Analysis of shear-wave splitting to infer the seismic anisotropy of the lithosphere-asthenosphere system – inversion ambiguities, automatization, and machine-learning approaches

Seismic anisotropy provides a unique link between directly observable surface structures and the more elusive dynamic processes in the mantle below. The ability to infer the vertically- and laterally-varying anisotropic structures is of great significance for the geodynamic interpretation of surface-recorded waveform effects.

In the first part of this presentation, we assess the capabilities of different observables for the inversion XKS phases to uniquely resolve the anisotropic structure of the upper mantle. For this purpose, we perform full-waveform calculations for simple models of upper-mantle anisotropy. In addition to waveforms, we consider the effects on apparent splitting parameters and splitting intensity. The results show that, generally, it is not possible to fully constrain the anisotropic parameters of a given model, even if complete waveforms are considered. We also discuss advantages and disadvantages of using the different observables.

Recent technological advances have prompted implementations of large-scale seismic experiments producing huge amounts of seismic data. Standard processing procedures, thus, require automatization to facilitate fast and objective data processing. This also applies to the analysis of shear-wave splitting. A recent extension of the SplitRacer software code allows for an automatization of the analysis by choosing a time window based on spectral analyses and by categorization of results based on different splitting methods.

Finally, we will present new results from the application of Neural Networks to the analysis of shear-wave splitting. Our initial approach involves training based on synthetic data and deconvolution of the real waveforms. Current limitations and possibilities for extension will be discussed.