Page 269 - Acquisition and Processing of Marine Seismic Data

260 5. PREPROCESSING

much narrower filter operator in the time In practice, however, shorter operator lengths
domain. Take the inverse Fourier transform are preferred because they require less computa-
to obtain the final filter operator in the time tional time during the applications and hence
domain (Fig. 5.18C). they are more economical.
Frequency filtering is generally used to
Using the filter operator obtained as shown in improve the vertical resolution of the seismic
Fig. 5.18, filtering of the seismic data can be done data. It is generally believed that high frequen-
either in the time or frequency domain. In the cies ensure vertical resolution. On the other
frequency domain application, the amplitude hand, Yılmaz (1987) showed that both low-
spectra of the input seismic trace and the filter and high-frequency components are required
operator are multiplied (Fig. 5.19A). In the time to enhance the vertical resolution. Therefore, it
domain, however, a convolution process is per- is desired that the seismic data have a wide fre-
formed and coefficients of the filter operator are quency band involving both low and high fre-
convolved by seismic trace amplitudes quencies. In the interpretation stage, the
(Fig. 5.19B). Even though both applications pro- frequency content of any particular reflection
duce the same results (Fig. 5.19C), the time is important by means of its continuity on the
domain application is preferred in practice since intersecting seismic lines when the sections are
the convolution process is more economical than tied. Therefore, application of a band-pass filter
computing the forward and then inverse Fourier with similar pass-bands for all vintages of a
transform before and after filtering. particular prospect is important to correlate
After the application, the filtered output trace different seismic datasets.
will contain amplitudes only in the frequency
band of the filter operator. Filtering does not 5.5.2 Band-Pass Filtering of Marine
affect the phase spectrum; only the amplitude Seismic Data
spectrum becomes band limited. If the pass-
band widens in the frequency domain, the filter A spectral analysis of the raw seismic data is
operator in the time domain narrows, and hence required for a correct determination of the cut-
contains a fewer number of nonzero filter coeffi- off frequency and transition band slope param-
cients, which makes the filter operation compu- eters before filtering the marine seismic data. In
tationally faster. The time domain length of general, very high-amplitude swell noise domi-
the operator is also important. Fig. 5.20 shows nates in the low-frequency band of the spectrum
filter operators of different lengths and their of raw marine shots records. Fig. 5.21A shows a
corresponding amplitude spectra, that is, the schematic illustration of an amplitude spectrum
pass-band in the frequency domain. The compu- of a shot gather from a high-resolution marine
tations in Fig. 5.20 indicate that increasing the seismic survey visualized between 0 and the
operator length makes the desired and actual Nyquist frequency (here, 500 Hz for 1 ms sam-
operator spectra similar. The operator lengths pling rate). In addition to its high amplitude
shorter than 400 ms leave some residual ripples content, the most significant characteristic of this
in the pass-band region, while operator lengths noise is its low-frequency content. The fre-
longer than 400 ms do not provide further quency band of the swell noise is generally lim-
improvement on the spectral shape of the ited to a 0–10 Hz band and the noise can easily
pass-band. When the operator is truncated too be removed by application of a band-pass filter
much, the amplitude spectrum of the operator with an approximately 10-Hz low-frequency
degrades, although the slopes of the spectrum cut-off. The 0–10 Hz frequency band, however,
are not affected from the truncation process. may also contain signal amplitudes from deeper