Effects of Subgrid-Scale Mixing Parameterizations on Simulated Distributions of Natural 14C, Temperature, and Salinity in a Three-Dimensional Ocean General Circulation Model

Abstract

The effects of parameterizations of subgrid-scale mixing on simulated distributions of natural 14C, temperature, and salinity in a three-dimensional ocean general circulation model are examined. The parameterizations studied are 1) the Gent?McWilliams parameterization of lateral transport of tracers by isopycnal eddies; 2) horizontal mixing; 3) a parameterization of vertical mixing in which the amount of mixing depends on the local vertical density gradient; and 4) prescribed vertical mixing. The authors perform and analyze four ocean GCM simulations that use different combinations of these parameterizations. It is confirmed that the Gent?McWilliams parameterization largely eliminates the tendency of GFDL-based models to overestimate temperatures in the thermocline. However, in the authors? simulations with the Gent?McWilliams parameterization the deep ocean is too cold, in places by more than 3 degrees. Our results are the first known to assess the effects of the Gent?McWilliams parameterization on the simulated distribution of natural 14C. The most important change (compared to results obtained with horizontal mixing) is that interior ocean 14C values are lower; that is, the water is ?older? with Gent?McWilliams. In most locations in the deep North Atlantic, simulated ?14C values are much too low with horizontal mixing and are even lower with Gent?McWilliams. Both this problem and the problem of the simulated deep ocean being too cold are probably due, at least in part, to insufficient downward penetration of NADW, resulting in the deep North Atlantic in the model being ventilated primarily via AABW. This problem exists when Gent?McWilliams is not used, but Gent?McWilliams makes the symptoms it presents (an overly cold and old deep North Atlantic) worse. Gent?McWilliams also results in a dramatic reduction in convective adjustment in the model, compared to results obtained with horizontal mixing; as a result, simulated tracer distributions are improved at high latitudes. Finally, Gent?McWilliams increases the susceptibility of the authors? model to some types of numerical problems. The stability-dependent vertical mixing parameterization causes relatively small changes in simulated distributions of temperature and natural 14C (compared to results with a prescribed uniform vertical diffusivity), but these changes tend to improve agreement with observations. Assuming they are based on correct physical premises and are properly calibrated, both the stability-dependent vertical mixing parameterization and the Gent?McWilliams parameterization should give the model more predictive capability than simpler parameterizations do in that they allow the amount or direction of mixing to change in response to changes in ocean density.