Full-waveform inversion in the frequency-ray parameter domain

Wenyong Pan, Kristopher A. Innanen

Full-waveform inversion (FWI) promises high-resolution estimates of the subsurface model properties by iteratively minimizing the difference between modeled and observed data. Its computational cost remains an obstacle in practical applications, and research is active in developing efficient FWI implementations. We describe an efficient frequency-ray parameter (f-p) domain FWI equipped with linear phase-encoding in this paper. The linear phase-encoding is performed by constructing the super-gathers by summing densely distributed individual shots with linear phase shifts. A slant update strategy with varied ray parameters is proposed to further reduce the computation burden. The proposed strategies can reduce the computation burden significantly but also unfortunately introduce strong cross-talk artifacts. We demonstrate that a partial overlap-frequency strategy is important to suppress these cross-talk artifacts. The frequency-ray parameter domain FWI is implemented with gradient-based methods, quasi-Newton l-BFGS method and a truncated Gauss-Newton method. The f-p domain FWI is then enacted on a Marmousi-II model to demonstrate the effectiveness and efficiency of the combined strategies on reconstructing the velocity model. Different optimization methods with the proposed strategies are examined and compared. The resistivity to noisy data is finally analyzed and discussed.