Geoscience Reference
In-Depth Information
In many implementations there are further parameters to be adjusted to arrive at an
optimal sparse spike inversion product. For example, it may be possible to set con-
straints on the impedance values within each of the macrolayers, based on the well data
and geological knowledge of lateral variability. There may be a parameter that sets the
sparseness of the series, determining the degree to which the inverted product tries to
follow all the detail of the seismic traces. As discussed above, there is a wavelet scaling
parameter that affects the integration of reflectivity to yield impedance. This is often a
crucial parameter; in effect, it determines how much of the final inverted section comes
from the low-frequency model and how much is derived from the seismic traces them-
selves. In principle it can be determined from a well tie, but sometimes there are quite
large variations in scaling from one well to another, perhaps because of variation in ab-
sorption. A useful approach is to look at the results of various plausible choices on a trial
piece of seismic section, comparing the results with well data and geological knowledge
of lateral variability of different lithologies. This is an interpretive process, and the so-
lution will probably be appropriate only over a limited range horizontally and vertically.
Different stratigraphic levels may be best imaged on different inverted datasets.
Often, the most difficult part of the inversion process is the construction of the low-
frequency model. This depends on information other than the seismic trace data, very
often on well information. One approach is to make a model by interpolating well
impedance values within the macrolayers
(fig. 6.4)
. A low-frequency band-limited ver-
sion of this model is added to the band-limited product from the seismic trace inversion,
using the spectrum of the extracted wavelet to decide exactly what frequency range
should be taken from the model; this might typically be 0-6 Hz. Where there is good
Fig. 6.4
Low-frequency model from interpolated well data.
