Anisotropy and XKS splitting from geodynamic models of double subduction: testing the limits of interpretation

The analysis of the splitting signature of XKS phases is crucial for constraining seismic anisotropy patterns, especially in complex subduction settings such as outward-dipping double subduction. A natural example of this is found in the Central Mediterranean, where the Apennine and the Dinaride slabs subduct in opposite directions, with the Adriatic plate separating them. To assess the capability of XKS-splitting analysis in revealing anisotropic seismic properties, such as fast polarization directions and shear wave anisotropy (in per cent), we use three-dimensional numerical geodynamic models combined with texture evolution simulations. In these models, two identical outward-dipping oceanic plates are separated by a continental plate. Using the full elastic tensors - directly derived from the texture evolution simulations - we compute anisotropic seismic properties and synthetic teleseismic waveforms. From these waveforms synthetic observables are determined, including apparent splitting parameters (fast polarization directions and delay times) and splitting intensities. Based on these observables, we (1) derive models for a single anisotropic layer (one-layer model), (2) identify regions with significant depth-dependent anisotropic seismic properties, and (3) perform inversions at selected locations in terms of two anisotropic layers (two-layer model).We consider two geodynamic models: one with a strong (M1) and one with a weak (M2) continental plate. Model M1 exhibits significant retreat of the subducting plates with no horizontal stretching of the continental plate, whereas Model M2 shows less retreat, substantial horizontal stretching, and detachment of the subducting plates. These different subduction styles result in distinct flow and deformation patterns in the upper mantle, which are reflected in the anisotropic seismic properties. In Model M1, the fast polarization directions below the continental plate are predominantly trench-parallel, whereas in Model M2, they are mostly trench-normal. In most regions of both models, the one-layer models are sufficient to resolve the anisotropic seismic properties, as these properties are nearly constant with depth. However, for both models, we identify some isolated regions - primarily near the tips of the subducting plates and beneath the continental plate - where fast polarization directions exhibit significant variations with depth. Inverting the apparent splitting parameters in these regions yields multiple two-layer models at each location that excellently fit the observables. However, their anisotropic seismic properties can vary significantly, and not all these two-layer models adequately approximate the true depth variations. This ambiguity can be partially reduced by selecting two-layer models in which the summed shear wave anisotropy closely matches that of one of the one-layer models, as these models better capture the true variations.