North Atlantic Deep Water and Antarctic Bottom Water: Their interaction and influence on the variability of the global ocean circulation
Abstract
[1] Interhemispheric signal transmission in the Atlantic Ocean connects the deep water production regions of both hemispheres. The nature of these interactions and large-scale responses to perturbations on timescales of years to millennia have been investigated using a global three-dimensional ocean general circulation model coupled to a dynamic-thermodynamic sea ice model. The coupled model reproduces many aspects of today's oceanic circulation. A set of experiments shows the sensitivity to changes in different surface boundary conditions. Buoyancy changes in the Weddell and Labrador Seas exert a direct effect on the overturning cells of the respective hemisphere. They influence the density structure of the deep ocean and thereby lead to alterations in the strength of the ACC. Changing the wind stress south of 30°S influences the magnitude of the deep water production of both hemispheres. The interhemispheric effect in these experiments cannot be explained solely by advective mechanisms. Switching off the wind stress over the latitude band of the Drake Passage leads to a slow gradual decrease of the water mass transport in the ACC, resulting in an almost complete cessation. The model results prove the necessity to continue integrations over thousands of years until new equilibria are established.
1. Introduction
[2] A number of mechanisms for long term interhemispheric reactions of the climate system have been discussed. Concentrating on ocean processes, Crowley [1992] argues that high NADW production would lead to Southern Hemisphere temperature decrease. In the NADW circulation relatively cold water in the deeper layers is transported to the south while in the upper layers warm water travels north resulting in a northward heat transport. The reduction in heat loss of the South Atlantic and the Southern Ocean in case of a weaker NADW overturning exceeds the heat loss by reduced upwelling of NADW by a factor of 3 to 6. In the opposite case large NADW production would lead to cooling of the Southern Hemisphere.
[3] These ideas have been confirmed in a number of model studies (among others by Mikolajewicz and Maier-Reimer [1990] and Maier-Reimer et al. [1993]). Observational evidence for this kind of interrelations is sparse. Hall et al. [1997] analyzed a moored array in the western basin of the equatorial Atlantic, finding quasi-annual cycles in Antarctic Bottom Water (AABW) and Lower North Atlantic Deep Water (LNADW) transport rates, with AABW maxima coinciding with LNADW minima and vice versa.
[4] In the framework of paleoclimate changes, Vidal et al. [1999] found evidence for an opposite phase relationship of sea surface temperatures (SST) in the northern and southern Atlantic during the last glacial period. They infer an interhemispheric connection via the global ocean circulation. Broecker [1998] and Broecker et al. [1999] analyzed δ18O, 14C and PO4* inventories for the deep ocean. They propose an alternation between intense deep water formation in the northern North Atlantic and in the Southern Ocean as explanation for the opposite phases of SST in the Northern and Southern Hemispheres. The climatic consequence was depicted as a bipolar seesaw with stronger heat release from the Southern Ocean and warming in Antarctica during periods of strong deep water formation in the Southern Hemisphere accompanied by weak heat release and cooling in the North and vice versa. Cox [1989] analyzing four idealized global ocean experiments, Goosse et al. [1997] varying freshwater fluxes and Fieg [1996] and Fieg and Gerdes [2001], comparing the results of over twenty numerical experiments with both LGM (Last Glacial Maximum) and recent boundary conditions, find a similar seesawing behavior in the strength of NADW and AABW.
[5] Studies with a coarse-resolution OGCM [Toggweiler and Samuels, 1993, 1995] suggest a dependence of the formation rate of NADW on Southern Hemisphere winds at the latitude band of the Drake Passage, a teleconnection Toggweiler and Samuels call the “Drake Passage effect”. Strong westerly winds cause northward Ekman drift in the circumpolar zone. The water pushed north must return southward at depth, below the depth of topographic ridges at the Drake Passage's latitude. Above the ridges no mean zonal pressure gradient can exist, which implies the impossibility of net southward geostrophic flow there. According to Toggweiler and Samuels, the only place for deep downwelling and the formation of a compensating deep flow is the North Atlantic as the ocean is too stratified elsewhere. The enforced separation between northward flows at the surface and southward flows at depth constitute an overturning of the deep ocean that is independent of vertical mixing. The “Drake Passage effect” has been confirmed by investigations with an OGCM of comparable resolution coupled to a simple atmospheric feedback model [Rahmstorf and England, 1997]. The latter, though, found the strength of the NADW cell being modified by the influence of Southern Hemisphere winds instead of being controlled by them.
[6] The present study investigates numerical experiments where main processes affecting deep and bottom water formation are modified. The analysis concentrates on global modes of the ocean circulation, trying to identify timescales and mutual interaction between its parts. The interplay of North Atlantic Deep Water, Antarctic Bottom Water and the Antarctic Circumpolar Current is addressed applying a global ocean general circulation model coupled to a dynamic-thermodynamic sea ice component. Section 2 gives a description of the model components, their coupling and the basic experimental setup. Section 3 illustrates the main features of the model in equilibrium state. Examining the reaction to alterations in the systems boundary conditions (section 4) reveals mechanisms of the thermohaline circulation and serves as a basis for a comparison of the model's behavior with the results of other studies in section 5. Finally, section 6 summarizes the results of the present investigation and gives an outlook.
2. Model Description and Coupling
[7] The ocean component used here is a general circulation model based on the second version of the “Modular Ocean Model” (MOM-2) developed at the “Geophysical Fluid Dynamics Laboratory” in Princeton [Pacanowski, 1995]. It applies the primitive equations of thermodynamics and hydrodynamics. Its main features are based on works by Bryan [1969] and Cox [1984]. As this model has been used widely and is fully documented the reader is referred to Pacanowski [1995] for details.
[8] The horizontal resolution of the model is 4° in longitude and 3° in latitude. There are 20 vertical levels, monotonically increasing from about 50 m spacing near the surface to about 450 m near the bottom at 5000 m depth (see Table 1). The model's topography is taken from Danabasoglu and McWilliams [1995]. Smoothing and a minimum depth requirement in their model removes shelf seas and leaves the Greenland-Iceland-Scotland ridge area too deep. To avoid unrealistic circulation patterns a shelf area in the Barents Sea and a shallower transition zone between the North Atlantic and the Arctic Ocean have been introduced into the model (following Meissner and Gerdes [2002]). The resulting topography is displayed in Figure 1.
| Level | Δza, m | zb, m |
|---|---|---|
| 1 | 51.23 | 25.00 |
| 2 | 56.13 | 77.46 |
| 3 | 65.79 | 137.52 |
| 4 | 80.00 | 209.05 |
| 5 | 98.39 | 297.25 |
| 6 | 120.51 | 405.82 |
| 7 | 145.82 | 538.27 |
| 8 | 173.70 | 697.47 |
| 9 | 203.45 | 885.67 |
| 10 | 234.36 | 1104.38 |
| 11 | 265.64 | 1354.38 |
| 12 | 296.55 | 1635.67 |
| 13 | 326.30 | 1947.47 |
| 14 | 354.18 | 2288.27 |
| 15 | 379.49 | 2655.82 |
| 16 | 401.61 | 3047.25 |
| 17 | 420.00 | 3459.05 |
| 18 | 434.21 | 3887.25 |
| 19 | 443.87 | 4327.46 |
| 20 | 448.77 | 4775.00 |
- a Layer thickness.
- b Depth of scalar and vector points.
[9] Sub-grid-scale mixing for momentum is implemented with constant values for the horizontal and vertical friction coefficients as listed in Table 2 for the control run (referred to as “CTRL” in the following). Small grid distances at high latitudes due to the convergence of the meridians often require the filtering of prognostic variables. The truncation of the series expansions involved can cause nonphysical values that are especially troublesome in the tracer fields. On the other hand, tracer time steps are limited by relatively slow advection and filtering of tracers might not be necessary to the same extent as filtering of velocity components. To account for the need to filter those and to avoid the problem of nonphysical tracer values, filtering is here applied only to velocities and the stream function north of 79.5°N using a symmetric finite impulse response filter. Tracer advection employs the FCT algorithm [Boris and Book, 1973; Zalesak, 1979], which eliminates undesirable computational “noise” within the models tracer fields while reducing diffusive effects on them as far as possible [see, e.g., Gerdes et al., 1991; Griffies et al., 2000]). Convection is parameterized applying the Rahmstorf scheme (as explained by Rahmstorf [1993]). Table 2 also states the different time steps used for tracers (temperature, salinity) and internal and external mode velocities. This asynchronous integration technique decreases the speed of gravity and external rotational waves but leaves baroclinic Rossby waves almost unaffected [Bryan, 1984].
| Description | Parameter | Value | Unit |
|---|---|---|---|
| Vertical eddy viscosity coefficient | κm | 10−3 | m2/s |
| Horizontal eddy viscosity coefficient | Am | 106 | m2/s |
| Vertical eddy diffusivity coefficient | κh | 0 | m2/s |
| Horizontal eddy diffusivity coefficient | Ah | 0 | m2/s |
| Restoring time constant for SSS | ΔtrS | 96 | days |
| Length of time step for tracers | τ | 43,200 | s |
| Length of time step for internal mode velocities | τ | 1728 | s |
| Length of time step for external mode velocities | τ | 864 | s |
[10] A dynamic-thermodynamic sea ice model [Harder, 1996; Harder et al., 1998] is the second climate system component used for the present study. It employs a viscous-plastic rheology [Hibler, 1979] and the Parkinson and Washington [1979] thermodynamics with the Semtner [1976] zero-layer approach for heat conduction. The model includes a prognostic snow layer [Owens and Lemke, 1990] and accounts for the effect of flooding [Fischer, 1995]. The sea ice is considered a two-dimensional continuum and is described by the mean sea ice thickness, the ice concentration (i.e. spatial coverage) and the ice drift velocity. The thermodynamic evolution of the ice cover is determined from the energy budget at the ice surface and from heat conduction through the ice. Atmospheric and oceanic drag govern the evolution of the ice drift velocity. The internal ice forces are parameterized using a viscous-plastic rheology with an elliptic yield curve. For a more detailed discussion of the model, see Brix [2001].
[11] The ice model is run on the same grid as the ocean model. Its domain was divided into two parts calculating separately on both hemispheres poleward from 45°. The time step chosen was the same as the tracer time step of the MOM. Parameterizations used in the ice model are listed in Table 3. The ice strength parameter P* has different values for the Arctic and Antarctic because sea ice conditions differ substantially [Harder and Fischer, 1999]. Therefore it was necessary to tune these values separately for a realistic simulation. The albedo, i.e. the capacity to reflect incoming radiation, also differs between the hemispheres to account e.g. for the effects of meltponds on multiyear ice in the Arctic. The albedos chosen for the Arctic are identical to those used by Harder [1996], values for the Antarctic are taken from Fischer [1995].
| Description | Parameter | Value | Unit |
|---|---|---|---|
| Thermodynamical Parameters | |||
| Density of sea ice | ρI | 910 | kg m−3 |
| Density of snow | ρs | 300 | kg m−3 |
| Density of water | ρw | 1000 | kg m−3 |
| Specific melt energy for sea ice | LI | 3.34 × 105 | J kg−1 |
| Lead closing parameter | h0 | 1 | m |
| Dynamical Parameter | |||
| Drag coefficient for the ocean | cw | 5.5 × 10−3 | |
| Rheological Parameters | |||
| Ice strength NH | P* | 15,000 | N m−2 |
| Ice strength SH | P* | 22,500 | N m−2 |
| Ice concentration parameter | C | 20 | |
| Eccentricity of the yield curve | ϵ | 2 | |
| Regime parameter | Δmin | 2 × 10−9 | s−1 |
| Radiation Parameters/Albedos | |||
| Open water | αw | 0.1 | |
| Ice (without snow) NH | αI | 0.65 | |
| Ice (without snow) SH | αI | 0.75 | |
| Melting ice (without snow) NH | αm | 0.60 | |
| Melting ice (without snow) SH | αm | 0.66 | |
| Snow NH | αs | 0.80 | |
| Snow SH | αs | 0.85 | |
| Melting snow NH | αsm | 0.70 | |
| Melting snow SH | αsm | 0.75 | |
- a NH: Northern Hemisphere; SH: Southern Hemisphere.
[12] The ocean supplies the sea ice with SST and values for velocity at the second model level that are used to calculate the sea surface tilt and the ice ocean drag. The ice model, in turn, feeds MOM with internal stresses, information about evaporation, fresh water fluxes and net heat fluxes. The exchange between the two components takes place every 12 hours. For a more detailed description of sea ice-ocean coupling see Hibler and Bryan [1987].
[13] The surface forcing fields consist of monthly mean data sets which are interpolated linearly for the actual time step within the model. The sea surface salinity (SSS) fields were obtained from the climatology of Levitus [1982]. The data was averaged over the upper 50 m to match the first vertical level thickness and was transferred onto the model grid by horizontal averaging. The tracer equation for salinity at the shallowest grid level is linearly restored to the surface forcing fields with a time constant ΔtrS as listed in Table 2. All other forcing fields have been derived from a validated 15 year set of assimilated data for the period 1979 to 1993 provided by the reanalysis project of the European Center for Medium-Range Weather Forecasts (ECMWF) [Gibson et al., 1999]. The data sets used here (2-m temperature, 2-m dew point temperature, total cloud cover, precipitation, scalar winds and the horizontal wind stress components) are monthly means interpolated onto the model's grid.
3. Control Run
[14] For validation purposes it is important that the model is capable to reproduce the main circulation systems and water masses, i.e. the present-day's climate state. To this end the model's equilibrium state (meant to be the state in which large-scale advective adjustments have been completed) will be presented in this section. It has been reached after about 3500 years of integration. The run then was continued for another 1500 years as a control case for the experiments with altered boundary conditions that will be described in the next section. The results presented here are taken from a 500 year mean for the model years 3500 to 3999.
[15] The overall structure of surface salinities and temperatures in the world ocean (not shown) is reproduced realistically by the model, also covering regional structures like the intrusion of fresh Pacific waters through the Indonesian Passage into the Indian Ocean. Deviations from reality are found in the big river estuaries, where the rivers' contribution to the freshwater budget is not sufficiently reproduced. However, this effect is limited to the model's first level. Surface salinities in the Arctic along the Siberian and Canadian coast are too high. Around Antarctica the belt between 50° and 60°S shows salinities that exceed observed values by about 1 psu. The most pronounced shortcoming in the SST distribution is found near the West African shore. There, too weak upwelling takes place in the model, leading to exaggerated temperatures.
[16] As an illustration of the latitude-depth distributions temperature and salinity sections across the Atlantic are shown in Figure 2. Compared to climatological values the thermocline is sharper, but a little too shallow. At intermediate depths the Atlantic is too warm, while the bottom water is too cold. The signature of Antarctic Intermediate Water is simulated well for a coarse resolution model. Nevertheless, the intermediate water tongue is shallower than in reality and its salinity values are 0.3 to 0.4 psu higher. The whole Atlantic basin is too salty, the lower thermocline structure is less sharp than observed. The reason for this water mass being too homogeneous may be found in the deep water formation south of the Greenland-Scotland-Ridge where salinities are higher than in the observed deep water formation regions of the Nordic and Labrador seas leading to subduction of too warm and too salty water masses.
[17] The Northern Hemisphere's big current systems, the Kuroshio in the Pacific Ocean and the Gulf Stream in the Atlantic Ocean are reproduced properly (compare second level velocities in Figure 3). The pathway of warm surface waters along the east coast of North America, across the basin and into the Nordic Seas compares well to observations. However, as in most ocean models, the Gulf Stream is too weak. In the Indian Ocean the southward western boundary current that represents the Mozambique and East Madagascar currents is very pronounced. The overall structure of the subtropical and subpolar gyres is close to reality. The flow field of the Antarctic Circumpolar Current (ACC) with its northward deviation east of the Drake Passage is also properly reproduced. Its volume transport of about 235 Sv is high compared to estimates based on velocity observations of 130 to 140 Sv [Whitworth and Petersen, 1985; Ganachaud and Wunsch, 2000]. However, it is in the range of recent model results (for instance the 240 Sv in the coupled model run of Boville and Gent [1998]). Sensitivity experiments that have been performed with an uncoupled version of the ocean model using other parameterizations (i.e. the isopycnal thickness diffusion scheme after Gent and McWilliams [1990]) yielded more realistic values for the water mass transport in the ACC. These improvements, however, had the downside of deteriorating other model features and diagnostic values as for instance deep water production rates. As the emphasis in the present work was to be laid on the water masses involved in the overturning circulation the volume transport in the ACC has been regarded as of minor importance compared to the qualitative features that are met satisfyingly.
[18] Meridional overturning stream functions for the Atlantic and the global ocean are presented in Figure 4. The most intense features are the shallow low latitude cells with upwelling at the equator. They are driven by Ekman transport due to the easterly trade winds. Most of this water sinks within 30° of the equator. Ekman transports in the subpolar zones produce shallow cells with equatorward surface flows. In the Northern Hemisphere this cell is quite weak as it is opposed by the strong North Atlantic thermohaline circulation. The corresponding cell in the Southern Hemisphere is called the Deacon cell. It is connected in intermediate depths to the thermohaline cell intruding from the North. The combined effect of both cells leads to strong upwelling near 60°S. The southernmost cell represents the formation of AABW. It extends to the seafloor and spreads far north across the equator to 50°N. The large-scale overturning is driven by downwelling north of 50°N in the Atlantic. The pronounced Atlantic cell has a maximum strength of more than 24 Sv and is centered in about 1000 m depth. Its southward branch represents the transport of NADW. About 16 Sv enter the Southern Hemisphere, while 8 Sv upwell before the equator is reached. The meridional overturning stream function of Figure 4 is in qualitative agreement with, for example, the investigations by Danabasoglu and McWilliams [1995] and Meissner and Gerdes [2002], even though the circulation cell is slightly stronger than in their cases. For the export of deep water out of the Atlantic basin estimates vary between around 13 Sv [Schlitzer, 1993; Schmitz and McCartney, 1993; Schmitz, 1995] and 17 Sv [Rintoul, 1991].
[19] We further evaluate our OGCM with respect to its ability to reproduce major water mass transport pathways. To follow the path of deep water a passive tracer was introduced into the model in year 3490. Its concentration was set to a constant value of one at the surface level. Figure 5 (top) gives the concentration of the tracer in 1950 m depth 300 years after it has been introduced into the model. Concentration values of one indicate that the whole water at this grid box has been replaced by water that has been in contact with the surface within the 300 years. The maximum concentrations are found in the Atlantic close to the Greenland-Scotland-Ridge and spreading southward from there along the western boundary. At 30°S the concentration plume turns east and follows the northern edge of the ACC circling the entire Southern Ocean. Further concentration maxima exist in the Arctic Ocean, which is filled with subducted water. The entire Pacific Ocean north of 30°S shows concentration values smaller than 0.1. 300 years are not enough to fill this basin with water masses originating from the surface layers.
[20] Other maxima are located in the Ross Sea and the Weddell Sea. High concentrations are confined to very small areas, indicating that the waters sink to greater depths. At 3800 m depth (Figure 5 (bottom)) no signals exist that point to waters of Northern Hemisphere origin. Hence, NADW is entirely confined to higher levels. But waters subducted in the Southern Ocean have filled the adjacent basins and slowly proceed north. While during 300 years water of Northern Hemisphere origin has already spread around the globe with the ACC, Southern Hemisphere bottom water has only reached the equator in the Atlantic and Indian Oceans. The pathways of NADW and AABW in the model are well represented and are confirmed by ideas derived from observations.
[21] Monthly means of sea ice concentration are shown in Figure 6 for February and August. Comparison with satellite data [National Snow and Ice Data Center (NSIDC), 1996] reveals that the winter concentrations are simulated well in the central Arctic basin. The ice-free region along Scandinavia is a result of the warm North Atlantic Current. The observed winter sea ice cover along the East coast of Greenland is reproduced only in its northern part; south of the Greenland-Scotland-Ridge the concentrations are too low. The ice cover in the Labrador Sea is realistic. In the North Pacific the ice concentrations are underestimated. In wintertime the Northern Hemisphere's sea ice covers 8.8 × 106 km2, while Gloersen et al. [1992] name an observational value of 14.5 × 106 km2. (Ice extent as used here is defined as the area poleward of a line of grid cells with an ice concentration of at least 10%. For the observations the sea ice extent is defined as the area enclosed by the ice edge, which in turn is defined as the 15% concentration contour. Winter is used here for the mean value over the months December to February (DJF) in the Northern Hemisphere and June to August (JJA) in the South. Summer values span the same months for the other hemisphere, respectively.) In summer, the difference between modeled and observed values is about one third less with 7.9 × 106 km2 against 11.5 × 106 km2. These differences are not solely due to a deficit in modeling the sea ice cover. Due to the resolution of the model grid the model underestimates the Arctic Ocean's size by 15–20%. Large parts of the shelf areas and the Canadian Archipelago are not resolved by the model because of their shallow depths. This leads to an overall Arctic ice extent that is lower than figures derived from observations. In boreal summer the ice cover is restricted to the central Arctic Ocean and the northeastern Greenland shore, both in model and reality. The concentrations in the area of the Chukchi Sea and the Canadian Basin are too high. SSM/I data [NSIDC, 1996] give values reaching from one in the central basin to zero on the shelves. The annual cycle of the ice extent with its extremes in March and September is realistic. Summarizing, it can be stated, that the Northern Hemisphere ice concentrations and extent are modeled satisfactorily regarding the models coarse resolution.
[22] Around Antarctica the model's ice cover exhibits a strong annual cycle. In winter the continent is surrounded by a closed ice cover, while in summer, there are only a few small ice covered areas left. The model's ice extent reaches values of 16.2 × 106 km2 in winter (JJA) and 5.87 × 106 km2 in summer (DJF). Gloersen et al. [1992] report 16.0 × 106 km2 and 7.0 × 106 km2, respectively. The annual cycle in the model reaches its maximum in September and October with 22.9 × 106 km2. Caused by the high temperatures in this area, the spring melting in the model is very intense, leaving almost no ice in late summer and fall (0.7 × 106 km2 in March). Compared to the observed ice covered regions [NSIDC, 1996] in the Weddell Sea and between the Antarctic Peninsula and the Ross Sea, the modeled areas are too small in summer. In winter the sea ice extends farther to the North than in reality. Especially in the Weddell Sea region and the Indian Ocean sector the modeled ice edge reaches 55°S.
[23] Winter sea ice thickness patterns in the Arctic (not shown) are unrealistic. This is due to two reasons: the models resolution is too coarse and the existence of an island at the North Pole for numerical reasons partially blocks the transpolar drift. The sea ice tends to pile up north and northwest of Greenland as the passage between the island and the North Pole is too small to let the ice drift through. This “ice jam” reaches back to the Chukchi Sea. This problem is not unique to this model; it has also been reported, for example, by Stössel et al. [1998] and Fritzsch et al. [2000]. The ice thickness distribution around Antarctica is within the range of observed values [Jeffries, 1998]. In the western Weddell Sea along the Antarctic Peninsula ice thickness appears to be too low compared to measurements [Strass and Fahrbach, 1998]. This is due to the model's convective activities that are concentrated in the southwestern edge of the Weddell Sea and bring too warm water to the surface.
4. Sensitivity Experiments
[24] Perturbations of the surface forcing have been introduced to study the temporal evolution of anomalies revealing mechanisms of interaction and information transfer. Two sets of experiments have been performed: changes of the surface buoyancy forcing with its possible direct influence on the thermohaline circulation and alterations of the wind forcing.
[25] In all of the following experiments the disturbances were introduced into the model in the year 3490 of the control run. Then the model was integrated for another 1500 years keeping these disturbances. As a reference the control run was continued unchanged over the same time.
4.1. Buoyancy Experiments
[26] The convective process is very sensitive to perturbations in certain areas as shown, for example, by Lenderink and Haarsma [1994]. They identified potentially convective regions, where convection can easily be triggered or stopped. The idea here was to use such sensitive areas to strengthen or slow down convective activities and deep water formation. Two regions that are well known for their sensitivity with respect to convective disturbances have been chosen: the Labrador Sea and the Weddell Sea. Nine grid boxes were selected in each region (compare Figure 1). There, the values to which the salinity was restored, were manipulated by adding or subtracting 1 psu to or from the climatological values as indicated in Table 4. These changes lie within the range of observed anomalies as, for example, during the great salinity anomaly [Dickson et al., 1988].
| Experiment | Salinitya, psu | ACCb | NADWc | AABWd | |||
|---|---|---|---|---|---|---|---|
| Sverdrups | Δ,e % | Sverdrups | Δ,e % | Sverdrups | Δ,e % | ||
| CTRL | ±0 | 236.0 | 0 | 24.81 | 0 | 8.11 | 0 |
| LAB+ | +1 | 220.9 | −6.3 | 30.83 | 24.3 | 6.40 | −21.1 |
| LAB− | −1 | 237.4 | 0.6 | 23.78 | −4.2 | 7.19 | −11.3 |
| WED+ | +1 | 267.5 | 13.3 | 22.95 | −7.5 | 13.56 | 67.2 |
| WED− | −1 | 229.2 | −2.9 | 25.12 | 1.2 | 5.49 | −32.3 |
- a Restoring changes as explained in the text.
- b ACC: Drake Passage throughflow.
- c NADW: maximum of the stream function for the zonally integrated volume transport below the surface layer in the North Atlantic.
- d AABW: amount of the minimum of the global stream function for the zonally integrated volume transport below 300 m depth.
- e Deviation from the control run in percent.
[27] In Figure 7 the total kinetic energy per unit volume is displayed. From this graph, four conclusions can be drawn immediately: the runs with the positive salt anomalies show stronger deviations from the control run than those with negative ones. The deviations have opposite signs between the hemispheres, i.e. positive salt anomalies in the Weddell Sea lead to a positive energy deviation, while positive salt anomalies in the Labrador Sea show negative energy deviations and vice versa. The amplitude of the Weddell Sea experiments (referenced as “WED” in the following) are bigger than those of the Labrador Sea runs (“LAB”) with the same anomaly strength and the system does not instantly switch to another state, but is slowly adjusting to a new equilibrium.
[28] The magnitude of the ACC dominates the global kinetic energy budget as it is by far the strongest ocean current system on earth. Table 4 reflects this connection; runs with positive kinetic energy deviations from the control run also show positive deviations in the Drake Passage throughflow and vice versa. At the same time positive (negative) anomalies in the ACC strength are connected to negative (positive) anomalies in the NADW strength. The reaction of AABW is nonuniform. The reason for different behavior of the LAB and WED runs could possibly lie in different mechanisms and processes that are set off by introducing the anomalies either in the northern or the Southern Hemisphere.
4.1.1. Labrador Sea Salinity
[29] It seems straightforward that adding salt to the surface waters of the Labrador Sea triggers a strengthening of the North Atlantic meridional overturning circulation (MOC). It weakens the stratification, rises density and hence leads to a stronger deep water production. The amount of salt needed to increase deep water production appears to be rather low. There was no significant difference in the strength of the NADW cell between LAB+ (+1 psu) and a run with a positive anomaly of +0.5 psu (not shown) indicating an obviously nonlinear connection between salt anomalies and strength of deep water production. Both experiments exhibit a small cell with positive anomalies close to the southern slope of the Greenland-Scotland-Ridge, with an adjacent small negative anomaly (Figure 8 (top)). This indicates a northward shift of the overturning cell for these experiments. The entire cell south of 40°N is strengthened resulting in an intensified cross-equatorial transport. At the same time the cell extends to greater depths (almost 3.5 km instead of little more than 2.5 km in the control run (see Figure 4)). The altered overturning cells have a direct effect on the temperature and salinity fields of the Atlantic Ocean. As can be seen from the zonally averaged sections of Figure 9 the ocean at intermediate depths becomes colder and fresher. This is due to a subduction of Labrador Sea surface water that is cold and fresh compared to the normal conditions at these depths. As the overturning cell reaches deeper, Labrador Sea surface water is transported also into depths below 3 km. There it causes the deep ocean to become warmer and saltier. The effect of the strengthening of the Atlantic part of the AABW cell in the LAB+ case (approximately 0.3 Sv) can not be detected in temperature and salinity of the deep ocean as the increase of the NADW cell (by 6 Sv) is predominant there. The changes in the characteristics of the upper 1000 m can be explained by a shift of the pycnocline induced by changed upwelling and the stronger intrusion of subpolar mode waters. The global AABW cell (not shown) exhibits a slight weakening in the course of the model run. For LAB− the deviations from the control run are small. There is a negative anomaly in the meridional overturning south of the Greenland-Scotland-Ridge (Figure 8 (bottom)). This indicates a weaker overturning, which in turn leads to a weak temperature decrease in deeper layers (not shown). There is no appreciable salinity signal for that case.
[30] In run LAB+ the changes of the kinetic energy as shown in Figure 7 are dominated by the weakening of the ACC transport. There is no instantaneous reaction to the alteration in salinity restoring, but a gradual weakening. Parts of the Gulf Stream and the North Atlantic Current are deflected in LAB+ (Figure 10). The surface waters are transported more directly into the Labrador Sea to feed the strengthened NADW cell. In the LAB− experiment only minor changes can be observed (not shown). The overall weak response to the negative salinity anomaly in LAB− occurs because the freshwater cap does not directly cover the deep water formation site of the control experiment. The NADW cell weakens somewhat because fresher waters are advected to the adjacent convection sites with subsequent stabilization of the stratification there.
[31] The mechanism revealed by the LAB+ experiment can be described as follows: The density anomalies induced at depth by convection are advected southward with the Atlantic overturning motion and subsequently change the water mass properties in the Southern Hemisphere. The altered structure of the deep density field with decreased meridional gradients affects the transport of the ACC [Gnanadesikan and Hallberg, 2000; Borowski et al., 2002].
4.1.2. Weddell Sea Salinity
[32] In the Weddell Sea experiments the changes in the Drake Passage throughflow and the strength of the AABW cells are much more pronounced than in the NADW cells. In run WED+ (Figures 11a and 11c) the global AABW cell strengthens with its deep extremum in the Southern Hemisphere increasing from about 8 to more than 13 Sv. The AABW cell becomes thicker and extends now from 2600 m to the bottom. The positive anomaly along the continental slope of Antarctica indicates that the sinking into the deeper levels now takes places farther North. The Atlantic Ocean carries only a small share (less than one Sv) of this intensification in the abyssal levels. On the other hand, the negative anomaly in the global overturning extends to the North at intermediate levels. This part of the signal is dominated by the Atlantic MOC which shows a negative maximum of 3 Sv close to the Greenland-Scotland-Ridge and a weakening of the cross-equatorial transport by more than one Sv. In the WED− case (Figures 11b and 11d) the reaction of the ocean is far weaker. The global overturning (Figure 11d) shows a positive anomaly in the upper 3 km along the Antarctic continental slope indicating decreased sinking. This weakening, however, has almost no influence on the abyssal layers and the rest of the overturning stream function. Especially in the Atlantic there are no relevant changes (Figure 11b).
[33] An intensified deep water production south of the ACC results in subduction of cold (T < −1°C) and salty (34.9–35.1 psu) water in run WED+, thus changing the abyssal water mass properties accordingly (compare Figures 11e and 11g). The meridional density gradient of the Atlantic Ocean increases. At depths between 1 and 2.5 km temperature and salinity are almost unchanged. The WED− experiments yield slightly higher temperatures in the abyssal ocean, while there is only little change in the salinity properties there (Figure 11f and 11h).
4.2. Wind Stress Experiments
[34] Several studies with models of different degrees of complexity [Toggweiler and Samuels, 1993, 1995, 1998; Cai and Baines, 1996; McDermott, 1996; Gnanadesikan, 1999; Hasumi and Suginohara, 1999] have shown the fundamental sensitivity of deep water production and the global circulation to variations in Southern Ocean wind stress. The sensitivity to altered wind forcing is investigated here by analyzing the results of three experiments. To portray the manipulations carried out, the mean zonal component of the annual mean wind stress is displayed in Figure 12. The wind stress is amplified to 1.5 times of its amplitude for run TAU+, to half of its strength for TAU− and to zero for TAU 0 south of 50°S with a smooth transition to its regular strength at 30°S.
[35] The production rate of NADW (Figure 13), the cross-equatorial volume transport of the Atlantic deep circulation (not shown) and the extrema of the Atlantic AABW cell are approximately proportional to the factor by which the wind stress is multiplied. The maximum of the global AABW cell does not fit into the picture of a quasi-linear decrease with reduction of the wind stress at all: it has its maximum in the control run and values that are smaller by 20% for the TAU+ and TAU− cases.
[36] If one only takes into account the cases with nonvanishing wind stress (TAU+, CTRL and TAU−), a quasi-linear relation is also valid for the strength of the Drake Passage through flow (Figure 13). For TAU 0 the ACC stops almost completely (6.5 Sv), while a linear extrapolation of the other runs' behavior would yield about 50 Sv. The quasi-cessation of the ACC is in contrast to the results of Cai and Baines [1996], who report ACC strengths between 24 and 85 Sv and England [1993] (57 Sv) for experiments without wind forcing. A possible reason for this discrepancy could be a too short integration time in their model runs. In the experiment presented here the strength of the ACC is subject to a gradual decrease (compare Figure 17 (bottom)) that can be explained by a slow reduction of the potential energy stored in the ocean. As the timescale for the adjustment of the ACC's strength is larger than 1000 years, the use of rates of change of the global mean temperature and salinity as criterion for an equilibrium state as used by Cai and Baines [1996] may be misleading. On the other hand, potential energy is resupplied to some degree by the surface buoyancy fluxes. In our case, these fluxes are modified by altered ice conditions around Antarctica. The missing offshore wind components lead to a tendency for ice piling up at the coast. The thick ice cover, reinforced by continuous snow accumulation on the ice floes, prevents local freezing and thus the brine release into the ocean that is important for Antarctic bottom water production and its subsequent influence on the deep density gradients driving the ACC [Borowski et al., 2002].
[37] Figure 14 shows deviations of the North Atlantic surface velocities from the control run. For TAU+ (left) the Gulf Stream/North Atlantic Current system intensifies and penetrates further north, while in the TAU− run it is weakened and does not reach the Nordic Seas. For TAU 0 the Gulf Stream is weakened further compared to TAU− and takes a more zonal course (not shown).
[38] The altered circulation in the northern North Atlantic Ocean is also reflected in the meridional overturning. Figure 15 shows the temporal evolution of the global (left) and Atlantic (right) cells for run TAU+. The annual means for the first year after the wind anomalies have been applied show a strong signal with a positive anomaly in the Deacon cell, i.e. centered at 50°S and reaching down to the bottom of the ocean. After 10 years a strong positive anomaly at the Greenland-Scotland-Ridge has developed, while the anomaly in the Southern Hemisphere extends farther to the north. Another 10 years later the anomaly in the North Atlantic Ocean has weakened, broadened and moved southward, merging with the cell expanding from the South. Thus, the entire ocean basin up to 42°N is covered by a positive anomaly. The small cell vanishes within the next years (not shown). In year 510 the overturning anomalies have reached a “final” state that is characterized by a positive cell in the Atlantic Ocean with strongly enhanced sinking at the Greenland-Scotland-Ridge and regions of massive upwelling in the Southern Ocean.
[39] The changed deep water formation influences the water mass distribution in the Atlantic Ocean. The zonally averaged potential temperature and salinity deviations in the Atlantic Ocean show a warming of almost the entire basin north of 60°S for run TAU+ (Figure 16). Nearly the same region has also becomes altier. Both quantities exhibit basin-wide maxima in 1 km and 3.5 km depth. The only exceptions are the intermediate water tongues that show negative temperature anomalies and lower salinities, as well as the surface layers.
5. Discussion
5.1. Drake Passage Effect
[40] Long and short timescales in the models reaction to the wind stress alterations can be distinguished (Figure 17). An initial intensification of the NADW cell lasting for only 10 to 15 years corresponds with the appearance of the small-scale cell near the Greenland-Scotland-Ridge in Figure 15. The intense positive anomaly of year 10 at the Greenland-Scotland-Ridge can be interpreted as evidence for a thermohaline signal leading to a locally altered stratification. This contradicts the idea of the “Drake Passage effect” as causing a “purely passive transit” of wind induced mass transport into the Northern Atlantic with subsequent sinking there due to the presence of stable stratification elsewhere. After some years of weaker overturning the NADW cell develops continuously until it reaches a steady strength after approximately 300 years. The long term reaction of the North Atlantic deep water production can be explained by the “Drake Passage effect”. In run TAU+ stronger winds lead to an enhanced northward Ekman drift and stronger upwelling in the latitude of the Drake Passage (as can be deduced from the intensified Deacon cell). The volume transport out of the Southern Ocean into lower latitudes increases. In the Atlantic these waters move across the equator into the region of the Gulf Stream and the North Atlantic Current until they reach the deep water production regions. There, they contribute to an intensified sinking. The downwelling signal has arrived in the northern North Atlantic within 20 years (Figure 15). After this rapidly propagating signal, slower changes due to advection of different properties become important. For TAU− this mechanism works with reversed signs, thus being the basis of the quasi-linear part of the relationship outlined before.
[41] In the latitudes of Drake Passage, changes in the upwelling alter the meridional density gradient and thus the strength of the ACC on a very short timescale. The fact that its magnitude continues to increase slowly over hundreds of years in run TAU+ can be regarded as a further proof that the ACC is not primarily driven by direct vertical momentum transfer from the atmosphere to the ocean that would have an instantaneous effect. Instead, the mechanism works indirectly via changes of the water mass properties in the deep Southern Ocean, consistent with the findings of Gnanadesikan and Hallberg [2000] and Borowski et al. [2002].
[42] A somewhat different mechanism describing the enhancement of the thermohaline circulation due to increased wind stress has been proposed by Tsujino and Suginohara [1999] as result of their study with an idealized ocean basin model. They propose that a strengthened Ekman upwelling in wind driven gyres lifts the pycnocline there leading to enhanced downward heat conduction. This results in deep water of lower density that in turn causes enhanced buoyancy loss (cooling) in their deepwater formation region finally increasing the meridional overturning. Tsujino and Suginohara [1999] integrated their model for 6000 years and their results reflect a thermally and dynamically steady state in a system with one overturning cell. In our more realistic two cell scenario it is hard to distinguish the effects of the altered density structure on either of the two cells and/or the ACC. While we cannot rule out that a mechanism of the kind described by Tsujino and Suginohara [1999] is at work in our model, we doubt that besides the mentioned influence of the meridional density structure on the ACC these processes could be clearly attributed.
[43] The slow increase of the strength of the AABW cell in case TAU+ of Figure 17, though, could be caused by a mechanism of this kind. The AABW values still continue to change after the adaptations associated with the “Drake Passage effect” have led to new stable conditions in the magnitudes of NADW and ACC. The AABW cell does not reach a new equilibrium within 1500 years after the changes were applied. There is still a trend that needs to be explained by the global nature of the adjustment processes in the deep ocean. With new boundary conditions the model needs to adapt to a new equilibrium; it took the control run more than 3000 years to reach such an equilibrium state. It is not surprising, thus, that these adjustments are still to be completed after 1500 years for these strong changes in the model's boundary conditions.
[44] Besides the centennial and millennial timescales the model results also exhibit variability with higher frequencies. These are most prominent in the oscillations of the strength of the AABW cell. Also, the single NADW events that are present in all three wind stress runs shortly after the disturbances have been introduced into the model cannot be explained by advective processes. In this context, the theory of the “Drake Passage effect” by Toggweiler and Samuels [1993, 1995] can account only for a part of the results obtained here. There is no concise explanation we can offer for the identical sign of the anomalies in all three cases. The solution might just be attributed to a wave signal triggered by the “shock” of an abrupt change of the wind stress magnitude causing a remote change of the local stratification in the Northern Atlantic's deep water production regions.
[45] There are a number of uncertainties and limitations concerning the “Drake Passage effect” that should be kept in mind: the Deacon cell, for instance, is only present in a global stream function integrated zonally along depth levels. Döös and Webb [1994] pointed out that this cell virtually disappears if the integration is performed along density layers. They found in their model study that there is no mass transport across isopycnals associated with the Deacon cell. Furthermore, it can be doubted that the “Drake Passage effect” is robust at higher resolutions than the one used here; studies with high resolution models for the Southern Ocean [Olbers and Ivchenko, 2001] showed that the residual of the Deacon cell is very shallow (a few hundred meters) when including the effects of eddy mixing. Moreover, a more realistic topography than the one used here would allow a southward return flow at shallower depths eroding the constraint that the sinking of the water masses transported northward is only possible in the northern North Atlantic. Besides that, the prohibition of any net geostrophically balanced flow across the latitude band of the Drake Passage holds only for the zonal mean. There may well be bidirectional basin-scale geostrophic flows and small-scale eddies. It should be remembered also that in the present study no atmospheric feedbacks exist that would weaken the control that the “Drake Passage effect” exerts on the NADW according to Rahmstorf and England [1997].
5.2. NADW-AABW Seesaw and ACC Strength
[46] For a positive salinity anomaly in the Labrador Sea the intensified NADW production causes a decreased meridional density gradient in the Southern Ocean and thus weakens the ACC. The situation is different for the positive Weddell Sea anomaly. Here the density gradient and the ACC strength increase. As the Weddell Sea is closer to the latitude of the Drake Passage throughflow the reaction takes place faster and its amplitudes are bigger. The connection between strengthened NADW production and a weaker ACC and vice versa as present in the LAB and WED runs is in accordance with the model results of Mikolajewicz and Maier-Reimer [1990], who find a negative correlation between the two quantities. England [1993] attributes this interplay to adjustments in the deep density field on either side of the Drake Passage. Hirst and Cai [1994] and Cai and Baines [1996] report increases in both NADW and ACC strengths when manipulating the vertical diffusivities in their experiments. Their results do not contradict the statements above. The strengthening of the ACC in both cases is due to the deepening of the permanent pycnocline while the properties of the NADW don't change substantially.
[47] From the point of view of interhemispheric interactions, there are important differences between buoyancy forces affecting NADW and AABW. The LAB experiments show a distinct influence on the Southern Hemisphere, even for the LAB− run in which deep convection was not slowed down directly. This is due to the export of modified NADW into the South Atlantic and its subsequent incorporation into the global water cycle. For the WED runs the situation is different. AABW is, at least in part, exported across the equator, but it recirculates without intense effects onto the circulation or water mass properties of the Northern Hemisphere. The surface layers north of 30°S are hardly affected in the WED experiments, nor is the NADW cell. Thus, the interplay of NADW and AABW is surely not a linear one in the sense, that an intensification of one leads to a weakening of the other and vice versa. This connection could have been suspected from values for NADW and AABW in Table 4 for the buoyancy experiments (neglecting the exception LAB−, where the deep water production area has not been influenced directly as discussed before). This “seesawing” behavior of deep water production has been proposed by Stocker [1998] and Broecker et al. [1999] and was confirmed by the findings of Fieg and Gerdes [2001] in their model studies. The present model results contradict this paradigm. First, the cells evolve with different speeds. The NADW cell adjusts to a new equilibrium within about 500 years, while the AABW cell needs longer than the integration time performed here. Second, in the global zonal average the AABW cell behaves in a different way than in the Atlantic average. In the LAB+ run, for instance, the global cell intensifies, while the Atlantic part weakens. Reducing this reflection to the global AABW branch (as has been done in Table 4), the postulated relation of positive NADW anomalies taking place together with negative AABW and vice versa is only true after long integration times. In the temporal evolution of the LAB runs, there are extended periods when anomalies share the same sign. Third, the TAU runs show no relationship of the suspected kind at all.
[48] A possible mechanism for an interaction between NADW and AABW could involve the ACC. It has been demonstrated that changes in the deep water production rate of both cells alter the water mass transport of the ACC. On the other hand, it is difficult to detect an influence of the ACC onto NADW and AABW. However, an altered ACC will definitely influence its surroundings, e.g., by intensifying or weakening of the adjacent Ross and Weddell Sea gyres. A strengthening of the cyclonic Weddell Sea gyre, for instance, would lead to an elevation of the pycnocline, thus preconditioning the region for convection (following the argumentation of Marshall and Schott [1999]). Hence, an influence of the NADW cell via the ACC on the AABW cell appears to be possible. A similar conclusion has been drawn by Goodman [1998], who performed experiments with an OGCM with idealized ocean basins. For his run with an existing NADW cell he finds, that the AABW production does not depend on the NADW production. The strength of the AABW formation and the northward flow of bottom water is, however, affected by the transport in the ACC in his experiment. To what extent these conclusions reflect reality cannot be said from the results of the model runs performed here. The deep water production is parameterized in the model. In this respect the real process and the modeled process are very different. The model does not resolve the Antarctic shelf that is very important for the convective process. Therefore, this model does not permit a detailed conclusion on the nature of this specific feature, the interrelation of the ACC's magnitude and deep water formation in the Southern Hemisphere. Nevertheless, the possibility exists that the proposed mechanism is at work in reality.
6. Conclusions
[49] In the present work a coupled ocean-sea ice model is used to investigate the variability and interactions of water masses in the Atlantic Ocean, its thermohaline circulation and the role of the Southern Ocean as the link between ocean basins. A set of experiments manipulating the ocean's wind and thermohaline forcing is used to study the system's reaction to changes in boundary conditions.
[50] Changes that involve the abyssal ocean and its hydrographic properties can take far more than thousand years as the ocean needs to adapt to a new equilibrium state. This can be seen for example from the results of the model runs with altered wind stress forcing. A strengthened wind stress field over the Southern Ocean leads to increased upwelling due to stronger Ekman pumping. This causes a stronger northward transport of surface waters. In the Northern Atlantic Ocean the circulation patterns are altered, the waters protrude further North feeding the deep water formation regions. This mechanism thus causes changes in the strength of the NADW cell that in turn influence the magnitude of the ACC's water mass transport. The adjustment of the abyssal ocean associated with this mechanism was not completed within the 1500 years of integration in this model run. In the model run without wind stress over the Southern Ocean an almost complete cessation of the ACC was found. This contradicts the results of Cai and Baines [1996]. As the strength of the water mass transport in the ACC is subject to a slow gradual decrease after turning off the forcing, the integration of this experiment needs to be continued for more than thousand years, even if the deep ocean's global mean water mass properties (used as criterion for a new equilibrium by Cai and Baines [1996]) only show minor changes. The ongoing changes in the vicinity of the Antarctic continent that are due to changes in sea ice production rates are sufficient to significantly affect the strength of the ACC.
[51] The studies performed here show remote effects of the key quantities influencing the Atlantic Ocean. The experiments in which the buoyancy forcing is manipulated show the common feature that a strengthening of the NADW cell is associated with a weakening of the ACC. This is due to the altered water mass structure in the deep ocean causing a weakening of the meridional density gradients in the Southern Ocean. There are hints for a connection between the ACC strength and the AABW formation rate, that involves a change in the gyre systems south of the ACC.
[52] It should be emphasized, that the interaction between the components of the circulation system is not a mechanistic one: An intensification of the NADW cell does not automatically cause a weakening of the AABW cell, as could have been suspected from the results of previous studies [Stocker, 1998; Broecker et al., 1999; Fieg and Gerdes, 2001]. For example, in the model runs with altered wind stress the long term reaction of the NADW and AABW cells contradicts the behavior found in the above studies. Both cells show common strengthening (weakening) for intensified (decreased) wind stresses. Thus, a distinction must be made concerning the nature of perturbations: altered wind stress forcing in the Southern Ocean influences both Atlantic overturning cells directly via a change in the surface circulation patterns and their strength. Buoyancy forcing changes in the present experiments influence deep water production rates of one hemisphere leading to altered water mass properties in the interior of the oceans, that in turn exert an (indirect) effect on the overturning cell of the other hemisphere.
[53] Despite the insights gained from analyzing the results of the model runs, there are a number of remaining questions: why does the NADW cell's fast reaction show the same sign for positive as well as for negative wind stress anomalies? What causes the global and the Atlantic AABW cell to be out of phase in most of the experiments? What is the role of other ocean basins in this context? It also remains to be seen whether the features detected in this study are robust in further experiments with finer resolution models. However, it has become obvious that the interplay of the water masses in the Atlantic Ocean cannot be understood if the influence of the ACC is neglected.
Acknowledgments
[54] The authors would like to thank Gokhan Danabasoglu for providing the bottom topography data from his model. This work is based on parts of the Ph.D. thesis of the first author. We would like to thank Dirk Olbers for his support of this work. Remarks from Stefan Rahmstorf and an anonymous reviewer helped to improve the manuscript.
