a fast implementation of stochastic 1d elastic seismic full-waveform inversion

We present a fast implementation of a 1d elastic full-waveform inversion for reconstructing the elastic structure of the subsurface. the fwi is an inversion algorithm that directly models the full seismic wavefield by solving a semi-analytical form of the elastic wave equation for a 1d layered earth model known as the reflectivity method. the input seismic data are pre-conditioned angle-gathers. the inversion is done using a stochastic algorithm known as psoes, a fast hybrid stochastic optimization algorithm. our work primarily contributes to accelerating the computation times required for the inversion, with the goal of developing a strategy that enables code implementation at production scales. additionally, we are working on creating a framework for conducting joint inversions with potential field data. the computational cost of the fwi is directly proportional to the number of unknowns in the inversion problem, which correlates with the vertical resolution of model (i.e., the layer thicknesses in the 1d earth model) and the maximum depth of the study. however, the relationship is not linear because increasing the number of unknowns affects the run time of both the forward and inverse problems. on the other hand, the quality of the solution highly depends on the vertical resolution since modeling the higher frequencies in the data requires a relatively small vertical thickness. an optimum implementation of the fwi could result in calculating 1d elastic profiles of the subsurface which could be used for constraining the inversion of potential field data over sedimentary basins.

a fast implementation of stochastic 1d elastic seismic full-waveform inversion

we present a fast implementation of a 1d elastic full-waveform inversion for reconstructing the elastic structure of the subsurface. the fwi is an inversion algorithm that directly models the full seismic wavefield by solving a semi-analytical form of the elastic wave equation for a 1d layered earth model known as the reflectivity method. the input seismic data are pre-conditioned angle-gathers. the inversion is done using a stochastic algorithm known as psoes, a fast hybrid stochastic optimization algorithm. our work primarily contributes to accelerating the computation times required for the inversion, with the goal of developing a strategy that enables code implementation at production scales. additionally, we are working on creating a framework for conducting joint inversions with potential field data. the computational cost of the fwi is directly proportional to the number of unknowns in the inversion problem, which correlates with the vertical resolution of model (i.e., the layer thicknesses in the 1d earth model) and the maximum depth of the study. however, the relationship is not linear because increasing the number of unknowns affects the run time of both the forward and inverse problems. on the other hand, the quality of the solution highly depends on the vertical resolution since modeling the higher frequencies in the data requires a relatively small vertical thickness. an optimum implementation of the fwi could result in calculating 1d elastic profiles of the subsurface which could be used for constraining the inversion of potential field data over sedimentary basins.