Cross-gradient regularization for multiparameter FWI of physically modeled data

Scott Keating, Kristopher A. Innanen, Joe Wong

In 2022 an in-plane, surround acquisition ultrasonic survey was acquired in the physical modelling lab. Buzzer sources, operating at the low end of the ultrasonic spectrum were employed. These data were inverted using a variable density acoustic, frequency domain full waveform inversion. Results demonstrated relatively robust capability to characterize the longer wavelength features of target cross-sections. Shorter wavelength features were more difficult to recover. This effect is likely linked to the lack of complete reflection measurements in the data, which may have impeded accurate characterization of reflection behaviour. A cross-gradient regularization was able to enforce geometric similarity between the density and velocity models recovered by the inversion.