Velocity model building by waveform inversion of early arrivals and reflections: A 2D case study with gas-cloud effects

Abstract

Joint full-waveform inversion (JFWI) combines reflection waveform inversion (RWI) and early-arrival waveform inversion to build a large-scale velocity model of the subsurface from long-offset data. The misfit function of JFWI requires an explicit separation between the short-spread reflections and early arrivals, the feasibility of which is illustrated with a real data case study. JFWI is alternated with a waveform inversion/migration of short-spread reflections to provide a short-scale impedance model. This model is needed for building the sensitivity kernel of RWI along the two-way reflection paths. The large-scale velocity macromodel built by JFWI can be used as the initial model for classic FWI to enrich the high-wavenumber content of the subsurface model. We have developed an application of this workflow to a real 2D ocean bottom cable (OBC) profile across a gas cloud in the North Sea to review its main promises and pitfalls. Viscoacoustic VTI seismic modeling allows us to account for attenuation and anisotropy effects in a passive way during JFWI and FWI. Using a smoothed version of an existing traveltime tomographic model as the initial model, we first find that the JFWI velocity macromodel is more accurate than the RWI counterpart thanks to the key contribution of the diving waves. Second, we find that the large-scale velocity model updated by JFWI provides a more accurate initial model for classic FWI than does the original smoothed tomographic model. However, because a data difference-based misfit function is used, 2D JFWI still suffers from cycle skipping when a crude 1D velocity model is used as the initial model; therefore, more robust misfit function should be designed to mitigate cycle skipping.