The theory of FWI is well-established. However its practical application to 3D seismic datasets is still a subject of intense research. This technique has shown spectacular results in quantitatively extracting P-wave velocities in the shallow near surface at depths of less than 1 km, using wide-angle OBC datasets. This study deals with establishing a robust methodology for the application of FWI that can be routinely applied to analogous field datasets, both in the shallow near surface and at deeper reservoir depths. A practical strategy for anisotropic 3D acoustic FWI was developed and implemented. The stratergy is tested on a series of 3D datasets: (1) a synthetic Marmousi dataset, (2) an OBC field data and (3) a streamer data.

A 3D synthetic Marmousi data is used to compare FWI implementations in both the time domain and the frequency domain. In both domains, it was possible to recover an almost ‘perfect’ model with complete data coverage, no noise, and few iterations. Both approaches were useful and competitive, and ideally both should be available within a comprehensive suite of inversion tools.

The anisotropic time-domain FWI strategy was successfully implemented to complex OBC field data set with long offsets, full-azimuthal coverage and low frequencies. The FWI quantitatively recovered p-wave velocities in the shallow near surface, at intermediate depths where the sediments are gas bearing, and at deeper reservoir depths. The velocities are indeed realistic and are consistent with an independent reflection PSDM volume, well data and pressure data. The synthetic FWI data better match the field data, with the phase residuals between the two datasets significantly reduced to low values. The gathers are flatter and the depth-migrated images are more resolved and focused.

The strategy was also successfully implemented to complex streamer field data set with short offsets, narrow-azimuthal coverage and reduced signal at the low frequencies. The FWI quantitatively recovered P-wave velocities down to depths of 750 m. A complex series of high and low velocity channels are recovered. These are consistent with an independent reflection PSTM volume. The synthetic FWI data better match the field data, with the phase residuals between the two datasets significantly reduced to low values. The depth-migrated images are more resolved and focused in the shallow section.