Simple Multiple

Key Takeaways

• Water-bottom simple multiples in marine seismic surveys are predictable in their travel time (exactly twice the two-way water bottom reflection time), their moveout velocity (equal to the water velocity, approximately 1,500 meters per second, which is slower than the stacking velocity of any genuine deep primary reflection), and their amplitude and waveform (which replicate the water-bottom primary reflection modified by two reflections at the water surface, introducing a polarity reversal per water-surface reflection): the predictability of water-bottom simple multiples enables their identification on unstacked CMP gathers by the distinctive low-velocity moveout (the hyperbolic curvature of the multiple in the offset-time domain is flatter than for primary reflections at the same two-way time because the multiple's stacking velocity is the water velocity rather than the higher velocity of deep primaries), and their removal by velocity-discriminant stacking (applying NMO with the primary velocity and stacking, which automatically attenuates the multiple because its slower moveout means it is not correctly flattened by the primary velocity NMO correction); the water-bottom multiple is most problematic in shallow-water environments where the water bottom is strong and the water layer acts as an efficient resonator for the multiple energy, and where the multiple's two-way time (twice the shallow water bottom time) falls within the target depth range for deep reservoir reflections; in deep water (water depths greater than approximately 2,000 meters), the water-bottom multiple arrives at travel times exceeding 4 seconds (for 2000 m at 1500 m/s), which is typically deeper than the exploration target depths and therefore does not interfere with the primary reflection from most deep-water reservoirs.
• Surface-related multiple elimination (SRME) is the industry-standard method for predicting and subtracting simple multiples in marine seismic data, using the data itself (without a subsurface model) to predict the multiple wavefield through a convolution of the data with itself: the fundamental observation underlying SRME is that any seismic trace can be convolved with another trace to predict the multiple that would be generated by the surface reflection between the two recording positions, because the multiple's travel path is equivalent to a primary from the first trace followed by a downward reflection at the surface and a second primary from the second trace; by summing these cross-convolutions over all source-receiver pairs that contribute to a given multiple travel path, SRME constructs a prediction of the full multiple wavefield for each common-midpoint location, which is then adaptively subtracted from the data to remove the multiples while preserving the primaries; SRME requires a continuous spatial distribution of sources and receivers to accurately predict all multiple travel paths, which is why 3D SRME (which uses the 3D shot-receiver geometry of a modern marine 3D survey) is more effective than 2D SRME (which uses only the in-line receiver positions) for removing multiples that have out-of-plane travel paths in areas with complex 3D bathymetry or geological structure; the adaptive subtraction step (which finds the amplitude and phase filter that minimizes the energy of the residual multiple after subtraction) is critical for handling the amplitude and waveform differences between the predicted and actual multiples that arise from propagation effects not captured by the SRME convolution model.
• Radon demultiple (also called parabolic Radon or hyperbolic Radon transform demultiple) is a complementary method for attenuating simple and interbed multiples in seismic CMP gathers by exploiting the moveout velocity difference between primaries and multiples: the Radon transform converts the seismic data from the offset-time domain into a curvature-moveout domain (the Radon parameter space), where primary reflections and multiples that have different moveout velocities appear at different positions in the Radon domain; a mute in the Radon domain that keeps the primary-moveout Radon parameters and rejects the multiple-moveout parameters, followed by inverse transformation back to the offset-time domain, produces a primary-only seismic dataset from which the multiple-contaminated portions have been removed; the Radon demultiple is most effective when the velocity separation between primaries and multiples is large (for example, the water-bottom multiple in shallow water, which has a stacking velocity of 1,500 m/s while the deepest primary reflections have stacking velocities of 2,500 to 4,000 m/s), and least effective when primaries and multiples have similar moveout velocities (for example, intrabed multiples between two closely spaced reflectors that have nearly the same moveout as the primaries below them); Radon demultiple is typically applied after NMO correction and in conjunction with SRME to achieve maximum multiple attenuation, with SRME removing the surface-related components and Radon demultiple handling the residual multiples and interbed multiples that SRME does not predict.
• Predictive deconvolution for simple multiple attenuation exploits the periodic nature of water-bottom simple multiples (which arrive at regular time intervals equal to the water-bottom two-way time) to design a filter that predicts and subtracts the periodically repeating multiple energy from the seismic trace: the autocorrelation of a seismic trace containing periodic multiples shows peaks at the multiple lag time (the water-bottom two-way time and its multiples), and the deconvolution filter designed from this autocorrelation has zeros at these lag times, effectively removing the periodic multiple energy while preserving the non-periodic primary reflection content; predictive deconvolution is most effective when the multiple period (the lag between the multiple and its primary) is constant over the data (constant water depth), the multiple is strictly periodic (no variation in the reflection coefficient or travel path between multiple bounces), and the primary wavefield is not periodic at the same lag times; the limitations of predictive deconvolution arise from the assumptions of a stationary, minimum-phase wavelet and a purely periodic multiple that are violated in practice by lateral velocity variation (changing water depth changes the multiple period), noise, and the overlap of multiple and primary energy in the autocorrelation that causes the deconvolution filter to imperfectly separate them; predictive deconvolution was the primary demultiple method in 2D marine seismic processing before the development of SRME and Radon methods, and continues to be used as a preprocessing step and as a supplement to the more sophisticated modern methods.
• Simple multiple identification in seismic interpretation requires distinguishing multiples from primary reflections using geological plausibility, velocity analysis, and synthetic seismogram comparison, because multiple attenuation processing never completely eliminates all multiple energy and residual multiples in the stacked section can appear as genuine reflections if not recognized: the first test for a potential multiple is the moveout velocity check, where the velocity spectrum shows whether the reflection has a stacking velocity consistent with primary moveout (higher velocity, more curvature at near offsets) or multiple moveout (lower velocity, flatter curvature); if the velocity is ambiguous, the synthetic seismogram (generated from the sonic and density logs in nearby wells convolved with a source wavelet) provides a direct prediction of the primary reflection events, and any event in the seismic that does not appear in the synthetic is a candidate for a multiple; in areas with no well control, the time-period test (checking whether the event occurs at a time equal to twice the two-way time of a strong shallower reflection) is the primary identification tool; the consequences of misidentifying a multiple as a primary reflection include drilling wells targeting the phantom "reflector" and finding only empty formation at the predicted depth, which is why the demultiple processing workflow and the residual multiple quality control in seismic processing directly affect exploration success rates.