A demigration-based reflection full-waveform inversion workflow

Abstract

Reflection FWI is effective at providing low-wavenumber velocity updates for deep areas beyond the penetration depth of diving waves and significantly improves seismic imaging. However, the tomographic term of the FWI gradient that is good for low-wavenumber velocity updates can be contaminated by the stronger high-wavenumber migration term. We present a reflection FWI workflow that is based on Born modeling to explicitly generate the tomographic term and thus is minimally impacted by migration contamination. In addition, we propose to use a set of partial stacks as reflectivity models for Born modeling to reduce the risks of cycle-skipping and incorrect update sign and to use a traveltime cost function to mitigate the negative impact from amplitude mismatch between input data and modeled synthetic data. Finally, we demonstrate the benefit and effectiveness of our approach using a field data set from the Gulf of Mexico.

A review on reflection-waveform inversion

“…Since the migration component usually is much stronger than the tomography component, this leakage does not affect the result of RTM very much, but the leakage in the tomography component could lead to RWI updating background velocity along the wrong direction significantly. Wang et al (2018) from CGG demonstrated that the leakage can cause inversion to update the model toward the wrong direction at the steep flanks of salt bodies. In order to mitigate this issue, Irabor and Warner (2016) proposed to split the wavefield into two parts, one of which travels along the vertical axis and the other of which propagates along the horizontal axis, and then applied Eqs.…”

“…13, which shows the width of the convex region in the waveform-based objective function is several times wider than that of conventional FWI. Wang et al (2018) pointed out the likelihood can be further reduced by splitting the longoffset seismic data gather into two parts and then inverting them separately.…”

“…So far, the concept and theory of RWI have become mature. There are many successful applications on marine seismic data (Sun et al 2017;Wang et al 2018). The preprocesses of seismic data include de-multiples and deghost.…”

“…The separation of tomography and migration components is achieved using Born modeling. Both waveform-based (Wang et al 2018) and travel timebased (Sun et al 2017) objective functions succeeded to recover background velocity. Similar to conventional FWI, the industry is trying to apply RWI on land seismic data.…”

Efficient reflection waveform inversion using a locally normalized zero‐lag correlative objective function

“…Although the low‐wavenumber velocity update extracted with vertical Hilbert transformation (equation ()) is overall effective, some contamination occurs around high‐dip structures from the high‐wavenumber components (see Fig. 3e) due to the intrinsic limitation of the vertical Hilbert transform (Wang et al ., 2018). Therefore, in the case of high‐dip structures, the high‐wavenumber artefacts should be suppressed by smoothing the gradient (Yao et al ., 2019).…”