Reducing the influence of remnant noise on elastic FWI with misfit modification

Luping Qu, Scott Keating, Kristopher A. Innanen

Seismic data collected in the field are mostly elastic data including body waves, surfacewaves, and unwanted noises from various sources. Compared with acoustic FWI, noisesuppression or estimation in a elastic engine are of more practical significance. Various typesof noise may exert different influences on the multi-parameter estimation in elastic FWI. Inthis study, we analyzed the influence of random and correlated noises on the estimation ofmodel parameters Vp, Vs, and density. To mitigate the influence of the noises, we adoptedthe modified FWI misfit, in which the data covariance matrix is incorporated, to invert the elastic model parameters while estimating the noises in seismic data. The data covariancematrices calculated from the data residuals, which were obtained from a first round ofFWI, were generated and applied in the following iterative inversion. As the elastic FWIwas conducted in the frequency domain, the dimension size of the data covariance matrix iscalculated by each frequency, and is not likely to cause the out-of-memory problem. Two types ofnoises including random noises and correlated noises were added to the true spectra andestimated in the modified FWI. The inversion results were compared with the results of theconventional FWI with the same iteration number.