Oz Yilmaz
Oz Yilmaz has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
- Hossein Jodeiri Akbari Fam
Laurentian University, Harquail School of Earth Sciences, Sudbury, Ontario, Canada.;;;;;;;;;;
3 shared publications - Mostafa Naghizadeh
Laurentian University
2 shared publications - André J.-M. Pugin
Geological Survey of Canada, Ottawa, Ontario, Canada.
1 shared publication - Han Wang
SINOPEC 4 , Petroleum Exploration and Production Research Institute, Beijing, China
1 shared publication - Jianghai Xia
Zhejiang University of China, Hangzhou, Zhejiang, China.
1 shared publication - Jie Zhang
School of Management, University of Science and Technology of China, Hefei 230026, China
1 shared publication
Latest Publications
- 7 January 2025
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 90 (4) , U31-U46
Root-mean-square velocity estimation by a deep-learning method
Migration velocity analysis is often a crucial yet labor-intensive step in seismic data processing, traditionally involving iterative manual picking of semblance spectra. This process is subjective and can lead to inconsistent results from different analysts. Despite advancements in automated velocity estimation, most methods process a single common-midpoint gather or semblance spectrum individually, neglecting lateral continuity between midpoints. We develop a novel approach using a 3D U-Net to extract the entire 2D root-mean-square (rms) velocity field from a 3D constant-velocity migration (CVM) volume of any size. The 3D convolutions in the U-Net enable feature extraction across three dimensions of the CVM volume — velocity, time, and midpoint. The training and validation processes use a diverse collection of synthetic and field data to ensure robustness and real-world applicability. The initial tests on synthetic examples indicate that this method can handle the rms velocity field generated by horizontal layers, undulating layers, and simple faults. Leveraging the transfer learning strategy, we apply our model to the Marmousi data set and two field data sets. The predicted velocity fields and corresponding prestack time migration images demonstrate that this deep-learning method efficiently produces stable and geologically plausible rms velocity fields, closely matching conventional manual picking results. These findings suggest that neural networks have the potential to process large-scale and complex field seismic data sets.
- 7 January 2025
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 90 (4) , U31-U46
Root-mean-square velocity estimation by a deep-learning method
Migration velocity analysis is often a crucial yet labor-intensive step in seismic data processing, traditionally involving iterative manual picking of semblance spectra. This process is subjective and can lead to inconsistent results from different analysts. Despite advancements in automated velocity estimation, most methods process a single common-midpoint gather or semblance spectrum individually, neglecting lateral continuity between midpoints. We develop a novel approach using a 3D U-Net to extract the entire 2D root-mean-square (rms) velocity field from a 3D constant-velocity migration (CVM) volume of any size. The 3D convolutions in the U-Net enable feature extraction across three dimensions of the CVM volume — velocity, time, and midpoint. The training and validation processes use a diverse collection of synthetic and field data to ensure robustness and real-world applicability. The initial tests on synthetic examples indicate that this method can handle the rms velocity field generated by horizontal layers, undulating layers, and simple faults. Leveraging the transfer learning strategy, we apply our model to the Marmousi data set and two field data sets. The predicted velocity fields and corresponding prestack time migration images demonstrate that this deep-learning method efficiently produces stable and geologically plausible rms velocity fields, closely matching conventional manual picking results. These findings suggest that neural networks have the potential to process large-scale and complex field seismic data sets.
- 5 October 2022
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 88 (1) , T13-T31
3D generalized spherical multifocusing seismic imaging
We introduce a 3D generalized spherical multifocusing (GSMF) algorithm to generate a high-resolution 3D stacked volume that is equivalent to a synthesized 3D zero-offset wavefield for crooked-line/3D seismic data. The proposed algorithm can be applied to arbitrary recording geometry from areas with irregular topography, a complex near the surface, and complex subsurface. The 3D GSMF method simultaneously corrects for elevation statics, nonhyperbolic moveout associated with reflections beneath complex overburden structures, and azimuth-dependent dip-moveout effects. In addition, the formulation is dually generalized for the optical domain and the effective medium. The optical domain and effective medium parameterizations account for heterogeneity either by shifting the reference time to project the problem into the optical image space or by adjusting the velocity of an effective overburden, respectively. We test the performance of our method using 3D synthetic data with 3D and crooked-line surveys. The numerical tests have shown that the accuracy of the new approximation is significant for gently to highly curved interfaces beneath low to relatively high heterogeneous overburden with rugged topography, even at large offsets and midpoint separations. In addition, we rigorously evaluate the method using 3D real seismic data acquired over a complex thrust-belt area with rugged terrain. Compared with conventional 3D stacking, the new formulation yields a high resolution and accurate seismic stacked volume from land seismic data collected with arbitrary 3D geometries.
- 5 October 2022
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 88 (1) , T13-T31
3D generalized spherical multifocusing seismic imaging
We introduce a 3D generalized spherical multifocusing (GSMF) algorithm to generate a high-resolution 3D stacked volume that is equivalent to a synthesized 3D zero-offset wavefield for crooked-line/3D seismic data. The proposed algorithm can be applied to arbitrary recording geometry from areas with irregular topography, a complex near the surface, and complex subsurface. The 3D GSMF method simultaneously corrects for elevation statics, nonhyperbolic moveout associated with reflections beneath complex overburden structures, and azimuth-dependent dip-moveout effects. In addition, the formulation is dually generalized for the optical domain and the effective medium. The optical domain and effective medium parameterizations account for heterogeneity either by shifting the reference time to project the problem into the optical image space or by adjusting the velocity of an effective overburden, respectively. We test the performance of our method using 3D synthetic data with 3D and crooked-line surveys. The numerical tests have shown that the accuracy of the new approximation is significant for gently to highly curved interfaces beneath low to relatively high heterogeneous overburden with rugged topography, even at large offsets and midpoint separations. In addition, we rigorously evaluate the method using 3D real seismic data acquired over a complex thrust-belt area with rugged terrain. Compared with conventional 3D stacking, the new formulation yields a high resolution and accurate seismic stacked volume from land seismic data collected with arbitrary 3D geometries.
- 1 January 2022
- Published by Society of Exploration Geophysicists in The Leading Edge
- Vol. 41 (1) , 40-46
A reality check on full-wave inversion applied to land seismic data for near-surface modeling
We evaluate the performance of traveltime tomography and full-wave inversion (FWI) for near-surface modeling using the data from a shallow seismic field experiment. Eight boreholes up to 20-m depth have been drilled along the seismic line traverse to verify the accuracy of the P-wave velocity-depth model estimated by seismic inversion. The velocity-depth model of the soil column estimated by traveltime tomography is in good agreement with the borehole data. We used the traveltime tomography model as an initial model and performed FWI. Full-wave acoustic and elastic inversions, however, have failed to converge to a velocity-depth model that desirably should be a high-resolution version of the model estimated by traveltime tomography. Moreover, there are significant discrepancies between the estimated models and the borehole data. It is understandable why full-wave acoustic inversion would fail — land seismic data inherently are elastic wavefields. The question is: Why does full-wave elastic inversion also fail? The strategy to prevent full-wave elastic inversion of vertical-component geophone data trapped in a local minimum that results in a physically implausible near-surface model may be cascaded inversion. Specifically, we perform traveltime tomography to estimate a P-wave velocity-depth model for the near-surface and Rayleigh-wave inversion to estimate an S-wave velocity-depth model for the near-surface, then use the resulting pairs of models as the initial models for the subsequent full-wave elastic inversion. Nonetheless, as demonstrated by the field data example here, the elastic-wave inversion yields a near-surface solution that still is not in agreement with the borehole data. Here, we investigate the limitations of FWI applied to land seismic data for near-surface modeling.
- 1 January 2022
- Published by Society of Exploration Geophysicists in The Leading Edge
- Vol. 41 (1) , 40-46
A reality check on full-wave inversion applied to land seismic data for near-surface modeling
We evaluate the performance of traveltime tomography and full-wave inversion (FWI) for near-surface modeling using the data from a shallow seismic field experiment. Eight boreholes up to 20-m depth have been drilled along the seismic line traverse to verify the accuracy of the P-wave velocity-depth model estimated by seismic inversion. The velocity-depth model of the soil column estimated by traveltime tomography is in good agreement with the borehole data. We used the traveltime tomography model as an initial model and performed FWI. Full-wave acoustic and elastic inversions, however, have failed to converge to a velocity-depth model that desirably should be a high-resolution version of the model estimated by traveltime tomography. Moreover, there are significant discrepancies between the estimated models and the borehole data. It is understandable why full-wave acoustic inversion would fail — land seismic data inherently are elastic wavefields. The question is: Why does full-wave elastic inversion also fail? The strategy to prevent full-wave elastic inversion of vertical-component geophone data trapped in a local minimum that results in a physically implausible near-surface model may be cascaded inversion. Specifically, we perform traveltime tomography to estimate a P-wave velocity-depth model for the near-surface and Rayleigh-wave inversion to estimate an S-wave velocity-depth model for the near-surface, then use the resulting pairs of models as the initial models for the subsequent full-wave elastic inversion. Nonetheless, as demonstrated by the field data example here, the elastic-wave inversion yields a near-surface solution that still is not in agreement with the borehole data. Here, we investigate the limitations of FWI applied to land seismic data for near-surface modeling.
- 28 July 2021
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 86 (6) , S355-S369
2.5D multifocusing imaging of crooked-line seismic surveys
Two-dimensional seismic surveys often are conducted along crooked-line traverses due to the inaccessibility of rugged terrains, logistical and environmental restrictions, and budget limitations. The crookedness of line traverses, irregular topography, and complex subsurface geology with steeply dipping and curved interfaces could adversely affect the signal-to-noise-ratio of the data. The crooked-line geometry violates the assumption of a straight-line survey that is a basic principle behind the 2D multifocusing (MF) method and leads to crossline spread of midpoints. In addition, the crooked-line geometry can give rise to potential pitfalls and artifacts and thus leads to difficulties in imaging and velocity-depth model estimation. We develop a novel MF algorithm for crooked-line seismic data and revise the traveltime equation accordingly to achieve better signal alignment before stacking. Specifically, we have developed a 2.5D MF reflection traveltime equation that explicitly takes into account the midpoint dispersion and cross-dip effects. The new formulation corrects for normal, inline, and crossline dip moveouts (DMOs) simultaneously, which is significantly more accurate than removing these effects sequentially. Applying normal moveout, DMO, and cross-DMO separately tends to result in significant errors, especially for large offsets. The 2.5D MF method can perform automatically with a coherence-based global optimization search on data. We investigated the accuracy of the new formulation by testing it on different synthetic models and a real seismic data set. Applying our approach to the real data led to a high-resolution seismic image with a significant quality improvement compared to the conventional method. Numerical tests indicate that the new formula can accurately focus the primary reflections at their correct location, remove anomalous dip-dependent velocities, and extract true dips from seismic data for structural interpretation. Our method efficiently projects and extracts valuable 3D structural information when applied to crooked-line seismic surveys.
- 28 July 2021
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 86 (6) , S355-S369
2.5D multifocusing imaging of crooked-line seismic surveys
Two-dimensional seismic surveys often are conducted along crooked-line traverses due to the inaccessibility of rugged terrains, logistical and environmental restrictions, and budget limitations. The crookedness of line traverses, irregular topography, and complex subsurface geology with steeply dipping and curved interfaces could adversely affect the signal-to-noise-ratio of the data. The crooked-line geometry violates the assumption of a straight-line survey that is a basic principle behind the 2D multifocusing (MF) method and leads to crossline spread of midpoints. In addition, the crooked-line geometry can give rise to potential pitfalls and artifacts and thus leads to difficulties in imaging and velocity-depth model estimation. We develop a novel MF algorithm for crooked-line seismic data and revise the traveltime equation accordingly to achieve better signal alignment before stacking. Specifically, we have developed a 2.5D MF reflection traveltime equation that explicitly takes into account the midpoint dispersion and cross-dip effects. The new formulation corrects for normal, inline, and crossline dip moveouts (DMOs) simultaneously, which is significantly more accurate than removing these effects sequentially. Applying normal moveout, DMO, and cross-DMO separately tends to result in significant errors, especially for large offsets. The 2.5D MF method can perform automatically with a coherence-based global optimization search on data. We investigated the accuracy of the new formulation by testing it on different synthetic models and a real seismic data set. Applying our approach to the real data led to a high-resolution seismic image with a significant quality improvement compared to the conventional method. Numerical tests indicate that the new formula can accurately focus the primary reflections at their correct location, remove anomalous dip-dependent velocities, and extract true dips from seismic data for structural interpretation. Our method efficiently projects and extracts valuable 3D structural information when applied to crooked-line seismic surveys.