Modeling evidence for large, ENSO-driven interannual wintertime AMOC variability

Over the past decade, moored observational arrays stretching across the Atlantic Ocean, such as RAPID/MOCHA, OSNAP and MOVE (Frajka-Williams et al 2019), have been monitoring the Atlantic Meridional Overturning Circulation (AMOC) at high temporal resolution (Cunningham et al 2007, Kanzow et al 2010, Smeed et al 2014). These unprecedented data sets have revealed a large seasonal cycle and significant high-frequency variability in the AMOC, previously unobserved due to temporal and spatial sparsity of measurements (Cunningham et al 2007, Kanzow et al 2010). Here, high-frequency AMOC variability refers to variability on intra-annual to interannual time scales.

While observations of the AMOC have been limited until recently, exploration of the AMOC with general circulation models (GCMs) has been an active area of research. The majority of GCM studies have tended to focus on low-frequency AMOC variability, and its role in driving decadal-to-centennial timescale climate variability (e.g. Knight et al 2005, Zhang et al 2007, Danabasoglu 2008, Danabasoglu et al 2012, Delworth et al 2017), and thus limited their analysis to the annual mean (or low-pass filtered) AMOC. Nonetheless, high-frequency variability of the AMOC in models has also been investigated (Buckley and Marshall 2016, and references therein): it has been largely attributed to local wind forcing and this has informed the conclusion that the observed high-frequency variability (e.g. in the RAPID time series) is likewise driven by local wind forcing, particularly that associated with the North Atlantic Oscillation (NAO) (Roberts et al 2013, Zhao and Johns 2014, Elipot et al 2017). Questions remain, however, as to the sources of the NAO variability, and whether such sources can be linked directly to the seasonal AMOC variability.

The aim of this study is to examine the extent to which remote sources of variability, specifically the El Niño-Southern Oscillation (ENSO), can influence the wind-driven, high-frequency AMOC circulation via atmospheric teleconnections and how this influence can be highly seasonal. Improving our understanding of seasonal AMOC variability will aide in our interpretation of ongoing observational programs (Frajka-Williams et al 2019), and, thus, advance our ability to detect changes in the AMOC and predict how the AMOC will evolve in response to increasing greenhouse gases.

Here, analyzing two configurations of the Community Earth System Model (CESM), we find robust Ekman-induced variability in the wintertime AMOC in the North Atlantic that is significantly linked to ENSO via surface wind-stress anomalies associated with the NAO. We further show that this pattern of variability projects strongly onto the leading mode of the wintertime AMOC, and is distinct from the leading mode of the annual mean AMOC.

2.1. Global climate models

2.2. Integrations

2.3. Analysis

Before examining the effect of ENSO on the AMOC, we begin by presenting the wintertime (JFM) ENSO-NAO teleconnection in CAM5. While the dynamical connection between ENSO and the NAO remains an area of ongoing research (Mezzina et al 2020), several studies have identified a relationship between ENSO and the NAO in winter via the north-eastward propagation of planetary waves excited by anomalous convection in the tropical Pacific region (Rodríguez-Fonseca et al 2016, and references therein). The teleconnection pathway can be either tropospheric or stratospheric, with El Niño events being associated with the negative phase of the NAO and La Niña events being associated with the positive phase of the NAO (see Polvani et al 2017, for example).

Here, using large sample sizes, we find a clear NAO-like sea level pressure (SLP) pattern associated with ENSO. Figures 1(a) and (b) show El Niño and La Niña composites for CAM5 of wintertime SLP anomalies. In the North Pacific region, we clearly see the anomalous low and high pressure teleconnection patterns associated with El Niño and La Niña, respectively. In the North Atlantic region, we see the characteristic dipole SLP anomaly pattern of the NAO. In a composite mean sense, a negative NAO pattern (positive SLP anomalies in the subpolar North Atlantic and negative SLP anomalies in the subtropical North Atlantic (Hurrell et al 2013)) accompanies El Niño events and a positive NAO pattern accompanies La Niña events.

Zoom In Zoom Out Reset image size

Although many GCMs and Earth System Models (ESMs) have biases in their representation of ENSO (Karamperidou et al 2017, Feng et al 2020), we note that the ENSO teleconnections in CAM5 are reasonably well-represented in the North Atlantic region with a slight underestimation of the associated NAO pattern (Deser et al 2017). We find a similar connection between ENSO and the NAO in WACCM4 (see figure S1. Note that the main body of the paper shows plots for CAM5, while complimentary plots for WACCM4 are included in the supplementary material (available online at stacks.iop.org/ERL/16/084038/mmedia)).

These atmospheric NAO anomalies are associated with characteristic patterns of air-sea coupling. Figures 1(c) and (d) show El Niño and La Niña composites of wintertime surface wind stress ($\vec{\tau}$; vectors) and turbulent (sensible and latent) heat flux anomalies in the North Atlantic region. The anomaly patterns are typical of those associated with the NAO (Frankignoul 1985, Cayan 1992, Deser and Timlin 1997, Wen et al 2005, Visbeck et al 2013). For El Niño (figure 1(c)), we find anomalously westward surface wind stress north of 40^∘ N and anomalously eastward surface wind stress south of 40^∘ N. This is associated with a reduced turbulent heat flux from the ocean to the atmosphere north of 40^∘ N, i.e. anomalously positive turbulent heat flux into the ocean. We see slightly weaker and oppositely signed patterns for La Niña (figure 1(d)). The SST anomalies (figure S2) that accompany these wind stress and turbulent heat flux anomalies exhibit the characteristic tripolar pattern associated with the NAO; for El Niño, we find positive SST anomalies in the subpolar and tropical regions and negative SST anomalies in the subtropics. Note that the SST anomalies along the coast of Labrador are associated, in part, with negative and positive sea ice concentration anomalies for the El Niño and La Niña composites, respectively (not shown) (Strong et al 2009).

Given the significant impact of ENSO via the NAO on the air-sea exchange of heat and momentum over the North Atlantic, we now explore the possibility that ENSO may actually be able to impact the AMOC (via the NAO). Previous studies have examined both the impact of ENSO and the NAO on the AMOC, but have focused on the annual mean AMOC and have found little evidence of a significant effect on interannual timescales (Mignot and Frankignoul 2005, Delworth and Zeng 2016). Here we restrict our analysis to the wintertime AMOC and show that there is indeed a significant effect of ENSO on the wintertime AMOC.

We begin by examining the effect of ENSO on the mixed-layer depth, using the mixed-layer depth definition based on the buoyancy gradient approach described in Large et al (1997). Figures 1(e) and (f) show El Niño and La Niña composites of wintertime mixed-layer depth anomalies in the North Atlantic region. We clearly see significant anomalies in the Labrador Sea region, the primary region of convection and deep water formation in the North Atlantic. During El Niño events, when the mixed layer is heated via turbulent heat flux anomalies and momentum is extracted from the mixed-layer via surface wind stress anomalies, we find that the mixed layer shoals by approximately 50–100 m in the Labrador Sea region. We see the opposite during La Niña events, a deepening of the mixed layer by approximately 25–50 m in the Labrador Sea region. The question remains whether the influence of these wind stress, heat flux and mixed-layer depth anomalies is restricted to the surface ocean or is associated with large-scale changes in the AMOC itself.

To address this question, we now present our key finding. Figures 2(a) and (b) show El Niño and La Niña composites of wintertime AMOC anomalies (see also figure S3 for WACCM4). We find significant anomalies in the large-scale circulation that extend well-below the Ekman layer. Specifically, during El Niño events the AMOC significantly weakens to the south of approximately 40^∘ N and significantly strengthens north of 40^∘ N. The southern anomalies extend throughout the depth of the Atlantic Basin, while the northern anomalies are restricted to the upper 2–3 km. Slightly weaker and opposite-signed AMOC anomalies occur during La Niña events. The AMOC anomalies associated with ENSO approximately represent a shift about the climatological winter maximum streamline (see black contours in figure 2). The anomalies are largest at the surface and decay with depth.

Zoom In Zoom Out Reset image size

Given that we are considering interannual wintertime anomalies rather than long-term averages, the vertical structure of the AMOC anomalies suggests a process of oceanic adjustment. Near the surface, we find wind-driven Ekman transport anomalies and at depth, corresponding compensating flow anomalies with a baroclinic structure (Tandon et al 2020). Notably, the AMOC anomalies at high-latitudes are opposite in sign to what might be expected if changes in Labrador Sea deep water formation were the dominant factor (figures 1(e) and (f)), demonstrating the key role of the wind stress anomalies. The opposing effects of deep water formation and wind forcing at high-latitudes may contribute to the weaker circulation of the northern cell compared to its southern counterpart.

As a point of comparison, we also show El Niño and La Niña composites of the annual mean AMOC in figures 2(c) and (d) (see also figure S3 for WACCM4). Very little of the winter ENSO signal is seen in the annual mean AMOC, highlighting that detection of an interannual link between ENSO and the AMOC will fail if only the annual mean AMOC is considered.

To assess the importance of this wintertime pattern of AMOC variability, we perform a PC analysis and extract the leading mode of wintertime AMOC variability in CAM5. The spatial pattern of the leading mode, i.e. the first empirical orthogonal function (EOF), (figures 3(a) and S4(a) for WACCM4) is strikingly similar to the pattern of the ENSO composite AMOC anomalies shown in figures 2(a) and (b) and reminiscent of the NAO-associated mode described in Elipot et al (2017). In CAM5, the first EOF accounts for more than half of the variance in the wintertime AMOC with 52% of the variance explained (the variance explained by the second and third EOFs are 12% and 9%, respectively). It is important to note that the nature of this EOF pattern is quite distinct from the leading EOF of the annual mean AMOC, which shows a basin-scale pattern (Danabasoglu et al 2012). Complementary statistics for WACCM4 are included in the supplementary material (text S1).

Zoom In Zoom Out Reset image size

The role of ENSO in contributing to this mode of variability can be illustrated by isolating years of the PC time series that correspond to El Niño and La Niña events. Figure 3(b) (figure S4(b) for WACCM4) shows the probability density distributions (PDFs) of these PC values (in units of standard deviation) for the El Niño (red) and La Niña (blue) winters. We clearly see that the PDFs are well-separated and this is confirmed by performing a Kologorov–Smirnov test, which indicates that the two distributions are significantly distinct.

We also find that the correlation between the leading PC time series of the winter AMOC and the winter ENSO index is quite high and statistically significant at the 95% level (R² = 25.6%), while the correlation between the leading PC time series of the annual mean AMOC and the winter ENSO index is very low, but also statistically significant given the large sample size (R² = 3.5%). This helps to explain why studies of the annual mean AMOC may have overlooked this significant link between ENSO and the AMOC in winter. Furthermore, we find that the correlation between the leading PC time series of the winter AMOC and the winter NAO index is somewhat lower, but statistically significant (R² = 17.5%) than for the winter ENSO index. Thus, we find that the winter AMOC appears to be more strongly influenced by the ENSO-NAO teleconnection rather than the NAO itself. Given that the NAO is not entirely independent of ENSO, this may be related to certain characteristics of the NAO, such as the persistence of the NAO in the presence of ENSO due to local atmosphere-ocean feedbacks (Alexander and Scott 2008, Hu et al 2011). This finding requires further study; however, we do not have sufficient daily model output to examine this at this time.

Using long, fully-coupled control integrations of CAM5 and WACCM4, we have here shown that the wintertime AMOC exhibits a distinct mode of variability that appears to be strongly influenced by ENSO via the NAO. The dipole pattern of variability about the wintertime AMOC maximum resembles an Ekman-induced baroclinic adjustment (Elipot et al 2017, Tandon et al 2020).

Although these findings are model-based and restricted to only two GCMs, they support the growing observational evidence for substantial high-frequency AMOC variability. This variability that has been primarily attributed to local wind forcing, but its distinct seasonal nature and connection to remote sources of variability, such as ENSO, have been largely overlooked in previous studies (e.g. Mignot and Frankignoul 2005, Balan Sarojini et al 2011, Delworth and Zeng 2016).

Further work is required to establish the climate implications of such a mode of wintertime AMOC variability (Delworth et al 2016). However, our finding will help with the interpretation of the ongoing AMOC measurements being collected by trans-Atlantic arrays (Frajka-Williams et al 2019). Distinguishing anthropogenic trends from internal variability is a complex, yet very important, task when analyzing in-situ observations. Work such as this highlight that there may be significant sources of internal variability that are yet to be fully explored.

K L S and L M P gratefully acknowledge support from the NSF Division of Polar Programs, PLR-1603350. We also owe tremendous thanks to Gokhan Danabasoglu for valuable discussions and the CESM Large Ensemble Community Project and for the associated supercomputing resources provided by NSF/CISL/Yellowstone that have provided the research community access to the CESM1(CAM5) LENS integrations. The CESM1(WACCM4) integration is stored on the NCAR High Performance Storage System and is available upon request. NCAR is sponsored by the National Science Foundation.

The data that support the findings of this study are openly available at the following URL/DOI: www.cesm.ucar.edu/projects/community-projects/LENS/data-sets.html.