Geology Reference
In-Depth Information
data and well ties. Such approaches are relatively
straightforward to implement with modern software,
but given that the inversion results are constrained to
the seismic bandwidth the results may not be reliable
when the reservoir comprises thin beds.
A similar approach has been used by Michelena
et al.(
2009
) to estimate the probability of channels in
a fluvial system characterised by sands with low por-
osity and permeability and a large degree of vertical
and lateral compartmentalization. Following charac-
terisation of the well logs in terms of dominant facies
(e.g. channels (multi- and single-story) and flood-
plains (sandy and shaley)), a conditional probability
model was generated from upscaled P and
S impedance logs and applied to the inverted data.
The probability model is estimated from crossplots as
shown in
Fig. 10.3a
. Depth slices through the 3D
channel probability volumes show credible channel
patterns (
Fig. 10.3b
). The results also illustrate the
sensitivity to the choice of well data used in the
modelling, with a more optimistic result being
obtained when the wells chosen for the probability
model have a relatively high N:G (
Fig. 10.3b
).
In areas with enough well control, the thin-bed
limitations of deterministic inversions might be
addressed with geostatistical inversion techniques
(
Chapter 9
).
Figure 10.4
shows an example of sand
probability predicted from geostatistical impedance
realisations. In this case the probability at each sample
point is simply the proportion of the realisations that
have an impedance below a particular cut-off. These
realisations can also be used to investigate the poten-
tial connectivity of reservoirs. For example,
Fig. 10.5
illustrates three stochastic realisations with imped-
ance values below a threshold value and connected
to well penetrations.
Another approach to generating facies realisations
from seismic is the use of indicator simulation tech-
niques (e.g. Doyen et al.,
1994
, Perez et al.,
1997
).
Figure 10.6
shows an example where channel
and non-channel facies have been predicted on the
basis of:
(1) seismic amplitude distributions for the two facies,
(2) estimated proportions of channel sands and non-
channel shales,
(3) a spatial covariance model.
0.1
Shales
0
Brine sands
-0.1
-0.2
-0.3
Oil sands
-0.4
-0.5
-0.1
-0.05
0
0.05
0.1
0.15
0.2
R(0)
Figure 10.1
An AVO crossplot generated from well data and
based on a single interface AVO model, showing different facies with
iso-probability lines defining the probability density function for
each (from Avseth
et al
.,
2001
).
seismic. The application of these techniques should
give superior discrimination compared to AVO
projections. Avseth et al.(
2001
,
2003
) describe a
statistical AVO approach which uses a single inter-
face model for generating intercept and gradient
pairs for given boundary types (see
Fig. 5.51
).
A drawback to this type of modelling of course is
that the effects of reflector interference or of seismic
noise have not been accounted for. However, this
could be addressed by generating the interpretation
model from pseudo-wells in which seismic noise has
been included (e.g. Sams and Saussus,
2007
).
Given that deterministic inversion reduces the
effects of the wavelet and attempts to back out
absolute values of impedance, a popular approach to
the facies identification problem is the use of cross-
plot templates of
properties such
as AI vs EI (e.g. Mukerji et al.,
2001
), or angle-
independent properties such as AI and SI (Vernik
et al.,
2002
), AI and V
p
/V
s
(e.g.
'
impedance type
'
degaard and Avseth,
2004
; Lamont et al.,
2008
)or
(e.g. Goodway
et al.,
1997
). These templates can provide the basis for
probabilistic interpretation in a similar way to that
shown in
Fig. 10.1
.
Figure 10.2
illustrates an example
of the use of a Bayesian classification scheme to esti-
mate the probability of gas sands from pre-stack
simultaneous inversion results. Key issues to address
in this type of approach are the upscaling of the log
data to create the appropriate rock model for a given
target zone and the quality of the pre-stack seismic
λρ
vs
μρ
It is clear from the illustration that the connectivity of
channel sands is highly dependent on the covariance
range.
223
