Geoscience Reference
In-Depth Information
Fig. 3.18
Autotrack (in blue) of a low-amplitude horizon with no consistent character.
the window, tries to find a match for the trace shape by cross-correlating traces; this is
computationally more demanding but will give better results on low-amplitude events.
The basic idea can obviously be extended to pick a negative maximum or a zero-
crossing, or to take account of a pre-defined horizon dip when moving the picking
window from one trace to the next. A more interesting extension is to apply the method
across the entire 3-D volume. As soon as we attempt to do this, the algorithm needs
to have some way of checking consistency of picks on a particular trace arrived at by
different routes. This is analogous to the tying of loops in manual interpretation; if we
start at a particular trace and pick a horizon until we reach another trace on a different
inline and crossline, we want to get the same answer independent of the path through
the data. Such a
volume autotracker
needs to check continuously for consistency as it
picks away from an initial manually interpreted point, called a
seed point
. Differences
in the way that the path through the dataset is controlled mean that different autotrackers
will give different results on the same dataset in areas where the pick is difficult. In
practice it is not possible to interpret a horizon across an entire 3-D dataset starting
from an initial pick on just one trace. It is more realistic to use the autotracker as a
sophisticated interpolation device, starting from a manual interpretation of a coarse grid
of lines. Faults can be a particular problem. The autotracker may be able to find its own
way correctly across minor faults, but there are often changes in reflection character
across major faults, perhaps because of genuine changes in the geology. In such a case,
it may be best to divide the area into panels between the major faults and autotrack
each panel separately using its own seed picks. In any case, it is essential to review
the results produced by the autotracker and to be prepared for several iterations of the
process. The first pass will probably pick the simple areas correctly, but some areas will
have to be erased and given an increased density of seed data until a satisfactory pick is
achieved. Some software packages allow this erasure-reseeding-retracking process to
