On the detectability of local, high-velocity anomalies in acoustic crosshole full-waveform inversion

In this study, two-dimensional, time-domain full-waveform inversion (FWI) is applied to crosshole seismic data acquired in the glacially overdeepened Tannwald Basin (ICDP site 5068_1), located north of Lake Constance, to derive a high-resolution P-wave velocity model. The acoustic waves were excited by a sparker source in borehole B and recorded by a 24-station hydrophone string in borehole C, 28 m away. Data preprocessing included extensive tube wave suppression using fk-filtering. Existing traveltime tomography models were used as initial models for the acoustic waveform inversion. Although the two applied workflows yield very similar waveform fits and final misfit values, their outcomes differ: the run using an anisotropic gradient smoothing filter produces a background model with a high vertical resolution, while the isotropic gradient smoothing results in the inversion of local high-velocity anomalies that introduce X-shaped artifacts. A comparison with sonic logs confirms that both models significantly improve the initial model, but also display discrepancies beyond the sonic logs’ error range. We conclude (1) that gradient smoothing is a critical parameter in FWI and should be thoroughly tested, especially when FWI including uncertainty quantification cannot be performed, and (2) that two extremal smoothing realizations facilitate to infer a comprehensive conceptual geological interpretation.