entrant channel that includes three velocity grid points and four tracer grid points.
The lateral walls and bottom all have insulating boundary conditions. A no-slip condition is applied at the walls, and the bottom is assumed to be impermeable. There is no bottom friction. The model uses the rigid-lid approximation at the surface in order to filter out high-frequency external gravity waves. The imposed surface wind stress, which is based on Hellerman and Rosenstein (1983), has only a zonal component. The zonal wind stress is shown in Figure 2b. The wind stress curl is anticyclonic between 15° and 45° latitude in each hemisphere and cyclonic elsewhere.
The temperature is restored to prescribed values with a linear damping coefficient, with a time constant of 50 days. The reference temperature profile for the restoring boundary condition in Experiment 1 was obtained by averaging the world ocean values for Northern and Southern Hemisphere sea surface temperatures from Levitus (1982). In Experiment 2, the values for the Southern Ocean south of 50°S are lowered slightly in order to investigate the sensitivity of the circulation to the temperature of the sinking water. Both temperature profiles are shown in Figure 2a. In Experiments 1 and 2 the surface salinity is similarly restored to the idealized reference salinity values shown in Figure 2a.
To evaluate the salinity flux, Experiment 3 (mixed boundary conditions) used the equilibrium state of Experiment 1 but allowed it to run for an additional 50 years after the initial 7000-year integration utilized in Experiments 1 and 2. The surface salinity flux is averaged for that time period, and this average value is used to restart the model from the equilibrium state of Experiment 1. (This salinity flux is shown in Figure 9a.) The acceleration techniques of Bryan (1984) are used to speed the convergence of the model. The baroclinic velocity and barotropic vorticity equations are integrated with a time step of 2400 seconds, and the tracer equations are integrated with time steps of 4 days. The vertical eddy viscosity and vertical heat and salt diffusivity are set to 1.0 cm2/s everywhere. Horizontal eddy viscosity is set to 6 × 109 cm2/s, and horizontal tracer diffusivity is set to 1 × 107 cm2/s. The method of complete convection, the asymptotic limit of the diffusive convection parameterization, is used for all experiments. It has been described by Yin and Sarachik (1994).
Figure 3a shows the zonally averaged stream function for Experiment I under restoring boundary conditions on both temperature and salinity. Experiment 1 shows three distinct thermohaline cells: an Arctic cell 16 Sv in strength with sinking at the northern wall, an Antarctic cell over 6 Sv in strength with sinking at the southern wall, and a cell 6 Sv in strength with sinking on the north side of the re-entrant current and southward surface flow all the way from the equator. Surface-confined cells on either side of the equator, locally driven by the equatorial Ekman divergence caused by easterly trade winds, are also evident.
Note that the surface stream function between 30°N and 40°N is oriented northward when the winds at those latitudes are westerly, implying southward Ekman flow near the surface. Examination of the surface currents (Figure 4a) indicates that the southward branch of the zonally averaged stream function in Figure 3 is dominated by a southward boundary current, while the interior flow is northward in