A-Team

The Anisotropy Team (A-Team), under the direction of Dr. Ilya Tsvankin, is focused on modeling, inversion, and imaging of seismic reflection and borehole data from anisotropic media. The current A-Team research portfolio emphasizes full-waveform inversion (FWI) for elastic anisotropic models, time-lapse monitoring of CO₂ sequestration and hydrocarbon production, and novel applications of machine-learning techniques in seismology.

Our Team

Ahmed Ahmed

Maksat Jazbay

Ashish Kumar

Ilya Tsvankin

The Problems We Solve

How can we obtain high-resolution seismic images of a fractured formation?

High-resolution orthorhombic velocity models obtained by elastic full-waveform inversion

Estimation of the parameters of azimuthally anisotropic media (e.g., orthorhombic) poses serious challenges, one of which is the large computational cost of processing and inversion of 3D wide-azimuth data. Full-waveform inversion (FWI) of surface seismic data for elastic orthorhombic media also suffers from parameter trade-offs that cannot be overcome without constraining the model-updating procedure. We have developed an FWI methodology that incorporates geologic constraints to reduce the inversion nonlinearity and enhance the resolution of parameter estimation. These constraints are obtained from well logs, which can provide rock-physics relationships for different geologic facies. Because the locations of the available well logs are usually sparse, a machine-learning (ML) algorithm is employed to account for lateral heterogeneity in building the lithologic constraints. The top figure on the right is the P-wave vertical velocity estimated by unconstrained FWI for an orthorhombic model, and on the bottom is the FWI result obtained using facies-based constraints (Singh et al., 2020). The developed algorithm achieves a much higher spatial resolution of several inverted parameters compared to unconstrained FWI, even in the absence of recorded frequencies below 2 Hz.

How can we reconstruct time-lapse parameter variations with a distorted source wavelet?

Source-independent methodology incorporated into the time-lapse FWI workflow for VTI media

Full-waveform inversion (FWI) requires an accurate estimate of the source wavelet. However, reconstruction of the source wavelet from field data is often problematic. The non-repeatability of source signals in time-lapse surveys makes the source-wavelet estimation even more challenging. To address this problem, we incorporate a source-independent methodology into the time-lapse FWI workflow for VTI (transversely isotropic with a vertical symmetry axis) media. The figure shows that our algorithm makes it possible to reconstruct the time-lapse variations in the P-wave vertical velocity even using a substantially distorted source wavelet, and the achieved spatial resolution is close to that obtained with the actual wavelet.

How can we efficiently monitor CO2 injection in an elastic medium?

CNN with multitask and transfer learning

Monitoring time-lapse changes in subsurface reservoirs is crucial for effectively managing oil and gas production and CO2 injection. It is particularly important to accurately estimate fluid movement and time-varying saturation level. Aiming for a more precise estimation of reservoir changes due to injected CO2, we develop a seismic monitoring algorithm based on machine learning. For the baseline survey, the model is obtained from seismic data using a physics-based method, such as full-waveform inversion (FWI). However, FWI is computationally expensive and faces problems related to parameter crosstalk and trade-offs. Also, it is often challenging to estimate density and fluid saturation from reflection seismic data. We take advantage of recent advances in computational sciences and machine learning to solve nonlinear inverse problems that arise in time-lapse monitoring. We leverage the strengths of convolutional neural networks with multitasking and transfer learning to predict the temporal variations in multiple reservoir parameters.

Some Recently Completed Projects

• Time-lapse elastic FWI for transversely isotropic media with a vertical (VTI) and tilted (TTI) symmetry axis
• Seismic monitoring of CO2 sequestration using time-lapse FWI of seismic data combined with convolutional neural networks
• High-resolution seismic reservoir monitoring using multitask and transfer machine learning
• Elastic FWI for orthorhombic models with lithologic constraints incorporated using machine learning
• Application of mimetic finite-difference propagators to FWI of multicomponent ocean-bottom seismic data
• Seismic wave propagation and wavefield decomposition for anisotropic attenuative media