Selvi Misnia Irawati, S.Si., M.T.
KRIGING
INTRODUCTION
• Matheron (1962) introduced a correction by
weighting the sampling values by means of
variogram function.
• The name of ”kriging” adopted from the name of a
mine engineer (statistician) from South Africa D.G.
Krige who firstly thought about the matter since
1950.
DEFINITION
• Kriging assumes that the distance or direction between
sample points reflects a spatial correlation that can be
used to explain variation in the surface.
• The Kriging tool fits a mathematical function to a
specified number of points, or all points within a
specified radius, to determine the output value for each
location.
• Kriging is a multistep process. It includes exploratory
statistical analysis of the data, variogram modeling,
creating the surface, and (optionally) exploring a
variance surface.
• Kriging is most appropriate when you know there is a
spatially correlated distance or directional bias in the
data. It is often used in soil science and geology.
DEFINITION
Best Linear Unbiased Estimator (BLUE):
• Best : minimum error variance
• Linear : linear combination
• Unbiased : error expectation is zero
DEFINITION
• Kriging is an exact interpolator (it preserves the
observations) when there’s no nugget effect.
• Kriging weights depend only on locations and
covariance function, not on data.
• Kriging is a smooth predictor (if two points are close,
then their kriging estimates are close).
EQUATION
EQUATION
• The Interpretation of the Unknown Value z(u) and
the Data Values z(ua) as Realizations of the RV’s Z(u)
and Z(ua) Allows the Estimation Error to be Defined
As
RVEstimation _ Error = Z * (u) - Z (u)
where Z(u) = True Value
• The Estimation or Error Variance is Given by
s Est
2
_ Error ( u) = Var {Z *
(u) - Z (u)}
• All Kriging Varieties Share the Objective of
Minimizing the the Estimation or Error Variance, that
is E {Z * (u) - Z (u)} = 0
VARIATION OF KRIGING
• Kriging Varieties Can Be Distinguished According to the Model
Used for the Trend m(u)
• Simple Kriging (SK) Considers the Mean m(u) Known and
Constant In the Study Area. That is,
m(u) = m
• Ordinary Kriging (OK) Accounts for Local Fluctuations of the
Mean by Limiting the Domain of Stationarity of the Mean to the
Local Neighborhood W(u). That is,
m(u’) = Constant but Unknown
• Kriging with Trend (KT) Considers that the Unknown Local Mean
(m(u’)) Smoothly Varies Within Each Local Neighborhood W(u).
The Trend Component Is Modeled as a Linear Combination of
Functions fk(u). That is,
m(u’) = S ak(u’) fk(u’) with ak(u’) = ak Constant but Unknown
SIMPLE KRIGING
• Simple Kriging Equation
n(u)
é n(u)
ù
Z *
SK = å la SK
( u) Z ( ua ) + ê1 - å la ( u) ú × m
SK
a=1 ë a =1 û
• Unbiased Because the Error Mean = 0
• Weight of Mean
• Rewrite SK Estimator
n( u)
Z *
SK = å
a
la
=1
SK
(u) Z (ua ) + lmSK (u)m
• where
é n(u)
ù
l SK
m ( u) = ê1 - å laSK ( u) ú
ë a =1 û
SIMPLE KRIGING
• SK In Terms of C (Covariance Function)
n(u)
å
b
lb
=1
SK
(u) CR (ua - ub ) = CR (ua - u)
a = 1® n
• Error Variance
n( u)
s SK
2
(u) = CR (0) - å laSK (u)CR (ua - u)
a =1
SIMPLE KRIGING
Matrix Notation
C Matrix D Matrix
é C11 C12 C13 .. C1n ù é l1 ù é C10 ù
êC C22 .. .. .. ú êl2 ú ê C20 ú
ê 21 ú ê ú ê ú
êC23 .. .. .. .. ú ´ ê .. ú = ê .. ú
ê ú ê ú ê ú
ê .. .. .. .. .. ú ê .. ú ê .. ú
ê
ëCn1 .. .. .. Cnn ú
û ê ëln úû ê ëCn 0 ú
û
Weights
Solve for Weights (l) by Inverting C Matrix and
Multiplying Inverse by D Matrix
EXAMPLE
EXAMPLE SIMPLE KRIGING - 1
• Consider the data configuration, 𝑥 𝑢! =10, 𝑥 𝑢" =20,
m = 18. Further, assume that the isotropic variogram,
representing the spatial relationship, is given as
• Estimate the value at location 𝑢#. Also, estimate the
error variance. If the spatial model is given as
• How would the estimate and error variance
change?
SOLUTION – 1
• Assume the variogram is represented
by
• In this example, we have 2 sample • Because the covariance and the
points, or n = 2. Therefore, the matrix variogram are symmetric, the variogram
equation can be written as between 2 given points is the same,
irrespective of the directions.
• Relationship between the covariance
and the variogram
• In the previous matrix, the only • Distance between 𝑢! and 𝑢! is 30 ft,
unknows are λ1 and λ2. and between 𝑢" and 𝑢" is 50 ft.
• Substitute all the values,
• Because the distances between 𝑢!
and 𝑢! , 𝑢" and 𝑢" =0. C(0) is the sill of
the variogram.
SOLUTION – 1
• The error variance is
calculated
• The error variance is not
an absolute measure of
uncertainty in estimation;
however, it does indicate
how well the sample
data are configures
estimated values at the surronding the
unsampled location. unsampled location.
EXAMPLE SIMPLE KRIGING - 2
Example Data Set. Calculate Estimate for Location X.
Well X Y Value Distance to X
1 61 139 477 4.47
2 63 140 696 3.61
3 64 129 227 8.06
4 68 128 646 9.49
5 71 140 606 6.71
6 73 141 791 8.94
7 75 128 783 13.45
X 65 137 ????
SIMPLE KRIGING
• Simple Kriging Example = 607.3 at (64, 138)
• Note Estimate Consists of Two Parts
• Sum of (Weights x Sample Values)
• (1 - Sum of Weights) x Mean
Well X Y Value Distance to X Well 1 2 3 4 5 6 7
1 61 139 477 3.16 1 0.00 2.24 10.44 13.04 10.05 12.17 17.80
2 63 140 696 2.24 2 2.24 0.00 11.05 13.00 8.00 10.05 16.97
3 64 129 227 9.00 3 10.44 11.05 0.00 4.12 13.04 15.00 11.05
4 68 128 646 10.77 4 13.04 13.00 4.12 0.00 12.37 13.93 7.00
5 71 140 606 7.28 5 10.05 8.00 13.04 12.37 0.00 2.24 12.65
6 73 141 791 9.49 6 12.17 10.05 15.00 13.93 2.24 0.00 13.15
7 75 128 783 14.87 7 17.80 16.97 11.05 7.00 12.65 13.15 0.00
64 138 ???? 10.00 5.11 0.44 0.20 0.49 0.26 0.05
5.11 10.00 0.36 0.20 0.91 0.49 0.06
607.328 Estimate 0.44 0.36 10.00 2.90 0.20 0.11 0.36
7.107 Uncertainty 0.20 0.20 2.90 10.00 0.24 0.15 1.22
0.49 0.91 0.20 0.24 10.00 5.11 0.22
0.26 0.49 0.11 0.15 5.11 10.00 0.19
0.05 0.06 0.36 1.22 0.22 0.19 10.00
D Matrix C Matrix
3.873 0.136 -0.069 -0.003 0.000 0.000 0.000 0.000
5.113 -0.069 0.136 -0.002 -0.001 -0.009 0.000 0.000
0.672 -0.003 -0.002 0.109 -0.032 -0.001 0.000 0.000
0.395
Inverse of 0.000 -0.001 -0.032 0.111 -0.002 0.000 -0.012
1.126 C Matrix 0.000 -0.009 -0.001 -0.002 0.136 -0.069 -0.001
0.581 0.000 0.000 0.000 0.000 -0.069 0.135 -0.001
0.116 0.000 0.000 0.000 -0.012 -0.001 -0.001 0.102
603.7142857
0.169 80.430 0.653
0.417 290.530
Weights x Sample 2.134
0.039 8.869 Values 0.026
C -1
x D = 0.014 9.252 0.006
Weights 0.066 39.894 0.074
-0.001 -0.922 -0.001
0.004 2.784 0.000
430.838 Sum of (weights x Sample Values
176.490 (1-Sum of Weights) x Mean
Sum of Weights 0.708 607.328 The Estimate 7.107
The Uncertainty
