Inversão da forma de onda acústica completa usando Monte Carlo Hamiltoniano

Abstract

Full Waveform Inversion (FWI) is a high-resolution technique used to estimate a model that describes the physical structure of Earth’s interior considering the entire seismic wavefield. The proper solution is given in terms of an ensemble of plausible models due to the ill posedness of the problem. Hamiltonian Monte Carlo (HMC) method has acquired great attention in solving seismic inversion problems due to its ability of sampling on high-dimensional parameter spaces by avoiding the random walk behavior. In this work, we verify the feasibility of using the HMC method on the FWI problem under the acoustic wave approximation. We consider as laboratory two synthetic velocity models under different data variance scenarios. The results suggest that the inversion process is strongly dependent on data variance and on the choice of mass matrix, implying that more efforts are necessary for real data applications.