The SEAM Phase I Subsalt Earth Model, which is a 3D representation of a deep water Gulf of Mexico salt domain with its high geological complicity—such as realistic faults, overturned beds, overhanging salts—is thus widely applied to validate the imaging and velocity model building (VMB) techniques.
As a method subsequently used to provide subsurface velocity model, our new, anisotropic full waveform inversion (AFWI) that combines TV regularization and automated salt-flooding is capable of inverting complex structures with salts. The 3D SEAM model is used to validate the new TTI AFWI, the salt structures are well inverted by TV regularization and salt flooding in the first few iterations with isotropic FWI, and then the anisotropic parameters are added as input to invert more detailed structures. The inverted model is thus used to provide the velocity model for seismic imaging.
Alternatively, a more time-saving artificial intelligence VMB (AI-VMB), which involves a hybrid physics-guided neural network (H-PGNN) that involves both data-driven, deep learning, and physics-based waveform modelling is proposed. We further expanded it into 3D and applied the SEAM model to validate the new velocity model building technique. In the real data application, the 3D SEAM is further cropped into small velocity cubes as additional training samples. Therefore, a more practical synthetic training set is thus applied in the real data validation.
The SEAM Phase I Subsalt Earth Model, which is a 3D representation of a deep water Gulf of Mexico salt domain with its high geological complicity—such as realistic faults, overturned beds, overhanging salts—is thus widely applied to validate the imaging and velocity model building (VMB) techniques.
As a method subsequently used to provide subsurface velocity model, our new, anisotropic full waveform inversion (AFWI) that combines TV regularization and automated salt-flooding is capable of inverting complex structures with salts. The 3D SEAM model is used to validate the new TTI AFWI, the salt structures are well inverted by TV regularization and salt flooding in the first few iterations with isotropic FWI, and then the anisotropic parameters are added as input to invert more detailed structures. The inverted model is thus used to provide the velocity model for seismic imaging.
Alternatively, a more time-saving artificial intelligence VMB (AI-VMB), which involves a hybrid physics-guided neural network (H-PGNN) that involves both data-driven, deep learning, and physics-based waveform modelling is proposed. We further expanded it into 3D and applied the SEAM model to validate the new velocity model building technique. In the real data application, the 3D SEAM is further cropped into small velocity cubes as additional training samples. Therefore, a more practical synthetic training set is thus applied in the real data validation.
The updated velocity model either from A-FWI or AIVMB can thus be applied into the 3D anisotropic reverse time migration (RTM), which is also validated by the SEAM 3D model.
Speaker Bio
Junxiao Li
Junxiao Li obtained his PhD in geophysics from the CREWES projects in University of Calgary in 2017. During his graduate studies, Junxiao did internships with ADNOC, PETRONAS, and Geotech Beijing. After graduation, he did one year postdoc project funded by Canada First Research Excellence Fund (CFREF) and then joined in R&D at PETRONAS research Sdn Bhd in Kuala Lumpur. Junxiao’s research focus is mostly on full waveform inversion, seismic forward modeling, and deep learning geophysics. He has authored two patents and more than 30 technical publications.