In marine data processing, it is a common practice to run residual multiple attenuation as one of the later steps -- after migration, for instance. The multiple-attenuation tests do not access every line of 3D data, and the evaluation of the efficacy of the attenuation is performed along control lines. Since the multiple-attenuation parameters are optimized for the control lines, unfortunately, sometimes, a surprising amount of residual-multiple energy is left and is spread by migration over the entire migrated data set. Sometimes, the evaluation of the multiple attenuation is performed by inspecting post-stack migrated data, or even stacked volumes. Since this ignores the effect of pre-stack migration on the multiple-attenuated data, it can allow residual multiples to appear, manifesting as crosshatched patterns and high-frequency migration noise resembling overmigrated events.

Another very common process after pre-stack migration is automatic residual-moveout correction to mitigate the effects of velocity inaccuracies, which aims to align reflectors horizontally to improve the stacking power and better constrain the data for AVA analysis. Consequently, a method which relies on the flatness of the gathers is a natural process to attenuate residual multiples after automatic residual-moveout correction. However, the problem of residual moveout will not be addressed in this work.

Weglein (1999) classified multiple-attenuation methods into two main categories: (1) filtering methods and (2) prediction methods. The most widely used filtering methods rely on differences in traveltime between primaries and multiples to separate them in appropriate domains where multiples can be filtered out Hampson (1987). Since, regardless of the domain, primaries and multiples normally overlap, particularly at small reflection angles and offsets, the filtering is not perfect; therefore, some of the primary energy can be attenuated, or some of the multiple energy is left out (crosstalk). Prediction methods simulate multiples by auto-convolution of shot gathers and then subtract them from the original data using nonlinear adaptive subtraction Verschuur et al. (1992). As prediction methods became popular, filtering methods inherited the nonlinear adaptive subtraction to attack the crosstalk problem. So, filtering methods can be used to estimate multiples by exploring the same differences between them and primaries as before, but instead of simply filtering them out, a possible approach is to subtract the estimated multiples from the original data using nonlinear adaptive-subtraction schemes. This provides more flexibility to solve specific problems.

Generally, residual-multiple attenuation is performed by filtering methods, by zeroing the residual-multiple energy (along with some primary energy) or by using nonlinear adaptive subtraction. Choo et al. (2004) devised a method to perform residual-multiple attenuation in a different manner. They apply AVA inversion to simulate the primaries and then estimate the residual multiples using nonlinear adaptive subtraction. Although many details of their work are unpublished, the method I present here is, in essence, very similar to the one proposed by them. The method simulates the primaries according to an AVA curve, adjusts amplitude and phase of the simulated primaries to the amplitude and phase of the original data by applying filters computed in a nonlinear adaptive strategy, subtracts the adjusted primaries from the original data to get an estimate of the residual multiples, and, finally, obtains the attenuated data by nonlinear adaptive subtraction. This approach solves the AVA modeling problem globally.

The method is applied on angle-domain common-image gathers (ADCIGs) of a Gulf of Mexico (GOM) 2D data subjected to multiple attenuation after migration. The data set contains very complex multiple patterns Alvarez et al. (2004).

5/6/2007