Vous êtes sur la page 1sur 3

Seismic AVO: Modeling and Inversion

WednesdayMorning, September26th

AVO Inversion:Theory and Practice so1.1


Daniel Hampton* and Brian Russell,Hampson-Russell
SoftwareServicesLtd.. Canada

interpreted as bright-spot anomalies.


SUMMARY Indeed, both anomalies were drilled.
However, only the line shown in Figure 1
The process of AVO inversion is an attempt actually contained a gas-filled Colony sand.
to extract quantitative lithologic The line of Figure 2, when drilled, was
information from conventional P-wave found to contain a coal at the Colony level.
reflection data. The basic assumption is
that the amplitude variation due to Obviously, Figures 1 and 2 tell us that a
reflection at a range of incident angles can Colony coal cannot be distinguished from a
be used to deduce changes in Poisson's ratio Colony gas sand using conventional bright-
across the interface (Shuey, 1985). To spot techniques with the stacked seismic
date, most analyses have been qualitative, data.
in that an increase or decrease in amplitude
with offset is correlated with the presence
or lack of gas on a sand (Ostrander, 1994). However, we do know that the Poisson's ratio
In the quantitative inversion process, a for a coal should not vary significantly
lithologic model is derived which has the from the surrounding formations, whereas the
property that the synthetic response Poisson's ratio for a gas sand should
calculated from it matches the real data, display a negative change. We should
within some tolerance level. therefore be able to use forward modelling
and inversion to distinguish between the
In this paper, two methods for deriving the coal and the gas sand on the basis of the
quantitative model are analyzed. The first observed variation of amplitude with offset.
method is a generalized linear inversion
approach in which layer thicknesses and rock LINEARIZED INVERSION
properties are perturbed in the vicinity of
some initial guess until an optimum fit to In the linearized inversion approach, a
the observed data is obtained. This method lithologic model is derived by perturbing an
suffers from the problem that the final initial guess model. It is assumed that the
result may depend critically on the initial earth is one-dimensional with a set of
guess. The second method is a Monte Carlo layers whose number and approximate
approach in which model values are generated properties are derived from well-log
randomly, and the correlation function information. Typically, a small number of
calculated. This method suffers from the layers (less than 10) is used to model a
problem that one is never sure if enough short time sequence. The layers are
models have been calculated, but offers the described by the following 4 parameters:
possibility of escaping from the reliance on
the initial guess. R = zero-offset reflection
coefficient at the interface
Both of these methods have been applied to a above the layer
case study from Alberta in which the
objective is to differentiate between a gas To = zero-offset two-way travel-
sand and a coal seam which produce similar time to the interface above
bright-spot responses. the layer

GEOLOGYOF STUDY AREA Ar = change in Poissonls ratio at


interface above the layer
The Cretaceous Colony sand of southern
Alberta is an excellent reservoir for Ap = residual (parabolic) moveoc
hydrocarbons, containing high porosity and on reflection from interfac
often displaying good stratigraphic trapping
conditions. The seismic data also displays Note that the forward model must reproduce
classic bright-spot anomalies, which are not only the amplitudes of the observed
indicative of gas sands. Figures 1 and 2 events, but their times as well. The
show seismic sections with what could be

1450
2 AVO Inversion - Theory and Practice

inversion in particularly sensitive to the INVERSION RESULTS


exact travel-time at the far-offset traces,
and the fourth parameter allows for the The results of applying linearized inversion
possibility of an error in that travel-time are shown in Figures 3 and 4. Both of these
which is second order in offset, figures display the well logs on the left
(in the order: unblocked P-wave, blocked P-
The inversion consists of minimizing the wave, density, S-wave, and blocked Poisson's
objective function of the form: ratio), the forward model in the centre, and
the real data commonoffset stack on the
TRACEERROR2 IMPEDANCEERROR' right. Figure 3 shows the results for the
J = + gas sand. Notice that a significant drop in
UT 01 Poisson's ratio was needed to produce a
match between the synthetic and the real
POISSONRATIO ERROR' data.
+
UR This drop was from a background level of 0.4
down to 0.16. The acoustic impedance
where: decreases at the top of the sand and
increases at the base. This produces an
TRACE ERROR increase in absolute amplitude at both top
and bottom of the layer. One disturbing
= error in the forward model part of the fit was that the event at 660
msec on the model did not appear on the real
IMPEDANCEERROR data. This is probably due to a spurious
spike on the sonic log.
r deviation between final
impedance result and the The same sequence of panels is shown for the
initial guess coal model in Figure 4. Notice that an
excellent fit was obtained over the complete
POISSONRATIO ERROR time zone. In this case, the best fit
Poisson's ratio was one that showed a slight
= deviation between the final increase over the background value. The
Poisson ratio and the initial amplitudes do not display a significant
guess increase or decrease with offset.
0 = expected error of each type The Monte Carlo analysis tended to Confirm
these general results, but showed that a
The equation is solved by linearizing it in significant range in the combination of
the vicinity of the initial guess, using Poisson#s ratio and gas sand thickness was
Shuey#s approximation for the Zoeppritz possible. This confirms the observation
equations (Shuey, 1985). that the non-uniqueness is inherent in the
trade-off between time and amplitude.
MONTECARLOANALYSIS
In summary, the two AVO inversion techniques
A major limitation of the linearized have given us a tool with which to
inversion approach is the dependance on the distinguish true from false "bright-spots"
initial guess. This is a particular problem in the Colony play, and it is hoped that
in the AVO case because of the this can be applied in other areas as well.
interdependance between the amplitude and
time parameters, and because of the multi- REFERENCES
modal nature of the objective function. The
Monte Carlo technique and its extension to Ostrander, N.J., 1984, Plane-wave reflection
simulated annealing (Rothman, 1985) offers coefficients for gas sands at nonnormal
the possibility of overcoming this angles of incidence: Geophysics, 49, 1637-
limitation. In these methods, trial 1648.
solutions are generated randomly, and
accepted if they are improvements over the Rothman, D.H., 1985, Nonlinear inversion,
previous solution (and in some cases if they statistical mechanics, and residual statics
are not). These methods are particularly estimation: Geophysics, 50, 2784-2796.
attractive in the AVO problem because it is
such a small problem and a large number of Shuey, R.T., 1985, A simplification of the
solutions may be generated very rapidly. In Zoeppritz equations: Geophysics, 50, 609-
addition to searching out the global 614.
optimum, information is stored concerning
&Q trial solutions and subsequent plots
show the variation in the objective function
as various combinations of parameters are
modified. This allows the investigator to
visually assess the impact of the non-
uniqueness problem.
1457
AVO Inversion - Theory and Practice 3

Vous aimerez peut-être aussi