Reflection FWI with an augmented wave equation and quasi-Newton adaptive gradient scheme

Conventional Full Waveform Inversion (FWI) produces high-resolution model updates from diving wave energy in recorded seismic data. However, the penetration depth of refraction energy is limited by the maximum recorded offset and the Earth’s velocity field, so we rely on reflected energy to update the deeper parts of the model. The inclusion of reflections in FWI requires special care due to the relatively weak amplitude of the “rabbit ear” tomographic term. We propose a novel reflection FWI (R-FWI) simultaneous inversion for velocity and derivative quantities which may include properties related to amplitude variation with angle of incidence (AVA). Such multi-parameter inversions face difficulties with parameter scaling, Hessian estimation and slow convergence rates. Therefore, we also propose an efficient adaptive gradient extension to a conventional quasi-Newton optimization scheme to complement our multi-parameter approach. We demonstrate these new approaches on data from the North-West shelf of Australia.