30‐Oct‐24

Covariance
• The mutual relationship between two random variables (X and Y) is computed
GLC518: Principles and Applications by their covariance.
• It is defined as the expected product of their deviations around their
Geostatistics respective means.

• For a limited number of samples:

Variogram Modelling

8
Correlation 51 Basic properties of covariance
• The covariance can be standardized to give a number which varies between ‐1 For given random variables X, Y, and Z, and constants a, b, and c
and 1 in order to quantify the degree of association of two variables.
• This is achieved by dividing it by the standard deviations of X and Y.
LC
G

Basic properties of variances Spatial Correlation

For given random variables X and Y, and constants a, b • Spatial correlation is the relationship between values of a variable at different
locations in space.
• Two data close to each other are more likely to have similar values than two data
that are far apart.

• Spatial Autocorrelation is a correlation of a variable with itself through space.

• “Everything is related to everything else, but near things are more related than
distant things” –Waldo Tobler

• Fundamental concept in spatial analysis as it impacts accuracy of predictions and

inferences.

1
 30‐Oct‐24

Importance of Outliers
Random variable and Regionalized variable
• Classical Statistics: Random variable, no
reference to spatial location and variability
• A random function is a family of random
variables that gives the function's values at
the various points in the function's domain.
• Geostatistics: Regionalized variable ‐ A
variable which has a deterministic component
Box and whisker plot (spatial location) and a random component
(value at any particular location)
• Examples: Spatial data (elevation,
temperature, climate, elemental
concentrations, etc.) or temporal data
(gradual varying information)

Spatial interpolation

8
Variables that can be modelled by random functions

sands, industrial minerals

51
• Metal grades for precious metals, uranium, base metals, coal, diamonds, beach

• Quality parameters e.g. for iron ore, silica, alumina; for gold, arsenic; for coal,
calorific value, ash & sulphur content; for cement, iron content, magnesium
Stationarity
• A spatial process is stationary, if its statistical properties such as mean and
variance are independent of absolute location, but dependent on the distance
and direction between two locations.
• Intrinsic stationarity: The variance
oxide, moisture of difference is same between
• Topographic variables such as seam thickness, overburden thickness, depth to a any two points that are at same
geological horizon, position of the sea floor distance and direction apart.
LC
• Rock type indicators e.g. for distinguishing between sandstone and shale in oil • Second‐order stationarity: The
reservoirs, or between different facies stationarity assumption is often
• Porosity and permeability, for both oil reservoirs and aquifers, hydraulic head too strong in practice. The mean
and transmissivity in hydrology can change over area of interest,
• Geochemical trace element concentrations in soil samples and stream sediments and variance may not be
bounded when this area of
• Pollutant concentrations in soil & water and in the atmosphere interest grows.
G

Variogram Variogram
• A variogram is a function that measures the variance of the difference between Suppose we have a total of N(h̅) pairs of samples separated by a vector h̅, then we
two variables at two locations. will estimate the variogram by:
• For two grades Z(x) and Z(x + h) at two points x and (x+h), h feet away from x, the
difference will be:

• Considering the average squared differences as dissimilarity function:

where V is the volume of the deposit and 2y(h̅̅) is the variogram, a function of a
vector (a distance and the orientation of that distance) If n samples are regularly distributed along a line at an interval of d feet
We will have (n‐1) pairs to compute γ(d), (n‐2) to compute γ(2d) and so on…
It expresses how grades differ in average according to the distance in that direction

2
 30‐Oct‐24

Variogram Variogram
Let us consider the following set of gold assay values (grade in units of dwts/ton of Let us consider the following set of gold assay values (grade in units of dwts/ton of
ore) of samples, each separated by a distance of 3 ft. taken from a segment of a ore) of samples, each separated by a distance of 3 ft. taken from a segment of a
gold bearing lode ‘O’ of gold field I, southern India. Compute the variogram with a gold bearing lode ' O ' of gold field I, southern India. Compute the variogram with a
interval of 3 ft and 6 ft. interval of 3 ft and 6 ft.

8
Variogram 51 Variogram
How will the variogram be different if we consider a set‐up where some values say
4th, 19th and 22nd are missing.
LC
G

Variogram
Calculate the experimental variogram for the first three distance classes for the
data given below. The samples are regularly spaced every 5m in 1D.

Calculate the experimental variogram for the first three lags and plot it. To see the
impact of the outlier, compare this variogram to the one obtained earlier.

3
 30‐Oct‐24

Variogram Variogram (E‐W)

56 grades arranged in a unit square grid have been presented below. Use them to
calculate the experimental variograms in the four main directions for distances up
to 4 lags. For each distance, note the number of pairs of points used. All pairs of
points a particular distance apart in a specified direction should be used, not just
those in the same row or column.

8
Variogram (NW‐SE) 51 Variogram
LC
G

Sill, Range and Nugget Properties of Variogram: Continuity

• SILL: The value at which the model first
flattens out.

• RANGE: The distance at which the model

first flattens out.

• NUGGET: The value at which the variogram

Good continuity High‐low‐scale variability
intercepts the y‐value.

• Partial Sill is the sill minus the nugget.

• Nugget effect can be attributed to The continuity is reflected

measurement errors or spatial sources of by the rate of growth of
variation at distances smaller than the y(h) for small values of h.
sampling interval or both.
No continuity

4
 30‐Oct‐24

Properties of Variogram: Zone of influence Properties of Variogram: Behavior near the origin
• Behavior of variogram for small values of h is also
important as it is related to continuity and spatial
regularity of the variable
a) Quadratic: This indicates that the regionalized
variable is highly continuous. A quadratic shape can
also be associated with the presence of a drift.
b) Linear: The regionalized variable is continuous, and
is less regular than above.
Range: 35' Range: 200' c) Discontinuous: It does not tend to zero as h tends to
0. The variable is highly irregular at short distances.
The zone of influence d) Flat: Pure randomness or white noise. The
(range) is the zone beyond regionalized variables Z(x+h) and Z(x) are
which the influence of a uncorrelated for all values of h, no matter how close
sample disappears. they are.
Range: ~0

8
Geometric (elliptic)
Properties of Variogram: Anisotropies 51 Types of anisotropy

1 = E—W
2 = N—S
3 = vertical
4 = cube diagonal

Variogram with Linear variogram Main axes coincide Main axes do not coincide
LC
Horizontal and vertical the same sill with different slopes with coordinate axes with coordinate axes
Identical porphyry
molybdenum deposit variograms in an iron mine Zonal or stratified
• Complex type of anisotropy
The anisotropies are easily depicted by computing the variogram in different directions. • More variation between strata than within them
• Sills of variograms are not same in all directions
If the variogram depends only on the magnitude of the distance between points (h), it is • Variogram is split into two components: isotropic one plus another depending
said to be isotropic. only on vertical component
G

Presence of a Drift Proportional Effect and Nested structure

• From a theoretical point of view, for large
distances, the variogram must increase very • A variogram is said to have a proportional effect
slowly. when its value (particularly its sill) is proportional
to the square of the local mean grade.
• This occurs with lognormally distributed data.
• In practice, we often find variograms which • The variograms for different zones have same
increase more rapidly than h2 shape but the sill in rich zones is much higher
than in poor ones.
• This indicates the presence of a drift
• Nested Structures indicate presence of variations
at different scales such as sample collection,
petrographic analysis, etc.

5
 30‐Oct‐24

Hole effects and periodicity Common Variogram Models

• In some cases, experimental variogram rises above its sill 1. Power functions: linear model
value, then drops down.
• As this "hump" in the variogram corresponds to a hole in the
covariance, the effect is called a hole effect.
• This is caused by less number of points used in computing
the experimental variogram value for a specific distance or
due to natural fluctuations in the variogram. 2. Spherical model
• Variograms, like covariances, can exhibit periodic behaviour.
• Ex: Sedimentary gold deposits
• Variograms calculated perpendicular to the ridges and valleys
can show periodicity but those parallel to ridges do not.
where C = sill and a = range
If there is nugget effect (Co), the same is added to the model.

8
Common Variogram Models 51 Variogram in Hydrogeology
3. Exponential model

4. Gaussian model
LC
5. Cubic model Variograms for a sample set of Variograms for sample set of
logarithmic values of Transmissivity logarithmic values of Specific capacity
G

Structural Analysis
Estimation vs Calculation
1. Step 1: Collect and check data
• Types of sampling and analyses used
• Different geological zones, faulting, etc.
• Estimate: an approximate calculation
• Preferential sampling, etc.
• Estimation: process of approximately calculating
• Whether to work with grades in 3D or accumulations in 2D
• Limits of area to study
• Calculation: This is an exact number
• Basic statistics (means, variances, correlations) are calculated and look for
• 2 * 5 =10
outliers or abnormal values and nonhomogeneous data (mixed populations)
2. Step 2: Calculate experimental variograms
3. Step 3: Fit a variogram model
4. Step 4: Kriging or simulation

6
 30‐Oct‐24

Interpolation Techniques
Interpolation vs. extrapolation
Nearly all the methods of prediction, including the simpler forms of kriging, can
• Interpolation: Estimation of a value at be seen as weighted averages of data. Thus we have the general prediction
an un‐sampled location from the known formula:

values within the neighborhood

• Extrapolation: estimation of a value at

an un‐sampled location from the known where x0 is a target point for which we want a value; the z(xi), i = 1, 2, . . . N,
at places xi are the measured data; and yi are the weights assigned to them.
values beyond the range of influence of
the known values in the neighborhood • How many nearby points should we include for a given un‐visited site?
• How to select these nearby points?
• How to allocate the weight for each nearby point?

8
Interpolation Techniques

Non geostatistical

Polygonal

Nearest‐neighbor
Geostatistical

Kriging (OK, SK)

Indicator Kriging (IK)

51
Inverse distance power Multiple Indicator Kriging (MIK)
Spline
Interpolation
Conditional simulations
LC
Deterministic Probabilistic

Polygonal IK

Nearest‐neighbor MIK

Inverse distance power Conditional simulations

Thiessen Polygon
Kriging (OK, SK) Interpolation
G