Geoscience Reference
In-Depth Information
These will be discussed further in the last section
of this chapter, but are introduced below.
simple consequence of the limits apparent from
the (statistically insufficient) well data set.
2.5.3.1 Faulting
With the exception of some (relatively) spe-
cialist structural modelling packages, large
scale structural features are strongly determin-
istic in a reservoir model. Thought is required
as to whether the structural framework is to be
geophysically or geologically led, that is, are
only features resolvable on seismic to be
included, or will features be included which are
kinematically likely to occur in terms of struc-
tural rock deformation. This in itself is a model
design choice, introduced in the discussion on
model frameworks (Sect.
2.3
) and the choice
will be imposed deterministically.
2.5.3.5 Seismic Conditioning
The great hope for detailed deterministic control
is exceptionally good seismic data. This hope is
often forlorn, as even good quality seismic data is
not generally resolved at the level of detail
required for a reservoir model. All is not lost,
however, and it is useful to distinguish between
hard
and
soft
conditioning.
Hard conditioning
is applicable in cases where
extremely high quality seismic, sufficiently
resolved at the scale of interest, can be used to
directly define the architecture in a reservoir
model. An example of this is seismic geobodies in
cases where the geobodies are believed to directly
represent important model elements. Some good
examples of this have emerged from deepwater
clastic environments, but in many of these cases
detailed investigation (or more drilling) ends up
showing that reservoir pay extends sub-seismically,
or that the geobody is itself a composite feature.
The more generally useful approach for rock
modelling is
soft conditioning
, where information
from seismic is used to give a general guide to the
probabilistic algorithms (Fig.
2.20
). In this case,
the link between the input from seismic and the
probabilistic algorithm may be as simple as a
correlation coefficient. It is the level of the coeffi-
cient which is now the deterministic control; and
the decision to use seismic as either a hard or soft
conditioning tool is also a deterministic one.
One way of viewing the role of seismic in reser-
voir modelling is to adapt the frequency/amplitude
plot familiar from geophysics (Fig.
2.21
). These
plots are used to show the frequency content of
a seismic data set and typically how improved
seismic acquisition and processing can extend
the frequency content towards the ends of the
spectrum. Fine scale reservoir detail, often sits
beyond the range of the seismic data (extending
the blue area in Fig.
2.21
). The low end of the
frequency spectrum - the large scale layering - is
also typically beyond the range of the seismic
sample, hence the requirement to construct a
low frequency 'earth model' to support seismic
inversion work.
2.5.3.2 Correlation and Layering
The correlation framework (Sect.
2.3
) is deter-
ministic, as is any imposed hierarchy. The prob-
abilistic algorithms work entirely within this
framework - layer boundaries are not moved in
common software packages. Ultimately the
flowlines in any simulation will be influenced
by the fine layering scheme and this is all set
deterministically.
2.5.3.3 Choice of Algorithm
There are no hard rules as to which
geostatistical algorithm gives the 'correct'
result yet the choice of pixel-based or object-
based modelling approaches will have a pro-
found effect in the model outcome (Sect.
2.7
).
The best solution is the algorithm or combina-
tion of algorithms which most closely reflects
the desired reservoir concept, and this is a
deterministic choice.
2.5.3.4 Boundary Conditions
for Probabilistic Algorithms
All algorithms work within limits, which will be
given by arbitrary default values unless imposed.
These limits include correlation models,
object dimensions and statistical success criteria
(Sect.
2.6
). In the context of the concept-driven
logic described above these limits need to be
deterministically chosen, rather than left as a
