Seismic anisotropy inside the Earth from a model space search approach

Abstract

For theoretical and computational convenience, the Earth has long been modelled as an isotropic medium for wave propagation. There is, however, considerable evidence of seismic anisotropy at different depths and different scales throughout the Earth, which can provide insight into Earth's dynamic processes. Since computational power has greatly increased during the last decades, it has now become possible to account for seismic anisotropy in tomographic modelling. Current methods to map seismic anisotropy inside the Earth are often based on the inversion of seismological data but these inversions can be highly non-unique, yielding discrepancies among the models produced in different studies. Since detection of seismic anisotropy is of great importance for the understanding of mantle mineralogy and deformation processes that take place inside the Earth, it is necessary to better understand tomographic models and the origin of the differences among them. In this thesis, we aimed to study seismic anisotropy in a more robust and systematic way than in previous studies, using a novel technique that describes better the model space and the complete ensemble of solutions, with reliable uncertainty estimates. The method employed in this thesis offers a way to characterize the model space better, and to give a complete description of resolution, trade-offs and uncertainties on the model parameters. We used the Neighbourhood Algorithm (NA), a direct search approach developed by Sambridge from which robust information on Earth's properties can be obtained without having to introduce unnecessary a priori information on the model space. The Gaussian hypothesis imposed in traditional inverse techniques can give rise to solutions that depend strongly on the starting model in highly underdetermined inverse problems, where the distribution of likely models can be strongly non-Gaussian. With the NA we can deal with ill-posed problems without assuming a priori Gaussian statistics for the model space. Another advantage of the NA is that it gives an overview of all the models compatible with the data, rather than choosing one with some subjective regularisation. In addition, likelihoods and trade-offs are obtained for the various model parameters, which gives a powerful tool to estimate true resolution and uncertainties. The NA was applied to surface wave phase velocity maps and normal mode data to establish the necessity to introduce seismic anisotropy in reference Earth models, and to study lateral variations in radial anisotropy in the upper mantle, transition zone and top of the lower mantle. A study of inner core anisotropy was also performed using recent measurements of anomalously split normal modes. It produced models of inner core anisotropy that remove a longstanding controversy arisen from the use of two different kinds of data, i.e. normal modes and travel-times.