Suprasalt model building using full-waveform inversion

Abstract

The application of full-waveform inversion (FWI) to bring high resolution to the velocity model is becoming a standard approach in the velocity model-building workflow. Diving wave FWI in conjunction with reflection FWI (RFWI) has been widely used in the Gulf of Mexico (GOM) to optimize the suprasalt model. Accuracy of a velocity model from tomography is dependent on residual moveout (RMO) picking accuracy. In a good signalto-noise ratio area, the confidence of RMO picking is high. But gathers in areas affected by gas exhibit poor event continuity, which makes it difficult to get accurate RMO picks. In such a geologic regime, FWI can improve the velocity model and therefore the final image quality. There are two main components of a velocity model from the GOM area: the first is the sediment, and the second is salt geometry. In the beginning of the modelbuilding cycle, it is most likely that salt geometry is not accurately defined. This inaccuracy leads to a big mismatch between synthetic and observed data for both diving wave FWI and RFWI. One way to handle this situation is to start with the salt model and iteratively adjust the salt interpretation as FWI model building progresses from lower to higher frequencies. Another approach could be eliminating the salt-related energy from the input and then using the sediment-only model for FWI. We are proposing a desalt approach in which we try to eliminate or reduce the salt-related energy from the input data and then use a sedimentonly velocity model as a starting model for the entire suprasalt FWI workflow. We will present a case study in which, by adapting the desalt workflow, we could manage to do more FWI iterations by eliminating salt interpretation. Introduction Wide-azimuth (WAZ) data in the Gulf of Mexico (GOM) area have proven to be advantageous over conventional narrowazimuth data in terms of better subsurface illumination and fold coverage. Deriving a velocity model using the tomography approach is still the most cost-effective and widely acceptable approach across the industry. Over the past couple of years, the industry has derived and adopted an advanced model-building and imaging approach to improve the overall velocity model and final seismic image. For example, when in the presence of orthogonal WAZ surveys, extending model building from vertical transverse isotropy/tilted transverse isotropy (TTI) to orthorhombic to resolve the azimuthal variation of seismic velocity properties associated with slow and fast velocity direction is a wise choice to realize the best results from expensive orthogonal WAZ acquisition. Reliability of derived models using tomography depends on the accuracy of residual moveout (RMO) picked on the commonimage gather (CIG), and therefore these methods tend to suffer Dhananjay Tiwari1, Jian Mao1, and James Sheng1 in areas with poor signal-to-noise ratio (S/N). Tomography cannot bring high resolution to the velocity model even when S/N is good for RMO picking. Gas-charged sediment and shallow channels are a few geologic scenarios in which the model can be improved by using full-waveform inversion (FWI). We often see seismic-obscured areas in the GOM with little to no signal. These seismic-obscured areas, especially in the shallow zone, hamper deeper imaging due to unresolved velocity anomalies during the tomography workflow. Extending a model-building approach from conventional tomography to FWI can help bring the low-velocity anomaly associated with the gas-charged sediment into the velocity model as well as update the seismicobscured areas we often see in the GOM. FWI works to minimize the differences between observed and synthetic data in terms of amplitude and phase (Lailly, 1983; Tarantola, 1984) by updating the velocity model. If the observed and synthetic differences are within the half cycle of wavelet, then FWI can still provide a desired update to the model without cycle skipping. It is in the best interest of diving wave FWI (DFWI) and reflection FWI (RFWI) to begin with a good background velocity model to mitigate cycle skipping in the beginning. A lack of signal at low frequency limits the FWI capabilities to derive long-wavelength updates for the velocity model. Tomography does an excellent job of updating the long-wavelength corrections to the velocity model. It is also able to derive the long-wavelength background velocity, compensating for the limitations of FWI due to the lack of low frequency in the input. Interaction of reflection and refraction energy from the salt boundary with the suprasalt sediment creates a big mismatch between observed and synthetic data if the starting model is sediment. Possibilities of cycle skipping for FWI updates increase as the mismatch goes more toward a half wavelength between observed and synthetic seismic data. A salt model can be used as a starting model to minimize this effect to some extent. We can use the salt model as a starting model for FWI, and then an iterative approach of updating the salt model at every FWI update (Wang et al., 2015) can be used. But this approach is expensive when used for medium to large 3D seismic projects, especially if the target is to optimize the suprasalt velocity model. Another approach to handle the mismatch near the salt is to remove or attenuate the salt-related energy from the input data. Attenuating the energy associated with the salt boundary from the acquired seismic at the beginning of the project can help reduce the mismatch between observed and synthetic data while using the sediment model. Using this approach can save several iterations of adjusting the salt interpretation during FWI iterations and thus the overall cost. We have demonstrated a robust multistage FWI for high-resolution model building (Mao et al., 2016). Each shot in the WAZ data covers more area; therefore, a coarse-grid shot 1TGS, Houston, Texas, USA. E-mail: [email protected]; [email protected]; [email protected]. https://doi.org/10.1190/tle38030214.1. D ow nl oa de d 03 /0 1/ 19 to 2 05 .1 96 .1 79 .2 37 . R ed is tr ib ut io n su bj ec t t o SE G li ce ns e or c op yr ig ht ; s ee T er m s of U se a t h ttp :// lib ra ry .s eg .o rg /