Kinematics and flow patterns in deep mantle and upper mantle subduction models: Influence of the mantle depth and slab to mantle viscosity ratio

Three-dimensional fluid dynamic laboratory simulations are presented that investigate the subduction process in two mantle models, an upper mantle model and a deep mantle model, and for various subducting plate/mantle viscosity ratios (eta(SP)/eta(M) = 59-1375). The models investigate the mantle flow field, geometrical evolution of the slab, sinking kinematics, and relative contributions of subducting plate motion and trench migration to the total rate of subduction. All models show that the subducting plate is always moving trenchward resulting from slab pull. Furthermore, all deep mantle models show trench retreat, as do upper mantle models in the initial stage of subduction before slab tip-transition zone interaction. Upper \mantle models with a low (eta(SP)/eta(M) (66-217) continue to show trench retreat after interaction. Upper mantle models with a high eta(SP)/eta(M) (378, 709) show a period of trench advance after interaction followed by trench retreat. Upper mantle models with a very high eta(SP)/eta(M) (1375) show continued trench advance after interaction. The difference in trench migration behavior and associated slab geometries is attributed to both eta(SP)/eta(M) and the mantle depth to plate thickness ratio T-M/T-SP, which both affect the slab bending radius to mantle thickness ratio r(B)/T-M. Four subduction regimes can be defined: Regime I with r(B)/T-M <= similar to 0.3, trench retreat, slab draping, and a concave trench; Regime II with similar to 0.3 < r(B)/T-M < similar to 0.5, episodic trench migration, slab folding, and a concave trench; Regime III with r(B)/T-M approximate to 0.5, trench advance, slab rollover geometries, and minor trench curvature; and Regime IV with r(B)/T-M >= similar to 0.8, trench retreat, slab draping, and a rectilinear trench. In all models, slab-parallel downdip motion induces poloidal mantle flow structures. In addition, trench retreat and rollback motion of the slab induce quasi-toroidal return flow around the lateral slab edges toward the mantle wedge. Rollback-induced poloidal flow around the slab tip is not observed in any of the experiments. Finally, comparison between the slab geometries observed in the upper mantle models and slab geometries observed in nature imply that the effective viscosity ratio between slab and ambient upper mantle in nature is less than 10 3 and of the order 1-7 x 10(2) , with a best estimate of 1 - 3 x 10(2).