Present-day stress field in subduction zones. Insights from 3D viscoelastic models and data

3D viscoelastic FE models were performed to investigate the impact of geometry and kinematics on the lithospheric stress in convergent margins. Generic geometries were designed in order to resemble natural subduction. Our model predictions mirror the results of previous 2D models concerning the effects of lithosphere-mantle relative flow on stress regimes, and allow a better understanding of the lateral variability of the stress field. In particular, in both upper and lower plates, stress axes orientations depend on the adopted geometry and axes rotations occur following the trench shape. Generally stress axes are oriented perpendicular or parallel to the trench, with the exception of the slab lateral tips where rotations occur. Overall compression results in the upper plate when convergence rate is faster than mantle flow rate, suggesting a major role for convergence. In the slab, along-strike tension occurs at intermediate and deeper depths (> 100 km) in case of mantle flow sustaining the sinking lithosphere and slab convex geometry facing mantle flow or in case of opposing mantle flow and slab concave geometry facing mantle flow. Along-strike compression is predicted in case of sustaining mantle flow and concave slabs or in case of opposing mantle flow and convex slabs. The slab stress field is thus controlled by the direction of impact of mantle flow onto the slab and by slab longitudinal curvature. Slab pull produces not only tension in the bending region of subducted plate but also compression where upper and lower plates are coupled. A qualitative comparison between results and data in selected subductions indicates good match for South America, Mariana and Tonga–Kermadec subductions. Discrepancies, as for Sumatra–Java, emerge due to missing geometric (e.g., occurrence of fault systems and local changes in the orientation of plate boundaries) and rheological (e.g., plasticity associated with slab bending, anisotropy) complexities in the models.