Estimation of primaries via sparse inversion with reciprocity

Accurate removal of surface related multiples is a key step in seismic data processing. The industry standard for removing multiples is SRME, which involves convolving the data with itself to predict the multiples, followed by an adaptive subtraction procedure to recover the primaries (Verschuur and Berkhout, 1997). Other methods involve multidimensional division of the up-going and down-going wavefields (Amundsen, 2001). However, this approach may suffer from stability problems. With the introduction of the ‘‘estimation of primaries by sparse inversion’’(EPSI), van Groenestijn and Verschuur (2009) recentely reformulated SRME to jointly estimate the surface-free impulse response and the source signature directly from the data. The advantage of EPSI is that it recovers the primary response directly, and does not require a second processing step for the subtraction of estimated multiples from the original data. However, because it estimates both the primary impulse response and source signature from the data EPSI must be regularized. Motivated by recent successful application of the curvelet transform in seismic data processing (Herrmann et al., 2007), we formulate EPSI as a bi-convex optimization problem that seeks sparsity on the surface-free Green’s function and Fourier-domain smoothness on the source wavelet. Our main contribution compared to previous work (Lin and Herrmann, 2009), and the contribution of that author to the proceedings of this meeting(Lin and Herrmann, 2010), is that we employ the physical principle of as source-receiver reciprocity to improve the inversion.