Résumé
Plate kinematics in the vicinity of subduction zones, as well as seismic tomography provide insights into the deep dynamics of subducting slabs. Velocities at which subducting plates are consumed at the trench (the subduction velocities) typically exceed 3-4 cm/yr at present-day. Absolute trench velocities (relative to a lower-mantle reference frame) are lower, between -2 and 2 cm/yr [Heuret and Lallemand, 2005]. This implies that the "accommodation space" created by the slab rollback associated with lateral trench migration is not nearly sufficient for accommodating the length of incoming slab in the horizontal dimension. In the vertical dimension, even the fastest estimates for slab sinking rates over long time scales amount to only a fraction of 3-4 cm/yr [Butterworth et al. 2014, van der Meer et al. 2010, Sigloch & Mihalynuk 2013]. Hence the rates at which the lithosphere typically subducts cannot be accommodated by fast vertical sinking either. Seismic tomography confirms the "traffic jam" conditions for slabs in the mantle that are implied by these numbers, with slab thickening imaged in and beneath the mantle transition zone (MTZ). These highly visible, thickened, slabs have been interpreted as the result of folding [Ribe et al., 2007], and their relative localization (massive, near-vertical "slab walls") supports the notion of near-stationary trenches over long time scales [Sigloch and Mihalynuk, 2013]. Buoyancy-driven analog and numerical models of subduction have commonly produced subduction and trench velocities that differ from the first-order observations above. Their subduction velocities typically drop below 1-2 cm/yr once the modelled slab enters the high-viscosity lower mantle, and their trench migration velocities remain almost equal to subduction velocities, thus accommodating the slab mainly in the horizontal direction. In addition, these models tend to produce trench retreat and slab "rollback", unless the latter is very weak and/or the overriding plate is very strong [Goes et al., 2017]. These modelling results have led to the conclusion that near-vertical slab sinking and folding at the MTZ is an end-member regime restricted to very specific subduction set-ups. We have added a weak asthenospheric layer to typical 2-D thermo-mechanical models of subduction zones with a complex rheology [e. g., Garel et al., 2014], which partly reconciles the models and the observations. A weak asthenosphere appears as an intuitive candidate for increasing subduction velocity because a reduced mantle drag at the base of the subducting plate lowers the mantle's resistance to the plate's trench-ward motion. We further found that the models with a weak asthenospheric layer lessens the trench motion and thus tend to produce prominent vertical folding of slabs at the MTZ. Subduction velocities remain higher than trench velocities long after the slab reaches the MTZ, so that 300-to-400-km wide "slab walls" are continuously produced in the lower mantle over a relatively wide range of model parameters. The presence of a weak asthenosphere has often been speculated to explain seismic properties beneath oceanic plates, but seldom modelled. This study contributes to a quantification of its potential effects on subduction dynamics. [Copyright Author(s) 2022. CC Attribution 4.0 License: https://creativecommons.org/licenses/by/4.0/legalcode]