3-D time-domain VTI viscoacoustic full waveform inversion: Application to Valhall field data

This work presents a first 3D field data application to the OBC Valhall dataset of a recently developed vertical transverse isotropic (VTI) viscoacoustic full waveform inversion (FWI) engine, incorporating attenuation as a passive parameter. Our implementation uses a checkpointing-assisted reverse-forward simulation (CARFS) algorithm to efficiently build the FWI gradient, appearing more efficient than standard checkpointing techniques. The reverse propagation of the incident field is performed stably in attenuating medium with low memory cost, thanks to the decimation and interpolation strategy. We follow a frequency-content continuation workflow: starting from low frequencies, each FWI takes the refined velocity model and re-estimates a new source wavelet at given frequency band, combined with a random shots subsampling. The inverted velocity model clearly captures channel features and gas cloud, in agreement with former studies (Sirgue 2010, Operto 2015}. Estimated source wavelets using this final inverted model show good consistency over shots. The synthetic data can match the observed seismograms quite well with an anisotropic visco-acoustic forward problem engine. This encouraging single-parameter inversion for a real case motivates multiparameter inversion in the time domain while keeping cross-talk effects small.