1. Introduction
The North Atlantic Ocean has been shown to be a key region for the initialization of decadal forecasts (Dunstone et al. 2011), and sea surface temperatures (SSTs) in this region are likely important for the climates of the nearby continents of North America and Europe (Rodwell et al. 1999). SSTs in the North Atlantic show large multidecadal variability (Knight et al. 2005), which has been linked to drought in the Sahel region (Folland et al. 1986; Zhang and Delworth 2006), Atlantic hurricane formation (Goldenberg et al. 2001; Smith et al. 2010), precipitation over northern Europe (Sutton and Hodson 2005), and the growth and persistence of Arctic sea ice, which could also affect the climate of northern Europe (Screen 2013). In addition to the response to globally increasing greenhouse gas emissions, understanding the natural variability of this region is important in helping to improve the veracity of decadal predictions, which rely in particular on the North Atlantic subpolar gyre (NA SPG; Dunstone et al. 2011). Indeed, the NA SPG could be an important region in regulating decadal (Häkkinen and Rhines 2004) and longer-time-scale climate cycles (Kleppin et al. 2015). Because of the paucity of long-term observational records within the NA SPG, much of the mechanistic understanding must be gained through analysis of climate models, some of which is now summarized.
There have been many studies investigating the interannual/decadal variability of the NA SPG, which may be useful in adding value to predictions made up to a decade ahead. Given the number of such studies, we present here a very brief review. One of the first studies on NA SPG decadal variability proposed a mechanism related to temperature-induced gyre changes that advect salinity into the model’s sinking regions and have a periodicity of 50 yr in an early coupled climate model (Delworth et al. 1993). Following on from the idealized-ocean work of Frankignoul et al. (1997), this mechanism was ocean only, with the atmosphere providing white noise forcing. However, the agreement between idealized models and fully coupled general circulation models was better in the subtropics than in subpolar regions (Frankignoul et al. 1997). Later work found the mechanism of Delworth et al. (1993) in the HadCM3 model (Dong and Sutton 2005) but with a reduced period of just 25 yr, with this reduction attributed in part to the removal of flux corrections and the improved representation of surface temperature gradients in the ocean.
Using the ECHAM3/Large Scale Geostrophic (LSG) ocean model, Timmermann et al. (1998) searched for the observationally based salinity-dominated mechanism of Wohlleben and Weaver (1995) related to great salinity anomalies. A periodicity of 35 yr was reported, and this time the atmosphere was postulated to play a direct, coupled role. However, in parallel work using a coupled model with the same atmosphere but a different ocean (ECHAM3/HOPE), Grötzner et al. (1998) also suggested a coupled decadal mode but with the now even shorter time scale of 17 yr and this time with temperature changes playing an important role. To try to reconcile these differences, Eden and Willebrand (2001) investigated the relative importance of heat and freshwater fluxes related to the North Atlantic Oscillation (NAO) in an ocean-only model and found that, of the two, heat fluxes were more important than freshwater fluxes for the interannual/decadal variability of the NA SPG. However, the coupling between the ocean and atmosphere on multiannual time scales appears to go in both directions (Rodwell et al. 1999; Battisti et al. 1995), suggesting that coupled models are required in order to fully capture the interactions.
Disagreements remain over the main contribution to density changes in the NA SPG, along with questions about the degree to which the atmosphere plays a coupled role and the key processes that set the time scale. However, the general periodicity of simulated multiannual but subcentennial variability has begun to crystallize (Frankcombe et al. 2010). In addition to the aforementioned works, other studies of this variability within the NA SPG continually find periodicities of about 20 yr (Visbeck et al. 1998; Watanabe et al. 1999; Holland et al. 2001; Eden and Greatbatch 2003; Dai et al. 2005; Álvarez-García et al. 2008; Danabasoglu 2008; Born and Mignot 2012; Tulloch and Marshall 2012; Sévellec and Fedorov 2013; Escudier et al. 2013; Kwon and Frankignoul 2014). This time scale is sometimes attributed to the basin-crossing time scale of Rossby waves in the NA SPG (Frankcombe et al. 2010), though many studies attribute it to the time to build up sufficient temperature anomalies. Indeed, the role of the forcing from Rossby waves has recently been called into question (MacMartin et al. 2013), and the importance of wave processes in controlling decadal variability is still unclear (Février et al. 2007; Roussenov et al. 2008). This approximately 20-yr variability is generally separate from centennial variability in the Atlantic, which relies on the long advective time scales to bring anomalies from the tropical Atlantic or Arctic and in which salinity is more consistently the dominant driver of the related density changes (Vellinga and Wu 2004; Jungclaus et al. 2005; Menary et al. 2012). Indeed, the role of salinity in either weakening or strengthening circulations in the NA SPG may depend on time scale (Deshayes et al. 2014).
As previously noted, limited instrumental observations within the NA SPG make it hard to detect the existence of decadal variability. However, paleo-reconstructions do suggest increased variance at decadal time scales (Mann et al. 1995), and indeed 20-yr variability can be detected on the outskirts of the NA SPG in paleo-proxies (Chylek et al. 2012). Additionally, the relative importance of temperature or salinity variability in real-world overturning circulation changes has been investigated. On multiannual time scales, Curry and McCartney (2001) found that the Labrador Sea potential energy anomaly and the overturning were thermally driven—insofar as temperatures changed twice as much as salinities in the sinking regions (after scaling by the thermal and haline expansion coefficients).
To bring together this previous work with climate models, Fig. 1 schematically depicts the studies mentioned so far along with the reported period and whether the proposed mechanism is coupled or ocean only. Additionally, we note whether the time scale is reported to be primarily set by either wave processes, the mean circulation strength and the integration of anomalies within the NA SPG, and the interaction with the deep western boundary current. Also noted is whether density changes in the Labrador Sea (or model-equivalent sinking region) are reported to be temperature or salinity dominated. In short, there is much disagreement between the models on the key processes, the details of the mechanism (see above for some examples; the reader is referred to the specific studies for further details), the degree of atmospheric interaction, and the dominant driver of density changes.
A summary of some of the literature on simulated decadal variability in the NA SPG, with a particular emphasis on studies that found self-sustaining cyclical behavior. Key regions of the NA SPG are marked. The figure legend (right) denotes the studies that we have attempted to synthesize and an associated numerical identifier. We noted where these studies report a significant peak in the power spectrum on decadal time scales as well as whether the mechanism is primarily ocean only or inherently coupled. Studies where the atmosphere is postulated to amplify—but not transmute—the signal are marked with an asterisk. For each study the feedback or process that is reported as crucial in setting the time scale is marked on the map using a simple numbering system: 1) feedbacks relating to the deep-water pathways and their interaction with the northward-flowing western boundary current, 2) Rossby wave (or sometimes “geostrophic self advection”) transit times across the NA SPG, and 3) the mean advection time scale for anomalies to propagate into the NA SPG from the tropics, or for small anomalies to integrate over time. Last, using the same numerical key, the studies are split into whether temperature or salinity is reported to control decadal-time-scale density changes in the Labrador Sea. Studies in parentheses appear in more than one category. This represents a drastic simplification of each of these studies and the reader is referred to the original works for further details. In particular, the reported feedback/process that sets the overall time scale to some degree also reflects the precise focus of the particular study.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0106.1
As analyzing decadal variability requires many decades/centuries of integration, these previous studies generally use low-resolution coupled models (>1° ocean resolution, >2° atmosphere resolution) or higher-resolution ocean-only models. There are reasons to suppose that improved atmospheric resolution could affect the amplitude of decadal variability (Danabasoglu 2008), while improved ocean resolution and associated representation of the Gulf Stream and other boundary currents may affect the precise time scales of multiannual/decadal variability (Grötzner et al. 1998; Gelderloos et al. 2011; Hodson and Sutton 2012). Higher-resolution topography may also be expected to affect the efficacy of wave processes as compared to idealized ocean models with smoothed or no topography (Roussenov et al. 2008; Zhang and Vallis 2007) and improve deep-water pathways (Spence et al. 2012). At high ocean resolution, eddy-induced mixing can be left explicit rather than parameterized (or the parameterization significantly turned down), which may impact the magnitude and variability of ocean heat and freshwater transports (Volkov et al. 2008; Treguier et al. 2014). Stronger sea surface temperature gradients, associated with higher ocean resolution, may improve the strength of atmosphere–ocean coupling (Brayshaw et al. 2008). Ultrahigh resolution within the Agulhas region has also been shown to affect the variability of the simulated low-latitude Atlantic overturning (Biastoch et al. 2008b).
In this study, we document the drivers of NA SPG variability in a new, high-resolution coupled model that represents a rare combination of high resolution (in both ocean and atmosphere) and the multicentury length integration required to analyze decadal time-scale modes. We examine whether high resolution, and the associated processes it allows, affects the nature of simulated decadal variability.
The paper is structured as follows: Section 2 describes the model and data used. We then briefly characterize the model in section 3 before exploring the mechanism of decadal variability in some depth throughout section 4. The implications of our findings are discussed in section 5 followed by conclusions in section 6.
2. Model description and experimental setup
We examine a prototype of the Met Office Hadley Centre’s state-of-the-art coupled ocean–atmosphere–land ice global environment model, HadGEM3 (hereafter referred to as HG3). A high-resolution, near-present-day control simulation has been run for 460 years. The atmosphere component is the Met Office unified model, version 7.7 (Walters et al. 2011). It has a horizontal resolution of N216 (92 km at the equator) and 85 levels in the vertical with a model top at 85 km with at least 30 levels in or above the stratosphere. The ocean is resolved on the NEMO tripolar grid (0.25°, 75 depth levels, version 3.2; Madec et al. 2008), with a pole under Antarctica and poles on either side of the Arctic Ocean in Asia and North America. The ocean in HG3 was initialized from rest on 1 December using the 2004–2008 time-mean EN3 (Ingleby and Huddleston 2007) December-time climatology and subsequently allowed to freely evolve with repeating 2000 external forcings in the atmosphere. The year 2000 was chosen as it combined a well-sampled and recorded set of external forcings with relatively neutral conditions in major climate indices, such as El Niño; for further details of the model configuration and other simulations, see Walters et al. (2011).
We use observed data from the EN4 objective analysis (Good et al. 2013), which provides infilled, optimally interpolated fields of temperature and salinity on a 1° × 1° grid from 1900 to present day. EN4 is an updated version of EN3, with improved quality control and error estimates, but was not available when the climate model was initialized. We use the period 1900–2013 to construct a simple climatology for comparison with HG3 and note that the biases in HG3 are large enough (see section 3a) that the method used to construct the climatology is unlikely to be of first-order importance (i.e., the results are not sensitive to choosing 1900–2013 or 1960–2013 climatologies). Unlike the HG3 model, which is run with interannually constant forcings appropriate for the year 2000, these observational data also include the effects of all other natural and anthropogenic forcings.
HG3 is a precursor to the model used in the Met Office global seasonal forecast (GloSea5; MacLachlan et al. 2014). GloSea5 will also be similar to the new decadal prediction model. However, there are some differences between the HG3 and GloSea5 models, as GloSea5 underwent additional development while the HG3 control was running. Most importantly for the present study of the NA SPG is the more diffuse thermocline in the HG3 ocean (NEMO, version 3.2) as compared to GloSea5 (NEMO, version 3.4; see discussion section) (Megann et al. 2014). Despite this the NA SPG biases in upper-ocean temperature and salinity (compared to EN4) are small compared to many other coupled climate models used to study NA SPG variability [e.g., Escudier et al. (2013) and Wang et al. (2014) for SSTs; see section 3a]. Further details of global mean-state biases within the atmosphere and ocean in HG3 can be found in Walters et al. (2011).
3. Characterizing the model
We now examine the North Atlantic mean-state biases and signal of decadal variability in HG3 in some more detail as a precursor to investigating the mechanisms of variability that exist on top of these biases. In all cases, decadal variability refers to 5-yr smoothed data, unless otherwise stated.
a. NA SPG mean state
Mean-state biases in top-500-m depth-averaged temperatures (T500), salinities (S500), and densities (ρ500) in the NA SPG are less than ±3°C, ±0.4 psu, and ±0.1 kg m−3 in the interior NA SPG, with larger +4°C, +0.6-psu, and ±0.2 kg m−3 biases in the boundary current regions (Fig. 2). The temperature and salinity biases are close to being density compensating in the NA SPG, but in the subtropical gyre (not the focus of this study) temperature biases dominate, resulting in lighter waters. The anomalously cold region in the western SPG, often attributed to the simulated Gulf Stream being too zonal (Kwon et al. 2010), is not as large as in many coupled climate models (Scaife et al. 2011). Warm anomalies exist all along the NA SPG northern boundary currents. These anomalies are associated with reduced ice distribution around southern Greenland and in the Labrador Sea (not shown). Within the NA SPG, simulated deep convection, as estimated from the annual standard deviation in March mixed layer depths [using the mixed layer estimation method of Kara et al. (2000)], is located in the Labrador Sea and Irminger Current.
(a) T500, (b) S500, and (c) ρ500 biases in HG3 (computed from full time series) compared to EN4. Gray shading is used for regions shallower than 500 m. (d) Standard deviation in March mixed layer depths (Kara et al. 2000), to highlight where deep convection occurs.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0106.1
The Atlantic meridional overturning circulation (AMOC) streamfunction in the model is shown in Fig. 3a. The zero streamfunction line sits at a depth of 2–3 km, with the maximum overturning occurring at a depth of approximately 1 km. The deeper overturning cell, representing Antarctic Bottom Water (AABW) and Lower North Atlantic Deep Water (LNADW), has a strength of around 3 Sv (1 Sv ≡ 106 m3 s−1), whereas the shallower AMOC cell, representing the western boundary current and Upper North Atlantic Deep Water (UNADW), has a mean strength of 17 Sv for the last 200 yr of the simulation.
(a) Time-mean Atlantic overturning streamfunction in HG3. The contour interval is 2 Sv and the zero line is marked with a gray contour. At 26.5°N the profile from the RAPID array is overlaid on the same color/contour scale. The depth of the maximum in the RAPID profile is marked with a cross. Note that the latitudes north of 45°N are approximate (within 1°) because of the increasingly curved nature of the model grid toward the two northern poles. (b) Time-mean NA SPG barotropic streamfunction in HG3. Contour interval is 10 Sv. (c) Time series of the overturning streamfunction at 26.5°N and 1000 m in HG3 (red). Also shown are the time-mean and annual mean standard deviation from the 10 years of RAPID data (black). (d) Time series of the minimum (multiplied by −1) of the barotropic streamfunction in the NA SPG in HG3.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0106.1
At 26°N it is possible to directly compare the streamfunction in the model to the Rapid Climate Change (RAPID) observations. The depth of the RAPID overturning maximum is marked with a cross and is approximately 200 m deeper than in the simulations, which at these depths represents a single model grid cell in the vertical. The depth of the RAPID zero streamfunction line is around 4 km, much deeper than simulated. This is not uncommon in models and may be partly an artifact of computing the simulated overturning using the full three-dimensional velocities (Roberts et al. 2013), although some models do represent a much deeper upper cell (Yeager and Danabasoglu 2012). Nevertheless, using a “RAPID-style” calculation as in Roberts et al. (2013), with a depth of no motion at 4740 m, yields a zero streamfunction depth approximately only 250 m deeper than using the full three-dimensional velocities. The structure and variability of the streamfunction shallower than this are essentially unchanged. Finally, the NA SPG barotropic streamfunction and associated time series are also shown (Figs. 3b,d) and broadly compare well to observational estimates and high-resolution models (Treguier et al. 2005).
Although the depth (1000 m) and strength (17 Sv) of the maximum of the upper AMOC cell are consistent with observations, the simulated annual variability in this index is weaker than observed. The simulated annual mean AMOC streamfunction at 26.5°N and 1000-m depth has a standard deviation of 1.2 Sv (0.9 Sv if first detrended), compared to an annual standard deviation of 2.3 Sv from the 10 years of RAPID data available (Fig. 3c). Additionally, the simulated index begins at a low value and then takes several centuries to spin up to a more stable state more favorably comparable to the observed mean. Although this represents an improvement in this index of the NA circulation, the spinup of the overturning circulation also results in an increase in northward heat and salt transport within the Atlantic Ocean, causing the NA SPG to drift away from its initialized state to a warmer and saltier state, seen in Fig. 2.
The simulated AMOC index also shows some evidence of multiannual/decadal variability, particularly at the more northerly latitudes of the NA SPG (not shown) in addition to 26°N, as in other models (Zhang 2010). The maximum correlation between the simulated AMOC indices at 26.5° and 50°N occurs when the 50°N index leads by 1 yr (correlation of 0.63, with a correlation of 0.12 required for significance at the 95% level), suggesting the lower latitude variability is responding to variability farther north in the NA SPG. We now move on to examine the decadal variability of the NA SPG in more detail.
b. Signal of decadal variability
The time-mean T500 simulated in HG3 is shown in Fig. 4a along with contours at 6° and 10° to mark the general shape of the NA SPG. A comparison with observations (EN4) again shows the general warm bias of the NA SPG, particularly toward the edges of the gyre. A power spectrum for T500 over the whole region reveals a significant peak at a period of 16–17 yr (Fig. 4b). This periodicity exists whether using the entire simulation or alternatively removing the first 200 yr (not shown), suggesting that it is not merely an adjustment process, and so we use the entire time series to maximize the available data. Additionally, the periodicity is not unique to any of the four individual subregions within the NA SPG (dashed regions in Fig. 4a); all show a significant peak at 16–17 yr, as do the North Atlantic Current (NAC) and Irminger regions (not shown). Indeed, in HG3 many other large-scale ocean indices in the NA SPG also reveal peaks in their power spectra at periods of 16 and 17 yr, such as SSTs, depth-averaged salinities, the AMOC at 50°N, and the strength of the NA SPG itself (as defined by the barotropic streamfunction; cf. Fig. 3).
(a) Time-mean T500 in HG3. Contours at 6° and 10°C are also marked (black) to show the shape of the gyre and for comparison with equivalent contours from EN4 (gray). Areas in white are shallower than 500 m. The dashed gray box denotes the four quadrants and fifth overall region for which power spectra of T500 were produced. The Irminger Current region (red box) and Gulf Stream/NAC region (blue box) analyzed in the text are also marked. The dashed black line stretching south from the Grand Banks denotes the transect location for dynamic height analysis. (b) The T500 power spectrum for the whole subpolar gyre region (combination of all four quadrants). An estimate of significance is given by the 5%–95% confidence intervals for a red noise process with the same mean and standard deviation. Periods of 16–17 yr are highlighted with the blue shading. (c) As in (b), but for the NAO index, defined as the difference between simulated sea level pressures over the Azores and Iceland.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0106.1
In addition to these ocean indices, the NAO index also shows periodicity at 16–17 yr in its otherwise much whiter spectrum (Fig. 4c). This is suggestive of a link from ocean to atmosphere in the region of the North Atlantic in which the ocean can impart some of its long-term memory to the atmosphere. Such a feedback might in general be expected to be weaker than similar atmosphere-to-ocean processes and related to the strength of the ocean circulation and SST gradients (Nonaka and Xie 2003), and thus detection of this feedback is perhaps at least in part due to the increased signal-to-noise ratio resulting from the length of the control simulation (though we note that this is still short compared to many previous studies with lower-resolution models). The mechanistic drivers behind this 17-yr mode in the ocean and atmosphere, and the reasons for the particular time scale, are investigated in the next sections, initially characterizing the variability in the NA SPG as a whole before analyzing the specific processes in different regions.
4. Mechanism of decadal variability in the NA SPG
We now diagnose the mechanism of decadal variability within the NA SPG, beginning with a heat budget for the region before investigating how temperature anomalies propagate around the gyre.
a. Heat budget
To begin to understand the variability of T500 in HG3, a heat budget of the NA SPG is diagnosed (appendix A). There is considerable variability in the net heat flux into the NA SPG, the majority of which appears to be attributable to the advective heat fluxes from the south, which results in decadal-time-scale heat content changes within the NA SPG. Annual and decadal correlations between the total heat flux and net advective fluxes are 0.75 and 0.69 (for annual and decadal data the 95% significance levels, assuming a two-tailed t test, are 0.12 and 0.37, respectively), whereas the same for the total heat flux and net surface fluxes are 0.63 and 0.29 (the regression gradients scale similarly), respectively, suggesting that particularly on decadal time scales advective heat fluxes dominate the variability. Once within the NA SPG, how do these heat content anomalies evolve?
b. Lagged regression analysis
To investigate the spatial characteristics of the heat content variability, lagged regressions of NA SPG T500 on SST spatially averaged over the NA SPG were performed (Fig. 5). T500 anomalies can be seen propagating around the NA SPG: eastward along the southern boundary while spreading into the interior with a time scale of around 4–6 yr (notably slower than implied by the mean circulation speed in this region); westward along the northern edge but south of the Greenland–Iceland–Norwegian (GIN) Seas; and into the central Labrador Sea as opposite-sign anomalies form in the Gulf Stream region. A similar evolution of anomalies was also found when regressing T500 on T500 spatial averages over the eastern SPG, NAC region, or Labrador Sea (not shown). The remainder of Fig. 5 will be discussed in section 4f. Features such as the Reykjanes Ridge can be seen diverting the flow. Although not shown here, there is little evidence of significant amounts of the signal diverting into the GIN Seas in the far northern part of the SPG. The heat content anomalies reach the Labrador Sea from the eastern SPG within a couple of years, but several more years are required for the anomalies to spread into the interior SPG. As the heat content anomalies in the Labrador Sea build up, so does a cold anomaly in the Gulf Stream/NAC region. The opposite phase of the cycle now begins.
Regressions between basinwide North Atlantic (45°–65°N) SSTs and (left)–(right) SST, T500, SHF, SSS, December–February (DJF) MSLP, and ice fraction. (top)–(bottom) The SST index lags then leads the fields from −6 to +4 and then +11 yr. The same color palette is used for each regression map with the units and scale for each regression slope shown below.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0106.1
The underlying essence of the cycle is captured by regressing T500 indices in the northern and southern edges of the NA SPG against each other [Fig. 6; the same result is also found when removing the spinup phase (not shown)]. This shows the southern edge of the NA SPG leading the northern edge by 4–6 yr and subsequently lagging changes in the northern edge by 0–2 yr with opposite sign, yielding a half period of 4–8 yr and a full period of 8–16 yr (constrained here to be even by the use of annual data). The time scale is further increased by 2 yr (putting the 16–17-yr spectral peak more toward the center of this range) if a third location in the eastern SPG is added to the regression model, forcing the signal to go via the eastern SPG (by regressing the southern index with an index of the eastern NA SPG and then regressing the index of the eastern NA SPG with the northern index; not shown), suggesting that the spread in time scales is perhaps related to the superposition of various advective pathways. This decadal mode is generally confined to the top 500 m–1 km, with the exception of the central Labrador Sea where it extends to around 2 km (not shown). Decadal variability in the band 10–30 yr, encompassing the spectral peak at 17 yr, explains >15% of the interannual variability in T500 within the NA SPG, with this value rising to >30% in the center of the gyre.
The lagged correlation between the Irminger Current and NAC T500. Regions are as marked in Fig. 4a. Time series have been filtered with a bandpass filter of 5–150 yr to highlight the decadal correlations by removing annual variability and the long-term drift. An estimate of the significance is provided by the 95% (red) and 99% (blue) confidence intervals estimated by creating 40 000 random time series with the same mean, standard deviation, and applied filtering.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0106.1
The lagged regression analysis leads to two key questions: First, what is controlling the apparent propagation of the heat content anomalies in both the Gulf Stream extension/NAC and the northern boundary currents/Irminger Current? Second, what is the negative feedback that forms the opposite sign anomaly in the NAC, resulting in a cyclical mechanism and a spectral peak in NA SPG temperatures?
c. Heat content anomalies in the NAC region
To determine what controls the heat content changes on the southern boundary of the NA SPG, the heat budget of the NAC region is examined in more detail. A region was chosen where simulated zonal currents are much stronger than meridional or vertical currents (see Fig. 4a, blue box). This simplifies the later interpretation of the decomposition of advective heat fluxes into circulation and temperature components. As noted in appendix A, it is not possible to close the heat budget precisely, which becomes more apparent for smaller subregions. Table 1 shows the time-mean advective components and net surface heat fluxes (SHF) for the top 500 m of the NAC. Note that the choice of reference temperature becomes irrelevant when considering the net transport through all faces combined but not when considering open sections (Schauer and Beszczynska-Möller 2009). The most important advective heat fluxes are from the east and west, associated with the mean volume transport through the region from east to west. These advective heat fluxes are approximately balanced by the surface heat fluxes, but the sum of the two is not identical to the actual heat content change implied by the in situ temperatures. This is due to missing diagnostics (see appendix A) and the use of monthly means, rather than at each model time step, when computing υT. However, although the means are slightly different, the variability in both time series is well correlated on all time scales at monthly or longer-term sampling (Table 1). Thus, in the ensuing analysis of the variability we treat the budget as sufficiently closed.
Time-mean simulated heat fluxes into the NAC and Irminger Current regions (TW; referenced to 0°C).
The annual and decadal time-scale correlations (regression gradients; W = watts) between the advective heat fluxes and net actual heat content changes in the NAC are 0.82 (0.92 WdOHC/Wadv) and 0.54 (0.40 WdOHC/Wadv), respectively, compared to 0.43 (0.92 WdOHC/Wsurf) and 0.20 (0.20 WdOHC/Wsurf) for the correlation between surface heat fluxes and the net heat content change (for annual and decadal data the 95% significance levels, assuming a two-tailed t test, are 0.12 and 0.37, respectively). Thus, much of the annual and decadal variability in the heat content changes in the Gulf Stream is associated with advective heat fluxes, but there is a role for surface fluxes to modulate these changes, even on decadal time scales. See appendix B for the full heat transport breakdown. The remaining question regarding these advective heat fluxes is whether they are due to the anomalous circulation or anomalous temperature.
For the NAC region it can be seen that slightly more of the advective heat flux variability arises from anomalous circulation advecting mean temperature anomalies (υ′T0; Table 3). Although the magnitudes are similar between υ′T0 and υ0T′ components, their relationships with the net ocean heat transport (OHT; i.e., υT) are not, with υ0T′ having a higher positive correlation with OHT. Correlations between υ′T0 and OHT are 0.29, 0.36, and 0.42 on monthly, annual, and decadal time scales, respectively, compared to 0.00, −0.16, and −0.23 for υ0T′ (Table 2). This holds throughout the western half of the southern edge of the NA SPG (not shown) and is associated with a strong background temperature gradient. Thus, υ′T0 appears to be the dominant advective heat flux in the NAC region on all time scales.
Correlations (regression slopes in brackets) between υT and advective heat flux components in the NAC and Irminger Current at various time scales (TW). The 95% significance levels, assuming a two-tailed t test and accounting for some missing data, are 0.03, 0.12, and 0.37 for monthly, annual, and decadal data, respectively.
Standard deviations of advective heat flux components in the NAC and Irminger Current at various time scales (TW).
d. Heat content anomalies in the Irminger Current region
The same breakdown of heat content changes in a particular region was applied to the Irminger Current at the entrance to the Labrador Sea (Fig. 4a, red box). Similarly to the NAC region, this region was chosen where horizontal circulation was well defined in a particular direction and much larger than all orthogonal circulations. The breakdown of heat fluxes into surface, net advective, and advective subcomponents is shown in Table 1. Similarly to the Gulf Stream region, the net surface and net advective heat fluxes approximately balance but do not fully explain the directly calculated heat content change. However, as before, the correlation between the sum of the surface and advective components and the flux implied by the actual heat content change is very good on all time scales, and so we again treat the budget as sufficiently closed.
For the individual fluxes, on annual time scales, the correlation (regression gradient) between the advective heat fluxes and net heat content changes is 0.56 (0.56 WdOHC/Wadv), again marginally greater than the correlation between surface heat fluxes and net heat content changes at 0.47 (0.52 WdOHC/Wsurf). On decadal time scales these drop to 0.21 (0.08) and 0.19 (0.09) for advective and surface fluxes, respectively. Despite these low decadal correlations, there is still a very large correlation between their sum and the actual net heat content change (Table 1), suggesting that on these decadal time scales no single component of the heat budget can be considered the controlling influence. This is also indicated by the strong anticorrelation between advective and surface heat fluxes of −0.87 on decadal time scales (for annual and decadal data the 95% significance levels, assuming a two-tailed t test, are 0.12 and 0.37 respectively).
In contrast to the Gulf Stream region, for the Irminger Current the most important advective heat flux is that due to the mean circulation advecting anomalous temperature (υ0T′; Table 3). The value υ0T′ has slightly greater variability on all time scales than υ′T0 and also shows larger correlations (and regression gradients) with the actual OHT changes on all time scales. Correlations between OHT and υ0T′ for monthly, annual, and decadal variability are 0.83, 0.34, and 0.29, respectively, whereas correlations between OHT and υ′T0 are much smaller (Table 2). In our Irminger Current box the zonal currents are an order of magnitude larger than in all other directions, and so we suggest that it is the zonal mean circulation that is playing an important role in moving heat content anomalies from east to west on the northern edge of the NA SPG. Additionally, the simulated mean temperature of the core of the inflow and outflow waters differs by 0.2 K in the Irminger Current region, compared to 1.6 K for the NAC region, which may help to explain the smaller standard deviations in advective heat fluxes through the Irminger Current region.
In summary, the heat budget for the NA SPG as a whole has been diagnosed, and it has been seen that advective heat fluxes play an important role on decadal time scales but that the relative contributions of circulation and temperature anomalies to the OHT are region specific. We now investigate the remaining question of what controls the negative feedback between Labrador Sea and NAC temperature anomalies.
e. Negative feedback between Labrador Sea and Gulf Stream T500
The anomalous temperatures in the Labrador Sea, which are related to the increased heat flux into the region, affect deep-water formation in this region. As noted in section 1, an assessment of related studies suggests an approximately even split between temperature and salinity control of the Labrador Sea density changes related to increased deep-water formation on decadal time scales. Following Delworth et al. (1993), we decompose the simulated density changes in the Labrador Sea into those due to temperature and those due to salinity (Fig. 7a). This analysis suggests that in HG3 simulated density changes in the Labrador Sea are due to temperature-induced density changes (annual correlation with actual density is 0.64; correlation required for significance at the 95% level, assuming a two-tailed t test, is 0.12) rather than salinity-induced density changes (annual correlation with actual density is −0.06). A lagged correlation analysis confirms that on both annual and decadal time scales density changes are temperature controlled (Fig. 7b). We hypothesize that simulated dense water formation in the Labrador Sea in HG3 contributes to circulation anomalies in the NAC region via the creation of an anomalous north–south dynamic height gradient and as such acts as a negative feedback on NA SPG temperatures.
(a) 5-yr smoothed Labrador Sea (55°–62°N, 50°–60°W) top-500-m mean density (black) and contributions from temperature (by keeping salinity at the time mean in the density equation of state; blue) and from salinity (by keeping temperature at the time mean in the density equation of state; red). Temperature and salinity were both linearly detrended prior to computing density. (b) Lagged correlation of temperature-induced (blue) and salinity-induced (red) density against actual density for detrended data, either unsmoothed (dashed) or presmoothed with a 5-yr running mean [solid; as in (a)]. Temperature-/salinity-induced density leads actual density at negative lags.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0106.1
To examine this hypothesis we calculate a composite difference in the density in a cross section through the NAC, which lags the density upstream in the Labrador Sea (Fig. 8a). To the north the connection between surface and deep water is revealed, with the signal sinking below the surface as it progresses southward. The north–south density gradient is associated with a change in the local dynamic height (Fig. 8b). Despite the negative density anomaly in the south, it can be seen that a large part of the dynamic height anomaly is controlled by the northern, positive density anomaly.
Transect south from the Grand Banks through the NAC at 47.5°W, as shown in Fig. 4a. Density (referenced to 0 m) composite of high-minus-low densities in the Labrador Sea, computed by averaging all cases where Labrador Sea volume mean density (computed over the region 56°–61°N, 47°–55°W, 0–1000 m) was at least one standard deviation larger than the time mean and subtracting the average of all cases where density was at least one standard deviation less than the time mean. Cross sections lag the Labrador Sea index by 1 yr. (b) As in (a), but for dynamic height composites (relative to 1500 m). (c) As in (a), but for the geostrophic circulation (relative to 1500 m). (d) As in (c), but first removing the NAO signal from the density field as in Polo et al. (2014) (see text). Cross-sectional time-mean density is indicated by the black contours. All data have been detrended and 5-yr smoothed. Only data significant at the 99% level are shown. Significance is estimated using a moving blocks bootstrap approach (Wilks 1997) reconstructing the composites 10 000 times by resampling the data with a block length estimated from the autocorrelation in each original composite.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0106.1
As the signal of anomalous density spills out of the Labrador Sea, this dynamic height gradient increases and is balanced by anomalous shear in the geostrophic velocities (Fig. 8c). The mean geostrophic velocity anomaly between the surface and 500 m for the pictured transect is 1.2 cm s−1, increasing to 1.6 cm s−1 for the top 200 m only. Thus, an increase in density in the Labrador Sea, associated with a cooling in this region, is followed by a strengthening of the circulation in the NAC and thus an increase in northward OHT into the NA SPG (with likely also some additional contribution from υ′T′ as the anomalous circulation acts on anomalously warm, low-density surface water; cf. Fig. 8a). This acts as a negative feedback on the NA SPG temperatures. We now discuss the atmospheric contribution to these ocean feedbacks.
f. The role of the atmosphere
Although the proposed mechanism of decadal (17 yr) variability in HG3 has been described mostly in terms of ocean dynamics, there are regions where the atmosphere directly forces, or acts as a positive feedback on, the ocean variability.
For example, the negative feedback dipole between Labrador Sea and NAC temperatures is reminiscent of the Ekman response to NAO forcing. To quantify the instantaneous (i.e., zero lag) impact of the NAO, we attempt to isolate its signal, similarly to the analysis of Polo et al. (2014). Specifically, the annual mean three-dimensional ocean density field was regressed onto the wintertime NAO index (both unfiltered; not shown). The direct impact of the NAO was then removed from the density field by scaling the regression pattern by the NAO index and removing the pattern from the density at each time point. Removing the instantaneous NAO-related signal weakens the density/dynamic height and thus the geostrophic current response (Fig. 8d) calculated in section 4e, suggesting that some of the proposed ocean negative feedback is forced by the atmosphere and not merely an ocean-only process. On annual time scales the magnitude of the current response is reduced by 45%, but on longer, decadal time scales the reduction is less stark (13% reduction; shown in Fig. 8d). This analysis assumes that the instantaneous impact of the NAO is annually independent and can be linearly separated. We discuss below the extent to which the NAO and ocean temperatures/densities can be seen as one-way forcing from atmosphere to ocean and the extent to which this is actually a coupled feedback (i.e., some of the NAO signal is itself forced by the ocean, implied by the spectral peak in the NAO power spectrum in Fig. 4c). However, the reduction in anomalous circulation response when removing the NAO suggests that atmospheric forcing/the NAO may act to reinforce this ocean feedback.
In the northern NA SPG we have previously shown a role for ocean advection in moving heat content anomalies westward via the mean circulation (section 4d). At the same time, surface heat fluxes were also shown to be nonnegligible. In Fig. 5 the SST, T500 (discussed in section 4b), SHF, sea surface salinity (SSS), mean sea level pressure (MSLP), and sea ice are regressed at various lags against NA SPG mean SSTs. The SHF is directed into the ocean and has a cooling effect in the eastern SPG at lag = 0 and a warming effect in the western SPG at lag = +2 (i.e., it is effectively moving heat content anomalies from east to west). This is likely related to the concomitant strongly negative NAO anomaly in the MSLP field at the same lags. The actual magnitude of the SHF contribution to the Irminger Current ocean heat content (OHC) change is similar to the contribution from advective fluxes (Table 1), but, as noted in section 4d, both are individually quite poorly correlated with the OHC change on multiannual time scales. This is consistent with a mechanism whereby the ocean integrates the interannually independent forcing from the atmosphere/NAO, resulting in decadal time-scale variability in ocean heat content. However, the spectral peak in the NAO index (Fig. 4c) also implies some ocean-to-atmosphere forcing.
Additionally, in the eastern SPG, the SSTs are anticorrelated with the NAO index, seen both at the lag = 0 regression and with the opposite phase at lag = −6. These SSTs are likely a combination of the direct forcing of both 1) the NAO via SHFs and anomalous Ekman and gyre circulation (Häkkinen and Rhines 2004; Sarafanov et al. 2008) and 2) the advective heat flux associated with the diagnosed mechanism of decadal variability. As noted previously, the simulated NAO shows a spectral peak at 17 yr, similar to ocean indices within the NA SPG. It would appear most likely that this atmospheric memory must come from the ocean, but unfortunately, sufficiently long-term atmosphere-only experiments with this model are not available to further test this hypothesis.
At lag = 0 (and with opposite phase at lag = −6), the anomalous NAO-related SHFs show the same sign change over both the Labrador Sea and Gulf Stream/NAC, but over the Gulf Stream/NAC they are of the wrong sign to explain the heat content changes (both at the surface and throughout at least the top 500 m of the water column). This is consistent with advective heat fluxes playing a much more dominant role in the heat budget of the NAC region (see section 4c) than the Irminger Current/Labrador Sea region (section 4d). However, as noted at the beginning of this section, in the NAC region there is a significant portion of the ocean geostrophic circulation (and associated heat transport) response that is itself related to the NAO (cf. Figs. 8c,d). In short, it is impossible to completely separate the effects of either the atmosphere or the ocean.
SSS evolves similarly to SST in the NA SPG, although the largest changes are associated with movement of the ice edge in the GIN Seas (first, fourth, and sixth columns in Fig. 5). In general in the NA SPG, positive salinity anomalies covary with positive temperature anomalies in both space and time, again suggesting a role for advective fluxes. NAO-related surface freshwater fluxes are also proposed to be of only secondary importance because of the fact that simulated SSS anomaly magnitudes are independent of the amplitude of the NAO (not shown).
Similarly to other large-scale variables within the NA SPG, ice edge changes exhibit decadal variability with a spectral peak at a period of 17 yr (not shown). However, unlike in similar work with the IPSL model (Escudier et al. 2013), negative sea ice anomalies do not appear to lead cooling in the NA SPG (Fig. 5, sixth column). We suggest that in our simulations ice edge changes are primarily a passive response to the temperature-dominated decadal variability within the NA SPG, perhaps again via the NAO (Deser et al. 2000), rather than a direct driver of this variability.
g. Summary of the proposed mechanism
The mechanism of decadal (17 yr) variability simulated in the NA SPG T500 and SSTs is summarized in Fig. 9. Positive circulation anomalies in the southern part of the SPG move heat eastward and northward into the eastern SPG with a time scale of around 5 yr (orange). These heat content anomalies are then transported by the mean circulation around the northern edge of the SPG with a time scale of around 2 yr (red). In the Labrador Sea these anomalies affect the stability of the water column. These negative density anomalies, associated with reduced deep-water formation, spill out from the Labrador Sea into the SPG, deepening as they go. In the region north of the Gulf Stream, these negative density anomalies affect the north–south density gradient and induce geostrophic circulation anomalies weakening the NAC. The weaker circulation reduces ocean heat transport and acts to cool the NA SPG (blue). The phase of the oscillation is thus reversed.
A schematic of the proposed mechanism. The various processes in different regions, the time scales, and the postulated role of the atmosphere are as described in the text. Dashed gray lines denote the approximate location of NA SPG and subtropical gyres, with bathymetry of particular interest marked brown. Also marked are regions where OHT correlates with circulation anomalies (orange) or is dominated by temperature anomalies (red) and where the negative feedback is suggested to occur (blue). Additionally, black dashed lines denote regions where the atmosphere is postulated to play a role in the forcing or feedback of ocean anomalies.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0106.1
The postulated role of the atmosphere is also noted (black dashed lines in Fig. 9). As temperature anomalies build up in the eastern SPG, the atmosphere acts to strengthen these anomalies. When the east of the NA SPG is anomalously warm or cold, SHFs also act to move the ocean heat content anomaly westward. Last, in the region of the Labrador Sea/Gulf Stream temperature (density) dipole, the NAO is associated with around 13% of the ocean-circulation feedback (calculated in section 4e and shown in Fig. 8d).
We now discuss the implications of our work and similarities between it and previous studies.
5. Discussion
In the context of the brief literature summary in section 1 and the schematic illustration presented in Fig. 1, our simulations broadly fall into a temperature-dominated regime in the Labrador Sea in which the mechanism could be described as Ocean* (where the asterisk implies that a positive feedback between the NAO and SSTs may be amplifying the mode; see Fig. 1). The time scale is set in part by mean circulation speeds in the northern SPG but with a transition to anomalous circulation in the southern SPG—although it is not clear from the simulations precisely where this transition occurs.
The simulated time scales between changes in the Labrador Sea, NAC, and eastern SPG have been attributed to advective processes. However, confounding this are wave processes, which are also weakly detectable within the model. Analysis of the deep density field (1500–3000 m; not shown) reveals signals characteristic of boundary waves propagating from the Labrador Sea to the equator, along the equator to the eastern boundary, and subsequently north and south along the eastern boundary, all the while radiating Rossby waves westward. The evolution is very similar to that found in the idealized model of Johnson and Marshall (2002) and yields a lag between the Labrador Sea and eastern SPG of 5 yr, broadly similar to that due to advection. Although detectable, these wave signals require heavy filtering of the deep density field while the proposed mechanism exists mainly in the top 1 km. We can conclude only that wave processes may play an additional role in our simulated variability, but the magnitude of this is unclear. We also note that the relatively diffuse thermocline in HG3 (Megann et al. 2014) may act to dampen these wave processes (Grötzner et al. 1998) as compared to the updated seasonal forecast model, GloSea5 (which will be similar to the new Met Office decadal prediction model).
Despite the lagged regression analysis used in this study, and its ubiquity within studies of decadal variability within climate models, there are some hints from the present work that the proposed mechanism may be asymmetric. This asymmetry manifests in the time scales of various phases of the cycle being also dependent on the sign of the anomaly; that is, the same processes are at work in opposite phases of the mechanism but may evolve with different time scales. Some evidence for this can be seen in Fig. 5 in which all the fields reverse sign over 6–8 yr, implying a periodicity of 12–16 yr, and yet the spectral peak occurs at the upper end of this at 16–17 yr. If we construct lagged composites of the T500 (or SST) field based on the top/bottom 10% of phases of the SST index (not shown), we find a reversal time scale of 10 yr following a high SST phase but a reversal time of 6 yr following a low SST phase. This asymmetrical time scale does not appear to be directly due to the effect of heat transport by the anomalous circulation in the southern SPG (υ′T0; see section 4c), as the lags between the NAC and eastern SPG are the same time scale in both phases. It is important to note, however, that constructing composites, which use only 20% of the total data, reduces the number of degrees of freedom.
This asymmetry also appears evident in the coupled simulation when compositing MSLP based on high/low phases of the SST index. Although both MSLP patterns, composited against positive SSTs (Fig. 10a) and negative SSTs (Fig. 10b), show significant MSLP anomalies, the magnitude and precise structure are clearly different, with only the negative SST composite associated with the canonical NAO pattern (Fig. 10b). Additionally, atmosphere-only sensitivity experiments (not shown) suggest a stronger coupling in the NA SPG between anomalously positive NAO/negative SSTs than anomalously negative NAO/positive SSTs. We plan to investigate the asymmetry further in a separate study.
MSLP composites created using area-averaged eastern NA SPG SSTs (50°–65°N, 10°–30°W) to highlight the asymmetry between positive and negative phases of the proposed mechanism. (a) Composite created using the highest 10% of SST anomalies. (b) As in (a), but for the lowest 10% of SST anomalies. Only data significant at the 99% level are shown. Significance is estimated using a moving blocks bootstrap approach (Wilks 1997) reconstructing the composites 10 000 times by resampling the data with a block length estimated from the autocorrelation in each original composite.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0106.1
Comparison with observations and other models
It is difficult to prove that the mode of variability reported here is inconsistent with observational data because of the paucity of observational records in the NA SPG, particularly in the northern half, and the presence of confounding additional transient forcings in the observational record. However, paleo-proxies from the NA SPG suggest that there is 20-yr variability in some indices in the region (Sicre et al. 2008; Chylek et al. 2012), although it must be noted that there is disagreement on the spectral characteristics of all proxies (Mann et al. 1995). The specific elements of our proposed mechanism (anomalous circulation OHT in the southern part of the NA SPG, mean circulation OHT in the northern part, and a negative feedback between Labrador Sea and NAC temperatures) are also broadly consistent with the observational literature. For example, there are some similarities to the anticorrelated relationship between Labrador Sea and NAC temperatures/transports seen in observations (Curry and McCartney 2001). This observational work also highlights the significant role of the NAO in this relationship as well as the dominant role for temperature (as opposed to salinity) in driving these changes. We note that as a result of the northern NA SPG warm bias in HG3 there is less ice in the mean, which may detrimentally affect the ability of ice/freshwater fluxes to affect the decadal variability. In models where the NA SPG mean-state bias is cold, feedbacks involving ice and freshwater fluxes have been shown to be crucial to the diagnosed decadal variability (Escudier et al. 2013).
Although it is difficult to isolate the precise mechanisms by which increased ocean or atmosphere resolution may have altered our results —without a parallel set of low-resolution simulations within the same model framework—there are specific features of the decadal variability that are likely to be affected by enhanced resolution. For example, our proposed mechanism of NA SPG decadal variability suggests a prominent role for boundary currents, which may be improved by higher resolution (Grötzner et al. 1998; Gelderloos et al. 2011). Additionally, the increased atmospheric resolution (which represents the main computational burden for the coupled model) may affect the innate atmospheric variability over the North Atlantic (Matsueda et al. 2009), while the role of the atmosphere may also be modulated by the improved ocean resolution (Scaife et al. 2011). Recent work comparing 1°-, 0.25°-, and 
Similar to our findings, recent ultrahigh-resolution (
The mechanism we have presented has a time scale of 17 yr, similar to the 20-yr time scale found in the IPSL-CM5A-LR model recently investigated by Escudier et al. (2013). However, a similar time scale does not imply the same mechanism; see for example an identical 17-yr time scale but different mechanism reported by Born and Mignot (2012). The present study reports a mode of variability where temperature dominates the density budget, whereas Escudier et al. (2013) report a mode in which freshwater/salinity fluxes have an important role. Indeed, salinity advection within the SPG has been proposed as a cause of bistability in the SPG (Born et al. 2013), albeit on longer time scales. It is intuitive that whether the density budget is dominated by temperature or salinity would affect whether a strengthening northward circulation acted as a positive or negative feedback—but why are NA SPG density changes differently controlled in the two models?
One hypothesis is that the nature of the biases (compared to observations) affects the variability as the nonlinear equation of state for density becomes increasingly salinity dominated at cooler temperatures. To estimate this effect we compute the density change in the Irminger Current region, mechanistically important in both studies, for a one-standard-deviation change in temperature or salinity (while keeping the other at climatological values) in both HG3 and the IPSL-CM5A-LR model as well as an observational estimate from EN4 (Table 4). In HG3, such a temperature change has double the impact on density than a change in salinity. This is not the case in the IPSL model, where salinity changes are found to be more important (the relative magnitudes are unchanged if we remove the spinup in HG3; not shown). The EN4 data suggest that the real world may be in a temperature-dominated regime, similar to HG3. This points to there being some relationship between the NA SPG mean-state biases of a given model and the subsequently diagnosed mechanisms of decadal variability. Note that this cursory analysis merely compares mean states and variability and does not explicitly investigate whether density variability is controlled by temperature or salinity. Nevertheless, one implication of this would be that decadal prediction studies using anomaly-assimilation methods, in which the mean-state biases are implicitly assumed to be independent of the variability, would need to reevaluate the validity of this assumption (Robson 2010). We plan to investigate this relationship in more detail in a forthcoming study (Menary et al. 2015).
Characteristic magnitudes of density changes (kg m−3) in different simulated/estimated temperature or salinity (T/S) regimes. Mean states are volume-averaged temperature and salinity (in the models defined as the observed mean plus a model bias; e.g., EN4 + HG3 bias) in the Irminger Current (58°–60°N, 43°–45°W, top 500 m). The density changes are calculated by estimating the decadal standard deviation (sd) in temperature or salinity (by bandpass filtering the data to allow only periods in the range 10–30 yr) and recalculating the densities with these T/S perturbations added. As there is limited raw data from EN4 to reliably estimate decadal variability in the Irminger Current, and to simplify the experimental design and interpretation, we use HG3 estimates of the decadal variability in temperature and salinity in all cases.
6. Conclusions
We have analyzed a decadal mode of variability in the near surface (top 500 m) of the North Atlantic subpolar gyre (NA SPG) in a 460-yr control simulation with a version of the high-resolution coupled climate model HadGEM3 (HG3).
The mode of variability involves the propagation of heat content anomalies around the NA SPG with a periodicity of around 17 yr.
Simulated decadal variability (between 10 and 30 yr) in the NA SPG explains more than 15% of the annual mean variance in top-500-m depth-averaged temperatures. This rises to >30% of the variance within the interior NA SPG and Labrador Sea.
The simulated NA SPG heat budget is dominated by advective, rather than surface, heat fluxes on decadal time scales, with advection from the subtropics playing the primary role. For the specific regions of interest, namely the Irminger Current and North Atlantic Current (NAC), advective fluxes were also found to dominate. The large depth extent of the mode is also consistent with an important role for advection (Saravanan and McWilliams 1998).
The role of mean or anomalous circulation in transporting heat content anomalies was found to vary with region. Anomalous circulation dominated the variability in the NAC, whereas mean circulation, and hence temperature anomalies, dominated in the Irminger Current region.
A negative feedback, required for the mechanism to result in a spectral peak, occurs between the Labrador Sea and NAC. Here, density anomalies spill out of the Labrador Sea resulting in a dynamic height gradient across the NAC/Labrador Sea that induces vertical shear in the geostrophic currents. These current anomalies result in heat transport anomalies that reverse the cycle. The density changes are driven by temperature rather than salinity.
Variability in the NAO directly contributes to various stages of the mechanism as well as showing signs of responding to ocean variability. Removing the North Atlantic Oscillation (NAO) signal from the negative feedback between Labrador Sea and NAC temperatures/densities (see section 4f) shows that about 45% of the geostrophic current speed feedback is related to the NAO on annual time scales but that on decadal time scales the ocean feedback still dominates. The atmosphere also acts to reinforce temperature anomalies in the eastern NA SPG and aid their westward propagation in the northern SPG. The proposed mechanism is summarized in Fig. 9.
Whether density changes are temperature or salinity controlled affects where, and how, negative feedbacks can occur. This may also be expected to affect the particular mechanism simulated in the model. This could have important implications for decadal prediction studies that use the method of anomaly assimilation and prediction, in which the future evolution of the model is assumed to be independent of the mean state—an assumption that we suggest may not be valid.
A modified version of the model presented here will be used as part of the Met Office decadal prediction system, and analyses such as that we have presented will be important in developing and evaluating such systems. Given the relationship between resolution and the improved realization of particular processes, as well as mean-state biases, further high-resolution coupled model studies would be valuable in testing whether these results are robust.
Acknowledgments
Matthew Menary and Richard Wood were supported by the Joint DECC and Defra Hadley Centre Climate Programme, DECC/Defra (GA01101). Daniel Hodson and Rowan Sutton were supported by NERC through the National Centre for Atmospheric Science (NCAS). Jon Robson was supported by the Projet Prévisibilit Climatique Decennale (PRECLIDE) and the Seasonal-to-Decadal Climate Prediction for the Improvement of European Climate Services (SPECS) project GA 308378. Data from the RAPID–WATCH MOC monitoring project are funded by the Natural Environment Research Council and are freely available at www.rapid.ac.uk/rapidmoc. The authors thank Matthew Mizielinski for initially setting up the simulations. For access to, and time on, the MONSooN supercomputer we acknowledge support from the Met Office and the Natural Environment Research Council.
APPENDIX A
Heat Budget
The basinwide, full-depth NA SPG heat budget is shown in Fig. A1 for the latitude range 53°–73°N. Because of the lack of availability of the correct ocean diagnostics at high enough output frequency (precluded by the expense of storing high-resolution atmosphere and ocean data), the heat budget of the NA SPG does not close perfectly (cf. red and black lines in Fig. A1a). However, the error is negligible—less than 1% of the net surface fluxes of the region. Sensitivity tests where all output diagnostics were computed online and stored revealed that horizontal diffusion was the most important missing heat flux. The heat budget of the NEMO ocean model is further complicated by the use of a linear free surface, with variable volume that sits on top of the fixed-volume ocean grid cells and a heat flux between the two. For further details on the precise formulation of the heat budget within the NEMO ocean model, see Madec et al. (2008).
The full-depth heat budget of the NA SPG (53°–73°N) volume plotted using 9-yr running means for clarity and to highlight decadal variability. Positive is into the specified region. (a) Individual components of the heat budget as denoted in the legend. (b) The anomalous heat budget (referenced against years 22–42) to highlight the trends in latent heat fluxes and advective heat fluxes through the southern boundary. Note that the heat content change (dT/dt; black) and sum of heat fluxes (red) do not match prior to the year 100 as instantaneous ocean temperatures (used to calculate dT/dt) were stored with intermittent frequency during this period.
Citation: Journal of Climate 28, 19; 10.1175/JCLI-D-15-0106.1
This reveals that OHTS dominates the variability in advective heat fluxes. Using annual data, the standard deviation of OHTS is 28 PW, compared to 17 PW for OHTN. The variability in OHTS is split between vertical AMOC and horizontal gyre heat transport variability at these latitudes (annual correlation between OHTS and OHTAMOC is 0.74 and between OHTS and OHTgyre is 0.88). The surface fluxes (directed into the ocean) are dominated by shortwave (solar) heating of the NA SPG, whereas longwave, latent, and sensible heat fluxes represent net heat loss from the NA SPG.
To investigate the relative magnitudes of their variability, and any trends, the mean of each heat flux over the years 22–42 is removed (Fig. A1b). Rather than remove the full time mean, removing the mean from just the period soon after the model was initialized serves to additionally show how the heat fluxes diverge. Net advective heat fluxes into the region are increasing throughout the period, balanced largely by increasing surface heat flux loss but with some residual heating of the NA SPG. The advective heat flux trend is dominated by the increase in heat flux from the south, which is due to the strengthening AMOC (Fig. 3c), with much of this heat lost via latent heat loss as well as longwave emission. The rate of net warming is highest in the first century, which is also why the net heat flux appears to be below zero for the remainder of the time (i.e., the net warming rate is slower in the subsequent years).
APPENDIX B
Heat Budget Breakdown
REFERENCES
Álvarez-García, F., M. Latif, and A. Biastoch, 2008: On multidecadal and quasi-decadal North Atlantic variability. J. Climate, 21, 3433–3452, doi:10.1175/2007JCLI1800.1.
Battisti, D., U. Bhatt, and M. Alexander, 1995: A modeling study of the interannual variability in the wintertime North Atlantic Ocean. J. Climate, 8, 3067–3083, doi:10.1175/1520-0442(1995)008<3067:AMSOTI>2.0.CO;2.
Biastoch, A., C. W. Boening, and J. R. E. Lutjeharms, 2008b: Agulhas leakage dynamics affects decadal variability in Atlantic overturning circulation. Nature, 456, 489–492, doi:10.1038/nature07426.
Born, A., and J. Mignot, 2012: Dynamics of decadal variability in the Atlantic subpolar gyre: A stochastically forced oscillator. Climate Dyn., 39, 461–474, doi:10.1007/s00382-011-1180-4.
Born, A., T. F. Stocker, C. C. Raible, and A. Levermann, 2013: Is the Atlantic subpolar gyre bistable in comprehensive coupled climate models? Climate Dyn., 40, 2993–3007, doi:10.1007/s00382-012-1525-7.
Brayshaw, D. J., B. Hoskins, and M. Blackburn, 2008: The storm-track response to idealized SST perturbations in an aquaplanet GCM. J. Atmos. Sci., 65, 2842–2860, doi:10.1175/2008JAS2657.1.
Burkholder, K. C., and M. S. Lozier, 2011: Mid-depth Lagrangian pathways in the North Atlantic and their impact on the salinity of the eastern subpolar gyre. Deep-Sea Res. I, 58, 1196–1204, doi:10.1016/j.dsr.2011.08.007.
Burkholder, K. C., and M. S. Lozier, 2014: Tracing the pathways of the upper limb of the North Atlantic meridional overturning circulation. Geophys. Res. Lett., 41, 4254–4260, doi:10.1002/2014GL060226.
Chylek, P., C. Folland, L. Frankcombe, H. Dijkstra, G. Lesins, and M. Dubey, 2012: Greenland ice core evidence for spatial and temporal variability of the Atlantic multidecadal oscillation. Geophys. Res. Lett., 39, L09705, doi:10.1029/2012GL051241.
Curry, R., and M. McCartney, 2001: Ocean gyre circulation changes associated with the North Atlantic Oscillation. J. Phys. Oceanogr., 31, 3374–3400, doi:10.1175/1520-0485(2001)031<3374:OGCCAW>2.0.CO;2.
Dai, A., A. Hu, G. Meehl, W. Washington, and W. Strand, 2005: Atlantic thermohaline circulation in a coupled general circulation model: Unforced variations versus forced changes. J. Climate, 18, 3270–3293, doi:10.1175/JCLI3481.1.
Danabasoglu, G., 2008: On multidecadal variability of the Atlantic meridional overturning circulation in the Community Climate System Model version 3. J. Climate, 21, 5524–5544, doi:10.1175/2008JCLI2019.1.
Delworth, T., S. Manabe, and R. Stouffer, 1993: Interdecadal variations of the thermohaline circulation in a coupled ocean–atmosphere model. J. Climate, 6, 1993–2011, doi:10.1175/1520-0442(1993)006<1993:IVOTTC>2.0.CO;2.
Deser, C., J. E. Walsh, and M. S. Timlin, 2000: Arctic sea ice variability in the context of recent atmospheric circulation trends. J. Climate, 13, 617–633, doi:10.1175/1520-0442(2000)013<0617:ASIVIT>2.0.CO;2.
Deshayes, J., R. Curry, and R. Msadek, 2014: CMIP5 model intercomparison of freshwater budget and circulation in the North Atlantic. J. Climate, 27, 3298–3317, doi:10.1175/JCLI-D-12-00700.1.
Dong, B., and R. Sutton, 2001: The dominant mechanisms of variability in Atlantic Ocean heat transport in a coupled ocean-atmosphere GCM. Geophys. Res. Lett., 28, 2445–2448, doi:10.1029/2000GL012531.
Dong, B., and R. Sutton, 2005: Mechanism of interdecadal thermohaline circulation variability in a coupled ocean–atmosphere GCM. J. Climate, 18, 1117–1135, doi:10.1175/JCLI3328.1.
Dunstone, N., D. Smith, and R. Eade, 2011: Multi-year predictability of the tropical Atlantic atmosphere driven by the high latitude North Atlantic Ocean. Geophys. Res. Lett., 38, L14701, doi:10.1029/2011GL047949.
Eden, C., and J. Willebrand, 2001: Mechanism of interannual to decadal variability of the North Atlantic circulation. J. Climate, 14, 2266–2280, doi:10.1175/1520-0442(2001)014<2266:MOITDV>2.0.CO;2.
Eden, C., and R. J. Greatbatch, 2003: A damped decadal oscillation in the North Atlantic climate system. J. Climate, 16, 4043–4060, doi:10.1175/1520-0442(2003)016<4043:ADDOIT>2.0.CO;2.
Escudier, R., J. Mignot, and D. Swingedouw, 2013: A 20-year coupled ocean-sea ice-atmosphere variability mode in the North Atlantic in an AOGCM. Climate Dyn., 40, 619–636, doi:10.1007/s00382-012-1402-4.
Février, S., J. Sirven, and C. Herbaut, 2007: Interaction of a coastal Kelvin wave with the mean state in the Gulf Stream separation area. J. Phys. Oceanogr., 37, 1429–1444, doi:10.1175/JPO3062.1.
Folland, C., T. Palmer, and D. Parker, 1986: Sahel rainfall and worldwide sea temperatures, 1901–85. Nature, 320, 602–607, doi:10.1038/320602a0.
Frankcombe, L. M., A. von der Heydt, and H. A. Dijkstra, 2010: North Atlantic multidecadal climate variability: An investigation of dominant time scales and processes. J. Climate, 23, 3626–3638, doi:10.1175/2010JCLI3471.1.
Frankignoul, C., P. Müller, and E. Zorita, 1997: A simple model of the decadal response of the ocean to stochastic wind forcing. J. Phys. Oceanogr., 27, 1533–1546, doi:10.1175/1520-0485(1997)027<1533:ASMOTD>2.0.CO;2.
Gelderloos, R., C. A. Katsman, and S. S. Drijfhout, 2011: Assessing the roles of three eddy types in restratifying the Labrador Sea after deep convection. J. Phys. Oceanogr., 41, 2102–2119, doi:10.1175/JPO-D-11-054.1.
Goldenberg, S. B., C. W. Landsea, A. M. Mestas-Nuñez, and W. M. Gray, 2001: The recent increase in Atlantic hurricane activity: Causes and implications. Science, 293, 474–479, doi:10.1126/science.1060040.
Good, S. A., M. J. Martin, and N. A. Rayner, 2013: EN4: Quality controlled ocean temperature and salinity profiles and monthly objective analyses with uncertainty estimates. J. Geophys. Res. Oceans, 118, 6704–6716, doi:10.1002/2013JC009067.
Grötzner, A., M. Latif, and T. Barnett, 1998: A decadal climate cycle in the North Atlantic Ocean as simulated by the ECHO coupled GCM. J. Climate, 11, 831–847, doi:10.1175/1520-0442(1998)011<0831:ADCCIT>2.0.CO;2.
Häkkinen, S., and P. B. Rhines, 2004: Decline of subpolar North Atlantic circulation during the 1990s. Science, 304, 555–559, doi:10.1126/science.1094917.
Hodson, D. L., and R. T. Sutton, 2012: The impact of resolution on the adjustment and decadal variability of the Atlantic meridional overturning circulation in a coupled climate model. Climate Dyn., 39, 3057–3073, doi:10.1007/s00382-012-1309-0.
Holland, M., C. Bitz, M. Eby, and A. Weaver, 2001: The role of ice–ocean interactions in the variability of the North Atlantic thermohaline circulation. J. Climate, 14, 656–675, doi:10.1175/1520-0442(2001)014<0656:TROIOI>2.0.CO;2.
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, doi:10.1016/j.jmarsys.2005.11.019.
Johnson, H., and D. Marshall, 2002: Localization of abrupt change in the North Atlantic thermohaline circulation. Geophys. Res. Lett., 29, 1083, doi:10.1029/2001GL014140.
Jungclaus, J., H. Haak, M. Latif, and U. Mikolajewicz, 2005: Arctic–North Atlantic interactions and multidecadal variability of the meridional overturning circulation. J. Climate, 18, 4013–4031, doi:10.1175/JCLI3462.1.
Kara, A., P. Rochford, and H. Hurlburt, 2000: An optimal definition for ocean mixed layer depth. J. Geophys. Res., 105, 16 803–16 821, doi:10.1029/2000JC900072.
Kleppin, H., M. Jochum, B. Otto-Bliesner, C. A. Shields, and S. Yeager, 2015: Stochastic atmospheric forcing as a cause of sudden Greenland climate transitions. J. Climate, doi:10.1175/JCLI-D-14-00728.1, in press.
Knight, J., R. Allan, C. Folland, M. Vellinga, and M. Mann, 2005: A signature of persistent natural thermohaline circulation cycles in observed climate. Geophys. Res. Lett., 32, L20708, doi:10.1029/2005GL024233.
Kwon, Y.-O., and C. Frankignoul, 2014: Mechanisms of multidecadal Atlantic meridional overturning circulation variability diagnosed in depth versus density space. J. Climate, 27, 9359–9376, doi:10.1175/JCLI-D-14-00228.1.
Kwon, Y.-O., M. A. Alexander, N. A. Bond, C. Frankignoul, H. Nakamura, B. Qiu, and L. A. Thompson, 2010: Role of the Gulf Stream and Kuroshio–Oyashio systems in large-scale atmosphere–ocean interaction: A review. J. Climate, 23, 3249–3281, doi:10.1175/2010JCLI3343.1.
MacLachlan, C., and Coauthors, 2014: Global seasonal forecast system version 5 (GloSea5): A high-resolution seasonal forecast system. Quart. J. Roy. Meteor. Soc, 141, 1072–1084, doi:10.1002/qj.2396.
MacMartin, D. G., E. Tziperman, and L. Zanna, 2013: Frequency domain multimodel analysis of the response of Atlantic meridional overturning circulation to surface forcing. J. Climate, 26, 8323–8340, doi:10.1175/JCLI-D-12-00717.1.
Madec, G., and Coauthors, 2008: NEMO ocean engine. IPSL Tech. Rep. 27, 357 pp. [Available online at http://www.nemo-ocean.eu/About-NEMO/Reference-manuals.]
Mann, M., J. Park, and R. Bradley, 1995: Global interdecadal and century-scale climate oscillations during the past five centuries. Nature, 378, 266–270, doi:10.1038/378266a0.
Marzocchi, A., J. J.-M. Hirschi, N. P. Holliday, S. A. Cunningham, A. T. Blaker, and A. C. Coward, 2015: The North Atlantic subpolar circulation in an eddy-resolving global ocean model. J. Mar. Syst., 142, 126–143, doi:10.1016/j.jmarsys.2014.10.007.
Matsueda, M., R. Mizuta, and S. Kusunoki, 2009: Future change in wintertime atmospheric blocking simulated using a 20-km-mesh atmospheric global circulation model. J. Geophys. Res., 114, D12114, doi:10.1029/2009JD011919.
Megann, A., and Coauthors, 2014: GO5.0: The joint NERC–Met Office NEMO global ocean model for use in coupled and forced applications. Geosci. Model Dev., 7, 1069–1092, doi:10.5194/gmd-7-1069-2014.
Menary, M. B., W. Park, K. Lohmann, M. Vellinga, M. D. Palmer, M. Latif, and J. H. Jungclaus, 2012: A multimodel comparison of centennial Atlantic meridional overturning circulation variability. Climate Dyn., 38, 2377–2388, doi:10.1007/s00382-011-1172-4.
Menary, M. B., D. L. R. Hodson, J. I. Robson, R. T. Sutton, R. A. Wood, and J. A. Hunt, 2015: Exploring the impact of CMIP5 model biases on the simulation of North Atlantic decadal variability. Geophys. Res. Lett., 42, 5926–5934, doi:10.1002/2015GL064360.
Nonaka, M., and S.-P. Xie, 2003: Covariations of sea surface temperature and wind over the Kuroshio and its extension: Evidence for ocean-to-atmosphere feedback. J. Climate, 16, 1404–1413, doi:10.1175/1520-0442(2003)16<1404:COSSTA>2.0.CO;2.
Polo, I., J. Robson, R. Sutton, and M. A. Balmaseda, 2014: The importance of wind and buoyancy forcing for the boundary density variations and the geostrophic component of the AMOC at 26°N. J. Phys. Oceanogr., 44, 2387–2408, doi:10.1175/JPO-D-13-0264.1.
Roberts, C., and Coauthors, 2013: Atmosphere drives recent interannual variability of the Atlantic meridional overturning circulation at 26.5°N. Geophys. Res. Lett., 40, 5164–5170, doi:10.1002/grl.50930.
Robson, J. I., 2010: Understanding the performance of a decadal prediction system. Ph.D. thesis, University of Reading, 233 pp.
Rodwell, M., D. Rowell, and C. Folland, 1999: Oceanic forcing of the wintertime North Atlantic Oscillation and European climate. Nature, 398, 320–323, doi:10.1038/18648.
Roussenov, V. M., R. G. Williams, C. W. Hughes, and R. J. Bingham, 2008: Boundary wave communication of bottom pressure and overturning changes for the North Atlantic. J. Geophys. Res., 113, C08042, doi:10.1029/2007JC004501.
Sarafanov, A., A. Falina, A. Sokov, and A. Demidov, 2008: Intense warming and salinification of intermediate waters of southern origin in the eastern subpolar North Atlantic in the 1990s to mid-2000s. J. Geophys. Res., 113, C12022, doi:10.1029/2008JC004975.
Saravanan, R., and J. C. McWilliams, 1998: Advective ocean–atmosphere interaction: An analytical stochastic model with implications for decadal variability. J. Climate, 11, 165–188, doi:10.1175/1520-0442(1998)011<0165:AOAIAA>2.0.CO;2.
Scaife, A. A., and Coauthors, 2011: Improved Atlantic winter blocking in a climate model. Geophys. Res. Lett., 38, L23703, doi:10.1029/2011GL049573.
Schauer, U., and A. Beszczynska-Möller, 2009: Problems with estimation and interpretation of oceanic heat transport–conceptual remarks for the case of Fram Strait in the Arctic Ocean. Ocean Sci., 5, 487–494, doi:10.5194/os-5-487-2009.
Screen, J. A., 2013: Influence of Arctic sea ice on European summer precipitation. Environ. Res. Lett., 8, 044015, doi:10.1088/1748-9326/8/4/044015.
Sévellec, F., and A. V. Fedorov, 2013: The leading, interdecadal eigenmode of the Atlantic meridional overturning circulation in a realistic ocean model. J. Climate, 26, 2160–2183, doi:10.1175/JCLI-D-11-00023.1.
Sicre, M.-A., and Coauthors, 2008: A 4500-year reconstruction of sea surface temperature variability at decadal time-scales off north Iceland. Quat. Sci. Rev., 27, 2041–2047, doi:10.1016/j.quascirev.2008.08.009.
Smith, D. M., R. Eade, N. J. Dunstone, D. Fereday, J. M. Murphy, H. Pohlmann, and A. A. Scaife, 2010: Skilful multi-year predictions of Atlantic hurricane frequency. Nat. Geosci., 3, 846–849, doi:10.1038/ngeo1004.
Spence, P., O. A. Saenko, W. Sijp, and M. England, 2012: The role of bottom pressure torques on the interior pathways of North Atlantic deep water. J. Phys. Oceanogr., 42, 110–125, doi:10.1175/2011JPO4584.1.
Sutton, R., and D. Hodson, 2005: Atlantic Ocean forcing of North American and European summer climate. Science, 309, 115–118, doi:10.1126/science.1109496.
Timmermann, A., M. Latif, R. Voss, and A. Grötzner, 1998: Northern Hemispheric interdecadal variability: A coupled air–sea mode. J. Climate, 11, 1906–1931, doi:10.1175/1520-0442-11.8.1906.
Treguier, A. M., S. Theetten, E. P. Chassignet, T. Penduff, R. Smith, L. Talley, J. Beismann, and C. Böning, 2005: The North Atlantic subpolar gyre in four high-resolution models. J. Phys. Oceanogr., 35, 757–774, doi:10.1175/JPO2720.1.
Treguier, A. M., and Coauthors, 2014: Meridional transport of salt in the global ocean from an eddy-resolving model. Ocean Sci., 10, 243–255, doi:10.5194/os-10-243-2014.
Tulloch, R., and J. Marshall, 2012: Exploring mechanisms of variability and predictability of Atlantic meridional overturning circulation in two coupled climate models. J. Climate, 25, 4067–4080, doi:10.1175/JCLI-D-11-00460.1.
Vellinga, M., and P. Wu, 2004: Low-latitude freshwater influence on centennial variability of the Atlantic thermohaline circulation. J. Climate, 17, 4498–4511, doi:10.1175/3219.1.
Visbeck, M., H. Cullen, G. Krahmann, and N. Naik, 1998: An ocean model’s response to North Atlantic Oscillation-like wind forcing. Geophys. Res. Lett., 25, 4521–4524, doi:10.1029/1998GL900162.
Volkov, D. L., T. Lee, and L.-L. Fu, 2008: Eddy-induced meridional heat transport in the ocean. Geophys. Res. Lett., 35, L20601, doi:10.1029/2008GL035490.
Walters, D., and Coauthors, 2011: The Met Office unified model global atmosphere 3.0/3.1 and JULES global land 3.0/3.1 configurations. Geosci. Model Dev. Discuss., 4, 1213–1271, doi:10.5194/gmdd-4-1213-2011.
Wang, C., L. Zhang, S.-K. Lee, L. Wu, and C. R. Mechoso, 2014: A global perspective on CMIP5 climate model biases. Nat. Climate Change, 4, 201–205, doi:10.1038/nclimate2118.
Watanabe, M., M. Kimoto, T. Nitta, and M. Kachi, 1999: A comparison of decadal climate oscillations in the North Atlantic detected in observations and a coupled GCM. J. Climate, 12, 2920–2940, doi:10.1175/1520-0442(1999)012<2920:ACODCO>2.0.CO;2.
Wilks, D., 1997: Resampling hypothesis tests for autocorrelated fields. J. Climate, 10, 65–82, doi:10.1175/1520-0442(1997)010<0065:RHTFAF>2.0.CO;2.
Wohlleben, T., and A. Weaver, 1995: Interdecadal climate variability in the subpolar North Atlantic. Climate Dyn., 11, 459–467, doi:10.1007/BF00207195.
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, doi:10.1175/JCLI-D-11-00149.1.
Zhang, R., 2010: Latitudinal dependence of Atlantic meridional overturning circulation (AMOC) variations. Geophys. Res. Lett., 37, L16703, doi:10.1029/2010GL044474.
Zhang, R., and T. L. Delworth, 2006: Impact of Atlantic multidecadal oscillations on India/Sahel rainfall and Atlantic hurricanes. Geophys. Res. Lett., 33, L17712, doi:10.1029/2006GL026267.
Zhang, R., and G. K. Vallis, 2007: The role of bottom vortex stretching on the path of the North Atlantic western boundary current and on the northern recirculation gyre. J. Phys. Oceanogr., 37, 2053–2080, doi:10.1175/JPO3102.1.
