Rapid Three-Dimensional Subsurface Imaging with Data-Driven Full Waveform Inversion

Non-invasive geophysical methods offer the potential to cost-effectively characterize large volumes of the subsurface to inform geotechnical design. Among geophysical methods, three-dimensional (3D) seismic full waveform inversion (FWI) shows strong potential, but is currently too time-consuming for general use. In response, a new method is proposed for pseudo-3D (P-3D) imaging. P-3D imaging combines multiple two-dimensional images of shear wave velocity (Vs), each generated by a pre-trained data-driven model, from seismic data collected using closely spaced geophone arrays to create a 3D image of the subsurface. We evaluate the method on a real dataset, where seven lines of seismic data were used to characterize a 50 m by 30 m area to a depth of 20 m. The entire data collection and P-3D model generation took only three hours to complete by a seven-person team. The predicted P-3D model captures a variable soil-rock interface that is consistent with local geology and independent horizontal-to-vertical spectral ratio measurements.