Frequency-domain Elastic Full Waveform Inversion Based On Time-domain Numerical Simulation

Full waveform inversion is such a crucial measure to quantify media parameters and recover the velocity structure in the subsurface,which make full use of the kinematic and dynamic information existing in the prestack seismic data.Owe to the high ability to provide reliable velocity models for prestack depth migration,lithology judgment and oil/gas reservoir identification.It has gradually attracted more and more attention of geophysicists,become a hot topic in the field of geophysical prospecting.In this paper,we first give the brief introduction and analysis of the major steps of time-domain elastic full waveform inversion,including reconstruction strategies of wavefield,gradient calculation and estimation of update quantity.Numerical testing proves conventional elastic full waveform inversion fails in convergence on the global optimal solution accurately,on the condition of the initial model without low frequency information.Elastic full waveform inversion shows a strongly dependence on the initial model.To alleviate this problem,in this paper we extend the envelope inversion method to elastic medium.Moreover,to reduce the instability of the envelope inversion method when the seismic data with higher dominate frequency,we implement the process of time-scale decomposition on seismic data before the envelope inversion.So it self-adaptively achieves a multi-scale inversion strategy starting from ultra-low frequency.The applicability of the inversion strategy is analyzed and compared through a set of numerical experiments.In order to optimize the relationship between storage cost and calculation quantity in time-domain FWI,we provide the hybrid domain inversion strategy by properly combining the high efficiency of time-domain forward modeling with the multi-scale advantage of frequency-domain inversion method.The gradient vectors corresponding to different frequency components are calculated by extracting the corresponding information from time domain wavefield between the extrapolation.Besides,for the improvement of the deep image result,we introduce the frequency domain compensation factor into the gradient precondition step to balance the energy contribution of the gradient field.To further improve the stability of the inversion,envelope inversion results are used as the initial model of the hybrid inversion method.Therefore,a multi-scale strategy in frequency-domain based on time-domain modeling is achieved.We testify to the validity and stability of the proposed theory and program through Marmousi model and Sigsbee2 A model.