ABSTRACT

The ocean distributions of chlorofluorocarbons (CFCs) have been measured extensively in order to determine the mechanisms, rates, and pathways associated with thermohaline deep-water formation. Model temperature, salinity, and CFC-11 fields from the National Center for Atmospheric Research (NCAR) global ocean climate model are compared against observations with emphasis on the patterns of Antarctic Bottom Water (AABW) production, properties, and circulation in the Southern Ocean. The model control simulation forms deep water as observed in both the Weddell and Ross Seas, though not along other sectors of the Antarctic coast. Examination of the deep water CFC-11 distribution, total inventory, and profiles along individual observational sections demonstrates that the decadal-scale deep-water ventilation in the model Southern Ocean is both too weak and too restricted to the Ross and Weddell Sea source regions. A series of sensitivity experiments is conducted to determine the factors contributing to these deficiencies. The incorporation of a simple bottom boundary layer (BBL) scheme leads to only minor reductions in overall model–data error. The limited impact of the BBL may reflect in part other model large-scale circulation problems, for example, the lack of saline Circumpolar Deep Water along the Antarctic slope, and the coarse vertical resolution of the model. The surface boundary conditions in the permanent sea-ice-covered regions are a more major factor, leading to inadequate formation of dense, cold, and relatively saline shelf water, the precursors of AABW. Improved model–data agreement is found by combining the BBL parameterization with reasonably small adjustments in the surface restoring salinities on the Weddell and Ross Shelfs, justified by undersampling of winter conditions in standard climatologies. The modified salinities result in increased AABW production and enhanced signature of shelf water properties in the deep Southern Ocean similar in character to the effect of coupling with an active sea ice model.

1. Introduction Return to TOC

The observation of transient chemical tracers in the oceans has contributed greatly to our understanding of deep and bottom water formation and circulation (e.g., Schlosser et al. 1991a ; Smethie 1993 ; Doney and Jenkins 1994 ). Modeling studies are increasingly using this transient tracer data as an important measure with which to assess the skill of regional and global ocean circulation models (England 1995 ; England and Maier-Reimer 2001 ; England and Rahmstorf 1999 ; Roether and Doney 1999 ). The chlorofluorocarbons CFC-11 and CFC-12 are particularly useful in this regard because of relatively well-known atmospheric time histories and air–sea boundary conditions, low detection limits and inertness in seawater, and a growing global database (Bullister 1989 ; Doney and Bullister 1992 ; Roether and Doney 1999 ).

The continental shelves along Antarctica are one of the two main sources of deep water to the open oceans, the other being the Nordic seas in the northern North Atlantic. The production rate of the Southern Hemisphere deep-water component is a matter of some debate, with values ranging from approximately 2–5 Sverdrups (Sv 10⁶ m³ s¹) for the Weddell Sea [e.g., Carmack and Foster 1975 ; Gordon 1998 ) to basinwide geochemical estimates of 8.1–9.4 Sv using CFCs (Orsi et al. 1999 )] and 15 Sv using PO^*₄ and ¹⁴C (Broecker et al. 1998 ). Broecker (1999) , in fact, hypothesizes that systematic differences among the tracer estimates may reflect a slowdown in Southern Ocean deep-water production over the twentieth century. The paleoclimate record and coupled ocean–atmosphere climate model simulations (Manabe and Stouffer 1993 ; Murphy and Mitchell 1995 ) demonstrate that the deep Southern Ocean is a potential key element in modulating and/or driving the response of the full climate system. Recent hypotheses for explaining the glacial atmospheric CO₂ drawdown emphasize the interaction of Southern Ocean circulation, sea ice, and productivity (e.g., Stephens and Keeling 2000 ; Moore et al. 2000 ). Sarmiento et al. (1998) suggest that decreases in Southern Ocean deep-water ventilation under climate warming scenarios will reduce the projected oceanic uptake of anthropogenic CO₂, leading to a small positive climate feedback.

Orsi et al. (1999) present a recent synthesis, including much of the World Ocean Circulation Experiment (WOCE) hydrographic and chlorofluorocarbon data, describing the large-scale patterns of Southern Ocean bottom water circulation. A wide array of more regional Southern Ocean deep-water tracer studies have also been carried out (e.g., Schlosser et al. 1991b ; Roether et al. 1993 ; Bayer et al. 1997 ; Rintoul and Bullister 1999 ). Antarctic Bottom Water (AABW) is formed at a limited number of locations along the coast where very dense, cold ( < 1.7°C) near-surface shelf waters with high CFC concentrations are created by ocean–atmosphere and ocean–ice interactions. The shelf waters then sink down the continental slope as narrow, thin bottom boundary currents (tens of kilometers wide and 100–200 m thick), the dynamics of which are controlled locally by the balance of bottom friction, geostrophy, and entrainment (Smith 1975 ; Price and Baringer 1994 ). The AABW overflows entrain on average about an equal amount of warmer ( > 0°C), low CFC Circumpolar Deep Water (CDW) (Orsi et al. 1999 ). Distinct varieties of AABW, distinguishable by their temperature–salinity characteristics, are thought to form in the western Weddell Sea, Amery Ice Shelf, Adéle Coast, and western and eastern Ross Sea (Fig. 1 ). The newly ventilated AABW with accompanying high transient tracer burden are then incorporated into the deep, cyclonic subpolar gyre circulations in the Weddell–Enderby, Australian–Antarctic, and Southeast Pacific Basins.

The current generation of global ocean general circulation models show a wide range of behavior in terms of the strength and pattern of Southern Ocean deep-water ventilation (Dutay et al. 2002 ), with a corresponding disparity in model-estimated anthropogenic CO₂ uptake (Orr et al. 2001 ). The local mesoscale processes of shelf water formation, downslope flow, and entrainment are not resolved spatially in this class of models (e.g., Winton et al. 1998 ; Zhang et al. 1999 ) and must be incorporated via subgrid-scale parameterizations. Sea ice dynamics are included only implicitly in many cases, and the representation of ocean–ice exchange of heat and freshwater is often problematic. Further problems may arise because of interactions with imperfect large-scale circulation and water mass property fields.

Here we compare simulated temperature, salinity, and chlorofluorocarbon distributions from the National Center for Atmospheric Research (NCAR) Community Climate System Model (CCSM) global ocean climate model (Gent et al. 1998 ) against observations with emphasis on the patterns of AABW formation and circulation in the Southern Ocean. The NCAR control simulation forms deep water as observed in both the Weddell and Ross Seas, though not along other sectors of the Antarctic coast. But, based on the CFCs, the control simulation underpredicts the ventilation of the deep Southern Ocean on decadal timescales. The inclusion of a simple bottom boundary layer (BBL) scheme (Beckmann and Döscher 1997 , hereafter BD) improves the simulations, but by itself has a relatively small overall impact. This may be related to other large-scale circulation errors, for example, the water properties of the entrained intermediate water, the weak currents, and low vertical resolution characteristic of coarse resolution climate models. The surface boundary conditions in the permanent sea-ice-covered regions are a more major factor, leading to inadequate formation of dense, cold, and relatively saline shelf waters, the precursors of AABW. Improved model–data agreement is found by combining the BBL parameterization with reasonably small adjustments in the surface restoring salinities along the Antarctic coast and shelf regions. The resulting changes in deep-water properties and circulation are shown to be similar to the effect of coupling with an active sea ice model (Weatherly et al. 1998 ).

2. The NCAR CCSM Ocean Model Return to TOC

a. Control experiment

The global ocean model is the lower resolution configuration of the ocean component of the National Center for Atmospheric Research Community Climate System Model (Boville and Gent 1998 ; Gent et al. 1998 ). It is a z coordinate descendant of the well-known Bryan and Cox models of the Geophysical Fluid Dynamic Laboratory (Bryan 1969 ; Cox 1984 ; Pacanowski et al. 1993 ). In most respects the model is configured as in Large et al. (1997) . The model bathymetry for the Southern Ocean is shown in Fig. 1 , along with the tracks of several hydrographic sections used for model–data comparison.

Forcing is provided by monthly mean International Satellite Cloud Climatology Project (ISCCP) cloud fraction and surface solar radiation (Bishop and Rossow 1991 ; Rossow and Schiffer 1991 ), monthly mean precipitation [microwave sounding unit (MSU): Spencer 1993 ], and daily averages of National Centers for Environmental Prediction (NCEP) reanalysis surface atmospheric wind, air temperature, and humidity (Kalnay et al. 1996 ; note that monthly NCEP averages were used in Large et al. 1997 ). The momentum flux, turbulent and net longwave surface heat fluxes, and evaporation are calculated from the reanalysis data and model sea surface temperature (SST) using standard bulk formula (NCAR Oceanography Section 1996 ; Doney et al. 1998 ). A repeating 4-yr cycle of forcing, from years 1985 through 1988, is applied.

In the CONTROL integration, sea ice is not modeled explicitly, but its extent is diagnosed from the monthly SST climatology of Shea et al. (1990) ; similar to the simulations of Large et al. (1997) , sea ice cover is assumed to increase from zero to 100% as the climatological SST goes from 0.8° to 1.8°C. Ice–ocean heat and freshwater fluxes are included through a strong restoring of model surface temperature and salinity to the Levitus et al. (1994) climatology (average from 0 to 25 m) under regions of diagnosed sea ice. In regions away from the diagnosed sea ice extent, where surface forcing can drive SST below the freezing point, a restoring term is included to maintain SST at or above the freezing point. The resulting positive heat flux is treated as local sea ice formation with a corresponding brine rejection salinity flux. The local sea ice volume is accumulated over time (but not advected horizontally) and melts when (if) the SST goes above freezing.

Zonal and vertical resolutions remain the same as in the previous work, with 3.6° zonal resolution and 25 vertical levels spanning the 5-km vertical domain. Meridional resolution has been enhanced in the Tropics, with a meridional grid length of 0.86° between ±12° of the equator, increasing smoothly to 1.85° poleward of ±32°.

Reduced horizontal viscosities are applied in the tropics, resulting in higher velocities in the equatorial undercurrent (Large et al. 2001 ). Whereas a full anisotropic horizontal viscosity is used in Large et al. (2001) (and the variable viscosity, VAR_VISC, case below), the horizontal viscosity in the CONTROL case represents an earlier, intermediate formulation in which the viscosity is spatially dependent but isotropic. Open-ocean horizontal viscosity is 8 × 10³ m² s¹ within 10° of the equator, rising to 2 × 10⁵ m² s¹ poleward of ±30°; the viscosity takes on the higher value near land boundaries, irregardless of latitude, in order to resolve the dissipative Munk layer.

Other subgrid-scale mixing parameterizations include the K-profile parameterization (KPP) vertical mixing scheme (Large et al. 1994 ) and the Gent–McWilliams (GM) isopycnal mixing parameterization (Gent and McWilliams 1990 ). The incorporation of GM mixing tends to reduce the overall Southern Ocean deep ventilation rate by damping or removing spurious open ocean convection observed in earlier solutions (e.g., Toggweiler et al. 1989a ; Toggweiler et al. 1989b ), leaving behind only the smaller, more realistic coastal AABW formation path (Danabasoglu et al. 1994 ; England and Hirst 1997 ). The GM isopycnal diffusion and thickness coefficients are set at 0.8 × 10³ m² s¹. For KPP, minimum background levels of 5 × 10⁴ m² s¹ and 5 × 10⁵ m² s¹ are used for vertical viscosity and diffusion, respectively.

The physical initial condition for the CONTROL experiment is the result of several centuries of integration with depth-dependent acceleration (factor of 50 in the deep water), applied to potential temperature and salinity at depths below 1000 m, similar to that described in Bryan (1984) and Danabasoglu et al. (1996) . A further 31 years of synchronous integration are performed before starting the CFC control simulation to allow for transient adjustments following the switch from the asynchronous (accelerated) dynamics. The initial CFC-11 concentration is set to zero throughout the model domain. Air–sea exchange of CFCs begins, in both simulations, with the year of the model clock reading 1939 and is continued through the end of year 1997 (see below). An upper-ocean time step of 1 h is used in both accelerated and synchronous modes.

b. Sensitivity experiments

A set of sensitivity experiments (Table 1 ) is conducted to investigate the impact of the BD BBL scheme, enhanced shelf surface salinities, and an active sea ice model (Weatherly et al. 1998 ). The first two cases, bottom boundary layer BBL and enhanced surface salinity HIGH_SAL, have the same base physics and surface forcing as the CONTROL integration. The latter pair VAR_VISC and SEA_ICE differ somewhat from the CONTROL, including fully anisotropic horizontal viscosity and some minor changes in surface forcing and other physical parameters. These cases, which did not include transient CFC simulations, are included to illustrate the effect on the steady-state hydrography and circulation of switching from strong under sea ice restoring of temperature and salinity to a fully active sea ice model.

One well-known problem of z-coordinate ocean models such as the NCAR model, with its stepwise representation of topography, is the crude and unrealistic representation of downslope flow of dense waters (Winton et al. 1998 ). As shown schematically in Fig. 2 , the offshore advection of dense waters results in local convective instability and mixing with the ambient water in the grid box below. Thus the downslope transport of dense shelf waters occurs in the model as a series of mixing events, rather than advection as a bottom boundary current, greatly reducing the effective downward transport and magnitude of the density (and tracer) anomalies.

The tendency to excessively dilute dense waters in the process of overflow is one of the principal concerns that must be weighed when determining the appropriate class of model for a particular problem. Isopycnal (Bleck and Smith 1990 ; Oberhuber 1993 ) and sigma coordinate models (e.g., Haidvogel et al. 1991 ; Gerdes 1993 ; Danabasoglu and McWilliams 2000 ) represent topography and overflows in a more smooth and natural framework and, for the case of the isopycnic coordinates, better conserve water mass properties in the adiabatic interior regions. But these benefits may be offset for global applications by their own unique set of problems, namely the melding of the isopycnal interior and diabatic surface layer and vertical resolution in high-latitude deep convective regions for isopycnal models and the determination of the pressure gradient over strong topography for sigma models.

The Cox–Bryan (Cox 1984 ) Geophysical Fluid Dynamics Laboratory (GFDL) class of z-coordinate ocean models is widely used and retains a number of advantages, including lack of error in the calculation of the pressure gradient. Winton et al. (1998) suggest that z-coordinate models could adequately simulate downslope, density driven flow if the model resolved both the horizontal bottom slope and the bottom Ekman layer. However, the resulting constraints on the grid resolution, 3–5 km in horizontal and 30–50 m in vertical, are severe for global climate applications. Smith (1975) and Price and Baringer (1994) have shown some success with streamtube models in capturing the downslope entrainment, width, trajectory, and velocity for a number of overflows, but the coupling of local streamtube overflow models with a coarse-resolution GCM raises a number of physical and technical issues. Considerable effort, therefore, has been invested toward improving the representation of dense water overflows through simple subgrid-scale bottom (or benthic) boundary layer (BBL) parameterizations.

The BBL parameterization of BD97 is a zeroth-order approach to this problem. The BD scheme, modified slightly as in Döscher and Beckmann (2000 , hereafter DB), approximates a simple terrain following bottom boundary layer by allowing adjacent bottom grid cells that are separated in the vertical to communicate directly under specific conditions (Fig. 2 ). Advective tracer fluxes departing from bottom cells are rotated from the horizontal plane such that they enter the adjacent bottom cell, conditional upon the density in the upper cell being greater than that in the lower cell and the flow being offslope. A supplemental BBL diffusivity is also applied between bottom cells, subject to only the former condition. The modifications are applied only to the tracer terms (potential temperature, salinity, and any passive tracers), though secondary effects on the density field will subsequently alter the model velocities as well.

Our BBL implementation improves upon the original in terms of cost (an approximate 2.5% overhead relative to our standard model on a Cray C90, as compared with 10%) and in terms of memory usage. We chose a BBL diffusive coefficient of 3 × 10³ m² s¹, a value somewhat greater than the GM isopycnal mixing term. The advective component of the BBL is completely suppressed at the high northern latitudes where Fourier filtering is applied (beyond 75°N), as the filtering only acts in the zonal direction and cannot stabilize advective fluxes that have a vertical component.

A second major issue is the under-ice restoring salinities in deep-water formation regions (Large et al. 1997 ). Climatological estimates tend to be biased toward local summer months, when surface waters are likely to be too fresh due to seasonal ice melt. Some authors (e.g., Hirst and Cai 1994 ) argue that the restoring salinities on the Weddell and Ross Sea shelves should be adjusted upward to account for poor winter observational coverage and unresolved sea ice formation.

Two changes to the under-ice surface restoring are implemented in the HIGH_SAL case, which also has the BBL scheme active. First, the Levitus et al. (1994) climatology is replaced by the more recent Levitus et al. (1998) World Ocean Database data release, which is updated with somewhat better data coverage in the Southern Ocean. The change in surface forcing climatology is seen primarily at regional scale, with for example more saline conditions in the western Weddell Sea and fresher conditions along the Adéle coast west of the Ross Sea. Second, the surface restoring salinity field is enhanced almost everywhere along the Antarctic coast and shelf regions during the austral winter months (June–August). Rather than restore to the sea surface salinities (0–25 m), the highest salinities found within the uppermost 300 m of the column and for potential temperatures less than 1.5°C over the year are used based on the argument that those water properties are more representative of surface winter shelf water.

The modifications, which are not allowed to exceed a maximum of 0.15 ppt, are performed on the original 1° Levitus grid and then interpolated to the model grid. The full salinity adjustment is applied to all ocean points adjacent to the Antarctic coast and in shallow shelf regions, defined as bottom depths less than 300 m, poleward of 65°S such that the tip of the Antarctic Peninsula and Drake Passage are excluded. Away from the coast and shelfs, the modified restoring salinities are blended with the original surface field over two grid points in the north, northwest, and northeast directions and four grid points in the east–west direction with weights of 2/3 and 1/3, respectively; (two grid spaces in the zonal direction equal one in the meridional direction at 60°S).

The VAR_VISC case differs from the CONTROL solution by having fully anisotropic horizontal viscosity (Large et al. 2001 ) and reduced background vertical viscosity and diffusion, 10⁴ m² s¹ and 1.5 × 10⁵ m² s¹, respectively. The VAR_VISC case uses a single composite year (1985–88) of NCEP, ISCCP, and MSU surface forcing rather than the repeat 4-yr cycle, and the under-ice surface Levitus et al. (1994) restoring fields are 0–10 m averages, rather than 0–25 m. The overall effect of these differences is somewhat warmer and saltier deep water in the Southern Ocean and alterations in the regional patterns of bottom water formation.

In the SEA_ICE case, the under ice restoring is replaced by the CSM 1.0 active sea ice model. The CSM ice model is a cavitating fluid rheology, dynamic/thermodynamic ice model that prognostically computes local ice formation and melt, brine rejection, and horizontal ice advection. The ice model is implemented following Weatherly et al. (1998) with the corrected air–ice drag coefficient as in Bryan (1998) .

The BBL and HIGH_SAL cases branch from the CONTROL simulation near the end of the accelerated spin up. After activating the new physics, the simulations are integrated for an additional 40 accelerated years (2000 deep-water years) followed by 12 years of fully synchronous integration prior to the introduction of CFCs. The acceleration technique for VAR_VISC and SEA_ICE differed from the others cases, with a factor of 6 acceleration of tracers relative to momentum everywhere compounded by a depth-dependent acceleration reaching 5 in the deep ocean. The two cases are spun up in accelerated mode for 150 years (4500 deep-water tracer years) from a common equilibrium initial condition.

c. Chlorofluorocarbon tracer simulations

The CONTROL, BBL, and HIGH_SAL model cases include an additional passive tracer CFC-11 (Table 1 ). Air–sea gas exchange for the chlorofluorocarbon (CFC-11) is specified as in the Ocean Carbon-Cycle Model Intercomparison Project [OCMIP; Orr (1999) ; www.ipsl.jussieu.fr/OCMIP/]. A standard gas transfer formulation is used relating the flux F_CFC to a specified atmospheric partial pressure P^air_CFC, the model surface CFC concentration C^sea_CFC, and a gas transfer velocity k_w:

F_CFC = k_w(_CFCP^air_CFC C^sea_CFC), (1)

The rate of gas exchange is parameterized using the Wanninkhof (1992) quadratic wind speed relationship with the monthly mean wind speed squared ² computed by J. Boutin and J. Etcheto (1995, personal communication) from the SSM/I (Special Sensor Microwave Imager) satellite dataset following Wentz (1992) and Boutin and Etcheto (1996) . Gas exchange is suppressed under sea ice, diagnosed from the climatologies of Walsh (1978) and Zwally et al. (1983) .

The atmospheric time histories (1939–97) of CFC-11 used in OCMIP protocols were compiled by Walker et al. (2000) from reconstructions of integrated industrial release data and atmospheric loss rates (pre-1975) and atmospheric CFC observations (post-1975) (Bullister 1984 ; Doney and Bullister 1992 ; Doney et al. 1997 ). Separate CFC time histories were constructed for the Northern and Southern Hemispheres by Walker et al. (2000) to account for the predominant Northern Hemisphere industrial release and finite atmospheric mixing time across the intertropical convergence zone; in the NCAR CCSM simulations, the hemispheric time histories are assumed to be representative poleward of 10° latitude and are blended linearly between 10°S and 10°N. The atmospheric concentration of CFC-11 increases approximately exponentially following its introduction in the 1930s and 1940s. The atmospheric growth rate decreased and was approximately linear in the 1970s and 1980s, and more recently has decreased to zero or negative in the 1990s following the implementation of the Montreal Protocol.

Advection of passive tracers in the model is accomplished with centered differences, as has been customary with this lower resolution configuration of the model. We have experimented with a third-order upwind-weighted advection scheme, as described in Holland et al. (1998) , for the transport of CFC-11. Oscillations in the CFC-11 concentration of two gridpoint scale are much reduced with the alternative advection scheme in certain places (e.g., at Drake Passage), but errors of larger spatial scale, including unphysical undershoots of zero concentration, differ less between the two schemes. Sign-preserving advection schemes that limit extrema could be expected to modify the solution more significantly, and we intend to implement such a scheme with optional flux correction in an efficient manner, as outlined by Hecht et al. (1998) (at which time further exploration of the effect of the choice of advection scheme on the dynamical tracers T and S will be performed).

3. Control simulation and model–data comparisons Return to TOC

a. Deep-water CFC and hydrographic distributions

In the CONTROL simulation, AABW forms in the model Ross and Weddell Seas, transporting newly ventilated CFC-tagged water down the continental slope into the deep Southern Ocean (Fig. 3 ). As it is advected away from its various source regions, AABW is transformed by mixing with the overlying warmer, less dense, and low-CFC deep waters. Similar to the data, the model AABW produced in the Weddell Sea is colder, fresher, and more dense than its Ross Sea counterpart. Traced by the T–S signature and elevated CFCs, the model low salinity Weddell Sea bottom water spreads as suggested by field results, east along the Antarctic Circumpolar Current into the northern regions of the Weddell–Enderby and Australian–Antarctic Basins (Orsi et al. 1999 ) and northward into the South Atlantic Argentine Basin (Smythe-Wright and Boswell 1998 ) and the Crozet Basin in the southwest Indian Ocean (Haine et al. 1998 ). Plumes of relatively warm, saline AABW spread along the bottom equatorward of the Ross Sea source east of New Zealand and along the ridge separating the Southwest and Southeast Pacific Basins, also in broad agreement with the data (Orsi et al. 1999 ). Because of the relatively coarse horizontal resolution and smooth bottom topography, the model captures the large-scale patterns but not the details of deep-water flow, particularly in regions with considerable small-scale topographic channeling of the deep-water flow.

As shown in more detail in section 3c, a subsurface CFC maximum is often found at mid depth in the model, particularly for the CONTROL case, in contrast with the observations where the AABW CFC signal is typically bottom intensified. The horizontal extent of CFCs averaged over the deep water (not shown) is slightly greater than that along the bottom, particularly over the central Weddell and Ross Sea gyres, but does not change the broad patterns in Fig. 3 . Relative to the observations (Fig. 4 ), the CFC-tagged bottom waters near the model Ross and Weddell Sea sources are generally too low in concentration and too limited in horizontal extent. The model solution has no AABW source in the Indian sector, and the deep-water CFC levels in that ocean are essentially zero. This is in sharp contrast with the synthesis of Orsi et al. (1999) that shows elevated deep-water CFC levels propagating westward along the slope from the Amery Ice Shelf and Adéle Coast and significant AABW CFC concentrations (>0.01 pmol kg¹) throughout almost the entire Southern Ocean. Note that the CFC data used to construct the Orsi et al. (1999) spatial maps is normalized in time by basin (Atlantic 1987; Indo–Pacific 1993) while Fig. 3 displays the model CFC bottom field at the midpoint, 1990. Piecewise construction of the model maps using the 1987 and 1993 fields does not qualitatively alter the large-scale patterns or the model–data comparison.

The apparent westward flow of low-level CFC-tagged water in the model from the Atlantic through Drake Passage into the very eastern end of the Southeast Pacific Basin is also anomalous relative to the data. Only a small amount of the most dense components of AABW in the observations pass through Drake Passage (Roether et al. 1993 ), which, together with deep topographic flow restrictions south of the Kerguelen Plateau (75°E) and Pacific–Antarctic Ridge (160°E), allow the different varieties of deep AABW to retain relatively distinct –S compositions. Closer inspection of the model bottom layer –S properties suggests that this feature is not simply a problem arising from overly smooth model topography or counterflow under the Antarctic Circumpolar Current. The model Ross Sea AABW is excluded from the deeper reaches of the Southeast Pacific Basin (>4000 m) by a dense, cool (0.4°C), and saline water mass with a point source in Drake Passage. The origin of this water mass does not correspond to any known deep-water source and is likely a numerical artifact arising from the interplay of mixing, the centered difference advection scheme, and the high velocities in Drake Passage.

The model bottom-water hydrographic fields (Fig. 3 ) and model data difference maps (CONTROL minus Levitus) (Fig. 4 ) show a uniform warm and light bias relative to the observations within about 20° latitude of Antarctica, with largest discrepancies in the Weddell–Enderby and Australian–Antarctic Basins. An exception is the region to the north of the Antarctic Peninsula in the eastern Drake Passage where the model outflows from the Weddell Sea are not blocked by topography as in the ocean, leading to colder and denser bottom waters in the model. The outflows of Ross and Weddell Sea bottom water are too fresh both near the source regions on the continental slope and their poleward extensions while the interior of the adjacent cyclonic gyres are too salty. These findings are consistent with weak overall AABW production in the model and suggest that the source waters on the shelf are too light and too fresh and/or that there is excessive mixing with intermediate water during the overflow process—two questions which will be explored further in the sensitivity experiments.

The warm bias in the deep waters of the CONTROL case (Fig. 4 ) is also quite evident in the Southern Ocean horizontally averaged potential temperature plotted as a function of depth in Fig. 5 . The high salinities at mid depth and low salinities near the bottom in the horizontal average profile (Fig. 5 ) reflect a persistent error (Large et al. 1997 ) in the NCAR model whereby the export of North Atlantic Deep Water (NADW) out of the Atlantic occurs at too shallow a depth and the mixing between NADW and AABW is insufficient (also see the strong negative salinity bias in the South Atlantic; Fig. 4 ). Another concern is the model fresh bias in the upper half kilometer or so. Similar to the results of Toggweiler and Samuels (1995) , the CONTROL case does not simulate well the shallow salinity maximum associated with the upwelling and subsequent modification of Circumpolar Deep Water into shelf water.

b. Southern Ocean CFC inventory

The CFC deficit in the model Southern Ocean deep waters can be quantified directly by comparing the model deep-water CFC inventories with the observational estimates of Orsi et al. (1999) . The bounding volume in the model differs somewhat from that in Orsi et al. (1999) , who define AABW for the inventory estimates as all waters offshore of the 2500-m isobath and more dense than a neutral density surface ⁿ > 28.27 kg m³. The model inventories are computed by integrating CFC-11 concentrations offshore of the 2500-m isobath as well but with an upper surface set by the first local CFC vertical minimum below 1000 m. The well-defined CFC minimum in the simulations is a reasonable upper boundary for model bottom waters that generally matches the density surface criteria of Orsi et al. (1999) . By avoiding a fixed density surface, we allow for better intercomparison across the model sensitivity experiments, where the model deep-water density field can differ considerably from experiment to experiment.

The results for the Atlantic and Indo–Pacific sectors are presented in Table 2 (note that the model Indian sector deep water has essentially zero CFC inventory in the CONTROL and BBL simulations). Because of differences in the time of the field sampling, Orsi et al. (1999) normalized the Atlantic and Pacific data to different years, 1987 and 1993, respectively, before computing the inventories, and the model has been sampled in a similar fashion. The model inventories also are subdivided by latitude (total and south of 60°S) to highlight regional differences.

The CONTROL model inventories are a factor of 5 to 10 smaller than the data values, demonstrating clearly that the AABW formation processes in the model do not lead to sufficient ventilation of the Southern Ocean deep waters. Similar conclusions are reached by examining the CONTROL model meridional overturning streamfunction (not shown) with less than about 1–2 Sv penetrating below 2500 m near the Antarctic slope.

c. Chlorofluorocarbon sections

The model tendency toward too weak AABW ventilation is also borne out by comparisons with some of the individual shipboard CFC sections that went into the Orsi et al. (1999) synthesis. To facilitate each separate comparison, the model is sampled along the ship transects for the simulated month at the time of the actual shipboard occupation. Figure 6 displays the observed and model CFC distributions for the CONTROL (along with BBL and HIGH_SAL) case from WOCE line S4P, along approximately 67°S, across the northern Ross and Bellingshausen Seas [occupied March 1992; data from J. L. Bullister and M. J. Warner (2000, personal communication)]. The highest CFC deep-water concentrations in the simulation flow out from the western Ross Sea at mid depth (1000–2500 m), with peak concentrations about a factor of 2 lower than the observations and the CFC maximum displaced off the bottom by a couple of kilometers in the middle of the section. Not surprisingly, the core of old unventilated water at intermediate depth is underresolved by the model. The simulated deep water CFCs along the meridional WOCE section P15S, at approximately 170°W [Fig. 7 ; occupied February 1996; J. L. Bullister and M. J. Warner (2000, personal communication)] are qualitatively similar to the data south of about 50°S but poorly represent the northward extension of the bottom CFC signal, which reaches as far as 20°S in the data compared with only 55°S in the CONTROL case.

Figures 8 and 9 show a comparable model–data CFC comparison for the Atlantic using zonal and meridional sections from the AJAX program [section lines at 60°S, 0° from January to February 1984; Weiss et al. (1990) ]. Note that the model bathymetry differs considerably from the actual station depths, a reminder that only large-scale topographic features are resolved in a coarse resolution global model. The model simulates reasonably well the deep CFC maximum on the western end of the AJAX II zonal section but does not have sufficient eastward transport of deep CFC-tagged water along the northern edge of the Weddell Sea cyclonic gyre. This is particularly evident for the AJAX prime meridian line where the model deep CFCs do not penetrate to this longitude at all by 1984, and the model thus fails to represent the observed deep-water CFC signal between 45° and 60°S (Fig. 9 ).

The deep circulation patterns in the model Atlantic are generally correct though too slow since by the end of the simulations in 1998 the model does predict small though detectable CFC concentrations at 45°–60°S along the AJAX line (Fig. 3 ). It is perhaps notable that the major axis for eastward spread at this longitude in the model occurs where the Mid-Atlantic Ridge (MAR) blocks deep flow beneath the Antarctic Circumpolar Current but that an ample, deep channel between the Maud Rise and the southernmost limit of the MAR exists to the south. The deep CFC core along the Antarctic continental slope on the AJAX line (Fig. 3 ) is attributed by Orsi et al. (1999) to westward flowing AABW produced along the Adéle Coast and, thus not surprisingly, is absent from the model solution, which has no AABW formation in the Indian Ocean sector.

The model–data CFC agreement in the upper 2000 m along the AJAX (Fig. 9 ) and the P15S (Fig. 7 ) lines is considerably better than for the deep fields. The model captures reasonably well the strength, location and depth of the subpolar mode and Antarctic Intermediate Waters on both sections (Dutay et al. 2002 ). An exception is near the coast of Antarctica on the AJAX line, however, where waters with CFC concentrations above 4.5 pmol kg¹ extend to 800 m in the model, as contrasted with the observed depth of 100–200 m.

4. Sensitivity experiments Return to TOC

a. Bottom boundary layer parameterization

The incorporation of the bottom boundary layer parameterization in the BBL case should tend to increase the direct downslope transport of more dense, CFC-tagged shelf water into the deep waters and reduce the mixing and entrainment of intermediate depth waters. Figure 10 shows the resulting bottom water CFC-11 distribution and hydrographic difference fields from observations (cf. with Figs. 3 and 4 ). The largest impacts are found near the deep-water source regions, resulting in increases in the slope water densities and salinities and decreases in CFCs around the southern edge of the Weddell and Ross Sea gyres. The bottom water on the adjacent Weddell Sea abyssal plain becomes colder and fresher, with CFCs penetrating all the way to the bottom in contrast to the CONTROL case where they remained above 4000 m. The reverse pattern is found near the northern tip of the Antarctic Peninsula. In the CONTROL case, dense shelf waters flow all the way around the gyre to the tip of the peninsula, where they are forced offshore down the slope. The model AABW then moves northward into Drake Passage causing anomalous water properties in the model. Switching on the BBL allows the dense water to leave the shelf much sooner, leading to the dipole pattern in the difference maps. But as shown by the CFC inventories, this does not appear to significantly affect the overall rate of deep-water formation in the Weddell Sea. Increased ventilation of the southern deep Ross Sea is also evident in the CFCs and hydrographic properties but less dramatic. The inclusion of the BBL scheme does not significantly alter the boundary in the central and eastern Ross Sea with the dense, salty, CFC-free bottom water remaining to the east.

The result on the large-scale model AABW fields is positive but relatively small, not enough to overcome the original errors in the CONTROL simulation. The horizontally averaged salinity in case BBL increases over that of case CONTROL, in good agreement with observations, but the average bottom water temperature remains unchanged at about 0.5°C too warm (Fig. 5 ). The changes in the deep-water ventilation rate derived from the CFC inventories are also rather minimal with no shift in the Atlantic inventory and an enhancement of about 20% in the Indo–Pacific (Table 2 ). The BBL scheme impact is seen more in concentrating CFCs and recently ventilated AABW in the deep Ross and Weddell Sea gyres while diminishing the northern deep plumes.

As shown in the CFC ship sections (Figs. 6–9 ), the BBL parameterization tends to improve the simulation relative to the observations by enhancing the CFC concentrations right along the bottom and increasing the offshore penetration. But the effects are typically small, on the order of a few hundred meters (one to two grid levels) and 5°–10°, compared with the errors in the CONTROL case. In particular, the BBL case does not capture the bottom intensification of the CFCs in the data.

To facilitate comparison with DB and Dutay et al. (2002) we include a zonal section along 24°N in the North Atlantic (Dec 1997/Feb 1998). The observations show upper and lower deep-water CFC maxima associated NADW formed in the Labrador Sea and Nordic seas, respectively (Smethie et al. 2000 ). The model exhibits only a single CFC core at an intermediate depth. With the BBL scheme active, Dengg et al. (1999) find dramatic improvements in the simulated overflow depth of North Atlantic Deep Water, but in each case the source water characteristics were explicitly set atop the sill. By contrast, we see a more modest improvement in Fig. 11 , with a strengthening of the CFC maximum but little change in the outflow depth.

b. Enhanced surface restoring salinities

Another potential cause for the weak AABW ventilation in the CONTROL and BBL simulations is that the model is unable to form sufficient precursor shelf waters of the proper density. This point is illustrated in regional, winter season temperature and salinity scatterplots for the Ross and Weddell Seas and southern Indian Ocean, displayed in Figs. 12a, 12c, and 12e , respectively. In the BBL model and Levitus et al. (1994) climatologies, the deep-water potential temperature and salinity points fall roughly on a straight mixing line with near-surface waters at the seawater freezing temperature of 1.8°C. Comparing the endpoints, however, one sees that the inferred shelf water sources in the BBL are about 0.2 to 0.3 ppt too fresh, and even then one does not see a strong indication in the model for source waters at the expected salinity. By extrapolation, the model source waters have a ₂ potential density anomaly of about 37.05–37.10 kg m³ compared with 37.15–37.20 kg m³ in the data.

The near-freezing, saline seawater (S > 34.5 psu for the Weddell Sea and S > 34.3 psu for the Ross Sea) in the Levitus et al. (1994) observations are typically found below 200 m depth on the shelf, with a freshwater lens above, and represent the core of recently ventilated shelf waters (Fig. 5 ). The model surface boundary conditions under sea ice, however, are not capturing these water masses because the surface T and S are restored to the climatological values averaged over only the upper 25 m. Simply put, the period of active formation of shelf waters at the ocean surface is not well represented in Levitus et al. (1994) , and therefore the CONTROL model does not form water of the appropriate density on the shelf. The HIGH_SAL case was motivated to solve this problem.

As shown in Fig. 12 , the adjustment of the restoring boundary conditions improves the agreement between the model and observed surface salinities. The surface to deep mixing curves are now more similar, with projected surface endpoints having higher salinities in the Weddell Sea and to a less degree in the Ross Sea. While the changes in the surface forcing are to salinity, the main impacts with respect to density are due to cooler deepwater temperatures, consistent with enhanced downward transport of the dense, more saline shelf waters. Relative to the BBL case, more negative effective freshwater fluxes (not shown) are found in the HIGH_SAL case in deep-water formation regions (western Weddell Sea, eastern Ross Sea, Amery Ice Shelf, and Adéle coast). These negative fluxes, which mimic brine rejection in the absence of a dynamic sea ice model, are equivalent to about 0.5 m of sea ice production in those regions, roughly comparable to estimates of net sea ice production and export. The differences in the surface freshwater fluxes vary regionally, with areas of positive freshwater flux as well, reflecting the combination of the enhanced shelf salinities, the switch from the Levitus et al. (1994) to Levitus et al. (1998) climatology, and the nonlocal effects of circulation.

The impact on the deep-water CFC and hydrographic fields is significant (Fig. 13 ), pointing toward a broad enhancement in the AABW ventilation in both the Pacific and Atlantic basins and initiation of bottom-water formation in the Indian sector. Bottom-water CFC concentrations increase sharply as does the poleward and westward boundary of the CFC ventilated regions extending from the Ross and Weddell Seas and Amery Ice shelf. The deep-water CFC inventories increase by a factor of roughly 2–3 relative to the BBL and CONTROL cases. The individual shipboard section comparisons (Figs. 6–9 ) also show improvements, particularly with respect to the bottom intensification of the signal. A number of problems remain, however, for example on the AJAX section, where deep CFCs still do not reach the prime meridian by 1984 (though they do by 1990), and on the WOCE S4P zonal section across the Ross Sea, where the model CFC maximum is still displaced off the bottom by about 1000 m.

The CONTROL simulation exhibits consistent warm and light biases over almost the entire Southern Ocean from 50°S to the Antarctica coast relative to Levitus et al. (1994) (Fig. 4 ). In the HIGH_SAL case, the enhanced AABW production causes the Southern Ocean deep water to become uniformly colder and more dense, and the model–data temperature errors with Levitus et al. (1994) are smaller, more regional in pattern, and of both signs (Fig. 13 ). The model–data salinity difference maps do not change dramatically in the Ross and Weddell Seas, reflected in the temperature–salinity scatterplots as well (Fig. 12 ) as mostly a vertical shift in the deep-water T–S curves. The formation of deep water in the Indian sector reduces the model hydrography biases considerably, though the deep waters remain somewhat too warm, salty, and light along the coast from 80° to 150°E. North of 50°S, the HIGH_SAL simulation deep-water model–data differences grow, and the deep waters become uniformly too cold and dense as reflected in the Southern Ocean horizontal averages (Fig. 5 ). This may reflect more problems in the model downstream deep-water circulation and mixing of northern and southern source water rather than AABW formation per se.

c. Active sea ice

Chlorofluorocarbons have not yet been run in a version of the NCAR ocean model with an active ice model, but we can examine the effect on the deep-water hydrographic patterns by comparing the VAR_VISC and SEA_ICE cases. In particular, we would like to examine the degree to which the surface restoring salinity adjustments discussed above mimic the effect of a fully coupled ocean–ice system. The SEA_ICE case is based on a different control simulation, and while the CONTROL and VAR_VISC cases are derived from the same basic model framework, the two solutions differ in detail on their circulation and water mass properties due to a set of factors noted in section 2b. In general the pathways of deep-water formation in the NCAR model are sensitive to model configuration, and during the course of mapping the parameter space of, for example, the new anisotropic viscosity parameterization (Large et al. 2000), we have found that Southern Ocean deep-water formation rates can change considerably and in a nonintuitive manner.

Similar to the CONTROL case, the Southern Ocean bottom-water temperatures in the VAR_VISC solution are too warm (Fig. 14 ) and light (not shown) compared to the Levitus data, with the largest anomalies in the Atlantic and Indian sectors. Differing from the CONTROL, the VAR_VISC mean deep-water salinities are on average too high, and the regional error patterns are reversed with too low salinity in the Pacific and too high in the Atlantic. The VAR_VISC temperature–salinity scatterplots demonstrate that the same types of problems with respect to the shelf water end-members occur as in the CONTROL (Figs. 12b, 12d, and 12f ). Further, the mid-depth salinity maximum associated with NADW (an important mixing component for AABW) is stronger than observed, which together with weak AABW ventilation particularly in the Atlantic sector leads to the observed positive deep water salinity bias. The horizontal average salinity in the VAR_VISC case also contains a weak near-surface salinity maximum (Fig. 5 ), perhaps the result of the reduced background vertical mixing.

The sea ice model and modified restoring salinities behave similarly in that they both tend to increase the end-member salinity composition of the Shelf Waters and enhance overall AABW production rates. This is apparent in the temperature–salinity scatterplots (Fig. 12 ) as well-defined mixing curves more like observed between the shelf and deep-water end-members that extend to colder temperatures than found in the original control solutions. The difference between the HIGH_SAL and SEA_ICE cases is that the water mass properties of the shelf water end-members in the SEA_ICE case are fresher and lighter than observations, leading to T–S mixing curves with much shallower slopes than in either HIGH_SAL or Levitus et al. (1994) .

In terms of deep-water hydrography, the increased AABW formation and ventilation of the deep Southern Ocean is expressed as a decrease in deep-water temperature and an increase in potential density (not shown) over the Southern Ocean south of 50°S (Fig. 5 ). Averaged over the basin, both solutions overshoot the observations, resulting in an average cold, dense bias of about 1.0°C (Fig. 5 ). Spatially, the elevated ventilation (and thus temperature decrease) is more concentrated in the SEA_ICE case in the Indian and Atlantic sectors (e.g., cf. Fig. 4b and Fig. 13b relative to Fig. 14a and 14c ). The resulting SEA_ICE model deep-water temperature fields are somewhat too warm in the Weddell Sea and close to observed in the Ross Sea. Large negative temperature differences are found along Indian sector coast compared to observed.

The basin-mean deep salinity profile does not change dramatically from the BBL to HIGH_SAL cases, in part because the T–S mixing curve is approximately vertical as observed. In the SEA_ICE case, the vigorous shelf water production results in much stronger shelf water signatures in the deep water (e.g., colder temperatures). But because the end-member composition is incorrect, the deep water salinity decreases sharply, partially offsetting the effect of the temperature drop on density. In better agreement with the data, the SEA_ICE case produces a stronger intermediate salinity maximum. The basin salinity patterns show the smallest decreases in the Southeast Pacific Basin and the largest modifications in the Atlantic and Indian Oceans, where significant amounts of AABW appear to form in the coupled, ocean sea ice model.

5. Discussion and conclusions Return to TOC

Simulating AABW formation and circulation in a global ocean climate model is a significant challenge, requiring accurate representations of a number of individual processes including shelf water production, overflow and entrainment, and abyssal circulation. Comparison of the simulated CFC and hydrographic fields from a control integration of NCAR ocean climate model against field data indicates several deficiencies. The model deep-water ventilation in the Ross and Weddell Seas is too weak and too restricted close to the source regions. Deep water is not produced in the Indian Ocean, in sharp contrast with the CFC observations that support additional AABW formation sites near the Amery Ice Shelf and Adéle Coast. The model deep-water CFC maximum tends to be found at mid depth, rather than along bottom in the observations. Deep-water masses are warmer and lighter than in the observations throughout the Southern Ocean, and the salinities are too fresh near the source regions. The model deep-water errors are consistent with reduced production of too low salinity shelf waters caused by poor representation of winter formation events in the surface T and S climatology used to force the model under sea ice.

The incorporation of the BD and DB bottom boundary layer scheme leads to improvements, though relatively minor in extent, in the penetration of CFC-11 and the model deep-water mass properties. The BBL scheme acts in the expected direction to increase the downslope flow and decrease entrainment along the continental slopes, providing a more direct conduit for bottom-water ventilation, but insufficient to greatly alter the effective formation of AABW. Rather more dramatic BBL-induced changes have been found in realistic geometry, North Atlantic Basin models by DB , and Dengg et al. (1999) when the Nordic seas source water properties behind Denmark Strait are properly prescribed. Further, Campin and Goosse (1999) using a similar BBL scheme generate quite large offshelf flows in the Weddell (8.2 Sv) and Ross (8.6 Sv) Seas, leading to reduction in the trapping of cold water on the Antarctic shelves and an improvement in the zonal average density profile with depth. The weaker than hoped impact of the BBL in the NCAR simulations may reflect less the skill of the actual parameterization than the complex interaction of the BBL model with the biased mean state of the model. In addition to errors in the formation of the AABW precursors on the shelf, major contributing factors may be the lack of saline Circumpolar Deep Water along the model Antarctic slope where entrainment into the overflows occurs and the coarse vertical resolution of the model (and thus the effective BBL thickness in the BD scheme), 400 m in the deep water.

The BD and DB BBL model is a zeroth-order approach to the problem, and a number of more sophisticated alternatives have been put forward recently that have the potential for improved behavior. For example, Gnanadesikan et al. (2000, manuscript submitted to J. Phys. Oceanogr.) incorporate an explicit, embedded 50-m bottom boundary layer, much closer to the observed bottom Ekman depth and the thickness of overflow plumes, reducing the effective dilution of the overflow density and tracer anomalies in the BD scheme in coarse vertical resolution implementations. Their method also computes a pressure gradient term for the BBL and applies those velocities, rather than the horizontal velocities as in BD , but reintroduces the potential for topographic pressure gradient term errors as in sigma models. Killworth and Edwards (1999) and Song and Chao (2000) go a step further, embedding a turbulent bottom boundary layer slab model that allows the thickness to vary in space and time. These parameterizations extend the BBL treatment to the momentum equations as well, but have been evaluated only in idealized situations. More thorough investigation of these schemes in a global climate modeling context is warranted, focusing in particular on their interactions with sea ice treatment, partial bottom cells, and interior mixing parameterizations.

More substantial improvements in the NCAR model deep-water properties (particularly temperature and potential density) occur when the winter surface salinities along the Antarctic coast and shelf regions are adjusted to better match those of observed shelf water precursors. Improvements to the model solution include the initiation of bottom-water formation in the Indian Ocean and a factor of 2–3 increase in the deep-water CFC inventory. By contrast, Toggweiler and Samuels (1995) found only a minimal impact on modeled AABW salinities when a small surface salinity correction was adopted in this manner, equivalent to the brine rejection salt flux from 0.5 m sea ice per year. They suggest that the persistent model deep Southern Ocean salinity bias in their simulation arises instead from a physical circulation deficiency, namely the lack of saline Circumpolar Deep Water along the slope resulting in the entrainment of too fresh intermediate water by the overflows. Our results are more promising, but only partially solve the problem of overly weak model bottom water ventilation that is shown in Dutay et al. (2002) to be a common problem in z-coordinate models, especially those with Gent–McWilliams isopycnal mixing. As an interim step, we recommend for models using surface salinity restoring under sea ice the adoption of more realistic winter boundary conditions in the deep-water formation regions. This can be accomplished either through salinity adjustments to the Levitus et al. (1998) climatology or the use of higher resolution, regional salinity climatologies such as that compiled for the Arctic by Steele and Morley (2000) .

The effect of these surface salinity modifications are similar to those found with coupling to an active sea ice model. Both processes increase the end-member salinity composition of the shelf waters and enhance overall AABW production rates, leading to stronger shelf water signatures and colder temperatures in the deep waters. However, in the coupled sea ice case shown here, the shelf water end-member composition is too fresh and too light, and the strong bottom-water formation results in large, low salinity biases in the deep waters relative to observations. This finding highlights two aspects of the problem. First, the interactions of the ocean and ice components and surface forcing in a coupled ocean sea ice model are complex and are not guaranteed (or even likely on the first attempt) to generate realistic water properties for the shelf waters. Second, simultaneous matching of the Southern Ocean deep-water temperature, salinity, and CFC distributions requires adequate representation of both the rates of production and the near-surface water properties.

In the longer term, the problems associated with the under sea ice fluxes will require better tested and more robust active sea ice models. A number of groups are already spinning up their global ocean climate models coupled with an active sea ice model (thermodynamics and dynamics), but results from the OCMIP CFC comparison (Dutay et al. 2002 ) suggest that most of the coupled ocean–sea ice models tend to have too much ventilation of the deep Southern Ocean. The Livermore National Laboratory (LNL) model is one exception. Duffy and Caldeira (1997) report large improvements in the LNL simulation of both Southern Ocean deep-water salinity and radiocarbon fields in a coupled ocean sea ice calculation. Their results, however, depend on the inclusion of a subgrid-scale parameterization that distributes the model brine rejection at the base of the mixed layer rather than at the surface. Their model solutions are quite sensitive to the prescribed fraction of brine rejected at depth. More thorough comparison is needed of the relationship among deep-water CFC and hydrography fields, surface forcing, sea ice dynamics, and shelf water properties across the OCMIP models. Improvement of source water characteristics and the interactions of sea ice models and upper-ocean process would now appear to be a high priority, though this is considerably more challenging in global ocean or coupled ocean–ice models designed for application to climate study than in a regional process-study model.

Acknowledgments. We thank Ralf Döscher for his assistance in implementing the BBL parameterization, and for many useful discussions. Peter Gent provided a careful reading of the manuscript as well as broader advice over the duration of this work. Keith Lindsay contributed to the model visualizations and Christine Shields contributed to the model integrations. The CFC observations were kindly provided by John Bullister. We would also like to acknowledge the international OCMIP group, who provided the atmospheric CFC time series and surface gas exchange forcing. This work was supported in part by NASA through the U.S. OCMIP program and the U.S. JGOFS Synthesis and Modeling Project (NASA Grant W-19,274). The National Center for Atmospheric Research is sponsored by the National Science Foundation.

TABLE 1. Simulations.

TABLE 2. Southern Ocean deep-water CFC-11 inventories offshore of the 2500-m isobath. See text for details of how the inventories are computed for the observations (Orsi et al. 1999) and model cases.

Click on thumbnail for full-sized image.

FIG. 1. Model bathymetry (legend labeled in meters) in the Southern Ocean with section lines and geographical designations. One of the section lines follows close to the prime meridian, at the boundary of the figure. The color bar intervals coincide with the actual model vertical grid.

Click on thumbnail for full-sized image.

FIG. 2. In the BBL formulation of Beckmann and Döscher (1997) , if dense water (D) overlies lighter water (L) a supplemental bottom layer diffusion is applied. If, in addition to this condition, the orientation of the advective flux is off-slope, then it is rerouted along the bottom as shown.

Click on thumbnail for full-sized image.

FIG. 3. Southern Ocean distributions of bottom water properties for 1990 of the CONTROL simulation showing (a) CFC-11 (pmol kg¹), (b) potential temperature (°C), (c) salinity, and (d) potential density anomaly ₂ (kg m³) referenced to 2000 dbars. The fields are averaged over the deepest model layer (400 m thick) and are shown only where the model depth exceeds 2000 m.

Click on thumbnail for full-sized image.

FIG. 4. Southern Ocean distributions of (a) observed bottom-water CFC-11 (pmol kg¹; Orsi et al. 1999 ) and model–data differences for the CONTROL simulation minus the Levitus et al. (1994) climatology for (b) potential temperature (°C), (c) salinity, and (d) potential density anomaly ₂ (kg m³) referenced to 2000 m. The fields are averaged over the deepest model layer (400 m thick) and are shown only where the model depth exceeds 2000 m.

Click on thumbnail for full-sized image.

FIG. 5. Potential temperature and salinity vs depth, horizontally averaged over the Southern Ocean, from model cases, as indicated (see Table 1 for elucidation of case names) and from the Levitus climatology (Levitus et al. 1994 ).

Click on thumbnail for full-sized image.

FIG. 6. CFC-11 concentrations (pmol kg¹) along the WOCE S4P section [section data from J. L. Bullister and M. J. Warner (2000, personal communication); section line at 67°S, as seen in Fig. 1 ] for (a) observations, (b) CONTROL, (c) BBL, and (d) HIGH_SAL cases.

Click on thumbnail for full-sized image.

FIG. 7. CFC-11 concentrations (pmol kg¹) along the WOCE P15S section [section data from J. L. Bullister (2000, personal communication); section line at 170°W, as seen in Fig. 1 ] for (a) observations, (b) CONTROL, (c) BBL, and (d) HIGH_SAL cases.

Click on thumbnail for full-sized image.

FIG. 8. CFC-11 concentrations (pmol kg¹) along the AJAX II zonal section [Weiss et al. (1990) ; section line at 60°S, as seen in Fig. 1 ] for (a) observations, (b) CONTROL, (c) BBL, and (d) HIGH_SAL cases.

Click on thumbnail for full-sized image.

FIG. 9. CFC-11 concentrations (pmol kg¹) along the AJAX meridional section [Weiss et al. (1990) ; section line at 0°W, as seen in Fig. 1 ] for (a) observations, (b) CONTROL, and (c) HIGH_SAL. Case BBL does not differ appreciably from the CONTROL case on this section line.

Click on thumbnail for full-sized image.

FIG. 10. Southern Ocean distributions for the BBL case for 1990 of (a) bottom-water model CFC-11 (pmol kg¹) and model data differences for the BBL simulation minus the Levitus et al. (1994) climatology for (b) potential temperature (°C), (c) salinity, and (d) potential density anomaly ₂ (kg m³) referenced to 2000 m. The fields are averaged over the deepest model layer (400 m thick) and are shown only where the model depth exceeds 2000 m.

Click on thumbnail for full-sized image.

FIG. 11. CFC-11 concentrations (pmol kg¹) for the A24N section [section data from J. L. Bullister (2000, personal communication) along 24°N in the Atlantic] for (a) the observations, (b) CONTROL, and (c) BBL cases.

Click on thumbnail for full-sized image.

FIG. 12. Water column (0–5000 m) scatterplots of potential temperature against salinity for austral winter from the Levitus et al. (1994) climatology (black) and four model cases (red and green) within three geographic regions, as labeled. The Ross Sea includes points poleward of 50°S and between 155°E and 148°W. The Weddell Sea, including the southern half of Argentine Basin, encompasses points poleward of 45°S and between 58° and 0°W. The Indian Ocean sector includes points poleward of 60°S and between 40° and 130°E. Isopleths of potential density anomaly referenced to 2000 dbars are overlain.

Click on thumbnail for full-sized image.

FIG. 13. Southern Ocean distributions of (a) bottom-water modeled CFC-11 and model data differences for the HIGH_SAL case minus the Levitus et al. (1994) climatology for (b) potential temperature (°C), (c) salinity, and (d) potential density anomaly ₂ (kg m³) referenced to 2000 m. The fields are averaged over the deepest model layer (400 m thick) and are shown only where the model depth exceeds 2000 m.

Click on thumbnail for full-sized image.

FIG. 14. Southern Ocean distributions of bottom-water model data difference from the Levitus climatology (Levitus et al. 1994 ) for potential temperature (°C) and salinity for case VAR_VISC in (a) and (b) and for case SEA_ICE in (c) and (d). The fields are averaged over the deepest model layer (400 m thick) and are shown only where the model depth exceeds 2000 m.