Initial gradient optimization for elastic full-waveform inversion by means of spectral recomposition

Abstract

Predicting velocity models accurately using FWI technique is strictly related to an enhanced initial estimate. A reliable mean to build a more befitted initial model for FWI provides a more accurate velocity model building and/or fewer iterations — and, therefore, a lower processing time — for building the velocity model. In light of these considerations, we propose to use an approach based on spectral recomposition of elastic seismic data and to implement it in order to perform the optimization of the gradient in initial estimations for FWI focusing in S-waves. This approach employs an inversion scheme aimed at reconstructing the seismic spectrum of wavelets associated with a reflection event. This allows us to estimate the temporal position of each wavelet within a seismogram. Once the temporal position of each wavelet is determined, we can guide the calculated wavelet to better fit the corresponding observed signal, starting from a closer initial point. The utilization of this approach results in notably accurate predictions of elastic velocity models with a significant reduced number of iterations, especially for lower frequencies.