Abstract

The Atlantic meridional overturning circulation (AMOC) is an important component of the North Atlantic climate system. Here, simulations from 10 coupled climate models are used to calculate patterns of sea surface temperature (SST) and subsurface density change associated with decadal AMOC variability. The models are evaluated using observational constraints and it is shown that all 10 models suffer from North Atlantic Deep Water transports that are too shallow, although the biases are least severe in the Community Climate System Model, version 4 (CCSM4). In the models that best compare with observations, positive AMOC anomalies are associated with reduced Labrador Sea stratification and increased midocean (800–1800 m) densities in the subpolar gyre. Maximum correlations occur when AMOC anomalies lag Labrador Sea stratification and subsurface density anomalies by 2–6 yr and 0–3 yr, respectively. In all 10 models, North Atlantic warming follows positive AMOC anomalies, but the patterns and magnitudes of SST change are variable. A simple detection and attribution analysis is then used to evaluate the utility of Atlantic midocean density and Labrador Sea stratification indices for detecting changes to the AMOC in the presence of increasing CO2 concentrations. It is shown that trends in midocean density are identifiable (although not attributable) significantly earlier than trends in the AMOC. For this reason, subsurface density observations could be a useful complement to transport observations made at specific latitudes and may help with the more rapid diagnosis of basin-scale changes in the AMOC. Using existing observations, it is not yet possible to detect a robust trend in the AMOC using either midocean densities or transport observations from 26.5°N.

1. Introduction

The Atlantic meridional overturning circulation (AMOC) is associated with a climatically significant northward heat transport of ~1.2 × 1015 W (Johns et al. 2011) that is expected to weaken in response to global warming (Gregory et al. 2005; Weaver et al. 2012). The potential consequences of such a weakening have been a key motivator for recent observations designed to monitor different aspects of this circulation (Bryden et al. 2005; Cunningham et al. 2007; Kanzow et al. 2006; Willis 2010). However, the substantial internal variability of the AMOC means it will likely take several decades to identify an anthropogenic trend using transport observations alone (Baehr et al. 2008; Roberts and Palmer 2012). Previous studies have shown that AMOC detection times could be reduced by using well-observed hydrographic properties that have a characteristic “fingerprint” associated with changes in the AMOC but higher signal-to-noise ratios (Vellinga and Wood 2004; Roberts and Palmer 2012). Such fingerprints also have the potential to be used as proxies to extend records of the AMOC back in time.

In the absence of multidecadal AMOC observations, coupled climate models have been the tool of choice for identifying AMOC fingerprints in a variety of surface and subsurface ocean properties (Baehr et al. 2007; Bingham and Hughes 2009; Knight et al. 2005; Latif et al. 2004; Mahajan et al. 2011; Roberts and Palmer 2012; Vellinga and Wood 2004; Zhang 2007, 2008). One pattern of climate variability that is often linked to the AMOC is a mode of multidecadal North Atlantic sea surface temperature (SST) variability that has been observed in both instrumental and proxy records (e.g., Kushnir 1994; Schlesinger and Ramankutty 1994; Delworth and Mann 2000). This phenomenon is often termed the Atlantic multidecadal oscillation (AMO) [or Atlantic multidecadal variability (AMV)] and has been linked with impacts on Atlantic hurricane activity, Asian monsoons, and rainfall in North Africa and Europe (Folland et al. 1986; Sutton and Hodson 2005; Sutton and Dong 2012; Zhang and Delworth 2006). In addition, several modeling studies have used preindustrial control simulations to demonstrate that there is a strong association between the AMOC and a pattern of North Atlantic SST variability that resembles the AMO (Knight et al. 2005; Latif et al. 2004).

We expect a relationship between the AMOC and SSTs to exist because the overturning circulation is the dominant contributor to ocean heat transports in the subtropical North Atlantic (Johns et al. 2011). However, the partitioning of ocean heat transports between overturning and horizontal gyre components may change in the future if there is a significant weakening of the AMOC. This could cause the relationship between the AMOC and SSTs to change under global warming conditions. In addition, the attribution of recent Atlantic SST variability to changes in the AMOC requires assumptions to be made about the SST response to other climate forcings (Mann and Emanuel 2006; Trenberth and Shea 2006). For example, some recent studies have suggested that variations in anthropogenic and volcanic aerosols could have played a role in driving recent Atlantic SST variability (Booth et al. 2012; Otterå et al. 2010). Given these caveats, it is not clear that fingerprints derived from SSTs are always a reliable indicator of historical and/or future AMOC change.

In contrast, subsurface hydrographic properties are less likely to be sensitive to surface forcings and are more likely to have high signal-to-noise ratios for the detection of trends (Vellinga and Wood 2004). To minimize the impact of external forcings, several studies have used models to identify fingerprints of AMOC variability using subsurface ocean properties (e.g., Mahajan et al. 2011; Roberts and Palmer 2012; Vellinga and Wood 2004; Zhang 2007, 2008). For example, Baehr et al. (2007) used the ECHAM5 model to define a fingerprint of subsurface zonal density gradients at 26.5°N and concluded that there was no trend in the AMOC during the last 50 yr. However, these studies have mostly relied on results from a single climate model (or related models from a single institute). In this study, we present a multimodel comparison of AMOC fingerprints derived from SSTs and subsurface densities using data from 10 climate models that include dynamic representations of the ocean and atmosphere.

Horizontal gradients in density (and thus pressure) are dynamically linked to ocean transports, and previous studies using general circulation models have found linear relationships between the AMOC and meridional density/pressure gradients (e.g., Thorpe et al. 2001; Griesel and Maqueda 2006). In addition, arguments based on geostrophy and simplifying scaling assumptions have suggested that the AMOC should be positively associated with changes to meridional density gradients, although the nature of the relationship (linear or nonlinear) depends on whether or not a vertical length scale is assumed to be constant or derived assuming advective–diffusive buoyancy balance in the thermocline (e.g., Stommel 1961; Park 1999). Other studies have shown that processes unrelated to buoyancy fluxes, such as wind forcing in the Southern Ocean, can exert a strong control on the AMOC (e.g., Toggweiler and Samuels 1995, 1998). For example, using a series of ocean general circulation model experiments run to steady state (~3000 yr), de Boer et al. (2010) demonstrated that it was possible to get a negative correlation between the AMOC and meridional density gradients in the Atlantic when wind stress in the Southern Ocean was perturbed.

In a review of the physical mechanisms that provide the energy to sustain large-scale overturning circulations, Kuhlbrodt et al. (2007) concluded that both wind-driven upwelling in the Southern Ocean and turbulent vertical mixing in the ocean interior are responsible for driving the AMOC on long time scales (millennia). However, the authors emphasized that the driving processes identified using energetic considerations do not fully determine the spatial extent or maximum strength of the overturning in the Atlantic. In particular, processes that modify the density of the North Atlantic can have a substantial impact on the AMOC by influencing the formation of North Atlantic Deep Water (NADW). In this paradigm, the driving forces of wind-driven upwelling and vertical mixing set the rate of deep-water upwelling on steady-state time scales, but buoyancy fluxes exert a strong control on transient anomalies in the AMOC. For the reasons outlined above, we consider subsurface densities to be a suitable candidate for the identification of robust fingerprints of AMOC change on multidecadal time scales, although we note that factors such as wind stress in the Southern Ocean may become important on much longer time scales.

We focus our analysis on large-scale patterns of subsurface densities in two depth ranges: 0–800 m and 800–1800 m. The rationale for choosing the range 800–1800 m (rather than an averaged or integrated quantity all the way from the surface) is to isolate the density changes associated with the AMOC from the higher-frequency variability driven by changes to surface fluxes and wind-driven gyre dynamics. The specific range of 800–1800 m is chosen such that there are at least two discrete vertical levels selected in all models and so that the lower boundary is not significantly deeper than the 2000-m maximum depth observable by the Argo network of autonomous profiling floats (Roemmich et al. 2009; http://www.argo.ucsd.edu/).

This paper is organized as follows: The models and data used in this study are described in section 2. Simulated profiles of the AMOC at 26.5°N and hydrographic properties from the North Atlantic subpolar gyre are evaluated in section 3. Spatial patterns of SST and subsurface density change associated with AMOC variability in preindustrial control simulations (and associated lead–lag relationships) are presented in section 4. The response of the identified AMOC fingerprints to changing CO2 concentrations, as well as their utility for detecting changes in the AMOC, is presented in section 5. Finally, a reconstructed index of midocean densities calculated from temperature and salinity observations is presented in section 6.

2. Model experiments and data

We use data from 10 different coupled climate models (Table 1), and data from 9 of these models (CCSM4, CNRM-CM5, HadGEM2-ES, MPI-ESM-LR, MPI-ESM-MR, MPI-ESM-P, MRI-CGCM3, NorESM1-M, and NorESM1-ME) are freely available online as part of phase 5 of the Coupled Model Intercomparison Project (CMIP5) archive (http://cmip-pcmdi.llnl.gov/cmip5/). We limit our selection of CMIP5 models to those that have Atlantic overturning streamfunction data and ocean grid information available to download (as of mid-2012). In addition, we also use data from HadCM3, a coupled climate model that was used in support of CMIP3 and that has previously been used to document a relationship between the AMOC and North Atlantic SSTs (Knight et al. 2005). AMOC index time series from each model are defined as the value of the Atlantic overturning streamfunction at the latitude nearest 30°N and depth nearest 1100 m. This index is chosen such that it is near the latitude of maximum Atlantic Ocean heat transport (Ganachaud and Wunsch 2003) and also close to the depth of the overturning transport maximum at 26.5°N (Kanzow et al. 2010). All AMOC transports are quoted in units of Sverdrups (1 Sv ≡ 106 m3 s−1).

Table 1.

Details of the models used in this study.

Details of the models used in this study.
Details of the models used in this study.

3. Evaluation of the AMOC in preindustrial control simulations

To evaluate the simulation of the AMOC in each model, we compare against observed vertical profiles of the AMOC at 26.5°N and climatologies of temperature, salinity, and density in the North Atlantic subpolar gyre. Figure 1 shows mean AMOC profiles from preindustrial control simulations compared with the mean overturning profile measured by the Rapid Climate Change–Meridional Overturning Circulation and Heatflux Array (RAPID–MOCHA) array at 26.5°N between April 2004 and March 2011 (Cunningham et al. 2007; McCarthy et al. 2012). In the upper ocean, the strength and depth of the Atlantic overturning maximum is in close agreement with observations in CCSM4, MPI-ESM-LR, MPI-ESM-P, and MPI-ESM-MR. In contrast, the overturning maximum is much too strong in NorESM1-M and NorESM1-ME, too weak in HadGEM2-ES and MRI-CGCM3, and too shallow in HadCM3 and CNRM-CM5. Observations indicate that the southward return flow of the AMOC at 26.5°N is composed of ~60% upper NADW (depth range of 1100–3000 m) and ~40% lower NADW (depth range of 3000–5000 m) (McCarthy et al. 2012). In comparison, all the models used in this study overestimate the fraction of southward AMOC return flow occurring in the 1100–3000-m-depth range and most have an overturning cell that is too shallow (Table 2). The model with the least severe biases in the vertical structure of the AMOC is CCSM4, which has an overturning cell that extends to a depth of 4750 m and is also in close agreement with the observed strength of the AMOC.

Fig. 1.

AMOC streamfunction profiles at 26.5°N from preindustrial control simulations (black and gray lines) compared with the mean overturning profile from the RAPID array (stars). Overturning maxima are indicated by diamonds and values are given in the legend.

Fig. 1.

AMOC streamfunction profiles at 26.5°N from preindustrial control simulations (black and gray lines) compared with the mean overturning profile from the RAPID array (stars). Overturning maxima are indicated by diamonds and values are given in the legend.

Table 2.

NADW layer transports at 26.5°N from model control simulations and observations. Layer transports are specified in Sverdrups with negative values indicating net southward flow in the specified depth range. The vertical extent of the AMOC overturning cell ZAMOC is given in meters. Observed layer transports are from McCarthy et al. (2012, their Table 1).

NADW layer transports at 26.5°N from model control simulations and observations. Layer transports are specified in Sverdrups with negative values indicating net southward flow in the specified depth range. The vertical extent of the AMOC overturning cell ZAMOC is given in meters. Observed layer transports are from McCarthy et al. (2012, their Table 1).
NADW layer transports at 26.5°N from model control simulations and observations. Layer transports are specified in Sverdrups with negative values indicating net southward flow in the specified depth range. The vertical extent of the AMOC overturning cell ZAMOC is given in meters. Observed layer transports are from McCarthy et al. (2012, their Table 1).

Inaccurate simulation of NADW transports in coupled climate models has been linked to poor representation of Nordic Sea overflow processes (Legg et al. 2009). Improvements to many aspects of the simulated Atlantic Ocean in CCSM4, including the depth structure of the AMOC, have been attributed to the inclusion of parameterized Nordic Sea overflow physics (Danabasoglu et al. 2010; Yeager and Danabasoglu 2012). Table 3 compares observational estimates of Nordic Sea overflow transports across the Greenland–Scotland ridge with values from CCSM4, HadCM3, HadGEM2-ES, MPI-ESM-LR, and MPI-ESM-MR. The simulated transports in HadCM3 and HadGEM2 are excessive in the Denmark Strait and too weak in the Faroe Bank Channel, whereas transports in CCSM4, MPI-ESM-LR, and MPI-ESM-MR compare well with observations. However, the rate of production (and subsequent transport) of dense source waters at the Greenland–Scotland ridge is only one of several processes that are important for accurate representation of overflows. The final depth, density, and volume transport of overflow product waters are strongly influenced by the entrainment of ambient water, which is governed by subgrid-scale processes in coupled climate models (e.g., shear instability and bottom friction; see overview in Legg et al. 2009). Despite the availability of dense source waters in HadCM3, HadGEM2-ES, MPI-ESM-LR, and MPI-ESM-MR, all four models suffer from shallow overturning cells. This bias is likely related to spurious mixing that is known to be an issue for prognostic flow over steep topography in level coordinate models (Winton et al. 1998). In contrast, the parameterized entrainment fluxes in CCSM4 are too low, leading to product water that is too dense and product water transports and variability that are too low (Yeager and Danabasoglu 2012).

Table 3.

Overflow transports in the Denmark Strait (DS), the Faroe Bank Channel (FBC) and across the Iceland–Faroe ridge (IF) from preindustrial control simulations and observations. Transports are quoted in Sverdrups, with positive values indicating transport into the Atlantic basin.

Overflow transports in the Denmark Strait (DS), the Faroe Bank Channel (FBC) and across the Iceland–Faroe ridge (IF) from preindustrial control simulations and observations. Transports are quoted in Sverdrups, with positive values indicating transport into the Atlantic basin.
Overflow transports in the Denmark Strait (DS), the Faroe Bank Channel (FBC) and across the Iceland–Faroe ridge (IF) from preindustrial control simulations and observations. Transports are quoted in Sverdrups, with positive values indicating transport into the Atlantic basin.

Figure 2 shows profiles of North Atlantic subpolar gyre temperature, salinity and density from preindustrial control simulations compared with climatological values (1971–2000) calculated from the Ensemble-Based Predictions of Climate Changes and their Impacts project (ENSEMBLES) 1° × 1° gridded analysis of quality-controlled subsurface temperature and salinity observations (EN3 v2a; Ingleby and Huddleston 2007; http://www.metoffice.gov.uk/hadobs/en3/). In general, the models with stronger overturning (e.g., NorESM1-ME, NorESM1-M) have weaker stratification in the subpolar gyre and the models with weaker overturning (e.g., HadGEM2, MRI-CGCM3) have stronger stratification (Fig. 2). Similarly, CCSM4, MPI-ESM-MR, MPI-ESM-LR, and MPI-ESM-P represent the vertical structure of temperature, salinity, and density in the subpolar gyre relatively well and also do a good job at simulating the depth and magnitude of the overturning maximum at 26.5°N. The intermodel differences in subpolar gyre stratification are mostly due to salinity-driven density biases in the upper 200 m, with some models also exhibiting compensating temperature biases.

Fig. 2.

Mean profiles of subpolar gyre (45°–65°N, 70°W–0°) density, temperature, and salinity from preindustrial control simulations (black and gray lines) compared with climatological profiles calculated from the EN3 v2a objective analysis of temperature and salinity observations (stars).

Fig. 2.

Mean profiles of subpolar gyre (45°–65°N, 70°W–0°) density, temperature, and salinity from preindustrial control simulations (black and gray lines) compared with climatological profiles calculated from the EN3 v2a objective analysis of temperature and salinity observations (stars).

Although biases in the spatial structure of temperature and salinity at depth are not evident in the area-averaged profiles presented in Fig. 2, maps of subpolar gyre temperature and salinity at depth (not shown) reveal that the differences between models are dominated by the spatial mean. However, it is worth noting that some models have horizontal property gradients that are in direct conflict with those in the EN3 climatology. For example, HadCM3, HadGEM2-ES, MPI-ESM-LR, MPI-ESM-MR, and MPI-ESM-P are warmer and saltier south of the Denmark Strait than elsewhere in the subpolar gyre when they should be relatively cold and fresh because of the influence of Denmark Strait overflow water. In contrast, deep subpolar gyre temperature and salinity gradients in CCSM4, CNRM-CM5, and MRI-CGCM3 are in much better agreement with observations.

4. Fingerprints of the AMOC in preindustrial control simulations

a. Sea surface temperature fingerprints

Pre-industrial control simulations are used to calculate spatial fingerprints of the AMOC by regressing decadal mean AMOC time series against decadal means of SST and vertically averaged potential density at each Atlantic grid point. Figure 3 shows regression maps of changes in SST associated with an increase in the AMOC. There is a consensus among models that increases in the AMOC are associated with warming in the North Atlantic, although the magnitudes and locations of maximum warming vary. Some models (e.g., MRI-CGCM3) have maximum warming focused in the Greenland–Iceland–Norwegian (GIN) Seas and the eastern subpolar gyre, whereas other models (e.g., HadGEM2-ES) show more distributed warming over the entire subpolar gyre. These differences could be a consequence of intermodel variations in the locations of convection and deep-water formation, differences in the importance of surface heat flux feedbacks, or dynamic changes associated with the wind-driven gyre circulation. There is no consensus among models regarding the sign and/or magnitude of SST change in the Southern Ocean.

Fig. 3.

(a)–(j) Maps of slope coefficients for linear regressions between decadal means of SST and the AMOC in preindustrial control simulations (contours; °C Sv−1). The number of decades used for each model is specified in the titles of each subplot. Regions without stippling correspond with regression coefficients that have a p value of <0.1 for a two-tailed t test (i.e., these regions are inconsistent with the null hypothesis of zero correlation at a 90% level of significance). The regions used to define a dipole index of Atlantic SST variability are indicated by the boxes in (b).

Fig. 3.

(a)–(j) Maps of slope coefficients for linear regressions between decadal means of SST and the AMOC in preindustrial control simulations (contours; °C Sv−1). The number of decades used for each model is specified in the titles of each subplot. Regions without stippling correspond with regression coefficients that have a p value of <0.1 for a two-tailed t test (i.e., these regions are inconsistent with the null hypothesis of zero correlation at a 90% level of significance). The regions used to define a dipole index of Atlantic SST variability are indicated by the boxes in (b).

To examine the strength of the association between SSTs and the AMOC in control simulations, a simple “north minus south” Atlantic SST dipole index is defined as the area average of SSTs in the subpolar North Atlantic (45°–80°N, 70°W–30°E) minus the area average of SSTs in the South Atlantic (45°S–0°, 70°W–30°E). We define a dipole index, rather than an area average of North Atlantic SSTs, so that it is insensitive to spatially uniform changes in global temperature. Figure 4 shows decadal means of the SST dipole index plotted against decadal means of the AMOC in preindustrial control simulations (crosses) and the associated linear regression statistics are presented in Table 4. Note that in these plots SSTs are presented as the independent x variable, as this is the assumption one would have to make if attempting to infer a change in the AMOC from SST observations. This regression analysis indicates that the slopes describing linear relationships between the AMOC and SSTs vary between model control simulations, and in two models (NorESM1-M and NorESM1-ME) regression slopes are not significantly different from zero (Table 4). However, this analysis does not consider the potential for lagged relationships between the AMOC and SSTs.

Fig. 4.

Atlantic SST dipole indices plotted against the AMOC at 30°N in control simulations (crosses) and selected CO2-forced experiments (diamonds). All data points are plotted using decadal means. Least squares linear regression lines for control simulations are plotted in gray. The corresponding regression coefficients, standard errors, and p values are documented in Table 4.

Fig. 4.

Atlantic SST dipole indices plotted against the AMOC at 30°N in control simulations (crosses) and selected CO2-forced experiments (diamonds). All data points are plotted using decadal means. Least squares linear regression lines for control simulations are plotted in gray. The corresponding regression coefficients, standard errors, and p values are documented in Table 4.

Table 4.

Linear regression (y = ax + b) statistics for the SST dipole index and AMOC data plotted in Fig. 4. Regression coefficients a and b are quoted as the estimated value ± 1.96 × the standard error (SE), where a is in units of Sv °C−1 and b is in units of Sverdrups. The number of decades used in the regression is specified by n and the p values are for a two-tailed t test of zero correlation.

Linear regression (y = ax + b) statistics for the SST dipole index and AMOC data plotted in Fig. 4. Regression coefficients a and b are quoted as the estimated value ± 1.96 × the standard error (SE), where a is in units of Sv °C−1 and b is in units of Sverdrups. The number of decades used in the regression is specified by n and the p values are for a two-tailed t test of zero correlation.
Linear regression (y = ax + b) statistics for the SST dipole index and AMOC data plotted in Fig. 4. Regression coefficients a and b are quoted as the estimated value ± 1.96 × the standard error (SE), where a is in units of Sv °C−1 and b is in units of Sverdrups. The number of decades used in the regression is specified by n and the p values are for a two-tailed t test of zero correlation.

To explore the importance of time lags, we calculate lead–lag correlations between the AMOC and the SST dipole index defined above (Fig. 5). In all 10 models, maximum correlations between the AMOC and SST dipoles occur when SSTs lag the AMOC by 1–5 yr. Our analysis indicates that surface warming in the North Atlantic subpolar gyre following an increase of the AMOC is a robust feature of AMOC variability in a variety of coupled climate models. These results are consistent with the idea that predictability of the AMOC could lead to predictability of North Atlantic surface temperatures on decadal time scales (e.g., Collins et al. 2006). However, our results also suggest that the time scales and spatial patterns of SST predictability will be model dependent.

Fig. 5.

Lead–lag correlations between the AMOC at 30°N and the Atlantic SST dipole (dashed black), midocean density dipole (solid black), and Labrador Sea stratification (dotted black) indices defined in section 4. The autocorrelation function of the AMOC is plotted as thick dashed gray line and the AMOC index leads at positive lags. Correlations are calculated using detrended annual mean data that has also been smoothed using a 10-yr low-pass Butterworth filter. To estimate 95% significance levels (dashed horizontal lines), threshold correlation values were calculated using a two-tailed t test and an estimate of the effective number of independent data points Neff = N/τ, where τ is an estimate of a decorrelation time scale defined as the positive lag at which the AMOC autocorrelation function first crosses zero and N is the length of the time series.

Fig. 5.

Lead–lag correlations between the AMOC at 30°N and the Atlantic SST dipole (dashed black), midocean density dipole (solid black), and Labrador Sea stratification (dotted black) indices defined in section 4. The autocorrelation function of the AMOC is plotted as thick dashed gray line and the AMOC index leads at positive lags. Correlations are calculated using detrended annual mean data that has also been smoothed using a 10-yr low-pass Butterworth filter. To estimate 95% significance levels (dashed horizontal lines), threshold correlation values were calculated using a two-tailed t test and an estimate of the effective number of independent data points Neff = N/τ, where τ is an estimate of a decorrelation time scale defined as the positive lag at which the AMOC autocorrelation function first crosses zero and N is the length of the time series.

b. Upper-ocean density (0–800 m) fingerprints

Figure 6 shows maps of decadal mean AMOC regressed against vertically averaged potential densities in the upper ocean (defined as the depth range of 0–800 m). In most locations the sign and/or magnitude of upper-ocean density change is variable across models. These differences are likely due to the influence of surface fluxes of heat and freshwater, wind-driven changes to isopycnal slopes, and shifts in the location of ocean fronts that could have a model-dependent association with changes in the AMOC. However, in all 10 models, increases in the AMOC are associated with positive density anomalies in (or near to) the Labrador Sea.

Fig. 6.

(a)–(j) Maps of slope coefficients for linear regressions between decadal means of vertically averaged upper-ocean (0–800 m) density and the AMOC in preindustrial control simulations (contours; kg m−3 Sv−1). Stippling is the same as described in Fig. 3. The region used to define an index of Labrador Sea stratification is indicated by the box in (b).

Fig. 6.

(a)–(j) Maps of slope coefficients for linear regressions between decadal means of vertically averaged upper-ocean (0–800 m) density and the AMOC in preindustrial control simulations (contours; kg m−3 Sv−1). Stippling is the same as described in Fig. 3. The region used to define an index of Labrador Sea stratification is indicated by the box in (b).

Previous modeling studies have documented the importance of Labrador Sea convection as a driver of AMOC variability (Yang 1999; Eden and Willebrand 2001; Getzlaff et al. 2005). In addition, observed variations in Labrador Seawater thickness have been linked to variability in water mass properties in the deep North Atlantic (Curry et al. 1998) and transport variability in the North Brazil Current (Zhang et al. 2011). To investigate the importance of convection in the Labrador Sea, we define a Labrador Sea stratification index as the volume average of potential densities in the depth range of 0–800 m minus the volume average of potential densities in the depth range of 800–1800 m for the region 45°–80°N, 70°–45°W. When this index is high (i.e., less negative), stratification is weak and conditions are more favorable for convection. The relationship between the Labrador Sea index and the AMOC at different time lags is shown in Fig. 5. In all 10 models, maximum correlations between the AMOC and the Labrador Sea index occur when the Labrador Sea index leads the AMOC by 2–7 yr, with correlations reduced or insignificant at zero lag. These results point toward a mode of AMOC variability that is robust across a range of models, where positive density anomalies in the Labrador Sea cause reduced stratification and convection, ultimately leading to an increase of the AMOC. However, we note that previous work has shown that lags between AMOC anomalies in the North Atlantic subpolar gyre and farther south are resolution dependent (Getzlaff et al. 2005), with eddy-permitting models having faster signal propagation speeds than coarser-resolution models (e.g., the coupled climate models considered in this study).

c. Midocean density (800–1800 m) fingerprints

Figure 7 shows maps of decadal mean AMOC regressed against vertically averaged potential densities in the depth range of 800–1800 m. Nine models show a remarkably consistent pattern of midocean density change in the North Atlantic associated with decadal changes in the AMOC. The large-scale nature of this pattern also makes it particularly appealing as a fingerprint of AMOC variability that could be identifiable from observations. Between 25° and 65°N, each model shows a coherent pattern of increased midocean density in the subpolar gyre and along the western boundary of the North Atlantic. This fingerprint of density change is consistent with previous work using an ocean general circulation model that described a similar pattern of vertically integrated pressure change from 1000- to 2500-m depth in the North Atlantic associated with increased transport into the deep western boundary current (Griesel and Maqueda 2006). The only model to show a very different pattern of midocean density change in the North Atlantic is HadCM3.

Fig. 7.

(a)–(j) Maps of slope coefficients for linear regressions between decadal means of vertically averaged midocean (800–1800 m) density and the AMOC (contours; kg m−3 Sv−1). Stippling is the same as described in Fig. 3. The regions used to define a dipole index of Atlantic midocean density variability are indicated in (b).

Fig. 7.

(a)–(j) Maps of slope coefficients for linear regressions between decadal means of vertically averaged midocean (800–1800 m) density and the AMOC (contours; kg m−3 Sv−1). Stippling is the same as described in Fig. 3. The regions used to define a dipole index of Atlantic midocean density variability are indicated in (b).

Figure 8 shows decadal means of a midocean density dipole index plotted against decadal means of the AMOC in preindustrial control simulations (crosses). The midocean density dipole index is defined as the volume average of potential densities in the North Atlantic subpolar gyre (45°–65°N, 70°W–0°) minus the volume average of potential densities in the South Atlantic (45°S–0°, 70°W–30°E) in the depth range of 800–1800 m. In all 10 models, there is a statistically significant linear relationship between the AMOC and this index (Table 5). The fraction of AMOC variance explained by the index varies across models, and the relationship in HadCM3 is particularly weak (although still statistically significant). Inspection of the regression residuals (not shown) gives no indication that the assumption of linearity is invalid.

Fig. 8.

Atlantic midocean density dipole index plotted against the AMOC at 30°N in control simulations (crosses) and selected CO2-forced experiments (diamonds). All data points are plotted using decadal means. Least squares linear regression lines for control simulations are plotted in gray. The corresponding regression coefficients, standard errors, and p values are documented in Table 5.

Fig. 8.

Atlantic midocean density dipole index plotted against the AMOC at 30°N in control simulations (crosses) and selected CO2-forced experiments (diamonds). All data points are plotted using decadal means. Least squares linear regression lines for control simulations are plotted in gray. The corresponding regression coefficients, standard errors, and p values are documented in Table 5.

Table 5.

Linear regression (y = ax + b) statistics for the midocean density dipole index and AMOC data plotted in Fig. 8. Regression coefficients a and b are quoted as the estimated value ± 1.96 × the SE, where a is in units of Sv−1 kg−1 m3 and b is in units of Sverdrups. The number of decades used in the regression is specified by n and the p values are for a two-tailed t test of zero correlation.

Linear regression (y = ax + b) statistics for the midocean density dipole index and AMOC data plotted in Fig. 8. Regression coefficients a and b are quoted as the estimated value ± 1.96 × the SE, where a is in units of Sv−1 kg−1 m3 and b is in units of Sverdrups. The number of decades used in the regression is specified by n and the p values are for a two-tailed t test of zero correlation.
Linear regression (y = ax + b) statistics for the midocean density dipole index and AMOC data plotted in Fig. 8. Regression coefficients a and b are quoted as the estimated value ± 1.96 × the SE, where a is in units of Sv−1 kg−1 m3 and b is in units of Sverdrups. The number of decades used in the regression is specified by n and the p values are for a two-tailed t test of zero correlation.

The relationship between midocean densities and the AMOC at different time lags is shown in Fig. 5. In all but two of the models analyzed, maximum correlations between the midocean density index and the AMOC occur when the midocean density index leads the AMOC by 0–3 yr. The two models that do not exhibit this same pattern are HadCM3 and CNRM-CM5, which both show maximum correlations when the midocean density dipole index lags the AMOC by ~20 yr.

To investigate the differences noted above, we calculate correlations between decadal means of the AMOC and area-averaged density at different depths in the North Atlantic subpolar gyre (Fig. 9). Note that, for comparison across models with different length control simulations, all values in this plot have been divided by a reference correlation such that values with a magnitude greater than unity are significant at the 95% level. All 10 models indicate that maximum correlations between the AMOC and subpolar gyre densities occur in the subsurface ocean. However, the model that best simulates the vertical structure of the AMOC at 26.5°N (CCSM4) has a maximum correlation at a depth of ~1400 m. In contrast, maximum correlations in HadCM3 and CNRM-CM5 occur at depths of 700 and 400 m, respectively. This suggests that a depth range of 800–1800 m is not the optimal choice for identifying density changes associated with the AMOC in these two models. When lead–lag correlations are recalculated for HadCM3 and CNRM-CM5 using the 0–800-m depth range (not shown), both models have maximum correlations when the upper-ocean density dipole index leads the AMOC. For this reason, we interpret the differences in lead–lag behavior in HadCM3 and CNRM-CM5, as well as the different pattern of midocean density change associated with the AMOC in HadCM3, to reflect a bias toward shallower depths of overturning in these models (see Fig. 1).

Fig. 9.

Correlations between decadal means of the AMOC at 30°N and density for different depths in the subpolar gyre (45°–65°N, 70°W–0°) in preindustrial control simulations. All correlations are standardized such that values with a magnitude greater than unity are significant at the 95% level for a two-tailed t test. The depths of maximum positive correlations are indicated by diamonds.

Fig. 9.

Correlations between decadal means of the AMOC at 30°N and density for different depths in the subpolar gyre (45°–65°N, 70°W–0°) in preindustrial control simulations. All correlations are standardized such that values with a magnitude greater than unity are significant at the 95% level for a two-tailed t test. The depths of maximum positive correlations are indicated by diamonds.

Given the general bias toward shallow overturning in the models investigated in this study (Fig. 1), it is possible that the model fingerprints identified in Fig. 8 are spuriously dominated by changes in upper NADW transports driven by Labrador Sea convection. If so, other patterns of density change, perhaps at greater depths, could be more important for identifying AMOC change associated with Nordic Sea overflow variability. However, Yeager and Danabasoglu (2012) found only a modest relationship between Nordic Sea overflow variability and the AMOC in the deep ocean (>2 km) in CCSM4. The dominant way that increased overflow transports impacted the AMOC in CCSM4 was to increase stratification in the Labrador Sea and suppress convection. If similar mechanisms to those documented in CCSM4 are dominant in the real world, then density fingerprints associated with changes in Labrador Sea stratification should also respond to changes in overflow transports.

In summary, our lead–lag correlation analysis indicates a consistent mode of variability composed of the following events: 1) positive density anomalies and reduced stratification in the Labrador Sea, 2) increased subpolar gyre densities in the subsurface waters that feed the deep western boundary current, 3) an increase of the AMOC and its associated heat transport, and 4) increased SSTs in the subpolar gyre.

5. AMOC fingerprints in forced experiments

a. Idealized CO2-forcing experiments

To explore the impact of a sustained change in the AMOC on spatial patterns of SSTs and subsurface densities, as well as evaluate their utility for detecting changes in the AMOC, we examine idealized CO2-forced experiments from CCSM4, MPI-ESM-MR, and HadGEM2-ES. We use standard CMIP5 experiments in which simulations are initialized from a preindustrial control state and CO2 is increased by 1% yr−1 to 4 times preindustrial levels. In HadGEM2-ES, this experiment is extended using data from an additional simulation in which CO2 is subsequently reduced by 1% yr−1, followed by a period of stabilization at preindustrial levels. Results from this extended HadGEM2-ES experiment are described more fully in Boucher et al. (2012).

To track the temporal evolution of Atlantic SSTs, Labrador Sea stratification, and midocean densities, we use the indices defined in sections 4a4c. Figure 10 shows the SST dipole, Labrador Sea stratification, and midocean density dipole indices calculated from annual means compared with the AMOC in the idealized CO2-forced experiments. Each experiment shows a weakening of the AMOC associated with increasing CO2 concentrations and, in the extended HadGEM2-ES experiment, a recovery of the AMOC to preindustrial levels as CO2 is reduced.

Fig. 10.

Standardized dipole indices compared with the AMOC in idealized CO2-forced experiments. AMOC (thick black lines), SST (thin black lines), Labrador Sea stratification (thick dashed gray lines), and midocean (800–1800 m) density dipole indices (thick gray lines) are standardized by removing the mean and dividing by the standard deviation. (a) HadGEM2-ES is forced with a 1% yr−1 increase of CO2 up to 4 times preindustrial levels, followed by a 1% yr−1 reduction of CO2 and stabilization at preindustrial levels. (b) CCSM4 and (c) MPI-ESM-MR are forced with a 1% yr−1 increase of CO2 up to 4 times preindustrial levels. The vertical dashed gray lines enclose the range of start dates used to estimate the time taken to detect significant trends.

Fig. 10.

Standardized dipole indices compared with the AMOC in idealized CO2-forced experiments. AMOC (thick black lines), SST (thin black lines), Labrador Sea stratification (thick dashed gray lines), and midocean (800–1800 m) density dipole indices (thick gray lines) are standardized by removing the mean and dividing by the standard deviation. (a) HadGEM2-ES is forced with a 1% yr−1 increase of CO2 up to 4 times preindustrial levels, followed by a 1% yr−1 reduction of CO2 and stabilization at preindustrial levels. (b) CCSM4 and (c) MPI-ESM-MR are forced with a 1% yr−1 increase of CO2 up to 4 times preindustrial levels. The vertical dashed gray lines enclose the range of start dates used to estimate the time taken to detect significant trends.

In all three models, the Labrador Sea and midocean density dipole indices capture the low-frequency changes of the AMOC. Regression analysis of the CO2-forced experiments confirms that there is a robust linear relationship between the AMOC and the midocean density dipole index in all three models (diamonds in Figs. 8b,c,f and Table 5).

In contrast, the SST dipole indices in the CO2-forced experiments show different behaviors in each model (diamonds in Figs. 4b,c,f, and Table 4) and have more noise at high frequencies, which makes them less useful as an index for the detection of trends. In MPI-ESM-MR, the SST dipole index behaves as would be predicted from analysis of the preindustrial control simulation. In CCSM4, there is a strong positive association between the SST dipole index and the AMOC, but the relationship is nonlinear. In HadGEM2-ES, there is no significant relationship between the AMOC and SST dipole index. The absence of a strong association between SSTs and the AMOC in the HadGEM2-ES experiment indicates that processes other than meridional heat transport by the overturning circulation are dominating the evolution of meridional SST gradients in the Atlantic.

Further analysis of this apparent inconsistency reveals that, despite a strong correlation between the AMOC and its associated ocean heat transport (OHT), the evolution of the total OHT at 30°N is dominated by the heat transport of the horizontal gyre circulation (Fig. 11). Observations indicate that the meridional OHT by the Atlantic gyre circulation at 26.5°N accounts for only 0.15 PW out of a total meridional OHT of 1.35 PW (Johns et al. 2011). In contrast, the gyre OHT in the CO2-forced HadGEM2-ES experiment starts at ~0.7 PW and peaks at ~1.5 PW when CO2 reaches 4 times preindustrial levels. Given the large bias in the gyre OHT at the start of the simulation, it is likely that the simulated changes in HadGEM2-ES are unrealistic. However, this experiment serves to illustrate the potential for changes in the relationship between the AMOC and SSTs if changes to the total heat transport are dominated by changes in gyre circulations.

Fig. 11.

Decomposition of northward Atlantic Ocean heat transports at 30°N from the CO2-forced HadGEM2-ES experiment plotted in Fig. 10.

Fig. 11.

Decomposition of northward Atlantic Ocean heat transports at 30°N from the CO2-forced HadGEM2-ES experiment plotted in Fig. 10.

In addition to the contribution from changes in ocean heat transports, the SST response in the North Atlantic could be affected by a variety of confounding factors, including external forcings, atmospheric feedbacks, and other modes of internal variability. For this reason, we argue that reconstructions of the AMOC based on SSTs alone should be treated cautiously for periods of Earth's history when external climate forcings are known to have evolved rapidly.

b. Detecting changes to the AMOC using subsurface density fingerprints

In the previous section, we showed that indices derived from densities in the Labrador Sea and middepth (800–1800 m) Atlantic Ocean can be used to define measures of the AMOC that are robust in the presence of external forcings. In this section, we examine the signal-to-noise properties of trends in these indices and trends in the AMOC. In particular, we quantify the time taken to detect trends that are 1) statistically different from zero and 2) significantly different from internal variability as simulated in preindustrial control simulations.

Tables 6 and 7 summarize the results of this analysis for each CO2-forced simulation. Trend magnitudes are calculated for trend lengths of between 5 and 100 yr for 40 different start dates (indicated by the dashed gray lines in Fig. 10) and are determined to be significantly different from zero when p values for a two-tailed t test are less than 0.05 (Table 6). This test is a measure of the signal-to-noise ratio of each time series, as linear trends with large residuals will take longer to be identified as significant. Detection times for trends that are statistically different to internal variability (Table 7) are estimated using the same bootstrap methodology described by Baehr et al. (2008) and Roberts and Palmer (2012). For each trend length, percentiles for the internal variability of trends are estimated from the distribution of trends in 104 random segments of data from the corresponding control simulations. Forced trends are considered to have been detected at the 95% level if they fall outside either the 2.5 or 97.5 percentiles estimated from the control simulations (i.e., p = 0.05 for such a trend to have been observed in the control simulation).

Table 6.

Average number of years taken to identify trends that are significantly different from zero in the 1% yr−1 CO2-forced experiments. Multimodel mean detection times are given as an average ±1.96 × the standard error.

Average number of years taken to identify trends that are significantly different from zero in the 1% yr−1 CO2-forced experiments. Multimodel mean detection times are given as an average ±1.96 × the standard error.
Average number of years taken to identify trends that are significantly different from zero in the 1% yr−1 CO2-forced experiments. Multimodel mean detection times are given as an average ±1.96 × the standard error.
Table 7.

Average number of years for detection of an anthropogenic trend in the 1% yr−1 CO2-forced experiments. Multimodel mean detection times are given as an average ± 1.96 × the standard error.

Average number of years for detection of an anthropogenic trend in the 1% yr−1 CO2-forced experiments. Multimodel mean detection times are given as an average ± 1.96 × the standard error.
Average number of years for detection of an anthropogenic trend in the 1% yr−1 CO2-forced experiments. Multimodel mean detection times are given as an average ± 1.96 × the standard error.

Our results indicate that nonzero trends in indices of midocean density are identifiable significantly earlier than trends in the AMOC and Labrador Sea stratification (Table 6) because of the reduced high-frequency noise in the midocean density index. However, the times taken to detect forced trends in the AMOC, Labrador Sea stratification, and midocean density dipole indices are statistically indistinguishable (Table 7). In other words, observed changes to the midocean density index may provide a useful constraint for the more rapid identification of large-scale changes to the overturning circulation, but they may not provide any additional benefit for determining whether such changes have an anthropogenic origin or are part of internal variability.

6. Observed changes to midocean densities since 1960

Baehr (2011) previously used zonal density gradients from 26.5°N to define an observation-based AMOC detection variable but concluded that it was not possible to identify significant changes with only 5 yr of available data. In this section, we attempt to reconstruct large-scale density changes associated with the AMOC using temperature and salinity observations from the EN3 v2a dataset. We use the same dipole index and regions for the North Atlantic subpolar gyre and South Atlantic as defined in section 4c. Initially, temperature and salinity data from the full-field gridded analysis are used to calculate anomalies of potential density relative to a monthly climatology. These data are then masked such that only grid boxes with one or more salinity and temperature profiles to a depth of at least 1700 m are kept in the analysis. We use this masked version of the data to estimate volume-averaged density anomalies in the depth range of 800–1800 m, assuming that the average anomaly of available data is representative of the entire volume. Months with less than five grid boxes containing observations (in either the North or South Atlantic regions) are not included in the analysis.

To estimate the uncertainty associated with changes in sampling density, we apply the same methodology to monthly data from the HadGEM2-ES preindustrial control simulation. For each sampling distribution of the monthly EN3 analysis, a midocean density dipole index is calculated from a masked version of HadGEM2-ES temperature and salinity data and a difference is calculated by comparison with the model “truth.” This process is repeated 1000 times for data randomly selected from 5240 HadGEM2-ES monthly mean fields. The resulting distribution of errors is used to define the 2.5% and 97.5% percentile limits of empirically derived uncertainty. These limits should be considered a minimum estimate of uncertainty as they do not include estimates of instrumental precision/bias and will be, to some extent, model dependent.

Figure 12a shows the reconstructed density dipole index with empirical estimates of uncertainty plotted as vertical bars. Our estimated uncertainties are very large during the twentieth century, but are much smaller by 2004 because of the expansion of the Argo array and increased number of observations from depths greater than 1000 m. Because of the large uncertainties in the density dipole index prior to the mid-2000s, we have limited our discussion to the 2004–12 period during which midocean densities are well observed and can be compared to observed changes in the AMOC at 26.5°N (Figs. 12b,c).

Fig. 12.

(a) Midocean density dipole index estimated from monthly EN3 v2a analysis fields with empirical estimates of 95% confidence levels plotted as vertical black bars. (b) Midocean density dipole index and 95% confidence levels (diamonds with vertical bars) overlain with a linear trend (dashed line) fitted to annual mean data (filled circles). (c) Observed AMOC transports from the RAPID–MOCHA 26°N array (thin black line) overlain with a linear trend (dashed line) fitted to annual mean data (filled circles). The P values indicate the probability of zero correlation for a two-tailed t test.

Fig. 12.

(a) Midocean density dipole index estimated from monthly EN3 v2a analysis fields with empirical estimates of 95% confidence levels plotted as vertical black bars. (b) Midocean density dipole index and 95% confidence levels (diamonds with vertical bars) overlain with a linear trend (dashed line) fitted to annual mean data (filled circles). (c) Observed AMOC transports from the RAPID–MOCHA 26°N array (thin black line) overlain with a linear trend (dashed line) fitted to annual mean data (filled circles). The P values indicate the probability of zero correlation for a two-tailed t test.

From 2004 to present, the observed AMOC time series exhibits substantial intra-annual variability and a notable minimum during 2009/10 (McCarthy et al. 2012). Over the same period, the reconstructed density dipole index does not change by more than ±10%. This result is consistent with previous studies attributing the large-magnitude seasonal cycle in the AMOC observed at 26.5°N to a local wind-driven forcing (Chidichimo et al. 2010; Kanzow et al. 2010) and indicates that such variability is not associated with changes in the large-scale meridional density structure in the midocean. Linear trends fit to annual mean data indicate that both time series exhibit a weakening trend from 2004 to present. (Note that annual means are used to prevent autocorrelation on seasonal time scales.) However, the trend in the density dipole index is only marginally significant (n = 8, p = 0.058) and the trend in the AMOC at 26.5°N is not significant (n = 7, p = 0.13). We conclude that it is not yet possible to detect a robust trend in either the observed AMOC or a midocean density dipole index.

7. Conclusions

We have presented a multimodel analysis of spatial fingerprints of the AMOC derived from sea surface temperatures and subsurface densities. Preindustrial control simulations were evaluated using observations of the vertical structure of the AMOC at 26.5°N and climatological profiles of temperature, salinity, and density in the North Atlantic subpolar gyre. All models exhibited a significant relationship between the AMOC and subsurface densities in the subpolar gyre, with maximum correlations occurring at depths greater than 400 m. In the models that compared most favorably with observations, positive anomalies in the AMOC were associated with reduced stratification in the Labrador Sea and increased densities at midocean depths (800–1800 m) in the subpolar gyre. Maximum correlations occurred when anomalies in the AMOC lagged Labrador Sea stratification and midocean density anomalies by 2–6 yr and 0–3 yr, respectively. In all 10 models, positive AMOC anomalies were associated with warming in the North Atlantic, with maximum correlations occurring when SST anomalies lagged changes in the AMOC by 1–5 yr. However, the locations and magnitudes of maximum warming varied between models. In addition, in an idealized CO2-forced experiment, unanticipated changes in the partitioning of ocean heat transport between overturning and horizontal gyre components caused a break down in the expected relationship between the AMOC and SSTs. In contrast, indices of midocean density and Labrador Sea stratification were shown to accurately capture low-frequency AMOC changes in the presence of an evolving CO2 forcing. In addition, trends in midocean density were found to be detectable (although not attributable) significantly earlier than trends in the AMOC. For this reason, we suggest that subsurface density observations could be a useful complement to transport observations made at specific latitudes and may help with the more timely diagnosis of basin-scale changes in the AMOC. Finally, because of limited observations prior to 2004, we conclude that it is not yet possible to detect a robust trend in either the observed AMOC or an observation-based reconstruction of the midocean density dipole index.

Acknowledgments

We acknowledge the World Climate Research Programme's Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups (see Table 1) for producing and making available their model output. For CMIP, the U.S. Department of Energy's Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. Data from the RAPID-WATCH MOC monitoring project are funded by the Natural Environment Research Council and are freely available online (from http://www.noc.soton.ac.uk/rapidmoc). This work was supported by the Natural Environment Research Council and the Joint DECC/Defra Met Office Hadley Centre Climate Programme (GA01101). We would also like to thank J. Kettleborough, I. Edmond, and J. Gregory for the development of tools and code for the downloading, archiving, and analysis of CMIP5 data at the Met Office. Finally, we thank J. Baehr, two anonymous reviewers, and the editor for comments that helped significantly improve this manuscript.

REFERENCES

REFERENCES
Baehr
,
J.
,
2011
:
Detecting changes in the meridional overturning circulation at 26°N in the Atlantic from observations
.
Deep-Sea Res. II
,
58
,
1833
1836
.
Baehr
,
J.
,
H.
Haak
,
S.
Alderson
,
S.
Cunningham
,
J.
Jungclaus
, and
J.
Marotzke
,
2007
:
Timely detection of changes in the meridional overturning circulation at 26°N in the Atlantic
.
J. Climate
,
20
,
5827
5841
.
Baehr
,
J.
,
K.
Keller
, and
J.
Marotzke
,
2008
:
Detecting potential changes in the meridional overturning circulation at 26°N in the Atlantic
.
Climatic Change
,
91
,
11
27
.
Bentsen
,
M.
, and
Coauthors
,
2012
:
The Norwegian Earth System Model, NorESM1-M—Part 1: Description and basic evaluation
.
Geosci. Model Dev. Discuss.
,
5
,
2843
2931
.
Bingham
,
R.
, and
C.
Hughes
,
2009
: Signature of the Atlantic meridional overturning circulation in sea level along the east coast of North America. Geophys. Res. Lett.,36, L02603, doi:10.1029/2008GL036215.
Booth
,
B.
,
N.
Dunstone
,
P.
Halloran
,
T.
Andrews
, and
N.
Bellouin
,
2012
:
Aerosols implicated as a prime driver of twentieth-century North Atlantic climate variability
.
Nature
,
484
,
228
232
.
Boucher
,
O.
, and
Coauthors
,
2012
: Reversibility in an Earth System model in response to CO2 concentration changes. Environ. Res. Lett., 7, 024013, doi:10.1088/1748-9326/7/2/024013.
Bryden
,
H.
,
H.
Longworth
, and
S.
Cunningham
,
2005
:
Slowing of the Atlantic meridional overturning circulation at 25°N
.
Nature
,
438
,
655
657
.
Chidichimo
,
M.
,
T.
Kanzow
,
S.
Cunningham
,
W.
Johns
, and
J.
Marotzke
,
2010
:
The contribution of eastern-boundary density variations to the Atlantic meridional overturning circulation at 26.5°N
.
Ocean Sci.
,
6
,
475
490
.
Collins
,
M.
, and
Coauthors
,
2006
:
Interannual to decadal climate predictability in the North Atlantic: A multimodel-ensemble study
.
J. Climate
,
19
,
1195
1203
.
Cunningham
,
S.
, and
Coauthors
,
2007
:
Temporal variability of the Atlantic meridional overturning circulation at 26.5°N
.
Science
,
317
,
935
938
.
Curry
,
R. G.
,
M. S.
McCartney
, and
T. M.
Joyce
,
1998
:
Oceanic transport of subpolar climate signals to mid-depth subtropical waters
.
Nature
,
391
,
575
577
.
Danabasoglu
,
G.
,
W.
Large
, and
B.
Briegleb
,
2010
: Climate impacts of parameterized Nordic Sea overflows. J. Geophys. Res.,115, C11005, doi:10.1029/2010JC006243.
de Boer
,
A.
,
A.
Gnanadesikan
,
N.
Edwards
, and
A.
Watson
,
2010
:
Meridional density gradients do not control the Atlantic overturning circulation
.
J. Phys. Oceanogr.
,
40
,
368
380
.
Delworth
,
T.
, and
M.
Mann
,
2000
:
Observed and simulated multidecadal variability in the Northern Hemisphere
.
Climate Dyn.
,
16
,
661
676
.
Dickson
,
R. R.
, and
J.
Brown
,
1994
: The production of North Atlantic Deep Water: Sources, rates, and pathways. J. Geophys. Res.,99 (C6), 12 319–12 341.
Eden
,
C.
, and
J.
Willebrand
,
2001
:
Mechanism of interannual to decadal variability of the North Atlantic circulation
.
J. Climate
,
14
,
2266
2280
.
Folland
,
C.
,
T.
Palmer
, and
D.
Parker
,
1986
:
Sahel rainfall and worldwide sea temperatures, 1901–85
.
Nature
,
320
,
602
607
.
Ganachaud
,
A.
, and
C.
Wunsch
,
2003
:
Large-scale ocean heat and freshwater transports during the World Ocean Circulation Experiment
.
J. Climate
,
16
,
696
705
.
Gent
,
P. R.
, and
Coauthors
,
2011
:
The Community Climate System Model version 4
.
J. Climate
,
24
,
4973
4991
.
Getzlaff
,
J.
,
C. W.
Böning
,
C.
Eden
, and
A.
Biastoch
,
2005
:
Signal propagation related to the North Atlantic overturning
.
Geophys. Res. Lett.
,
32
, L09602, doi:10.1029/2004GL021002.
Girton
,
J. B.
, and
T. B.
Sanford
,
2003
:
Descent and modification of the overflow plume in the Denmark Strait
.
J. Phys. Oceanogr.
,
33
,
1351
1364
.
Gordon
,
C.
,
C.
Cooper
,
C.
Senior
,
H.
Banks
,
J.
Gregory
,
T.
Johns
,
J.
Mitchell
, and
R.
Wood
,
2000
:
The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments
.
Climate Dyn.
,
16
,
147
168
.
Gregory
,
J. M.
, and
Coauthors
,
2005
:
A model intercomparison of changes in the Atlantic thermohaline circulation in response to increasing atmospheric CO2 concentration
.
Geophys. Res. Lett.
,
32
, L12703, doi:10.1029/2005GL023209.
Griesel
,
A.
, and
M.
Maqueda
,
2006
:
The relation of meridional pressure gradients to North Atlantic Deep Water volume transport in an ocean general circulation model
.
Climate Dyn.
,
26
,
781
799
.
Hansen
,
B.
, and
S.
Østerhus
,
2007
:
Faroe Bank Channel overflow 1995–2005
.
Prog. Oceanogr.
,
75
,
817
856
.
Ingleby
,
B.
, and
M.
Huddleston
,
2007
:
Quality control of ocean temperature and salinity profiles—Historical and real-time data
.
J. Mar. Syst.
,
65
,
158
175
.
Johns
,
W.
, and
Coauthors
,
2011
:
Continuous, array-based estimates of Atlantic Ocean heat transport at 26.5°N
.
J. Climate
,
24
,
2429
2449
.
Jungclaus
,
J.
, and
Coauthors
,
2013
:
Characteristics of the ocean simulations in the Max Planck Institute Ocean Model (MPIOM) the ocean component of the MPI-Earth system model
.
J. Adv. Model. Earth Syst.
,
5
,
422
446
,
doi:10.1002/jame.20023
.
Kanzow
,
T.
,
U.
Send
,
W.
Zenk
,
A.
Chave
, and
M.
Rhein
,
2006
:
Monitoring the integrated deep meridional flow in the tropical North Atlantic: Long-term performance of a geostrophic array
.
Deep-Sea Res. I
,
53
,
528
546
.
Kanzow
,
T.
, and
Coauthors
,
2010
:
Seasonal variability of the Atlantic meridional overturning circulation at 26.5°N
.
J. Climate
,
23
,
5678
5698
.
Knight
,
J. R.
,
R. J.
Allan
,
C. K.
Folland
,
M.
Vellinga
, and
M. E.
Mann
,
2005
:
A signature of persistent natural thermohaline circulation cycles in observed climate
.
Geophys. Res. Lett.
,
32
, L20708, doi:10.1029/2005GL024233.
Kuhlbrodt
,
T.
,
A.
Griesel
,
M.
Montoya
,
A.
Levermann
,
M.
Hofmann
, and
S.
Rahmstorf
,
2007
:
On the driving processes of the Atlantic meridional overturning circulation
.
Rev. Geophys.
,
45
, RG2001, 10.1029/2005GL024233.
Kushnir
,
Y.
,
1994
:
Interdecadal variations in North Atlantic sea surface temperature and associated atmospheric conditions
.
J. Climate
,
7
,
141
157
.
Latif
,
M.
, and
Coauthors
,
2004
:
Reconstructing, monitoring, and predicting multidecadal-scale changes in the North Atlantic thermohaline circulation with sea surface temperature
.
J. Climate
,
17
,
1605
1614
.
Legg
,
S.
, and
Coauthors
,
2009
:
Improving oceanic overflow representation in climate models: The Gravity Current Entrainment Climate Process Team
.
Bull. Amer. Meteor. Soc.
,
90
,
657–670
.
Macrander
,
A.
,
U.
Send
,
H.
Valdimarsson
,
S.
Jónsson
, and
R. H.
Käse
,
2005
:
Interannual changes in the overflow from the Nordic Seas into the Atlantic Ocean through Denmark Strait
.
Geophys. Res. Lett.
,
32
, L06606, doi:10.1029/2004GL021463.
Mahajan
,
S.
,
R.
Zhang
,
T.
Delworth
,
S.
Zhang
,
A.
Rosati
, and
Y.
Chang
,
2011
:
Predicting Atlantic meridional overturning circulation (AMOC) variations using subsurface and surface fingerprints
.
Deep-Sea Res. II
,
58
,
1895
1903
.
Mann
,
M. E.
, and
K. A.
Emanuel
,
2006
:
Atlantic hurricane trends linked to climate change
.
Eos, Trans. Amer. Geophys. Union
,
87
,
233–241
.
Martin
,
G. M.
, and
Coauthors
,
2011
:
The HadGEM2 family of Met Office Unified Model climate configurations
.
Geosci. Model Dev.
,
4
,
723
757
.
Mauritzen
,
C.
,
J.
Price
,
T.
Sanford
, and
D.
Torres
,
2005
:
Circulation and mixing in the Faroese Channels
.
Deep-Sea Res. I
,
52
,
883
913
.
McCarthy
,
G.
, and
Coauthors
,
2012
:
Observed interannual variability of the Atlantic meridional overturning circulation at 26.5°N
.
Geophys. Res. Lett.
,
39
,
L19609
,
doi:10.1029/2012GL052933
.
Otterå
,
O.
,
M.
Bentsen
,
H.
Drange
, and
L.
Suo
,
2010
:
External forcing as a metronome for Atlantic multidecadal variability
.
Nat. Geosci.
,
3
,
688
694
.
Park
,
Y.
,
1999
:
The stability of thermohaline circulation in a two-box model
.
J. Phys. Oceanogr.
,
29
,
3101
3110
.
Roberts
,
C. D.
, and
M. D.
Palmer
,
2012
:
Detectability of changes to the Atlantic meridional overturning circulation in the Hadley Centre Climate Models
.
Climate Dyn.
,
39
,
2533
2546
.
Roemmich
,
D.
, and
Coauthors
,
2009
:
The Argo program: Observing the global ocean with profiling floats
.
Oceanography
,
22
,
34
43
.
Schlesinger
,
M.
, and
N.
Ramankutty
,
1994
:
An oscillation in the global climate system of period 65-70 years
.
Nature
,
367
,
723
726
.
Stommel
,
H.
,
1961
:
Thermohaline convection with two stable regimes of flow
.
Tellus
,
13
,
224
230
.
Sutton
,
R.
, and
D.
Hodson
,
2005
:
Atlantic Ocean forcing of North American and European summer climate
.
Science
,
309
,
115
118
.
Sutton
,
R.
, and
B.
Dong
,
2012
:
Atlantic Ocean influence on a shift in European climate in the 1990s
.
Nat. Geosci.
, 5, 788–792,
doi:10.1038/ngeo1595
.
Thorpe
,
R.
,
J.
Gregory
,
T.
Johns
,
R.
Wood
, and
J.
Mitchell
,
2001
:
Mechanisms determining the Atlantic thermohaline circulation response to greenhouse gas forcing in a non-flux-adjusted coupled climate model
.
J. Climate
,
14
,
3102
3116
.
Tjiputra
,
J. F.
,
C.
Roelandt
,
M.
Bentsen
,
D.
Lawrence
,
T.
Lorentzen
,
J.
Schwinger
,
Ø.
Seland
, and
C.
Heinze
,
2013
:
Evaluation of the carbon cycle components in the Norwegian Earth System Model (NorESM)
.
Geosci. Model Dev.
, 6, 301–325.
Toggweiler
,
J.
, and
B.
Samuels
,
1995
:
Effect of Drake Passage on the global thermohaline circulation
.
Deep-Sea Res. I
,
42
,
477
500
.
Toggweiler
,
J.
, and
B.
Samuels
,
1998
:
On the ocean's large-scale circulation near the limit of no vertical mixing
.
J. Phys. Oceanogr.
,
28
,
1832
1852
.
Trenberth
,
K. E.
, and
D. J.
Shea
,
2006
: Atlantic hurricanes and natural variability in 2005. Geophys. Res. Lett.,33, L12704, 10.1029/2006GL026894.
Vellinga
,
M.
, and
R.
Wood
,
2004
: Timely detection of anthropogenic change in the Atlantic meridional overturning circulation. Geophys. Res. Lett.,31, L14203, doi:10.1029/2004GL020306.
Voldoire
,
A.
, and
Coauthors
,
2011
:
The CNRM-CM5.1 global climate model: Description and basic evaluation
.
Climate Dyn.
, 40, 2091–
2121
.
Weaver
,
A. J.
, and
Coauthors
,
2012
:
Stability of the Atlantic meridional overturning circulation: A model intercomparison
.
Geophys. Res. Lett.
, 39, L20709,
doi:10.1029/2012GL053763
.
Willis
,
J. K.
,
2010
: Can in situ floats and satellite altimeters detect long-term changes in Atlantic Ocean overturning? Geophys. Res. Lett.,37, L06602, doi:10.1029/2010GL042372.
Winton
,
M.
,
R.
Hallberg
, and
A.
Gnanadesikan
,
1998
:
Simulation of density-driven frictional downslope flow in z-coordinate ocean models
.
J. Phys. Oceanogr.
,
28
,
2163
2174
.
Yang
,
J.
,
1999
:
A linkage between decadal climate variations in the Labrador Sea and the tropical Atlantic Ocean
.
Geophys. Res. Lett.
,
26
,
1023
1026
.
Yeager
,
S.
, and
G.
Danabasoglu
,
2012
:
Sensitivity of Atlantic meridional overturning circulation variability to parameterized Nordic Sea overflows in CCSM4
.
J. Climate
,
25
,
2077
2103
.
Yukimoto
,
S.
, and
Coauthors
,
2012
:
A new global climate model of the Meteorological Research Institute: MRI-CGCM3: Model description and basic performance.
J. Meteor. Soc. Japan
,
90
,
23
64
.
Zhang
,
D.
,
R.
Msadek
,
M. J.
McPhaden
, and
T.
Delworth
,
2011
: Multidecadal variability of the North Brazil Current and its connection to the Atlantic meridional overturning circulation. J. Geophys. Res.,116, C04012, doi:10.1029/2010JC006812.
Zhang
,
R.
,
2007
: Anticorrelated multidecadal variations between surface and subsurface tropical North Atlantic. Geophys. Res. Lett.,34, L12713, doi:10.1029/2007GL030225.
Zhang
,
R.
,
2008
: Coherent surface-subsurface fingerprint of the Atlantic meridional overturning circulation. Geophys. Res. Lett.,35, L20705, doi:10.1029/2008GL035463.
Zhang
,
R.
, and
T.
Delworth
,
2006
: Impact of Atlantic multidecadal oscillations on India/Sahel rainfall and Atlantic hurricanes. Geophys. Res. Lett.,33, L17712, doi:10.1029/2006GL026267.