Long gone are the
days when faults appeared only as steps on vertical seismic sections.
If we use today's better data and exploit modern workstation tools available,
we should do a much better job of recognizing and understanding faults
in 3-D seismic data.
Faults cause
breaks in continuity of seismic horizons. These discontinuities generate
diffraction patterns and, before the days of seismic migration, diffraction
patterns were what the seismic interpreter sought as an indication of
faulting.
Migration in
2-D will collapse diffractions to some extent, whereas migration in 3-D
should do much better. Major faults are still recognized on vertical sections
and their throw estimated by offset in character correlation. For this,
double-gradational color is the best mode of display.
Spatial patterns
of faulting are revealed on time slices (or depth slices). These horizontal
sections must be used in conjunction with vertical sections to establish
sensible fault geometries.
Composite and
chair displays are established ways of combining these orthogonal sections
together. In a chair display, one looks at a horizontal slice where it
intersects a vertical section. You are able to see a fault's map pattern
along with its offset in a cross-section view. Various other kinds of
volumetric display also help to study and visualize faults.
Much of the science
of fault detection concerns the recognition of subtle faults. On a normal
vertical section a single-gradational color scheme, such as gradational
gray (Figure 1), is usually best, as this
type of display enhances the terminations of low amplitude events.
The detection
of subtle faults, however, is highly dependent on good data quality and
high signal-to-noise ratio. Some extra care and attention in data collection
and processing is always beneficial.
Coherence is
an invention of five years ago that has had a beneficial impact on fault
recognition. The coherence transformation suppresses the continuity of
seismic reflections and emphasizes discontinuities such as faults.
Coherence data
is best viewed as time slices or as a whole coherence cube. In good quality
data, faults can be strikingly evident and spatial patterns of faulting
can be clearly discerned.
Figure
2 is a time slice through a salt dome showing the common pattern of
radial faults. The upper half of the time slice is in coherence and the
lower half is the normal time slice display in amplitude. Note how the
faults are more clearly visible in coherence.
Coherence is
of less benefit in poor data and can sometimes be quite ambiguous. Different
algorithms from various vendors can give different results. Certain versions
are designed to overcome particular data problems such as high dip and
poor signal-to-noise ratio.
Once the major
faults have been recognized and the tectonic framework established, machine
autotracking should be used to complete horizon surfaces. The autotracker
follows the crest of an identified peak or trough with very high precision.
The resultant time (or depth) values contain information on subtle faults
-- but also noise.
An important
part of fault recognition is then scrutinizing these horizon surfaces
in an attempt to distinguish geology from noise.
Time-derived
horizon attributes are used for this purpose. Several of these are available
on modern workstations and the most important are dip, azimuth, edge and
residual.
Figures
3a and 3b show dip and residual for the same horizon. Dip is the magnitude
of dip of the local surface dip vector. Residual is the difference between
the horizon surface and its spatially-smoothed equivalent. We look at
these attributes in map form and judge what appears geologic.
In Figure
3 combination, pairs of brown anomalies on the left (3a) correspond
to blue anomalies on the right (3b). The strength of the anomalies over
background noise and the arcuate pattern support the interpretation of
fault grabens.
The edge map
of Figure 4 clearly distinguishes the short
north-south faults from the long arcuate one. The arcuate fault is about
seven kilometers long and looks impressive on the edge map -- however,
it has negligible throw and is barely visible on any vertical seismic
section. It was first recognized on this edge display and appears to be
caused by an igneous intrusion.
As shown, the
use of time-derived attributes can be a primary method of fault recognition.
We do not have to observe a clear break on a vertical section. However,
the interpreter typically looks at an appropriately oriented vertical
section and may see a minor interruption at the anomaly position. Commonly
this was not recognized during the mainstream of the interpretation.
Distinguishing
subtle faults from various kinds of noise is always a value judgement,
so experience is useful. Interpreters tend to look at more than one type
of time-derived attribute and seek the same feature on each as cross-validation.
The two panels
of Figure 3 show the grabens on both the dip
and residual displays; some of the minor wiggly features, probably noise,
occur on only one.
Three-dimensional
seismic data today typically contains an enormous amount of geologic detail.
Faults are clearly an important part of this information.
The modern interpreter
must use all the interpretation tools available to find and understand
the faults affecting the reservoir. With practice and experience, one
can extract the subtle but valuable details inherent in the data.