0200-0169/1 °
C 2002 International Association for Mathematical Geology
P1: GCR/GGT/GFQ P2: GVG
Mathematical Geology [mg] PP376-matg-367314 February 15, 2002 9:37 Style file version June 30, 1999

170 Ortiz and Deutsch

We show how to calculate the pointwise uncertainty in the variogram. This

pointwise uncertainty must be translated to joint uncertainty, that is, into uncer-
tainty in the variogram model. Within the bounds of pointwise uncertainty, we
propose to establish different scenarios, ranging from small continuity to great
continuity. The importance of the variogram can be assessed by creating realiza-
tions and passing them through a transfer function.

POINTWISE VARIOGRAM UNCERTAINTY

The variogram is defined as

2 · γ (h) = Var{Z (ui ) − Z (ui + h)} (1)

where Z (·) is an element of a random field {Z (u) : u ∈ D}.

A method of moments estimator of the variogram 2γ (h) is the average of
squared differences between data separated exactly by that distance vector h
(in practice, we define angle and lag tolerances, so that n(h) is the number of
pairs approximately h apart):

1 X n(h)
2 · γ̂ (h) = · [Z (ui ) − Z (ui + h)]2 (2)
n(h) i=1

where n(h) is the number of data pairs approximately h apart.

Consider X i = [Z (ui ) − Z (ui + h)]2 , the squared difference between the val-
ues at locations (ui ) and (ui + h). The variogram is the mean of the X i s:

1 X n(h)
X̄ = 2 · γ̂ (h) = · Xi (3)
n(h) i=1

From classical statistics, we know that the uncertainty in the mean X̄ is defined as
2
Var{ X̄ } = E{( X̄ − E{ X̄ })2 } = E{ X̄ } − (E{ X̄ })2 (4)

Now, using expression (4) we can calculate the uncertainty in the variogram as-
suming that we have a “reference” variogram model fitted to the experimental
points. X̄ is replaced by 2 · γ̂ (h) and the variance of squared differences around
the model is calculated as follows:

σ2·2 γ̂ (h) = E{(2 · γ̂ (h))2 } − (E{2 · γ̂ (h)})2

Ã !2 
 1 X n(h) 
=E · [Z (ui ) − Z (ui + h)]2 − (E{2 · γ̂ (h)})2
 n(h) i=1 
P1: GCR/GGT/GFQ P2: GVG
Mathematical Geology [mg] PP376-matg-367314 February 15, 2002 9:37 Style file version June 30, 1999

Calculation of Uncertainty in the Variogram 171

(Ã
1 X
n(h) X
n(h)
=E · [Z (ui ) − Z (ui + h)]2
n(h)2 i=1 j=1
!)
× [Z (u j ) − Z (u j + h)]2 − (2 · γ̂ (h))2 (5)

1 X
n(h) X
n(h)
σ2·2 γ̂ (h) = · E{[Z (ui ) − Z (ui + h)]2 · [Z (u j ) − Z (u j + h)]2 }
n(h)2 i=1 j=1

− (2 · γ̂ (h))2 (6)

This can be simplified by using the definition of the covariance:

Ci j (h) = Cov{X i , X j } = E{(X i − E{X i }) · (X j − E{X j })}

= E{X i · X j } − E{X i } · E{X j }
2
= E{X i · X j } − X̄ (7)

Now, replacing X i and X j by the squared differences [z(ui ) − z(ui + h)]2 and
[z(u j ) − z(u j + h)]2 , respectively, and X̄ by the variogram 2 · γ (h),

Ci j (h) = E{[Z (ui ) − Z (ui + h)]2 · [Z (u j ) − Z (u j + h)]2 } − (2 · γ̂ (h))2 (8)

A simple formula for the variance of a particular variogram value is obtained

replacing the covariance (8) in expression (6):

1 X
n(h) X
n(h)
σ2·2 γ̂ (h) = · Ci j (h) (9)
n(h)2 i=1 j=1

where Ci j (h) is calculated as in Eq. (8). To avoid confusion, note that Ci j (h) is the
covariance between pair i[Z (ui ) − Z (ui + h)]2 and j[Z (u j ) − Z (u j + h)]2
(Fig. 1).
Expression (9) tells us that the uncertainty in the variogram at a distance h
is the average covariance between “pairs of pairs” used to calculate the variogram
for that particular lag.
The covariance between “pairs of pairs” can be calculated theoretically under
a multi-Gaussian assumption. The following section presents this approach. The
next sections present the local and global simulation methods to check the results
given by the theoretical approach. The global simulation method is more general
in the sense that it gives the whole distribution of uncertainty in the variogram
P1: GCR/GGT/GFQ P2: GVG
Mathematical Geology [mg] PP376-matg-367314 February 15, 2002 9:37 Style file version June 30, 1999

172 Ortiz and Deutsch

Figure 1. Calculation of fourth order covariances Ci j (h). For a

given lag vector h, the fourth order covariance corresponds to the
covariance between the squared differences of pairs i and j.

values for each lag. Although the shape of the pointwise uncertainty distribu-
tion is unknown and we know that the variogram values must be nonnegative, a
Gaussian shape was assumed to present the confidence intervals calculated using
the variance in the theoretical approach and the local simulation method. Theory
says that if all the squared random variables are independent (which is clearly
not the case) the distribution of uncertainty in a variogram point should be χ 2
(chi square). The global simulation method shows in few cases asymmetric distri-
butions; however, a Gaussian distribution is a good approximation in most of the
cases.
The following steps are required for all three methodologies
1. Transform data to normal space: Any data distribution can be easily trans-
formed to a Gaussian univariate distribution. In the following examples the
program nscore in GSLIB (Deutsch and Journel, 1998) was used to per-
form the transformation. This transformation is commonly done to allow
Gaussian simulation.
2. Check multigaussianity: To fulfil the multi-Gaussian condition, one should
assure that not only the univariate distribution is Gaussian, but also the
bivariate and all multivariate distributions. In practice, some tests can be
done to the transformed distribution in order to accept bigaussianity; how-
ever, they are not often applied, especially in presence of sparse data.
P1: GCR/GGT/GFQ P2: GVG
Mathematical Geology [mg] PP376-matg-367314 February 15, 2002 9:37 Style file version June 30, 1999

Calculation of Uncertainty in the Variogram 173

3. Calculate the experimental variogram: The location of the sampled points

and the values of the variable under study at these locations are used to
calculate the experimental variogram, 2 · γ̂ (h).
4. Fit a variogram model: The fitted variogram model is critical for subse-
quent stages of uncertainty evaluation. The requirement for a variogram
model to assess uncertainty in the variogram is of some concern. Never-
theless, a model assumption is required to proceed.
The difference between the theoretical approach and the numerical methods lies
in how the variance for each lag is calculated.

THEORETICAL APPROACH

Assuming that the regionalized variable is multi-Gaussian, the variogram un-

certainty can be calculated from theory. Expanding expression (8), the covariance
can be written as a sum of fourth order moments:

Ci j (h) = E{[Z (ui ) − Z (ui + h)]2 · [Z (u j ) − Z (u j + h)]2 } − (2 · γ̂ (h))2

= E{Z (ui )2 · Z (u j )2 − 2 · Z (ui )2 · Z (u j ) · Z (u j + h)
+ Z (ui )2 · Z (u j + h)2 − 2 · Z (ui ) · Z (ui + h) · Z (u j )2
+ 4 · Z (ui ) · Z (ui + h) · Z (u j ) · Z (u j + h)
− 2 · Z (ui ) · Z (ui + h) · Z (u j + h)2 + Z (ui + h)2 · Z (u j )2
− 2 · Z (ui + h)2 · Z (u j ) · Z (u j + h) + Z (ui + h)2 · Z (u j + h)2 }
− (2 · γ̂ (h))2 (10)

This covariance is called a quadratic covariance (Matheron, 1965), and it can be

calculated if Z (ui ), Z (ui + h), Z (u j ), and Z (u j + h) have a multivariate Gaussian
distribution. In such case, any fourth order moment can be calculated using the
pairwise covariance values as follows:

E{Z 1 · Z 2 · Z 3 · Z 4 } = C12 · C34 + C13 · C24 + C14 · C23 (11)

Notice that those pairwise covariances are different than the Ci j (u) presented
earlier, which are fourth order statistics, since they correspond to the covariance
between pairs of squared differences (i.e. “pairs of pairs”). Then, the variogram
variance is calculated as a sum of fourth order moments minus two times the
variogram squared.
A simple program can perform these calculations. For each lag, the location
of pairs considered in the experimental variogram calculation is used to determine
P1: GCR/GGT/GFQ P2: GVG
Mathematical Geology [mg] PP376-matg-367314 February 15, 2002 9:37 Style file version June 30, 1999

174 Ortiz and Deutsch

the fourth order moment as follows:

E{z(ui ) · z(ui + h) · z(u j ) · z(u j + h)}
= C(z(ui ), z(ui + h)) · C(z(u j ), z(u j + h))
+ C(z(ui ), z(u j )) · C(z(ui + h), z(u j + h))
+ C(z(ui ), z(u j + h)) · C(z(ui + h), z(u j )) (12)
A valid, positive definite, covariance model is required to perform the calculation
presented above. That is the reason to require a first guess of the variogram model.

SIMULATION ALTERNATIVE

Local Simulation Method

The idea is to simulate each set of four-point locations in turn and evaluate the
fourth order moments in expression (10) by simple averages. Again, the assumption
of multigaussianity simplifies the simulation. A matrix or LU simulation approach
is very fast and efficient since only four points are considered at a time and there are
no conditioning data. All fourth order moments in expression (10) are estimated
as averages of products using the simulated values, and the variogram variance is
calculated with formula (9).

Global Simulation Method

The basic idea is to generate nonconditional realizations of the domain using

the variogram model, and then calculate the variogram using only the values at
the sampled locations. The variance between the variogram values at each lag
calculated using these realizations should converge to the same value obtained
through any of the other approaches; however, the advantage of this approach
is that we can estimate the entire uncertainty distribution of all variogram lags
simultaneously, without assuming its shape.
This approach was implemented using the GSLIB program sgsim, that is,
unconditional realizations are generated. The sequential path in the program could
be modified to only simulate the locations of the original data. Uncertainty in the
variogram is directly evaluated by the variability between multiple realizations.
This global simulation method can be viewed as a “spatial bootstrap” or
resampling from geostatistical realizations (Journel, 1994).

VALIDATION OF THEORETICAL APPROACH BY SIMULATION

The theoretical approach has the following advantages over the two
simulation-based methods (1) implementation is easier since the fourth order
P1: GCR/GGT/GFQ P2: GVG
Mathematical Geology [mg] PP376-matg-367314 February 15, 2002 9:37 Style file version June 30, 1999

Calculation of Uncertainty in the Variogram 175

moments are calculated analytically and directly, (2) computer speed is much
improved since there is no need for random number generation or multiple real-
izations, and (3) the simulation methods are approximate, although they converge
to the correct result provided the implementation is correct.
The global simulation method has the advantage that the entire distribution
of uncertainty is simulated.

EXAMPLE 1: CLUSTER.DAT

Consider the database cluster.dat available in GSLIB (Deutsch and

Journel, 1998). The sample locations are on a pseudoregular grid, with clusters in
the high value zones (Fig. 2). After normal score transformation, the north–south
variogram is calculated for five lags, using a lag separation distance of 4.0 and a
lag tolerance of 2.0.
An isotropic spherical variogram model with range 15 m and 90% of variance
contribution is fitted to the experimental variogram. The nugget effect is 0.1 (10%
of variance contribution):
µ ¶
h
γ (h) = 0.1 + 0.9 · Sph (13)
15

Figure 2. Location map of samples taken from Cluster database.

P1: GCR/GGT/GFQ P2: GVG
Mathematical Geology [mg] PP376-matg-367314 February 15, 2002 9:37 Style file version June 30, 1999

176 Ortiz and Deutsch

Table 1. Pointwise Variogram Uncertainty Calculated Using the Three Methods Presented

Variance Variance local Variance global

Lag Experimental Fitted theoretical simulation simulation
Lag distance variogram variogram approach method method

2 1.395 0.262 0.225 0.004 0.004 0.004

3 4.361 0.431 0.481 0.021 0.019 0.014
4 7.906 0.716 0.746 0.046 0.042 0.038
5 11.876 1.191 0.946 0.080 0.068 0.068
6 15.796 1.198 1.000 0.096 0.083 0.130

The variogram uncertainty is assessed theoretically, using local simulation, and

through the global simulation method. The variance has been calculated for each
lag using the three methodologies presented above. In the local simulation approach
(using LU simulation), 100 realizations were performed. The results are presented
in Table 1.
Results show that with a reasonable number of LU simulations, the local
simulation method gives a variance very close to the theoretical result. Assuming
normality in the uncertainty distribution, the confidence intervals can be calculated.
The variogram, its model and the central confidence intervals at 95, 75, 50, and
25% for each lag are shown in Figure 3.

Figure 3. The experimental variogram, along with the variogram model fitted and the
central confidence intervals at 95, 75, 50, and 25% for each lag (Cluster database).
P1: GCR/GGT/GFQ P2: GVG
Mathematical Geology [mg] PP376-matg-367314 February 15, 2002 9:37 Style file version June 30, 1999

Calculation of Uncertainty in the Variogram 177

EXAMPLE 2: RED.DAT

This database contains samples of a vertical north–south tabular deposit,

where thickness and gold, silver, copper, and zinc concentrations were measured.
The variogram uncertainty is calculated for thickness and gold content using the
theoretical approach and both numerical methods. The sample locations are pre-
sented in Figure 4. The normal score transformation is performed for each variable.
The following isotropic variogram model is fitted to the omnidirectional experi-
mental variogram of thickness:
µ ¶
h
γ (h) = 0.15 + 0.85 · Exp (14)
250

For gold content, the variogram model is:

µ ¶
h
γ (h) = 0.45 + 0.55 · Sph (15)
250

Figure 4. Location map of samples and gold content taken from the
database red.dat.
P1: GCR/GGT/GFQ P2: GVG
Mathematical Geology [mg] PP376-matg-367314 February 15, 2002 9:37 Style file version June 30, 1999

178 Ortiz and Deutsch

Figure 5. The experimental variogram, along with the variogram model fitted and the central confi-
dence intervals at 95, 75, 50, and 25% for each lag (red.dat database). Left: Variogram for thickness;
Right: Variogram for gold content.

The calculation of confidence intervals was performed for each variable, and the
results are shown in Figure 5.
The global simulation method was used to obtain the entire uncertainty dis-
tribution for each lag. 100 nonconditional realizations of a Gaussian random vari-
able were generated using sgsim. The simulated values at the sampled locations
(obtained from the database red.dat) were extracted for each realization. The
experimental variogram was calculated using the simulated values at the sampled
locations and the same parameters that were used to find the experimental points
shown in Figure 5.
The experimental variograms calculated for each realization using the entire
simulated field (showing ergodic fluctuations) and those calculated using only
the simulated data at the sample locations (now considering the effect of ergodic
fluctuations and “sampling fluctuations”) are shown in Figure 6 for thickness and
gold content.
Table 2 shows the variogram variance for each variable and lag, calculated
using the theoretical approach, the local simulation method, and the global simu-
lation method. Hundred realizations were generated for the numerical methods.
The results obtained from the theoretical approach and the local simulation
method are similar; however, the global simulation method gives lower variance
for all the lags. The main difficulty of this approach is to ensure correct use of the
variogram for all distances when a limited number of nearby samples is used (Tran,
1994). The variogram calculated for each realization (using all the simulated nodes)
was presented in Figure 6 (Left). The variability in the variograms calculated using
all the nodes in the grid is lower than the expected variability.
Histograms showing the entire uncertainty distribution for the corresponding
lags are presented in Figure 7. All the histograms generated through the global
simulation method are slightly asymmetric with a tail to the right. This asymmetry
was expected since the variogram is nonnegative.
P1: GCR/GGT/GFQ P2: GVG
Mathematical Geology [mg] PP376-matg-367314 February 15, 2002 9:37 Style file version June 30, 1999

Calculation of Uncertainty in the Variogram 179

Table 2. Theoretical Approach to Calculate the Variogram Confidence Intervals

Variance Variance local Variance global

Lag Experimental Fitted theoretical simulation simulation
Lag distance variogram variogram approach method method

Variable: Thickness
2 17.497 0.332 0.311 0.013 0.012 0.004
3 51.119 0.687 0.540 0.008 0.001 0.007
4 99.311 0.669 0.742 0.044 0.041 0.024
5 148.627 0.871 0.857 0.092 0.089 0.052
6 197.746 0.957 0.921 0.152 0.150 0.085
7 250.436 1.178 0.958 0.176 0.177 0.112
8 297.843 0.969 0.976 0.264 0.258 0.160
9 345.356 0.992 0.986 0.289 0.270 0.193
Variable: Gold content
2 17.497 0.493 0.554 0.044 0.041 0.014
3 54.099 0.706 0.712 0.015 0.001 0.008
4 99.435 0.715 0.833 0.030 0.005 0.015
5 149.221 0.865 0.908 0.053 0.043 0.028
6 198.912 1.065 0.949 0.078 0.075 0.056
7 249.254 1.216 0.972 0.096 0.092 0.066
8 297.879 0.961 0.985 0.134 0.140 0.079
9 345.618 1.088 0.991 0.160 0.161 0.110

TRANSFERRING POINTWISE UNCERTAINTY

INTO THE JOINT MODEL

Several alternative variogram models could be fitted within the confidence

limits generated above. In order to achieve more realistic predictions, we can as-
sume different scenarios within those confidence limits. It is important to note
that variogram models fitted using the 97.5 and the 2.5 quantile variogram values
for all lags (Fig. 8) do not fairly represent extreme cases in the joint uncertainty.
The correlation between the lags and the “continuity” of alternative variogram
models should be accounted for when fitting models to represent extreme “joint”
cases.
Our proposal is to evaluate the consequences of using our first guess (the one
used to calculate the pointwise uncertainty), plus two extreme scenarios showing
high and low continuity, within the pointwise confidence limits (Fig. 9). Simulation
can be done using those three scenarios to determine the sensitivity of the results
to variogram uncertainty. Notice that we do not just have to modify the parameters
(range and sill contribution) of the variogram model, but the type of structure to
account for high and low continuity scenarios.
Uncertainty in the variogram sill can be addressed by fitting models with
different sill. This uncertainty can be due to uncertainty in the reference statistics.
P1: GCR/GGT/GFQ P2: GVG
Mathematical Geology [mg] PP376-matg-367314 February 15, 2002 9:37 Style file version June 30, 1999

180 Ortiz and Deutsch

Figure 6. The experimental variogram values for each lag calculated using (Left) all the simulated
data and (Right) only the simulated values at sampling locations (red.dat database). Top: Thickness;
Bottom: Gold.

COMMENTS

A variogram model is required in all approaches. Ideally, one could determine

the uncertainty using the experimental points before fitting a model. The assess-
ment of uncertainty, however, requires a positive definite covariance model (i.e. a
nonnegative variogram model); therefore a variogram must be fitted before evalu-
ating the uncertainty. This seems circular, however, it is the only way to solve the
problem: the authorized model is assumed as the expected value of the variogram
at each lag and then the variance is calculated.
The variogram uncertainty can be transferred to subsequent stages of a geo-
statistical study. The theoretical approach and the local simulation method gener-
ate the same results. The global simulation method requires more computer time
and should give the same result, since the idea is basically the same as the local
method; however, it is difficult to honor the variogram precisely for large distances
and consequently, the variance may be lower. The advantage of the global simula-
tion method is that it estimates the shape of the joint uncertainty in the variogram
for all lags.
 P1: GCR/GGT/GFQ
Mathematical Geology [mg]
P2: GVG
PP376-matg-367314

Calculation of Uncertainty in the Variogram

February 15, 2002
9:37

Figure 7. An example of the uncertainty distribution of the pointwise variogram values: Histograms of variogram
values for lags 6, 7, 8, and 9 for Gold (red.dat database).
181
Style file version June 30, 1999
P1: GCR/GGT/GFQ P2: GVG
Mathematical Geology [mg] PP376-matg-367314 February 15, 2002 9:37 Style file version June 30, 1999

182 Ortiz and Deutsch

Figure 8. An example of an incorrect interpretation of joint uncertainty, given the point-

wise uncertainty. Scenarios 1 and 2 do not represent quantiles 97.5 and 2.5 in the joint
model.

Figure 9. An example of a correct interpretation of joint uncertainty: Scenarios 1 and 2

represent low and high continuity (extremes of the joint model).
 P1: GCR/GGT/GFQ P2: GVG
Mathematical Geology [mg] PP376-matg-367314 February 15, 2002 9:37 Style file version June 30, 1999

Calculation of Uncertainty in the Variogram 183

Confidence intervals for each experimental variogram value can be deter-

mined from the variance assuming normality. This is approximate since the his-
togram of variogram values obtained for each lag must be nonnegative. All methods
require multigaussianity, which could be relaxed with non-Gaussian simulation
methods. This has not been explored in this article.
The difference between the pointwise uncertainty and joint uncertainty must
be addressed: the procedures presented in this article allow calculation of the
pointwise uncertainty. Within this uncertainty, several variogram models (joint
models) can be fitted. The confidence intervals for the joint model will be different
since we are interested in finding the uncertainty in the continuity of the variable.
Several joint models with different degrees of continuity should be used in the
subsequent kriging and simulation stages of a geostatistical study to account for
the uncertainty in the variogram model.

REFERENCES

Cressie, N., and Hawkins, D., 1980, Robust estimation of the variogram: Math. Geology, v. 12, no. 2,
p. 115–126.
Cressie, N., 1991, Statistics for spatial data: John Wiley & Sons, New York, 900 p.
Deutsch, C., and Journel, A., 1998, GSLIB: Geostatistical software library and user’s guide, 2nd ed.:
Oxford University Press, New York, 369 p.
Genton, M. G., 1998, Highly robust variogram estimation: Math. Geology, v. 30, no. 2, p. 213–221.
Goovaerts, P., 1997, Geostatistics for natural resources evaluation: Oxford University Press, New York,
483 p.
Journel, A. G., 1994, Resampling from stochastic simulations: Environ. Ecol. Stat., v. 1, p. 63–84.
Matheron, G., 1965, Les variables régionalisées et leur estimation: Masson et Cie. Editeurs, Paris.
Olea, R. A., 1995, Fundamentals of semivariogram estimation, modeling, and usage, in Yarus, J. M.,
and Chambers, R. L., eds., Stochastic modeling and geostatistics: Principles, methods, and case
studies. American Association of Petroleum Geologists, Tulsa, OK, p. 27–36.
Omre, H., 1984, The variogram and its estimation, in Verly, G., and others eds., Geostatistics for natural
resources characterization: Reidel, Dordrecht, Holland, Vol. 1, p. 107–125.
Tran, T. T., 1994, Improving variogram reproduction on dense simulation grids: Comput. Geosci., v. 20,
no. 7, p. 1161–1168.

View publication stats