A multi-model study of the restratification phase in an idealized convection basin

International audienceThe representation of baroclinic instability in numerical models depends strongly upon the model physics and significant differences may be found depending on the vertical discretization of the governing dynamical equations. This dependency is explored in the context of the restratification of an idealized convective basin with no external forcing. A comparison is made between an isopycnic model including a mixed layer (the Miami Isopycnic Coordinate Ocean Model, MICOM), its adiabatic version (MICOM-ADIAB) in which the mixed layer physics are removed and the convective layer is described by a deep adiabatic layer outcropping at the surface instead of a thick dense mixed layer, and a z-coordinate model (OPA model).In the absence of a buoyancy source at the surface, the mixed layer geometry in MICOM prevents almost any retreat of this layer. As a result, lateral heat exchanges in the upper layers are limited while mass transfers across the outer boundary of the deep convective mixed layer result in an unrealistic outward spreading of this layer. Such a widespread deep mixed layer maintains a low level of baroclinic instability, and therefore limits lateral heat exchanges in the upper layers over most of the model domain. The behavior of the adiabatic isopycnic model and z-coordinate model is by far more satisfactory although contrasted features can be observed between the two simulations. In MICOM-ADIAB, the more baroclinic dynamics introduce a stronger contrast between the surface and the dense waters in the eddy kinetic energy and heat flux distributions. Better preservation of the density contrasts around the dense water patch maintains more persistent baroclinic instability, essentially associated with the process of dense water spreading. The OPA simulation shows a faster growth of the eddy kinetic energy in the early stages of the restratification which is attributed to more efficient baroclinic instability and leads to the most rapid buoyancy restoring in the convective area among the three simulations. Dense water spreading and warm surface capping occur on fairly similar time scales in MICOM-ADIAB although the former is more persistent that the latter. In this model, heat is mainly transported by anticyclonic eddies in the dense layer while both cyclonic and anticyclonic eddies are involved in the upper layers. In OPA, heat is mainly brought into the convective zone through the export of cold water trapped in cyclonic eddies with a strong barotropic structure. Probably the most interesting difference between the z-coordinate and the adiabatic isopycnic model is found in the temperature distribution ultimately produced by the restratification process. OPA generates a spurious volume of intermediate water which is not seen in MICOM-ADIAB where the volume of the dense water is preserved