3D Transdimensional Bayesian tomography of the Earth’s inner core

Abstract:

The inner core has been observed to contain regional heterogeneity in both isotropic and anisotropic seismic velocities, extending from the surface to the centre of the inner core. The major features include hemispherical variations, whereby the western hemisphere has a lower isotropic velocity than the eastern hemisphere, anisotropy oriented with the fast axis aligned to Earth’s rotational axis, and the presence of an innermost inner core. Further regional scale variations have been detected. Whether these features are truly required by the body wave data, or artefacts of the applied hemispherical and radial parameterisations, is a current topic of debate. Transdimensional Bayesian inversion circumvents this problem since the parameterisation is automatically adapted to information in the observations during inversion.

We present results of seismic tomography of the inner core with a transdimensional Bayesian methodology, applying 3D Voronoi cells to a high quality data set of 5296 PKPbd-df, PKPab-df, and PKPcd-df differential travel times, and 2344 absolute PKIKP travel times. Our dataset has excellent regional coverage, and samples the inner core from the inner core boundary to the centre. By taking a transdimensional Bayesian approach, we calculate the best fitting anisotropic and isotropic velocity structures in addition to their probabilities. This allows us to identify both well and poorly constrained regions of the models. We recover high resolution first order observations including a hemispherical structure, an isotropic outer inner core (OIC) and an increasing anisotropic velocity with depth without having these features present in the initial parameterisation. We further conduct inversions on a subset of our data without South Sandwich Islands (SSI) events; these data are proposed to be unreliable due to contamination from mantle structure. However, the large scale inner core structure remains unaffected, indicating that the SSI data does not significantly alter the general interpretations: only increasing the mean anisotropy in an isolated region.