Efficient full-waveform inversion with marine acquisition geometry

Full-waveform inversion (FWI) is a nonlinear data fitting procedure based on seismic data to derive a accurate velocity model. With the increasing demand for high resolution images in complex geological settings, the importance of improvements in acquisition and inversion become more and more critical. However, these improvements will be obtained at high computational cost, as a typical marine survey contains thousands of shot and receiver positions, and FWI needs several passes through massive seismic data. Computational cost of FWI will grow exponentially as the size of seismic data and desired resolution increase. In this paper we present a modified Gauss-Newton (GN) method that borrows ideas from compressive sensing, where we compute the GN updates from a few randomly selected sequential shots. Each subproblem is solved by using a sparsity promoting algorithm. With this approach, we dramatically reduce the size and hence the computational costs of the problem, whilst we control information loss by redrawing a different set of sequential shots for each subproblem.