The algorithm

is described in detail in Appendix A. A second difference in NEMO-SHELF is the use of a non-linear free surface formulation with variable volume (Levier et al., 2007) which is advantageous for this study as it allows to account for the injection of dense water using the model’s river scheme. The ‘river’ injection grid cells are arranged Y 27632 over a 50 m-thick layer above the bottom at 115 m depth in a 3 km-wide ring around a central ‘island’ of land grid cells (Fig. 2(a)). The island’s vertical walls avoid a singularity effect at the centre of rotation and prevent inflowing water from sloshing over the cone tip. A constant flow rate Q   (in m3s-1) of water at a given salinity S is evenly distributed over all injection grid cells. The inflowing water is marked with a passive tracer ‘PTRC’ (using the MYTRC/TOP module) by continually resetting the PTRC concentration to 1.0 at the injection grid cells. Thirdly, NEMO-SHELF includes the Generic Length Scale (GLS) turbulence model (Umlauf and Burchard, 2003) which we this website use in its k-∊ configuration with

parameters from Warner et al., 2005 and Holt and Umlauf, 2008. The scheme’s realistic vertical diffusivity and viscosity coefficients give confidence to the accurate representation of the frictional Ekman layer within the plume. The advection scheme in the vertical is the Piecewise Parabolic Method (vPPM, by Liu and Holt (2010)).

The high precision Pressure Jacobian Bcl-w scheme with Cubic polynomial fits which is particularly suited to the s-coordinate system is used as the horizontal pressure gradient algorithm (kindly made available by H. Liu and J. Holt, NOCL). For the parametrisation of the subgrid-scale horizontal diffusion of tracers and momentum we use the Laplacian (harmonic) operator with constant diffusivity coefficients ( Aht=Ahm=3.0m2s-1 for tracers and momentum respectively). Care is taken to separate the large lateral diffusion from the tiny diffusion in the diapycnal direction (see Griffies, 2004, for a discussion) by activating the rotated Laplacian operator scheme. For this study we modify the calculation of the slope of rotation to blend the slope of isopycnal surfaces with the slope of surfaces of constant geopotential depending on the intensity of the background stratification. This approach, which is described in detail in Appendix B, was especially devised for our ambient conditions where the calculation of isopycnal surfaces within a well-mixed ambient layer may lead to unphysical slope angles that cause lateral diffusion to ‘leak’ into the sensitive vertical diffusion.