Subduction zone trench migration: Slab driven or overriding-plate-driven?

Subduction zones on Earth, and their associated trenches and hinges, migrate with respect to the overriding plate, as indicated by overriding plate deformation (e.g. backarc extension or backarc shortening), and migrate with respect to hotspot and no-net-rotation reference frames (global or "absolute" reference frames). Three geodynamic models exist that attempt to explain how "absolute" trench migration velocity correlates with overriding plate deformation. In one model, trench and hinge migration result from lateral migration of the slab, with trench retreat causing overriding plate extension, and trench advance causing overriding plate shortening. In the second model, trench migration is forced by the overriding plate, with trenchward motion causing trench retreat and overriding plate shortening, and motion away from the trench causing trench advance and overriding plate extension. In the third model, the trench and subduction hinge are thought to be static, while overriding plate extension/shortening is accommodated by landward/trenchward overriding plate motion. In this paper, the conflicting geodynamic predictions made by these three models are tested by using global kinematic calculations of trench migration velocity (v_{T ⊥}), overriding plate velocity (v_{OP ⊥}) and overriding plate deformation in different global reference frames. The dependence between v_{T ⊥} and overriding plate deformation, as well as v_{OP ⊥} and overriding plate deformation was investigated. Correlation coefficients (R) and confidence limits were calculated using a quantitative approach with overriding plate deformation velocity (v_{OPD ⊥}), using a semi-quantitative approach with an overriding plate strain-classification approach, and using a Spearman rank correlation approach. For the quantitative approach the correlation between v_{T ⊥} and v_{OPD ⊥} is consistently positive (R = 0.33-0.68), where trench retreat corresponds to extension, and statistically significant at 95% confidence level for all but one reference frame. For v_{OP ⊥} and v_{OPD ⊥} the correlation is consistently negative but much less significant (R = -0.29 to -0.17) and statistically not significant at 95% confidence level for all reference frames. These findings indicate that trench migration results predominantly from lateral migration of the slab rather than from overriding plate motion. Such lateral slab migration is most likely driven by the negative buoyancy force of the slab itself, as demonstrated by geodynamic models.