Page 390 - Acquisition and Processing of Marine Seismic Data
P. 390

7.2 PREDICTIVE DECONVOLUTION                        381































           FIG. 7.16  (A) A high-resolution shallow seismic line with a first-order multiple (M 1 ), and (B) output of the predictive
           deconvolution for n ¼ 160 ms with a water depth-dependent prediction lag. Autocorrelograms of the sections are given in
           the bottom panels.


           prediction lag of α ¼ 40 ms. The most suitable  multiples and primaries interfere, become much
           operator length for the shot in Fig. 7.15 is  more distinct.
           n ¼ 80 ms, which removes almost all of the mul-  The theoretical foundation of predictive
           tiple energy from autocorrelograms of the    deconvolution is based on the periodicity of
           deconvolution output.                        the multiples in shot gathers or CDPs. The peri-
              Fig. 7.16A shows a shallow seismic line along  odicity, however, is only completely preserved
           which the water depth slightly changes. Uplift-  in zero-offset sections, and therefore predictive
           ing of the acoustic basement results in a trans-  deconvolution may not be perfectly successful
           parent and reflection-free zone at the center, in  on nonzero offset data, such as shot or CDP
           which the first-order multiple (M 1 ) is clearly vis-  gathers. In some cases, predictive deconvolution
           ible. Autocorrelograms of the input section indi-  is applied to the stack sections. On the other
           cate that the period of the multiple also changes  hand, a suitable result may not be achieved since
           as the water depth increases. Therefore, a pre-  the stacking modifies the amplitude relations
           diction lag varying from 240 to 340 ms depend-  between the primaries and the multiples. In fact,
           ing on the water depth is used with an operator  the only domain where the periodicity and
           length of 160 ms. The output of predictive   amplitude relationships are preserved is the
           deconvolution shows that most of the first-order  slant stack (τ-p) domain, and the multiples can
           multiple energy is suppressed (Fig. 7.16B), and  be better suppressed by predictive deconvolu-
           the primary reflections of the deeper parts of  tion applied in the slant stack domain. In the fol-
           the righthand side of the section, where the  lowing section, the theory and practice of the
   385   386   387   388   389   390   391   392   393   394   395