15 MARCH 2009
TING ET AL.
1469
Forced and Internal Twentieth-Century SST Trends in the North Atlantic*
MINGFANG TING, YOCHANAN KUSHNIR, RICHARD SEAGER, AND CUIHUA LI
Lamont-Doherty Earth Observatory, Columbia University, Palisades, New York
(Manuscript received 7 April 2008, in final form 26 September 2008)
ABSTRACT
In recent years, two alarming trends in North Atlantic climate have been noted: an increase in the intensity
and frequency of Atlantic hurricanes and a rapid decrease in Greenland ice sheet volume. Both of these
phenomena occurred while a significant warming took place in North Atlantic sea surface temperatures
(SSTs), thus sparking a debate on whether the warming is a consequence of natural climate variations,
anthropogenic forcing, or both; and if both, what their relative roles are. Here models and observations are
used to detect and attribute long-term (multidecadal) twentieth-century North Atlantic (NA) SST changes to
their anthropogenic and natural causes. A suite of Intergovernmental Panel on Climate Change (IPCC)
twentieth-century (C20C) coupled model simulations with multiple ensemble members and a signal-to-noise
maximizing empirical orthogonal function analysis are used to identify a model-based estimate of the forced,
anthropogenic component in NA SST variability. Comparing the results to observations, it is argued that the
long-term, observed, North Atlantic basin-averaged SSTs combine a forced global warming trend with a
distinct, local multidecadal ‘‘oscillation’’ that is outside of the range of the model-simulated, forced component and most likely arose from internal variability. This internal variability produced a cold interval
between 1900 and 1930, followed by 30 yr of relative warmth and another cold phase from 1960 to 1990, and a
warming since then. This natural variation, referred to previously as the Atlantic Multidecadal Oscillation
(AMO), thus played a significant role in the twentieth-century NA SST variability and should be considered
in future, near-term climate projections as a mechanism that, depending on its behavior, can act either
constructively or destructively with the region’s response to anthropogenic influence, temporarily amplifying
or mitigating regional climate change.
1. Introduction
The extremely active and destructive hurricane season in 2005 sparked an intense debate as to whether or
not the intensification of hurricane activity during recent decades was due to natural variability or anthropogenic forcing. At the center of the debate is the cause
and impact of the concomitant warming over the North
Atlantic (NA). A few recent studies (e.g., Emanuel
2005; Webster et al. 2005; Santer et al. 2006) linked the
increase in the intensity of the Atlantic hurricanes to
the rise in tropical Atlantic SST and suggested that the
latter is due to global warming. Other studies argued
* Lamont-Doherty Earth Observatory Contribution Number
7204.
Corresponding author address: Mingfang Ting, Lamont-Doherty
Earth Observatory, 61 Rt. 9W, Columbia University, Palisades,
NY 10964.
E-mail: ting@ldeo.columbia.edu
DOI: 10.1175/2008JCLI2561.1
Ó 2009 American Meteorological Society
that naturally occurring multidecadal SST variability is
the main source of the recent increases in Atlantic
hurricane activity (Goldenberg et al. 2001; Landsea
2005).
The recent warming trend in tropical North Atlantic
sea surface temperature (SST) is consistent with a coherent, North Atlantic basinwide SST warming, as can
be clearly seen in the North Atlantic basin average and
annual mean SST anomaly index (NASSTI) shown in
Fig. 1.1 The solid black line in this figure displays an
overall gradual warming culminating in a rapid upward
trend from 1975 to the present. The trend is obviously
not linear and includes an ‘‘oscillatory’’ component with
a relatively cold episode from 1900 to 1925, followed by
1
The annual mean NA SST anomaly in Fig. 1 is defined as deviations of the basinwide average SST from its long-term mean for
the entire twentieth century. It was also subjected to a recursive
Butterworth filter with a half power point at a period of 10 yr, so
that variability with time scales shorter than 10 yr has been
strongly reduced.
1470
JOURNAL OF CLIMATE
VOLUME 22
FIG. 1. NASSTI averaged over the ocean grids from the equator to 608N and from 7.58 to 758W. Black
solid line: observations; color lines: coupled ocean–atmosphere models of the IPCC twentieth-century
simulations averaged over multiple realizations starting from different initial conditions; dashed black line:
average of all models. The index is defined as the deviation from long-term climatological mean for the
entire twentieth century and the time series are subject to a low-pass tangent Butterworth filter with a 10-yr
cutoff.
a warmer period from 1930 to 1960, another relatively
cold period from 1970 to 1990, and finally the recent
rapid warming, all superimposed on the general rise of
temperatures. The combination of an upward trend plus
a multidecadal oscillatory component indicates the possible superposition of an externally forced component
and an internally generated one. The latter is consistent
with the North Atlantic multidecadal SST variation that
was identified in several previous studies (e.g., Kushnir
1994; Schlesinger and Ramankutty 1994; Enfield et al.
2001) and is commonly referred to as the Atlantic
Multidecadal Oscillation (AMO) after Kerr (2000).
The characteristics of the observed NASSTI trend in
Fig. 1 raise several questions. First, is it possible to
confirm that the observed temporal dependence results
from the superposition of significant internal variability
(AMO) on the response to external forcing? If so, what
is the best way to quantify the two components in observations? Second, what is the climatic impact of each
component in different regions of the world? Answers
to these questions are important for designing a useful
‘‘near term’’ climate prediction system to help plan and
prepare for climate change in the coming few decades.
Figure 2 shows the application of two of the previously proposed approaches designed to remove the
forced signal associated with both anthropogenic and
other natural (volcanic and solar) forcing from the total
observed NASSTI, with the purpose of uncovering the
internal component of the variability. The first commonly used method is to remove the linear trend from
the observed North Atlantic SST index, as shown in Fig.
2a (e.g., Enfield et al. 2001; Sutton and Hodson 2005;
Knight et al. 2006). This method assumes that the forced
trend is linear and uniform over time. The linear detrending method suggests that the positive anomaly in
NASSTI at the end of the twentieth century (0.48C) is
equally divided between the externally forced trend and
the internal AMO variability (amplitude 0.28C) and that
the latter is currently at a peak state, similar to its state
in the middle of the twentieth century. A second method
is to use the global mean sea surface temperature as a
proxy for the externally forced signal (Trenberth and
Shea 2006; Mann and Emanuel 2006). When subtracting
the global mean SST anomalies from the tropical North
Atlantic SST to remove the forced signal, Trenberth and
Shea (2006) concluded a predominant contribution from
the anthropogenically forced warming to the total North
Atlantic SST anomalies. In this study, we regress the twodimensional SST field on the time series of globally averaged SST (SSTg) and obtain an estimate of the internal
component as the local difference between the total field
and the regression pattern. The North Atlantic average
of both the regressed NASSTI and the residual is shown
in Fig. 2b. The regression method used here accounts for
the fact that the forced SST is not uniform spatially,
which differs from that used in Trenberth and Shea
(2006).
Comparing Figs. 2a and 2b, one sees that the two
methods imply considerable differences in the amplitude and temporal properties of the forced and internal
15 MARCH 2009
TING ET AL.
1471
FIG. 2. (a) The linear trend (solid black line) and detrended NASSTI (shaded). (b) NASSTI regressed
onto the global mean SST (SSTg regression, solid black line) and the difference between the observed
NASSTI shown in Fig. 1 and the SSTg regression (shaded). (c) NASSTI regressed onto the global mean
surface temperature (Tg regression, solid black line) and the difference between the observed NASSTI
shown in Fig. 1 and the Tg regression (shaded).
variability. Unlike linear detrending, regression on the
global mean SST implies that the positive NASSTI
anomaly at the end of the twentieth century is largely due
to the forced signal (;0.348C) and only a small portion is
caused by internal AMO variability (;0.068C), consistent with Trenberth and Shea (2006). Furthermore, although linear detrending might suggest that the AMO is
at its peak amplitude and that the internal variability in
the next 2 decades would stay at the same amplitude or
decrease, regression on the global mean SST suggests
that the internal component of the AMO could cause
even warmer north Atlantic SST in the coming years.
Another commonly used measure of the anthropogeni-
cally forced variability is the global mean surface temperature (Tg), as shown in Fig. 2c. This method suggests
an even weaker recent warming due to internal variability than when global mean SST is used, leaving the
externally forced signal to explain almost all of the observed change during the late twentieth century. In addition to the difference in relative contribution to forced
and internal components of NASSTI, the overall amplitude of the AMO is about 20% weaker using the global
mean SST and global mean surface temperature as a
proxy for forced trend. Given these differences, it is important to find an objective, quantitative way to measure
the realism of each method. We attempt this by making
1472
JOURNAL OF CLIMATE
use of both observations and model estimates of externally forced climate change.
2. Ratio of forced and internal variability in coupled
models
In this study, we separate the forced and internal
components of the North Atlantic decadal SST variability
by using all the Intergovernmental Panel on Climate
Change Fourth Assessment (IPCC AR4) twentiethcentury simulations that have multiple realizations with
a single model. The following six IPCC AR4 models all
have at least four realizations of the twentieth-century
simulations and are used in this study: the National
Center for Atmospheric Research Community Climate
System Model, version 3 (NCAR CCSM3) with eight,
the Geophysical Fluid Dynamics Laboratory Climate
Model version 2.1 (GFDL CM2.1) with five, the Goddard Institute for Space Studies Models E-H (GISS-EH)
with five and E-R (GISS-ER) with nine, the Meteorological Research Institute Coupled General Circulation Model version 2.3.2a (MRI CGCM2.3.2a) with
five, and the NCAR Parallel Climate Model (PCM)
with four realizations.
Shown in Fig. 1 in colored lines are the NASSTI for
the twentieth century from the six IPCC AR4 models
with the known and estimated forcing of the greenhouse
gas concentrations, aerosol, and natural solar and volcanic forcing prescribed. Each color line is the ensemble
average of, variously, four to nine realizations of the
twentieth century as simulated by each of the models.
The trend of increasing SST over the North Atlantic
basin in these models is similar to that observed (black
line), but the amplitude of the oscillatory component is
less than in observations. All of the models capture the
rapid increase in temperature in the recent 2 decades.
Because averaging over multiple realizations will tend
to isolate the forced signal and suppress internally generated variations that are uncorrelated between realizations,
this visual comparison suggests that observed, forced
North Atlantic SST change was embedded within a large
internal oscillation but that the recent warming was
largely externally forced. However, the relatively small
number of realizations in each of the model ensembles
makes it difficult to draw a firm conclusion on the relative
roles of internal and forced change in North Atlantic SST
variations of the twentieth century.
One way to quantify the relative contribution of externally forced variability to total variability is the analysis
of variance (ANOVA) method, which computes the ratio
of the forced variance and the total variance. This method
has been commonly used in separating the SST-forced
variability from the total variability in prescribed SST
VOLUME 22
experiments with multiple realizations (e.g., Harzallah
and Sadourny 1995). In this study, we extend this method
to estimate the internally and externally forced surface
temperature variances in each of the coupled ocean–
atmosphere models with prescribed anthropogenic, solar, and volcanic forcing.
If sI2 and sa2 represent the biased estimates of the
internal and ensemble averaged variances, respectively,
for the coupled ocean–atmosphere model, then
sI 2 5
1
MN
sa 2 5
1
M
å
å
m n
å
m
1
N
Tsmn
ån Tsmn
1
N
ån Tsmn
1
MN
2
å
å Tsmn
m n
2
,
where Tsmn is the surface temperature for year m and
ensemble member n, M is the total number of years,
and N is the total number of ensemble members. The
ratio of forced variance and total variance can be obtained as
sF 2
r5
5
sT 2
sF 2 5 s a 2
1
sI 2
(N 1)
,
sT 2
sa 2
1
(N
1)
with
sI 2
sT 2 5 s I 2 1 sF 2 .
The second term in the forced variance estimate (sF2)
removes the effect of internal variability contained in
the ensemble average variance due to the small ensemble
size. For a relatively large ensemble (say, N . 20), sF2
should be well approximated by sa2. In a relatively small
ensemble and in the presence of large internal variability,
sF2 can be overcorrected and even result in negative
value in some areas.
Figure 3 shows the variance ratio for decadal time
scale variations averaged across the six models (variances are computed after subjecting the data to Butterworth filter). Most of the tropics (308S–308N) show
that forced variance can account for 70% or more of the
total variance. The largest ratio is found over Indian
Ocean, indicating that decadal changes in the Indian
Ocean SST are largely a response to external (radiative)
forcing (Hurrel et al. 2004; Knutson et al. 1999). Over
the eastern tropical Pacific, there is a local minimum in
the forced variance ratio, suggesting that the model internal variability associated with tropical SST variability
on decadal time scales is significant. In the extratropics
and over land, forced variance accounts for as much as
15 MARCH 2009
TING ET AL.
1473
FIG. 3. Ratio of externally forced variance and the total variance averaged for the six IPCC AR4 coupled
models with at least four ensemble members for the twentieth century. See details in the text.
50% of the total surface temperature variance. The
lowest ratio is found over the extratropical North Atlantic, with ratio as low as 2%–10%, indicating the existence of strong internally forced decadal SST variations there. Small variance ratios are also found over
other extratropical ocean regions, between 308 and 608N
in the Pacific, and in the Southern Ocean near 608S. The
variance ratios may not be well represented over the
extratropical oceans because of the large internal variability there and the small number of ensemble members available. Figure 3 does, however, indicate clearly
that internal variability is important over the extratropical oceans, in particular the North Atlantic, which
is the subject of our investigation (note that the effect of
the North Atlantic is felt into Europe as well).
To get the ratio of variance for the North Atlantic
SST index, we repeated the same variance ratio calculation with the North Atlantic basin-averaged SST (averaged over the entire North Atlantic from 08 to 608N).
The ratio varies from 34% to 87% depending on the
model, indicating that although external forcing is responsible for a large portion of spatially coherent, decadal surface temperature variations there, nonetheless
internal variability is significant. For a North Pacific basin
average, the variance ratio ranges from 70% to 91%
depending on the model, whereas for the tropical SST
between 308S and 308N, the numbers are above 93%. In
the next section, a quantitative method will be used to
extract the forced variability from the total and determine its spatial and temporal pattern and, as a residual,
the dominant pattern of internal variability in the Atlantic Basin.
3. Forced and internal variability using
signal-to-noise maximizing EOF analysis
A rigorous technique to define the forced variability,
given multiple realizations of the coupled model simulations, is the signal-to-noise maximizing EOF analysis
(Allen and Smith 1997; Venzke et al. 1999; Chang et al.
2000). The method applies a spatial prewhitening transformation to the model output, which removes the spatial
correlations in the internal atmospheric variability (i.e.,
‘‘climate noise’’) contained in the ensemble average.
Thus, the spatial covariance in the ensemble average is
purely due to the forced responses. Figure 4 shows the
multimodel average of the spatial pattern (Fig. 4a) and
the corresponding individual-model principal components (PCs) for the dominant mode (Fig. 4b) of the
signal-to-noise (S/N) maximizing EOF analysis, using
the six models listed above (section 2). This first EOF
explains 55%–72% of the total model variance except
the GISS-EH model, which only explains 37%. However,
the second mode explains only 3%–6% in all models.
This indicates that on decadal time scales, the externally
forced variability can be represented rather decisively
by a single, globally synchronous pattern. The spatial
structure of the first mode is rather similar from model to
model (not shown). Averaged over all models, it displays
a largely global warming over both land and ocean areas.
Note, however, that the pattern exhibits considerable
spatial variation and even several patches of cooling over
the North Atlantic, North Pacific, and in the Southern
Hemisphere near 608S. These variations are likely caused
by such factors as local ocean dynamics and/or the uneven
1474
JOURNAL OF CLIMATE
VOLUME 22
FIG. 4. (top) The multimodel mean spatial structure of the first mode of the S/N-maximizing
EOF analysis averaged over the six IPCC AR4 models used in this study. Shown are regressions of
annual-mean, low-pass-filtered surface temperature on the S/N-maximizing PC1. (bottom) S/Nmaximizing PC1 for each of the six models. The colored lines are for the individual coupled ocean–
atmosphere models; the dashed black line shows the six-model average PC1; the solid black line is
the standardized global mean surface temperature from the GISS surface temperature dataset.
distribution of clouds and aerosol effects. In particular, a
similar S/N-maximizing EOF analysis applied to model
integrations with 1% year21 CO2 concentration increase
but without any aerosol forcing (not shown) displays no
negative centers over the North Pacific but does show a
similar cooling over extratropical North Atlantic, indicating that the North Pacific cooling in Fig. 4a might be
due to aerosol forcing, whereas the North Atlantic cooling is likely a result of the combined effect of aerosol
forcing and internal ocean dynamics.
The first principal component (PC1) of each of the
models (Fig. 4b) show a similar temporal trend with almost linear increases from the beginning of the twentieth
century to 1960 and a small dip in the sixties followed by
a sharper increase from the 1970s to the present. The
black dashed line in Fig. 4b is the average PC for the six
models and the solid black line is the standardized,
global-mean observed surface temperature (air temperature over land and SST over the oceans). The
similarity among the model PCs and between model
15 MARCH 2009
1475
TING ET AL.
TABLE 1. Correlation coefficient between S/N-maximizing PC1 and ensemble average global mean surface temperature index (Tg),
global mean sea surface temperature index (SSTg), global mean surface temperature over land (TLg), Indian Ocean SST index (IOSST),
and the North Atlantic SST index (NASSTI) for each of the participating models and the observations.
Tg
SSTg
TLg
IOSST
NASSTI
NCAR
GFDL
GISS-EH
GISS-ER
MRI
PCM
Model ensemble ranges
Observed
0.996
0.998
0.996
0.987
0.927
0.983
0.987
0.982
0.993
0.927
0.977
0.981
0.962
0.965
0.680
0.997
0.994
0.991
0.995
0.914
0.998
0.997
0.993
0.987
0.989
0.991
0.986
0.973
0.974
0.851
0.840–0.988
0.921–0.988
0.816–0.988
0.916–0.979
20.07–0.959
0.945
0.926
0.897
0.881
0.658
and observations in Fig. 4b indicates that the method
effectively isolates the global warming signal; although
the global mean surface temperature is a good approximation for the observed forced signal, the AMO signature is discernable in the global mean surface temperature (e.g., the peak in the 40s and the dip in the 70s).
To quantify the extent to which the S/N-maximizing
PC1 represents the ensemble mean variability of each
model’s surface temperature averaged over different
regions of the globe, we computed its correlation coefficient with the following indices: global mean surface
temperature, global mean SST, global mean surface
temperature over land (TLg), North Atlantic SST, and
Indian Ocean SST (IOSST) in Table 1. The Indian
Ocean SST is included here because of the uniformly
large forced variance ratio across all models in Fig. 3.
Clearly, the S/N-maximizing PC1 is highly correlated
with the global mean SST and global mean surface
temperature in all models. The global mean land surface
temperature and the Indian Ocean SST are slightly less
well correlated with PC1 but still have correlations
above 0.96 for all models. These high correlations are
evidence of the global nature of the first EOF, as shown
in Fig. 4a. Also shown in Table 1 are the correlations
between PC1 and the North Atlantic SST. As expected
(from Fig. 3), these correlations are lower than that with
the other indices, consistent with the notion that there
are large internally generated multidecadal variations in
this region, and ensemble averaging with limited ensemble size cannot effectively remove all the internal
variability. Two of the models (GISS-EH and NCAR
PCM, which contain five and four ensemble members
respectively) show particularly low correlations for the
NASSTI index (0.68 and 0.85). It is interesting to compare the corresponding correlations for models that are
close to these two in configurations—the GISS_ER (0.91)
and NCAR CCSM (0.93), which have nine and eight
ensemble members respectively—thus confirming the
importance of large ensemble size in removing the
internal variability.
The last column in Table 1 shows the correlation
between the same SST indices as derived from obser-
vations and the model-averaged PC1. These correlations are expectedly lower than the corresponding ones
for the models, likely because there is only one realization for the observations compared to at least four
independent realizations for each of the IPCC models.
It is also possible that the lower correlations are due to
the inconsistency between forced change in the models
and that in observations. The eighth column in Table
1 shows the range of correlation values between each of
the indices discussed above and the models’ PC1 using a
single ensemble member instead of the ensemble mean.
With the exception of Indian Ocean SST index, the
correlations for observed SST indices (last column) are
always within the range of values corresponding to
a single model realization. The correlations between
model-average PC1 and the observed SST indices decrease in the same way as in models, with the highest
correlations obtained for global mean surface temperature and global mean SST. It illustrates that the separation of forced and natural North Atlantic SST using
the global mean surface temperature and global mean
SST, as shown in Fig. 2, is a good approximation for
deriving the true forced variability. The observed North
Atlantic SST index correlates much less well with PC1
compared to the other indices in Table 1, indicating a
strong influence of internal variability there; this will be
explored further below.
Next we projected each model’s NASSTI onto the
corresponding PC1 and defined that as the forced contribution to the secular change in the basin. A similar
approach has been taken by Kravtsov and Spannagle
(2008), who used multimodel average regional surface
temperature as an indication for the externally forced
signal. Figure 5a shows the forced trend of NASSTI so
defined (color lines) along with the observed NASSTI
(black line). Although there is a large spread for the
forced NASSTI trends among the six models, it is clear
that the observed NASSTI oscillates outside of the
uncertainties of the model forced trend, consistent with
the large internal component of the NASSTI in observations. One notices that the spread among the six
models’ forced NASSTI is larger at the beginning and
1476
JOURNAL OF CLIMATE
VOLUME 22
FIG. 5. (top) Projection of NASSTI onto the S/N-maximizing PC1 in each of the participating models
(ensemble averaged, color lines) and observed counterpart (black line). (bottom) Observed internally
generated AMO index constructed by subtracting from the observed index the model estimates of the
forced NA SST shown in the top panel. The black dashed line in the bottom panel is the average across all
six models.
the end of the twentieth century and much smaller in the
middle of the century. This occurs because each model
predicts a slightly different rate of North Atlantic SST
increase. For example, the GISS-EH model has a much
smaller rate of increase compared to that of MRI. Because the mean over the entire record was removed
from each model’s output, the plotted time series fan
out at the ends. The Atlantic forced trends also show a
larger spread among different models compared to that
for the global mean SST and the Indian Ocean SST (not
shown), but they are comparable to that for the North
Pacific (not shown). The larger spread for the North
Atlantic and North Pacific may reflect the uncertainties
in model estimates of forced trends over the ocean basins where internal variability is large (Fig. 3). Given the
uncertainties, we note that the observed temperature
increases from the 1920s to the 1940s and during the
most recent decade, as well as the cooling trend between
1960 and the mid-1970s, are larger than any of the
forced trends in the models. Thus, Fig. 5a indicates
clearly that the observed decadal variations in NASSTI
cannot be explained by the response to external forcing
15 MARCH 2009
TING ET AL.
alone. An internal oscillatory component must be part
of the North Atlantic SST variability.
To remove the model-based estimate of the forced
change from the observed North Atlantic SST record,
we averaged the six models’ forced changes (the black
dashed line in Fig. 5a) and subtracted it from the observed time series. The uncertainty in this estimate is
represented by the spread generated when each model’s
forced component is separately removed from the data
(see Fig. 5b). The amplitude of the oscillation, to which
we hereafter refer to as AMO, is between 20.38 and
10.28C, which is comparable to the detrended NASSTI
in Fig. 2a but larger than those in Figs. 2c and 2e. In
terms of the phase of the oscillation, Fig. 5b indicates
that the AMO so defined is similar to that using the global
mean surface temperature or global mean sea surface
temperature as the forced signal (and shown in Fig. 2). In
all of these definitions, the AMO crosses to the positive
phase near the end of the twentieth century. Although
the global averages pinpoint the inflections of the AMO
well, they underestimate the amplitude by about 20%.
To relate the results to multidecadal Atlantic hurricane intensity variations, it is important to examine the
twentieth-century tropical North Atlantic SST variations in the so-called hurricane main development regions (MDRs) as defined in earlier studies (Goldenberg
and Shapiro 1996; Goldenberg et al. 2001; Emanuel
2005). Figure 3 indicates that the forced variance over
the tropical North Atlantic accounts for a much larger
portion of the total variance compared to the northern
North Atlantic in all models. This is broadly consistent
with Mann and Emanuel (2006), who argued that there
is no internal oscillation in MDR SST variability after
removing the forced signal due to both greenhouse
warming and aerosol effect. The observed situation is
depicted in Fig. 6a, which shows the projections of
MDR-averaged SST for August–October (ASO) onto
the S/N-maximizing PC1 for each model (colored lines),
the model average (dashed black line), and the observations (solid black line). Figure 6a shows that the recent warming of the MDR is mainly due to external
forcing and is similar in range to that of the North Atlantic basin-wide averages. Santer et al. (2006) analyzed
the twentieth-century SST trend in the Atlantic and
Pacific tropical cyclone regions and concluded that the
twentieth-century trend cannot be explained by internal
variability alone and that a larger share of the variability
is explained by external forcing. Viewed over the entire
century, our findings are consistent with their results.
However, an examination of Fig. 6b, showing the difference between the observed MDR SST and the forced
component as represented by the color lines in Fig. 6a,
indicates that even in the MDR the contribution of in-
1477
ternal variability is important. The latter is responsible
for the sharp 1930s temperature rise and the 1970s
temperature drop in the tropical North Atlantic, consistent with arguments made by, for example, Goldenberg
et al. (2001). However, as far as the temperature rise in
the last 2 decades or so is concerned, the contribution
due to internal variability is not negligible, but it is not
the dominant contributor. We stress that the MDR SST
variability does not directly infer Atlantic hurricane activities and the number of landfalling hurricanes. Other
factors such as vertical wind shear and static instability
are known to play important roles in hurricane variability, which is not included in the discussion of this paper
but may be associated with the Atlantic SST fluctuations
(Wang et al. 2008).
4. Climate impacts of forced and natural North
Atlantic SST variability
Recent studies (Enfield et al. 2001; McCabe et al.
2004; Sutton and Hodson 2005) have emphasized the
impact of the AMO on North American and European
precipitation. These studies defined the AMO as the
low-pass, linearly detrended, North Atlantic–averaged
SST anomaly. We have argued in this study that the
AMO should be defined differently and that the resulting AMO phase changes then differ from the linearly detrended result. In the following, we contrast the
climatic impacts of the externally forced North Atlantic
SST trend and the AMO as defined in this study. To that
end, we computed the regression between the observed
global surface temperature and land precipitation with
the forced NASSTI as defined by the multimodel average projection of the model’s NASSTI onto S/Nmaximizing PC1 (black dashed line in Fig. 5a) and the
AMO (black line in Fig. 5b). A Monte Carlo statistical
significance test2 is applied to these regressions and only
those with regression values significant at the 5% level
are shown in Fig. 7.
Figure 7a is the externally forced ‘‘global warming’’
pattern for the twentieth century. By definition this is
the same pattern that would emerge if the global field
were regressed on the multimodel average of PC1. It is
interesting to contrast this pattern with Fig. 4a, which
displays the multimodel averaged depiction of the
2
In the Monte Carle significance test here, the index time series
is first randomized in temporal ordering and then applied the same
recursive Butterworth filter before computing regression with
precipitation and surface temperature. The regressions that are at
or above the 97.5% or at or below the 2.5% level of all the randomized regressions were shown at the 5% statistical significance
level.
1478
JOURNAL OF CLIMATE
VOLUME 22
FIG. 6. Same as Fig. 5, but for MDR SST averaged over the 3-month hurricane season, August–October.
twentieth-century warming. As in the models, the observations displays a significant warming trend almost
everywhere except the Southern Greenland coast in the
North Atlantic, where a significant cooling occurred.
However, the intensity of the observed warming over
the tropical oceans is quite different from the simulated
one. In particular, tropical Pacific Ocean SSTs have not
warmed as much as in the model simulations. This may
be due to the fact that there is considerable internal
variation of tropical Pacific SSTs on decadal time scales
and that the single observed ‘‘realization’’ of this variability masks the forced signal. It is also possible that
response mechanisms not captured by the model reduce
the impact of radiative forcing in this region (Cane et al.
1997). The influence of the tropical Pacific on the global
climate is significant and hence it is important to investigate the ramification and causes of these differences, but this is beyond the scope of this paper.
For global land precipitation (Fig. 7c), however, the
regression with the forced component is less significant
than that for temperature. There are some indications of
increased precipitation over northern high latitudes, but
there is very little significant signal over the tropical and
subtropical latitudes. A very slight hint of a drying trend
over the western Sahel and the Mediterranean region can
be noticed in Fig. 7c. Comparison with the same regression but using the coupled models’ twentieth-century
simulations (not shown) indicates a much stronger
15 MARCH 2009
TING ET AL.
1479
FIG. 7. Spatial pattern of regression between forced NA SST and the global surface (a) temperature and (c) precipitation, and between
the naturally oscillating NA SST (AMO) and the global surface (b) temperature and (d) precipitation for the twentieth century. The land
surface temperature and precipitation data are taken from CRU’s 0.58 3 0.58 version and the SST over ocean is taken from GISS. Only
the regression values that exceed the 5% statistical significance level based on Monte Carlo method (details in text) is plotted. Units are
degrees Celsius per standard deviation of SST index for surface temperature and millimeters per month per standard deviation of SST
index for precipitation.
subtropical drying, including Southern Mexico, the
Caribbean, the Mediterranean, and the Sahel region
(see Solomon et al. 2007). This apparent difference
between the model and observations may be explained
by the stronger tropical warming in coupled models,
particularly in the Pacific, as discussed above. As shown
in previous studies (e.g., Yin 2005), one of the robust
responses of the atmospheric circulation to greenhouse
forcing is northward-shifted storm tracks, which enhance precipitation in the high latitudes and drying in
the subtropics (see also Held and Soden 2006). Figure 7
suggests that this mechanism may not be as advanced in
reality as predicted in the coupled models, perhaps because of the lesser warming of the Pacific Ocean in
observations compared to the model simulations. Further analysis is needed to confirm such connections.
The internally generated, AMO-related patterns in
temperature and precipitation are generally consistent
with the findings of previous studies. The temperature
pattern (Fig. 7b) is characterized by basinwide warming
over the North Atlantic and its surrounding regions. For
precipitation (Fig. 7d), the most dominant feature is the
positive anomaly over the Sahel associated with the
warming phase of the AMO (Knight et al. 2006; Zhang
and Delworth 2006), opposite to that associated with
the externally forced warming. This is not surprising
considering that the Atlantic SST patterns associated
with external forcing (Fig. 7a) and the AMO (Fig. 7b)
imply different polarities of the Atlantic interhemispheric SST gradient in the warm phase of the AMO
and during global warming. The interhemispheric gradient is a key factor in determining the seasonal position
of the Atlantic intertropical convergence zone (ITCZ),
which governs rainfall over tropical Africa. Thus, when
a global warming trend occurs with a cooling trend of
the AMO, one would expect the Sahel to experience
extreme drying conditions, such as was the case in 1960–
70. Other features of the AMO-related precipitation
anomalies are less significant, indicating drying of parts
of North and South America and Australia and an enhanced Indian monsoon and rainfall over northern Asia.
Another interesting finding in Fig. 7 is the opposite
impact of a warm AMO and the externally forced
warming trend on South Greenland temperature. Although the externally forced trend is negative along the
South Greenland coast, the AMO warming trend there is
positive. This is consistent with the simulation of models
examined in this study (not shown). A recent study
1480
JOURNAL OF CLIMATE
(Chylek et al. 2006) found that the warming of coastal
Greenland during 1920–30 is much stronger than the
recent warming trend, consistent with Fig. 7. Specifically, in 1920–30, the externally forced trend was weak
and the AMO trend dominated. One should be aware
that the temperature trend over central Greenland
where the ice sheet lies is positive in both the forced and
the internal component; thus, both contribute to a positive trend of surface temperature there in the recent
decade. The recent rapid decrease in Greenland ice sheet
volume (Serreze and Francis 2006) may be due to the
additive effect of the forced warming trend and the
transition to a positive phase in the internal component
of the AMO. Given the short observational records, there
may be overfitting in the regression analysis. Similar
analysis based on model simulations is needed to confirm the observational relationship.
5. Summary
Using six of the IPCC AR4 twentieth-century simulations with multiple ensemble members, we are able to
effectively separate the externally forced component
of North Atlantic Ocean SST variations and the internally varying component using signal-to-noise (S/N)maximizing EOF analysis. We further show that the
observed North Atlantic SST variability is well outside
of the uncertainty of the model-simulated forced trend,
indicating the existence of an internal component in
observations. The S/N-maximizing PC1 is found to be
highly correlated with the ensemble mean and globally
averaged SST and surface air temperature, indicating
that these indices can be used, alternatively, to represent forced variations in the models and in observations.
Taking the model-averaged S/N PC1 as the forced
North Atlantic SST trend, we found that the internal
variability of North Atlantic SST, or the AMO, was at
an upswing tendency at the end of the twentieth century
and thus in phase with the forced tendency. If the AMO
trend continues in this direction, the North Atlantic will
experience much faster warming in the coming years
than the rest of the world. It is further shown that the
hurricane main development region SSTs for August,
September, and October show a dominance of external
forcing in the recent warming trend. However, the
earlier warming trend in the 1930s and the cooling trend
in the 1970s were connected mainly to internal variability.
The spatial pattern of surface temperature and precipitation associated with the externally forced trend
and the AMO are examined using regression analysis.
The observed externally forced anomalies show a global
warming trend everywhere except the northern North
Atlantic. However, this pattern differs from the mod-
VOLUME 22
eled pattern in its details, particularly over the tropical
Pacific. There are indications in the model simulations
that the cooling trend over the northern North Pacific
may be related to aerosol forcing. Because most of the
IPCC AR4 model projections show a 25% slowing of
the Atlantic meridional overturning circulation over the
twenty-first century (Solomon et al. 2007), the strong
cooling over the North Atlantic may be a result of the
slowdown of ocean circulation. Consistent with previous
observational studies on the impact of the AMO on
precipitation around the Atlantic basin, we find that the
largest impact is over the Sahel region, where the
warming trend of the AMO is associated with increased
precipitation. The Sahel drought of the 1970s and 1980s
is associated with a cooling trend in the AMO. There
are also indications of drought conditions associated
with a warming trend in the AMO over South America
and Australia.
The results presented here do not lead to dramatically
different conclusions from the earlier studies dealing with
the same issue. We believe, however, that our rigorous
statistical analysis puts the claim that the North Atlantic
displayed in the twentieth century an internal ‘‘oscillation’’ of considerable magnitude (compared to overall
externally forced trend) on a more robust footing. We
were also able to show that this internal variation led
to sharp decadal changes in temperature, but due to
its oscillatory nature these transitions led to an overall
compensation on a century time scale. Moreover, we
also argue that the globally averaged surface temperature appears to be a good proxy for the temporal march
of externally forced variability and that most of the latter
is globally synchronous, albeit nonuniform spatially. The
smoothed, local expression of externally forced variability can therefore be represented by the regression of
local variables such as temperature and precipitation on
the smoothed time series of global mean surface temperature. This expression should obviously be quantified by an error estimate based on standard approaches
to error analysis, either parametric or nonparametric.
Our results point at the importance of internal variability, specifically the AMO, in determining future
changes in climate. The AMO may continue its upward
trend and contribute to positive North Atlantic SST
anomalies in the near future, given its past temporal
evolution and the fact that it crosses to the positive phase
at the end of the twentieth century. Assuming a linear
superposition of the forced and natural responses, the
North Atlantic may experience unprecedented warming
in the next decade or so when combined with a strong
externally forced, anthropogenic global warming trend.
However, if the AMO trend is reversed in the coming
years (N. Keenlyside et al. 2008, personal communication),
15 MARCH 2009
TING ET AL.
the warming of most of the North Atlantic will lag other
regions and this will similarly influence northeastern
North America, western Europe, and the Mediterranean.
The accurate prediction of the AMO phase transition is
thus important for the future, near-term climate change
prediction. Determining the nature and realism of the
AMO in coupled ocean–atmosphere models is an important next step leading to a better understanding of the
AMO dynamics and its predictability.
Acknowledgments. We thank Dr. Isaac Held for many
useful discussions and suggestions for this work. Thanks
are also due to Drs. Michela Biasutti, Mark Cane, Tom
Delworth, Alessandra Giannini, and Lisa Goddard for
helpful discussions and to two anonymous reviewers for
their many constructive comments. The work was supported by NOAA Grant NA030AR4320179 on abrupt
climate change.
REFERENCES
Allen, M. R., and L. A. Smith, 1997: Optimal filtering in singular
spectrum analysis. Phys. Lett., 234, 419–428.
Cane, M. A., A. C. Clement, A. Kaplan, Y. Kushnir, D. Pozdnyakov,
R. Seager, S. E. Zebiak, and R. Murtugudde, 1997: Twentiethcentury sea surface temperature trends. Science, 275, 957–960.
Chang, P., R. Saravanan, L. Ji, and G. C. Hegerl, 2000: The effect
of local sea surface temperatures on atmospheric circulation
over the tropical Atlantic sector. J. Climate, 13, 2195–2216.
Chylek, P., M. K. Dubey, and G. Lesins, 2006: Greenland warming
of 1920–1930 and 1995–2005. Geophys. Res. Lett., 33, L11707,
doi:10.1029/2006GL026510.
Emanuel, K. A., 2005: Increasing destructiveness of tropical cyclones over the past 30 years. Nature, 436, 686–688.
Enfield, D. B., A. M. Mestas-Nuñez, and P. J. Trimble, 2001: The
Atlantic multidecadal oscillation and its relation to rainfall
and river flows in the continental U.S. Geophys. Res. Lett., 28,
2077–2080.
Goldenberg, S. B., and L. J. Shapiro, 1996: Physical mechanisms
for the association of El Niño and West African rainfall with
Atlantic major hurricane activity. J. Climate, 9, 1169–1187.
——, C. Landsea, A. M. Mestas-Nuñez, and W. M. Gray, 2001: The
recent increase in Atlantic hurricane activity: Causes and
implications. Science, 293, 474–479.
Harzallah, A., and R. Sadourny, 1995: Internal versus SST-forced
atmospheric variability as simulated by an atmospheric general circulation model. J. Climate, 8, 474–495.
Held, I. M., and B. Soden, 2006: Robust responses of the hydrological cycle to global warming. J. Climate, 19, 5686–5699.
Hurrel, J. W., M. P. Hoerling, A. S. Phillips, and T. Xu, 2004:
Twentieth century North Atlantic climate change. Part I:
Assessing determinism. Climate Dyn., 23, 371–389.
View publication stats
1481
Kerr, R. A., 2000: A North Atlantic climate pacemaker for the
centuries. Science, 288, 1984–1986.
Knight, J. R., C. K. Folland, and A. A. Scaife, 2006: Climate impacts of the Atlantic Multidecadal Oscillation. Geophys. Res.
Lett., 33, L17706, doi:10.1029/2006GL026242.
Knutson, T. R., T. L. Delworth, K. W. Dixon, and R. J. Stouffer,
1999: Model assessment of regional surface temperature
trends (1949–1997). J. Geophys. Res., 104, 30 981–30 996.
Kravtsov, S., and C. Spannagle, 2008: Multidecadal climate variability in observed and modeled surface temperatures. J.
Climate, 21, 1104–1121.
Kushnir, Y., 1994: Interdecadal variations in North Atlantic sea
surface temperature and associated atmospheric conditions. J.
Climate, 7, 141–157.
Landsea, C. W., 2005: Hurricanes and global warming. Nature, 438,
E11–E12, doi:10.1038/nature04477.
Mann, M. E., and K. Emanuel, 2006: Atlantic hurricane trends
linked to climate change. Eos, Trans. Amer. Geophys. Union,
87, 233–244.
McCabe, G. J., M. A. Palecki, and J. L. Betancourt, 2004: Pacific
and Atlantic Ocean influences on multi-decadal drought frequency in the United States. Proc. Natl. Acad. Sci. USA, 101,
4136–4141, doi:10.1073/pnas.0306738101.
Santer, B. D., and Coauthors, 2006: Forced and unforced ocean
temperature changes in Atlantic and Pacific tropical cyclogenesis regions. Proc. Natl. Acad. Sci. USA, 103, 13 905–13 910.
Schlesinger, M. E., and N. Ramankutty, 1994: An oscillation in
the global climate system of period 65–70 years. Nature, 367,
723–726.
Serreze, M. C., and J. A. Francis, 2006: The Arctic on the fast track
of change. Weather, 61, 65–69.
Solomon, S., D. Qin, M. Manning, M. Marquis, K. Averyt, M. M.
B. Tignor, H. L. Miller Jr., and Z. Chen, Eds., 2007: Climate
Change 2007: The Physical Science Basis. Cambridge University Press, 996 pp.
Sutton, R. T., and D. L. R. Hodson, 2005: Atlantic Ocean forcing
of North American and European summer climate. Science,
309, 115–118.
Trenberth, K. E., and D. J. Shea, 2006: Atlantic hurricanes and
natural variability in 2005. Geophys. Res. Lett., 33, L12704,
doi:10.1029/2006GL026894.
Venzke, S., M. R. Allen, R. T. Sutton, and D. P. Rowell, 1999: The
atmospheric response over the North Atlantic to decadal
changes in sea surface temperature. J. Climate, 12, 2562–2584.
Wang, C., S.-K. Lee, and D. B. Enfield, 2008: Climate response to
anomalously large and small Atlantic warm pools during the
summer. J. Climate, 21, 2437–2450.
Webster, P. J., G. J. Holland, J. A. Curry, and H.-R. Chang, 2005:
Changes in tropical cyclone number, duration, and intensity in
a warming environment. Science, 309, 1844–1846.
Yin, J. H., 2005: A consistent poleward shift of the storm tracks in
simulations of 21st century climate. Geophys. Res. Lett., 32,
L18701, doi:10.1029/2005GL023684.
Zhang, R., and T. Delworth, 2006: Impact of Atlantic multidecadal
oscillations on India/Sahel rainfall and Atlantic hurricanes.
Geophys. Res. Lett., 33, L17712, doi:10.1029/2006GL026267.