Which physical processes effectively determine the stability regime of the Atlantic meridional overturning circulation (MOC) is not yet fully understood. We investigate the role of the oceanic freshwater transport into the basin, employing a coupled model of intermediate complexity. By modifying the longitudinal variation of surface salinities near the southern border at 33°S, the amount of salt flowing out of the Atlantic via the Brazil Current can be regulated. In turn, this will influence whether the MOC exports or imports salt water. The latter is associated with a basin-scale salinity-overturning feedback, which can be either positive or negative. Pulse experiments strongly suggest that its sign determines the existence of a monostable or bistable regime in our model.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 One of the key issues in climate studies is how the Atlantic MOC will respond to global warming and the associated changes in the hydrological cycle. These may effectively lead to fresher surface waters at high northern latitudes and weaken or possibly even shut down the MOC [Cubasch et al., 2001]. How to understand these varying model results is not an easy task. Simpler model experiments of imposing well-defined freshwater perturbations in the North Atlantic show that some coupled models [Manabe and Stouffer, 1988; Schmittner et al., 2002] exhibit a stable MOC shut down after removing the freshwater perturbation, while other models [Manabe and Stouffer, 1997; Schiller et al., 1997; Vellinga et al., 2002] show a restoration of the MOC on a timescale of hundreds of years. For box models the existence of regimes having either one or two stable steady states has been shown theoretically [Stommel, 1961; Rahmstorf, 1996]. Exactly why more realistic models show either one or two stable steady states is not yet fully understood.
 In the present paper we address this issue by examining the freshwater budget of the Atlantic basin, which consists of an atmospheric component (net evaporation) and an oceanic component (transport through the boundaries). Employing a coupled atmosphere/ocean/sea-ice model of intermediate complexity four different steady states are prepared having comparable net evaporation over the entire Atlantic. They exhibit varying relative magnitudes of the overturning and azonal components of the oceanic freshwater transport through the southern boundary. It will be shown that the states differ in the existence of a stable MOC shut down and that the Atlantic freshwater budget can be used as a diagnostic for the MOC stability regime.
2. Freshwater Budget of the Atlantic Basin
 First, we briefly discuss the components of the Atlantic freshwater budget. In equilibrium, the net basin-integrated result of evaporation E, precipitation P and continental run-off R is balanced by freshwater transports of the ocean at the southern border at 33°S and through Bering Strait,
The atmospheric contribution [E–P–R] is thought to carry a net freshwater transport out of the basin. The meridional overturning component Mov and azonal component Maz are defined by Rahmstorf  as:
where S0 is a reference salinity, the overbar and the brackets 〈.〉 denote zonal integration and zonal averaging, respectively, and v′ and S′ are deviations from zonal means. The other terms of equation (1) represent contributions from diffusion at 33°S and from the Bering Strait.
 The overturning component can be understood as the net freshwater transport carried by the MOC. It can be approximated as Mov ≈ −(1/S0)ΨΔS, where Ψ is the MOC strength and ΔS is the salinity difference between upper and deep waters at 33°S. The azonal component Maz incorporates both the export of surface and thermocline waters via the subtropical gyre as well as the flows coming in from the Indian ocean (“warm” water path) and Drake Passage (“cold” water path). The relative salinities and strengths of these flows will determine the magnitude of Maz and may vary considerably for different models. Using inverse-model data, Weijer et al.  have suggested that the present ocean has Maz = 0.38 Sv and Mov = −0.20 Sv. For negative values of Mov upper waters flowing northward freshen and the MOC exports freshwater, even though the Atlantic is a net evaporative basin. The “excess” salt is removed via the southern subtropical gyre.
3. Sensitivity Runs With Different Oceanic Transports at 33°S
 In the following we consider four sensitivity runs with the global coupled model ECBilt/CLIO. The atmospheric component ECBilt is a dynamic model (in T21L3 resolution), containing simplified parameterizations of the small-scale physical processes [Opsteegh et al., 1998]. CLIO consists of a free-surface ocean component (with 3° × 3° horizontal resolution and 20 vertical levels), including isopycnal diffusion, parameterizations of meso-scale eddies and dense-water overflows, and a thermodynamic/dynamic sea-ice component [Goosse and Fichefet, 1999]. The standard model (run 0) includes a freshwater flux reduction of about 0.2 Sv in the Atlantic area to compensate for excessive precipitation.
 In the four sensitivity runs (A–D) we aim to vary Maz by modifying the longitudinal distribution of the southern sea surface salinity (SSS). In the standard run, the SSS maximum near the South American continent is not very pronounced in contrast to observations. This is due to an over-estimation of the local precipitation. In runs A–D the basin-scale reduction was replaced by constant local corrections (summing up to 0.2 Sv) over the southern gyre, more specifically over the west of the Atlantic 17–33°S belt. In addition, run C (D) includes a west-east dipole correction of ±0.15 (0.25) Sv applied over the same region. States A–D were run for at least 1300 yr after a tuning phase using state 0.
 Modifying southern surface salinities clearly affects Maz (see Table 1). At the same time, net evaporation [E–P–R] (and its latitudinal variation) are very comparable for the four runs. As the azonal transport increases, Mov decreases and even changes sign in order to close the budget equation (1). Also the MOC weakens, as more salt flows out of the Atlantic via the Brazil Current and less salt flows equatorward inhibiting the NADW production. The diffusion freshwater transport is lower for runs C–D, because the imposed freshwater corrections smooth a SSS gradient along the African coast. The Bering Strait contribution varies little among the four runs. Our approach to vary Maz involved changing the southern salinity distribution. Similar changes in Maz and Mov could be achieved by applying modifications elsewhere in the global ocean and inducing changes in waters coming from the Indian ocean or from Drake Passage. For example, Seidov and Haupt  have shown that freshwater teleconnections are relevant for the working of the MOC by applying different inter-basin SSS gradients.
Table 1. The Atlantic Freshwater Budget According to Equation (1), the MOC Strength at 20°S and the Basin Maximum in the Standard Run 0 and in the Four Sensitivity Runs A–Da
All terms in Sv; averages over the last 100 years of the run.
 Contributions to Maz and Mov from the different water masses at various depths are depicted in Figure 1. Very saline surface and thermocline waters are found below which relatively fresh Antarctic intermediate waters lie (Figure 1a). The upper “limb” of the MOC consists of these waters flowing northward while undergoing transformations due to, for example, the surface forcing. NADW flowing southward lies between approximately 1200 m and 3300 m depth. Antarctic bottom waters (AABW) slightly fresher than NADW flow at the bottom. For increasing Maz NADW freshens, while upper level salinities increase (so that ΔS undergoes a sign change). “Excess” salt flowing out via the Brazil Current recirculates awhile in the subtropical gyre thereby increasing the local salinity. Significant contributions to Maz (Figure 1b) come from depths above 600 m. This is in accordance with the longitudinal variations of salinity being appreciable only at those depths. The imposed dipole freshwater corrections for runs C and D clearly result in a pronounced enhancement of Maz. For the different runs NADW and upper flow give a varying positive and negative contribution to Mov, respectively (Figure 1c). The AABW contribution to Mov is ≈0.03 Sv. Indeed, as AABW flows into the basin its salinity (Figure 1a) is lower than when it exits, so AABW exports salt water. The contribution at the bottom, instead of being negative for a single flow in the positive direction, is also positive. This mismatch is due to the chosen value for the reference salinity (S0 = 34.7 PSU) which is larger than the incoming AABW salinity. The obtained Mov for AABW is therefore an upper bound. Eddy transports account only for at most ±0.003 Sv.
4. Transition to a Stable Weak MOC
 The MOC imports freshwater (Mov > 0) in runs A–B, whereas it exports freshwater (Mov < 0) in runs C–D. This is also evident from Figure 1a, where the upper limb of the MOC is fresher (more saline) than NADW flowing southward in runs A–B (C–D). This implies that a MOC shut down will tend to salinify the basin in runs A–B, thus opposing the existence of a stable shut down. In contrast, a stable shut down may exist in runs C–D. We hypothesize that the sign of Mov indicates whether the MOC is in the monostable or bistable regime in our simplified general circulation model, as already argued by Rahmstorf  in the context of a box model.
 We examine this issue by applying a constant 0.4 Sv freshwater pulse over the Atlantic 50–70°N belt during 100 yr, then reducing the pulse strength instantly to 0.1 Sv and holding it constant for another 900 yr. After that, at a rate of 0.05 Sv/1000 yr the strength is further reduced to zero. The total amount of freshwater added to the system is comparable to that in earlier model experiments [e.g., Manabe and Stouffer, 1997; Schiller et al., 1997], but it is applied over a longer time period. As these experiments take a considerable amount of computing time, they are only carried out for states B and D. After the pulse experiment the runs are continued for another 1500 years for the states to achieve equilibrium. Both runs attain a collapsed MOC (Figure 2), with a shallow reversed cell in the southern Atlantic (of ca. −3 Sv). Run B starts to exhibit a gradual restoration during the last 1000 yr of the pulse experiment, when the freshwater perturbation is still positive. The MOC increases very rapidly at t = 2700 yr and then recovers fully. On the other hand, run D evolves toward a stable “weak” overturning state. The remaining overturning of 1 Sv flows at about 1000 m depth below the reversed cell typical of a collapsed MOC. Only a negative freshwater perturbation (that is, adding salt to the 50–70°N belt) results in a restoration to the initial strong overturning state (not shown).
 The different responses in run B and D can be explained from the net result of a number of feedbacks in our model. Most importantly, the cessation of MOC freshwater transport results in a positive SSS anomaly in the southern Atlantic in run B and a negative anomaly in run D. This is shown in Figure 3c (solid lines; mean response over yr 101–2000). The positive (negative) SSS anomaly in run B (D) gives rise to a positive (negative) anomaly in Maz at 33°S. The MOC shut down itself results in a negative (positive) anomaly in Mov of −0.05 (+0.11) Sv for run B (D), to which a positive contribution should be added of ca. +0.04 Sv due to the reversed cell exporting salt water out of the basin. The separate anomalous components ΔMov and ΔMaz are shown for run B and D in Figures 3a and 3b. The total anomalous oceanic transport is positive at first in both runs, which stabilizes the MOC shut down.
 The positive salinity anomaly in the southern Atlantic builds up in run B during the pulse experiment, while saline water is also transported northward by the gyre circulation. Starting at t ≈ 1500 yr ΔMov becomes increasingly more negative (exports more and more freshwater out of the basin), until at t ≈ 2500 yr the total anomalous transport becomes negative. The northward salinity advection becomes stronger as indicated by the SSS anomalies in Figure 3c (dashed line; mean response over yr 2600–2700) and there is a very strong MOC increase at t = 2700 yr.
 In both runs there are no significant changes in the basin-mean net evaporation. There is, however, a net evaporation dipole anomaly across the Atlantic equator in the collapsed state (Figures 3a and 3b), similar to earlier results [Schiller et al., 1997; Vellinga et al., 2002]. This atmospheric feedback provides an important MOC-restoring mechanism by forming a positive SSS anomaly in the northern region (Figure 3c). From here it may be transported northward by the subtropical gyre, thus favouring the restart of NADW production. In shorter pulse experiments (0.4 Sv during 100 yr) this feedback seems to cause a MOC recovery in either a few years (A–B) or about 100 yr (C–D) after termination of the pulse, see Figure 2 (inset). Interestingly, the progressive restoration times of each run match the decreasing sequence of Mov in Table 1. However, the time interval of the long freshwater pulse is sufficient to suppress this atmospheric feedback in favour of the feedback associated with the basin-wide salinification (freshening) in run B (D).
 An alternative method to determine the stability regime of the MOC is to compute hysteresis curves. Recently it was shown that, for a system in the bistable regime, the collapsed state is only found when the pulse is strong enough to make the system cross the unstable steady state [Dijkstra et al., 2004]. The present experiments illustrate that this is clearly not the case for the shorter pulse experiments. Here the system remains in the domain of attraction of the on-branch and returns to the strong-overturning state soon after the pulse is switched off. The long freshwater pulses do push the system onto the off-branch, reaching a quasi-equilibrium state after ca. 1000 yr (compare Figures 3a and 3b). We believe that the following slow decrease of the freshwater forcing is equivalent to computing part of the off-branch of the hysteresis curve. The transition to the on-branch (the first saddle-node bifurcation point) is reached for a positive (negative) forcing for run B (D), indicating that the initial state was in the monostable (bistable) regime for run B (D).
5. Summary and Conclusions
 We examined the role of the oceanic freshwater transports at the Atlantic basin's southern border in determining the stability regime of the MOC. Varying the azonal component Maz we obtained four steady states with similar basin-integrated net evaporation, but different meridional components Mov. The latter transport is a diagnostic for a basin-scale salinity-overturning feedback that may be either positive (Mov > 0) or negative (Mov < 0). This feedback has a longer timescale than the positive salt-advection feedback of the MOC “upper limb” waters in the northern Atlantic. Pulse experiments show that in our model a stable weak MOC steady state exists only when Mov < 0. This strongly suggests that the sign of Mov is closely linked to the system's dynamical regime. If this result holds, it would be possible to assess a priori the stability regime of the ocean simply by measuring or calculating Mov from the flow and salinity properties of waters at the latitude of Cape Hope.
 PdV was funded by the European Project EVK2-2001-00263.