Eliminating migration artifacts in angle domain based on one-way wave equation migration of multiples

The migration artifacts are mainly introduced by the crosscorrealtion of different seismic events. As they honor the imaging condition, they are difficult to be eliminated directly. However, only the traces of the true events are flat in angle domain common image gathers (ADCIGs) and the artifacts can be identified. To separate signal from noise more explicitly, high resolution parabolic Radon transform is applied. In numerical examples, we show that ADCIGs can be extracted from Fourier finite difference (FFD) migration more effectively than in reverse time migration (RTM). We followed the workflow proposed by Biondi and Symes (2004) to extract ADCIGs. In one-way wave equation migration, the crosscorrelation imaging condition is applied and the subsurface item is added to upgoing and downgoing wavefields. Then all the shots are stacked to obtain horizontal offset common image gathers (HOCIGs). Using the formula proposed by Sava and Fomel (2003), we can transform HOCIGs from offset domain to angle domain and obtain ADCIGs. Then the ADCIGs are processed with high resolution parabolic Radon transform. It takes maximum entropy distribution as the constraint condition and uses sparse constrained inversion to improve the resolution. If the migration velocity is correct, the energy of true seismic events is concentrated in the vicinity of zero curvature in the Radon domain, while the energy of artifacts is mapped to nonzero area. Save and Guitton (2005) present their method to suppress multiples in image domain. We modify the method in data to data migration and divide it into five steps: (1) Extract ADCIGs directly from data to data migration; (2) Apply high resolution parabolic Radon transform to ADCIGs; (3) Mute the nonzero components in the Radon domain; (4) Obtain ADCIGs without artifacts using adjoint Radon transform; (5) Stack all different angles to get the final image. In the numerical example, the result of conventional FFD migration using primaries only has no artifacts. It can be improved by muting the low frequency components at large angles. In data to data migration the same muting technique works while the artifacts still exist. After the application of our proposed workflow, most migration artifacts are eliminated and the signal-to-noise ratio of the imaging result is improved effectively. We also show that compared to RTM, FFD is much faster and more efficient to extract ADCIGs. Through the proposed method, the artifacts in the image of data to data migration are mostly eliminated. The final result is comparable to conventional FFD migration. As multiples prediction and wavelet estimation are not needed in data to data migration, it can be significant for real data. The proposed method is mainly aimed at one type of undesired crosscorrelation. How to eliminate the other type of artifacts needs further study. The possible solutions include wide azimuth acquisition technology, least squares migration method and so on. Another point to be addressed is that FFD is suitable to calculate HOCIGs fast while it cannot offer vertical offset common image gathers (VOCIGs). If we need to combine HOCIGs and VOCIGs to generate stable ADCIGs, the FFD operator should be modified.