Toward robust multicomponent FWI on land data: handling topography and data conditioning

Raul Cova, Bernard K. Law, Kristopher A. Innanen

Successful full-waveform inversion (FWI) studies using multicomponent marine data have been often reported in the literature. However, FWI applications to multicomponent land data remain limited. Among the challenges for a successful FWI in this setting we can find source repeatability, receiver coupling, rough topography, near-surface heterogeneities and strong elastic and attenuation effects. Due to the difficulty of including all these effects during the inversion process, it is a common practise to minimize their imprint on the data by conditioning the data before the inversion. Here we present a framework to study the effect of this conditioning on the FWI output. We first address the convenience of using a finite difference algorithm to include the topography during the forward modelling and its implications on the modelling of surface-waves. We compared this with the output of a spectral element method (SEM). The latter one showed a more accurate modelling of the surface-wave dispersion profile expected from the data. Also, less backscattered energy resulting from the discretization of the model was obtained. A benchmark dataset was also created to understand the effect of the data conditioning processes on the FWI output. Well log data available along the Hussar 2D-3C seismic line was used to build Vp, Vs and density models. A near-surface Vp model derived from first arrivals tomography was used to include velocity changes in the near-surface. Then, surface-consistent short-wavelength static corrections, amplitude corrections, and deconvolution operators, for the vertical and horizontal components, were derived from the real data and their inverse applied to the synthetic data. The goal was to obtain a synthetic dataset that would include some of the effects that are observed in the real multicomponent land data. This dataset will be used for understanding to what extent the conditioning of the data affects the inversion output and what strategies can be used to minimize their imprint in the inversion process.