Each iteration a of full-waveform inversion requires the migration of the difference between the real data and the new data created from an updated model. The migration process has typically used the reverse-time algorithm, though alternative algorithms are now being used. The reflectivity estimate from a migration may use a cross-correlation with forward modelled data and will contain artifacts that have a very low frequency and bias the reflectivity. The cause of these low frequency artifacts is identified and evaluated using reverse-time and downward-continuation wavefield propagation of energy using a wavelet on a one dimensional model. The model contains varying velocities that produce multiples that are displayed with a two dimensional array in space and time. The wavefields are propagated using finite difference and phase-shift algorithms, with various initial and boundary conditions. The resulting cross-correlations are then processed to evaluate their potential for representing the reflectivity of the model.
View full article as PDF (1.99 Mb)