1. Introduction
The Atlantic meridional overturning circulation (AMOC), defined here as the integrated meridional transport in the Atlantic as, for example, in Wunsch (2002), plays an essential role in the maintenance of the Northern Hemisphere climate as it transports a substantial amount of warm and saline waters poleward. Ganachaud and Wunsch (2000) estimated the maximum value of the northward heat transport to 1.3 PW around 25°N, associated to a volume of 15 ± 2 Sv (1 Sv ≡ 106 m3 s−1) of North Atlantic Deep Water circulating in the Atlantic. Talley et al. (2003) recently estimated this volume to be 18 Sv. Variations of this intensity are likely to significantly alter the climate in the North Atlantic (e.g., Manabe and Stouffer 1999). Direct observations are still lacking (Siedler et al. 2001), but modeling studies suggest that the AMOC substantially varies on the decadal to centennial time scales. The relative importance of atmosphere forcing, air–sea feedback, oceanic feedback, and nonlinearities in generating this variability remains however to be clarified. Decadal to centennial oscillations linked to oceanic processes have been found in oceanic general circulation models (OGCMs) using idealized surface boundary conditions or coupled to a simple energy balance atmospheric model (e.g., Mikolajewicz and Maier-Reimer 1990; Weaver et al. 1991; Chen and Ghil 1996; among many others). A substantial interdecadal variability is also found in more realistic climate models. Delworth et al. (1993) suggested that North Atlantic fluctuations with a dominant time scale of approximately 50 yr appeared in the Geophysical Fluid Dynamics Laboratory (GFDL) model because salinity anomalies advected toward the high latitudes were influencing the strength of the meridional overturning circulation. A phase lag between the horizontal and the meridional circulations was responsible for the quasi-oscillatory behavior, but the variability appeared to be primarily forced by the natural variability of the surface heat flux, without significant back action on the atmosphere (Delworth and Greatbatch 2000). On the other hand, Timmermann et al. (1998) proposed a coupled mechanism to explain a 35-yr mode of North Atlantic variability in the ECHAM3–LSG model. When the AMOC was anomalously strong, a warm sea surface temperature (SST) anomaly appeared in the North Atlantic, and its impact on the atmosphere resulted in a strengthened North Atlantic Oscillation (NAO) and hence decreased evaporation and Ekman transport in the Greenland Sea and off Newfoundland. This led to negative sea surface salinity (SSS) anomalies in the oceanic sinking regions, weakening the deep convection and subsequently the AMOC, thus reversing the changes. A different coupled mechanism was suggested by Vellinga et al. (2002) for the Third Hadley Centre Coupled Ocean–Atmosphere General Circulation Model (HadCM3). When the AMOC was anomalously weak, the North Atlantic was anomalously cold and the tropical South Atlantic was anomalously warm, which displaced the intertropical convergence zone southward. There was thus less precipitation north of the equator, which reduced the northward oceanic freshwater transport and led after about six decades to more saline surface conditions in the North Atlantic. This increased the AMOC and reversed the changes. Remote atmospheric forcing may also lead to a substantial AMOC variability, as shown by Latif et al. (2000) and Thorpe et al. (2001) in global warming conditions. In both models, the global warming led to a freshening and warming of the high latitudes, but positive SSS anomalies were created in the tropical Atlantic by enhanced large-scale air–sea interactions with the tropical Pacific. The AMOC was thus stabilized by the compensating advection of high salinity anomalies into the sinking region. In Latif et al. (2000), the interactions between the tropical Atlantic and Pacific were similar to those operating during present-day El Niño–Southern Oscillation (ENSO), but whether ENSO could have a direct impact on the AMOC in present-day conditions has not been established.
In summary, the strength and the nature of the AMOC coupling with the atmosphere remains an open question. In this paper, we focus on the response of the AMOC to the atmospheric forcing in a control run with the Bergen Climate Model (BCM). The simulation is presented in section 2. The AMOC variability is briefly described in section 3, as well as its link with deep convection in the North Atlantic. The influence of the NAO is investigated in section 4. In section 5, we show that ENSO could have a delayed impact on the model AMOC. Conclusions are given in section 6.
2. Data
The BCM is described in Furevik et al. (2003), where specific details on the model and experiment setup can be found, so that only a brief outline is given here. The model couples the Action de Recherche Petite Echelle Grand Echelle/Integrated Forecasting System (ARPEGE/IFS) atmospheric general circulation model (Déqué et al. 1994) in TL63 resolution (2.8° along the equator) with 31 vertical levels to a global, 24-layer version of the Miami Isopycnic Coordinate Ocean Model (MICOM; Bleck et al. 1992) that incorporates the dynamic and thermodynamic sea ice modules of Drange (1999). The oceanic grid has a 2.4° zonal resolution. The meridional resolution is 2.4° except equatorward of about 10° of latitude, where it gradually increases to 0.8°. The data on isopycnal levels were interpolated on the Levitus z-levels with a simple first-order remapping scheme. After spinup of 20 yr, the BCM was integrated for 300 yr with fixed heat and freshwater flux adjustments that contribute to the realism of the model climate state and reduce the model drift. As described by Furevik et al. (2003), the (seasonally varying) flux adjustment had been estimated in a prior run where SST and SSS were restored to the observed climatology. The time series of the maximum of the Atlantic meridional streamfunction in the North Atlantic between 10° and 70°N, and between 500 and 5000-m depth, hereafter called the Meridional Overturning Index (MOI; see section 3) is shown in Fig. 1. As in Mignot and Frankignoul (2004), we only considered the middle 200 yr of the simulation. The last 50 yr were not considered, as the meridional overturning streamfunction started drifting during the last 10 yr—presumably because of a slight imbalance in the flux adjustments in the northern North Atlantic—nor were the first 50 yr, in order to leave more time for the ocean to reach an approximate steady state (a very conservative choice).
The main characteristics of the AMOC have been discussed by Bentsen et al. (2004). Here, we focus on the AMOC variability on annual to multidecadal time scales in the 200 yr of data. As a drift may influence the lowest frequencies, a third-order polynomial was first removed by least squares fit from each variable at each grid point. This is illustrated for the MOI in Fig. 1, which shows that only centennial and longer time scales are affected by the detrending. The shorter (interannual) time scales are investigated below using the detrended annual means, and the longer (decadal) ones by further filtering periods smaller than 10 yr with the continuous wavelet transform technique (e.g., Daubechies 1992) from the detrended annual means. Monthly anomalies were also considered by subtracting the mean seasonal cycle from the detrended monthly averages. Significance of correlations between annual (decadal) time series are calculated using the Student’s t test assuming independent samples at a 1-yr (5-yr) interval.
3. The natural variability of the Atlantic meridional overturning circulation
The 200-yr mean Atlantic meridional streamfunction is shown in Fig. 2. It has features similar to the mean AMOC inferred from observational estimates (Ganachaud and Wunsch 2000; Talley et al. 2003): about 18 Sv of warm water flows northward in the upper ocean, mainly in the Gulf Stream and North Atlantic Current, sinks at high latitudes, and returns southward at depth as North Atlantic Deep Water (NADW). The simulated poleward transport of Atlantic Water across the Greenland–Scotland ridge (8.2 Sv) is very close to observation-based estimates, as discussed by Otterå et al. (2003). Note that the overflow of very deep waters over the ridges between Greenland and Scotland and the associated changes of water mass properties occurring northward appear more clearly when the streamfunction is represented as a function of density rather than depth (Gao et al. 2003). The deep circulation cell of Antarctic Bottom Water (AABW) that should lie beneath the NADW is not simulated. This is a common default of isopycnic models that use a reference pressure at the surface when computing potential density, which results in an incorrect representation of the cold deep waters from the Antarctic (DYNAMO Group 1997). Although interactions between NADW and AABW are still poorly known, we focus on the NADW cell variability and its link with the atmospheric forcing, and this deficiency should have limited influence.
In their analysis, Bentsen et al. (2004) characterized the natural variability of the AMOC using the time series [principal component (PC)] of its leading empirical orthogonal function (hereafter PC1). However, the latter may be too constraining to fully represent the AMOC variability as it refers to a fixed spatial pattern. Thus, we use instead the (detrended) time series of the maximum value reached by the AMOC between 10° and 70°N, and 500- and 5000-m depth, which can be seen in Fig. 1. In the BCM simulation, the latitude of the maximum is almost constant in time, around 23°N, and its value is strongly correlated (0.8) to PC1. The results below are thus very similar to both time series. Following Delworth et al. (1993), we refer to our index as the Meridional Overturning Index. Most of its variance is found at decadal time scales, but its frequency spectrum has no significant peak on time scales shorter than 50 yr (not shown). Note that the time series is too short to see oscillations on longer time scales.
To investigate the variability of the AMOC, we regressed it on the MOI as a function of time lag (Fig. 3). A significant anomalous positive overturning cell appears at high latitudes about 5 yr prior to the AMOC maximum and progressively expands to the south, occupying the whole North Atlantic basin by lag −1 (cf. lags −2 and 0 in the figure). The maximum anomaly keeps progressing southward and loses significance at high and midlatitudes after a lag of 4 yr. As the AMOC variability seems to originate from the high latitudes, it is likely to be linked to deep convection in the North Atlantic. To account for the intermittent character of the deep convection phenomenon, the deep convection sites were defined by a criterion on mixed layer depth variability, namely where the standard deviation of the mixed layer depth in March exceeds 320 m. Results are identical when using as a criterion that the maximal depth reached over the average annual cycle exceeds 1000 m (Fig. 4). Note that we use the mixed-layer depth directly coming from the isopycnal integration, as detailed in Furevik et al. (2003). As in Bentsen et al. (2004), three deep-water formation sites are identified: the Labrador Sea; the Greenland, Iceland, and Norwegian (GIN) Seas; and the Irminger Sea. Maximum mixed layer depths in the model are reached in the Irminger Sea. Observational evidence for deep convection in this area is recent (Bacon et al. 2003; Pickart et al. 2003b), and it seems to only reach depths of about 1000 m, versus 2000 m in the model. In the Labrador and GIN Seas, deep convection has often been observed (e.g., Clarke and Gascard 1983; Schott et al. 1993), reaching about a 2000- and 1500-m depth, respectively (Marshall and Schott 1999). In the model, deep convection is also found at these locations, but it is too shallow by about 400 m.
Dickson et al. (1996) showed in the observations that a positive phase of the NAO enhances deep convection in the Labrador Sea but reduces it in the GIN Seas, and vice versa. In the BCM simulation, the atmospheric patterns associated with deep convection in the two areas are indeed very similar to the NAO (see Bentsen et al. 2004, their Fig. 9), and the winter mixed layer depth is well correlated with the NAO time series (Fig. 5). The NAO index is defined here as the time series of the leading empirical function of the sea level pressure (SLP) over the North Atlantic sector, which compares well with the observations (Furevik et al. 2003). It typically varies on time scales of a few weeks, as in the observations (Feldstein 2000), and has only a little month-to-month persistence and none from one year to the next in the model. Monthly anomalies indicate that the NAO leads enhanced (weakened) deep convection in the Labrador (GIN) Sea by 1 month (not shown).
Deep convection in the Irminger Sea might also occur during positive NAO phases in the observations (Pickart et al. 2003a). In the model, however, it is only weakly related to the NAO (Fig. 5), presumably because of the limited spatial resolution. Instead it seems to respond to a complex series of atmospheric conditions in the North Atlantic, starting on the average 3 yr prior to a deep convection maximum (Fig. 6). The anomalous SLP pattern first resembles a negative NAO phase (lag −3), but it then mostly reduces to a low pressure monopole centered west (lag −2) and then east (lag −1) of Iceland. As shown by the arrows in Fig. 6, the cyclonic circulation induces significant anomalous cold northerly winds and evaporation over the Irminger Sea (see also Bentsen et al. 2004). The surface waters become colder and saltier, and they sink when dense enough. As opposed to the Labrador and GIN Seas where deep convection is forced by short-term changes in a standing atmospheric pattern (the NAO), the Irminger Sea deep convection is thus forced by a long-term succession of atmospheric patterns. It is interesting to note that in the GFDL coupled model, anomalously cold SSTs in the Irminger Sea are also associated with surface pressure and northerly wind anomalies that have their largest amplitude approximately 2 yr before (Delworth et al. 1997). In both models, the northerly winds increase the East Greenland current by lag −1, but SSS anomalies are advected from the Arctic in the GFDL model while no evidence of an advective mechanism is found in the BCM, where density anomalies are mostly created by the local atmospheric fluxes, as described above.
The strength of the cross correlations between the oceanic surface characteristics in the deep convection areas and the MOI shown in Fig. 7 indicates that in the BCM simulation, the AMOC is most sensitive to deep convection in the Irminger Sea, where an anomalously deep, dense, salty, and cold mixed layer leads the MOI by about 5 yr. The correlations between the MOI and the deep convection parameters in the Labrador and GIN Seas are much weaker, suggesting that they play a lesser role. Bentsen et al. (2004) have shown that the high-latitude AMOC anomalies propagate southward in the Atlantic basin as a coastally trapped wave, as in Kawase (1987) and Yang (1999). The time scale is however longer since Kelvin baroclinic waves are not present in coarse resolution models where the Rossby radius is not resolved but replaced by viscous boundary waves (e.g., Hsieh et al. 1983; Killworth 1985), which have a much slower propagation speed. This may explain why the perturbation takes about 6 yr to reach the equator. The exact time scale also probably depends on the resolution and the type of grid. Note that it is of the same order of magnitude as that found by Eden and Willebrand (2001) in a low-resolution OGCM.
4. The Atlantic meridional overturning circulation and the NAO
Previous modeling studies have suggested that the NAO has a strong impact on the thermohaline circulation (THC) through mechanical wind or buoyancy forcing (Timmermann et al. 1998; Delworth and Greatbatch 2000; Eden and Willebrand 2001). Since in the BCM simulation the NAO triggers deep convection in the Labrador and GIN Seas (Fig. 5), it is of interest to investigate its influence on the AMOC. Regressing the AMOC on the monthly NAO time series only shows significant correlation when the AMOC is in phase with the NAO or follows by a few months. The monthly AMOC anomaly that covaries with the NAO is very significant (Fig. 8) and is essentially due to Ekman pumping. During a positive NAO phase, the Ekman pumping is anomalously negative between 32° and 55°N as a consequence of the convergence of the Ekman currents due to the intensification of the westerlies and the trade winds, and anomalously positive elsewhere. The horizontal circulation is maximum at the surface and almost constant beneath the mixed layer, suggesting a barotropic response to the wind forcing, as in Eden and Willebrand (2001) and Hakkinen (1999). The barotropic response persists little, about 1 month at extratropical latitudes and up to 4 months in the subtropics, with a weak southward propagation.
To search for the baroclinic AMOC response that should appear at longer time lags, the same analysis is repeated with annual means (Fig. 9). At lag 0, we primarily detect the rapid wind-induced barotropic response (left). When the AMOC follows the NAO by 1 yr (middle), the AMOC response is composed of two cells that are almost opposite to the ones in phase. A large positive cell that is weak at the surface and maximum between 1000 and 2000 m is centered at 50°N and persists up to lag 3. A weak negative cell is located farther south around 20°N, with a maximum near 500 m, but significance is lost by lag 2. This pattern resembles the baroclinic response to the NAO winds found by Eden and Willebrand (2001), except that it only appears after 3 yr in their model. Note that the associated horizontal currents (not shown) resemble the gyre variability discussed by Curry and McCartney (2001) and Marshall et al. (2001). No significant signal is found when the AMOC leads.
In spite of their different structure, the NAO-induced AMOC patterns are not independent from that associated with the MOI, as defined by lag 0 in Fig. 3: the spatial correlation is 0.64 for the in-phase barotropic structure of Fig. 9 (left) and −0.56 for the baroclinic structure at lag 1 (Fig. 9, right). Consistently, the NAO index and the MOI are significantly correlated in phase and weakly anticorrelated at lag 1 (Fig. 10). When the data are low-pass filtered (section 2) to focus on the decadal time scales, the correlation between NAO and MOI is small (Fig. 10, dashed line), and correspondingly, the regression of the AMOC on the NAO time series shows no significant signal. We speculate that it is primarily due to a near compensation at low frequency between the barotropic and the baroclinic responses seen in Fig. 9. This differs from Timmermann et al. (1998) and Delworth and Greatbatch (2000), who suggested that low-frequency changes in the NAO force a basin-scale acceleration of the THC via persistent buoyancy fluxes at high latitudes that enhance deep convection. However, the deep convection in these models was occurring south of Greenland, where the buoyancy forcing of the NAO was strong, while in the BCM the main convection site is in the Irminger Sea and is not directly forced by the NAO. Note that we do not exclude that the baroclinic response in Fig. 9 may be driven in part by buoyancy forcing as well, but our two-cell response widely differs from the buoyancy forced basin-scale cell seen in the other models.
5. The Atlantic meridional overturning circulation and ENSO
a. ENSO in the BCM
In the BCM simulation, the ENSO phenomenon is strong and rather realistic. It is well described by the first mode of SST variability in the tropical Pacific (not shown) that explains 55% of the monthly SST anomaly variance between 12°N and 12°S, versus 64% in the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis. The mode time series, or ENSO index (positive for a warm El Niño phase) is anticorrelated at a 17-month lag, versus 20 in the NCEP–NCAR reanalysis, and is more strongly autocorrelated at a 3–4-yr lag. Its spectrum (Fig. 11, top) thus shows significantly enhanced variance around 35 months. In the Indo-Pacific, the associated SLP structure compares rather well with the observed Southern Oscillation even though the eastern Pacific lobe extends farther westward along the equator (Fig. 12). In the North Atlantic, however, the SLP signal is 2.5 times too strong in the BCM, and the observed southwest–northeast pattern is replaced by a dipole that has some similarity with a negative phase of the NAO, except that it is shifted northward and strongly dominated by the southern pole. The too-strong impact of ENSO on the North Atlantic is a known deficiency of the ARPEGE atmospheric model (Cassou and Terray 2001). It results in an unrealistic in-phase anticorrelation (r = −0.35) between the NAO and ENSO indices that makes it harder to distinguish between NAO and ENSO influences on the northern North Atlantic, as discussed below. On the other hand, the ENSO signal is fairly well reproduced in the tropical and subtropical Atlantic, and the ENSO-induced anomalous precipitation pattern is significant and realistic, as shown in Fig. 13 by a comparison with the NCEP–NCAR reanalysis (for clarity, we did not indicate statistical significance). During the late rainy season in the Caribbean Sea (August–November; Fig. 13, left) preceding a mature ENSO phase (December–January–February), a divergent surface flow dominates the eastern tropical Atlantic, contributing to anomalously weak precipitation over the Caribbean, as in the “dry Caribbean–ENSO relationship” of Hastenrath (1976). During the following months (middle), high pressure over the equatorial Atlantic reduces the meridional SLP gradient, yielding weaker trade winds and a gradual warming of the sea surface, consistent with Enfield and Mayer (1997). The warm SST anomaly reaches its maximum 4–6 months after the mature ENSO phase (i.e., in boreal spring of the ensuing year; right) and is associated with a wetter Caribbean at the start of the new rainy season, as discussed in Giannini et al. (2000). This realistic behavior of ENSO in the tropical Atlantic is of particular interest in view of the previous studies that have linked ENSO to the AMOC (e.g., Latif 2001).
b. Interannual time scales
Lagged regression based on annual means shows no significant link when the AMOC leads ENSO. In phase with ENSO events (Fig. 14, left), the AMOC is weaker at midlatitudes (typically by 0.3 Sv), as during a negative NAO phase but with less intensity (Fig. 9, left), consistent with the anticorrelation between the two atmospheric indices, and mechanical forcing by anomalous Ekman pumping (Fig. 14, top). The difference with the NAO case is that the AMOC southern cell extends less toward the Tropics, and there is an opposite cell at 15°S. The correlation loses significance 1 yr after ENSO, but 2 yr after (lag 2), a significant response of the opposite sign and slightly shifted southward appears with a similar intensity (Fig. 14, middle). There is some similarity with the baroclinic response to the NAO, but the pattern persists much longer and slowly propagates southward, so that 6 yr after ENSO (right), the weakening of the meridional circulation is centered at 35°N, resembling to some extent the lag-0 barotropic response, but without significant positive cells. Significance is lost by lag 7. Consistent with this picture, the MOI and ENSO index are significantly correlated in phase and also when the MOI lags by 6 yr (Fig. 15). The AMOC weakening following a positive ENSO event persists longer than that following a negative NAO phase because the ENSO forcing is more persistent and affects the deep convection sites differently. As shown in Fig. 16, the ENSO teleconnections simultaneously reduce the deep convection in the Labrador and Irminger Seas without significantly affecting the GIN Seas, while a negative NAO phase reduces the deep convection in the Labrador Sea and enhances it in the GIN Seas, without much effect in the Irminger Sea (see Fig. 5, right). The effect of ENSO might in addition be further prolonged by the advection of a negative salinity anomaly created in phase off Newfoundland and that propagates toward the Irminger Sea in 3–4 yr, as described in Mignot and Frankignoul (2004). In conclusion, ENSO forcing explains why Bentsen et al. (2004) found that regimes can exist in the BCM simulation where deep convection in the Irminger and Labrador Seas act together.
c. Decadal time scales
To filter out the fast response to the wind forcing that may obscure slower changes, we now consider the data filtered at decadal time scales. We thus now consider the response to decade-long ENSO fluctuations or to successions of shorter events of the same signs, both referred to as decadal changes of ENSO. Contrary to the NAO case where statistical significance was lost, low-pass filtering strongly enhances the AMOC weakening that follows ENSO events, and it makes its slow southward propagation more visible (not shown). The AMOC weakening is initiated at high latitudes, reaches its maximum (0.5 Sv) about 4 yr after decade-long warm ENSO events or decadal succession of short warm ENSO events, and expands to the Southern Hemisphere after about 8 yr, consistent with the AMOC propagation in Fig. 3.
The cross correlation between the low-pass-filtered ENSO and MOI time series confirms the strong anticorrelation when ENSO leads by about 4 yr (Fig. 17) but, in addition, shows a significant positive correlation when ENSO leads by about 39 yr. Note that this peak is significant (at the estimated 10% level) independently of the trend removal, although artificially large correlations are found at positive lags when no trend is removed. It is also significant when periods lower than 5 yr only are filtered out. To document the associated patterns, the global SST field was regressed on the MOI with the same time lag (Fig. 18, left), showing indeed a significant positive SST monopole in the tropical Pacific that resembles the model El Niño.
To confirm the relationship between tropical Pacific SST and the AMOC, we used a lagged maximum covariance analysis (MCA) based on a singular value decomposition. As in Frankignoul et al. (2001), statistical significance was estimated using a moving block bootstrap approach. The first covariance mode between the low-pass-filtered global SST and low-pass-filtered AMOC is highly significant when AMOC follows SST by a few years, confirming that a positive ENSO phase is followed by a basin-scale weakening of the AMOC (Fig. 19). No other significant first mode was found, but the second mode of covariability is significant when SST leads the AMOC by 36 yr. The associated patterns (Fig. 20) show that a positive ENSO phase is followed by a dipole-like AMOC fluctuation with anomalously strong meridional circulation at high latitudes and an anomalously weak one in the Tropics. This structure is very close to the second EOF of the AMOC (not shown), and it has some similarity with the lag −4 pattern in Fig. 3. Based on the latter, we can thus expect a basin-scale acceleration of the AMOC about 36 + 4 = 40 yr after a decade with warm ENSO events, as seen in Fig. 17. However, no significant covariance could be found near this lag, possibly because the simulation is too short for the signal to be robust when only using SST data.
We saw that warm ENSO events lead to a salinification of the western tropical Atlantic via anomalous freshwater fluxes (section 5a), and the surface waters are indeed anomalously salty 39 yr prior to the AMOC maximum (Fig. 18, right). The delayed ENSO influence could thus occur via salinity advection, as suggested in the observations by Latif (2001). To investigate this mechanism in the BCM, we released fictitious Lagrangian drifters in the surface mixed layer at the center of the positive haline anomaly in the tropical North Atlantic and followed their advection by the model climatological monthly currents, as in Mignot and Frankignoul [2004; see Blanke and Raynaud (1997) for a description of the Lagrangian advective tool]. The trajectories of the floats calculated over 40 yr are shown in Fig. 21 as a function of depth. The releases were distributed throughout the year, but, for clarity, we only show the trajectories of particles released in January and July. The time step used for the calculation is 10 days. Most drifters propagate twice around the subtropical gyre, first remaining in the mixed layer, then subducting in the center of the gyre and recirculating at greater depth in the Caribbean Sea and along the coast of the United States. Some drifters then escape the subtropical gyre along the North Atlantic Current and are seen to finally reach the high latitudes, essentially the Irminger and Labrador Seas, between 500- and 1500-m depth. We emphasize that these trajectories were calculated using the monthly climatological currents of the model and are thus independent of the filtering applied to the data. They show that a 40-yr delay between decade-long warm ENSO events or a decadal succession of short warm ENSO events and a weakening of the AMOC is consistent with the northward advection of salinity anomalies created by ENSO in the tropical Atlantic. The good correspondence between the time scales found in the two independent analyses gives credit to the proposed long-term link. The propagation of salinity anomalies was difficult to follow. However, the lagged regression of the anomalous potential density at 600 m on the ENSO time series indeed shows that the water becomes anomalously dense in the North Atlantic about 35 yr after the warm ENSO events (Fig. 22). This could destabilize the water column and, presumably, favor deep convection.
6. Conclusions
In the BCM, the natural variability of the AMOC is dominated by a basin-scale adjustment to changes in the deep convection in the Irminger Sea. Although the relevance of deep convection in the Irminger Sea has been confirmed by recent observations (Pickart et al. 2003b; Bacon et al. 2003), its relative importance seems overestimated, as the Labrador and GIN Seas are generally considered as the main areas of deep-water formation in the northern North Atlantic (e.g., Lazier et al. 2001). In the model, enhanced deep convection in the Irminger Sea is due to a succession of atmospheric patterns that bring anomalous northerly winds over the area, evaporation, and cooling. Unlike in the observations (Pickart et al. 2003a), the atmospheric forcing is not linked to the NAO. Note however that observations are partly influenced by noise and perhaps also by changes in radiative forcing, composition of the atmosphere, etc.
The influence of the NAO on the AMOC was explored. Although a positive NAO phase enhances deep convection in the Labrador Sea and inhibits it in the GIN Seas (and conversely), as in the observations, it primarily influences the AMOC through the mechanical action of the wind. The AMOC is indeed rapidly intensified south of 40°N and weakened to the north, consistent with a barotropic response to NAO winds. The baroclinic response appears 1 yr later and almost has the opposite polarity. At longer time scales, the barotropic and baroclinic responses thus largely compensate, and the AMOC response to the NAO is weak. We found no evidence of a significant low-frequency impact of the NAO through buoyancy forcing. This differs from previous studies of Timmermann et al. (1998) and Delworth and Greatbatch (2000), where the buoyancy forcing associated to the NAO caused a basin-scale change of the AMOC. The weaker NAO influence in the BCM is presumably due to the strength of the deep convection in the Irminger Sea, which is not directly affected by the NAO in the model. Although of questionable realism, this result emphasizes that the location of deep convection and the high-latitude atmospheric forcing must both be well represented in coupled models in order to correctly simulate the AMOC variability.
Finally, we investigated the link between ENSO and the AMOC. In the BCM, the ENSO phenomenon is fairly realistic, but its influence on the atmospheric circulation in the North Atlantic sector is too strong, and ENSO and NAO are significantly anticorrelated, unlike in the observations. As a result, the ENSO teleconnections affect the AMOC through both wind forcing and a direct impact on deep convection in the Labrador and Irminger Seas, resulting in a more persistent influence—a slower overturning following a positive ENSO phase by about 6 yr. However, as these impacts are directly linked to the artificial ENSO influence on the northern North Atlantic, they are unlikely to be realistic.
Some evidence of more realistic influence of ENSO on the AMOC was found at low frequency with a time lag of about 40 yr. The mechanism is based on the advection of salinity anomalies that appear in the tropical Atlantic during El Niño events, consistent with the observed ENSO impact that was suggested by Latif (2001). Their northward propagation is primarily due to the mean oceanic circulation and takes approximately 35 yr. The salinity anomalies would first circulate in the subtropical gyre, then subduct, and recirculate at depths along the coast of northern America. They would then be in part advected along the North Atlantic current, eventually reaching the Irminger and Labrador Seas, where they tend to destabilize the water column and thus presumably enhance deep convection, hence the AMOC. However, this delayed ENSO impact was difficult to detect in only 200 yr of model data, and it should be investigated further in longer simulations or through specific experiments in forced conditions. It is interesting to note that several sensitivity experiments with coupled models highlight a similar mechanism. Latif et al. (2000) described a scenario simulation where the AMOC is stabilized in global warming conditions by the advection of salinity anomalies from the tropical Atlantic that were created by strong ENSO events. A similar mechanism is suggested by Thorpe et al. (2001), except that it does not involve ENSO. In a BCM simulation perturbed by an artificial freshening in the high latitudes, Otterå et al. (2003) observed a stabilization of the AMOC because salinity anomalies that were created by the weakening of the Guyana Current in the tropical Atlantic were advected northward, countering after about 50 yr the imposed freshening. Subtropical salinity anomalies due to a shift of the ITCZ position acted similarly in two simulations with the HadCM3 model, one in control conditions (Vellinga and Wu 2004) and one perturbed by an anomalous freshwater anomaly in the high latitudes (Vellinga et al. 2002). Our study does not tell whether salinity advection from the tropical Atlantic would be sufficient to stabilize the AMOC in a global warming scenario with the BCM, but it emphasizes its possible role in the long-term variability of the thermohaline circulation.
Acknowledgments
We thank Helge Drange, Mats Bentsen, and Ina Karin Kindem for providing the BCM run, and Bruno Blanke and Daniele Iudicone for making available the Lagrangian tool. Thanks are also due to the reviewers, whose comments led to an improved presentation of the work. Useful discussions with Gilles Reverdin and Jean-Luc Mélice are acknowledged. The NCEP–NCAR reanalysis data were provided through the NOAA Climate Diagnostics Center (http://www.cdc.noaa.gov/). This research was supported in part by the PREDICATE project of the European Community, the PNEDC, and the Institut Universitaire de France. J. M. was supported during her doctoral thesis by a scholarship from the DGA–CNRS.
REFERENCES
Bacon, S., W J. Gould, and Y. Jia, 2003: Open-ocean convection in the Irminger Sea. Geophys. Res. Lett., 30 .1246, doi:10.1029/2002GL016271.
Bentsen, M., H. Drange, T. Furevik, and T. Zhou, 2004: Simulated variability of the Atlantic meridional overturning circulation. Climate Dyn., 22 , 701–720.
Blanke, B., and S. Raynaud, 1997: Kinematics of the Pacific Equatorial Undercurrent: An Eulerian and Lagrangian approach from GCM results. J. Phys. Oceanogr., 27 , 1038–1053.
Bleck, R., C. Rooth, D. Hu, and L T. Smith, 1992: Salinity-driven thermohaline transients in a wind- and thermohaline-forced isopycnic coordinate model of the North Atlantic. J. Phys. Oceanogr., 22 , 1486–1505.
Cassou, C., and L. Terray, 2001: Oceanic forcing of the wintertime low-frequency atmospheric variability in the North Atlantic European sector: A study with the ARPEGE model. J. Climate, 14 , 4266–4291.
Chen, F., and M. Ghil, 1996: Interdecadal variability in a hybrid coupled ocean–atmosphere model. J. Phys. Oceanogr., 26 , 1561–1578.
Clarke, R A., and J-C. Gascard, 1983: The formation of Labrador Sea Water. Part I: Large-scale processes. J. Phys. Oceanogr., 13 , 1764–1778.
Curry, R G., and M S. McCartney, 2001: Ocean gyre circulation changes associated with the North Atlantic Oscillation. J. Phys. Oceanogr., 31 , 3374–3400.
Daubechies, I., 1992: Ten lectures on wavelets. Society for Industrial and Applied Mathematics CBMS-NSF Lecture Notes 61, 357 pp.
Delworth, T L., and R J. Greatbatch, 2000: Multidecadal thermohaline circulation variability driven by atmospheric surface flux forcing. J. Climate, 13 , 1481–1495.
Delworth, T L., S. Manabe, and R J. Stouffer, 1993: Interdecadal variations of the thermohaline circulation in a coupled ocean–atmosphere model. J. Climate, 6 , 1993–2011.
Delworth, T L., S. Manabe, and R J. Stouffer, 1997: Multidecadal climate variability in the Greenland Sea and surrounding regions: A coupled model simulation. Geophys. Res. Lett., 24 , 257–260.
Déqué, M., C. Dreveton, A. Braun, and D. Cariolle, 1994: The ARPEGE/IFS atmosphere model: A contribution to the French community climate modelling. Climate Dyn., 10 , 249–266.
Dickson, R., J. Lazier, J. Meincke, P. Rhines, and J. Swift, 1996: Long term coordinated changes in the convective activity of the North Atlantic. Progress in Oceanography, Vol. 38, Pergamon, 241–295.
Drange, H., 1999: RegClim ocean modelling at NERSC. Regclim: Regional Climate Development under Global Warming, General Tech. Rep. 2, Norwegian Institute for Air–Sea Research, 93–102.
DYNAMO Group, 1997: DYNAMO—Dynamics of the North Atlantic Models: Simulation and assimilation with high resolution models. Tech. Rep. 294, Institüt für Meereskunde, Christian Albrechts Universität, Kiel, Germany, 334 pp.
Eden, C., and J. Willebrand, 2001: Mechanisms of interannual to decadal variability of the North Atlantic circulation. J. Climate, 14 , 2266–2280.
Enfield, D B., and D A. Mayer, 1997: Tropical Atlantic sea surface temperature variability and its relation to El Niño–Southern Oscillation. J. Geophys. Res., 102 , C1,. 929–945.
Feldstein, S B., 2000: Teleconnections and ENSO: The timescale, power spectra, and climate noise properties. J. Climate, 13 , 4430–4440.
Frankignoul, C., G. De Coëtlogon, T M. Joyce, and S. Dong, 2001: Gulf Stream variability and ocean–atmosphere interactions. J. Phys. Oceanogr., 31 , 3516–3529.
Furevik, T., M. Bentsen, H. Drange, I. K. T. Kindem, N G. Kvamstø, and A. Sorteberg, 2003: Description and evaluation of the Bergen Climate Model: ARPEGE coupled with MICOM. Climate Dyn., 21 , 27–51.
Ganachaud, A., and C. Wunsch, 2000: Improved estimates of global ocean circulation, heat transport and mixing from hydrographic data. Nature, 408 , 453–456.
Gao, Y., H. Drange, and M. Bentsen, 2003: Effects of diapycnal and isopycnal mixing on the ventilation of CFCs in the North Atlantic in an isopycnic coordinate OGCM. Tellus, 55B , 837–854.
Giannini, A., Y. Kushnir, and M C. Cane, 2000: Interannual variability of Caribbean rainfall, ENSO, and the Atlantic Ocean. J. Climate, 13 , 297–310.
Hakkinen, S., 1999: Variability of the simulated meridional heat transport in the North Atlantic for the period 1951–1993. J. Geophys. Res., 104 , C5,. 10991–11007.
Hastenrath, S., 1976: Variations in the low-latitude circulation and extreme climatic events in the tropical Americas. J. Atmos. Sci., 33 , 202–215.
Hsieh, W W., M K. Daney, and R C. Wajsowicz, 1983: The free Kelvin wave in finite-difference numerical models. J. Phys. Oceanogr., 13 , 1383–1397.
Kawase, M., 1987: Establishment of deep ocean circulation driven by deep-water production. J. Phys. Oceanogr., 17 , 2294–2317.
Killworth, P D., 1985: A two-level wind and buoyancy driven thermocline model. J. Phys. Oceanogr., 15 , 1414–1432.
Latif, M., 2001: Tropical Pacific/Atlantic ocean interactions at multi-decadal time scales. Geophys. Res. Lett., 28 , 539–542.
Latif, M., E. Roeckner, U. Mikolajewicz, and R. Voss, 2000: Tropical stabilization of the thermohaline circulation in a greenhouse warming simulation. J. Climate, 13 , 1809–1813.
Lazier, J. R. N., R. Pickart, and P. Rhines, 2001: Deep convection. Ocean Circulation and Climate: Observing and Modelling the Global Ocean, G. Siedler, J. Church, and J. Gould, Eds., International Geophysics Series, Vol. 77, Academic Press, 387–400.
Manabe, S., and R J. Stouffer, 1999: The role of thermohaline circulation in climate. Tellus, 51A , 91–109.
Marshall, J C., and F. Schott, 1999: Open-ocean convection: Observations, theory and models. Rev. Geophys., 37 , 1–64.
Marshall, J., H. Johnson, and J. Goodman, 2001: A study of the interaction of the North Atlantic Oscillation with the ocean circulation. J. Climate, 14 , 1399–1421.
Mignot, J., and C. Frankignoul, 2004: Interannual to interdecadal variability of sea surface salinity in the Atlantic and its link to the atmosphere in a coupled model. J. Geophys. Res., 109 .C04005, doi:10.1029/2003JC002005.
Mikolajewicz, U., and E. Maier-Reimer, 1990: Internal secular variability in an ocean general circulation model. Climate Dyn., 4 , 145–156.
Otterå, O H., H. Drange, M. Bentsen, N G. Kvamstø, and D. Jiang, 2003: The sensitivity of the present-day Atlantic meridional overturning circulation to freshwater forcing. Geophys. Res. Lett., 30 .1898, doi:10.1029/2003GL017578.
Pickart, R S., M A. Spall, M H. Ribergaard, G. W. K. Moore, and R F. Milliff, 2003a: Deep convection in the Irminger Sea forced by the Greenland tip jet. Nature, 424 , 152–156.
Pickart, R S., F. Straneo, and G. W. K. Moore, 2003b: Is the Labrador Sea Water formed in the Irminger basin? Deep-Sea Res., 50A , 23–52.
Schott, F., M. Visbeck, and J. Fischer, 1993: Observations of vertical currents and convection in the central Greenland Sea during the winter of 1988/1989. J. Geophys. Res., 98 , 14401–14421.
Siedler, G., J. Church, and J. Gould, and Eds., 2001: Ocean Circulation and Climate: Observing and Modelling the Global Ocean. International Geophysics Series, Vol. 77, Academic Press, 715 pp.
Talley, L D., J L. Reid, and P E. Robbins, 2003: Data-based meridional overturning streamfunctions for the global oceans. J. Climate, 16 , 3213–3226.
Thorpe, R B., J M. Gregory, T C. Johns, R A. Wood, and J. F. B. Mitchell, 2001: Mechanisms determining the Atlantic thermohaline circulation response to greenhouse gas forcing in a non-adjusted coupled climate model. J. Climate, 14 , 3102–3116.
Timmermann, A., M. Latif, R. Voss, and A. Grötzner, 1998: Northern Hemispheric interdecadal variability: A coupled air–sea mode. J. Climate, 11 , 1906–1931.
Vellinga, M., and P. Wu, 2004: Low-latitude freshwater influence on centennial variability of the Atlantic thermohaline circulation. J. Climate, 17 , 4498–4511.
Vellinga, M., R. Wood, and J. Gregory, 2002: Processes governing the recovery of a perturbed thermohaline circulation in HadCM3. J. Climate, 15 , 764–780.
Weaver, A J., E S. Sarachik, and J. Marotzke, 1991: Freshwater flux forcing of decadal and interdecadal oceanic variability. Nature, 353 , 836–838.
Wunsch, C., 2002: What is the thermohaline circulation? Science, 298 , 1179–1181.
Yang, J., 1999: A linkage between decadal climate variations in the Labrador Sea and the tropical Atlantic Ocean. Geophys. Res. Lett., 26 , 1023–1026.
Raw MOI time series over the 300 yr of the model run. The vertical dashed lines show the limits of the 200 yr that are considered in the present study. The smooth curve indicates the third-order polynomial fit of the time series.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
(top) Mean AMOC in the BCM. (bottom) Spectrum of the (detrended) MOI, in log–log units (solid line) and in log–linear units (variance preserving spectrum: dashed line). The spectrum was calculated by the multi-taper method using four tapers. The vertical line indicates the 90% confidence interval.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Regression of the AMOC on the MOI. AMOC leads at negative lags (in years). The contour interval is 0.2 Sv, continuous for positive and dashed for negative values. The thick contour denotes zero. Light (dark) shaded areas are negative (positive) and significant at the 10% level.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Maximal depth of the mixed layer over the annual cycle.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
(top) Cross correlation between the annual NAO time series and the mixed layer depth averaged over the areas of deep convection during the winter season (Jan–Apr). The NAO leads at positive lags. The thin line refers to the Labrador Sea, the dashed line to the Irminger Sea, and the thick line to the GIN Seas. The gray area limits the 10% significant domain. (bottom) Regression of the annual mixed layer depth on the normalized annual NAO time series. The contour interval is 10 m. The contours and shadings are as in Fig. 3.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Lagged regression of SLP (contours) and wind stress (arrows) anomalies onto the anomalous deep convection in the Irminger Sea. The atmosphere leads at negative lags (in years). Only the vectors with at least one significant component at the 10% level are shown. The contour interval is 0.1 hPa. The contours and shadings are as in Fig. 3.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Cross correlation between the annual MOI and the winter (Jan–Apr) mixed layer depth (thick line), sea surface density (thin line), SSS (dashed line), and SST (mixed line). The gray area delimits the 10% significant domain. The MOC leads at positive lags.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
(top) In-phase regression of the zonal mean Ekman pumping anomalies on the normalized time series of monthly NAO anomalies. Note that the Ekman pumping is not defined near the equator. (bottom) Same for the AMOC monthly anomalies. The contour interval is 0.4 Sv. The contours and shadings are as in Fig. 3, with the continuous line for positive and the dashed line for negative values.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Same as Fig. 8, but for annual means and various lags (in years). NAO leads at positive lags. The contour interval is 0.2 Sv.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Cross correlation between the NAO time series and the MOI, in annual means (solid line) and after low-pass filtering (dashed line). The dark (light) area delimits the 10% significant domain for the annual (decadal) data.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
(top) Same as in bottom of Fig. 2, but for the ENSO time series. (bottom) Normalized ENSO time series of annual mean.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Southern Oscillation associated with a positive phase of the leading tropical Pacific “El Niño” SST mode in (top) BCM and (bottom) NCEP–NCAR. The contour interval is 0.4 hPa. The contour and shadings are as in Fig. 3.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Lagged regression of precipitation on ENSO in Dec–Jan–Feb in (top) BCM and in (bottom) NCEP–NCAR. SON (0) indicates that precipitation is considered from Sep to Nov prior to the mature phase of ENSO and MJJ(+1) indicates that precipitation is considered from May to Jul after the mature phase. The contours correspond to 0, 2.5, 5, 12.5, 22.5, and 50 mm month−1, with continuous (dashed) line for positive (negative) values. The thick contour denotes zero.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Same as Fig. 9, but for the normalized annual ENSO time series when the latter is (left) in phase and leads by (middle) 2 and (right) 6 yr. The contours and shadings are as in Fig. 3.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Cross correlation between the annual MOI and ENSO index. The shaded area indicates the 10% two-sided significant level.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Regression of the annual mixed layer depth to the normalized annual ENSO time series. The mixed layer depth follows the ENSO time series by 1 yr. The contour interval is 10 m (contours and shadings are as in Fig. 3).
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Lagged correlation between the low-passed ENSO and MOI time series. The shaded area indicates the 10% significant level.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Lagged regression of (left) the global decadal SST (contour interval: 0.4 K) and (right) the North Atlantic SSS (contour interval: 0.15 psu) on the normalized low-pass-filtered MOI. The MOI lags by 39 yr in both cases. The contours and shadings are as in Fig. 3.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Leading mode of the MCA between low-pass-filtered SST and AMOC when the former leads the latter by 3 yr. (left) SST homogeneous field (projection of the SST on the MCA time series associated to the SST). The contour interval is 0.08 K. (right) AMOC heterogenous field (projection of the AMOC on the same time series). The contour interval is 0.2 Sv. The contours and shadings are as in Fig. 3. SC is the square covariance and r is the correlation of the two MCA time series. The value in parentheses indicates the estimated significance level.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Same as Fig. 19, but for the second mode of covariability. The SST leads the AMOC by 36 yr.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Lagrangian trajectories of the fictitious drifters launched in the Tropics (black diamonds) and advected by the model mean currents for 39 yr. The colors indicate the depth of the drifters.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
Regression at lag 35 of the potential density at 600 m on the ENSO time series (ENSO leads); all data are low-pass filtered.
Citation: Journal of Climate 18, 13; 10.1175/JCLI3405.1
