Kriging Method and Application
Kriging Method and Application
Kriging Method and Application
The relation between sampling (drillhole) grade and block grade shows a
sistematical scattering.
It means that drillhole sampling is not the best estimation for a block, so it is
need a correction.
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.
The correlation between grade of drillhole samples and true grade of blocks
which represented by the drillhole (obtained after mining the blocks) will give a
scatter plot which shows that the most of data (points) situated within the
ellipsoid as seen in the next figure.
Blocks
grade
Samples
grade
In case the grade analysis of samples is the right estimation to the grade of
each blocks, then the regression through the origin will be along line A-A.
The research from Krige for Au grade samples showed that in reality the
slope of regression line was a bit low as seen by line B-B (the next figure).
Blocks
grade
Samples
grade
Data scattering between samples grade vs.
blocks grade for Au (line B-B)
This means that the deviation is systematic and drillhole samples are not
the representative values for the blocks grade.
Samples grade higher than average value gives the higher value into the
blocks grade, if not corrected.
For example: Sample grade z1 gives block grade Z1 by line A-A which is
higher than the true block grade Z1 (line B-B).
On the other hand, the sample grade lower the average value gives the value
lower than block grade, for example: sample z2 by line A-A gives block grade
Z2 which is lower than the true block grade Z2 (line B-B).
Matherons correction by considering the variogram of regional data analysis
showed that estimation on blocks grade was not only influenced by the samples
within the blocks but also influenced by the samples around them in vicinity.
The correction gives:
- better estimated values,
- variance of estimation K
2
Samples Samples
(a) grade (b) grade
Data scattering between samples grade vs. Blocks grade which showed
mistaken mining
Sector 1: all blocks with grade > cog coincides to samples grade > cog !
have been mined
Sector 2: all blocks with grade < cog coincides to the samples grade < cog !
have not been mined
Sector 3: all blocks with grade < cog due to the mistaken samples grade > cog
! have been mined
Sector 4: all blocks with grade > cog due to the mistaken samples grade < cog
! have not been mined
If the regression line B-B which shows the correlation between samples and
blocks grade is plotted, then blocks with grade 5% is mineable although the
sample grade is 3,5% (Sector 1 in Fig. b).
Sector 4 in Fig. b which is not mineable due to the mistaken information will be
smaller, while Sector 3 which is mineable although the samples grade is low will
be larger. But in general, blocks with grade > cut-off grade (5%) and mineable
will be larger (Sector 1).
By means of Kriging method, we will not define the new better regression line,
but this method will correct the samples grade to be higher or lower until the
scattering within ellipsoid is tighter (see the next figure).
Blocks Blocks
grade grade
Samples Samples
grade grade
The change of ellipsoid of data scattering after
correction by Kriging method
Assume we have a group S1 of n samples with the same volume in a place xi.
Estimation of grade Z from the volume V is Z*. This estimated value obtained
by the weighting on samples grade z(xi):
n
Z* = i z ( x i )
i =1
n
By this way we will get estimated values without biased (unbiased), which
means the difference between Z and Z* insist to be 0.
E{Z Z *} = 0
By considering the weighting factor, we will get an estimation variance as:
E2 = Var [Z Z * ]
(x y )dx dy i j (x i x j )
2 n 1 n n
= i (x i y )dy
V i =1 V VV V V i =1 j=1
i =1 i =1 j=1
Estimation variance is the function of weighting factors which is known
their sum is 1. To choose the optimum weighting factors, the estimation
variance is set to be minimum.
Requirement that the sum of unkown i is 1 can be approximated by the
help of Lagrange multiplier () to minimize the following equation:
Q = E2 2 ( i 1) minimum
j (xi x j ) + =
n
1
( x xi )dx or
j =1 VV
(S , S ) + = (S , V )
n n
j i j i and
i =1
i =1
j=1
This system is used to define the values of i and which will generate
the minimum estimation variance.
Estimation variance (Kriging variance) is expressed by:
(x x j )dx or
n
1 1
2
K = dx ( x y )dy j
VV V V j =1 VV
= (V ,V ) + + j (S j ,V )
n
2
K
j =1
Notes:
The equations can be used if z(xi) for calculation of estimated values Z is
about equal to the average values of a number of samples in vicinity.
For example: (xi x j ) acts as the average of from the group of points Si
and Sj on locations xi and xj.
Following is the matrices of kriging system to calculate the unkown i and :
Matrices (S i S j ) is simetric.
The matrices system can be simplified to be:
[K ] [L] = [M ]
The matrices can be solved on L to get i and , so the equation is:
[L] = [K ] [M ]
1
= (V, V ) + [L] [M ]
2
K
t
The Effect of Geostatistical Parameters to the Weighting Factors and
Estimation Variance
Effects of geostatistical parameters will be explained by some simple example
below:
Known samples xi with grades z(xi) were taken with same distance (L=20 m) along
the line. Average grade of all sample is z = 1.0 . Variogram (Matheron model ) of the
data has parameters: C0 = 0; C = 1; a = 60 m
We will calculate the weighting factors, estimation variance (kriging variance), and
relative standard deviation for grade z * along the line L (i.e. on point x1).
In order to know how the effect of aureole samples and nugget variance, we will
consider the influence of one point (x1 itself), or influenced by three points (x1, x2,
x3) or by all points.
Nomogram of auxilliary functions (h), (h) and F(h) for Spherical Model, with
C0 = 0; C = 1; a = 1
a) Kriging system with considering only one sample
(S , S ) + = (S , L ) = (L , L ) + + j (S j , L )
n n n
j =1
j i j i
j =1
j =1 2
K
j =1
for n = 1 so 1 = 1, then: (L , L ) = C 0 + C F (L )
(S 1 S 1 ) = 0 = 0 + 1 F L ( a ) = F(20 60)
(S1L ) = C 0 + C. (
) = F(0,333) = 0,165
= C0 + C X L [ 2] = 0 ,124
= 0 + 1 X (10 )
60 1 (S1L ) = 1 0,124 = 0,124
= 0,124
K2 = 0 ,165 + 0 ,124 + 0 ,124 = 0 ,083
Kriging system:
1 (S 1 S 1 ) + 2 (S 1 S 2 ) + 3 (S 1 S 3 ) + = (S 1 L )
X (L' ) =
(L + L ) X (L + L ) (L ) X (L )
2 2 2 2
1 (S 2 S 1 ) + 2 (S 2 S 2 ) + 3 (S 2 S 3 ) + = (S 2 L ) L
1 + 2 + 0 = 1,000
K2 = 0 ,165 + 0 ,01 + 0 ,76 0 ,124 + 0 ,24 0 ,477
(S 1 S 1 ) = 0
= -0,165+0,01+0,208
(S 2 S 2 ) = 1 2 [C0 + C (2 L )] (
40 60
) = (0 ,667 ) = 0 ,851 K2 = 0 ,053 (the same as previous item)
= 1 [0 + 1 0,851] = 0,425
2
(S 2 S 1 ) = (S 1 S 2 ) = 0 ,481
(S 1 L ) = 0 ,124 (the same as previous item)
(S 2 L ) = 0 ,477
When we use three points as item b), all remaining sample points grouped to be
one sample with average grade z .
All average samples are assumed having large distance from points x1, x2, x3 and
slice L, therefore (h) and all auxiliary function of (h), F(h) are equal to 1 (=
total sill).
(S 1 S 3 ) = (S 3 S 1 ) = C0 + C (1,0 ) = 1,0
z ( x ) + z ( x )
3
Z* = 1 z ( xi ) + 2 2 + 3 {z (S 2 S 3 ) = (S 3 S 2 ) = C0 + C (1,0 ) = 1,0
{ 1442443 2 S
S 1
3
2S
(S 3 S 3 ) = C0 + C = 1,0
Kriging system:
All sample grouped in S3 have large
1 (S 1 S 1 ) + 2 (S 1 S 2 ) + 3 (S 1 S 3 ) + = (S 1 L )
position beyond (distance >> a), so
1 (S 2 S 1 ) + 2 (S 2 S 2 ) + 3 (S 2 S 3 ) + = (S 2 L ) all average value of all (x i x j )
1 (S 3 S 1 ) + 2 (S 3 S 2 ) + 3 (S 3 S 3 ) + = (S 3 L ) is equal to 1 (= total sill).
1 + 2 + 3 + 0 =1
(S 1 L ) = 0 ,124
(S 1 S 1 ) = 0 (S 2 L ) = 0 ,477
(S 2 S 2 ) = 0 ,425
As for example, we take samples with large
distance 6L = 120 m:
0 ,4812 + 0 ,266 1,116 2 = 0 ,124 [1]
X (L' ) =
(L + 6 L ) X (L + 6 L ) (6 L ) X (6 L )
0,142
L 2 = = 0,224
=
(
7L X 140
60
) 6L X(120 60) = 7 X(2,333) 6 X(2,0) 0,635
1 = 0,1162 + 0,734 = 0,760
L 3 = 0,226 1,1162 = 0,016
= 7 0,84 6 0,82 = 0,96 1,00
(3 can be ignored as the value
1 0 ,000 + 2 0 ,481 + 3 1,000 + = 0 ,124 [1] is too small)
1 0 ,481 + 2 0 ,425 + 3 1,000 + = 0 ,477
= (L , L ) + + j (S j , L )
n
[2] 2
K
1 1,000 + 2 1,000 + 3 1,000 + = 1,000 [3] j =1
(L , L ) = 0 ,165 = 0 ,000
1 + 2 + 3 + 0 = 1,000 [4]
0 ,760 0 ,124 = 0 ,094
0 ,4811 + 0 ,056 2 + 0 ,000 + 0 ,000 = 0 ,353 [1]-[2] 0 ,244 0 ,477 = 0 ,107
(S , L) = 0 ,217
n
(S , L ) = 0 ,652
n
11sample
conto 33samples
conto Semua conto
all samples
C0 0,0 0,3 0,5 0,0 0,0 0,3 0,5
1 1,0 1,0 1,0 0,76 0,76 0,54 0,47
2 0,12 0,22 0,34 0,34
3 0,12 0,02 0,12 0,25
K2 0,08 0,38 0,58 0,05 0,05 0,19 0,25
K 29% 62% 76% 23% 23% 43% 50%
Properties of Kriging Method
By means of Kriging method we obtain the best estimator based on the available
information of a mineral deposit. The weight factors is chosen to obtain the
minimum estimation variance.
So the Kriging considers about:
1. Structural and spatial correlation by means of function (h).
2. Relative geometrical relation among estimator data and volume by means of
(S i , S j ) (relation among data) and (S i ,V ) (relation between data and volume).
If the variogram is isotropic and data is regular, then kriging system will give
symmetrical data.
In many cases, samples inside and around the estimated block give estimation,
while sample far from the estimated block will have weights close to zero. In this
case the searching radius will not have influence (screened).
The screen effect will occur, if the nugget effect is zero or very small = C0 / C . The
nugget effect can reduce the screen effect. For the large nugget effect, all sample
will be considered having the same weight.
As explained previously that Kriging method utilize all information effectively
so we obtain the best linear estimation for the true values.
The main target of kriging estimator is to avoid the systematic error (over
estimate or under estimate) in reserve estimation.
This point will be indispensable when we have to use the reserve estimation to
decide the mineable or not mineable blocks.
Example of Kriging on Regular Grid
The calculation is performed for a block of mineral deposit with known variogram
parameter with Matheron model and C0 = 0; C = 1; a = 60 m; the average grade =
z = 1.0 .
Block is rectangular with size 20 m 30 m and there are 4 sample around it and 1
sample in the middle.
(S i , S j ) + = (S i , R )
3
j =1
j
(S 1 , S 1 ) = 0
1 40 (S 2 , R ) = C0 + C Q(R' )
(S 2 , S 2 ) = C0 + C = 0 ,5 0 ,852 = 0 ,426
2 60 30 15 10 15
3 Q , 1 Q ,
Q(R' ) = 60 60 60 60
1 60
(S 3 , S 3 ) = C0 + C = 0 ,5 1,000 = 0 ,500 2
2 60
3 1
= 0,638 0,241 = 0,517
20 2 + 30 2 2 2
(S 2 , S 3 ) = C 0 + C = (0 ,601) = 0 ,793
60
20
(S 1 , S 2 ) = C 0 + C = 0 ,481
60
30
(S 1 , S 3 ) = C 0 + C = 0 ,688
60
(S 3 , R ) = C0 + C Q(R' ' )
10 15
(S 1 , R ) = C 0 + C Q , = 0 ,241 45 10 15 10
60 60 3 Q , 1 Q ,
Q(R' ' ) = 60 60 60 60
2
3 1
= 0,536 0,241 = 0,683
2 2
30 20 Compare to the extension variance of:
(R , R ) = C 0 + C F , = 0 ,320
60 60
E2 ( ) = E2 30 , 20 = 0 ,158
60 60
Then kriging system being:
E = 40%
1 0 ,000 + 2 0 ,481 + 3 0 ,688 + = 0 ,241
1 0 ,481 + 2 0 ,426 + 3 0 ,793 + = 0 ,517 In general, the Kriging method for
1 0 ,688 + 2 0 ,793 + 3 0 ,500 + = 0 ,683 regular grid does not only
considered 4 samples around the
1 + 2 + 3 + 0 = 1,000
estimator block, but actually 8
samples as figured:
Solution for 4 equations with 4 variables:
8 2 9
1 = 0 ,57 2 = 0 ,26 3 = 0 ,17 = 0 ,00
= (R , R ) + + j (S j , R )
n
2
K
j =1
4 1 5
= 0,320 + 0,00 + (0,057 0,241 + 0,26 0,517 + 0,17 0,083)
= 0,320 + 0,387
= 0,067
Relative standard deviation:
7 3 6
K2
K = 100% = 26%
z
Estimated grade for the middle block (Block 1) is:
z (x2 ) + z (x3 ) z (x4 ) + z (x5 )
Z* = 1 z ( x 1 ) + 2 + 3 +
2 2
z(x 6 ) + z(x 7 ) + z(x 8 ) + z(x 9 )
4 + 5 z
4
In this case there are 6 linear equations system to define the weight i ( = 0 as
explained previously).
For the large nugget effect = C0 / C , we need to consider a group of sample beyond
the first ring.
Note:
The equation system for weighting each sample via i is also valid for all blocks
will be estimated with a condition that samples and blocks configuration are the
same.
To perform Kriging estimation on 66 blocks with
regular grid, we must calculate 4 weights factor
for 4 drillhole samples around the block.
Grade z* of a block can be estimated using sample grade in that block and samples
grade around the estimated block.
Because the samples in blocks are not positioned regularly, so the relation of
among points is replaced by relation of with estimated plane according to:
1
(S1S1 ) (R,R ) or (S1S2 ) (R, A )
n
1
1 (R , R ) + 2 (R , A ) + 3 (R , V ) + = (R , R )
n
1
1 ( A, R ) + 2 ( A, A) + 3 ( A,V ) + = ( A, R )
m
1 (V , R ) + 2 (V , A) + 3 (V ,V ) + = (V , R )
1 + 2 + 3 =1
On the other hand, we need to notice that the extension of deposit (V) is larger than
range a, so (V, R ) = (R , V ) = (V, A ) = (A, V ) = (V, V ) = C 0 + C = K and then = 0.
The relation of among planes which used to solve the equation can be obtained
by numerical calculation via integration as explained previously for the
differentiation of auxiliary function F.
Example of kriging application on mineral deposit (Royle, 1971)
sample grade
Conto di within
Nb of sample tengahblock aureol
Aureoles varians
Variance Simpangan baku
Standard deviation
1 8 3,68 1,9
1 7 3,99 2,0
1 6 4,25 2,1
0 6 8,43 2,9
Previous map/figure shows sample values (written as large font) and kriged values
below it (written as italic small font)
Histogram of original grades:
Minimum = 0.02; Maximum = 26.40; Average = 4.24; N = 85; Variance = 14.637;
Standard dev. = 3.826; Skewness = 2.820; Kurtosis = 15.090; Number of class = 12;
Interval = 2.5
Problems:
1. How many parts of 78 kriged blocks which have original sample grades
>3.0%?
2. How many parts of 78 kriged blocks which have kriged grades >3.0%?
3. Please remark the blocks with kriged grades >3.0% as mineable area!
References
Royle, A.G., A Practical Introduction to Geostatistics. Course Notes of the
University of Leeds, Dept. of Mining and Mineral Sciences, Leeds, 1971.
sample grade
The same mineral deposit is
calculated again using Kriging by
assuming that all drillholes exactly
located in the middle of grid/block.
The kriging result can be shown in the
figure beside.
( x1 x 2 ) = ( x 2 x1 ) = C 0 + C ( x1 x 2 )
( x1 x 3 ) = ( x 3 x1 ) = C 0 + C ( x1 x 3 )
X2
( x 2 x3 ) = (x 3 x 2 ) = C0 + C ( x 2 x3 )
X0
( x1 x0 ) = C 0 + C ( x1 x0 )
( x 2 x0 ) = C 0 + C ( x 2 x0 )
X3
( x 3 x0 ) = C 0 + C ( x 3 x0 )
x1 x3 x2
I-------------------I--------------------I
20 m x0 20 m
500 mN
x4
400
x3
300
V
?
x2
200
x1
100
There are 4 drillholes of x1, x2, x3, and x4 which located randomly in the middle of
regular grid where the grid spacing is 100100 m (see figure). Those drillholes
were derived from exploration of primary Au deposit. The Au grades for each
known points are: z(x1) = 10 ppm, z(x2) = 20 ppm, z(x3) = 5 ppm, and z(x4) = 15
ppm. If the variogram model for Au grades distribution is Spherical with range =
200 m, nugget variance (C0) = 5 ppm2, and sill (C) = 25 ppm2, then:
Continued.Exercise for Homework! Collected due to Final Test day!
T T
Seam 0m T Seam 0m
Seam 0m
-20 m -20 m
(m) -20 m (%) (%)
3.5 10 2.5
R
Seam
3.15 9.1 2.25
R
Seam 2.8 Seam
R 8.2 2
Q Q
Seam
1.75
Seam
2.45 7.3
-75 m Q -75 m
Seam
2.1 6.4 1.5
-75 m
P 1.75
P 1.25
Seam
5.5
-90 m P -90 m Seam -90 m
Seam
1.4 4.6 1
3.08125
VI-91
32.
II-47 VI VI-71
VI-28 VI-39
VI-17
SW
II-36
2.7625
NE
II-25
II-14
35. II
VII
Z (m)
2.125
.
1.80625
-6906.
-6916.
.
-6095
VI-91 VII-33
-6926.
.
1.4875
VI VII-24
-6085
30. 30.
VI-39 VII-3 VII-20
-6936.
VI-71
.
-6075
-6946.
VI-102
1.16875
.
VI-28
-6065
VI-17
-6055
II-47
II-36
0.85
25. II-14 a) 25. 0.53125 Isatis
0.2125
II-25
II
<0
-6100. -6090. -6080. -6070. -6060. -6050. N/A
X (m)
X (m)
-6100. -6090. -6080. -6070. -6060. -6050.
VII-3 VII >=0.338897 Fence diagram in SW-NE
40. 40.
VII-20 VII-24 VII-33
0.322302
VI-102
VI-71 VI-91
direction
II-47 VI-39 VI
VI-17 VI-28
II-25
II-36
0.305708
0.289113
II-14
35. 35.
VII
II
Z (m)
0.272519
Z (m)
VI-91 VII-33
0.255925
30.
VI-39 VI
VI-71
VI-102
VII-3 VII-20
VII-24
30. 0.23933
VI-17
VI-28
0.222736
0.206141
Sections of: (a)
II-47
II-36
25. II-14
II-25
b) 25. 0.189547
0.172953
II
<0.16189
-6100. -6090. -6080. -6070. -6060. -6050. N/A
estimated Ni+Co grades,
X (m)
X (m) (b) kriging standard
-6100. -6090. -6080. -6070. -6060. -6050.
40. II-47
VI-28
VI-39 VI VI-71 VI-91
VI-102 VII-3 VII-20 VII-24 VII-33 VII
40.
>=3.4
3.08125 deviation, and (c) profile
2.7625
VI-17
II-36
2.125
II
Z (m)
Z (m)
1.80625
1.4875
VI-91 VII-33
VI VII-24
test pit
VI-39
30. 30.
VII-3 VII-20
VI-71
1.16875
VI-102
VI-28
VI-17
II-36
II-47
0.85
25. 25. 0.53125
II
II-14
II-25 c) 0.2125
<0
-6100. -6090. -6080. -6070. -6060. -6050. N/A
X (m)
Application of Kriging using SGeMS (2004)
Kriged Pb in
HSZ (%)
Level 3025 m
1000 m
Krig.Var. of Pb
in HSZ
Level 3025 m
1000 m