Page 350 - Acquisition and Processing of Marine Seismic Data

6.5 DETERMINATION OF DECONVOLUTION PARAMETERS 341

amplitude spectrum and suppresses the ampli- the recorded seismogram, which represent the
tudes in the autocorrelograms is indicated by general characteristics of the wavelet’s autocor-
(*)in Fig. 6.26. relation (Fig. 6.11). The deconvolution operator
length can be determined from the autocorrela-
tion of the input seismogram, and must be cho-
6.5.3 Operator Length
sen as close as possible to the time length of the
The time span over which the autocorrelation first isolated amplitude package of the autocor-
of the input seismic trace approximately equals relogram (Fig. 6.29A). In practice, it may not be
the wavelet’s autocorrelation is defined as the easy to determine the proper operator length by
deconvolution operator length. In practice, the analyzing only one trace from one shot, and
operator length is one of the most crucial param- therefore it is a general approach to include sev-
eters of the deconvolution application and must eral traces from successive shots in the analysis
be properly determined, since it controls the effi- (Fig. 6.29B).
ciency of the deconvolution. The suitable opera- Fig. 6.30 shows an analysis of the deconvolu-
tor length for the seismic data is obtained from tion outputs for different operator lengths
the analysis of autocorrelations of input seismic along with their corresponding amplitude spec-
traces. tra and autocorrelograms on a marine shot
Figs. 6.27 and 6.28 show a minimum phase gather. Too-short operator lengths leave some
seismogram sampled at 1 ms and its determinis- residual energy in the autocorrelograms corre-
tic and statistical spiking deconvolution results sponding to the source wavelet and reverbera-
for different operator lengths, respectively. In tions, whereas operator lengths longer than
both deconvolution outputs, the effect of the 100 ms do not provide further improvement
seismic wavelet seems to be removed and the on the deconvolution output. For the example
major reflectivity peaks are properly located. shot gather in Fig. 6.30, the most suitable opera-
In the case of deterministic deconvolution, the tor length can be considered to be 80 ms.
analysis shows that better results are obtained
as the operator length increases (Fig. 6.27). For 6.5.4 Prediction Lag
statistical deconvolution, however, some minor
noise amplitudes as spikes trailing the major Prediction lag is definitely the most impor-
reflectivity peaks occur in the deconvolution tant parameter for predictive deconvolution. It
output, and do not diminish as the operator controls the resolution of the deconvolution out-
length increases (Fig. 6.28). Similar spiky noise put, and the multiple reflections can also be sup-
amplitudes are also visible in the deterministic pressed using properly determined prediction
deconvolution output for small operator lag values (Section 7.2). Shorter prediction lags
lengths. In both cases, the best deconvolution cause much more compression of the source
result is obtained for n ¼ 100 ms operator wavelet, and hence increase the bandwidth of
length. This is because the first transient zone the output; however, they may also boost the
of 100-ms length in the autocorrelograms in low- and high-frequency noise amplitudes. As
Figs. 6.27B and 6.28B resembles the autocorrela- the prediction lag increases, the effectiveness
tion of the source wavelet. of the deconvolution on the whitening of the
The suitable operator length approximately spectrum is reduced and the output autocorrelo-
equals the length of the wavelet’s autocorrela- grams include more energy in nonzero lags. If it
tion. Since we normally do not know the source equals the sampling rate, the process becomes a
wavelet embedded in the seismic trace, we can spiking deconvolution. Although the prediction
use the initial parts of the autocorrelation of lag can have any value, selection of the first or