TPE 5I7 GEOSTATISTICS AND

RESERVOIR MODELLING

PART II

1
 Introduction to Statistics

• In statistics, we generally want to study a population.

You can think of a population as a collection of persons,
things, or objects under study. To study the population,
we select a sample. The idea of sampling is to select a
portion (or subset) of the larger population and study
that portion (the sample) to gain information about the
population. Data are the result of sampling from a
population..

2
 Introduction to Statistics

• From the sample data, we can calculate a statistic. A statistic is

a number that represents a property of the sample. For
example, if we consider one math class to be a sample of the
population of all math classes, then the average number of
points earned by students in that one math class at the end of
the term is an example of a statistic. The statistic is an estimate
of a population parameter.
A parameter is a number that is a property of the population.
Since we considered all math classes to be the population, then
the average number of points earned per student over all the
math classes is an example of a parameter.

3
 Introduction to Statistics

• A variable, notated by capital letters such as X and Y, is a

characteristic of interest for each person or thing in a population.
• Variables may be numerical or categorical. Numerical variables
take on values with equal units such as weight in pounds and time
in hours. Categorical variables place the person or thing into a
category. If we let X equal the number of points earned by one
math student at the end of a term, then X is a numerical variable.
If we let Y be a person's party affiliation, then some examples of Y
include Republican, Democrat, and Independent. Y is a categorical
variable.
• Data are the actual values of the variable. They may be numbers
or they may be words.
4
 Introduction to Statistics

Sampling
• Gathering information about an entire population often costs too
much or is virtually impossible. Instead, we use a sample of the
population. A sample should have the same characteristics as
the population it is representing.
• Most statisticians use various methods of random sampling in an
attempt to achieve this goal. This section will describe a few of
the most common methods. There are several different methods
of random sampling. In each form of random sampling, each
member of a population initially has an equal chance of being
selected for the sample

5
 Introduction to Statistics

• Other well-known random sampling methods are the stratified

sample, the cluster sample, and the systematic sample.

• To choose a stratified sample, divide the population into groups

called strata and then take a proportionate number from each
stratum
• To choose a cluster sample, divide the population into clusters
(groups) and then randomly select some of the clusters.

• To choose a systematic sample, randomly select a starting point and

take every nth piece of data from a listing of the population
6
 Introduction to Statistics
• When you analyse data, it is important to be aware of sampling
errors and non sampling errors. The actual process of sampling
causes sampling errors. For example, the sample may not be large
enough. Factors not related to the sampling process cause non
sampling errors. A defective counting device can cause a non
sampling error.
• As a rule, the larger the sample, the smaller the sampling error.
• In statistics, a sampling bias is created when a sample is collected
from a population and some members of the population are not as
likely to be chosen as others.
• When a sampling bias happens, there can be incorrect conclusions
drawn about the population that is being studied.

7
 Introduction to Statistics

Data obtained during sampling can be analysed in two ways:

• Descriptive Statistics:
• Deals with the organization, presentation and summarization of data.
It allows for the understanding the characteristics of the information
and enables us to use the information more productively.

• Inferential Statistics:
• Deals with the procedures for deriving conclusions on the basis of the
data.

8
 Introduction to Statistics

• Frequency is the number of times a value of the data occurs.

• Relative Frequency is the ratio (fraction or proportion) of the

number of times a value of the data occurs in the set of all outcomes
to the total number of outcomes.

• Cumulative Relative Frequency is the accumulation of the previous

relative frequencies

9
 Introduction to Statistics

Probability normally is associated to a particular event of a random

experiment.
• A statement that there is 30% probability of finding oil at a location
where a new well is to be drilled can have two meanings:

• That in reservoirs with similar geology, 30% will produce oil. This is
related to the random nature of outcomes.
• That there is 30% probability that a well to be drilled will produce oil.
This is related to the incomplete knowledge of the reservoir.
Deterministic events can be treated as random events if we lack
sufficient knowledge about the events.
10
 Introduction to Statistics

There are Laws of Probability

• The probability value can never be less than zero or greater than 1.
• For mutually exclusive events, the probability of the union is equal to
the addition of the probability of individual events.
• P(A1) + P(A2) = P(A1 U A2)

• While for 2 events that are not mutually exclusive

• P(A U B) = P(A) + P(B) – P(A n B)

11
 Introduction to Statistics

Probability Function
• The probability function describes the probability that a random
variable will take a certain value.
• It is closely related to the relative distribution function.

Expected Value
• This is defined as some weighted average outcome of a random
experiment if the experiment is conducted a large number of times.
• E(x) = Xi. P(Xi)

12
 Introduction to Statistics

Important Distribution Functions

• In statistics, there are several important distribution functions with
known characteristics. These include:

• Uniform Distribution
• Triangular Distribution
• Beta Distribution
• Normal or Gaussian distribution
• Log-normal distribution

13
Introduction to Statistics

14
Introduction to Statistics

15
 Exercise

The probability of success is estimated to be 0.2 for an

exploratory well in Basin 1. For another exploratory
well in Basin 2, the probability of success is estimated
to be 0.3. If both wells are drilled, what is the
probability that both well will be successful?

16
 Solution to Exercise

P(A) = 0.2
P(B) = 0.3
Therefore,
P(A n B) = P(A) * P(B)
= 0.2 x 0.3 = 0.06
Therefore the probability of success in both wells is 6%

17
 Exercise

The probability of finding oil in an exploratory well is

estimated to be 0.2. One of the uncertainties in finding
oil in this well is the presence of source rock. The
probability of the presence of source rock is 0.7. After
the preliminary calculations were made, a well drilled
in a nearby area confirmed the presence of source rock
in the region. What is the probability of finding oil in
the first exploratory well given that the presence of
source rock is confirmed?

18
 Solution to Exercise
Let A = Event that oil is found in the exploratory well.
Let B = Event that the source rock is present.

For conditional probability, i.e. the probability of event A occurring given

that B has occurred is given by:

P(A|B) = P(A n B)/ P(A)

P(B) = 0.7, P(A n B) = 0.2
P(A|B) = 0.2/0.7 = 0.286

Therefore, the probability of finding oil improves to 0.286 because source

rock is present
19
 Introduction to Statistics

• Mean: The arithmetic average of sample data points.

• Median: Another measure of the central tendency. The sample point
that divides the sample into equal halves.
• Mode: Another measure of a central tendency. The observation that
occurs most frequently in the sample.
• Mean, median and mode coincide with each other if the distribution
is symmetric but exhibit different values if the distribution is skewed.
• Extremes: The maximum and minimum values
• Variance: Represents the spread of the data. It is a quantitative
measure of how widely the data are distributed.

20
 Measures of the Spread of the Data

Variance is a quantitative measure of how widely the data are distributed.

Mathematically,

Where, s2 is the sample variance, ẍ the mean sample and n the total number
of samples.
Variance can also be calculated as:

21
 Measures of the Spread of the Data

Coefficient of variation is expressed as:

Cv = S/ ẍ

Where, s2 is the sample variance, ẍ the mean sample and n the total number
of samples.

Standard deviation, s, is the square root of the variance

22
 Example Exercise

• The following data for pay-zone thickness (in feet) are

collected from all available wells in a reservoir: 6, 10,
20, 12, 20, 10, 15, 32, 27, 10, 18, 29, 8,17,23, 36, 19,
13, 33, 10, 26. Calculate the mean, median, mode,
25th percentile, 75th percentile, variance, Cv and
range.

23
 Solution

• First arrange the data in ascending order:

6, 8, 10, 10, 10, 10, 12, 13, 15, 17, 18, 19, 20, 20, 23,
26, 27, 29, 32, 33, 36. Calculate the mean, median,
mode, 40th percentile, 85th percentile, variance, Cv and
range.
Mean: (6+8+10+10+10+10+…………+33+36)/21 = 18.76
Median= (21+1)/2 or 11th term = 18
Mode = 10
24
 Solution

• Variance is given by:

• Cv = s/x = 8.95/18.76 = 0.477

25
 Solution

• Finding the kth percentile:

• i= k(n +1)/100

• When k=40, i= 40(21+1)/100 = 8.8 = Data in 9th position =

15.

• When k=85, i=85(21+1)/100 = 18.7 = Data in 19th position

= 32
26
 Finding Percentile

Finding the Percentile of a value in a data set:

Calculate 100(X+0.5Y)/N

• Then round up to the nearest integer

• X= No of data from start of list up to the value but not
including the value.
• Y=No of times the value is repeated
• N= Total no of data
27
 Finding Percentile

Listed below are the ages of 29 people:

18, 21, 22, 25, 26, 27, 29, 30, 31, 33, 36, 37, 41, 42, 47, 52,
55, 57, 58, 62, 64, 67, 69, 71, 72, 73, 74 , 76, 77

Find: a the 70th percentile and b. the 85th percentile:

28
 Finding Percentile

a. When k= 70:
i=70(29+1)/100 = 21.
21st position in the data set is 64.
i.e. 70th percentile is 64 years.
b. 85th percentile:
• i=85(29+1)/100 = 25.5
• Age in 25th position = 72 Age in 26th = 73
• Therefore 85th percentile = (72+73)/2 = 72.5 years.
29
 Spatial Data Set Analysis (Univariate)

Spatial data sets collected in petroleum reservoirs are unique in

several ways:
1. The data set do not represent random sampling . All wells are
not drilled on the basis of the same information available.
2. The data sets result in biased sampling. More samples
selectively come from regions with a higher potential.
3. The data sets may show local variability. A reservoir may
contain several different geological units.
4. Sample declustering is one technique of treating the influence
of biased sampling.
30
 Bivariate Statistics

There are many cases that involve the spatial analysis of data
sets of two variables. e.g porosity and permeability.
The commonest analysis tools are
Covariance, C(x,y)
Coefficient of correlation, r(x,y)
Linear regression

31
 Bivariate Statistics

Covariance is expressed as:

Coefficient of correlation is given as:

32
Variance and Correlation Coefficient: Case 1

33
 Solution: Case 1
Sample Porosity(X) K (Y) X2 Y2 XY
1 29.49 1156 869.6601 1336336 34090.44
2 26.79 531 717.7041 281961 14225.49
3 28.74 1059 825.9876 1121481 30435.66
4 27.65 822 764.5225 675684 22728.3
5 27.69 1014 766.7361 1028196 28077.66
6 22.69 109 514.8361 11881 2473.21
7 23.30 138 542.8900 19044 3215.4
8 23.81 166 566.9161 27556 3952.46
9 25.54 362 652.2916 131044 9245.48
Sum 235.70 5357 6221.544 4633183 148444.1

Mean 26.1888889 595.2222

S2 5.42478765 160508.6
S 2.32911736 400.6353
C(x,y) 905.580247
34
R(x,y) 0.97047956
Variance and Correlation Coefficient: Case 2

35
 Solution: Case 2
Sample Porosity(X) log K (Y) X2 Y2 XY
1 29.49 3.063 869.6601 9.381969 90.32787
2 26.79 2.725 717.7041 7.425625 73.00275
3 28.74 3.025 825.9876 9.150625 86.93850
4 27.65 2.915 764.5225 8.497225 80.59975
5 27.69 3.006 766.7361 9.036036 83.23614
6 22.69 2.037 514.8361 4.149369 46.21953
7 23.30 2.140 542.890 4.579600 49.86200
8 23.81 2.220 566.9161 4.928400 52.85820
9 25.54 2.559 652.2916 6.548481 65.35686
Sum 235.70 23.690 6221.544 63.69733 628.4016

Mean 26.1888889 2.632222

S2 5.42478765 0.148887
S 2.32911736 0.385859
C(x,y) 0.88742469
R(x,y) 0.98744140
36
 Linear Regression
• Oftentimes, it is desirable to know the relationship between two variables.
• The relationship between log K and φ in our previous example can be
expressed as:
y=mx +b
Where, m = slope of the curve and b the intercept.
It can be shown that the best fit can be obtained by defining m and b as:

m=C(x,y)/Sx2

b=ȳ - mẋ

37
 Linear Regression
• For our example,

C (log K, φ) = 0.8875
Sφ = 2.329
Sφ2 = 5.424
m= 0.8875/5.424 = 0.1636
b = log Ќ – m φ
= 2.6322-0.1636(26.19) = -1.652

The best fit equation is:

log K = 0.1636φ - 1.652

38
 Spatial Relationship

This is the relationship between a value measured at a point in one place,

versus a value from another point measured a certain distance away.

Describing spatial structure is useful for:

▪ Indicating intensity of pattern and the scale at which that pattern is exposed
▪ Interpolating to predict values at unmeasured points across the domain (e.g. kriging)
▪ Assessing independence of variables before applying parametric tests of significance

39
 Quantifying Spatial Relationship

Geological variables are often related to each other. The relationship

becomes stronger as the distance between neighboring values decreases
and becomes unrelated beyond a certain distance.

The qualitative information must be defined in a form suitable to be used

to estimate values at unsampled locations.

The variogram is the most commonly used measure of spatial correlation

40
 Areal Heterogeneity: Conventional Methods

• Conventional methods like inverse distance method weight the

value of each point by its distance to the cell being analyzed and
averages the values.
• They assume that an unknown value is influenced more by
nearby than far away points, but we can control how rapid that
decay is. Influence diminishes with distance.
• They have no method of testing for the quality of predictions, so
validity testing requires taking additional observations.
• They are sensitive to sampling, with circular patterns often
around solitary data points

41
Areal Heterogeneity: Conventional Methods

42
 Areal Heterogeneity: Conventional Methods

• In linear geostatistics, the estimate z*(u) at the unsampled

location u(x,y,z) is written as a linear combination of the n related
data z(uα):
Z*(u) = Σλα z(uα) (1.0)
Two critical questions arise:
Decision of stationarity: which data are retained to estimate the
unknown z(u)?
How should the weights λα be determined?

43
 Principles that govern determination of weights

• Data that are ‘closer’ to the target location should be more

informative and get more weight. Simply defining ‘closeness’ in
terms on distance could ignore geologic continuity. Geostatistics
incorporates information about spatial continuity into estimates
unlike IDW.
• Data values that are sampled close to each other are somewhat
redundant; hence each value of such a cluster is less valuable
than an isolated datum at the same distance from the unknown.
• Kriging is one of the geostatistical methods that take into account
these two principles (data correlation and redundancy) when
determining the weights λα in equation (1) above.
44
 Kriging Method
• Like Inverse Distance interpolation, Kriging forms weights from
surrounding measured values to predict values at unmeasured
locations.
• As with IDW, the closest measured values usually have the most
influence. However, the kriging weights for the surrounding
measured points are more sophisticated than those of IDW.
• IDW uses a simple algorithm based on distance, but kriging
weights come from a semivariogram that was developed by
looking at the spatial structure of the data. To create a
continuous surface or map of the phenomenon, predictions are
made for locations in the study area based on the semivariogram
and the spatial arrangement of measured values that are nearby.

45
 Kriging Method
• In other words, kriging substitutes the arbitrarily chosen weight
from IDM with a probabilistically-based weighting function that
models the spatial dependence of the data.
• The structure of the spatial dependence is quantified in the semi-
variogram
• Semivariograms measure the strength of statistical correlation as
a function of distance; they quantify spatial autocorrelation
• Kriging associates some probability with each prediction, hence it
provides not just a surface, but some measure of the accuracy of
that surface
• Kriging equations are estimated through least squares

46
 Kriging Method
• Kriging has both a deterministic, stochastic and random error
component
Z(s) = μ(s) + ε’(s)+ ε’’(s),
where
μ(s) = deterministic component
ε’(s)= stochastic but spatially dependent component
ε’’(s)= spatially independent residual error
• Assumes spatial variation in variable is too irregular to be modeled by
simple smooth function, better with stochastic surface
• Interpolation parameters (e.g. weights) are chosen to optimize fn
47
 Kriging Method

• Hence, the foundation of Kriging is notion of spatial

autocorrelation, or tendency of values of entities
closer in space to be related.
• This is a violation of classical statistical models, which
assumes that observations are independent.
• Autocorrelation can be assessed using a
semivariogram, which plots the difference in pair
values (variance) against their distances.
48
 Steps in Geostatistics
Application of Geostatistics is a three-step procedure:

1. Assumption of Stationarity
2. Spatial modelling of sample data using a variogram
3. Estimation of a variable value at unsampled locations
called kriging.

49
 Stationarity in Geostatistics
In addition to assuming that all locations are described by
random variables, we also have to consider the restrictions
associated with available sample data to predict values at
unsampled locations.

The assumption of stationarity requires that the proposed

model based on sampled data can adequately describe the
behaviour of the population.

The region of stationarity must not be too narrow (limited data)

or too broad especially where there is a lot of geological
variation

50
 Stationarity in Geostatistics
For geostatistical purposes, the first and second orders of
stationarity must be specified:

Or,

For first order of stationarity

And for second order of stationarity,

51
 Bivariate Relationships in Spatial Data

• The relationship between two variables can be quantified using

covariance:

• We can use the same bivariate relationship in spatial data. However,

we consider the same variable but examined at different spatial
locations.

52
 Bivariate Relationships in Spatial Data
• For two variables x and y, the covariance is given as:

• For value x, If we denote x(u) as a value at location u and x(u+L) as value at location u+L,
we can rewrite the equation above as:

L= distance between the two variables called lag distance

n= no of pairs located at a distance L apart.
53
 Bivariate Relationships in Spatial Data
• Similarly, the coefficient of correlation for the two variables x and y, is given as:

r(x,y) = c(x,y)/(SxSy)

• And, r[x(u), x(u+L)] = C[x(u), x(u+L)]/[Sx(u). Sx(u+L)]

• As in the case of covariance, the coefficient of correlation can be

written simply as a function of the lag distance:

r(L) = c(L)/(Sx(u)Sx(u+L)
54
Example

55
Example

56
Lag Distance of 1 ft (n=6)

57
 Lag Distance of 1 ft (n=6)

• Therefore, Sx = 1.638

• Similarly, S(x(u+1)) = 1.474

• Therefore,

• r(1) = c(1)/[Sx(u).S(x(u+1)] = 1.73/[1.638x 1.474]

= 0.7165

58
Lag Distance of 1 ft (n=6)

59
Lag Distance of 2 ft (n=5)

60
Lag Distance of 3 ft (n=4)

61
Effect of Lag Distance on Covariance

62
Effect of Lag Distance on Covariance

63
 Mathematical Expectation

• Expected value is defined as some weighted outcome of a random

experiment if the experiment is conducted a large number of times.
• For a discrete random variable X, the expected value of X is defined as:

• For a continuous variable,

• Where E[X] = expected value; xi = outcome of random variable,

• P[X=xi] = probability for the outcome.
64
 Mathematical Expectation

Example: What is the expected value of a rolling-a-die experiment?

Solution:
For a die experiment, probability of a random variable is 1/6

65
 Properties of Expected Value

• E[K] = K
• E[Ku(X)] = KE[u(X)]
• E[u1(X)+u2(X)] = E[u1(X)] + E[u2(X)]

Generally, the arithmetic mean µ, and variance σ2 of a single variable is:

µ = E[X]
σ2 = E[(X- µ)2]
• =E[X2]-{E[X]}2
• =E[X2] - µ2

66
 Properties of Expected Value
Example
Calculate the variance for the rolling-a-die experiment.
Solution
σ2 =E[X2] - µ2
µ = E[X]= 3.5

67
 The Variogram
The variogram is the most commonly used geostatistical
technique for describing the spatial relationship. It is half of
the variance between the two values located L distance apart.
Mathematically, it is defined as:

The value of the variogram is 0 at L=0 and increases as the lag

distance L increases, which is the exactly opposite of
covariance.

68
 The Variogram
Oftentimes, both the variogram and covariance can capture
spatial relationship adequately.
Why use the variogram instead of covariance?
• Tradition
• Estimation of the variogram requires restrictive assumptions
than the covariance:
variogram requires only an assumption that variance of the
difference between two values be finite, while calculation of
covariance requires that the variation of the data be finite.

69
Semivariance
•Semivariance (distance h) = 0.5 * average [ (value at location i–
value at location j)2] OR
n

  i
{ z ( x ) − z ( xi + h )}2

 ( h) = i =1
2n
•Based on the scatter of points, the computer (Geostatistical analyst)
fits a curve through those points
•The inverse is the covariance matrix which
shows correlation over space

by Austin Troy
 Variogram
• Plots semi-variance against
distance between points
• Is binned to simplify
• Can be binned based on just
distance (top) or distance
Binning based on distance only
and direction (bottom)
• Where autocorrelation
exists, the semivariance
should have slope
• Look at variogram to find
where slope levels
©2008 All lecture materials by Austin TroyBinning based
except where notedon distance and direction
 Variogram
• SV value where it flattens
out is called a “sill.” sill
• The distance range for which
there is a slope is called the
“neighborhood”; this is
where there is positive
spatial structure
• The intercept is called the
“nugget” and represents
random noise that is nugget
spatially independent range
©2008 All lecture materials by Austin Troy except where noted
Functional Forms

From Fortin and Dale Spatial Analysis:A Guide for Ecologists

 Modeling Estimated Variogram
There are two requirements for modelling estimated variogram:

1. Use a minimum number of parameters and models

1. Need to capture all the essential features observed in the estimated

variogram

74
 The Kriging Technique
The term kriging comes from Danny Krige, a South African geoscientist,
who first applied the technique to gold mines.
The kriging technique uses a linear estimation procedure to estimate a
value at unsampled locations.

Thus, the estimated value is a weighted average of the neighbouring

points.
All kriging techniques use this basic equation with minor variations
depending on the particular application

75
 The Kriging Techniques

• The goal, in the estimation procedure, is to calculate the

weights assigned to the individual neighbouring points. The
weights depend on the spatial relationships.

• The relationships are obtained from the modelling of the

variogram.

• The kriging algorithms use the minimum variance unbiased

estimation technique to estimate the weights.

76
Kriging Method
•The fitted variogram results in a series of matrices and vectors that
are used in weighting and locally solving the kriging equation.
•Basically, at this point, it is similar to other interpolation methods
in that we are taking a weighting moving average, but the weights
(λ) are based on statistically derived autocorrelation measures.
• λs are chosen so that the estimate z ( x0 ) is unbiased and the
estimated variance is less than for any other possible linear combo
of the variables.
Kriging Method
•Produces four types of prediction maps:
•Prediction Map: Predicted values
•Probability Map: Probability that value over x
•Prediction Standard Error Map: fit of model
•Quantile maps: Probability that value over certain quantile
Experimental Variogram
• Measures the variability of data with respect to spatial distribution
• Specifically, looks at variance between pairs of data points over a
range of separation scales

4/22/2021
Stationarity
• Stationarity implies that an entire dataset is
described by the same probabilistic process…
that is we can analyze the dataset with one
statistical model

4/22/2021
Stationarity and the Variogram
• Under the condition of stationarity, the
variogram will tell us over what scale the data
are correlated.
Correlated at any distance

Uncorrelated
(h)

Correlated at a max distance

h
4/22/2021 UB Geology GLY560: GIS
Variogram for Stationary Dataset
•Range: maximum
distance at which data
are correlated
Range •Nugget: distance over
Semi-Variogram

which data are

function

absolutely correlated or
Sill unsampled
•Sill: maximum
Nugget
variance ((h)) of data
Separation Distance pairs

4/22/2021 UB Geology GLY560: GIS

Variogram Models

4/22/2021 UB Geology GLY560: GIS