Applied Geostatistics

The Variogram
Deepak Devegowda

Geostatistical Analysis
Variography:
Geologic surfaces and features have
different scales and directions of
continuity.
A variogram is the metric which
describes the anisotropic behavior of a
regionalized variable.

Variograms can be calculated from

discrete, continuous or gridded data.
The analysis is done by calculating the
average squared difference between
pairs of measurements at different
separation intervals. The results are
presented as a graph of variance against
separation interval.

What is Geostatistics?
A set of mathematical tools that uses:
Deterministic (best estimate) and
Stochastic techniques

A set of methods that:

Can integrate different types of data
Can aid in reservoir characterization
Preserve heterogeneity and connectivity

For the purpose of:

Filling the interwell space
Assessing uncertainty

Stochastic what
(flipping a coin)?

Spatial Analysis & Modeling

Where it fits in the workflow?
QC
Data
Analysis

Spatial
Modeling

Data
Auditing

Structural
Model

HRGM

Stratigraphic
Model

Facies
Model

Petrophysical
Model

Variography

To introduce the concept of a variogram, it is first important

to understand the role of geostatistics. Geostatistics is a set
of statistical tools that allow us to analyze spatially
distributed data. In classical statistics, there is an underlying
assumption of data independence. This is not true in the
earth sciences and any science that uses data gathered from
a geographical coordinate system. The vary nature of a
surface requires dependency on distance and orientation.
The field of geostatistics was developed in mining industry by
geological engineers. It uses both deterministic and
stochastic methodologies to help us understand the behavior
of spatial data. It has the unique ability to not only integrate
different types of data, but also data with different scales of
volume support. It is useful in exploration geology as well as
reservoir characterization. It provides not only estimates of
values at unsampled locations, but provides the basis for
understanding the reliability and uncertainty of the estimate.

MIN=
3.1
P25=
6.2
P50=
8.4
P75=
11.5
MAX=
19.1
MEAN=
8.9
STD=
3.8
STD/MEAN= 0.43

Variography

When a variable is tied to geographic

space it is said to be regionalized. It
has the property of being predictable
with a high degree of certainty, but
not always.
For example, the most common
regionalized variable is a formation
top (structural elevation). With
reasonable data density, structural
elevation can be modeled accurately.
However, with decreasing data
density, non-uniformly distributed
data locations, and presence of
increasingly more complicated
structural patterns, the certainty of
our map decreases, or the degree of
uncertainty increases. Variograms are
models that quantify the scales and
directions of continuity in reionalized
variables.

Variables of interest: Tops, Phi, K,

Gravity, Magnetics, fracture
density, sand/shale volumes, etc.
Products of complex physical and
chemical processes.
Processes superimpose a spatial
pattern on the rock properties.
Important to understand and
quantify the spatial scales and
directional aspects.
Classical statistics tools do not
handle the spatial aspects
associated with the behavior of
these attributes.
Therefore, we need a new
approach to handle spatially
correlated data.

Spatial Analysis and Modeling

The variogram is a model of spatial

continuity that identifies and quantifies
the directions and scales of continuity.
That is, it identifies the orientations and
grain of the underlying geological
surface or body. It is ultimately used to
determine the weights in the Kriging
equations, the geostatistical estimation
method. Variography can be calculated
for any regionalized variable.

Identifies and quantifies directions

and scales of spatial continuity
Used to determine the weights
during interpolation or simulation
Applied to any Regionalized
Variable

( X i X (i h ) )
n

Variance

i 1

2n

Compute the average squared

difference between pairs of
measurements at different
separation intervals, known as
the Lag interval.

Distance (LAG)

(h)

Spatial Analysis & Modeling

Estimation Techniques:
Closest Point
Moving Average
Least Squares
Projected Slope
Triangulation
Polynomial Fit

Minimum Curvature
Masked Method
Adaptive Fitting
Inverse Distance
Kriging
Others

Interpolation
The estimated value is a weighted linear sum of
values at sampled locations:

Zo(estimate) = iZi
Zi: value at locations i

Z0

i : weight at locations Zi

i = 1, unbiasedness condition
Z

How are the weights () determined?

9

Z
2

Inverse Distance Weighting

Zo = aZa
1

d p

where:

d p

and dap = power of the distance from Za, (control points) to Zo, (grid node)
Control points are weighted by distance
Closer points have more influence

Lines of equal distance

10

What assumptions
are made when
using this Model?

Interpolation
Standard Methods
distance only

11

Geostatistical Methods
distance and direction

What is the Variogram?

( X i X (i h ) )
n

1
5

4
3

12

(h)

i 1

2n

What is the Variogram?

The Variogram

The progression of slides shows the concept of developing lags and the associated cloud of
points (squared differences) for each lag interval. Further, pairs of points can be compared
directionally to produce directional variograms.

( X i X (i h ) )
n

13

(h)

i 1

2n

Spatial Analysis and Modeling

Exercise: Compute the experimental semi-variogram for 5 lags

( X i X (i h ) )
n

Porosity %

First Lag

(h)

i 1

6
10
14
12
8
2

2
4

-42=16
-42=16
22=4
42=16
62=36

These are vertical well samples spaced 1-m apart,

which will be the Lag interval.
The objective is to compute the sum of the squared
differences between pairs of points separated by
the Lag interval.
What is the first pair of points in the group of
points that separated 1-m?
What is the first pair of points separated by 2h (2m)?
What is the first pair separated by 3h, 4h, 5h?
How many data pairs are in 1h, 2h, 3h, 4h and 5h?

02=0
-22=4
92/7=13.1

Only divide by n pairs because the calculation is done in one direction only.

Spatial Analysis and Modeling

Exercise:

Calculate the variogram values for lags 1, 2, 3, 4 and 5 meters

Plot the experimental variogram. Is there a pattern?

( X i X (i h ) )
n

(h)

i 1

n = the number of pairs

i = the first pair point
i+h = the second pair point

n
Porosity %
80

Note: only divide by n and

not 2n as you will do the
computations from top to
bottom only, even though
the computer does both
directions.

6
10
14
12
8

2
2
4

70
60
50

(h)

40
30
20
10
1

Lag (h)

Spatial Analysis and Modeling

Omnidirectional variogram: Should be computed first
Average scale for all directions, uses all data pairs
Best indicator of model type (e.g. Spherical, Exponential, Cubic, Gaussian)
Best indicator of a Nugget (discontinuity at the origin)

16

Spatial Analysis and Modeling

Model Types
All models have the same effective spatial scale: 3100-m

Exponential: least smooth

Cubic

Spherical

Gaussian: most smooth

Spatial Analysis & Modeling

Relationship Between the
Variogram & Covariance:

Distance = multiples of lag

intervals
Y axis = variance = mean squared
difference

(h)

Variogram

Distance

Distance = multiples of lag

intervals
Y axis = Cov = Sill - Variogram
Covariance is used in the kriging
equation because it is
computationally more efficient.

Cov (h)

Covariance

Distance

Spatial Analysis and Modeling

Sill = Data
variance

Long Scale

Short Scale

Anatomy of the Variogram:

Long Scale

Nugget Effect

Nugget Effect with Long Scale

Spherical Model

Nested Short and Long Scale

Spherical Models

Note: The Nugget acts as a low pass filter (removes short scale features) in kriging,
but adds short scale uncorrelated noise when performing simulation.

19

Impact of the Nugget Term

Effect of Increasing Nugget Values

Illustrates the impact of an increasing Nugget effect using a Unique (all data)
Neighborhood. Regardless of the amount of nugget the data values at the wells
are always honored if the well locations reside on a grid node. The maps are the
average porosity.

Impact of the Nugget Term

Effect of Increasing Nugget Values

Cross-section through the wells in the previous slide:

No Nugget
25% Nugget
50% Nugget
75% Nugget

100% Nugget

Illustrates the impact of an increasing Nugget effect using a Unique

Neighborhood. Regardless of the amount of nugget the data values at the wells
are always honored if the well locations reside on a grid node.

21

The Variogram in Perspective

Scales of Continuity

Directions of Continuity

22

Traditional Variograms and Variogram Map

Minimum direction
of Continuity

Maximum direction
of Continuity

23

Traditional Variograms and Variogram Map

24

Spatial Analysis and Modeling

Variogram Map Polar Plot

If you have enough well data or seismic data

then you can compute the Variogram polar
plot to determine the major and minor
directions of continuity and approximate
scales.
These parameters are used to compute and
model the anisotropic variogram.

25

Spatial Analysis and Modeling

Search Parameters
Because of irregular spacing of
input points, search criteria must be
defined to select points within
distance range given by the Lag.
When computing the horizontal
variogram from a 3D data set, the
vertical thickness (slicing thickness)
is equal to the layer thickness
(parallel bedding) or a single layer
(proportional bedding) to avoid an
apparent Nugget effect by mixing
across too many layers vertically.

What is the Variogram?

Search Parameters for the horizontal variogram

Because of irregular spacing of input points, search criteria must be defined to
select points within distance range given by the Lag.
North

Lag tolerance

Xi+h
Lag distance (h)
Angle

Points within this

area are accepted as
lying a distance h
from origin (at a
data location).

Angle
tolerance

Band
width

East

Xi

27

Spatial Analysis and Modeling

Vertical search thickness to compute horizontal variogram

Thickness interval (vertical cell thickness)

z
1 cell thick

Grid in stratigraphic space

XOY

28

Spatial Analysis and Modeling

Anisotropic, nested model

Structure 1 (Red)

Structure 2 (Green)

Spatial Analysis and Modeling

Nested Ellipse

Kriging Result

Simulation Result
Depending upon whether you will
perform kriging or simulation, the
images provide a visual image of the
result of applying a variogram model.

Spatial Analysis and Modeling

Why do I need a variogram model?

31

Kriging system requires

knowledge of correlation
function for all ranges and
azimuths
Smoothes experimental statistics
and introduces geological
information
To estimate the weights of
neighboring values
Ensures positive estimation
variance (only certain
mathematical functions satisfy
this condition)

Spatial Analysis and Modeling

Variograms - Special Considerations:
Behavior of the variogram is poor with few pairs
The data spacing and geobodies size can give a false impression of
a Nugget effect
Outliers adversely affect the variogram
The first few lags are most important for modeling (weights are
largest)
Avoid complex nested structures
The variogram should relate to a geological model
The variogram polar plot contains a great deal of information, but
requires a considerable amount of data

32

Savannah River Site, S.C.

Spatial Analysis and Modeling

Santee

Warely Hill
Green Clay

33

Congaree

Ellenton

Savannah River Site, S.C.

Maximum direction of continuity = N85oE
Minimum direction of continuity = N175oE
min

max

Santee

Variogram

Maximum intermediate direction of

continuity = N27oE
Minimum intermediate direction of
continuity = N117oE

34

Cleaning Up Gamma Ray Logs

Noise

35

Removing Random Noise

Filtering the Nugget Effect

Amplitude Horizon

36

Variogram Map

Removing Random Noise

D2 = 300

Filtering the Nugget Effect

D1 = 1500

Nugget = Noise

Variogram Map

Directional Variogram

Note: Nugget (noise) does not show up in the

variogram map.
37

Removing Random Noise

Filtering the Nugget Effect

Original
Noise

Filtered
38

Map Analysis
Factorial Kriging

39

The spatial pattern of many variables

results from the combination of different
factors which generally operate at distinct
spatial scales.
These scales can often be
distinguished on the experimental
semivariogram which appears as a
sum of nested structures.
Kriging analysis allows us to estimate
the spatial components related to the
different scales, and so separate local
and regional features of the
phenomenon under study or filter the
noise.
Often information is not only spatial but
also multivariate: samples are collected in
space and several measurements are
performed on each of them.
The different variables generally
correlate and, as in the classical factor
analysis, it is natural to presume that
they reflect some common
underlying factors.

As these factors often operate at

different scales and influence in
various ways the variables, the
variables must be expected to
correlate in a way that is different
according to the spatial scale
considered. Such attention to scaledependence may enhance a relation
between variables that is otherwise
blurred in an approach where all
different sources of variation are
mixed, leading to a better
understanding of the physical
underlying mechanisms controlling
spatial patterns.
Factorial kriging analysis allows one
to analyze relations between
variables at the spatial scales
detected and modeled from
experimental semivariograms.

Decomposition of the Variogram

Filtering spatial components

Original

Trend

Trend

Long &
Short
Scale

40

Long
Scale

Short
Scale

Spatial Analysis and Modeling

Answer:
80

70
69.0
60
60.0

60.0
50

(h)

40
40.7
30

20
13.1

10

Lag (h)

Applied Geostatistics
The Variogram
Deepak Devegowda