OCEAN MODELS
93
size locally at a strait: but this trick is possible only when the strait is
between land points because the grid size δx, δy does not depend on z.
It cannot be used for deep fracture zones in the middle of ocean basins.
The influence of overflow waters is especially important in the North
Atlantic, with North Atlantic Deep water (at depths of 2000-3000 m)
coming over Denmark Straits and the Faroe-Scotland ridge, Mediterranean water (1000 m depth) coming through Gibraltar, and Antarctic
bottom water (4000 m depth) spreading over sills in the mid-Atlantic
ridge (Romanche and Vema fracture zones, for example). The overflows
are very badly represented in z coordinate models with staircase topography, generally leading to excessive mixing (Willebrand et al., 2001).
There are simple models (such as streamtubes) to calculate exchange
between basins in simplified cases, that could be the basis for parameterizations (Price and Yang, 1998). The problem with this method is
that each overflow must be specified at a given grid cell or set of grid
cells, which is quite cumbersome in a world ocean. Another drawback
is that such a parameterization introduces grid scale sources and sinks
that may not be handled well by the numerics.
Modellers look for parameterizations valid everywhere in the domain,
such as the “Bottom boundary layer” (BBL) parameterization (Beckmann and D¨ oscher, 1997). Since this pioneering work, BBL parameterizations have been implemented in many ocean models. However, we
don’t have yet a complete picture of their efficiency, depending on model
characteristics. Dengg et al. (1999) find that a BBL parameterization
induces a large improvement in a 1/3 ◦ model of the Atlantic using a
centered advection scheme. However in the 1/6 ◦ model of the Atlantic
we find that the improvement is modest, although the BBL parameterization is similar and both models use the same isopycnal mixing of
tracers (Fig. 7). What causes the different performance of the BBL
parameterization in those two models is unclear; numerical details may
matter.
The FOAM model was developped from the HadCM3 ocean component at 1.8 ◦ resolution. In the latter, a variant of a diffusive BBL scheme
was implemented that dramatically improved the representation of the
Nordic seas overflows (Gordon et al., 2000). In this scheme, when bottom water at a grid cell is denser than the deeper water colums around,
the algorithm looks for the level of neutral buoyancy of that bottom
water and mixes the dense water into that model level. Lighter water
is then moved up in the water column to replace the dense water. The
behavior of this scheme in the 1/9 ◦ version of FOAM has no yet been
evaluated in detail.
93
size locally at a strait: but this trick is possible only when the strait is
between land points because the grid size δx, δy does not depend on z.
It cannot be used for deep fracture zones in the middle of ocean basins.
The influence of overflow waters is especially important in the North
Atlantic, with North Atlantic Deep water (at depths of 2000-3000 m)
coming over Denmark Straits and the Faroe-Scotland ridge, Mediterranean water (1000 m depth) coming through Gibraltar, and Antarctic
bottom water (4000 m depth) spreading over sills in the mid-Atlantic
ridge (Romanche and Vema fracture zones, for example). The overflows
are very badly represented in z coordinate models with staircase topography, generally leading to excessive mixing (Willebrand et al., 2001).
There are simple models (such as streamtubes) to calculate exchange
between basins in simplified cases, that could be the basis for parameterizations (Price and Yang, 1998). The problem with this method is
that each overflow must be specified at a given grid cell or set of grid
cells, which is quite cumbersome in a world ocean. Another drawback
is that such a parameterization introduces grid scale sources and sinks
that may not be handled well by the numerics.
Modellers look for parameterizations valid everywhere in the domain,
such as the “Bottom boundary layer” (BBL) parameterization (Beckmann and D¨ oscher, 1997). Since this pioneering work, BBL parameterizations have been implemented in many ocean models. However, we
don’t have yet a complete picture of their efficiency, depending on model
characteristics. Dengg et al. (1999) find that a BBL parameterization
induces a large improvement in a 1/3 ◦ model of the Atlantic using a
centered advection scheme. However in the 1/6 ◦ model of the Atlantic
we find that the improvement is modest, although the BBL parameterization is similar and both models use the same isopycnal mixing of
tracers (Fig. 7). What causes the different performance of the BBL
parameterization in those two models is unclear; numerical details may
matter.
The FOAM model was developped from the HadCM3 ocean component at 1.8 ◦ resolution. In the latter, a variant of a diffusive BBL scheme
was implemented that dramatically improved the representation of the
Nordic seas overflows (Gordon et al., 2000). In this scheme, when bottom water at a grid cell is denser than the deeper water colums around,
the algorithm looks for the level of neutral buoyancy of that bottom
water and mixes the dense water into that model level. Lighter water
is then moved up in the water column to replace the dense water. The
behavior of this scheme in the 1/9 ◦ version of FOAM has no yet been
evaluated in detail.