An overview of Hampson-Russells new Joint Inversion Program
Dan Hampson Brian Russell Keith Hirsche
10 August 2005 Theory 3-1
Objective
Objective of Joint Inversion:
To analyze pre-stack CDP gathers and invert for Zp, Zs, and (optionally) Density ().
10 August 2005
Theory 3-2
Introduction
Our current practice is to invert separately for Zp, Zs, and . An example of this procedure is LMR analysis:
Gathers AVO Analysis
RP Estimate RS Estimate Invert to ZP Invert to ZS
Transform to and Cross-plot
10 August 2005 Theory 3-3
Introduction
The problem with this approach is that it ignores the fact that Zp and Zs should be related. For example, we expect that from Castagnas equation, Vp and Vs should be more or less linearly related, with variations precisely where there are hydrocarbons. Similarly, should be related to Vp by some form of generalized Gardners equation.
ARCOs original mudrock derivation (Castagna et al, Geophysics, 1985)
10 August 2005 Theory 3-4
Introduction
The objective of joint inversion is to include some form of coupling between the variables. This should add stability to a problem that is ill-conditioned: - very sensitive to noise - very non-unique. A second objective is to create a joint inversion which is consistent with Strata for the case of zero-offset.
10 August 2005
Theory 3-5
Joint Inversion Theory
We start with the modification of Aki-Richards equation as per Fatti et al:
RPP ( ) = c1RP + c2 RS + c3 RD
where:
c1 = 1 + tan 2 c2 = 8 2 sin 2 1 c3 = tan 2 + 2 2 sin 2 2 V = S VP
10 August 2005
RP =
1 VP V + 2 P
1 VS + RS = 2 VS RD = .
Theory 3-6
Joint Inversion Theory
To simplify this theory, it is common practice to use the small reflectivity approximation. For example, the exact equation for Rp is:
RP (i ) =
But, if we define: we can show that:
Z P (i + 1) Z P (i ) Z P (i + 1) + Z P (i )
LP = ln( Z P ) (natural logarithm)
RP (i ) 1 2 [ LP (i + 1) LP (i )]
Similarly:
LS = log( Z S ) LD = log( )
10 August 2005
RS (i ) 1 2 [ LS (i + 1) LS (i )]
RD (i ) LD (i + 1) LD (i )
Theory 3-7
Joint Inversion Theory
In matrix notation for the P-wave reflectivity this is:
R P = (1 2 ) D L p
or:
RP (1) R (2) P =1 M RP ( N )
1 1 0 L LP (1) 0 1 1 0 L (2) P 2 0 0 1 1 M 0 0 0 L LP ( N )
10 August 2005
Theory 3-8
Joint Inversion Theory
We add the effect of the wavelet by defining the wavelet matrix:
T = WRp
T (1) T (2) =1 M T ( N ) 0 W1 0 W W 0 1 2 2 W3 W2 W1 0 W3 W2 L 1 1 0 L LP (1) L 0 1 1 0 LP (2) L 0 0 1 1 M L 0 0 0 L LP ( N )
Finally, Fattis equation looks like:
T ( ) = (1 2) c1W ( ) DLP + (1 2)c2W ( ) DLS + c3W ( ) DLD
Note that the wavelet can be different for each angle.
10 August 2005 Theory 3-9
Joint Inversion Theory
Now we want to make use of the fact that the resulting Zs and should be related to Zp. We use two relationships which should hold for the background wet trend:
VS VP = = constant ln( Z S ) = ln( Z P ) + ln( )
and:
Constant
= aVPb
ln( a ) b ln( Z P ) + ln( ) = 1+ b 1+ b
10 August 2005
Generalized Gardner
Theory 3-10
Joint Inversion Theory
More generally, we assume the following relationships for the background trend:
ln( Z S ) = k ln( Z P ) + kc + LS ln( ) = m ln( Z P ) + mc + LD
This assumes that the major trend is linear and that the outliers are the hydrocarbons:
Ln(Zs)
LS
10 August 2005
Ln(Zp)
Theory 3-11
Joint Inversion Theory
More generally, we assume the following relationships for the background trend:
ln( Z S ) = k ln( Z P ) + kc + LS ln( ) = m ln( Z P ) + mc + LD
Ln()
Ln(Zs)
LD
LS
10 August 2005
Ln(Zp)
Ln(Zp)
Theory 3-12
Joint inversion theory
This changes Fattis equation to:
% % T ( ) = c1W ( ) DLP + c2W ( ) DLS + c3W ( ) DLD
where:
% c1 = (1 2) c1 + (1 2) kc2 + mc3 % c2 = (1 2) c2
Finally, assume we have a series of traces at various angles. We concatenate the traces into a single vector to get the system:
% % c2 (1 )W (1 ) D c3 (1 )W (1 ) D T (1 ) c1 (1 )W (1 ) D T ( ) c ( )W ( ) D c ( )W ( ) D c ( )W ( ) D LP % %2 2 1 2 2 3 2 2 1 2 LS = M M M M % LD % T ( N ) c1 ( N )W ( N ) D c2 ( N )W ( N ) D c3 ( N )W ( N ) D
10 August 2005 Theory 3-13
Joint inversion theory
The algorithm looks like this: (1) Given the following information: - A set of N angle traces. - A set of N wavelets, one for each angle. - Initial model values for Zp, Zs, and . (2) Calculate optimal values for k and m using the actual input logs. (3) Set up the initial guess:
[ LP
LS
LD ] = [log( Z P ) 0 0]
T
(4) Solve the system of equations by conjugate gradients. (5) Calculate the final values of Zp, Zs, and :
Z P = exp( LP ) Z S = exp(kLP + kc + LS )
10 August 2005
= exp(mLP + mc + LD )
Theory 3-14
Synthetic and real data tests
We now show 2 tests of the joint inversion algorithm: (1) A synthetic data set, showing variations in fluid content from pure gas to pure brine. (2) A real data set from Western Canada.
10 August 2005
Theory 3-15
Gas/Wet Synthetic Tests
Vp
We produced a series of synthetic gathers corresponding to varying fluid effects:
Vs
Target Zone
100% Gas
100% Wet
10 August 2005
Theory 3-16
The result at the GAS location Initial guess: Zp Zs
Model
Input
Error
After 50 iterations:
10 August 2005
Theory 3-17
The result at the WET location Initial guess: Zp Zs
Model
Input
Error
After 50 iterations:
10 August 2005
Theory 3-18
Zp
Zs
100% Gas
10% Gas
0% Gas
10 August 2005
Theory 3-19
The synthetic test on a range of CDP gathers
18,000
Original offset gathers
0o
90o
Transformed to angle
10 August 2005
Theory 3-20
The synthetic test on a range of CDP gathers
Zp
Zs
10 August 2005 Theory 3-21
Zp
Vp/Vs
10 August 2005
Theory 3-22
Input gathers
The synthetic test on a range of CDP gathers
Synthetic gathers
Error
10 August 2005
Theory 3-23
Real Data Test Colony
This test applies the simultaneous inversion algorithm to the Colony data set from Western Canada:
10 August 2005
Theory 3-24
Real Data Test Colony
Transform to angle gathers:
10 August 2005
Theory 3-25
Real Data Test Colony
Using the known well, create cross plots to determine the optimum coefficients:
10 August 2005
Theory 3-26
Real Data Test Colony
Zp
Zs
10 August 2005
Theory 3-27
Real Data Test Colony
Zp
Vp/Vs
10 August 2005
Theory 3-28
Real Data Test Colony
Input gathers:
Synthetic data from inversion:
10 August 2005
Theory 3-29
Real Data Test Colony
Input gathers:
Synthetic error from inversion:
10 August 2005
Theory 3-30
Comparison between real logs and inversion result at well location
Zp Vp/Vs
10 August 2005
Theory 3-31
Cross plotting Vp/Vs against Zp using the log curves:
This zone should correspond to gas:
10 August 2005
Theory 3-32
Zp
Vp/Vs
Gas Zone from log cross plot
10 August 2005 Theory 3-33
Extension to PS data
Similarly to the Fatti equation, we can write down a linearized expression for the PS reflectivity (Stewart, 1990; Larson, 1999):
RPS ( , ) = c4 RS + c5 RD where : c4 = tan
4 sin 2 4 cos cos ,
tan 1 + 2sin 2 2 cos cos , c5 = 2 and :
= sin 1 ( sin ) .
10 August 2005
Theory 3-34
Extension to PS data
Combining the expression for the PS reflectivity with the relationships given earlier, we get:
% TPS ( ) = c4W ( ) DLP + ( c4 2 )W ( ) DLS + c5W ( ) DLD , % where : c4 = k ( c4 2 ) + mc5 .
Note that this is exactly the same form as the original equation for TPP:
$ $ $ TPP ( ) = c1W ( ) DLP + c2W ( ) DLS + c3W ( ) DLD
This means that we can (theoretically) handle any combination of PP and PS traces, at any number of angles.
10 August 2005
Theory 3-35
Joint Inversion Assumptions
(1) Aki-Richards 3-term equation as modified by Fatti. (2) Small reflectivity assumption. (3) Linear relationship between ln(Zs), ln(Zp) and ln() is reasonable. (4) A constant value of = Vs/Vp used in Fatti coefficients. (5) NMO-stretch can be handled by angle-dependent wavelets.
10 August 2005
Theory 3-36
