Before
seismic interpretation workstations, interpreters marked paper sections
with each horizon, then laboriously read off the time of the reflection
— usually to no better than ±5 ms — and at intervals of perhaps
every tenth trace.
Computers,
however, remember exactly where the interpreter picks — and often
pick more accurately.
Autotracking
on lines in either 2-D or 3-D data, and through a volume in 3-D,
picks with precision on every trace, so machine horizon tracking
today can reveal geology in greater detail than manual interpretation
can ever achieve.
The first
decision is what seismic event to pick.
Figure
1 shows a geological marker in two wells; a continuous reflection
approximately ties the two markers, but in one well the marker is
on a positive-to-negative zero crossing, while it is close to a
peak in the second well. A synthetic seismogram might help, but
that is beyond the scope of this article. Often the interpreter
can simply decide on a continuous event to pick close to the marker
to be mapped.
Most reflections
are composite; the perfect phase point to pick is uncertain. All
interpretation systems can "snap" to, or follow, maximum negative,
maximum positive or zero crossings (going negative with increasing
time, and going positive).
In Figure
1 the reflection is picked on the peak of an event, and on the
positive-to-negative zero crossing. Each horizon is an automatic
track along a line from a single seed point on one trace.
Any interpretation
system has parameters that can be set to specify the way in which
the correlation from trace to trace is done. These may include:
- Maximum
time (or depth) change from one trace to the next.
- Whether
to pick the largest event within this limit, or the closest.
- Maximum
amplitude change from one trace to the next.
- Whether
to follow an event, or to cross-correlate.
If the
parameters are too restrictive, the tracking leaves gaps. If the
parameters are too loose, it makes mistakes.
For 3-D
data, automatic tracking from one trace to the next can be extended
throughout the data volume. Figure 2
shows the result of automatic picking of the peak. Where the event
mapped becomes less obvious, the automatic picking breaks down.
The conventional
seismic data may not give the most accurate picture of geological
structure when you pick on a constant phase point (peak, trough
or zero crossing), especially if the reflection is weak. Seismic
attributes allow removing the amplitude information to make all
reflections the same. The instantaneous phase attribute does this,
but is discontinuous where the phase passes 180 degrees (it jumps
to -180 degrees).
A more
elegant attribute is cosine of instantaneous phase, which is -1.0
for both 180 degrees and -180 degrees (figure
3). Notice there are differences of up to 1.1 meters (1.5 meters,
or five feet — a significant depth error in many prospects) between
the two tracked horizons — green (picked on conventional data)
and red (picked on cosine-of-phase).
Workstations
interpolate between samples using a spline function, so the peak
or zero-crossing is picked with a much greater precision than the
sample interval (two meters in this data).
This difference
between the two horizons is real. Automatic tracking is at least
an order of magnitude more accurate than manual picking; such a
small time difference would never be detected with manual timing.
Along with
the structure map, we can get a reflection amplitude map, either
trace amplitude or the reflection magnitude. Reflection amplitude
measurements are impractical with manual picking.
With the
cosine-of-phase data volume, details differ (compare figure
4 and figure 2). Either of these
maps show an excellent picture of the structure in most places,
and we have achieved it by manually identifying the event on one
trace only.
The results
vary with which point on the reflection is used for picking. In
this example, automatic picking fails over a larger area if the
positive-to-negative zero crossing is used on the conventional data
(figure 5). There are fewer gaps with
cosine-of-phase (figure 6), but obvious
errors in several areas.
For 3-D,
an alternative to automatic tracking is to pick manually a subset
of the data, then interpolate. This is particularly attractive if
automatic tracking is unreliable because the event is weak, or if
the horizon is extensively faulted.
In this
case the steps to producing both a structure map and an amplitude
map are:
- Manually pick a subset
of the data, such as every tenth line and crossline, by displaying
the lines on the screen and picking exactly where you want the
horizon, using automatic tracking along the line, or point-to-point
interpretation.
- Interpolate between
the picked lines.
- Snap to a peak, trough
or zero crossing.
- Smooth if needed.
- Map and extract amplitudes.
This alternative
technique can be used over the whole of a 3-D survey, or only over
parts where automatic picking is unreliable, and the results merged
with automatic picking in areas where that technique is reliable.
Automatic
tracking of seismic horizons in good quality data from a few seed
points is a powerful tool for rapidly completing interpretation
of 3-D seismic data volumes. It reveals geology with much greater
precision and detail than manual interpretation can.
However,
there are pitfalls in its use; it is less than reliable unless the
interpreter understands the geology and restricts the automatic
picking by using parameters chosen to minimize mis-picking.
Interpolation
is often a good alternative for 3-D if automatic tracking will not
work reliably.