TA3201 Geostatistics for Resources Modeling

Concept on Estimation
Variance
Extension Variance and Estimation Variance

• Reserve estimation is characterized by an extension of one or

several known values to the area where the values are unknown.

• A known value (from a core sample or mining block) then extended

to the parts of orebody with unknown values.

Some estimation methods used in mining:

1. Estimation on average grade of an orebody based on assay data of
drilling or other sampling.
2. Estimation on a mine or mining blocks by means of polygon as
area of influence, inverse distance weighting, nearest point, etc.
 Sketch of
extension of one
point and
estimation of
several points

• The accuracy of each estimation is not always be interpreted using estimation

variance, but we can use statistics of the estimation values controlled by the
confidence interval.

• Estimation using one sample value where the value is extended to the larger
volume known as ”Extension”  Extension Variance.

• While estimation based on several values where the values are extended to the
volume known as “Estimation”  Estimation Variance.
Estimation of One Sample v and Extension to Volume V
(Extension Variance)

In a simple case, we assume that V is a volume of reserve (mining

block) with the average value of regionalized variable (grade,
thickness) is unkown.

Z V   Z x  dx  U
For V: 1
VV

While v is volume (drill core, mined ore blocks) with the values are
known.

Z v   Z x  dx  U *
1
For v: vv

The unknown U will be estimated using the known U*, it means

that value of v is extended to V.
 Through the extension, we will have an
error as:
Er  U *  U  Z v  Z V
This error is represented by a variance
known as Extension Variance of v to V.

The error variance or deviation is expressed by:

E (v to V) = Var Z v   
 Z V   E2 U*  U

According to Matheron (1971), the formula can be expressed as:

 E2 v to V   2 v, V    V, V    v, v 

 is the average of  x  y  where x and y are independent each
 v ,V 
other in volume v and V, i.e. x for volume v and y for volume V:

 v , V  
total  x  y  forallx in v dan all y in V 
totalpoints

dx   x  y dy
1
=

vV v V

 V ,V  is the average of  x  y  where x and y are the points

location independent each other in volume V:

 V, V  
total x  y  for allx in V andally in V 
numbertotalpoints

dx   x  y dy
1
= 
VV V V
 v ,v  is the average of  x  y , where x and y are points location independent each
other in volume v:

 v , v  
total x  y for allx in v andally in v 
numbertotalpoints

dx   x  y dy
1
= 
vv v v

In general, the basic relationship can be expressed as:

2
 E2 v to V  dx  x  y dy
vV v V
  covariancevV

dx   x  y dy
1
- 
VV V V
 varianceV

dx  x  y dy
1
vv v v
-  variancev

• The formula is valid for all domain, where v can be as: point, line, plane, or
volume, while V also can be as line, plane, or volume.
• Calculation on differential of the relation v to V can be represented by table or
nomogram.
Estimation Number of (N) Samples S and Extension to Volume V (Estimation
Variance)

In the field, we often know a number of samples values z(xi) at some locations xi.
The average value of block or reserve is estimated through the average sample
values: N

 z x 
1
U 
*
i
N i 1

Variance of deviation U* - U is known as estimation variance of V through N

samples.

Estimation of
some samples
points xi to the
volume V
Matheron (1971) gave the estimation variances as:
N

   x  y dy
2  covarianceNV
 2
N  
2
E i
NV i 1 V

dx   x  y dy
1
- 
VV V V
 varianceV

  x  yj 
N N
1  varianceN
- i
NN i 1 j 1

Or by means of S for N samples:  N

2
  E2  2 S ,V    V ,V    S , S 

The differentiation of that formula is analog to what we have done for extension
variance.
 S, S 
is influenced by the sample geometry,  V, V  is influenced by the
estimation geometry, while  S, V  is influenced by the relation between S and V.
 Spatial structure

Geometry S Relation S to V Geometry V

Sketch of estimation variance concept according to Huijbregts (1975)

Notes:

1. There is no different between extension variance and estimation variance.

Extension variance is related to the extension one sample to its area of influence,
while estimation variance is related to the extension number of samples to the
associated area of influence (ore block or ore reserve).

2. The problems of estimation can be solved by the available formula. In common,

sometimes the problem is solved by six order integration, where it is
approximated by summation method. Using table and nomogram will be helpful
in calculation.

Available data S Available data S

estimator Z s to unknown Z v

 
 E2  var error Z S  Z V  2 S ,V    V ,V    S , S 
3. Estimation variance is related so much to the relative position of S sample and
their geometrical relation to the estimated volume V. In the formula, we need
to notify the geometrical relation between S and V through  S,V  , and
geometrical relation within S samples through  S, S  , also within volume V
through  V ,V  .

The figure below shows the different geometrical relation between S samples
and volume V which will give the different estimation variance.

is better than

Relative position of S samples to the estimation volume V

4. The estimation variance is small if the position of samples close each others
and will result better volume estimation. It will produce more regular
variogram which means the regionalized variable within the deposit is more
continue.

5. Estimation variance has important meaning because it will predict how much
the reserve followed by the high accuracy, while we add some additional
drilling. Estimation variance  N2 depends so much on  h and geometrical
relation of samples.

6. Estimation variance is written as the square of average deviation. The

accuracy of estimation is quantified by estimation variance  N which
2

reflecting a measure on the dispersion of estimated values close to true values.

We can state about the dispersion type for the true values, then can give a
confidence interval to them. The confidence interval can be obtained from the
standard deviation of estimation  N   E   E2 or from the relative standard
deviation  N / U .
*
7. Some values below are used to give a conventional confidence interval to the
estimated values:
U = true values
U* = estimated values
 E2 = estimation variance
 = standard deviation of estimation
E

U  U* 2 E = confidence interval ~ 97.5% of confidence level on the

limitation of one side or with 2.5% of error factor.

U  U* 1 E = confidence interval ~ 85% of confidence level on the

limitation of one side or with 15% of error factor.

This condition is valid for normal distribution. For another distribution, the
interval is not changed much.
Calculation on Estimation Variance

  x  yj 
N N N
2
    1
  1
VV V V
Based on the basic formula:  
2
E  x i  y dy  dx  x  y dy  i
NV i 1 V NN i 1 j 1

or  E2  2 S, V   V, V   S, S

will be shown some examples on how to calculate the estimation variance.

Extension of Point Sample S along Line L (Point S in the Side of Line L)

 E2  2 S, L   L, L   S, S

 S, L    x i  ydy
1
1.L L

Integration can be replaced by summation, where length L divided into L for l.

Then N = 1, because the calculation if only for one point.
1 N
 S, L     i.L    L 
l i 1
The average values of  xi  y  where points xi and y dispersed along line L is
presented by the auxilliary function (L) or by the attached nomogram.

Auxilliary function of  L, L  F L has been discussed in the previous chapter
about dispersion variance:
 S , S   0 (because it is auto-correlation)
 x1  x1    0   0 (variogram of zero distance)

Example: L= 0.5 (0.5) = 0.359; F(0.5) = 0.245

σE2  (2)  (0.359)  0.245  0.0  0.473
Nomogram of auxilliary function (h), (h) and F(h) for variogram Spherical
Model with C0 = 0; C = 1
Extension of Point Sample S along Line L (Point S in the Middle of Line L)

 E2  2 S , L   L, L   S , S 

     
 S , L   1 2  L 2  1 2  L 2   L 2

 L , L   F L 
 S , S   0
 
 E2  2   L 2  F L 

Example: L = 0.5  0.5 2   0.186 ; F(0.5) = 0.245

 E2  (2)  (0.186)  0.245  0.0  0.127

We can conclude the result that one sample located in the middle of line L gives
better position compared in the side of line.
Defining variance estimation can be done easily using auxilliary functions of (L)
and F(L) either from the nomogram or table.
Extension of Point Samples S along Line L (Points S in Both Sides of Line L)

 E2  2 S , L   L, L   S , S 

1 N
• This example shows that
 S , L     x i  ydy
N .L i 1 V
with N = 2 positioning of one sample
= 1  2   L    L  located in the middle of line L
2
(although the number of sample
 L , L   F L 
is small), will give better result
  x i  y j 
1 N N
 S, S 
N.N i 1 j1 rather than the samples are
=
1
 x1  y 1    x1  y 2    x2  y1    x2  y 2  positioned in the sides of line.
22

=
1
 0    x1  y 2    x2  y1    0  • Defining variance estimation
22

=
1
 2   x1  y 2   1   x1  y 2  can be done easily using
22 2
auxilliary functions either from
in this case:  1 2   L the nomogram or table.
 E2  2   L  FL  1 2   L

Example: L = 0.5 (0.5) = 0.359; F(0.5) = 0.245;  0.5 = 0.688

 E2  (2)  (0.359)  0.245  (1 2)  (0.688)  0.129
Extension of Point Sample S within Plane R (Point S in the Side of Plane R)

 E2  2 S , R    R , R    S , S 

 S, R    x i  y dy
1
N  R R

The same as extension along line, the integral function can be replaced by
summation (S). Auxilliary function of Q(H,L) can be seen the next table.

 S , R   QH , L This function has been

discussed in the calculation of  S , S    0   0
  R , R   F H , L 
dispersion variance
Example: R  L  H  (0.4)  (0.8)

   
Q 0.4 0.8  0.629 ; F 0.4 0.8  0.451 ;  0  0.0

 E2  (2)  (0.629)  0.451  0.0  0.807

Auxilliary function of
point extension in
rectangular Q(h/a,l/a), for
variogram Spherical Model
with C0 = 0; C = 1
(Siemens, 1981)

Example: Q(0.4/0.8) = 0.629

Lateral Extension of Sample S within Plane R (Point S in the Middle of Plane R)

 E2  2 S , R    R , R    S , S 

This function
  
 S , R   ( 4 )  ( 1 4 )  Q H 2 , L 2  Q H 2 , L 2  has been
discussed in
 S , S    0   0

  R , R   F H , L  the calculation  
 E2  2  Q H 2 , L 2  F H , L 
of dispersion
variance
Example: R  L  H  (0.4)  (0.8)

 
Q 0.2 0.4  0.346 ; F 0.4  
0.8  0.451  0   0 ,0
;
 E2  (2)  (0.346)  0.451  0.0  0.241

This example shows that the sample position in the middle of plane gives much
lower estimation variance rather than the sample is positioned in the side of plane.
Extension in 3D (Space) of Point/Line Samples

Extension points in the space is similar as extension in lateral direction.

These extensions can be obtained by the similar way through some modification on
the basic formula.
Some nomograms for variogram Spherical Model has been available to define the
extension variance:

• Variance of extension point to line

• Variance of extension point to cubic plane
• Variance of extension point to rectangular plane
• Variance of extension line to rectangular plane
• Variance of extension line to box shape (3D)
• Variance of extension cubic plane to box shape (3D)
Nomogram of
extension/estimation
variance of point to
line for variogram
Spherical Model with
C0 = 0, C = 1 (Annels,
1991)

Example:   for h/a =

2
E

0.133 is 0.034
Nomogram for
extension/estimation
variance of point to
cubic plane for
variogram Spherical
Model with C0 = 0, C =
1 (Annels, 1991)

Example:   for h/a =

2
E

0.333 is 0.125
Nomogram of
extension/estimation
variance of point to
rectangular plane for
variogram Spherical
Model with C0 = 0, C =
1 (Annels, 1991)
Nomogram of
estimation variance of
points to rectangular
plane for variogram
Spherical Model with
C0 = 0, C = 1 (Annels,
1991)
Nomogram of
extension variance of
line to rectangular
plane for variogram
Spherical Model with
C0 = 0, C = 1 (Annels,
1991)
Nomogram of
extension
variance of line
to box shape for
variogram
Spherical model
with C0 = 0, C =
1 (David, 1977)
Nomogram of
extension
variance of plane
to box shape for
variogram
Spherical Model
with C0 = 0, C =
1 (David, 1977)
 Variance of Global Estimation

Total deviation of U *  U  in the process of estimation can be solved by summation

of each deviation as:

U   
N
*
 U   i U i*  U i
i 1

If we assume that each deviation are not dependant each other, we can obtain
global estimation variance as:

  Var U  U      Var U  U i    i2   E2 i 

N N
2 * 2 *
E i i
i 1 i 1

Global Estimation Variance for Samples along Line with Constant Spacing

Line with length L consists of a number of N segments l which is the area of

influence of point to segment: L  N  l
 Estimation variance of segment l has the weight as:
 E2 l   2   E2   2   E2  
  l L  l N l  1N 1
N
For all part of line L we get global estimation variance as:
N N
 L    li      E2      E2  
2 2 1 1
E E 2
i 1 i 1 N N

Problem:
On the traverse with length 160 m, the sampling is conducted with spacing 8 m. If the
distribution of Zn grade is represented by variogram of Spherical Model with C = 34%² and
a = 60 m, and average grade is z = 3.8%, then please calculate the global estimation
variance and its relative standard deviation!

Solution:
By the nomogram of estimation variance of Spherical Model variogram, we obtain:
   
 E2 l    E2 l a   E2 8 60   E2 0.133  0.034  C  0.034  34%2
Global estimation variance for N = 20 samples on line L is:
1 1
 E2 L    E2 l   0.034  34%2  0.058%2
N 20
Standard deviation =  E L   0.058%  0.24%
2

Relative standard deviation =  E L  / z  100%  0.24 3.8  100%  6.30%

Global Estimation Variance of Samples along Line with Different Spacing

If the spacing of samples is not same L  

i 1
li
, so the average calculation and
defining estimation variance must be performed using weighting.
li
Weighting factor : i 
L
1 N
Average values : z   li  zx i 
L i 1
N
 L   2   E2  
1
Global estimation variance : 2
E
L
l
i 1
i
2

Problem:
We take N = 10 samples along the traverse with length 160 m, the average grade
is z  3,8% Zn . The variogram of Zn grade is Spherical Model with C = 34%²
and a = 60 m (the same as previous problem).
Solution:
Samples spacing li are listed below:
II II / a  E2  
I i2 I i2   E2  

5 0,083 0,022 25 0,550

10 0,167 0,042 100 4,200
15 0,250 0,065 225 14,625
5 0,083 0,022 25 0,550
30 0,500 0,130 900 117,000
25 0,417 0,106 625 66,250
20 0,333 0,084 400 33,600
15 0,250 0,065 225 14,625
10 0,167 0,042 100 4,200
25 0,417 0,106 625 66,250
160 321,850

1
 E2 L   2
 321.85  34%2  0.43%2
L
 E L   0.43%2  0.65%
 E L  / z  100%  17.2%
Global Estimation Variance of Samples within Plane

For samples within the plane, we can use the same rule as we used for samples
located along the line.
Plane R consists of N parts with different
dimension ri. Each plane has one sample (i.e. drill
hole) located in the center part.

Estimation variance of each plane can be obtained

by reading the nomogram or table.

Extension of points to rectangular plane, and by

considering the weight factor, we can calculate the
global estimation variance as:
N N
 R   2  
1 1
2
E
R

i 1
r
i
2
 2
E ri 
R 2 r
i 1
i
2
  E2 ()

In case, the blocks have the same dimension, then the global estimation variance
is: 2
 E R     E2 ()
1
N
Problem:
In a phosphat deposit, 95 drill hole samples taken with grid 5050 m². The average
value of accumulation grade and thickness is z = 365 m%. The variogram model
is Spherical with sill C = 77,912 m² %², nugget variance C0 = 30,000 m² %², and
range a = 150 m.
Question:
Please calculate the global estimation variance and its relative standard deviation!
Solution:
With h/a = l/a = 50/150 = 0.333, we get extension variance of point to rectangular
plane 0.125  so:
 E2 r    E2 () = C0  C  0.125  30,000  (77,912)  (0.125)  39,739m2 %2

For the all part of plane we get the global estimation variance as:
1 1
 E2 R    E2 r    39,739m2 %2
N 95
 E2 R  418.3m2 %2
20.4
Relative standard deviation =   100%  5.6%
 E R  20.4 m% 365