3D full-waveform inversion in time-frequency domain: Field data application

Abstract

We present a 3D Gauss-Newton full-waveform inversion (3D GN-FWI) method in the time-frequency domain for detection of subsurface anomalies. The use of Gauss-Newton approach is particularly important for near-surface imaging, in which acquired seismic wavefields are dominated by Rayleigh waves. The inverse Hessian matrix utilized in this approach acts as a weighting function to reduce the dominancy of Rayleigh waves, increase the contribution of body waves, and thus help resolve deeper structures. However, the Gauss-Newton method requires a huge amount of computer memory to store derivative wavefields (Jacobian matrix). To address this issue, the presented 3D GN-FWI method exploits advantages of the combined time-frequency domain. Specifically, the forward wave simulation is done in the time-domain to generate wavefields at multiple frequencies simultaneously without the requirement of an inverse matrix solver, while inversion is conducted in the frequency-domain to significantly reduce the required computer memory. Synthetic and field experimental datasets are used to assess the capability of the presented waveform method. The synthetic result shows that a variable profile with a buried void is well recovered. For the field experiment, a large mobile shaker was used to induce wavefields at individual frequencies with consistent magnitudes required for the presented frequency-domain inversion. The wavefields were recorded with uniform 2D grids of sources and receivers on the ground surface, and analyzed to obtain 3D subsurface wave velocity profiles. The seismic result is consistent with an invasive standard penetration test (SPT), including the identified bedrock depth and buried void.