Résumé
Oceanic plate subduction implies tight interactions between converging lithospheres and surrounding mantle. Plate-mantle couplings can be modeled using thermo-chemical codes of mantle convection. But how to model correctly with a continuous fluid approach the subduction interface, characterised by strong and localised discontinuities? The present study aims at better deciphering the different mechanisms involved in the functioning of the subduction interplate, simply modeled by a weak crust layer, free to evolve. Pseudo-brittle and non-Newtonian behaviours are modelled. This study shows first that the numerical resolution is critical. If the subducting plate is 100 Myr old, subduction occurs for any crust strength. The stiffer the crust is, the shallower the interplate down-dip extent is and the hotter the fore-arc base is. Conversely, imposing a very weak subduction channel leads to an extreme mantle wedge cooling and inhibits mantle melting in wet conditions. If the incoming plate is 20 Myr old, subduction occurs only if the crust is either stiff and denser than the mantle, or weak and buoyant. These conditions lead notably to (1) fore-arc lithosphere cooling, and (2) partial or complete hindrance of wet mantle melting. Finally, subduction plane dynamics is intimately linked to the regime of subduction-induced corner flow: either focussed towards the mantle wedge tip and strongly warming the subduction plate, or, diffuse and favoring global cooling by the lengthening of the subduction plane. The thermal states simulated within the mantle wedge are compared with observations to decipher the best rheological ranges modelling the subduction channel. Two intervals of crustal activation energy are underlined: 345-385 kJ/mol to reproduce the slab surface temperature range inferred from geothermometry, and 415-455 kJ/mol to reproduce the hot mantle wedge core suggested by seismic tomographies. As these two intervals do not overlap, an extra process involved in subduction dynamics is needed. A moderate mantle viscosity reduction, caused by metasomatism in the mantle wedge, is proposed. From these results, it can be inferred that the subduction channel down-dip extent should vary with the subduction setting, consistently with the worldwide variability of sub-arc depths of the subducting plate surface.