Practical strategies for waveform inversion

The term “waveform inversion” refers to a collection of techniques that use the information from the times and waveform shapes of seismic data to derive high fidelity velocity models for seismic imaging. Waveform inversion was first introduced by Lailly, Tarantola, and Mora(Lailly, 1983; Tarantola, 1984; Mora, 1988). Since these pioneering efforts many researchers have attempted to use various strategies and computational schemes to make waveform inversion implemented either in the time or the frequency domain a processing tool for real data sets. The attractiveness of waveform inversion lies mainly in its lack of approximations, at least in a formal theoretical sense, in contrast to other traditional velocity determination techniques such as semblance or tomography. However, a whole raft of approximations must be made to make the technique viable with today’s computing technology and restrictions of seismic data acquisition. Some of these approximations are rather severe, such as restriction to acoustic waveform inversion while others are made simply to speed up the process. These are collectively referred to as “waveform inversion strategies”, which turn the whole process into a very manpower intensive art form. This paper discusses these various strategies and their influences on the velocity models that are obtained from waveform inversion. One cannot exhaustively test all the choices of strategies, and for that reason the paper focuses on the choices that in our experience create the most difficulties. These approaches will be illustrated on data from offshore Brazil.