| Popis: |
We m-examine the problem of pre-stack marine multiple suppression within the context of a predictive modelling and adaptive subtraction scheme. The multiple pra diction method is a compromise between single trace predictive deconvolution and full wave-equation modelling. Predictive decon fails to account for both diffractions and differential NM0 as well sz pegleg splitting due to different water depths at source and receiver positions. Our prediction technique reduces almost to predictive decon in shallow water yet accurately models diffraction effects in deeper water where diffractions become more dominant. A transition from shallow to deep water is characteristic of many lines in offshore continental margin exploration environments. Post-stack data comparisons show that the overall improvement in primary to multiple ratios can be spectacular. Although the method is certainly not a panacea for multiple problems, dramatic improvements in data quality can be realized at marginal incremental cost. As with all predictive multiple suppression techniques, the method complementa the stacking process’s ability to enhance real events. Despite the oft-noted fact that stack is a ‘powerful tool’, effective pm-stack multiple suppression improves the geophysicist’s ability to pick more optimal mutes and stacking velocities contributing to bottom line gains in seismic interpretability. INTRODUCTION: The suppression of water bottom generated multiples is a major problem in processing offshore California seismic data. The hard bottom here can be attributed to a high energy marine environment which has prevented accumulation of younger sediments on the seafloor. Conventional approaches to multiple attenuation fail in this area for a number of reasons. The relatively shallow water bottom (160-300 mils) makes it difficult to improve primary to multiple ratios by exploiting residual NM0 be cause the primary and multiple velocities are so similar. Also, the strong subsurface dips combined with a dipping seafloor decrease the stationarity of the multiples and degrade the performance of pre-stack gapped deconvolution. Effective multiple prediction requires a more accurate technique to properly account for these effects. Several authors have reviewed the use of the wave equation for improved pm-stack multiple prediction. Lowenthal et al (1974), Fourmannet al (1979), Morley (1982), Bernth & Sonneland (1983), Wiggins (1985) and, most recently, Berryhill & Kim (1986) have all examined various aspects of the problem. Most of the above papers have touted the use of this technique for deep water areas. Although the differences between wave equation prediction and single trace prediction are always more noticeable over deep than over shallow water, the expense of wave equation prediction can often be justified in water of shallow to intermediate depth. In these aress, the improvement in primary to multiple ratios on intermediate offsets as well as on near ‘offsets in the presence of structural dip can greatly aid the interpreter. MULTIPLE MODELLING dz DATA PRECONDITIONING: Our modelling method is a Kirchoff wavefield extrapolation technique similar to that described by Berryhill (1986). We use a best estimate of water depth along the line ss well as water velocity to forward-continue each recorded shot wavefield to an imaginary datum defined at the mirror image of the seafloor. These wavefields are then extrapolated back to the surface to simulate a roundtrip passage through the water layer. This process creates a pre-stack model of the seafloor-generated multiples. The modelling just described is more susceptible than most data processing to artifact generation. A considerable amount of care is required to pm-condition the data to get around truncation and aliszing problems. Positive offsets (simulated receivers ahead of the boat) need to be reconstructed using seismic reciprocity. This is particularly true when the shooting is downdip. Typically, this corresponds to the acquisition case where the boat is headed shoreward. In this case the apexes of the events in the shot gathers do not fall within the actual recording spread; the multiple model will thus tend to lose energy on the near traces. This is because the reflection energy tends to spread away from the reflection events’ apexes. To avoid model ahazing we pm-interpolate traces within the receiver spread and low-pass filter the data particularly at large offsets and shallow times. Subtraction of the lower frequency model from the data can leave high frequency residual multiples pre-stack. These residual events, however, usually respond well to stack cancellation. SUBTRACTION TECHNIQUES: A variety of filter design techniques are currently under development to fit the model to the data before subtraction. These methods range from space-time invariant |
