Block Estimation: Covariance Between A Point and A Block

Download as pdf or txt
Download as pdf or txt
You are on page 1of 33

2.

Block estimation

Covariance between a point and a block


n
X *
v ( u0 ) = 0 +  i X ( ui )
i =1

X v* ( u0 ) = the estimate of block valuelocated at u0


X ( ui ) = the po int values surrounding theunsample block

 n

o = m1 −  1 
 i =1 

  jC (ui , u j ) = Cv (ui , u0 )
n
   
j =1 for i = 1, … , n
The arithmetic average of all the poin-to-point covariance between the sampled
point, ui and all the point, u0j, located within the block .

  1 nb  
(
Cv (u0 , ui )   C u0 j , ui
nb j =1
) nb= selected number of
points within a block

     
 C (u1 , u1 )  C (u1 , un )  1   Cv (u0 , u1 )
     =  
    
       
C (un , u1 )  C (un , un )  n  Cv (u0 , un )

n
We have X ( u0 ) = 0 +  i X ( ui )
*
v
i =1

  2
 
ˆ = Cvv (u0 , u0 ) −  iCv (ui , u0 )
2
E
i =1
Numerical Example 4.2 porosity value
using block kriking page 109

Field Example 4.3 – Simple Block KrigingTo Generate


Gross Thickness Maps. Page 112
3. Ordinary Kriging

 n

X (uo ) = o +  i X (ui )
*

i =1

  
E X (uo ) − X (uo ) = 0
*

  n

o = m(uo )1 −  i 
 i =1 
n
We force the λo to be zero, so 
i =1
i =1
 n

X (uo ) =  i X (ui )
*

i =1

  jC (ui , u j ) +  = C (ui , uo )
   

     
 C (u1 , u1 )  C (u1 , un ) 1   1   C (u1 , u0 ) 
    
       
 C (u , u )  C (u , u )
 =  
1   n   C (un , u0 )
 n 1 n n
   
 1    
0      
 1 1 
The weight λi

−1
=C c

Estimeted value at uo

 n

X (uo ) = o +  i X (ui )
*

i =1

And the variance

  n
 
 = C (uo , uo ) −  1C (ui , uo ) − 
ˆ2
E
i =1
Numerical Example 4.3. page 113

Two configurations for Numerical Example 4.3


   
x(u1 ) = 20, x(u2 ) = 50, x(u3 ) = 30, and x(u4 ) = 100.

a. Isotropic Variogram
 
()
 L = 100 M s 500
(L ) ,
b. Isotropic Variogram
 
()
 L = 50 + 50 M s 500
(L )
and
c. Anisotropic Variogram
 
() ()
 L = 100 M s500 L east/west direction,

and
 
()
 L = 100 M s 250
(L ) north/south direction.
Estimate the value and the error variance at
location uo the previous three models for
both configurations !
Others exercises for Ordinary Kriging

Field Example 4.4 – Ordinary


Kriging CrossValidation

Field Example 4.5 – Ordinary KrigingTo Generate


Gross Thickness Maps.

Field Example 4.6 – Ordinary KrigingTo


Estimates Original Oil in Place
Cokriging
( ) ( )
n n
X * ( uo ) =  X* i X u X i +  Y*k Y uYk
i =1 k =1

Unbiased estimation
E  X * ( uo ) − X ( uo )  = 0
n n
mX  X i + mY  Yk − mX = 0
i =1 k =1
n n


i =1
Xi = 1 and 
i =1
yi =0
Minimum Variance
 
( ) ( )
n n
Var  X ( uo ) −  X i X u X i −  Yk Y uYk 
 i =1 k =1 

( ) ( ) ( )
n m


j =1
Xj C X u X i , u X j +  Yk CC u X i , uYk +  X = C X u0 , u X i for i = 1,..., n,
k =1

and

( ) ( ) ( )
n m


i =1
Xi CC u X i , uY j +  Yl CY uYk , uYl + Y = CC u0 , uYk for k = 1,..., m.
l =1
CX

(u X1 , u X1 ) (
C X u X1 , u X n ) (
CC u X1 , uY1 ) (
CC u X1 , uYm ) 1 0

 
 
C X (u X1 , uXn ) (
CX uX n , uX n ) (
CC u X n , uY1 ) (
CC u X n , uYm ) 1 0
 
 CC (u X1 , uY1 ) CC (u Xn , uY1 ) C (u
Y Y1 , uY1 ) C (u
Y Y1 , uYm ) 0 1
 
 
 
 CC (u X1 , uYm ) (
CC u X n , uYm ) (
CY uYm , uY1 ) (
CY uYm , uYm ) 0 1
 1 1 0 0 0 0
 
 0 0 1 1 0 0 

  X1   C X ( u , u ) 
0 X1
   
   
   C X
 Xn  
(u , u )
0 Xn 

 Y1   C
 = C
(u , u )
0 Y1 

   
   
 Ym   CC (u , u )
0 Ym 
 X   
   1


 Y  
 0 
Numerical Example 4.4

Fig. 4.33 – Sample configuration for Numerical Example 4.4

         
C X (u3 , u3 ) CC (u3 , u1 ) CC (u3 , u2 ) CC (u3 , u3 ) 1 0   X 1  C X (u3 , u0 )
 C (u , u )           
 C 3 1 CY (u1 , u1 ) CY (u1 , u2 ) CY (u1 , u3 ) 0 1  Y1   CC (u1 , u0 )

       
CC (u3 , u2 ) CY (u2 , u1 ) CY (u2 , u2 ) CY (u2 , u3 ) 0 1  Y2  CC (u2 , u0 )
          =   
 C 3 , u3 )
C (u CY (u3 , u1 ) CY (u3 , u2 ) CY (u3 , u3 ) 0 1  Y3   CC (u3 , u0 )
  
1 0 0 0 0 0   X   1 
    
 0 1 1 1 0 0  Y   0 
 
C X (u3 , u3 ) = C X (0) = 25,
   
CC (u1 , u3 ) = CC (0) −  C (u1 , u3 ) = CC (0) −  C (265)
 
= 4 − 1 + 3M s500 (265) = 0.838,

   
CY (u1 , u2 ) = CY (0) −  Y (u1 , u2 ) = 1.2 −  C (100)
 
= 1.2 − 0.2 + 1M s500 (100) = 0.704.

   
C X (u3 , u0 ) = C X (0) −  X (u3 , u0 )

= 25 − 15 + 10M s500 (200) = 4.32. 
 25.0 0.838 0.838 4.0 1 0    X 1   4.32 
    
 0.838 1.2 0.704 0.279 0 1   Y1   2.11 
 0.838 0.704 1.2 0.279 0 1   Y2   2.11 
  = 
 4.0 0.279 0.279 1.2 0 1   Y3   1.30 
 1  
 0 0 0 0 0    X   1 
 0 0   Y   0 
 1 1 1 0

X1 = 1.0, Y1 = 1.248, Y2 = 1.248, Y3 = - 2.496,


and X = - 12.78, Y = - 0.41


x* (uo ) = 1 5 + 1.248 0.8 + 1.248 0.6 − 2.496  1.2 = 3.75,

ˆ E2 = 25 − 1  4.32 − 1.248  2.11 − 1.248  2.11 + 2.496  1.3


+ 12.78 = 31.83
Cokriging

Two variables Z1(x) and Z2(x) (such as porosity & acoustic impedance)
Use of Z1 and Z2 data to get a better interpolation of Z1

Z cok (x0 ) =   Z (x ) +   Z (x )
1i  N Z1
i 1 i
1 j  N Z 2
j 2 j

 cok
2
(x0 ) = VarZ (x0 ) − Z cok (x0 )
Thanks to the extra information coming from Z2 (x):

0   cok
2
(x0 )   k2 (x0 )
Cokriging (Doyen, 1988)

Kriged porosity

Porosity by regression from


1/(acoustic impedance)

Porosity by
cokriging
Conventional Estimation Techniques

Universal Kriging
Universal Kriging
  
Samples value X (u ) = m(u ) + R(u )
 
m(u ) =  i)
(
Local mean is
X u
neighborhood

Subtract local mean from individual samples, we obtain


  
R(u ) = X (u ) − m(u )
We define drift
 L

m(u ) =  al fl (u )
i =0

f0(u) =1 so that for a linear trend


 
m(u ) = a0 + a1u

And for a quadratic trend


  
m(u ) = a0 + a1u + a2u 2
Begin with the ordinary Kriging

 n

X * (u0 ) =  i X (ui )
i =1

Application of the unbiased condition requares that

n
 
  f (u ) = f (u )
i =1
i l i l 0 for l = 0, … , L

This imposes L+1 constraints


Applying the minimum variance,
       
 C (u1 , u1 )  C (u1 , un ) f 0 (u1 )  f L (u1 )  1   C (u1 , u0 )
        
    
         
C (un , u1 )  C (un , un ) f 0 (un )  f L (un ) n C (un , u0 )
      =   
 0 1)
f (u  f 0 (un ) 0  0   0   f 0 (u0 ) 
          
      
 L 1 )
f (u  f L (un ) 0  0    L   f L (u0 ) 
n+L+1 Matrix
The error variance is estimated by

  n
  L

ˆ = C (u0 , u0 ) −  iC (ui , u0 ) −  l fl (u0 )
2
E
i =1 l =0

Another alternative uses the method of external drift. The trend in the data is
Provided by the secondary variable Y.
 
m(uo ) = a0 + a1Y (u0 )
Stationarity

The residual should have a constant variance

A variable with A variable with


• no trend and • a trend and
• a residual with varying dispersion • a residual with varying dispersion
Stationarity
Example Of Stationary And Nonstationary
Surfaces
The Basic Assumption Of Geostatistics
A geological variable z(x) is composed of a systematic
trend and a random component.

z(x) is the realization of a random function Z(x)


equal to the sum of a trend m(x) and of a random
stationary residual R(x) of mean 0:

Z (x) = R (x) + m (x)

But how can we calculate statistics from just


one outcome?
The Stationarity Assumption
Mean and variance are independent
of location:

E Z ( x ) = E Z (x + h ) = m

VarZ (x ) = VarZ (x + h) =  2

Covariance between values at two


locations only depends on distance:

CovZ (x ), Z (x + h ) = C (h )
Field Example 4.8 – Universal Kriging To Estimate Top of Reservoir

– Location of reservoir top well data


Reservoir top running east/west Reservoir top running north/south.
Cross-validation estimates for reservoir top using universal,
simple, and ordinary kriging
Map of reservoir top generated with universal kriging

You might also like