GEOPHYSICAL RESEARCH LETTERS, VOL. 28, NO. 10, PAGES 2077-2080, MAY 15, 2001
The Atlantic multidecadal oscillation and its relation to
rainfall and river flows in the continental U.S.
David B. Enfield
NOAA Atlantic Oceanographic and Meteorological Laboratory, Miami, Florida.
Alberto M. Mestas-Nu˜nez
Cooperative Institute for Marine and Atmospheric Studies, University of Miami, Miami, Florida.
Paul J. Trimble
South Florida Water Management District, West Palm Beach, Florida.
Abstract. North Atlantic sea surface temperatures for
1856-1999 contain a 65-80 year cycle with a 0.4
◦
C range,
referred to as the Atlantic Multidecadal Oscillation (AMO)
by Kerr [2000]. AMO warm phases occurred during 1860-
1880 and 1940-1960, and cool phases during 1905-1925 and
1970-1990. The signal is global in scope, with a posi-
tively correlated co-oscillation in parts of the North Pa-
cific, but it is most intense in the North Atlantic and cov-
ers the entire basin there. During AMO warmings most of
the United States sees less than normal rainfall, including
Midwest droughts in the 1930s and 1950s. Between AMO
warm and cool phases, Mississippi River outflow varies by
10% while the inflow to Lake Okeechobee, Florida varies by
40%. The geographical pattern of variability is influenced
mainly by changes in summer rainfall. The winter patterns
of interannual rainfall variability associated with El Ni˜no-
Southern Oscillation are also significantly changed between
AMO phases.
Introduction
Using a singular spectrum analysis on global surface
temperature records since the 1850s, Schlesinger and Ra-
mankutty [1994] identified a North Atlantic surface tem-
perature oscillation with a period of 65-70 years and sug-
gested that it arises from internal ocean-atmosphere vari-
ability. Andronova and Schlesinger [2000] conducted simu-
lations of the observed global temperature using six models
with varying combinations of external forcings due to an-
thropogenic (greenhouse) and solar variabilities plus injec-
tions of volcanic aerosols. The external forcings account for
the nonlinear secular increase in temperatures but fail to re-
produce the previously identified 65-70 year cycles that are
manifested in global temperature data. The residual oscilla-
tion is likely a natural cycle mediated by ocean-atmosphere
interactions that can’t be reproduced by the simple cli-
mate/ocean model. Similar oscillations in a 60-110 year
band are seen in paleoclimatic North Atlantic climate re-
constructions dating at least to 1650 A.D. [e.g., Delworth
and Mann, 2000]. In two independent, naturally forced in-
tegrations of the GFDL coupled ocean-atmosphere model,
Copyright 2001 by the American Geophysical Union.
Paper number 2000GL012745.
0094-8276/01/2000GL012745$05.00
Delworth and Mann [2000] have reproduced the observed
multidecadal patterns of variability. They demonstrate that
in both the model and observations SST appears to carry the
multidecadal signal and that the model evolution involves
fluctuations in the intensity of the Atlantic thermohaline
circulation. Consistent with this, Venegas and Mysak [2000]
find a multidecadal mode of variability between observed sea
ice concentration in the Greenland Sea and sea level pressure
over high northern latitudes that is more or less synchronous
with the AMO variability in SST.
Partly to distinguish it from wide-band variability as-
sociated with the atmospheric North Atlantic Oscillation
(NAO), the long time scale oceanic phenomenon has recently
been referred to as the Atlantic Multidecadal Oscillation
(AMO) [Kerr, 2000]. While anthropogenic factors appear
to have become dominant in the late 20th century, the os-
tensibly natural temperature swings of the AMO have alter-
nately disguised and accentuated the secular trend. Consid-
erable importance is now being placed on understanding and
predicting this natural cycle so that it may be correctly ac-
counted for in ongoing evolution assessments of greenhouse
warming. It is also important to understand the effects of
the AMO on the intensity and geographic coverage of in-
terannual impacts such as those of El Ni˜no-Southern Oscil-
lation (ENSO). In this study we examine both the multi-
decadal and interannual behaviors of precipitation over the
continental U.S. as they relate to the alternating phases of
the oceanic AMO.
Data and methods
Our study is based on three data sets: an updated (1856-
1999) version of the Kaplan et al. [1998] monthly reanalysis
of global SST anomalies (SSTA), monthly rainfall over the
continental United States summarized by climate divisions
(1895-1999) [National Climatic Data Center], and (as inde-
pendent hydrological checks) the records of Mississippi River
outflow and the indirectly estimated inflow into Florida’s
Lake Okeechobee. These are compared to the appropriate
area-weighted rainfall accumulations over the corresponding
catchments. Because net runoff goes as the difference be-
tween rainfall and evapotranspiration (unavailable for our
analysis) the comparisons with river flows were done by
rescaling the basin rainfall totals to the variance and mean
of the river flow data. We focus on the multidecadal charac-
ter of the data sets by applying a ten-year running mean to
2077
2078 ENFIELD ET AL.: MULTIDECADAL ATLANTIC SST VARIABILITY AND U.S. RAINFALL
1860 1880 1900 1920 1940 1960 1980 2000
-0.2
-0.1
0
0.1
0.2
0.3
AMO ==> mean N. Atlantic SSTA
a
0E 40 80 120 160 160 120 80 40 0W
40S
0
40N
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8
Corr(AMO,Rain)
c
130 110 90 70W
20
30
40
50
82W 80W
26W
28W
Div-4
Figure 1. (a) AMO index: the ten-year running mean of de-
trended Atlantic SSTA north of the equator. (b) Correlation of
the AMO index with gridded SSTA over the world ocean (all sea-
sons). The thick contour is zero and thin contours denote the
95% significance level. (c) Correlation of the AMO index with
climate division rainfall with the Mississippi basin highlighted by
light gray fill. The larger highlighted circles indicate correlations
above the 90% significance level. Inset diagram to the right is a
blow-up of Florida showing Lake Okeechobee and Florida climate
division 4. The colorbar applies to correlations in both panels.
linearly detrended time series of all the data. Quantitative
comparisons between SSTA and rainfall-related variables are
made using conventional linear correlation analysis. Due to
the high degree of serial correlation in the smoothed time
series, a specially designed Monte Carlo analysis based on
the randomization of phases in the frequency domain was
used to determine the significance of correlations [Ebisuzaki,
1997].
1900 1920 1940 1960 1980 2000
-10
-5
0
5
10
% of Annual Outflow
Mississippi River Outflow
Mississippi Basin Rainfall
a
1900 1920 1940 1960 1980 2000
-20
0
20
40
% of Annual Inflow
b
L.Okeechobee Inflow
Florida Div-4 Rainfall
Figure 2. (a) Ten-year running means (all seasons) of Missis-
sippi River outflow (heavy, solid) expressed as a percentage of
the long term annual mean, and the area-weighted Mississippi
basin rainfall (shaded departures), rescaled to the outflow. (b)
As above but for Lake Okeechobee inflow and Florida division 4
rainfall.
1900 1920 1940 1960 1980 2000
-1
-0.5
0
0.5
1
Residual NINO 3.4 (DJF) vs. Okeechobee Rain (JFM)
Correlation
SSTA (°C)
-0.25
0
0.25
N.Atl. avg. SSTA
Correlation
-95%
+95%
a
1900 1920 1940 1960 1980 2000
-1
-0.5
0
0.5
1
Residual NINO 3.4 (DJF) vs. Mississippi Rain (JFM)
Correlation
SSTA (°C)
-0.25
0
0.25
N.Atl. avg. SSTA
Correlation
-95%
+95%
b
Figure 3. (a) Ten-year running mean of the AMO index (shaded
departures) shown in comparison with the 20-year running cor-
relation between the NINO-3.4 SSTA index for Dec.-Feb. and
the unsmoothed Jan.-Mar. rainfall anomaly of Florida climate
division 4 (heavy, solid). (b) As above, but for the area-weighted
rainfall accumulation over the Mississippi basin.
Slow changes associated with the AMO
We index the AMO with a ten-year running mean of
Atlantic SSTA north of the equator (Fig. 1a). The tem-
poral variations reproduce the phases and periodicity previ-
ously ascribed to the AMO. The roughly 0.4
◦
C peak-to-peak
variations are larger than for comparable areas in all other
oceans. The high correlations of this index with North At-
lantic gridded SSTA (Fig. 1b) confirm that this is an effec-
tive index. This simple index is virtually identical to what
one obtains by smoothing the first rotated (North Atlantic)
EOF mode of Mestas-Nu˜nez and Enfield [1999, henceforth
ME99].
1920 1940 1960 1980 2000
-0.2
-0.1
0
0.1
0.2
10-year running mean of N.Atlantic SSTA
1930-1959
1965-1994
SSTA (°C)
a
1930-1959
b
130 110 90 70W
20
30
40
50
-0.75 -0.5 -0.25 0 0.25 0.5 0.75
1965-1994
c
130 110 90 70W
20
30
40
50
Figure 4. (a) The AMO index (1920-1995) showing two con-
trasting 30-year time periods for the calculation of ENSO-climate
connections. (b) The correlation between the NINO-3.4 SSTA
index for Dec.-Feb. and the unsmoothed divisional rainfall for
Jan.-Mar. during the 30 year period 1930-1959. (c) As in b,
but for the 30 year period 1965-1994. The Mississippi basin is
highlighted by light gray fill. The colorbar applies to both maps.
90% significance is indicated by enhanced circles and by dashed
vertical lines on the colorbar.
ENFIELD ET AL.: MULTIDECADAL ATLANTIC SST VARIABILITY AND U.S. RAINFALL 2079
Consistent with the North Atlantic mode of ME99,cor-
relations between the AMO index and SSTA elsewhere in
the world ocean are small, except for the Pacific, mainly
north of 40
◦
N. ME99 hypothesize that the covariability in
the North Pacific is passively linked to the North Atlantic
through fluctuations in the tropospheric polar vortex. This,
and our choice of referring to the variability as “Atlantic”,
are consistent with indications that the oscillation is driven
primarily by interactions in the Atlantic sector and that the
Atlantic thermohaline circulation is involved [Delworth and
Mann, 2000; Venegas and Mysak, 2000]. However, we note
that the variability is global in scope and that the pres-
ence of the signal in the North Pacific SST may augment
the AMO mode itself and certainly may contribute to the
climate impacts associated with the AMO, such as we de-
scribe in this paper. We also note that ME99 have found
other multidecadal modes of SSTA variability in the Pacific
but that they are temporally uncorrelated with the AMO
variability.
The correlations of the similarly smoothed climate divi-
sion rainfall with the AMO index display a robust continental-
scale pattern dominated by negative correlations (Fig. 1c).
Many of the 90% significant correlations, all negative, are
found in the Mississippi basin. A further clustering of nega-
tive correlations occurs west of the continental divide, except
for positive correlations in the Pacific Northwest. Positive
regional clusters also appear in the northeast and Florida.
As a check on the seasonality of the rainfall pattern the
analysis was repeated for three-month seasonal averages of
the rainfall data (not shown). In all but the summer sea-
son (July-August-September) the patterns are different from
Fig. 1c and have far fewer significant correlations. The sum-
mer season pattern is similar and has many significant cor-
relations. We therefore believe that multidecadal variations
in summer rainfall are mainly responsible for the observed
relationship.
The temporal variabilities of rainfall are displayed for two
representative hydrological provinces. The large distribu-
tion of negative correlations in the central U.S. is character-
ized by the area-weighted accumulation of division rainfall
within the Mississippi basin (Fig. 2a). The smoothed, di-
rectly measured time series of Mississippi River outflow to
the Gulf of Mexico is shown for comparison. To characterize
the positive correlation in a much smaller region, we show
the single Florida division 4 rainfall series (Fig. 2b), which
includes the entire catchment for the Lake Okeechobee in-
flow in south-central Florida (Fig. 1c). We also show the
estimated unmanaged Lake Okeechobee inflow computed as
the difference between measured lake volume changes and
the total metered outflows at sluice gates [South Florida
Water Management District]. For both catchments the flow
closely mimics the rainfall totals, in spite of the neglect of
evapotranspiration. We note that the peak-to-peak varia-
tions are 10% and 40% of the long-term mean for the Mis-
sissippi and Okeechobee flows, respectively. The former rep-
resents a very large amount of water annually, while the
latter significantly affects water management policy in the
hydrologically sensitive South Florida region.
Consistent with the significant divisional correlations in
Fig. 1c, these two very different hydrological regions track
the phases of the AMO very closely (Fig. 1a), except for a
brief period in the 1940s for the Mississippi basin. A more
detailed examination of SSTA in subregions of the North At-
lantic and allowing for shorter periodicities (less smoothing)
does not suggest an Atlantic source for the 1940s anomaly.
We can only speculate that the Mississippi basin is also sen-
sitive to and affected by one or more of the other slow climate
modes, such as occur in the Pacific sector.
Changes in ENSO variability
It is of interest to know whether the pattern of telecon-
nections of U.S. rainfall to tropical Pacific ENSO indices
changes significantly between phases of the AMO. It was re-
cently shown that changes in ENSO related rainfall anoma-
lies occur with alternating phases of the Pacific Decadal Os-
cillation (PDO). The PDO has shorter time scales than that
of the AMO [Mantua et al., 1997]. A north-south bipo-
lar distribution of correlations between western U.S. rainfall
and the Southern Oscillation Index (SOI) is stronger (more
significant correlations) when east Pacific SSTA is decadally
cool [McCabe and Dettinger, 1999]. During the high phase
of the PDO (east Pacific warm) El Ni˜no events exhibit a
more robust pattern of wetter (drier) winters in the southern
(northern) tier of the contiguous United States [Gershunov
and Barnett, 1998].
To test for analogous relationships with the AMO, we
first computed running 20-year correlations of the unsmoothed
Mississippi basin and Okeechobee rainfall totals with the av-
erage SSTA over the NINO-3.4 index region in the equato-
rial Pacific (5
◦
N-5
◦
S, 170
◦
W-120
◦
W). The correlations are
steadily positive and significant for the Lake Okeechobee
rainfall, i.e., south-central Florida is wetter (dryer) during
El Ni˜no (La Ni˜na) years, regardless of the AMO phase (Fig.
3a). For the Mississippi basin rainfall, however, we see a
very clear change (Fig. 3b). During the 1930-1960 warm
phase of the AMO the rainfall had a significant negative
correlation with NINO-3.4, whereas during the cool phases
before and after the correlations were insignificant.
To better understand the running correlations we corre-
lated the boreal winter NINO-3.4 index (December-January-
February) with the winter rainfall (January-February-March)
of every climate division, for two contrasting 30 year periods
(Fig. 4a): 1930- 1959 (Fig. 4b) and 1965-1994 (Fig. 4c).
The correlation patterns are similar in form but contrast
greatly in the size of regional clusters. For the AMO warm
phase, most of the eastern Mississippi basin is character-
ized by large negative correlations, while significant positive
correlations are confined to Florida and the southwestern
border with Mexico. For the more recent cool phase of the
AMO, the period for which most of our present knowledge
of ENSO impacts has been obtained, there is a larger dis-
tribution of positive correlations all along the southern tier
states. In contrast, the coverage of negative correlations
over the eastern Mississippi basin is half that of the warm
phase, while the Great Plains shows a large positive cluster.
The net result is that the Mississippi basin rainfall accumu-
lation is significantly impacted by ENSO (less winter rain-
fall during El Ni˜no events) during the AMO warm phase
(when negative correlations dominate) but not during the
cool phase (when positive correlations offset negative corre-
lations). The changes in the continental scale pattern are
not reflected in Florida rainfall, which has significant posi-
tive correlations for both phases of the AMO.
An analysis of how the AMO modulation of ENSO- con-
nected U.S. rainfall is related to the PDO modulation in the
western U.S. [McCabe and Dettinger, 1999] is beyond the
2080 ENFIELD ET AL.: MULTIDECADAL ATLANTIC SST VARIABILITY AND U.S. RAINFALL
scope of this paper. We note, however, that the PDO is
characterized by shorter time scales than the AMO. Hence,
both of these slow modes appear to modulate ENSO rainfall
and their effects may interact in complicated ways.
Discussion
To probe the explanation for the patterns we see, we
calculated the composite average distributions of 500 hPa
geopotential height from the NCEP/NCAR reanal-
ysis [Kalnay et al., 1996] for two periods, 1949-1969 and
1970-1994, and subtracted the average for 1949-1999 (not
shown). For the early period (AMO warm) the normal win-
ter ridge-trough pattern is flattened over the northern tier
of the U.S., ie., the ridge over the Pacific Northwest weak-
ens and the trough over the northern east-central region also
weakens. Over the southern tier the tendency is opposite,
i.e., 500 hPa heights tend to rise off the west coast and
decrease across the southeast. This can be interpreted as
a greater (lesser) frequency of winter cyclonic activity and
rainfall in the northwest (east-central) while the opposite
holds in the southwest (southeast). These mean tendencies
are clearly reflected in the correlations between the AMO
index and smoothed rainfall (Fig. 1c). They run counter to
the ENSO teleconnections in the west, weakening the ENSO
pattern there (Fig. 4b) while enhancing the El Ni˜no pattern
of dryness over the southern Ohio River drainage. For the
later period (AMO cool) the 500 hPa ridge-trough pattern
is strengthened. This accentuates the ENSO pattern in the
west and diminishes the area of dryness south of the Great
Lakes (Fig. 4c).
We note that the AMO index has been increasing since
about 1990 and became positive again circa 1995. Hence,
we may have once again entered a period such as 1930-1960,
and global temperatures can be expected to be greater than
they would be based only on greenhouse and other exter-
nal forcings [Andronova and Schlesinger, 2000]. However,
contrary to the general expectation of greater extratropical
rainfall under greenhouse warming scenarios [Houghton et
al., 1996], the effect of this new AMO warming should be
to decrease annual rainfall totals over the U.S., especially
over the eastern Mississippi basin. This implies that future
attempts to anticipate the impact of global warming on re-
gional rainfall may prove inaccurate if the models do not
reproduce the AMO variability and its impacts. This raises
the bar on the ability of coupled models to simulate the
climate of the 21st century.
The AMO-related rainfall variability has immediate prac-
tical implications for water management policies in the af-
fected regions of the United States. For example, during the
positive phase of the oscillation (1930-1964), net average an-
nual inflow to Lake Okeechobee was about double that dur-
ing the ensuing negative phase (1965-1994). This translates
into a near complete reversal in water management prior-
ities for multi-decadal periods. During the negative AMO
phase, inflow to the Lake is barely enough to meet the signif-
icant water needs of south Florida and management policy
must be biased in favor of water conservation. Included are
the hydrological demands of the Everglades, the minimum
freshwater flows required for the numerous productive es-
tuaries that populate the Florida coastlines, the demands
of agricultural industries that exploit the interior sections
of south Florida, and the water supplies of the rapidly de-
veloping coastal communities. During the positive phase
management priorities shift towards flood protection for the
region surrounding the Lake and minimizing the undesir-
able ecological impacts of high water levels on the Lake’s
littoral zone. This often requires large discharges of fresh-
water through the coastal estuaries that must be managed
carefully to minimize adverse effects that such discharges
have on the downstream ecosystems.
Finally, it is clear both from this study and that of others
that the slow variability (decadal or longer) in both northern
oceans renders ENSO teleconnections nonstationary over
the United States. Current methods of forecasting ENSO
climate impacts are based mainly on empirical relationships
involving observations taken during the recent AMO cool
phase. To the extent that we continue to use empirical re-
lationships, the shifts currently taking place might be ac-
counted for by using earlier observations and paleoclimate
findings. However, it is also clear that the best long-term
solution for climate prediction is to overcome the current
failure of coupled models to forecast rainfall impacts, and
for the models to account for interdecadal variabilities.
Acknowledgments.
We thank D. Goolsby (USGS) for
providing Mississippi River data and three AOML colleagues
(Drs. C. Wang, H. Willoughby and C. Landsea) for their com-
ments. This research was funded by NOAA’s Office of Global
Programs (PACS GC99-024) and the Inter-American Institute
for Global Change Research (IAI CRN-038).
References
Andronova, N. G., and M. E. Schlesinger, Causes of global tem-
perature changes during the 19th and 20th centuries, Geophys.
Res. Lett., 27, 2137-2140, 2000.
Delworth, T. L., and M. E. Mann, Observed and simulated mul-
tidecadal variability in the Northern Hemisphere, Climate Dy-
namics, 16, 661-676, 2000.
Ebisuzaki, W., A method to estimate the statistical significance of
a correlation when the data are serially correlated, J. Climate,
10, 2147-2153, 1997.
Gershunov, A., and T.P. Barnett, Interdecadal Modulation of
ENSO Teleconnections, Bull. Amer. Meteor. Soc., 80, 2715-
2725, 1998.
Houghton, J.T. et al., Climate Change 1995: The Scienc e of
Climate Change, Cambridge University Press, 1996.
Kalnay, E. et al., The NCEP/NCAR 40-year reanalysis project,
Bull.Amer.Meteor.Soc., 77, 437-471, 1996.
Kaplan, A., M. A. Cane, Y. Kushnir, and A. C. Clement, Analysis
of global sea surface temperatures 1856-1991, J. Geophys. Res.,
103, 18,567-18,589, 1998.
Kerr, R. A., A North Atlantic climate pacemaker for the centuries,
Science, 288 (5473), 1984-1986, 2000.
Mantua, N.J., S.R. Hare, Y. Zhang, J.M. Wallace, and R.C.
Francis, A Pacific decadal climate oscillation with impacts on
salmon production, Bull. Amer. Meteor. Soc., 78, 1069-1079,
1997.
McCabe, G. J., and M. D. Dettinger, Decadal variations in the
strength of ENSO teleconnections with precipitation in the
western United States, Int. J. Climatol., 19, 1399-1410, 2000.
Mestas-Nu˜nez, A. M., and D. B. Enfield, Rotated global modes
of non-ENSO sea surface temperature variability, J. Climate,
12, 2734-2746, 1999.
Schlesinger, M. E., and N. Ramankutty, An oscillation in the
global climate system of period 65-70 years, Nature, 367, 723-
726, 1994.
Venegas, S.A., and L.A. Mysak, Is there a dominant timescale
of natural climate variability in the Arctic?, J. Climate, 13,
3412-3434, 2000.
D. B. Enfield, NOAA Atlantic Oceanographic and Mete-
orological Laboratory, Miami, FL, 33149 USA. (e-mail: en
field@aoml.noaa.gov)
(Received December 12, 2000; revised February 16, 2001;
accepted February 26, 2001.)
