Geology Reference
In-Depth Information
L (ms)
Estimated wavelet
PEP
NMSE (%)
(b/B)
(bT)
Comment
60
0.5
3.5
0.85
28.7
Oversmoothed
100
0.6
3.8
0.63
17.4
Oversmoothed
140
0.65
4.3
0.50
12.5
About right
200
0.69
5.1
0.38
8.8
Undersmoothed?
248
0.71
5.7
0.32
7.2
Undersmoothed?
300
0.73
6.3
0.27
6.0
Too noisy
Figure 4.12
Extracted wavelets from marine seismic with varying wavelet length (L) from a 500 ms time segment. A wavelet of about 140 ms
is probably the best trade-off in this case between noise and oversmoothing (R. E. White, personal communication).
(in deg/Hz) × (1000/360), with negative values corres-
ponding to a time delay and positive values to a time
advance. In the high-bandwidth example in
Fig. 4.11a
there is little scope to vary the fit and the wavelet phase
is constrained to within about 10°, whereas in the low-
bandwidth example in
Fig. 4.11b
a wider range of fits
are possible and there is around 40° of phase ambiguity.
this is that the migration applied to the seismic data
may not have positioned the data correctly, particu-
larly in the presence of steep dips and significant
lateral velocity variation. There are two separate
possible causes of error. One of them is the use of
time migration rather than depth migration. Time
migration assumes that the overburden velocity is
laterally invariant on a length scale similar to the
migration aperture. This effectively means that in
the presence of dip the best match location is usu-
ally up-dip of the well with time-migrated data. The
other problem is that the velocity model used to
perform the migration is never perfectly deter-
mined. The positioning error is likely to be greater
as target depth and reflector dips increase; shifts of
100 m are quite possible. It is also possible that the
positioning is wrong, for example in old wells where
the deviation surveys are inaccurate, or with seismic
that has been loaded with the wrong coordinate
system.
An example of a best match location being differ-
ent from the well location is shown in
Fig. 4.13
. The
PEP map (
Fig 4.13a
) shows that the best match (red
area) is up-dip of the well, and at this location the
time delay map (
Fig. 4.13c
) shows zero delay. In
deciding whether to accept such a shift, it is important
4.4.1.3 How long is a wavelet?
The length of a seismic wavelet is a compromise
between goodness of fit (as expressed by the PEP or
the cross-correlation) and the estimated phase accur-
acy. An example of wavelets extracted from a 500 ms
time segment of seismic data is shown in
Fig. 4.12
.It
is evident that the long wavelets have been over-fitted
to the data with the incorporation of noise, whereas
the short wavelets have been oversmoothed. The
choice of a wavelet length of 140 ms in this case is
based on a number of factors including visual appre-
ciation and a compromise between the
cross-
correlation and phase error.
4.4.1.4 Best match location
It is usual to run the matching technique on a cube
of seismic around the well location rather than just
on the single trace at the location. The reason for
46
