KRIGING

by Danny Dorsel & Timothy La Breche

(These examples were created using statistical packages that utilize kriging.)

GEOSTATISTICS

Geostatistics is a branch of applied statistics developed by George Matheron of the

Centre de Morophologie Mathematicque in Fontainebleau, France. The original purpose
of geostatistics centered on estimating changes in ore grade within a mine. However, the
principles have been applied to a variety of areas in geology and other scientific
disciplines. A unique aspect of geostatistics is the use of regionalized variables which
are variables that fall between random variables and completely deterministic variables.
Regionalized variables describe phenomena with geographical distribution (e.g.
elevation of ground surface) The phenomenon exhibit spatial continuity; however, it is
not always possible to sample every location. Therefore, unknown values must be
estimated from data taken at specific locations that can be sampled. The size, shape,
orientation, and spatial arrangement of the sample locations is termed the support and
influences the capability to predict the unknown samples. If any of these characteristics
change, then the unknown values will change. The sampling and estimating of
regionalized variables are done so that a pattern of variation in the phenomenon under
investigation can be mapped such as a contour map for a geographical region.

SEMIVARIOGRAMS

Semivariance is a measure of the degree of spatial dependence between samples. The

magnitude of the semivariance between points depends on the distance between the
points. A smaller distance yields a smaller semivariance and a larger distance results in
a larger semivariance. The plot of the semivariances as a function of distance from a
point is referred to as a semivariogram. The semivariance increases as the distance
increases until at a certain distance away from a point the semivariance will equal the
variance around the average value, and will therefore no longer increase, causing a flat
region to occur on the semivariogram called a sill. From the point of interest to the
distance where the flat region begins is termed the range or span of the regionalized

variable. Within this range, locations are related to each other, and all known samples
contained in this region, also referred to as the neighborhood, must be considered when
estimating the unknown point of interest. Examples of possible semivariograms are
provided in the drawing below.

The center of the neighborhood is usually the unknown value. In order to determine this
value, all known values within the neighborhood are assigned weights using the
semivariogram. These weights and known values are then used to calculate the
unknown value. Problems with this method arise when the regionalized variable is not
stationary, but changes its average value from location to location. Defining a
neighborhood and assigning weights becomes impossible. Therefore, the regionalized
variable is divided into two parts, the residual and the drift. The drift is the weighted
average of points within the neighborhood around the unknown value. The residual is
the difference between the regionalized variable and the drift.
This residual then becomes a regionalized variable in itself which is stationary, and
thereby, allowing a semivariogram to be made. However, once again the problem of not
being able to define a neighborhood arises. Therefore, an arbitrary neighborhood is
chosen from which a drift can be calculated. The calculation includes the points within
the assumed neighborhood and a corresponding coefficient for each point which will be
explained in more detail in the kriging section. The only variable left in the calculation
is the semivariance; however, no semivariograms exist from which to obtain the
semivariances. Therefore, a reasonable semivariogram is assumed and compared to the
resultant residual semivariogram. If the two are the same, then the assumptions made
about the neighborhood and semivariogram are correct, and the unknown value can be
determined. If they differ, then another semivariogram must be used as an assumption
until they are the same.

KRIGING

Types of
Kriging

Theory of
Punctual Kriging

Example of
Punctual Kriging

Kriging is the estimation procedure used in geostatistics using known values and a
semivariogram to determine unknown values. It was named after D. G. Krige from
South Africa. The procedures involved in kriging incorporate measures of error and
uncertainty when determing estimations. Based on the semivariogram used, optimal
weights are assigned to unknown values in order to calculate unknown ones. since the
variogram changes with distance, the weights depend on the known sample distribution.

Types of kriging (as given in GEO-EAS 1.2.1 User's Guide)

Punctual Kriging (also termed Ordinary Kriging):

A variety of kriging which assumes that local means are not necessarily closely related
to the population mean, and which therefore uses only the samples in the local
neighborhood for the estimate. Punctual kriging is the most common method used in
environmental engineering.
Block Kriging: Estimating the value of a block from a set of nearby sample values
using kriging.
Point Kriging: Estimating the value of a point from a set of nearby sample values using
kriging. The kriged estimate for a point will usually be quite similar to the kriged
estimate for a relatively small block centered on the point, but the computed kriging
standard deviation will be higher. When a kriged point happens to coincide with a
sample location, the kriged estimate will equal the sample value.
Universal Kriging: Kriging similar to that of punctual kriging, but used when a trend,
or slow change in average values, in the samples exists.

Theory of Punctual Kriging

The equations in boxes can be clicked on in order to jump to their uses in the example
problem.
Punctual kriging is the simplest form of kriging. It uses dimensionless points to
estimate other dimensionless points, e.g. elevation contour plots. In punctual kriging,
the regionalized variable is assumed to be stationary, no drift exists. This assumption
allows for an estimate of an unknown value at point p, YE,P, to be calculated using a
weighted average of the known values or control points:

This estimated value will most likely differ from the actual value at point p, YA,P, and
this difference is called the estimation error:

If no drift exists and the weights used in the estimation sum to one, then the estimated
value is said to be unbiased. The scatter of the estimates about the true value is termed
the error or estimation variance,

or as its square root, known as the standard error of the estimate;

The estimate and estimation error depend on the weights chosen. Ideally, kriging tries to
choose the optimal weights that produce the minimum estimation error. In order to
derive the necessary equations for kriging, extensive use of calculus is required and will
not be included here; however, information about the derivation can be found in Clark
(1979) and Olea (1975). Optimal weights, those that produce unbiased estimates and
have a minimum estimation variance, are obtained by solving a set of simultaneous
equations. For simplicity and to illustrate the methodology of kriging, three unknown
values, Y1, Y2, and Y3, will be used to estimate an unknown value at point p, YE,P. Three
weights must be determined W1, W2, and W3, to make an estimate. The kriging
procedure begins with the following three simultaneous equations:

where
is the semivariance between control points i and j corresponding to the
distance between them , h. Since hij = hji, the left-hand matrix is symmetrical, with
zeroes long the diagonal since the distance from a point to itself is zero. The values of
the semivariances are taken from the known or estimated semivariogram. To assure that
the solution is unbiased, a fourth equation is used

Also, a fourth variable is introduced called the Lagrange multiplier,

, to assure that

the minimum possible estimation error is obtained. Therefore, the complete set of
simultaneous equations are:

Separating these equations into matrix form yields

or in general terms,

This matrix equation is solved for the unknown coefficients,

. The values in matrix
A and B are taken from the semivariogram or the mathematical expression describing
its form. Once the individual weights are known, an estimation can be made by

and an estimation variance can be calculated by

The methods used in kriging allow it to have an advantage over other estimation
procedures in that the estimated values have a minumum error associated with them and
this error is quantifiable.

Example of Punctual Kriging

The following example involves estimating the elevation of a water table at various
point knowing the elevation at three specified points. The following chart shows the

three known wells and their elevation in meters. The unknown elevation point is labeled
p.

The given information is provided in the following table:

Well x-coordinate y-coordinate Water Table Elevation
1

150

110

140

unknown

The distances between Wells and Location p

1 2

1 0 3.16 5.39 2
2- 0

3.61 1.41

3- -

3.61

Using a linear semivariance (this would have to be determined from an analysis of the
area under investigation, but for simplicity, a linear semivarianvce will be used) that has
a slope of 4 square meters per kilometer and a y-intercept of zero within a neighborhood

of 20 km, the following Semivariances for Distances between Wells and Loacation p
were calculated (i.e. semivarinace = 4 * distance):
1

1 0 12.65 21.54 8
2- 0

14.42 5.66

3- -

14.42

The equations that must be solved to find the weights in this example are

In matrix form, these equations become:

In order to determine the weights, the inverse of the left-hand matrix needs to be
determined, and in this example it is

By multiplying the inverse matrix by the right-hand matrix, the weights can be found,
and for this example they are

The estimate of the unknown value, YE,P, can now be calculated:

YE,P = 0.3805(150) + 0.4964(110) + 0.1232(140) = 128.9 meters

The estimation variance can also be calculated:

Se^2 = 0.3805(8.0) + 0.4964(5.66) + 0.1232(14.42) - 0.9319(1.0) = 6.70 meteres^2

The standard error of the estimate is the square root of the estimation variance and
equals:
Se = (6.70)^0.5 = 2.59 meteres

If it is assumed that the errors of estimation are normally distributed, then the standard
error can be used as a confidence band around the true value. Therefore, the probability
that the true elevation is within one standard error above or below the estimated value is
68%. Two standard errors away would give a confidence of 95%. For this example, the
water table elevation at location p is
YP = 128.9 +/- 5.18 meters, with 95% probability

If the elevation of another point, q, needs to be determined, then the same prodecure is
used with the new unknown coordinates. Therefore, for this example the coordinates of
point p, (3,2) would be changed to point q, (4,4). The weights assigned to the three
known samples using the new unknown point are

Applying the same procedure as above, the water table elevation at location q is
YP = 138.6 +/- 6.45 meters, with 95% probability

The weight changes at each well due to the location of the unknown point can be seen in
the following illustration, where the thickness of the arrow represents the weight
assigned to that particular well when determining the unknown elevation of the location
toward which the arrows point. Click the following line to view the changes and then
select BACK to return to this page.
Effect of unknown location on weights of known points

References
1. Statistics and Data Analysis in Geology, John C. Davis, 1973
2. GeoEAS Manual