LETTER doi:10.

1038/nature14491

Ocean impact on decadal Atlantic climate variability

revealed by sea-level observations
Gerard D. McCarthy1, Ivan D. Haigh2, Joël J.-M. Hirschi1, Jeremy P. Grist1 & David A. Smeed1

Decadal variability is a notable feature of the Atlantic Ocean and where sea-level fluctuations at any one point are influenced by
the climate of the regions it influences. Prominently, this is man- mesoscale variations17 even on long timescales, increasing the difficulty
ifested in the Atlantic Multidecadal Oscillation (AMO) in sea sur- of making estimates of ocean circulation that are coherent on large
face temperatures. Positive (negative) phases of the AMO coincide spatial scales. This is the case for sea level at Bermuda, whose decadal
with warmer (colder) North Atlantic sea surface temperatures. The fluctuations can be reproduced by considering a Rossby wave response
AMO is linked with decadal climate fluctuations, such as Indian to wind forcing16. To make estimates of ocean circulation that capture
and Sahel rainfall1, European summer precipitation2, Atlantic hur- the fluctuations in large-scale circulation and are less affected by eddy
ricanes3 and variations in global temperatures4. It is widely variability, measurements close to or on the western boundary are
believed that ocean circulation drives the phase changes of the necessary18. We account for this by focusing on the gradient of sea level
AMO by controlling ocean heat content5. However, there are no along the US east coast. Here the mean dynamic sea level decreases to
direct observations of ocean circulation of sufficient length to sup- the north (Fig. 1a) due to the transition from subtropical to subpolar
port this, leading to questions about whether the AMO is con- gyres. This dynamic gradient reflects a circulation that contains ele-
trolled from another source6. Here we provide observational ments not only of the Gulf Stream but also of cold, subpolar water from
evidence of the widely hypothesized link between ocean circulation the north, primarily associated with the overturning circulation19.
and the AMO. We take a new approach, using sea level along the Indeed, in model simulations, this meridional gradient of sea level along
east coast of the United States to estimate ocean circulation on this coast responds strongly to declines in the Atlantic overturning
decadal timescales. We show that ocean circulation responds to circulation20. Ultimately, it is the heat transport that we are interested
the first mode of Atlantic atmospheric forcing, the North in. And while the overturning circulation carries about 90% of the heat
Atlantic Oscillation, through circulation changes between the sub- at subtropical latitudes21, ocean heat transport at the latitude of the
tropical and subpolar gyres—the intergyre region7. These circula- intergyre region consists of similar contributions from both the over-
tion changes affect the decadal evolution of North Atlantic heat turning circulation and the gyre circulation22. For this reason, we do not
content and, consequently, the phases of the AMO. The Atlantic discuss separately overturning and gyre but only ocean circulation in
overturning circulation is declining8 and the AMO is moving to a this intergyre region, which contains elements of both mechanisms.
negative phase. This may offer a brief respite from the persistent Sea-level fluctuations from Florida to Boston divide into two coher-
rise of global temperatures4, but in the coupled system we describe, ent groups either side of Cape Hatteras23 (Extended Data Figs 2, 3). This
there are compensating effects. In this case, the negative AMO is large-scale coherence in sea level is driven by ocean circulation. North
associated with a continued acceleration of sea-level rise along the of Cape Hatteras, coherent sea-level fluctuations have been linked with
northeast coast of the United States9,10. fluctuations in the overturning circulation19,24. South of Cape Hatteras,
The difficulty in linking ocean circulation changes to decadal cli- fluctuations in the Gulf Stream from Florida to Cape Hatteras are
mate variations lies in the fact that long observational records of ocean reflected in sea-level fluctuations. As Cape Hatteras marks the bound-
transports are rare. Measurements such as those of the Florida Current ary between the subtropical and subpolar gyres on this coastline
since 198211 and the Greenland–Scotland ridge transports12 since the (Fig. 1a), we can construct a single sea-level composite representative
mid-1990s are some of the longest continuous ocean transport records of the subtropical (subpolar) circulation by averaging sea level from
available. Continuous, full-depth, basin-wide measurements of the linearly detrended, deseasonalized tide gauges, with the inverse baro-
Atlantic overturning circulation only began in 2004 with the RAPID meter effect removed, south (north) of the Cape (Fig. 1b, c). The dif-
monitoring project at 26u N (ref. 13). None of these records are long ference, south minus north (Fig. 1d), represents our circulation index.
enough to directly link ocean circulation with decadal climate varia- This index projects onto observed surface velocities during the satellite
tions such as the AMO. era in the intergyre region, with a positive index associated with more
Sea-level measurements from tide gauges provide an integrated northwards flow and a more northerly path of this circulation
measure of water column properties and offer timeseries of sufficient (Extended Data Fig. 4). Similarly, in a high-resolution ocean model,
length (Extended Data Fig. 1) to study decadal ocean circulation var- over timescales that contain both the cool phase of the AMO in the
iations. Investigating ocean circulation using tide gauges is not new: 1970s25 and the warm phase of the 1990s26, the sea-level index projects
the first attempt to estimate the Gulf Stream using tide gauges was onto a similar pattern of circulation, with a positive index associated
made in 193814. The principle is based on geostrophic dynamics: on with more northward heat transport (Extended Data Fig. 5).
timescales longer than a few days, ocean circulation is in geostrophic Ocean circulation is proportional to heat transport at both subtrop-
balance so, looking downstream, the sea level is seen to increase from ical and subpolar latitudes22. A number of recent studies (see, for
left to right in the Northern Hemisphere. example, ref. 27) have emphasized the dominant role of ocean heat
Estimates of the Gulf Stream using tide gauges have focused on the transport in heat content changes, relating the accumulation (in time)
American east coast, with an offshore estimate of sea level from either of heat transport to heat content. This suggests that the accumulation
an island gauge15 or a reconstructed sea level16. A weakness of this of our sea-level index across Cape Hatteras, as a proxy for ocean
method is that the offshore measurement lies in the eddy-filled ocean circulation, can be related to ocean heat content. The largest AMO
1
National Oceanography Centre, University of Southampton Waterfront Campus, European Way, Southampton SO14 3ZH, UK. 2Ocean and Earth Science, National Oceanography Centre, University of
Southampton Waterfront Campus, European Way, Southampton SO14 3ZH, UK.

5 0 8 | N AT U R E | VO L 5 2 1 | 2 8 M AY 2 0 1 5
G2015 Macmillan Publishers Limited. All rights reserved
 LETTER RESEARCH

400
a
55 b Northern sites
300 (7−30)
26
50
200

Sea-level anomalies (mm)

21
Latitude (ºN) 45 100 d South−North
30

Site number
40 0 16

7
35 –100 c Southern sites
11
6 (1−6)
–200
30
6
1 –300
25
–400 1
80 60 40 20 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
Longitude (ºW) Year

Figure 1 | Dynamic sea level and circulation along the western Atlantic of the AMO from 1995–2004 relative to from 1961–2012. b, c, Dynamic sea-
seaboard. a, Negative (positive) mean dynamic topography contours in blue level anomalies north (b; sites 7–30, 1200 mm offset) and south (c; sites 1–6,
(red) indicate cyclonic (anticyclonic) geostrophic streamlines. Contour values 2200 mm offset) of Cape Hatteras, with averages in black. d, The difference in
in metres shown in Fig. 2. The zero contour (dark blue) marks the boundary sea level, southern minus northern average, defines our sea-level index for
between the subtropical and subpolar gyres. Hatched areas indicate warm sea ocean circulation.
surface temperature anomalies of greater than 0.5 uC during the positive phase

signal is in the subpolar region (Fig. 1a), so we wish to show that, as a heat content change by 2 years with a maximum correlation of 0.58
measure of ocean circulation, our sea-level index is related to heat (significant at the 95% level). The reason that the accumulated sea-level
transport into the subpolar gyre and consequently heat content index leads the large rise in heat content from 40u N to 60u N in the early
changes there. Such a mechanism is supported by our model, in which 1990s can be interpreted by looking at maps of the heat content anom-
the sea-level index leads the heat transport into the subpolar gyre aly evolution. Heat content builds downstream of the intergyre region
at 40u N and, consequently, leads the heat content changes there from the mid-1980s to the mid-1990s (Fig. 2b). This heat content
(Extended Data Fig. 6). anomaly is then observed downstream in the subpolar gyre in the late
Although we do not have observations of heat transport, we can 1990s and early 2000s (Fig. 2c), indicating that the sea-level index could
relate our sea-level index directly to the heat content changes in the provide an early indication of subpolar heat content change.
subpolar gyre since 1960. Figure 2a shows the accumulated sea-level The first mode of atmospheric variability over the North Atlantic,
index (blue curve), together with a direct estimate of the heat content in the North Atlantic Oscillation (NAO), forces both buoyancy and
the area in the depth-weighted temperature anomaly in the top 500 m wind-driven ocean circulation7 and, we believe, is the major forcing
between 40u N and 60u N (black line). Heat content trends are similar of the circulation in the intergyre region. The 7-year NAO is signifi-
throughout the upper 1,000 m of the Atlantic, below which they reverse cantly correlated with (r 5 0.71 at the 98% level) and leads the 7-year
due to the depth structure of the Atlantic overturning circulation. The sea-level difference by approximately 1 year over the period 1950 to
cool subpolar upper ocean of the 1970s and 1980s and subsequent 2012. On extending the time period to 1920–2012, the correlation
warming in the 1990s is captured by the accumulated sea-level index, drops slightly but is still significantly correlated (r 5 0.61 at the 98%
observationally supporting the hypothesis that circulation changes and level, Extended Data Fig. 7). The correlation between the sea-level
not only air–sea fluxes were involved in these changes28. For the pur- difference and the NAO is higher and more significant than the
poses of statistical analyses, the timeseries have had a 7-year low-pass, correlation of the NAO with either the southern or northern sea-level
Tukey filter applied to them, which is referred to by the prefix ‘7-year’ (Fig. 1b, c) composites (r 5 20.5 at the 86% level for the southern
from here on. The 7-year sea-level index leads the 7-year rate of composite; r 5 20.43 at the 70% level for the northern); this supports

a b 1985 to 1994 ºC c 1995 to 2004

0.6 1 60
–0.5
–0.5

–0.1

3,000 0.8
–0.

55
Accumulated sea-level index (mm months)

0.4 0.6
50
2,000
Temperature anomaly (ºC)

0.4
0 0
45
0.2 0.2
Latitude (ºN)
0.1

1,000
0.1

0 40

0 Cape Hatt. –0.2 Cape Hatt.

0 35
–0.4
30
–0.2 –0.6
0.5

−1,000
0.5

25
–0.8

–0.4 –1 20
1960 1970 1980 1990 2000 2010 100 80 60 40 20 0 100 80 60 40 20 0
Year Longitude (ºW) Longitude (ºW)

Figure 2 | Relating the sea-level circulation index to heat content changes. accumulated NAO (red, dashed). b, Average temperature anomaly in the top
a, Accumulated sea-level index (nominally, in mm month) derived from 500 m for the periods 1985–94 relative to the average from 1958–2010.
accumulating the sea-level circulation index (blue), temperature anomaly in the Contours of mean dynamic topography (metres) defined in Fig. 1a are overlaid
upper 500 m of the subpolar North Atlantic from 40u to 60u N (black) and for reference. c, Same as b but for the period 1995–2004.

2 8 M AY 2 0 1 5 | VO L 5 2 1 | N AT U R E | 5 0 9
G2015 Macmillan Publishers Limited. All rights reserved
 RESEARCH LETTER

0.4 2. Sutton, R. T. & Dong, B. Atlantic Ocean influence on a shift in European climate in
the 1990s. Nature Geosci. 5, 788–792 (2012).
0.3 3. Goldenberg, S. B., Landsea, C. W., Mestas-Nuñez, A. M. & Gray, W. M. The recent
increase in Atlantic hurricane activity: causes and implications. Science 293,
0.2 474–479 (2001).
4. Chen, X. & Tung, K.-K. Varying planetary heat sink led to global-warming slowdown
Temperature (K)

0.1 and acceleration. Science 345, 897–903 (2014).

5. Delworth, T. L. & Mann, M. E. Observed and simulated multidecadal variability in
0 the Northern Hemisphere. Clim. Dyn. 16, 661–676 (2000).
6. Booth, B. B., Dunstone, N. J., Halloran, P. R., Andrews, T. & Bellouin, N. Aerosols
−0.1 implicated as a prime driver of twentieth-century North Atlantic climate variability.
Nature 484, 228–232 (2012).
−0.2 7. Marshall, J., Johnson, H. & Goodman, J. A study of the interaction of the North
Atlantic Oscillation with ocean circulation. J. Clim. 14, 1399–1421 (2001).
−0.3 8. Smeed, D. A. et al. Observed decline of the Atlantic Meridional Overturning
Circulation 2004 to 2012. Ocean Science 10, 39–38 (2014).
−0.4 9. Sallenger, A. H. Jr, Doran, K. S. & Howd, P. A. Hotspot of accelerated sea-level rise on
1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 the Atlantic coast of North America. Nature Clim. Change 2, 884–888 (2012).
Year 10. Boon, J. D. Evidence of sea level acceleration at US and Canadian tide stations,
Atlantic Coast, North America. J. Coast. Res. 28, 1437–1445 (2012).
Figure 3 | Sea-level circulation index, the NAO and the AMO on multi- 11. Meinen, C. S., Baringer, M. O. & Garcia, R. F. Florida Current transport variability:
decadal timescales. Shown are the accumulated sea-level index (blue), which is an analysis of annual and longer-period signals. Deep Sea Res. I 57, 835–846
representative of subpolar heat content evolution, the accumulated NAO (red, (2010).
dashed) and the AMO (black). The heat content proxy and the accumulated 12. Østerhus, S., Turrell, W. R., Jónsson, S. & Hansen, B. Measured volume, heat, and
NAO have been normalized. All timeseries have been 7-year low-pass filtered. salt fluxes from the Atlantic to the Arctic Mediterranean. Geophys. Res. Lett. 32,
The accumulated sea-level index and accumulated NAO have been detrended. L07603 (2005).
13. McCarthy, G. D. et al. Measuring the Atlantic Meridional Overturning Circulation at
26uN. Prog. Oceanogr. 130, 91–111 (2015).
our hypothesis that the NAO forces the ocean circulation and conse- 14. Montgomery, R. Fluctuations in monthly sea level on eastern US coast as
quently the ocean heat transport into the subpolar gyre. related to dynamics of western North Atlantic Ocean. J. Mar. Res. 1, 165–185
(1938).
In the past 90 years, the AMO has undergone three major transi- 15. Iselin, C. O. Preliminary report on long-period variations in the transport of the Gulf
tions: warming in the mid-1990s and 1920s, and a cooling in the 1960s. Stream system. Pap. Phys. Oceanogr. Meteorol. 3, 1–40 (1940).
From the early-1920s, when the tide gauge network along the east coast 16. Sturges, W. & Hong, B. Wind forcing of the Atlantic thermocline along 32uN at low
of North America was developed, robust comparisons of our sea-level frequencies. J. Phys. Oceanogr. 25, 1706–1715 (1995).
17. Wunsch, C. Mass and volume transport variability in an eddy-filled ocean. Nature
index to the AMO are possible (Fig. 3). The accumulated sea-level Geosci. 1, 165–168 (2008).
index and the accumulated NAO are linearly detrended and capture 18. Kanzow, T. et al. Basinwide integrated volume transports in an eddy-filled ocean.
much of the multi-decadal variation. The 7-year sea-level index leads J. Phys. Oceanogr. 39, 3091–3110 (2009).
19. Bingham, R. J. & Hughes, C. W. Signature of the Atlantic meridional overturning
the 7-year rate of change of the AMO by 2 years and is significantly circulation in sea level along the east coast of North America. Geophys. Res. Lett. 36,
correlated (r 5 0.51, significant at the 96% level, Extended Data Fig. 8). L02603 (2009).
This lead time of 2 years remains consistent when the timeseries is 20. Yin, J., Schlesinger, M. E. & Stouffer, R. J. Model projections of rapid sea-level rise on
broken into 60-year blocks (Extended Data Table 2). In recent years, the northeast coast of the United States. Nature Geosci. 2, 262–266 (2009).
21. Johns, W. E. et al. Continuous, array-based estimates of Atlantic Ocean heat
the sea-level index (Fig. 1d) indicates that the AMO is again transition- transport at 26.5uN. J. Clim. 24, 2429–2449 (2011).
ing to a negative phase, consistent with observations of a reduced 22. Grist, J. P. et al. The roles of surface heat flux and ocean heat transport convergence
overturning circulation8. in determining Atlantic Ocean temperature variability. Ocean Dyn. 60, 771–790
Using the sea-level difference between subtropical and subpolar (2010).
23. Thompson, P. & Mitchum, G. Coherent sea level variability on the North Atlantic
gyres, we have developed and validated a proxy for ocean circulation western boundary. J. Geophys. Res. 119, 5676–5689 (2014).
in the intergyre region. This represents a mechanism for ocean heat 24. Ezer, T. Sea level rise, spatially uneven and temporally unsteady: why the US east
transport to the subpolar gyre and heat content changes there. When coast, the global tide gauge record and the global altimeter data show different
observations exist, heat content changes have coincided with the major trends. Geophys. Res. Lett. 40, 5439–5444 (2013).
25. Hodson, D. L., Robson, J. I. & Sutton, R. T. An anatomy of the cooling of the North
phase transitions of the AMO, confirming that ocean circulation plays a Atlantic Ocean in the 1960s and 1970s. J. Clim. 27, 8229–8243 (2014).
key role in decadal Atlantic variability. The ocean responds to NAO 26. Häkkinen, S. & Rhines, P. B. Decline of subpolar North Atlantic circulation during
forcing with changes in ocean circulation: on decadal timescales, the the 1990s. Science 304, 555–559 (2004).
27. Bryden, H. L., King, B. A., McCarthy, G. D. & McDonagh, E. L. Impact of a 30%
ocean integrates NAO forcing and returns it to the atmosphere as the reduction in Atlantic meridional overturning during 2009–2010. Ocean Sci. 10,
AMO. This is implicitly the Bjerknes compensation that had previously 683–691 (2014).
been seen in air–sea fluxes29. The sea-level difference provides an indi- 28. Robson, J., Sutton, R., Lohmann, K., Smith, D. & Palmer, M. D. Causes of the rapid
cator of ocean circulation changes that precede phase changes in the warming of the North Atlantic Ocean in the mid-1990s. J. Clim. 25, 4116–4134
(2012).
AMO, thus explaining why, as the positive AMO declines4, sea-level 29. Gulev, S., Latif, M., Keenlyside, N. S. & Koltermann, K. North Atlantic Ocean control
rise is accelerating north of Cape Hatteras9,10. While Greenland ice sheet on surface heat flux at multidecadal timescale. Nature 499, 464–467 (2013).
melt has been linked with accelerating sea-level rise in recent years, the
Acknowledgements G.D.M. and D.A.S. are supported by the UK Natural Environment
fact that the period of accelerated sea-level rise from the 1950s to the Research Council (NERC) RAPID-WATCH programme. I.D.H. was partly supported by
1970s10 (as well as the current period of sea-level rise) coincides with a the UK NERC consortium project iGlass (NE/I009906/1). J.P.G. and J.J.-M.H. are
declining AMO indicates that multi-decadal fluctuations in ocean cir- supported by NERC National Capability funding.
culation play a key role. In this framework, sea-level rise along the US Author Contributions G.D.M. originated and developed the concept. I.D.H. provided the
east coast becomes entwined with the effects of the AMO on climate. tide gauge data analysis. J.P.G. and J.J.-M.H. provided the numerical model analysis.
D.A.S. carried out the statistical analysis. All authors contributed to the shaping and
Online Content Methods, along with any additional Extended Data display items production of the manuscript
and Source Data, are available in the online version of the paper; references unique
to these sections appear only in the online paper. Author Information Comma-separated data used in the manuscript are available to
download from http://bit.ly/1F7gtps. Reprints and permissions information is
Received 3 July 2014; accepted 8 April 2015. available at www.nature.com/reprints. The authors declare no competing financial
interests. Readers are welcome to comment on the online version of the paper.
1. Zhang, R. & Delworth, T. L. Impact of Atlantic multidecadal oscillations on India/ Correspondence and requests for materials should be addressed to G.D.M.
Sahel rainfall and Atlantic hurricanes. Geophys. Res. Lett. 34, L23708 (2006). ([email protected]).

5 1 0 | N AT U R E | VO L 5 2 1 | 2 8 M AY 2 0 1 5
G2015 Macmillan Publishers Limited. All rights reserved
 LETTER RESEARCH

METHODS whereas subpolar water can only enter the subtropics at depth (traditionally in the
Data. Monthly mean sea-level records were obtained from the Permanent Service deep western boundary current). Therefore we relate the heat transport into
for Mean Sea-level (www.psmsl.org) for tide gauges stretching from Florida to the subpolar gyre and heat content of the upper waters of the subpolar gyre
Boston (locations 1 to 30, Extended Data Fig. 1). Linear trends were removed from to the transport in the intergyre region:
each record. This removes the impact of glacial isostatic adjustment and other land HT400N !hs {hn
subsidence effects, which have time periods of thousands of years and are known
to affect tide gauges along this coastline. A 12-month low-pass filter removed the For exactly the reason that we need to use tide gauges as a proxy for heat
seasonal cycle. Southern (northern) composites of sea level were calculated by transport, we cannot validate the conceptual model directly due to the lack of
averaging records 1–6 (7–30). The meridional coherence of sea-level fluctuations direct observations. However, a global eddy-permitting (1/4u) ocean model
is such that using just a single tide gauge results in an r.m.s. error of only 5 mm (ORCA-025) provides the framework to investigate these balances. The heat
relative to the full composite. Finally, the sea-level index is simply the difference transport into the subpolar gyre has previously been shown in this model to be
obtained by subtracting the northern from the southern sea-level composite. The the dominant factor in setting upper ocean temperature in the subpolar gyre22.
high level of meridional coherence allows the interpretation of the sea-level gra- Here, we reproduce this result, showing that the accumulated heat transport across
dient as this simple index. 40u N captures the major decadal fluctuations in heat content of the subpolar gyre
Sources. Monthly NAO data from the National Center for Atmospheric Research (Extended Data Fig. 6). We can use these heat transport measurements to validate
‘‘The Climate Data Guide: Hurrell North Atlantic Oscillation (NAO) Index (PC- our circulation index. At this resolution there are shortcomings in the representa-
based)’’ (https://climatedataguide.ucar.edu/climate-data/hurrell-north-atlantic- tion of the Gulf Stream path: the Gulf Stream overshoots at Cape Hatteras
oscillation-nao-index-pc-based); monthly AMO index, based on the Kaplan sea and separates from the US coast too far north. However, we take account of
surface temperature (SST) data set (from http://www.esrl.noaa.gov/psd/data/ this in choosing the northern and southern sea-level points so that they
timeseries/AMO/); subsurface temperature data from the EN3 product (http:// straddle the separation point. Also, despite the model being eddy-permitting
www.metoffice.gov.uk/hadobs/en3/); geostrophic velocity anomalies were pro- rather than eddy-resolving, it does generate mesoscale variability. This is seen
duced and distributed by Aviso (http://www.aviso.altimetry.fr/) as part of the when including an offshore sea-level measurement (such as Bermuda) in a sea-
Ssalto ground processing segment. CNES-CLS09 Mean Dynamic Topography level circulation index. Such an index fails to reflect the large scale circulation. This
(v1.1 release) for the period 1993–99 was produced by the French Space Agency effect would be expected to be even larger in an eddy-resolving model. Extended
CNES. Data Fig. 5 shows that the model-derived sea-level index projects onto the inter-
Model validation. The multi-decadal oscillation of SSTs is most intense in the gyre velocities in a similar manner to the observed sea-level index. Extended Data
subpolar gyre (Fig. 1a). Modelling studies have shown that it is ocean heat trans- Fig. 6 shows the accumulated sea-level difference compared with the accumulated
port into the subpolar gyre (here we choose 40u N) that controls the heat content of heat transport across a section near 40u N and the volume averaged temperature of
the subpolar upper ocean and consequently the SST. The concept here is that the upper 500 m of the subpolar gyre (40u N to 60u N). The sea-level difference is
circulation in the intergyre region reflects the balance between warm subtropical significantly correlated with the heat transport into the subpolar gyre (r 5 0.62)
water entering the subpolar gyre and colder subpolar water being recirculated and leads by 5 years (as in the main text, we report statistics on unaccumulated
within the gyre. We show that the sea-level gradient along the US east coast is a timeseries).
good proxy for this circulation (Extended Data Figs 4 and 5). Code availability. Matlab code is available to download from http://bit.ly/
We can relate sea-level changes to ocean circulation in a reduced gravity geo- 1F7gtps.
strophic framework: Statistical analysis. Cross-correlations are calculated using annually averaged
g0 data after first removing the mean and linear trend from each variable. Two
v~ k|+h approaches are used to quantify the uncertainty in the correlation. First, we cal-
f
culated the parameter
where v is geostrophic velocity, k is the unit vector in the vertical direction, h is pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sea level, g 9 is reduced gravity and f is the Coriolis parameter. To estimate the r
T~ ðN{2Þ pffiffiffiffiffiffiffiffiffiffiffi
transport in the intergyre region, previous studies have considered the sea-level 1{r 2
difference between an onshore tide gauge and an offshore tide gauge, such as where r is the correlation and N is the number of samples. The distribution of T is
Bermuda. Reference 24, for example, relates the sea-level difference between assumed to have a t-distribution with N 2 2 degrees of freedom when the samples
Atlantic City and Bermuda to the Atlantic overturning circulation. are not autocorrelated. This is used with a one-sided test to estimate the likelihood
However, Bermuda is in the eddy-filled ocean interior17, which can disrupt that the correlation has not occurred by chance (that is, the certainty with which
spatially-coherent ocean transport signals. Our approach is to use sea-level esti- we can reject the null hypothesis). Our data are autocorrelated and the number of
mates south of Cape Hatteras instead of an offshore sea-level estimate. Dynamic independent samples (degrees of freedom) is therefore smaller than N. To cal-
topography along the US east coast also decreases to the north across the intergyre culate the effective number of degrees of freedom we follow ref. 30 by evaluating
boundary at Cape Hatteras much as it decreases from Bermuda to Atlantic City. the autocorrelation of each variable and the estimate N as
However, measurements on the coast do not suffer the same contamination due to
eddies as mid-ocean measurements18. Hence we estimate the transport along the ð1{a1 a2 Þ
Neff ~Nobs
intergyre boundary as: ð1za1 a2 Þ
vig !hs {hn where Neff is the degrees of freedom, Nobs is the number of observations and a1, a2
are the values of the autocorrelations at a lag of one year. We evaluated Neff over the
where the subscript ig refers to the intergyre region, s and n refer to south and longest time for each variable and then used the lowest value for all correlations.
north respectively. We can formulate the heat transport through a section strad- For the shorter time series Neff was reduced in proportion to the length of the
dling the intergyre boundary as: series. Degrees of freedom are reported in Extended Data Table 1.
ðð
HTig ~rcp Hvig dA In a second approach we applied the non-parametric method described in
ref. 31. A large number (we used 10,000) of simulated time series are con-
where r is density, cp is the specific heat capacity of seawater, H is conservative structed from the Fourier transform of one of the original data series by
temperature and A is the area of the section considered. In this study we assume preserving the modulus of each Fourier component but changing the phase
that the velocity fluctuations dominate the temperature fluctuations and so set the to a random value between 0 and 2p. The distribution of correlations between
heat transport directly proportional to the intergyre velocity. This is an assump- these random series and the second variable was then calculated. The percent-
tion that has proved true in direct heat transport estimates21. We note there is no age of simulated correlations that are less than the observed correlation indi-
dilemma in picking the location of the northern or southern points as the meri- cates the confidence that the true correlation is greater than zero. Because we
dional coherence of sea-level fluctuations allows us to use a simple average of all are considering lagged correlations we modify the technique of ref. 31 so that
sea-level records from Miami Beach to Cape Hatteras (Cape Hatteras to Boston) for each simulated time series we evaluate the maximum of cross-correlation
for hs (hn). across all lags rather than the correlation at zero lag only. This provides a more
In terms of upper ocean heat content, the heat transported in this intergyre stringent test of confidence.
region has a profound impact on the subpolar gyre. This is because warm water To estimate the uncertainty in the time lag of the maximum correlation we used
may be transferred from the upper waters of the subtropics to the subpolar gyre the times at which the correlation was equal to the maximum value less the

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

standard deviation of correlations derived from the simulated time series. The statistical analyses, the timeseries have had a 7-year, Tukey filter applied to them,
results are summarized in Extended Data Table 1. which is referred to in the text with the prefix ‘7-year’.
We have also evaluated the correlation over shorter periods to determine if Sample size. The timespan of the study was the maximum for which all of the
the lag has remained constant over time. Results from three overlapping 60-year necessary data were available. Therefore no statistical methods were used to pre-
periods are shown in Extended Data Table 2. For each the correlation is a max- determine sample size as we used all the samples available to us.
imum when sea-level difference leads the differentiated AMO by 2 to 3 years.
30. Bretherton, C. S., Widmann, M., Dymnikov, V. P., Wallace, J. M. & Bladé, I. The
The text refers to both accumulated and unaccumulated timeseries. Accumu-
effective number of spatial degrees of freedom of a time-varying field. J. Clim. 12,
lation of zero mean timeseries constrains the beginning and end of the accumu- 1990–2009 (1999).
lated timeseries to zero. To avoid this arbitrary constraint, we report all 31. Ebisuzaki, W. A method to estimate the statistical significance of a correlation when
our statistics on unaccumulated timeseries. As mentioned, for the purposes of the data are serially correlated. J. Clim. 10, 2147–2153 (1997).

G2015 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

30 30. Boston
42 27 29. Nantucket Island
25 26 28 29
(a) 23 24 (b) 28. Woods Hole
22
21
20 27. Providence
40 17 26. Newport
13 19 25. New London
11 12 18
14 16 24. Montauk
10 23. Bridgeport
38
9 15 22. Willets Point
87 21. New York
20. Sandy Hook
36
19. Atlantic City
18. Cape May
6 17. Philadelphia
Latitude

34 16. Lewes
15. Kiptopeke Beach
5 14. Cambridge II
32 4 13. Baltimore
12. Annapolis
3 11. Washington DC
2
30 10. Solomon’s Island
09. Gloucester Point
08. Sewells Point
07. Portsmouth
28
06. Wilmington
05. Charleston I
04. Fort Pulaski
26 1 03. Fernandina Beach
02. Mayport
01. Miami Beach
24
−80 −78 −76 −74 −72 −70 1900 1920 1940 1960 1980 2000
Longitude Year

Extended Data Figure 1 | Tide gauges used in this study. a, Locations and b, temporal coverage of the tide gauges used in this study.

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

Sea Level Anomalies Site No.

3 30

2.5 25

2
Level (m) (arbitrarily offset)

20

1.5
15

1
10

0.5
5

0
1900 1920 1940 1960 1980 2000
Year

Extended Data Figure 2 | Dynamic sea-level anomalies from the 30 stations tide gauges along this coastline. A seasonal cycle was removed using a
used in this study. Linear trends were removed from each record. This 12-month boxcar filter. From 1920, there are multiple tide gauges both north
removes the impact of glacial isostatic adjustment and other land subsidence and south of Cape Hatteras, so this is when we begin our study.
effects, which have time periods of thousands of years and are known to affect

G2015 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

Corr.
1
30
29 0.95
28
0.9
27
26 0.85
25
0.8
24
23 0.75
22
0.7
21
20 0.65
19
0.6
18
17 0.55
Site no.

16
0.5
15
14 0.45
13
0.4
12
11 0.35
10
0.3
9
8 0.25
7
0.2
6
5 0.15
4
0.1
3
2 0.05
1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Site no.
Extended Data Figure 3 | Correlation of tide gauges along the US east coast relative to one another. The dashed line indicates the location of Cape Hatteras.
There is high correlation between tide gauges grouped north and south of Cape Hatteras.

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

−0.1 −0.08 −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 0.1

Surf. Velocity Anom. (ms−1) Merid. Surf. Transport Anom. (10 m s )

3 2 −1

−6 −4 −2 0 2 4
60

55

(a) (b)
50

45
Latitude

40

35

30

25

20
−100 −90 −80 −70 −60 −50 −40 −30 −20 −10 0
Longitude
Sea−Level Index (mm)

100 (c)

−100

1994 1996 1998 2000 2002 2004 2006 2008 2010

Year

Extended Data Figure 4 | Surface velocity anomaly when the sea-level index with a more northerly circulation in the intergyre region and increased surface
is positive. a, Magnitude (m s21) and b, zonally integrated meridional velocity flow into the subpolar gyre. Velocities are geostrophic surface velocities derived
anomalies (103 m2 s21) for the time period 1993 to 2011, corresponding to from satellite altimetry.
when (c) the sea-level index is positive. A positive sea-level index is associated

G2015 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

−0.1 −0.08 −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 0.1

−1
Surf. Velocity Anom. (ms ) Heat Transport Change (%)
0 5 10
60

55

(a) (b)
50

45
Latitude

40

35

30

25

20
−100 −90 −80 −70 −60 −50 −40 −30 −20 −10 0
Longitude
Sea−Level Index (mm)

200
(c)
100

−100

−200
1960 1965 1970 1975 1980 1985 1990 1995 2000
Year
Extended Data Figure 5 | Model-derived surface velocity anomaly to 2001, corresponding to when (c) the model-derived sea-level index is
magnitude when the model-based sea-level index is positive. Similar to positive. Similar to the satellite observations, a positive sea-level index is
observed velocities, positive indices are associated with more northerly associated with a more northerly circulation in the intergyre region. Meridional
circulation in the intergyre region. a, Surface velocity magnitude (m s21) and heat transport change in both subtropical and subpolar gyres is positive when
b, percentage of meridional heat transport change (%) for the time period 1958 the sea-level index is positive.

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

Acc. SL diff Acc. HT40N Subpolar HCA

5 0.5
mm months

C
0 0

°
−5 −0.5
1960 1965 1970 1975 1980 1985 1990 1995 2000
Year
Extended Data Figure 6 | Model-derived sea-level index, heat transport gyre across 40u N (Acc. HT40N, black, normalized units). The heat transport
and subpolar heat content. The accumulated sea-level index (Acc. SL diff, into the subpolar gyre dominates the top 500 m temperature anomaly
blue, in mm months) leads the accumulated heat transport into the subpolar (Subpolar HCA, green, uC) in the subpolar gyre.

G2015 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

(a)
40 0.5

20

NAO index
0 0

−20

−40 −0.5
1920 1930 1940 1950 1960 1970 1980 1990 2000 2010

(b) (c)
1 1500

0.5
Frequency
Correlation

1000

500
−0.5

−1 0
−20 −10 0 10 20 0 0.5 1
Lag (years) Correlation
Extended Data Figure 7 | Relationship between sea-level index and the that would be expected from randomly generated timeseries with similar
NAO. a, 7-year sea-level difference (blue, cm) and 7-year NAO (green, spectral properties to the original timeseries. The red line indicates the
normalized units). b, Lagged correlations between the two quantities. maximum correlation between the two timeseries.
c, Scrambled correlation tests. The histogram indicates the typical correlations

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

(a)

Rate of change of AMO (°C yr )

40 0.1

20 0.05

0 0

−20 −0.05

−40 −0.1
1920 1930 1940 1950 1960 1970 1980 1990 2000 2010

(b) (c)
1 1500

0.5
Correlation

Frequency

1000

500
−0.5

−1 0
−20 −10 0 10 20 0 0.5 1
Lag (years) Correlation
Extended Data Figure 8 | Relationship between sea-level index and the rate that would be expected from randomly generated timeseries with similar
of change of the AMO. a, 7-year sea-level difference (blue) and rate of change spectral properties to the original timeseries. The red line indicates the
of the AMO (green). b, Lagged correlations between the two quantities. maximum correlation between the two timeseries.
c, Scrambled correlation tests. The histogram indicates the typical correlations

G2015 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

Extended Data Table 1 | Sea level, NAO and rates of change of the AMO statistics

Correlation, lags and significance between the southern (B) and northern (A) sea-level indices. First variable (Var X) and second variable (Var Y) are indicated in the first two columns. In the second column, Di refers
to the rate of change, and HC refers to subpolar heat content from 40u to 60u N. All timeseries are filtered with a 7-year low-pass filter (Filt). Degrees of Freedom (DoF), correlation (Corr) and significance (Sig.) are
shown. Significance is determined by a t-statistic (t-stat) or a scrambled test (Scrm). The RMS of a random correlation (column 9) together with the lag at maximum correlation (column 10) allows determination of
a range for the lag estimate (column 11).

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

Extended Data Table 2 | Sub-sampled statistics

Correlation, lags and lag range of sea-level index and the rate of change of the AMO over various time periods to investigate the consistency of the lags. Time periods under consideration are indicated in the first two
columns.

G2015 Macmillan Publishers Limited. All rights reserved