Full-waveform inversion of seismic data

Full-waveform inversion (FWI) is a data-driven imaging technique that seeks a model of physical properties of the subsurface that can explain the observed data, which is acquired in the field. In essence, FWI makes predictions match reality. Despite its high computational cost, FWI is routinely used to obtain high-resolution subsurface models of speed of sound that are crucial for geology interpretation, reservoir characterization, and reduction of the risks associated with well-drilling. This chapter gives a general overview of this technique, presents a best-practice FWI workflow for marine datasets with a practical field data example, discusses its implementation to other datasets, as well as potential pitfalls and future perspectives. The chapter is organized in three sections: introduction, FWI workflow and its implementation on a 3D field dataset, and an overview of some of the challenges and opportunities for FWI over the coming years. In the first section we introduce full-waveform inversion by reviewing its origins and main characteristics, discussing its superiority to conventional travel-time tomography, giving the necessary data requirements and reviewing the basic theoretical framework of FWI. In the second section, we demonstrate a practical application of a time-domain 3D acoustic anisotropic FWI to a marine field dataset over the Tommeliten Alpha field in the North Sea, which is applicable to a wide range of analogous datasets, and give a detailed workflow required for a successful implementation of FWI. We show that the recovered 3D P-wave velocity model accurately matches the sonic measurements at a well location and leads to the flattening of common image gathers as well as a significant uplift of the migrated image, which results in more continuous reflectors within the reservoir section. In the last section of the chapter we aim to: (1) give an overview of the latest and most relevant applications of FWI for crustal imaging and challenging land environment, (2) show that high-frequency FWI is useful not only for velocity model building but also for direct geological interpretation, as it has capacity to produce high-resolution PSDM-like sections, potentially eliminating the need for conventional processing and, finally, (3) briefly discuss how machine learning is being used to optimize algorithms and what impact FWI could have in other scientific fields.