Geophysical methods provide structural maps of the Earth’s subsurface and are used throughout exploration and production to guide the development of petroleum prospects. Of the various geophysical methods, seismology generally provides the most highly resolved spatial information. The processing and interpretation of seismic survey data has undergone a number of revolutions since industrial seismology began in the 1920′s: the most recent is the transition to Full Waveform Inversion, or FWI (model-driven data fitting), using numerical optimization methods and computational wave propagation to drive 3D mechanical models towards fitting observed data. The improvement in clarity and resolution gained from inversion can be stunning, often enough that every major oil and gas company and seismic contractor firm has deployed research groups to develop inversion algorithms, software, and workflows. This technology is still experimental, however, and beset by numerous challenges. This workshop will address several of the most important ones, for example:
- Inversion Physics: The physics of seismic waves are complex, encompassing at least anisotropic and anelastic behavior, yet most FWI is based on constant-density acoustics. How much earth structure is missed as a result? What is the importance of going beyond acoustics in FWI, towards realistic wave propagation (attenuation, anisotropy…), and how does one best parametrize and invert multi-mode, anisotropic, and anelastic models? How does one deal with parameter cross-talk in multi-parameter (joint) inversion, often very ill-conditioned problems? What can be done about the orders-of-magnitude cost differential between constant density acoustics and anisotropic viscoelasticity, on top of the “curse of dimensionality?” What numerical techniques deal well with gross scale disparities and nonlinearity in very high dimensions?
- Resolution and Uncertainty: Seismic imaging conventionally resolves structure at a fraction of a wavelength. Reservoir intervals are often smaller than that, and earth structure exists and influences seismic response at far sub-wavelength scales. Inversion yields uncertain results for these and many other reasons, such as parameter cross-talk, and this uncertainty can manifest as unreliable inference of earth structure. How can we understand the resolution of FWI, both in the conventional continuum sense (structural resolution limits as function of wavelength) and the way in which it interacts with model description (sampling, reduction, smoothing…) and encodes sub-wavelength and sub-cell earth structure? What is gained/lost from sparsity constraints of various types?
- Velocity Estimation: Inversion is a fantastically difficult optimization problem: descent algorithms tend to stall because of serendipitous destructive cancellation between predicted and observed data (“cycle skipping”). Many approaches to overcome the cycle skipping problem have been suggested: an incomplete list might include RWI/MBTT, various types of dataset comparison and misfit design (such as Laplace domain), source extensions (AWI, WRI…), medium extensions (DSO of various flavors…), and optimal transport. Are we anywhere near identifying best practices?
- Integrating Non-Seismic Data: Many sources of information about active prospects are nearly always available: structural geology, petrophysics, well logs, flow history, even other geophysical inversions such as passive and active source EM and gravimetry. How should we approach the multi-physics integration of various non-seismic constraints with FWI, combining direct and remote information and inevitably involving different resolution scales?
- Microseismicity and Complex Sources: With market forces dictating increased efficiency in unconventional field operations, microseismic FWI should receive serious attention. What can be gained by inversion of microseismic data for unconventional play imaging, and more generally by inversion of complex spatially extended sources?
This workshop will include a poster session; a request for posters will be sent to registered participants in advance of the workshop.
(King Abdullah Univ. of Science and Technology (KAUST))
(University of Texas at Dallas)
(University of Alberta)
William W. Symes
(Université de Grenoble I (Joseph Fourier))