Subduction of cold oceanic plates is a key driver of plate tectonics on Earth and has a wide range of surface expressions. The mechanisms governing subduction diversity are still debated. In this thesis, I investigate several key mechanisms using observations and numerical models of single plate subduction. First, I compile global subduction parameters along trenches from the last 120 Myr using recent global plate reconstruction models. The compilation highlights that trench motions generally contribute less than 20% to plate convergence, which has been attributed to low drag force at the base of the subducting plate. I investigate the role of basal drag using 2-D models of long and short plates including composite rheology. I show that even if basal drag force is low, it is expected to modify plate velocity significantly and that, in the Cenozoic record, its effect on velocity is masked because large-area plates experiencing significant basal drag have also been the ones with the highest driving slab pull. In particular for major subduction zones, some observed variability correlates with age-dependent variations of buoyancy and strength of the downgoing plate. Using a set of mechanical 3-D models, I show that to study Earth-like 'sinking' modes of subduction, the representation of plate buoyancy and strength needs to be carefully chosen. Next, I use these 3-D models to investigate the effect of lateral variations in plate buoyancy. Depending on the position of a buoyant ridge and whether the underlying plate is relatively old or young, the subduction of such a ridge modifies trench geometry and can either reduce or increase the shallow dip angle of the slab. These factors can explain why the impact of buoyant features is often different in West and East Pacific subduction zones. Thus variable plate strength and buoyancy and mantle drag significantly contribute to subduction diversity.