Waveform inversion schemes for 3D density structure

Abstract:

We develop waveform inversion schemes for density, based on numerical wave propagation, adjoint techniques and various non-seismological constraints to enhance resolution.

Density variations drive convection in the Earth and serve as a discriminator between thermal and compositional heterogeneities. However, classical seismological observables and gravity provide only weak constraints, with strong trade-offs. To put additional constraints on density structure, we develop waveform inversion schemes that exploit the seismic waveform itself for the benefit of improved density resolution.

Our inversion scheme is intended to incorporate any information that can help to constrain 3D density structure. This includes non-seismological information, such as gravity and the geoid, the mass and moment of inertia of the Earth, and mineral physical constraints on maximum density heterogeneities (assuming reasonable variations in temperature and composition).

In a series of initial synthetic experiments, we aim to construct efficient optimisation schemes that allow us to assimilate all the available types of information. For this, we use 2D numerical wave propagation combined with adjoint techniques for the computation of sensitivity kernels. With these kernels, we drive gradient-based optimisation schemes that incorporate our non-seismological constraints. Specifically, we assess the usefulness of an inversion strategy where additional information is used as hard constraints, as opposed to the optimisation of a single objective functional that incorporates all the information. Hard constraints may consist of the Earth's mass or moment of inertia, and are applied by solving a separate optimisation problem to project the initial (unconstrained) solution onto an allowed range.

These synthetic experiments will allow us to assess to what extent velocity and density structure need to be coupled in order to obtain useful and meaningful results to a density inversion.