ETD

• attenuation
• full waveform inversion
• standard linear solid
• viscoacoustic

Full Waveform Inversion (FWI) is a high-resolution seismic imaging technique that
exploits the full information content of recorded seismic waveforms to reconstruct subsurface physical parameters. In attenuating media, the accurate recovery of both P-wave velocity vp and the quality factor Qp requires a viscoacoustic forward operator and
a carefully designed inversion strategy. This thesis implements a sequential viscoacoustic FWI workflow based on the Standard Linear Solid rheological model within the Devito finite-difference framework, and evaluates its performance and robustness through
a series of controlled synthetic experiments. The workflow is first validated on a layered
reference model under ideal data conditions, confirming that the sequential strategy
is capable of recovering the large-scale velocity structure with partial attenuation recovery. Robustness experiments are then conducted on the BP gas cloud benchmark
model, where attenuation recovery is limited to the main gas cloud anomaly and largely
unresolved at depth due to the weak sensitivity of the seismic data to Qp contrasts in
low-attenuation regions. The results demonstrate that data noise is the least damaging
perturbation among those tested, though the random case preserves the velocity and
attenuation structure more accurately than the correlated case at equivalent noise level.
Source frequency mismatch proves to be the most damaging perturbation, producing
substantially compromised velocity models and near-complete failure of the attenuation inversion. Source phase errors occupy an intermediate position: small shifts are
absorbed by the inversion, whereas larger shifts introduce persistent artefacts in the
recovered models. The combined amplitude and phase error experiment shows that the
amplitude scaling introduces additional degradation on top of the phase error, further
limiting the Qp recovery by dominating the data residual with an irreducible amplitude component. Across all experiments, Qp recovery is consistently more sensitive to
data and source perturbations than vp recovery, reflecting both the sequential nature
of the inversion strategy and the intrinsically weaker sensitivity of the seismic data to
attenuation contrasts.