Geological interpretation - 3-D Seismic Interpretation - page 113

rock properties (Trappe & Hellmich, 2000) . A different way of training a net is to
ask it to cluster input data, without specifying in advance what the clusters should
be; the net looks for similarities in pattern of the input data and groups the input into
classes accordingly. For example, the input data to a net might be seismic trace data,
over a window hung off a reference picked horizon and large enough to encompass
several loops. The net then clusters the data, looking at the shape of the traces rather
than their amplitude. A map is produced, coloured according to the cluster that a
trace belongs to. An example is shown in fig. 4.6 . Here the data have been split into
12 classes, with characteristic trace shapes as shown in the lower part of the figure.
All these traces begin at the maximum of a trough, because the window used for
data selection was hung off an autotracked trough. The map shows a prominent lin-
eation (arrowed) to the north-east of which the traces are quite different from those
elsewhere in the map. This lineation is inferred from well control to be the edge of
a major Tertiary fan system; within the fan, to the south-west of the lineation, there
are additional variations which are not well understood for lack of well calibration.

4.4 Some examples

In this section we describe examples of some of the techniques described in the previous
section. Stratal slicing is demonstrated by fig. 4.7 (Zeng et al. , 2001) . This comes from
the Miocene-Pliocene section of offshore Louisiana. From well data, it is known that
in this area sands are acoustically softer than shale. The polarity convention for these
slices is that red soft, so in general we expect red to correspond to sand and blue to

shale. Slices at different levels show various features. In (a), we see moderately sinuous,
channel-like features. Based on comparison with well penetrations, these are thought
to be fluvial channels in a coastal plain environment, with fining-upward channel fill.
A deeper stratal slice, (b), shows a very different channel type, with low sinuosity.
Well penetrations show blocky log patterns. These are incised valley fills that contain
deposits of lowstand and transgressive systems tracts. A deeper slice still, (c), shows
soft red amplitudes with lobate to digitate plan-view geometry, grading into low- to
variable-amplitude lobes. Wells in the channels (e.g. log 1) show an upward-fining
distributary channel overlying upward-coarsening, prograding delta deposits. Delta-
front deposits (log 2) contain thin interbedded sands and shales, while prodelta wells
encounter shales (log 3). The overall system is interpreted as a highstand shelf delta.
Another example is shown in fig. 4.8 , this time from offshore Egypt (Wescott &
Boucher, 2000) . These are submarine delta-front channel complexes, formed during
a late Miocene-earliest Pliocene transgression, and are well imaged on horizon slices
through a coherency volume. The deeper Rosetta channel complex is well defined
because it is incised into underlying anhydrite; it is characterised by sharp channel
edges and low sinuosity. These channels are interpreted to be sediment bypass conduits,

=