For more than five decades, the Mediterranean Sea has been identified as a region of so-called thermohaline circulation, namely, of basin-scale overturning driven by surface heat and freshwater exchanges. The commonly accepted view is that of an interaction of zonal and meridional conveyor belts that sink at intermediate or deep convection sites. However, the connection between convection and sinking in the overturning circulation remains unclear. Here we use a multidecadal eddy-permitting numerical simulation and glider transport measurements to diagnose the location and physical drivers of this sinking. We find that most of the net sinking occurs within 50 km of the boundary, away from open sea convection sites. Vorticity dynamics provides the physical rationale for this sinking near topography: only dissipation at the boundary is able to balance the vortex stretching induced by any net sinking, which is hence prevented in the open ocean. These findings corroborate previous idealized studies and conceptually replace the historical offshore conveyor belts by boundary sinking rings. They challenge the respective roles of convection and sinking in shaping the oceanic overturning circulation and confirm the key role of boundary currents in ventilating the interior ocean.@en

The Atlantic Meridional Overturning Circulation (AMOC) plays an important role in climate. The classical view of an ocean conveyor belt with northward surface currents and southward return currents transporting convectively-formed waters from the subpolar North Atlantic Ocean to other ocean basins suggests a tight relation between convection and sinking. However, convection regions feature very little vertical mass transport. Instead, it has been argued that the sinking of waters must take place near boundaries where ageostrophic processes affect the flow. So far, this has been confirmed in highly idealized regional model studies and in laboratory experiments. It is, however, unclear how well the sinking of dense waters is represented in the current generation of global ocean models and climate models, and whether the factors driving and controlling the sinking in these models are in accordance with the developed theory. This is of crucial importance for our confidence in projections of the future behavior of the AMOC, which are based on this type of model. In this study, we analyze the characteristics of the vertical transport in two global ocean models: an eddy-permitting model at 0.25 degree resolution and its coarser 1.0 degree resolution counterpart. We show that the sinking indeed predominantly occurs in a narrow region close to the boundary in both model simulations, and not in deep convection regions. Notably, the amount of vertical transport that is found along the edges of the North Atlantic Ocean is highly variable in space, and large differences exist between the two model versions. In the eddy-permitting model, the magnitude of the local sinking appears to be governed by alongshore variations in density near the boundary, in line with theory@en