NS33A-03
Three-dimensional elastic full waveform inversion of land data using an envelope-based misfit function

Wednesday, 16 December 2015: 14:20
3024 (Moscone West)
Dmitry Borisov, Princeton University, Princeton, NJ, United States
Abstract:
Full waveform inversion (FWI) is an efficient data fitting technique used to estimate properties of the Earth (e.g., compressional and/or shear wavespeeds) from seismic data by minimizing the misfit between observed and simulated seismograms. Because of the very high computational cost, this technique has so far been used either with a 2D full elastic formulation or a 3D acoustic formulation applied to active-source surveys to image the shallow (i.e., the top few kilometers) subsurface. However, the Earth is three-dimensional, (visco)elastic and highly heterogeneous. Therefore, obtaining more accurate models requires solving a full 3D elastic wave equation.

In this study we apply elastic FWI to synthetic 3D datasets, generated for realistic land geology with complex shallow structures. The seismic traces recorded on the surface are dominated by high-amplitude and dispersive surface waves. Inversion of these seismograms makes conventional (waveform difference) FWI highly sensitive to the initial model due to increased nonlinear behavior of the misfit function. For such datasets we extend the envelope-based misfit function (Yuan et al. 2015) to 3D geometry while inverting surfaces waves at early stages of the FWI in order to improve the shallow parts of the P and S wavespeeds. We demonstrate that reliable near surface models allow us to better reconstruct deeper parts at later stages, while inverting body waves using a waveform difference misfit function. We also show that 2D inversion of 3D data collected in complex geological settings is not as accurate as full 3D FWI.