Postdepositional faulting results in
shifted but similar reflector patterns
across the fault. Syndepositional faults
are somewhat more complicated. In some
cases, greater accommodation space results
in layers that are thicker on the hanging wall
side of the fault. In other cases, there is little
deposition or even erosion on the footwall
side of the fault, resulting in a totally different
suite of reflectors on the handing wall.
Whatever the style of faulting, manual
interpretation of faults on vertical seismic
sections is traditionally a laborious and
time-consuming process. Seismic attributes
such as coherence, variance and Sobel filters
accelerate the fault interpretation process,
in particular by allowing the interpreter to
name and assign color to faults on time
slices before picking them on vertical slices
oriented perpendicular to the fault strike,
resulting in a suite of fault segments or
“sticks” on a grid of lines, which are then
merged into fault surfaces. Fault-sensitive
attributes can be further subjected to image
enhancement methods such as swarm
intelligence (for example, “ant tracking”),
Radon transforms, and smoothing and
sharpening filters aligned parallel and
perpendicular to the fault attribute anomaly.
One such development is the fault likelihood
attribute, which we will describe in this article.
Preconditioning By Filtering
Any fault interpretation exercise begins
with preconditioning of the input seismic
amplitude volume by use of structure-oriented
median or other filtering. Many
articles have been published in the
Geophysical Corner column demonstrating
the advantages of doing so (see October
2014, January 2016 and February 2017,
for instance). Another process that is
particularly beneficial for fault interpretation
is to bring in the sharpening of the fault
segments by running Kuwahara filtering on
the input seismic data (see the August 2016
installment of Geophysical Corner). Seismic
data after such preconditioning are amenable
to more accurate fault interpretation as can
be gauged from the two displays shown in
figure 1. The fault likelihood attribute overlaid
on the displays is described below.
A fault surface can be described by its
strike, dip and the direction of the slip vector
as illustrated in figure 2. In 2003, Dave
Hale introduced the computation of fault likelihood, strike and dip from the 3-D seismic
amplitude volume. In this approach, the first
step is to compute a reflector dip-oriented
coherence attribute volume. Next, the fault likelihood is computed by scanning over
the negative and positive fault dip angles
(where no fault angle corresponds to vertical
faults), as well as strike. An examination of
the results after this step usually exhibits the
fault likelihood coinciding with the faults on
the seismic data. Finally, the fault likelihood
images are again examined, preserving only
their local maxima, and setting other values
to zero. This process thins the fault likelihood
image, and thereafter the fault surfaces can
be extracted from the 3-D images of fault likelihood by using quadrilateral meshes and
linking them.
Although this approach performs well for
single fault extractions, intersecting faults
pose a problem. Near the fault intersections,
which reflectors should be correlated for
the estimation of the fault slip becomes
tricky, producing holes in the resulting fault
surfaces.
One solution to filling these holes is to
filter the likelihood images in the inline and
crossline directions and then add the two
filtered results.
In figure 3 we show a set of time slice
images at t=1900 milliseconds through a
fault likelihood attribute volume computed
from 3-D seismic survey acquired offshore
New Zealand. Figure 3a shows a system
of polygonal faults. A corendered display
of seismic amplitude and fault likelihood
is shown in figure 3b, where the breaks in
the seismic amplitudes are seen to align
with the fault likelihood lineaments. Figure
4a shows the fault surfaces, where block
arrows indicate holes near fault intersections.
Figure 4b shows the same surfaces after the
fault likelihood has been filtered parallel to
the inline and crossline directions. The fault likelihood and fault surfaces can in turn be
corendered with the seismic volume for ease
of interpretation, or these volumes could
be utilized elsewhere for other applications
such as making geological modeling more
meaningful, as well as reservoir modeling and
simulation.
Conclusion
The fault likelihood attribute helps in
the manual interpretation of faults on
workstations, and it provides a useful
input for software designed for automatic
extraction of fault planes. With the desired
orientation of linear geologic features
enhanced with such an attribute, their
interpretations can be carried forward to the
next step in terms of their correlations with
production data.