In Keathley Canyon, Gulf of Mexico, a 3D seismic data set was sharply focused using transmitted energies to update velocities using long and full azimuth streamer data. Authors S. Mothi, K. Schwarz, and H. Zhu of CGG share how it was done.
Because of recent advances in imaging technology, geophysical processors now look to resolution improvment of the underlying seismic velocity model to enhance overall subsurface image quality. In the last decade, almost all approaches were based on picking events from migrated offset/angle gathers and reducing gather curvature by ray tracing. Wave-equation-based full waveform inversion (FWI) is designed to minimize mismatch in the data space and can theoretically achieve a resolution of half the seismic data’s recorded wavelength (Tarantola, 1984). In the last few years, several successful applications of FWI have resulted in high-quality velocity models in various settings, like ocean bottom cable, towed streamer data in deep water, and land data.
Most applications of FWI rely on transmitted energy for determining velocities. The major shortcoming of this method is that velocity updating is limited by the offset range in the acquired data. Using towed-streamer acquisition, penetration depths of transmitted energy are limited to a couple of kilometers below the water bottom in deepwater regions of the Gulf of Mexico (GOM). To overcome this, FWI may be used with the seismic reflections as proposed in Tarantola’s pioneering work that requires knowledge of the densities. Migration-based waveform inversion that uses reflected energy appears promising. However, these methods have not yet been widely adopted for real data.
In this project, transmitted energies updated the velocities in Keathley Canyon, GOM using long offsets (up to 18km) and full azimuth (up to 10km) streamer data with variable depth tow. The staged methodology presented uses the wide range of offsets to investigate the impact of long offsets and full azimuth data on velocity inversion.
Study area
The study area is in the deepwater GOM, and is characterized by sedimentary basins with faults, carapaces, and complex salt structures. The staggered configuration of two streamer boats allows ultra-long offsets of over 18km in the inline direction, Figure 1a. The study uses east-west and north-south sail lines, Figure 1b, yielding full azimuth coverage, Figure 1c. Acquisition with variable-depth towed streamers results in diversity of the receiver ghost notch, creating significant lowfrequency content at far offsets that can aid initial FWI iterations.
FWI components
Acquired shot gathers are minimally processed with a mild denoise flow to remove low frequency swell noise and spikes. The shot records are muted to prevent any reflection energy and surface-related reflection multiples from interfering with the velocity updates. The far-field source signature is deghosted for the source ghost to obtain the source wavelet for the inversion. Then, a TTI acoustic model is applied and the FWI updates velocity along the tilt axis V0 using transmitted energy. The anisotropic parameters δ and ε are derived using well logs and 1D inversion, and the tilt axis is assumed to be perpendicular to the plane of the bedding. The starting velocity model is obtained using two iterations of ray-based tomography with beam migration to obtain a good starting velocity trend and prevent FWI from converging to a local minimum.
Offset stripping inversion
Starting with the lowest usable frequency content in the data, i.e. from 2-3Hz, the data is split into three offset classes: middle offsets (less than 8km), long offsets (8–12km) and ultra-long offsets (plus 12km). Structural horizons determined by the diving wave’s maximum penetration depth for the starting model and available offset ranges are used to constrain the velocity update regions. Shallow velocity updates are then done using several iterations of FWI on the middle-offset data at 3Hz. Following this, the long offsets are included in the inversion, and the updates cover the shallow and deeper sections. This process is repeated for the ultra-long offsets. At this stage, there is 3Hz inversion for the shallow overburden, using all the offsets and 3Hz data. Next a multi-scale approach is used to invert up to 7Hz in increments of 1Hz, using all the offsets.
Long offsets
.jpg)
Next, processing performs two inversion tests. Test #1 imitates a traditional wide azimuth acquisition (WAZ) receiver spread with maximum inline and crossline offsets, 7.5km and 4km respectively. Test #2 uses no offset restrictions. Test #1 produces velocity updates in deeper sections that are highly oscillatory in nature, and the carbonate section is incorrectly inverted as a slow-velocity section; this is attributed to insufficient diving ray penetration. From Test #2, the updates are more characteristic of the geology and do not have the same artificial oscillation. Furthermore, offset gathers from beam migrations also suggest that the perturbation from Test #1 results in over-corrected gathers in the area immediately above the carbonate section. Test #2 produces flatter gathers.
Full azimuth
Orthogonal acquisition provides data with a variety of azimuth information. With increased crossline offset, the transverse direction produces better illumination and transmission energy penetration, as well. To understand the benefits of this acquisition configuration, the above inversion scheme was applied to middle-offset data using 1) only N-S shotlines and 2) both N-S and E-W shotlines (full azimuth). From the 3Hz FWI results (after 1 iteration), the inversion suffers from a vertical striping pattern, i.e. an acquisition footprint, when using limited azimuth information. Sparse sampling of the acquisition in the crossline direction compared to the inline may cause these artifacts. This effect tends to be cumulative. In contrast, inversion using the full azimuth data does not suffer from these issues.
Inversion results
With the offset stripping and multiscale approach, processors perform the inversion using full azimuth and long-offset data for frequencies up to 7 Hz. The velocity models, Figures 2 and 3, clearly show that the inversion produces geological models, and tracks the faults and high-velocity condensed sections directly above the top of salt. The carbonates and the shale bodies are detected by the inversion. The pre-stack depth migration (PSDM) stack response is also improved after the inversion, see main image Map 1. Further, beam migrations show that the gather flatness is mostly improved and the detailed velocities result in more consistent curvature across the orthogonal azimuths. The improvement from the gathers is subtle because the starting velocity model had two iterations of reflection tomography.
These results clearly show the image improvement that additional offsets and azimuths bring to the inversion by illuminating more subsurface angles and deeper sections. This approach produces interpretive models that better characterize the different geological sections. OE
Acknowledgements
The authors thank CGG for permission to publish these results and acknowledge Yu Zhang, Andrew Ratcliffe, and Graham Conroy for their support with the inversion engine. We appreciate the input from Tony Huang, Kyle Huang, Qiaofeng Wu, Yunfeng Li and Kristin Johnston.
Authors
Sabaresan Mothi has worked for CGG for six years. Mothi earned an MS degree in electrical engineering from Texas A&M University and a MS degree in computational and applied mathematics from Rice University.
Katherine Schwarz has worked for CGG for four years. She is a senior seismic imager with experience in TTI and orthorhombic depth migration in the Gulf of Mexico and on land. Schwarz earned a PhD in physics from the University of California at Berkeley. Huifeng Zhu has worked for CGG for three years and is currently an imaging project coordinator. He has performed velocity model building work with both ray-based tomography and full waveform inversion. Zhu holds a B.Sc. in physics from the University of Science and Technology of China and a Ph.D. in physics from the University of Houston.
Subscribe
