Inversion - 3-D Seismic Interpretation - page 160

the variation of layer parameters from one seismic trace to the next along a seismic
line.
(5) Perhaps, add low-frequency information obtained from a model based on geological
data.
The additional information that has been taken into account is thus:
(a) the wavelet: removal of its effects is equivalent to a deconvolution, extending band-
width at the high-frequency end;
(b) the model: the geological input extends bandwidth at the low-frequency end.
To make all this more concrete, it is helpful to work through an actual processing flow.
A common approach to building a subsurface model is to split it up into macrolayers,
probably several hundreds of ms thick, bounded by the main semi-regional seismic
markers, and consisting of a single broad lithology. This makes it easier to construct
a geological model to constrain impedance variation within a macrolayer. Inside the
macrolayer, the subsurface is represented by means of a series of reflectivity spikes.
These spikes, when convolved with the wavelet, should reproduce the observed seismic
trace, and integration of the reflectivity will give the impedance variation within the
macrolayer. To prevent the software from trying to reproduce all the noise present in
the seismic section, it is usual to impose a requirement that the reflectivity spike series
is as simple as possible, either in the sense of using a small total number of spikes or
using spikes of small total absolute amplitude; the algorithm will trade off the misfit
between real and calculated seismic section against the complexity of the spike model,
usually under user control of the acceptable degree of misfit. The result is usually called
a 'sparse spike' representation of the subsurface.
It is obviously critical to the success of this process that we know the wavelet ac-
curately. This is usually obtained from a well-tie study. As discussed in chapter 3 ,
well synthetic or VSP information will tell us how zero-phase seismic ought to look
across the well; comparison with the real data tells us what wavelet is present. Figure 6.3
shows an example display from such a study. To the left, in red, is the candidate wavelet,
which in this case is close to being symmetrical zero-phase; to the right is a panel of six
(identical) traces showing the result of convolving this wavelet with the well reflectivity
sequence derived from log data. They should be compared with the panel of traces in the
middle of the figure, which are the real seismic traces around the well location. Various
geologically significant markers are also shown. There is clearly a very good match
between the real and synthetic data so far as the principal reflections are concerned,
so it is possible to have confidence that the wavelet shown is indeed that present in
the data. There are also some differences in detail, which will limit the accuracy of an
inversion result. These may arise from imperfections in the seismic processing, perhaps
the presence of residual multiples or minor imaging problems. With luck, the inversion
process will leave some of this low-energy noise out of the inverted image, because of
the sparseness of the spike reflectivity series. Another possible source of mismatch is
AVO, which will be discussed later in the chapter.