Points per wavelength analysis in global elastic FWI of surface waves: a synthetic case study

In this study, we analyze the effects of different spatial sampling rates on elastic modeling and inversion of seismic waves for near-surface applications. An empirical value of 5 points per wavelength (pxλ) is often used for the P body wave case, however, evidences suggests that this number is too small in the near-surface context, because of the low Vs velocity involved. Elastic data were computed with a finite difference (FD) modeling tool on a 2D synthetic model sampled at 5, 20, and 100 pxλ in which topography was added. Using the seismograms generated with 100 pxλ as observed data, we performed two global elastic FWI experiments employing genetic algorithms and an offset stripping procedure, in which the input data were the seismogram computed with 20 and 5 pxλ. Our results show that, in order to allow a proper surface wave propagation and to obtain an adequate model reconstruction, a higher spatial sampling rate than the canonical 5 pxλ is required, otherwise misleading artifacts can be introduced. Using 20 pxλ allows us to recover a significantly better Vs model at the end of the optimization, at the expense of a 30% increase in the computational cost.