Vietnam National University - Ho Chi Minh City

University of Technology
Faculty of Geology & Petroleum Engineering
Department of Drilling - Production Engineering

Course
Geostatistics in Petroleum Engineering
Trần Nguyễn Thiện Tâm
[email protected]
 Chapter 3

Chapter 3 Spatial Relationships -

Estimation and Modeling

11/19/2024 Geostatistics 2
 References
Mohan Kelkar, Godofredo Perez. Applied Geostatistics for Reservoir Characterization.
Society of Petroleum Engineers, Texas, 2002.

11/19/2024 Geostatistics 3
 Contents
❑ Random-Function Model
❑ Spatial Relationship
❑ Estimation of Variogram
❑ Modeling Variograms
❑ Cross Variograms

11/19/2024 Geostatistics 4
 Random-Function Model
Random experiment: an experiment that may result in several possible outcomes, none
of which can be predicted with certainty.
We can extend this concept to estimating values at unsampled locations. The goal of
geostatistics is to estimate values at locations where there is no information.
We use available sample data sets to develop certain models and use those models to
predict the values at unsampled locations.
In describing both the sampled data and variable values at unsampled locations in terms
of random variables, we are using what is called a random-function model.

11/19/2024 Geostatistics 5
 Random-Function Model
If we can develop a completely deterministic model based on the evolution of the
reservoir, we will be able to predict every reservoir property at every location with
certainty. However, we do not have this knowledge; therefore, our approach has to be
empirical and associated with uncertainty.
To reflect this uncertainty, we treat our estimates as random variables (denoted by
uppercase letters). For example, if an unsampled location is u0, the estimated value is
denoted as X(u0), where X is the variable.

11/19/2024 Geostatistics 6
 Random-Function Model
▪ Requirement of Stationarity

11/19/2024 Geostatistics 7
 Requirement of Stationarity
In addition to assuming that all locations are described by random variables, we also
have to consider the restrictions associated with using available sample data to
predict values at unsampled locations.
This assumption, in qualitative terms, requires that the proposed model based on our
sampled data can adequately describe the behavior of the population.

11/19/2024 Geostatistics 8
 Requirement of Stationarity
Mathematically, the first order of stationarity can be written as
𝑓 𝑋 𝑢 =𝑓 𝑋 𝑢+𝐿
where
f[ ] = any function of a random variable and
𝑢 and 𝑢 + 𝐿 define the two locations of the random variable.
The most commonly used function is the expected value. Therefore, we can write
𝐸 𝑋 𝑢 =𝐸 𝑋 𝑢+𝐿
That is, the expected value of a random variable at 𝑢 is the same as the expected value of
a random variable 𝐿 lag distance away.
The value of 𝐿 can vary from zero to the maximum distance between variables within
the region of interest.

11/19/2024 Geostatistics 9
 Requirement of Stationarity
The second order of stationarity can be mathematically defined as
𝑓 𝑋 𝑢1 , 𝑋 𝑢1 + 𝐿 = 𝑓 𝑋 𝑢2 , 𝑋 𝑢2 + 𝐿
This relationship indicates that any function of two random variables located 𝐿 distance
apart is independent of the location and is a function of only the distance and the
direction between the two locations. The arrows over the u and L indicate that locations
can be treated in terms of vectors rather than distances.

11/19/2024 Geostatistics 10
 Requirement of Stationarity
In practice, we can use covariance as one of the functions that relates two variables
located a certain distance and direction apart.
𝐶 𝑋 𝑢1 , 𝑋 𝑢1 + 𝐿 = 𝐶 𝑋 𝑢2 , 𝑋 𝑢2 + 𝐿
That is, the covariance within the region of stationary is a function of only the vector 𝑳,
not of the variable itself. This is an important assumption. It means that, as long as we
know the distance and direction between any two points, we can estimate the
covariance between the random variables at these two points; we do not need the
actual random variables at those locations. Eq. can therefore be written as
𝐶 𝑋 𝑢 ,𝑋 𝑢 + 𝐿 = 𝐶 𝐿
With the definition of covariance,
𝐶 𝑋 𝑢 ,𝑋 𝑢 + 𝐿 = 𝐸 𝑋 𝑢 ,𝑋 𝑢 + 𝐿 − 𝐸 𝑋 𝑢 .𝐸 𝑋 𝑢 + 𝐿
However, with the first-order stationary assumption, we can write
2
𝐶 𝑋 𝑢 ,𝑋 𝑢 + 𝐿 = 𝐸 𝑋 𝑢 ,𝑋 𝑢 + 𝐿 − 𝐸 𝑋 𝑢

11/19/2024 Geostatistics 11
 Spatial Relationship
▪ Covariance
▪ Correlation Coefficient
▪ Variogram

11/19/2024 Geostatistics 12
 Covariance
2
𝑛 𝐿 𝑛 𝐿
1 1
𝑐 𝐿 = ෍ 𝑥 𝑢 𝑖 . 𝑥 𝑢𝑖 + 𝐿 − ෍ 𝑥 𝑢𝑖
𝑛 𝐿 𝑖=1 𝑛
𝑖=1
where
𝑛 𝐿 = number of pairs at vector distance 𝐿;
𝑥 𝑢𝑖 , and 𝑥 𝑢𝑖 + 𝐿 = values ofthe variable at locations 𝑢𝑖 , and 𝑢𝑖 + 𝐿, respectively; and
n = total number of sample points.
The second term on the right side of Eg. represents the arithmetic mean of all the data
points. 𝑐 𝐿 = the estimated value based on the sampled data. We can rewrite Eq. as
𝑛 𝐿
1
𝑐 𝐿 = ෍ 𝑥 𝑢𝑖 . 𝑥 𝑢𝑖 + 𝐿 − 𝑥ҧ 2
𝑛 𝐿 𝑖=1
Where 𝑥ҧ the arithmetic mean of the sample data.

11/19/2024 Geostatistics 13
 Example 3.1
Table 3.1 gives the porosity data as a function of depth in a vertical well. For
convenience, we have used only seven porosity values; in practice, however, we will have
a lot more values. Estimate the covariance as a function of lag distance.

11/19/2024 Geostatistics 14
 Example 3.1 - Solution
We first need to calculate the arithmetic mean.
𝑛
1 1
𝑥ҧ = ෍ 𝑥 𝑢𝑖 = 8.25 + 9.0 + 6.25 + 5.00 + 5.3 + 4.75 + 5.0 = 6.22
𝑛 7
𝑖=1
Because the data are collected in only one direction (vertically), we do not need to worry
about the direction. For a lag distance of 1.0 ft, we have six pairs.
𝑐 1
1
= ሺ8.25 × 9.00 + 9.00 × 6.25 + 6.25 × 5.00 + 5.00 × 5.30 + 5.30 × 4.75 + 4.75
6
× 5.00ሻ − 6.22 2 = 0.8408 ϕ(u) ϕ(u + 1)
8.25 9.00
9.00 6.25
6.25 5.00
5.00 5.30
5.30 4.75
4.75 5.00

11/19/2024 Geostatistics 15
 Example 3.1 - Solution
Similarly, at a lag distance of 2.0 ft, we have five pairs. ϕ(u) ϕ(u + 2)
𝑐 2 8.25 6.25
1 9.00 5.00
= ሺ8.25 × 6.25 + 9.00 × 5.00 + 6.25 × 5.30 + 5.00 × 4.75
5 6.25 5.30
+ 5.30 × 5.00ሻ − 6.22 2 = −2.7009
5.00 4.75
Using similar calculations, we can calculate the value of 5.30 5.00
covariance at a lag distance of 3 ft as
𝑐 3
1 ϕ(u) ϕ(u + 3)
= 8.25 × 5.00 + 9.00 × 5.30 + 6.25 × 4.75 + 5.00 × 5.00
4 8.25 5.00
− 6.22 2 = −2.7790 9.00 5.30
6.25 4.75
5.00 5.00

11/19/2024 Geostatistics 16
 Example 3.1 - Solution
We calculate only to this point because the number of pairs becomes smaller at
additional lag distances. As a special case, the definition at 𝐿 = 0 is
𝑛
1
𝑐 0 = ෍ 𝑥 𝑢𝑖 . 𝑥 𝑢𝑖 − 𝑥ҧ 2
𝑛
𝑖=1
This is the definition of variance. For the present data,
1
𝑐 0 = 8.252 + 9.002 + 6.252 + 5.002 + 5.302 + 4.752 + 5.002 − 6.22 2
= 2.565
7

11/19/2024 Geostatistics 17
 Correlation Coefficient
𝐶ሺ𝐿ሻ
𝜌 𝐿 =
𝜎𝑢 𝜎𝑢+𝐿
where
𝜌 𝐿 = the correlation coefficient at lag distance 𝐿;
𝐶 𝐿 = covariance; and
𝜎𝑢 and 𝜎𝑢+𝐿 = standard deviations for the data allocations 𝑢 and 𝑢 + 𝐿, respectively.
However, if we make the assumption of the second order of stationarity, we can state
that
𝑉 𝑋 𝑢 =𝑉 𝑋 𝑢+𝐿 =𝐶 0
Therefore, we can also state that
𝜎𝑢 = 𝜎𝑢+𝐿 = 𝐶ሺ0ሻ

11/19/2024 Geostatistics 18
 Correlation Coefficient
𝐶ሺ𝐿ሻ
𝜌 𝐿 =
𝐶ሺ0ሻ
Using the sample data, we can write the estimated value of the correlation coefficient as
𝑐ሺ𝐿ሻ
𝑟 𝐿 =
𝑐ሺ0ሻ
where c(0) = sample variance.

11/19/2024 Geostatistics 19
 Example 3.2
Table 3.1 gives the porosity data as a function of depth in a vertical well. Calculate the
correlation coefficient as a function of lag distance.

11/19/2024 Geostatistics 20
 Example 3.2 - Solution
The values of covariance were computed earlier. Therefore, the correlation coefficient at
𝐿 = 1 ft is
𝑐ሺ1ሻ 0.8408
𝑟 1 = = = 0.327
𝑐ሺ0ሻ 2.57
Similarly,
𝑐ሺ2ሻ −2.7009
𝑟 2 = = = −1.05
𝑐ሺ0ሻ 2.57

𝑐ሺ3ሻ −2.7790
𝑟 3 = = = −1.08
𝑐ሺ0ሻ 2.57
and
𝑐ሺ0ሻ
𝑟 0 = =1
𝑐ሺ0ሻ

11/19/2024 Geostatistics 21
 Variogram
The variogram is the most commonly used geostatistical technique for describing the
spatial relationship. Mathematically, it is defined as
1
𝛾 𝐿 = 𝑉 𝑋 𝑢 −𝑋 𝑢+𝐿
2
It is half of the variance of the difference between the two values located 𝑳 distance
apart.
As the definition indicates, the value of the variogram is zero at 𝐿 = 0 because
1
𝛾 0 = 𝑉 𝑋 𝑢 −𝑋 𝑢 =0
2
Because of our assumption for the geoscience data, the difference between the two
values increases as the distance increases. That is, variance increases as lag distance
increases, which is exactly opposite of covariance. Unlike covariance, the variogram
starts with a zero value and increases as the lag distance between the two values
increases.

11/19/2024 Geostatistics 22
 Variogram
With the definition of variance,
1 2 2
𝛾 𝐿 = 𝐸 𝑋 𝑢 −𝑋 𝑢+𝐿 − 𝐸 𝑋 𝑢 −𝑋 𝑢+𝐿
2
However, the assumption of first-order stationarity requires that
𝐸 𝑋 𝑢 =𝐸 𝑋 𝑢+𝐿
Therefore, the second term on the right side of Eq. is equal to zero. Simplifying,
1 2
𝛾 𝐿 = 𝐸 𝑋 𝑢 −𝑋 𝑢+𝐿
2
With the definition of covariance, Eq. becomes
𝛾 𝐿 =𝐶 0 −𝐶 𝐿

11/19/2024 Geostatistics 23
 Variogram
As expected, the variogram is closely related to the covariance if the assumptions of the
first- and second-order stationarities hold. The variogram increases as the covariance
decreases. Both the variogram and covariance capture the spatial relationship. The
variogram increases as the lag distance between the values increases (the
relationship gets weaker); the covariance decreases as the lag distance between the
two values increases.

11/19/2024 Geostatistics 24
 Variogram
In practice, Eq. 3.18 can be written as
𝑛 𝐿
1 2
𝛾ො 𝐿 = ෍ 𝑥 𝑢𝑖 − 𝑥 𝑢𝑖 + 𝐿
2𝑛 𝐿 𝑖=1
where
𝑛 𝐿 = number of pairs at lag distance 𝐿;
𝑥 𝑢𝑖 and 𝑥 𝑢𝑖 + 𝐿 = data values for the ith pair located 𝐿 lag distance apart.
The accent (hat) over y indicates that it is an estimated value based on the sample
data.

11/19/2024 Geostatistics 25
 Example 3.3
Table 3.1 gives the porosity data as a function of depth in a vertical well. Estimate the
variogram as a function of lag distance.

11/19/2024 Geostatistics 26
 Example 3.3 - Solution
As in the case of covariance, first gather all the pairs at a given ϕ(u) ϕ(u + 1)
lag distance. For 𝐿 = 1 ft, there are six pairs. 8.25 9.00
𝛾ො 1 9.00 6.25
1 6.25 5.00
= ሾ 8.25 − 9.00 2 + 9.00 − 6.25 2 + 6.25 − 5.00 2
2×6 5.00 5.30
+ 5.00 − 5.30 2 + 5.30 − 4.75 2 + 4.75 − 5.00 2 ሿ = 0.845
5.30 4.75
4.75 5.00

11/19/2024 Geostatistics 27
 Example 3.3 - Solution
Similarly, for 𝐿 = 2 ft, we have five pairs ϕ(u) ϕ(u + 2)
𝛾ො 2 8.25 6.25
1 9.00 5.00
= ሾ 8.25 − 6.25 2 + 9.00 − 5.00 2 + 6.25 − 5.30 2
2×5 6.25 5.30
+ 5.00 − 4.75 2 + 5.30 − 5.00 2 ሿ = 2.11 5.00 4.75
5.30 5.00

For 𝐿 = 3 ft, we have four pairs; ϕ(u) ϕ(u + 3)

𝛾ො 3 8.25 5.00
1 9.00 5.30
= ሾ 8.25 − 5.00 2 + 9.00 − 5.30 2
+ 6.25 − 4.75 2
2×4 6.25 4.75
+ 5.00 − 5.00 2 ሿ = 3.31
5.00 5.00

11/19/2024 Geostatistics 28
 Estimation of Variogram
▪ Lack of Sufficient Pairs
▪ Instability
▪ Influence of Outliers
▪ Biased Sampling

11/19/2024 Geostatistics 29
 Lack of Sufficient Pairs
We calculate the “average” squared difference between two values located a certain
lag distance apart. For this average to be truly representative of a given lag distance,
sufficient data pairs are needed for that lag distance.
Obviously, the more pairs we have for a given lag distance, the more precise the
estimate of the variogram.

11/19/2024 Geostatistics 30
 Lack of Sufficient Pairs
• Number of Pairs
• Lag Tolerance

11/19/2024 Geostatistics 31
 Number of Pairs
A commonly used method to ensure sufficient pairs for a given lag distance is to restrict
the maximum distance at which the variogram is computed. For a given region of
stationarity, the possible number of pairs decreases as lag distance increases.
Numerical Examples 3.1 and 3.2, where the number of pairs decreased as lag distance
increased, illustrated this.

11/19/2024 Geostatistics 32
 Number of Pairs
Eventually, as the lag distance corresponds to the maximum possible distance
between any two sample points, only one data pair may be available at that lag
distance. Obviously, we may not be able to obtain a reliable estimate of the variogram
at that lag distance.

11/19/2024 Geostatistics 33
 Number of Pairs
To avoid this problem, a rule of thumb used in the geostatistical literature is to use half
the maximum possible distance within a region of interest as the maximum lag
distance at which the variogram is calculated.
For example, if 10,000 ft is the maximum distance between any two sample points
within the region of interest, the variogram estimation is restricted to a maximum lag
distance of 5,000 ft.

11/19/2024 Geostatistics 34
 Lag Tolerance
The second alternative to ensure sufficient pairs for a given lag distance is to define
the tolerance with respect to the distance as well as with respect to the direction. This is
especially useful for areal data sets. Typically, examination of areally distributed data
shows that the samples are not distributed at uniform intervals; therefore, a sufficient
number of pairs for a precise lag distance cannot be obtained.

11/19/2024 Geostatistics 35
 Lag Tolerance
For example, to estimate the variogram at a lag distance of 330 ft, we may have one
data pair at precisely 330 ft. However, we may have three additional pairs between
310 and 330 ft and five additional pairs between 330 and 350 ft. Obviously, one pair,
precisely at 330 ft, does not provide a reasonable estimate of the variogram.

11/19/2024 Geostatistics 36
 Lag Tolerance
However, we can capture additional pairs for a better estimate of the variogram by
defining a tolerance with respect to the lag distance. The variogram definition is
modified as
𝑛 𝐿
1 2
𝛾ො 𝐿 ± ∆𝐿 = ෍ 𝑥 𝑢𝑖 − 𝑥 𝑢𝑖 + 𝐿
2𝑛 𝐿 ± ∆𝐿 𝑖=1
where ∆𝐿 = tolerance with respect to lag distance.

11/19/2024 Geostatistics 37
 Lag Tolerance
To estimate the variogram at a lag distance of 𝐿,
we collect all the pairs within 𝐿 ± ∆𝐿 lag distance.
Fig. 3.3 illustrates tolerance with respect to
distance for an isotropic search, which is
concerned only with distance for all the angles.
When searching for sample points away from
𝒙 𝒖 at a given distance 𝑳, we can search for all
the data points within 𝑳 ± ∆𝑳 distance.
In Fig. 3.3, no sample point is precisely at a
distance L; however, by using a tolerance of AL,
we can obtain two pairs at that lag distance.

11/19/2024 Geostatistics 38
 Instability
The estimated variogram represents the arithmetic average of the squared differences
of variable pair values at a particular lag distance. Because it uses the square of the
difference, any large difference between a given pair is magnified. If pairs exhibit a
large difference, the squared difference may have a significant impact on the
arithmetically averaged variogram value. This effect may change the variogram value
disproportionately at a particular lag distance, resulting in instability of the estimated
variogram.

11/19/2024 Geostatistics 39
 Instability
This instability may prevent capturing the underlying variogram structure that may
be present and also causes fluctuations in the estimated variogram as lag distance
increases. The instability must be minimized to model the variogram.

11/19/2024 Geostatistics 40
 Instability
The two methods commonly used to minimize fluctuations are to increase the
possible number of pairs for a given lag distance or to remove certain pairs for a given
lag distance.

11/19/2024 Geostatistics 41
 Influence of Outliers
Outliers are hard to define. In a conventional sense, outlier data are data points that fall
outside the “norm.” For a normal distribution, a data point falling outside the mean plus
or minus three standard deviations can be considered an outlier.

11/19/2024 Geostatistics 42
 Influence of Outliers
The simplest way to deal with the outlier information that causes this instability is to
remove the data point from the estimation process. If sufficient physical reason exists
for removal, we can simply remove the data point or points and reestimate the
variogram.

11/19/2024 Geostatistics 43
 Influence of Outliers
A better way to deal with these variations in the sample data is to use some type of
nonlinear transformation to minimize the variation.

11/19/2024 Geostatistics 44
 Biased Sampling
When we analyze collected samples to describe the spatial relationship, we must
understand that the well density may be higher in regions where better prospects
exist and lower in regions where the potential is low. This type of biased sampling may
impact our ability to describe the spatial relationship. For example, if more wells are
drilled in high-porosity regions, the estimated variogram at small lag distances uses
information only in a high-porosity region.

11/19/2024 Geostatistics 45
 Modeling Variograms
▪ Modeling Requirements
▪ Models With Sill
▪ Anisotropic Models

11/19/2024 Geostatistics 46
 Modeling Requirements
In modeling the estimated variogram, we must consider two requirements.
o The first is the use of a minimum number of parameters and models to model the
variogram.
o The second requirement is the condition of positive definiteness.

11/19/2024 Geostatistics 47
 Modeling Requirements
Recall from Chap. 2 that, in the estimation process, we assume that the variable at the
unsampled location is linearly related to the surrounding samples. Mathematically, we
can write 𝑛

𝑥 ∗ 𝑢𝑜 = ෍ 𝜆 𝑖 𝑥 𝑢𝑖
𝑖=1
where
𝑥 ∗ 𝑢𝑜 = the estimated value at the unsampled location, 𝑢𝑜 ,
𝑥 𝑢𝑖 = the sample value at location 𝑢𝑖 , and
𝜆𝑖 = the weight assigned to sample 𝑥 𝑢𝑖 .

11/19/2024 Geostatistics 48
 Modeling Requirements
In other words, we assume that the value at the unsampled location is a weighted
linear combination of the nearby n samples. Because the estimated value is a random
variable with associated uncertainties, we can also estimate the variance associated
with it. Taking the variance on both sides of Eq. 𝑛3.40, we can write
𝑉 𝑥 ∗ 𝑢𝑜 = 𝑉 ෍ 𝜆 𝑖 𝑥 𝑢𝑖
𝑖=1
Expanding the right side of Eq.𝑛 3.41 gives 𝑛 𝑛

𝑉 ෍ 𝜆 𝑖 𝑥 𝑢𝑖 = ෍ ෍ 𝜆𝑖 𝜆𝑗 𝐶 𝑢𝑖 , 𝑢𝑗
𝑖=1 𝑗=1 𝑖=1
where 𝐶 𝑢𝑖 , 𝑢𝑗 = the covariance between the two sample points located at 𝑢𝑖 , and 𝑢𝑗 .

11/19/2024 Geostatistics 49
 Modeling Requirements
We know that variance must always be positive; i.e., the left and right sides of Eq. 3.42
must be positive. 𝑛 𝑛

෍ ෍ 𝜆𝑖 𝜆𝑗 𝐶 𝑢𝑖 , 𝑢𝑗 ≥ 0
𝑗=1 𝑖=1
The left side of Eq. 3.43 is zero only if all the weights, 𝜆𝑖 , are zero; otherwise, it should
always be positive. This is the condition of positive definiteness. The model should be
chosen so that Eq. 3.42 is always satisfied.

11/19/2024 Geostatistics 50
 Models With Sill
• Nugget-Effect Model
• Spherical Model
• Exponential Model
• Gaussian Model
• Combination Models

11/19/2024 Geostatistics 51
 Nugget-Effect Model
The simplest model is the nugget-effect model. In practice, it can be written as
0 if 𝐿 = 0
𝛾 𝐿 =ቊ
𝐶0 if 𝐿 > 0
where C0 = the sill value.
The variogram by definition has to be zero at a lag distance equal to zero. The nugget
effect assumes that the variogram abruptly increases to a value of C0 for any lag
distance greater than zero. Beyond that point, it remains constant.

11/19/2024 Geostatistics 52
 Nugget-Effect Model
Knowing the relationship between the variogram and the covariance,
𝛾 𝐿 =𝐶 0 −𝐶 𝐿
we can write the corresponding covariance equation as
𝐶0 if 𝐿 = 0
𝐶 𝐿 =ቊ
0 if 𝐿 > 0
Note that the sill value is equivalent to C(0).

11/19/2024 Geostatistics 53
 Nugget-Effect Model
The nugget effect indicates a total lack of information with respect to spatial
relationship. In practice, it may be hard to determine which reason is responsible for the
nugget effect. What is important is the fact that a high value of nugget indicates a lack
of information about the spatial relationship.

11/19/2024 Geostatistics 54
 Nugget-Effect Model
Fig. 3.32 shows the nugget-effect
model and other sill models. As
explained, the nugget effect is a
horizontal line with a constant sill
value of C0.

11/19/2024 Geostatistics 55
 Nugget-Effect Model
Fig. 3.33 shows an equivalent
covariance plot for the nugget-effect
model and the other models. Zero
value or a lack of correlation
indicates the nugget effect.

11/19/2024 Geostatistics 56
 Spherical Model
The spherical model is probably the most commonly used model to demonstrate
variograms with a sill. The equation for the spherical variogram can be written as
3
3 𝐿 1 𝐿
𝛾 𝐿 = 𝐶0 − if 𝐿 ≤ 𝑎
𝑀𝑆𝑎 𝐿 = 2 𝑎 2 𝑎
𝐶0 if 𝐿 ≥ 𝑎
where
𝑀𝑆𝑎 𝐿 = a spherical model with range a.
An equivalent covariance model can be written as
3
3 𝐿 1 𝐿
𝐶 1− + if 𝐿 ≤ 𝑎
𝐶 𝐿 = 0 2 𝑎 2 𝑎
0 if 𝐿 ≥ 𝑎

11/19/2024 Geostatistics 57
 Spherical Model
The spherical model is defined with a range of a and a sill value of C0. The variogram
reaches a sill value at a lag distance of a. Figs. 3.32 and 3.33, respectively, show the
spherical variogram and the associated covariance model. Note that, of all the models,
the spherical model has the highest slope at the origin. Therefore, if an estimated
variogram rises rapidly for a given range, the spherical model is a good option to try
to model the variogram.

11/19/2024 Geostatistics 58
 Exponential Model
The equation for the exponential model can be written as
−3𝐿
𝑀𝐸𝑎 𝐿 = 𝛾 𝐿 = 𝐶0 1 − exp
𝑎
if L ≥ 0,
where
a = effective range of the variogram and
𝑀𝐸𝑎 𝐿 = exponential model with a range a.
The corresponding covariance model can be written as
−3𝐿
𝐶 𝐿 = 𝐶0 exp
𝑎
if L ≥ 0

11/19/2024 Geostatistics 59
 Exponential Model
For the exponential model, the variogram reaches the sill value only asymptotically.
Therefore, for practical purposes, a range is defined as the lag distance where the
variogram reaches 95% of the sill value. Substituting L = a into Eq. 3.52 gives
𝛾 𝐿 = 𝐶0 1 − 𝑒 −3 = 0.95𝐶0 ≈ 𝐶0
and we obtain 𝛾 𝐿 equal to 95% of the sill value.

11/19/2024 Geostatistics 60
 Exponential Model
Figs. 3.32 and 3.33, respectively, show the variogram and corresponding covariance
models. As can be seen, for the same range, the slope at the origin for the exponential
model is smaller than that of the spherical model, indicating a much more gradual
change in the variogram value.

11/19/2024 Geostatistics 61
 Gaussian Model
The variogram equation for the Gaussian model can be written as
𝐿2
𝑀𝐺𝑎 𝐿 = 𝛾 𝐿 = 𝐶0 1 − exp −3 2 if 𝐿 ≥ 0
𝑎
where
𝑀𝐺𝑎 𝐿 = a Gaussian model with a range a.
The corresponding covariance model can be written as
𝐿2
𝐶 𝐿 = 𝐶0 exp −3 2 if 𝐿 ≥ 0
𝑎

11/19/2024 Geostatistics 62
 Gaussian Model
For the Gaussian model, the slope at the origin is zero, indicating an extremely smooth
variation in properties as a function of distance. The effective range for the Gaussian
model is a, which represents the lag distance at which the variogram reaches 95% of the
sill value.

11/19/2024 Geostatistics 63
 Gaussian Model
Figs. 3.32 and 3.33, respectively, show the Gaussian variogram model and the associated
covariance model. The figures show that the variogram changes very gradually at the
origin and exhibits an S-shaped curve before reaching a sill value.

11/19/2024 Geostatistics 64
 Combination Models
Using any of the four models described, we can create multiple linear combinations of
these models to model a given estimated variogram. Most commonly, the nugget-effect
model becomes an inherent part of the variogram model. For example, we can write a
variogram model as
𝛾 𝐿 = 𝐶0 + 𝐶1 𝑀𝑆𝑎 𝐿
where
C0 = the sill value of the nugget effect,
C1 = the sill value of a spherical model, and
a = the range of a spherical model.

11/19/2024 Geostatistics 65
 Combination Models
Fig. 3.34 shows this model schematically. The variogram starts at a value of C0,
representing a certain lack of information about the spatial relationship. The
variogram finally rises to a value of (C0 + C1), which is a total sill of the variogram. C1 is
the contribution of the spherical variogram to the total sill.

11/19/2024 Geostatistics 66
 Anisotropic Models
Anisotropic models are needed when the estimated variograms show different spatial
continuities in different directions.
For example, in a channel sand, the variograms of porosity may show better continuity
in the direction of the channel than across the channel. Anisotropic models need to be
used to capture these types of continuities.

11/19/2024 Geostatistics 67
 Anisotropic Models
• Geometric Anisotropy
• Zonal Anisotropy

11/19/2024 Geostatistics 68
 Geometric Anisotropy
Geometric anisotropy typically is observed
when the variograms in the directions of
maximum and minimum continuity show
a similar shape and sill but different
ranges (Fig. 3.45). In Fig. 3.45, the range in
the direction of maximum continuity, 𝑢, is
au. while the range in the direction of
minimum continuity is av. We assume that
the two directions are perpendicular to
each other.

11/19/2024 Geostatistics 69
 Geometric Anisotropy
To model the two variograms with the same sill, we have to use the same combination
of linear models in both directions except with different ranges.
For example, with a linear combination of nugget and spherical models, the model in
the 𝒖 direction is
𝛾𝑢 𝐿 = 𝐶0 + 𝐶1 𝑀𝑆𝑎𝑢 𝐿
and the model in the 𝒗 direction is
𝛾𝑣 𝐿 = 𝐶0 + 𝐶1 𝑀𝑆𝑎𝑣 𝐿
The models in both directions contain the same linear combination; however, the
range in each direction is different.

11/19/2024 Geostatistics 70
 Geometric Anisotropy
It is not necessary to restrict the model to a combination of only two models; we can
use as many as we want as long as the sills of the models in both directions are the
same and the types of models used are the same.
For example, we can model the variogram in the 𝒖 direction as
𝛾𝑢 𝐿 = 𝐶0 + 𝐶1 𝑀𝐸𝑎𝑢 𝐿 + 𝐶2 𝑀𝐺𝑎𝑢 𝐿
1 2

and the variogram in the 𝒗 direction as

𝛾𝑣 𝐿 = 𝐶0 + 𝐶1 𝑀𝐸𝑎𝑣 𝐿 + 𝐶2 𝑀𝐺𝑎𝑣 𝐿
1 2

Note that we have the same linear combination in both directions except that the
ranges differ in the two directions.

11/19/2024 Geostatistics 71
 Geometric Anisotropy
This type of modeling satisfies the requirement for a condition of positive definiteness.
Further, by defining the model in only two directions, we can estimate the variogram in
any other direction. Before we examine the method of estimating the variogram in any
direction, we must understand that any variogram model with a sill can be
normalized to convert to an equivalent model with a range equal to one.

11/19/2024 Geostatistics 72
 Geometric Anisotropy
For example, a spherical model can be written as
3
3 𝐿 1 𝐿
𝛾 𝐿 = 𝐶0 − if 𝐿 ≤ 𝑎
𝑀𝑆𝑎 𝐿 = 2 𝑎 2 𝑎
𝛾 𝐿 = 𝐶0 if 𝐿 ≥ 𝑎
This model has a range of a. If we define a variable
𝐿
𝐿𝐷 =
𝑎
we can write Eq. 3.50 as
3 1 3
𝐶0 𝐿𝐷 − 𝐿𝐷 if 𝐿 ≤ 𝑎
𝛾 𝐿𝐷 = ൞ 2 2
𝐶0 if 𝐿 ≥ 𝑎
where Eqs. 3.73a and 3.73b represent a variogram model with a unit range.

11/19/2024 Geostatistics 73
 Geometric Anisotropy
In Fig. 3.46, the directions of
maximum and minimum
continuities are shown as 𝑢 and 𝑣, Ԧ
respectively. These directions need
not coincide with the original axes, 𝑢′
and 𝑣Ԧ ′ , which are based on the
sampled information.

11/19/2024 Geostatistics 74
 Geometric Anisotropy
If the direction of maximum
continuity (𝑢) is 𝜃, then the direction
of minimum continuity (𝑣) Ԧ is (𝜃 +
90°). To calculate the variogram for
vector 𝐿 defined in an arbitrary
direction, we can first calculate the
components 𝐿𝑢′ and 𝐿𝑣 ′ based on the
original axes 𝑢′ and 𝑣Ԧ ′ through a
simple coordinate transformation. We
can write
𝐿𝑢 = 𝐿𝑢′ cos𝜃 + 𝐿𝑣 ′ sin𝜃
𝐿𝑣 = −𝐿𝑢′ sin𝜃 + 𝐿𝑣 ′ cos𝜃

11/19/2024 Geostatistics 75
 Geometric Anisotropy
We can then normalize the vector 𝑳
with the transformation
2 2
𝐿𝑢 𝐿𝑣
𝐿𝐷 = +
𝑎𝑢 𝑎𝑣

where au and av = ranges in the 𝑢 and

𝑣Ԧ directions, respectively. We can then
calculate the variogram at
dimensionless lag distance LD with the
appropriate equation.

11/19/2024 Geostatistics 76
 Geometric Anisotropy
This type of transformation satisfies
the limiting cases.
For example, when calculating the
variogram in the 𝒖 direction, 𝐿𝑣 = 0
and 𝐿𝐷 = 𝐿𝑢 /𝑎𝑢 . Substituting 𝐿𝐷 in
the variogram equation with a unit
range results in a variogram in the 𝑢
direction. Similarly, in calculating the
variogram in the 𝒗 direction, 𝐿𝑢 = 0
and 𝐿𝐷 = 𝐿𝑣 /𝑎𝑣 . Substituting in a
variogram equation with a unit range
results in calculating the variogram in
the 𝑣Ԧ direction.

11/19/2024 Geostatistics 77
 Example 3.4
From the estimated variograms in different directions, we observe that the direction of
maximum continuity is N30°E. The variogram model in that direction is defined as
𝛾 𝐿 = 0.2 + 0.4𝑀𝑆8,000 𝐿
The variogram in the direction of minimum continuity (N30°W) is defined as
𝛾 𝐿 = 0.2 + 0.4𝑀𝑆5,000 𝐿
Estimate the variogram value between two points located at (1,500, 4,000) and (4,000,
6,000).

11/19/2024 Geostatistics 78
 Example 3.4 - Solution
Typically, in practice, for a 2D system, we need to estimate a variogram between two
points whose coordinates based on the original axes are known. In this case, the (u’, v')
coordinates for the two points are given. Therefore, we can estimate the value of 𝐿𝑢′ and
𝐿𝑣 ′ , respectively, by
𝐿𝑢′ = 𝑢2′ − 𝑢1′ = 4,000 − 1,500 = 2,500
𝐿𝑣 ′ = 𝑣2′ − 𝑣1′ = 6,000 − 4,000 = 2,000

11/19/2024 Geostatistics 79
 Example 3.4 - Solution
These coordinates need to be transformed in the directions of maximum and minimum
continuities, respectively, before we estimate the variogram. With Eq. 3.74, knowing that
𝜃 = 30°,
𝐿𝑢 = 𝐿𝑢′ cos𝜃 + 𝐿𝑣 ′ sin𝜃 = 2,500 × 0.867 + 2,000 × 0.5 = 3,617.5
𝐿𝑣 = −𝐿𝑢′ sin𝜃 + 𝐿𝑣 ′ cos𝜃 = −2,500 × 0.5 + 2,000 × 0.867 = 484

11/19/2024 Geostatistics 80
 Example 3.4 - Solution
With Eq. 3.76, knowing that the range in the 𝑢 direction is 8,000 ft and the range in the 𝑣Ԧ
direction is 5,000 ft,
2 2 2 2
𝐿𝑢 𝐿𝑣 3,167.5 484
𝐿𝐷 = + = + = 0.4076
𝑎𝑢 𝑎𝑣 8000 5000

For L ≤ 1, substituting this value into an equation with a unit range gives
𝛾 𝐿𝐷 = 0.2 + 0.4𝑀𝑆1 𝐿𝐷
= 0.2 + 0.4 1.5 𝐿𝐷 − 0.5 𝐿3𝐷
= 0.2 + 0.4 1.5 × 0.4076 − 0.5 × 0.40763 = 0.4312

11/19/2024 Geostatistics 81
 Example 3.4 - Solution
The variogram in any given direction and at any given distance can be calculated in a
similar manner. As stated previously, we are not restricted to a particular set of models
in applying this technique.
For example, if our variogram model in N30°E is defined as
𝛾 𝐿 = 0.1 + 0.2𝑀𝑆8,000 𝐿 + 0.3𝑀𝐸5,000 𝐿
and the variogram model in N30°W is defined as
𝛾 𝐿 = 0.1 + 0.2𝑀𝑆5,000 𝐿 + 0.3𝑀𝐸7,000 𝐿
we can apply the same principles.

11/19/2024 Geostatistics 82
 Example 3.4 - Solution
In this case, we have to calculate two values of LD corresponding, to two different
models.
For the spherical model,
2 2 2 2
𝐿𝑢 𝐿𝑣 3,167.5 484
𝐿𝐷1 = + = + = 0.4076
𝑎𝑢 𝑎𝑣 8,000 5,000

and for the exponential model,

2 2 2 2
𝐿𝑢 𝐿𝑣 3,167.5 484
𝐿𝐷2 = + = + = 0.637
𝑎𝑢 𝑎𝑣 5,000 7,000

11/19/2024 Geostatistics 83
 Example 3.4 - Solution
The variogram equation with a unit range can be written as
0.1 + 0.2𝑀𝑆1 𝐿𝐷1 + 0.3𝑀𝐸1 𝐿𝐷2 for 𝐿𝐷1 ≤ 1, 𝐿𝐷2 ≤ 1
𝛾 𝐿 =൝
0.1 + 0.2 + 0.3 = 0.6 for 𝐿𝐷1 ≥ 1, 𝐿𝐷2 ≥ 1
Depending on the values of 𝐿𝐷1 , and 𝐿𝐷2 , the appropriate value of the variogram can be
calculated at any given lag distance and direction. Substituting,
𝛾 𝐿 = 0.1 + 0.2𝑀𝑆1 𝐿𝐷1 + 0.3𝑀𝐸1 𝐿𝐷2
= 0.1 + 0.2 1.5 × 0.408 − 0.5 × 0.4083 + 0.3 1 − exp −3 × 0.637 = 0.4712

11/19/2024 Geostatistics 84
 Zonal Anisotropy
Zonal anisotropy is observed where the
variogram in the directions of maximum
and minimum continuity show different
sills.
Typical zonal anisotropy is observed when
vertical vs. horizontal variograms are
compared. Changes in the vertical
direction are more rapid, representing
higher variance or sill value compared
with relatively smooth changes in the
horizontal direction, representing a
smaller sill value (Fig. 3.50).

11/19/2024 Geostatistics 85
 Zonal Anisotropy
Zonal anisotropy is more generalized than geometric anisotropy because it allows for
different sill values in two directions as well as different ranges. Zonal anisotropy,
however, does have some restrictions. It requires that the directions of maximum and
minimum continuity be perpendicular to each other and that linear combinations of
models used for modeling the two directions be the same.

11/19/2024 Geostatistics 86
 Zonal Anisotropy
For example, if we use a linear combination of a nugget and a spherical model in the 𝑢
direction as
𝛾𝑢 𝐿 = 𝐶0 + 𝐶1 𝑀𝑆𝑎𝑢 𝐿
We can write a model in the direction 𝑤 as
𝛾𝑤 𝐿 = 𝐶0 + 𝐶1 + 𝐶1′ 𝑀𝑆𝑎𝑤 𝐿
Note that both the range and the sill values differ in the 𝑢 and 𝑤 directions; however,
the same model is used in both directions.

11/19/2024 Geostatistics 87
 Zonal Anisotropy
As before, we are not to restrict the model to a combination of only two models; we
can use as many as we want as long as the sills of the models in both directions are the
same and the types of models used are the same.
For example, we can model the variogram in the 𝒖 direction as
𝛾𝑢 𝐿 = 𝐶0 + 𝐶1 𝑀𝑆𝑎𝑢 𝐿 + 𝐶2 𝑀𝐸𝑎𝑢 𝐿
1 2

and the variogram in the 𝒘 direction as

𝛾𝑤 𝐿 = 𝐶0 + 𝐶1 + 𝐶1′ 𝑀𝑆𝑎𝑤 𝐿 + 𝐶2 + 𝐶2′ 𝑀𝐸𝑎𝑤 𝐿
1 2

Note that we have the same linear combination in both directions except that the
ranges differ in the two directions.

11/19/2024 Geostatistics 88
 Zonal Anisotropy
By defining the variograms in only two directions, we can estimate the variogram in any
other direction using the same principle as in the case of geometric anisotropy. That is,
we convert the variograms into equivalent normalized variograms with a unit
range. After using the appropriate rotation transform, we define the magnitude of the
normalized vectors as
2 2
𝐿𝑢 𝐿𝑤
𝐿𝐷 = +
𝑎𝑢 𝑎𝑤

2
𝐿𝑤
𝐿′𝐷 =
𝑎𝑤

where au and aw = ranges in the u and w directions, respectively.

11/19/2024 Geostatistics 89
 Zonal Anisotropy
We then substitute the values of 𝐿𝐷 and 𝐿′𝐷 into a normalized variogram equation. For
example, the normalized variogram equation is
𝛾 𝐿𝐷 = 𝐶0 + 𝐶1 𝑀𝑆1 𝐿𝐷
and
𝛾 𝐿′𝐷 = 𝐶1′ 𝑀𝑆1 𝐿′𝐷
and the resulting variogram is
𝛾 𝐿 = 𝛾 𝐿𝐷 + 𝛾 𝐿′𝐷

11/19/2024 Geostatistics 90
 Zonal Anisotropy
Note that the two components are divided so that one component represents the
variogram that is common to both directions, and the other component represents the
variogram that is valid only in the 𝑤 direction. This type of separation ensures that the
variogram values in the limiting directions are satisfied.
For example, to calculate the variogram only in the 𝒖 direction,
𝐿𝑢
𝐿𝐷 =
𝑎𝑢
and 𝐿′𝐷 = 0
Substituting in Eq. 3.89 results in
𝐿𝑢
𝛾 𝐿𝐷 = 𝐶0 + 𝐶1 𝑀𝑆1
𝑎𝑢
and
𝛾 𝐿′𝐷 = 0
which is an equation for the 𝑢 direction only.
11/19/2024 Geostatistics 91
 Zonal Anisotropy
To calculate the variogram in only the 𝒘 direction,
𝐿𝑤
𝐿𝐷 =
𝑎𝑤
𝐿𝑤
and 𝐿′𝐷 =
𝑎𝑤
Substituting these values into Eqs. 3.89 and 3.90 gives
𝐿𝑤
𝛾 𝐿𝐷 = 𝐶0 + 𝐶1 𝑀𝑆1
𝑎𝑤
𝐿𝑤
and 𝛾 𝐿′𝐷 = ′
𝐶1 𝑀𝑆1
𝑎𝑤
Substituting into Eq. 3.91 gives
𝐿𝑤
𝛾 𝐿𝐷 = 𝐶0 + ሺ𝐶1 +𝐶1′ ሻ𝑀𝑆1
𝑎𝑤
which is the variogram equation for the 𝑤 direction.

11/19/2024 Geostatistics 92
 Example 3.5
The variogram in the 𝑢 direction is given as
𝛾𝑢 𝐿 = 5 + 20𝑀𝐸300 𝐿 + 30𝑀𝑆100 𝐿
and the variogram in the 𝑤 direction is given as
𝛾𝑤 𝐿 = 5 + 30𝑀𝐸50 𝐿 + 40𝑀𝑆30 𝐿
If the 𝑢 direction represents a horizontal plane and the 𝑤 direction represents a vertical
plane, estimate the variogram between two points located at (50, 100) and (120, 60).

11/19/2024 Geostatistics 93
 Example 3.5 - Solution
Because the major and minor axes coincide with the original axes, no rotation
transformation is needed. We can calculate
𝐿𝑢 = 120 − 50 = 70
and
𝐿𝑤 = 60 − 100 = −40
From the variogram equations, we can write
𝐶0 = 5
𝐶1 = 20
𝐶1′ = 30 − 20 = 10
𝐶2 = 30
𝐶2′ = 40 − 30 = 10

11/19/2024 Geostatistics 94
 Example 3.5 - Solution
For the exponential variogram,
2 2 2 2
𝐿𝑢 𝐿𝑤 70 −40
𝐿𝐷1 = + = + = 0.833
𝑎𝑢 𝑎𝑤 300 50
2
𝐿𝑤 40
𝐿′𝐷1 = = = 0.8
𝑎𝑤 50

11/19/2024 Geostatistics 95
 Example 3.5 - Solution
For the spherical variogram,
2 2 2 2
𝐿𝑢 𝐿𝑤 70 −40
𝐿𝐷2 = + = + = 1.503
𝑎𝑢 𝑎𝑤 100 30
2
𝐿𝑤 40
𝐿′𝐷2 = = = 1.33
𝑎𝑤 30

Using a normalized variogram equation gives

𝛾 𝐿𝐷 = 𝐶0 + 𝐶1 𝑀𝐸1 𝐿𝐷1 + 𝐶2 𝑀𝑆1 𝐿𝐷2
= 5 + 20 1 − expሺ−3 × 0.833ሻ + 30 = 53.36
Note that because 𝐿𝐷2 > 1, the spherical component contribution is equal to C2.
−3𝐿
𝑀𝐸𝑎 𝐿 = 𝛾 𝐿 = 𝐶0 1 − exp
𝑎

11/19/2024 Geostatistics 96
 Example 3.5 - Solution
Similarly,
𝛾 𝐿′𝐷 = 𝐶1′ 𝑀𝐸1 𝐿′𝐷1 + 𝐶2′ 𝑀𝑆1 𝐿′𝐷2
= 10 1 − expሺ−3 × 0.8ሻ + 10 = 19.1
Again, 𝐿′𝐷2 > 1, which makes the spherical component contribution equal to C2.
Therefore, the resulting variogram is
𝛾 𝐿 = 𝛾 𝐿𝐷 + 𝛾 𝐿′𝐷 = 53.36 + 19.1 = 72.46

11/19/2024 Geostatistics 97
 Cross Variograms
Cross variograms (and the corresponding cross covariance) represent the spatial
relationship between two variables located a certain lag distance apart.
For example, we may want to determine whether a permeability value at a given
location is spatially related to a porosity value a certain lag distance apart. If such a
relationship exists, we may be able to use it to improve the estimate of permeability
values at unsampled locations.

11/19/2024 Geostatistics 98
 Cross Variograms
The equation for the cross variogram is
1
𝛾𝑐 𝐿 = 𝐸 𝑋 𝑢 − 𝑋 𝑢 + 𝐿 𝑌 𝑢 − 𝑌 𝑢 + 𝐿
2
In practice, we can estimate the cross variogram between the variables x and y as
𝑛 𝐿
1
𝛾𝑐 𝐿 = ෍ 𝑥 𝑢𝑖 − 𝑥 𝑢𝑖 + 𝐿 𝑦 𝑢 𝑖 − 𝑦 𝑢𝑖 + 𝐿
2𝑛 𝐿 𝑖=1
Obviously, the estimation requires that both variable values be available at locations 𝑢𝑖
and 𝑢𝑖 + 𝐿. Unlike the standard variogram, the cross variogram does not need to be
strictly positive. If the two variables are negatively correlated with each other (as x
increases, y decreases), the cross variogram is also negative.

11/19/2024 Geostatistics 99
 Cross Variograms
The cross-covariance equation can be written as
𝐶𝑐 𝐿 = 𝐸 𝑋 𝑢 𝑌 𝑢 + 𝐿 − 𝐸 𝑋 𝑢 𝐸 𝑌 𝑢 + 𝐿
In practice, we can write Eq. 3.103 as
𝑛 𝐿 𝑛 𝐿 𝑛 𝐿
1 1 1
𝑐𝑐 𝐿 = ෍ 𝑥 𝑢𝑖 . 𝑦 𝑢𝑖 + 𝐿 − ෍ 𝑥 𝑢𝑖 × ෍ 𝑦 𝑢𝑖 + 𝐿
𝑛 𝐿 𝑖=1
𝑛 𝐿 𝑖=1
𝑛 𝐿 𝑖=1

𝛾𝑐 𝐿 = 𝐶𝑐 0 −𝐶𝑐 𝐿
where 𝐶𝑐 0 is the cross covariance at a lag distance of zero. This is calculated as
𝑛 𝑛 𝑛
1 1 1
𝑐𝑐 0 = ෍ 𝑥 𝑢𝑖 . 𝑦 𝑢𝑖 − ෍ 𝑥 𝑢𝑖 × ෍ 𝑦 𝑢𝑖
𝑛 𝑛 𝑛
𝑖=1 𝑖=1 𝑖=1
where n = total number of locations where both x and y variables are measured.

11/19/2024 Geostatistics 100

 Example 3.6
Table 3.8 shows data collected from a well. Estimate the cross variogram between the
permeability and the porosity data. Also estimate the corresponding cross covariance
between the same two variables.

11/19/2024 Geostatistics 101

 Example 3.6 - Solution
We use only three lag distances with Eq. 3.102
𝑛 𝐿
1
𝛾𝑐 𝐿 = ෍ 𝑥 𝑢𝑖 − 𝑥 𝑢𝑖 + 𝐿 𝑦 𝑢 𝑖 − 𝑦 𝑢𝑖 + 𝐿
2𝑛 𝐿 𝑖=1

11/19/2024 Geostatistics 102

 Example 3.6 - Solution
At L = 1 ft, there are six pairs and we can write
𝑦ො𝑐 1
1
= ሾ 8.25 − 9.00 35 − 50 + 9.00 − 6.25 50 − 21 + 6.25 − 5.00 21 − 16
2×6
+ 5.00 − 5.30 16 − 20 + 5.30 − 4.75 20 − 8 + 4.75 − 5.00 8 − 14 ሿ = 8.88

ϕ(u) ϕ(u + 1) k(u) k(u + 1)

8.25 9.00 35 50
9.00 6.25 50 21
6.25 5.00 21 16
5.00 5.30 16 20
5.30 4.75 20 8
4.75 5.00 8 14

11/19/2024 Geostatistics 103

 Example 3.6 - Solution
At L = 2 ft, there are five pairs and we can write
𝑦ො𝑐 2
1
= ሾ 8.25 − 6.25 35 − 21 + 9.00 − 5.00 50 − 16 + 6.25 − 5.30 21 − 20
2×5
+ 5.00 − 4.75 16 − 8 + 5.30 − 5.00 20 − 14 ሿ = 16.89

ϕ(u) ϕ(u + 2) k(u) k(u + 2)

8.25 6.25 35 21
9.00 5.00 50 16
6.25 5.30 21 20
5.00 4.75 16 8
5.30 5.00 20 14

11/19/2024 Geostatistics 104

 Example 3.6 - Solution
Similarly, at L = 3 ft, we have four pairs and can calculate
𝑦ො𝑐 3
1
= ሾ 8.25 − 5.00 35 − 16 + 9.00 − 5.30 50 − 20 + 6.25 − 4.75 21 − 8
2×4
+ 5.00 − 5.00 16 − 14 ሿ = 24.03

ϕ(u) ϕ(u + 3) k(u) k(u + 3)

8.25 5.00 35 16
9.00 5.30 50 20
6.25 4.75 21 8
5.00 5.00 16 14

11/19/2024 Geostatistics 105

 Example 3.6 - Solution
To estimate the cross covariance, we first need to estimate 𝑐𝑐 0 . Eq. 3.106 states that
𝑐𝑐 0
1
= ሺ8.25 × 35 + 9.00 × 50 + 6.25 × 21 + 5.00 × 16 + 5.30 × 20 + 4.75 × 8 + 5.00
7
× 14ሻ𝑛 𝑛
1 1
− ෍ 8.25 + 9.00 + 6.25 + 5.00 + 5.30 + 4.75 + 5.00 × ෍ሺ35 + 50 + 21 + 16
7 𝑛
𝑖=1 𝑖=1
+ 20 + 8 + 14ሻ = 20.53

11/19/2024 Geostatistics 106

 Example 3.6 - Solution
With the traditional relationship between the variogram and the covariance as stated by
Eq. 3.105,
𝛾𝑐 𝐿 = 𝐶𝑐 0 −𝐶𝑐 𝐿
for L = 1 ft,
𝑐𝑐 1 = 𝑐𝑐 0 − 𝑦ො𝑐 1 = 20.53 − 8.88 = 11.65
Similarly, for L = 2 and 3 ft, respectively,
𝑐𝑐 2 = 𝑐𝑐 0 − 𝑦ො𝑐 2 = 20.53 − 16.89 = 3.64
𝑐𝑐 3 = 𝑐𝑐 0 − 𝑦ො𝑐 3 = 20.53 − 24.03 = −3.50

11/19/2024 Geostatistics 107

 Example 3.6 - Solution
Fig. 3.51 shows a plot of the cross
variogram and the cross
covariance as functions of lag
distance. The behavior is similar
to the standard variogram and
covariance, respectively. This is
possible when the two variables
are positively related to each other.

11/19/2024 Geostatistics 108

 Example 3.6 - Solution
It is important to note that Eq. 3.104, which is a standard covariance equation, gives a
different result for the cross covariance from this result. With Eq. 3.104,
𝑛 𝐿 𝑛 𝐿 𝑛 𝐿
1 1 1
𝑐𝑐 𝐿 = ෍ 𝑥 𝑢𝑖 . 𝑦 𝑢𝑖 + 𝐿 − ෍ 𝑥 𝑢𝑖 × ෍ 𝑦 𝑢𝑖 + 𝐿
𝑛 𝐿 𝑖=1
𝑛 𝐿 𝑖=1
𝑛 𝐿 𝑖=1

11/19/2024 Geostatistics 109

 Example 3.6 - Solution
For L = 1 ft, we have six pairs and can write
1
𝑐𝑐 1 = 8.25 × 50 + 9.00 × 21 + 6.25 × 16 + 5.00 × 20 + 5.30 × 8 + 4.75 × 14
6
1 1
− 8.25 + 9.00 + 6.25 + 5.00 + 5.30 + 4.75 × 50 + 21 + 16 + 20 + 8 + 14
6 6
= 13.60
ϕ(u) k(u + 1)
8.25 50
9.00 21
6.25 16
5.00 20
5.30 8
4.75 14

11/19/2024 Geostatistics 110

 Example 3.6 - Solution
Similarly, for L = 2 and 3 ft, respectively, ϕ(u) k(u + 2)
𝑐𝑐 2 8.25 21
1
= ሺ8.25 × 21 + 9.00 × 16 + 6.25 × 20 + 5.00 × 8 + 5.30 9.00 16
5
1 6.25 20
× 14ሻ − 8.25 + 9.00 + 6.25 + 5.00 + 5.30 5.00 8
5
1 5.30 14
× 21 + 16 + 20 + 8 + 14 = 4.48
5

ϕ(u) k(u + 3)
1
𝑐𝑐 3 = 8.25 × 16 + 9.00 × 20 + 6.25 × 8 + 5.00 × 14 8.25 16
4
1 1 9.00 20
− 8.25 + 9.00 + 6.25 + 5.00 × 16 + 20 + 8 + 14 = 4.69 6.25 8
4 4
5.00 14

11/19/2024 Geostatistics 111