Two‐Layer Anisotropy Beneath Subduction Zones: Bayesian Inversion

Shear‐wave splitting measurements have the potential to constrain multiple layers of anisotropy and thereby enhance depth resolution. Using the formulation of Silver and Savage (1994, https://doi.org/10.1111/j.1365‐246x.1994.tb04027.x ), previous studies have employed deterministic grid‐search approaches to identify best‐fitting two‐layer models. In this study, we adopt a Bayesian inversion that incorporates a factor accounting for uncertainties in shear‐wave polarization orientations to resolve two‐layer anisotropic structures. We demonstrate this inversion using source‐side S and SKS splitting data from the Cocos subduction zone and then apply it to the South American and Hikurangi subduction zones. In the Cocos system, despite strong parameter tradeoff, our model resolves two layers without relying on a priori tectonic arguments, in contrast to previous deterministic inversions. The fast direction of the upper layer is perpendicular to the trench or parallel to the Cocos plate motion predicted by the no‐net‐rotation, rather than the hotspot, reference frame. The lower‐layer anisotropy is trench‐oblique, consistent with previous suggestions. In the other two subduction zones, our approach tightly constrains the upper‐layer anisotropy as trench‐subparallel and the lower‐layer anisotropy as trench‐normal, deviating from previously reported best‐fitting models. A three‐layer inversion suggests a transitional layer between the two layers for all subduction zones, but such models are not warranted by the data. The layering order in the South America and Hikurangi subduction zones is opposite to that in the Cocos and the poloidal–toroidal dual structure in the subslab mantle predicted by dynamic modeling, implying possible contributions from the slab and the overriding plate to the upper layer.