SUMMARY Gradient computations in full-waveform inversion (FWI) require calculating zero-lag cross-correlations of two wavefields propagating in opposite temporal directions. Lossless media permit accurate and efficient reconstruction of the incident field from recordings along a closed boundary, such that both wavefields propagate backwards in time. Reconstruction avoids storing wavefield states of the incident field to secondary storage, which is not feasible for many realistic inversion problems. We give particular attention to velocity–stress modelling schemes and propose a novel modification of a conventional reconstruction method derived from the elastodynamic Kirchhoff–Helmholtz integral. In contrast to the original formulation (in a previous related work), the proposed approach is well-suited for velocity–stress schemes. Numerical examples demonstrate accurate wavefield reconstruction in heterogeneous, elastic media. A practical example using 3-D elastic FWI demonstrates agreement with the reference solution.

Full Waveform Inversion (FWI) is a procedure used to determine the elastic parameters of the Earth by reducing the misfit between observed elastodynamic wavefields and their numerically modeled counterparts. The numerical solution of the elastodynamic wave equation is computationally expensive and its performance is typically bandwidth bound. Computing the gradient of the FWI misfit functional adds further complexity as it involves computing the zero-lag cross-correlation of two wavefields propagating in opposite temporal directions. In this paper, we utilize graphics processing units (GPUs) for their high memory bandwidth and combine two principal optimizations in order to compute FWI gradients on large models and for long simulation times. Wavefield reconstruction methods allow efficient gradient computations with minimal memory requirements and interconnection transfers. Time-space tiling techniques permit us to transcend the limited amount of GPU memory while avoiding dramatic slowdowns due to the low interconnection bandwidth. The implementation considers a task-oriented, hybrid usage of explicitly managed and Unified Memory in order to satisfy the requirements. Benchmarks demonstrate that the proposed approach is able to preserve 78 − 90% of the original performance, when oversubscribing the amount of physical memory available on GPUs. Comparison with existing methods highlights the benefits of the method.

Incorrect imaging of internal multiples can lead to substantial imaging artefacts. It is estimatedthat the majority of seismic images available to exploration and production companies have had nodirect attempt at internal multiple removal. In Part I of this article we considered the role of spar-sity promoting transforms for improving practical prediction quality for algorithms derived fromthe inverse scattering series (ISS). Furthermore, we proposed a demigration-migration approach toperform multidimensional internal multiple prediction with migrated data and provided a syntheticproof of concept. In this paper (Part II) we consider application of the demigration-migration approach to field data from the Norwegian Sea, and provide a comparison to a post-stack method (froma previous related work). Beyond application to a wider range of data with the proposed approach,we consider algorithmic and implementational optimizations of the ISS prediction algorithms tofurther improve the applicability of the multidimensional formulations.

The presence of internal multiples in seismic data can lead to artefacts in subsurface images ob-tained by conventional migration algorithms. This problem can be ameliorated by removing themultiples prior to migration, if they can be reliably estimated. Recent developments have renewedinterest in the plane wave domain formulations of the inverse scattering series (ISS) internal multipleprediction algorithms. We build on this by considering sparsity promoting plane wave transformsto minimize artefacts and in general improve the prediction output. Furthermore, we argue forthe usage of demigration procedures to enable multidimensional internal multiple prediction withmigrated images, which also facilitate compliance with the strict data completeness requirementsof the ISS algorithm. We believe that a combination of these two techniques, sparsity promotingtransforms and demigration, pave the way for a wider application to new and legacy datasets.