Eddy-induced variability of the meridional overturning circulation in a model of the North Atlantic


  • M. D. Thomas,

    Corresponding author
    1. School of Environmental Sciences, University of East Anglia, Norwich, UK
    2. Now at Laboratoire de Physique des Océans, CNRS-Ifremer-IRD-UBO, IUEM, Plouzané, France
    • Corresponding author: M. D. Thomas, Laboratoire de Physique des Océans, UMR 5623, CNRS-Ifremer-IRD-UBO, IUEM, Plouzané, France. (Matthew.Thomas@ifremer.fr)

    Search for more papers by this author
  • X. Zhai

    1. School of Environmental Sciences, University of East Anglia, Norwich, UK
    Search for more papers by this author


[1] Observations of the Atlantic Meridional Overturning Circulation (AMOC) show that it varies on all timescales. Here we isolate the contribution of eddies to AMOC variability using an eddy-permitting model of the North Atlantic driven by climatological and steady forcing. The eddy-induced AMOC variability is found to be ubiquitous and significant at all latitudes, with a magnitude comparable to the seasonal cycle in the subtropics. Furthermore, the eddy-induced AMOC variability is manifested not only at high frequencies but also at interannual and longer timescales. These results imply that a significant fraction of the AMOC variability is inherently unpredictable at seasonal to interannual timescales.

1 Introduction

[2] The Atlantic Meridional Overturning Circulation (AMOC) carries warm surface water to high northern latitudes and returns cold deep water to low latitudes. The approximately 1 PW of northward heat transport associated with the AMOC in the subtropical North Atlantic [Ganachaud and Wunsch, 2003] has important consequences on European climate [Vellinga and Wood, 2002; Wood et al., 2003]. Recent efforts using the RAPID hydrographic array [Kanzow et al., 2007] have revealed that large AMOC fluctuations, of a similar magnitude to its time-mean, are commonplace on weekly timescales at 26.5°N in the Atlantic [Cunningham et al., 2007; McCarthy et al., 2012]. The sources of this variability are not yet fully understood.

[3] Much of what is currently understood on AMOC variability comes from linear theory and coarse ocean models, which depict a linear and deterministic response of overturning to external forcing through wave adjustment or mean advection [e.g., Greatbatch and Peterson, 1996; Huang et al., 2000; Johnson and Marshall, 2002; Deshayes and Frankignoul, 2005]. In such frameworks, the AMOC variability is a superposition of approximately decadal period thermohaline forcing and higher frequency wind-driven variability [e.g., Biastoch et al., 2008]. The high frequencies are thought to be comprised mostly of Ekman transport fluctuations [Jayne and Marotzke, 2001]. Ocean adjustment to forcing, however, can be very different when achieved by nonlinear processes such as eddies [Dewar, 2003].

[4] Westward propagating eddies pervade the ocean and dominate the sea surface height variability [Chelton et al., 2011]. On arriving at the western boundary, the impact of these eddies on the overturning has been postulated to dominate the signal [Wunsch, 2008]. While the energy carried by the eddies has been shown to depreciate as they approach the western boundary [Zhai et al., 2010; Kanzow et al., 2009], the impact of eddies on the overturning remains unknown, although the volume anomalies carried westward by the eddies have to go somewhere.

[5] We investigate the impact of eddies on AMOC variability relative to the size of the seasonally forced AMOC. We extend recent work that has suggested eddies make an important contribution to short timescale AMOC fluctuations [Biastoch et al., 2008; Hirschi et al., 2013] by explicitly isolating the impacts of eddies throughout the North Atlantic in a 1/10° resolution regional model.

2 Model Description

[6] We used a North Atlantic configuration of the Massachusetts Institute of Technology general circulation model [Marshall et al., 1997]. The study domain lies between 14°S and 74°N and between 100°W and 20°E [Zhai and Marshall, 2013]. Horizontal resolution is 1/10°, and there are 33 vertical levels that are unevenly spaced to allow higher surface resolution. Bottom stress is linear with drag coefficient 1.1  ×10−3. Horizontal momentum is mixed according to a scale-selective biharmonic friction operator with Smagorinsky-like viscosity [Griffies and Hallberg, 2000], and tracers are horizontally mixed using a biharmonic operator with background viscosity of 1010 m4 s−1. No ice model is employed in our configuration.

[7] In this study, we have run two experiments. The first experiment, called Control, used climatological monthly mean forcing derived from the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) reanalysis [Kalnay et al., 1996]. Note that the NCEP wind stress does not account for the surface ocean velocities. Temperature and salinity fields at the model northern and southern boundaries and Mediterranean Sea were restored to monthly mean climatological values. The model was first spun up for 23 years at 1/5° resolution and then further run for another 30 years at 1/10° resolution, during which the model output has been saved as snapshots every 2 days. Saving the model output as 2 day averages makes no noticeable differences to our results. Results from the last 15 years are used for this study, the beginning of which is referred to as year 0. In the Control run, there is only a seasonal cycle in the external forcing; so AMOC variability at other frequency bands is generated internally by the model.

[8] In the second experiment, called Steady, the model was instead forced by the annual mean climatological forcing and restored to annual mean climatological temperatures and salinities at the model northern and southern boundaries and Mediterranean Sea. Calculated in this manner, the wind energy input in the two model runs will likely differ. Initialized from the Control run after 9 years of spin up at 1/10° resolution, the model was run for 20 years with the steady forcing, during which the model output was again saved every 2 days. Results from the last 15 years are presented, for which year 0 marks the beginning. In the Steady run, there is no variability in the external forcing at all, and the AMOC variability is due entirely to eddies resolved by the model. The Steady and Control runs can therefore be compared to estimate the relative contribution that eddies make toward the seasonally forced AMOC variability.

[9] The time mean eddy kinetic energy (EKE; math formula, where uand v are the zonal and meridional velocity anomalies respectively and the overbar represents a time average) and overturning stream function for the Control and Steady experiments are shown in Figure 1. The time mean maximum AMOC stream function at 26.6°N in Control is approximately 18 Sverdrups (Sv) (1 Sv = 106 m3 s−1), close to the estimation made from the RAPID array [McCarthy et al., 2012]. However, the model underestimates the range of the observed AMOC variability at 26.6°N by roughly half due to the lack of forcing variability at nonseasonal timescales [McCarthy et al., 2012]. The time mean overturning strength in Steady is weaker than in Control by approximately 6 Sv throughout the basin. The weakened AMOC in the Steady run is most likely due to the lack of strong wintertime convection though differences in wind energy input in the model runs may also have some impact. The basin average EKE strength (excluding the equatorial band) is similar in the two model runs, with about a 15% stronger EKE in Control. Local changes, however, can be large. Most notably, EKE patterns around the Gulf Stream and North Atlantic Current are generally broader and shifted to the east in Steady. However, the general similarity of the eddy field between Control and Steady lends credit to inferences made from Steady on the eddy-induced AMOC variability.

Figure 1.

(a, b) Eddy kinetic energy in units of log10(cm2 s−2) and (c, d) Atlantic meridional overturning stream function in units of Sv for the (a, c) Control and (b, d) Steady simulations.

3 Results

[10] Here we compare the output from the Control and Steady simulations to estimate the importance of eddy-induced overturning variability relative to seasonally forced variability. Time series of AMOC anomaly (defined here as the stream function at 1000 m depth minus its time-mean) are generated at each latitude and displayed as latitude-versus-time plots (Figure 2). The nature of AMOC variability in the Control run is very latitude dependent, with the largest variability seen in tropical latitudes and at approximately 35°N where the zonal extension of the Gulf Stream lies. This distribution of AMOC variability is broadly consistent with that reported from other eddy-permitting models [e.g., Hirschi et al., 2007]. To gain a better understanding of the temporal scales of AMOC variability, we have split the output into relative high (AMOChp) and low frequencies (AMOClp) by spectrally filtering with a threshold period of 170 days. At this threshold, the annual and semiannual harmonics of the seasonal variability are considered to be part of the low frequency variability (not shown). As shown by Chelton et al. [2011], Gaussian-shaped eddy signatures are not isolated in spectral space but are spread; so we consider the filter to only provide a first-order representation of the temporal scales of the data. However, the filter successfully separates out high frequency variability from a low frequency component that is dominated by the seasonal signal.

Figure 2.

Latitude-versus-time plots of the AMOC anomaly for the (a, e) unfiltered, (b, f) high pass filtered (AMOChp) and (c, g) low pass filtered (AMOClp) components of the (a–c) Control and (e–g) Steady simulations, and of the (d) zonal mean meridional Ekman transport (VEk) and (h) Control AMOClp minus VEk (subplot c minus subplot d). Units are Sv.

[11] A distinct seasonal signal exists in the tropical and subpolar latitudes of the Control run (Figure 2c). Between the latitude range of 20°N–30°N, however, the Control run seasonal signal is not strong. This distribution of the seasonal variability can be explained by the strength of the zonal mean meridional Ekman transport (VEk; Figure 2d), which displays a pronounced seasonal cycle at low and high latitudes but only a small signal in the subtropical latitude range of 20°N–30°N. On top of the seasonal signal is a high frequency component of variability (Figure 2b), which is spectrally broad band at all latitudes (not shown). As with the low frequency component, the largest high frequency activity is in the tropics and in the Gulf Stream region of approximately 35°N (Figure 2).

[12] In the Steady run, the unfiltered AMOC variability displays no regular seasonal signal and is in close consistency with the high pass filtered component of the Control run. Most of this variability is contained in the high frequencies, as can be seen from the filtered components (Figures 2f and 2g), though a non-negligible component of eddy-induced variability is manifested at seasonal and longer periods, especially in the Gulf Stream region and to the south of it. The coherent pattern to the south of the Gulf Stream at longer periods (Figure 2g) suggests that the Gulf Stream eddies may act as an important source for the low frequency AMOC variability at latitudes further to the south through, e.g., boundary wave propagation. It is, however, difficult to discern a direction of propagation from the plot. To investigate whether eddy-induced variability at seasonal and longer periods are similarly manifested in the Control run, we have subtracted the zonal mean meridional Ekman transport from the low pass filtered Control AMOC anomaly, thereby revealing low frequency variability that is otherwise masked by the stronger Ekman-induced seasonal signals (Control AMOClp-VEk; Figure 2h). As with Steady, the eddy-induced variability at these frequencies is coherent to the south of the Gulf Stream. The (non-Ekman) low frequency variability in Control (Figure 2h) is stronger than in Steady (Figure 2g), either because of the different distributions of EKE in the Control and Steady runs (Figures 1a and 1b) or because the seasonal signal might excite variability at interannual periods. The sharp stripes in Figure 2h derive from strong monthly changes in wind strength that are likely smoothed out by internal ocean variability.

[13] Our model results suggest that the contribution of eddies to AMOC variability is less significant at high latitudes, a result that may have implications for any future subpolar measurements of the AMOC. This result contrasts with the conclusions of Biastoch et al. [2008] who suggest eddy contributions to AMOC fluctuations are strongest at high latitudes. However, it should be noted that, unlike the methods employed here, a clean separation of eddy-induced variability cannot be determined using their method of comparing a high resolution (1/12°) with lower resolution (1/3°) models.

[14] Time series of the AMOC variability and its low pass filtered component are displayed in Figure 3 at latitudes 5°N, 26.6°N, and 42°N for the Control and Steady runs. The variance of the unfiltered and filtered components at these three latitudes are shown in Table 1 for both experiments. The three chosen latitudes represent fairly well the variance ratios in the three latitude ranges 0–20°N, 20°N–30°N, and 40°N–65°N. The variance of the two filtered components add up almost linearly to the variance of the unfiltered data and can therefore be used to estimate the relative differences.

Figure 3.

Time series of the unfiltered (blue line) and low pass filtered (AMOClp; grey line) component of the (a, c, and e) Control and (b, d, and f) Steady AMOC anomaly at latitudes (a, b) 42°N, (c, d) 26.6°N, and (e, f) 5°N. Units are Sv.

Table 1. Variance of the Unfiltered (AMOC), Low Pass Filtered (AMOClp), and High Pass (AMOChp) Filtered Components of the Control and Steady AMOC Anomaly at Latitudes 5°N, 26.6°N, and 42 °Na
  1. a

    Units are Sv2.


[15] In the Control run, the high frequencies dominate the variability, except at 42°N where the low frequencies are more than double the strength of the high frequencies (Table 1). In the Steady run, the missing seasonal variability brings about large reductions in the low frequency component, and as such, the high frequencies dominate everywhere. In the tropical and subpolar latitudes, where the seasonal signal is strongest in the Control run, low frequency variability is significantly weaker in Steady. However, the reduction is only approximately 50% where the seasonal signal is weak at 26.6°N, meaning that eddy-induced low frequency variability is of importance at the latitude of the RAPID array in our model. High frequencies in Steady are everywhere weaker than in Control: They are approximately half the magnitude of the Control run high frequencies in the tropics and two thirds of the magnitude everywhere else. This implies either that the eddy-induced AMOC variability is underestimated in the Steady run or that there is some nonlinear interaction between the seasonal forcing and the eddy field.

[16] An interesting feature of the subtropical AMOC variability in the Control run is a pronounced interannual to decadal period variability (Figures 2a and 2b). This is particularly evident in the time series at 26.6°N and is a feature of the low frequency variability (Figure 3c). In particular, a sharp reduction and subsequent recovery at approximately year 6 of the Control run is not dissimilar to that seen in 2009 in RAPID observations of the AMOC taken at the same latitude [McCarthy et al., 2012]. Since the model is driven solely by climatological monthly mean forcing, such a large event has to be generated by processes that are internal to the model. Therefore, our results suggest that the recently reported 2009 event may well be a random event and therefore not linked to changes in external forcing. In our Control run, the event occurs coherently throughout all latitudes to the south of the Gulf Stream, as seen in the low pass filtered Control AMOC when Ekman transports have been removed (Figure 2h).

[17] Of most interest here is the contribution that eddies make to AMOC variability relative to the seasonally forced signal. To determine this, we take the ratio of the variance of the unfiltered Steady run to the variance of the unfiltered Control run at every latitude (Figure 4). There are approximately three latitude regimes between which eddy-induced variability compares differently to the total Control variability, roughly split between tropical, subtropical, and subpolar latitudes. Between the equator and approximately 15°N, and again between approximately 40°N and 60°N, the Steady run variability is about one third of the magnitude of the Control run variability. Between approximately 20°N and 38°N, the Steady run is almost two thirds the magnitude of the Control run. This result, that eddies make the strongest contribution at subtropical latitudes, agrees well with the estimations made by Hirschi et al. [2013, their Figure 11]. Eddy-induced variability of the AMOC is therefore of importance throughout the North Atlantic, though it is of particular importance in the subtropics where it dominates over the seasonal variability. As discussed above, it should be noted that the eddy contribution to AMOC variability may be underestimated in the Steady run, particularly in the tropics. If we smooth the data at a period of 10 days, thereby bringing the temporal resolution of the data into consistency with that published from the RAPID array [McCarthy et al., 2012], then the ratios of the Steady variance to the Control variance undergo relatively little change.

Figure 4.

Variance of the (blue line) unfiltered and (black line) low pass filtered Steady run AMOC anomaly as a ratio to the unfiltered Control AMOC anomaly. Units are percentage.

[18] It is interesting to determine the contribution that eddy-induced low frequency variability makes towards the seasonally forced AMOC, which we have estimated by taking the ratio of the variance of the low pass filtered Steady data to the variance of the unfiltered Control data (Figure 4). In the tropical and subpolar latitudes of the North Atlantic, the eddy-induced low frequencies generally are weaker than 10% of the Control variability. In the subtropics between 20°N–30°N, on the other hand, the contribution at a number of latitudes is one quarter and more of the magnitude of the Control variability, a non-negligible amount. The relative increase in both the low and high frequency contributions to AMOC variability in these subtropical latitudes is due to the weak seasonal cycle of the AMOC manifested in these latitudes (Figures 2c and 2d). If we smooth the data at a period of 10 days, then the values are not greatly changed, though the ratio increases by approximately 4% in favor of a stronger eddy-induced contribution at 26.6°N and nearby latitudes.

4 Conclusions

[19] We have isolated the eddy-induced variability of the Atlantic Meridional Overturning Circulation (AMOC) using an eddying model that was forced with seasonally varying climatological forcing (Control run) and with unvarying time mean climatological forcing (Steady run). We find the eddy contribution to AMOC variability to be significant throughout the North Atlantic relative to the magnitude of the seasonally forced AMOC, particularly in the subtropics. In approximately tropical and subpolar latitudes, the eddy-induced variability is shown to be approximately one third of the magnitude of the Control variability, while in the subtropical Atlantic, the fraction is approximately two thirds. At subtropical latitudes, where seasonal wind forcing in the Control run is weak, the eddy-induced at variability at seasonal and longer periods makes a non-negligible contribution to AMOC fluctuations. We find that approximately one fourth of the seasonally driven AMOC variability can be explained by eddy-induced variability at these periods. Outside of the subtropics, where the Control run seasonal signal is large, the eddy-induced variability at seasonal and longer timescales makes only a small contribution of less than 10%. Our model results thus suggest that changes in the strength of the AMOC observed by a monitoring array at high latitudes are more likely to be associated with changes in external forcing. It should be noted, however, that the high frequency variability is weaker in the Steady run than in the Control run, potentially indicating that eddy-induced AMOC variability in Steady is an underrepresentation of their contribution in the real ocean. Our calculations of eddy contributions to AMOC variability should therefore be considered as a lower bound estimation.

[20] When we smooth the data with a 10 day running mean, the results are found to be largely unchanged. We therefore suggest that a large fraction of the overturning variability estimated using the RAPID array [McCarthy et al., 2012] is composed of stochastic eddy-induced variability. This would be true for any other future monitoring arrays at other latitudes that are similarly designed. Real ocean forcing is, of course, more than just seasonal in timescale and contains variability that is missing in our model, as evidenced by the larger range of variability observed in the RAPID array measurements [McCarthy et al., 2012] and simulated in interannually forced models [Blaker et al., 2012; Hirschi et al., 2013]. However, eddy-induced variability has been shown to make up a large fraction of the seasonally forced fluctuations, which itself can be expected to form a significant component of AMOC variability in realistically forced ocean models [Hirschi et al., 2007].

[21] The current understanding of AMOC variability has been largely based on linear theory and insights gained from coarse (non-eddy-resolving) ocean climate models. In such frameworks, the AMOC variability is deterministic; that is, changes in the strength of the AMOC can always be traced back to changes in external forcing, be it wind or thermohaline forcing. Our results from an eddying North Atlantic Ocean model show otherwise—a significant fraction of the AMOC variability at seasonal to interannual timescales can be caused by stochastic eddies in the ocean that are inherently unpredictable. Our study calls on future research on the role of eddies in the AMOC variability.


[22] The authors are grateful to The University of East Anglia School of Environmental Sciences for financial support. Computations were carried out on the High Performance Computing Cluster supported by the Research Computing Service at the University of East Anglia. The authors thank two reviewers for their helpful suggestions and are grateful to Ben Harden for useful comments on the manuscript.

[23] The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.