Particle swarm optimization for Rayleigh wave frequency-velocity spectrum inversion

Rayleigh wave exploration is capable of reconstructing seismic properties at high resolution. The dispersion curve inversion methods are mainly based on 1D layered -model assumptions and require picking the dispersion curves manually. The success is highly dependent on personal subjectivity. Full waveform inversion (FWI) has received particular attention in recent years due to the fact that it has no need for manual dispersion curve extraction and no restriction on medium distribution. However, there are still many challenges limiting the application of FWI in field data, e.g., cycle -skipping issues, dependence on the initial model, and difficulty in estimating the seismic source signature. We adapt a particle swarm optimization (PSO) algorithm for Rayleigh wave frequency -velocity spectrum (FVS) inversion. Compared with conventional dispersion curve inversion, we directly fit the FVS without manually extracting the dispersion curves. Compared with FWI, we avoid cycleskipping issues by transforming the observed data from the space-time domain to the frequency -velocity domain. We reduce the dependence on the initial model by adopting a global optimal inversion strategy, and improve the inversion efficiency by introducing GPU parallel computing. In addition, we find that the seismic source dependence of FVS inversion is much less than that of FWI. Both synthetic and field examples verify the effectiveness of the method, making it a valuable tool for retrieving the subsurface structure.