The impact of a parameterisation of submesoscale mixed layer eddies on mixed layer depths in the NEMO ocean model

A parameterisation scheme for restratification of the mixed layer by submesoscale mixed layer eddies is implemented in the NEMO ocean model. Its impact on the mixed layer depth (MLD) is examined in 30-year integrations of “uncoupled” ocean–ice (GO5) and “coupled” atmosphere–ocean–ice–land (GC2) 1/4° global climate configurations used by the Met Office Hadley Centre. The impact of the scheme on the MLD in GO5 is up to twice as large in subtropical and mid-latitudes when the mixed layer Rossby radius is not limited to guard against CFL-type instabilities and excessively strong volume overturning. Such a limit is not found to be necessary for stable integration of the scheme in NEMO. An alternative form of the scheme is described that approximates the mixed layer Rossby radius as a function only of latitude. This formulation is more generally robust to instability and has a comparatively larger impact on the MLD than the original formulation, but yields qualitatively similar results. The global mean impact of the scheme on the MLD is found to be almost twice as large in 1° and 2° configurations of GO5 as it is in the 1/4° configuration. This is shown to be the result of the scheme overcompensating for the decay in strength of resolved mixed layer density fronts in this model with decreasing horizontal grid resolution. The MLD criterion defining the depth scale of the scheme is shown to affect its global mean impact on the MLD by nearly a factor of 3 in GO5 and GC2, depending on whether the criterion is chosen to capture the actively mixing layer or the well-mixed layer. The parameterisation reduces the magnitude of deep MLD biases while increasing the magnitude of shallow biases. The globally averaged winter MLD bias is reduced from 17% to 9% of climatological values in GO5 but changes from +3% to −4% in GC2. Summer mixed layers are too shallow on average in both configurations and their average magnitude is increased by the parameterisation.