3D full waveform inversion of narrow-azimuth towed-streamer seismic data

Abstract

Full waveform inversion (FWI) is a powerful seismic technique, which aims to produce high resolution, high fidelity quantitative models of the subsurface. Most of the successful FWI applications have been limited to data collected using wide-angle ocean bottom cables (OBCs). The abundance of refracted arrivals help build more accurate low-frequency information into the model. However a large majority of seismic data has been collected using narrow-azimuth towed-streamers (NATS), which often lack these important arrivals. Successful applications of FWI also typically rely on accurate start models obtained from computationally intensive and time-consuming techniques. Therefore being able to successfully and repeatedly apply FWI to a wider range of datasets without the need for accurate start models is of great importance. The effect of acquisition geometry and the accuracy of the start model was investigated on a set of 3D synthetic datasets using acoustic isotropic FWI. It demonstrated that, for simple synthetic experiments with the same physics in the inversion as in the dataset, the model accuracy and acquisition geometry has minimal effect on the success of the inversion. However results from the inversions performed on datasets generated using more realistic physics demonstrated that datasets with longer offsets and wider azimuths did produce more accurate results but reflectors were misplaced in depth. FWI was also successfully applied to a challenging 3D NATS seismic dataset using a basic start model. An efficient strategy for improving the kinematic match of the water bottom arrival by adapting the density model as a proxy for correctly positioning the depth of the water bottom was demonstrated. It was able to improve the kinematic match of the water bottom arrivals which resulted in improved inversions with a better recovered velocity model. The models also produced synthetics with a better match to the observed data. Various strategies were then investigated to invert for both velocity and anisotropy. Local inversion methods for both anisotropy and velocity were first explored, then a combined method of global inversion for the initial anisotropy model and local inversion for the velocity model was demonstrated. Adding anisotropy into the inversion gave a better match of the velocities at the well location than the purely isotropic inversions and more correctly positioned the reflectors in depth. Overall the study shows that FWI can still produce promising results on datasets generated without optimal acquisition geometries, with limited early-arriving transmitted energy, and without a mature start model.