Content uploaded by Guojian Wang
Author content
All content in this area was uploaded by Guojian Wang on Jun 15, 2014
Content may be subject to copyright.
LETTER doi:10.1038/nature13327
Increased frequency of extreme Indian Ocean Dipole
events due to greenhouse warming
Wenju Cai
1,2
, Agus Santoso
3
, Guojian Wang
2,1
, Evan Weller
1
, Lixin Wu
2
, Karumuri Ashok
4
, Yukio Masumoto
5,6
& Toshio Yamagata
7
The Indian Ocean dipole is a prominent mode of coupled ocean–
atmosphere variability
1–4
, affecting the lives of millions of people in
IndianOceanrimcountries
5–15
.Initspositivephase,seasurface tem-
peratures are lower than normal off the Sumatra–Java coast, but higher
in the western tropical Indian Ocean. During the extreme positive-
IOD (pIOD) events of 1961, 1994 and 1997, the eastern cooling
strengthened and extended westward along the equatorial Indian
Ocean through strong reversal of both the mean westerly winds and
the associated eastward-flowing upper ocean currents
1,2
. This cre-
ated anomalously dry conditions from the eastern to the central Indian
Ocean along the Equator and atmospheric convergence farther west,
leading to catastrophic floods in eastern tropical African countries
13,14
but devastating droughts in eastern Indian Ocean rim countries
8–10,16,17
.
Despite these serious consequences, the response of pIOD events to
greenhouse warming is unknown. Here, using an ensemble of cli-
mate models forced by a scenario of high greenhouse gas emissions
(Representative Concentration Pathway 8.5), we project that the fre-
quency of extreme pIOD events will increase by almost a factor of
three, from one event every 17.3 years over the twentieth century to
one event every 6.3years over the twenty-first century. We find that
a mean state change—with weakening of both equatorial westerly winds
and eastward oceanic currents in association with a faster warming
in the western than the eastern equatorial Indian Ocean—facilitates
more frequent occurrences ofwind and oceanic current reversal. This
leads to more frequent extreme pIOD events, suggesting an increas-
ing frequency of extreme climate and weather events in regions affec-
ted by the pIOD.
In austral winter and spring, southeasterly trade winds that feed the
tropical convergence zone near the maritime continent are a feature of the
southern tropical Indian Ocean. During a pIOD event, an initial cooling off
Sumatra–Java, the eastern pole of theIndian Ocean dipole, suppresses local
convection, inducing easterly wind anomalies and a shallowing thermo-
cline.Thispromotes upwelling that in turn reinforces theinitial cooling
1,2,18
,
a process referred to as Bjerknes feedback. The growth of cool anomalies
causesanorthwestwardextensionofthesoutheasterlytradewinds
1,2,16
,with
anomalous easterlies along the equatorial Indian Ocean (Fig. 1a), where
weak westerlies normally prevail. Thechangeinwindpromotesconver-
gence, rainfall and warm anomalies in the equatorial western Indian Ocean.
The altered circulations induce droughts and bushfires in eastern Asia and
Australia
5–8
, floods in parts of the Indian subcontinent
11
and eastern
Africa
13,14
,coral reef death across western Sumatra
12
, and malaria outbreaks
in eastern Africa
15
. During extreme pIOD events, as occurred in 1961, 1994
and 1997, the anomalies, particularly the anomalous equatorial easterlies,
are far stronger (Fig. 1b), with commensurately greater impacts. During the
1997 event, devastating floods in Somalia, Ethiopia, Kenya, Sudan and
Uganda caused several thousand deaths and displaced hundreds of thou-
sands of people. In contrast, Indonesia suffered severe droughts and wild-
fires
2,16,17
made worse by the developing 1997 El Nin
˜o;theassociatedsmoke
and haze caused severe health problems to tens of millions of people in
Indonesia and surrounding countries
9,10
.
These dramatic impacts call for an urgent investigation into whether
extreme pIOD events will change in a warmer climate. Recent studies
have shown that greenhouse warming leads to a mean state change in the
equatorial Indian Ocean with an easterly wind trend and a faster warming
rate in the west than in the east, but referenced to the evolving mean state
there is no detectable change in either the overall frequency or amplitude
of pIOD events
19–21
. Here, using a suite of distinct process-based indica-
tors, we show that there is in fact a significant increase in the frequency of
the extreme pIOD events under greenhouse warming.
We characterize the observed extreme pIOD events in terms of their
contrast with moderate events, focusing on austral spring, the season in
which the IOD usually peaks. During extreme pIOD events, the cooling off
Sumatra is intensified by the large equatorial easterly anomalies through
generation of equatorial and coastally trapped upwelling Kelvin waves
22,23
,
enhanced evaporation
3
, and a weakening of the mean eastward oceanic
flows that transport heat eastward towards Sumatra
24
. The anomalous
convergence in the west, marked by increased rainfall and temperature, is
amplified through a series of processes: reduced wind speed and evapora-
tionassociatedwith the downstream extension of the southeasterly trades; a
deeper thermocline caused by the weaker eastward ocean heat transport
along the Equator
24
; and generation of equatorial downwelling Rossby
waves
3,4
. The warming in the west and cooling in theeast in turn strength-
ens the equatorial easterly anomalies, introducing a positive feedback
along the Equator that operates in addition to the Bjerknes feedback
centred off Sumatra–Java. The equatorial positive feedback, which is
far stronger during extreme pIOD events, leads to stronger equatorial
cooling (Extended Data Fig. 1a), and reversal of the equatorial winds and
ocean currents so that they flow towards the west (Extended Data Fig. 2f).
This creates a zone of atmospheric subsidence along the Equator char-
acterized by low rainfall and colder sea surface temperatures (SSTs) that
extend much farther to the west than during moderate pIOD events
(Fig. 1; Extended Data Fig. 1a).
Aheatbudgetanalysisfortheeastern-to-centralequatorialIndianOcean
during the IOD developing phase (July to October; Extended Data Figs 1
and 2 and Methods) clearly indicates the 1961, 1994 and 1997 events to be
the most extreme pIODs. The growth of equatorial SST anomalies during
these three events is dominated by nonlinear processes involving zonal
current anomalies. In particular, the nonlinear zonal advection, that is,
the product of the anomalous west-minus-east SST gradient with the
anomalous zonal currents (dark red bar, Extended Data Fig. 1c), sets these
three events apart from the rest. Essentially, the equatorial positive feedback
enhances anomalies of westward-flowing equatorial winds and currents,
allowing for an eventual reversal. This nonlinear process can be parame-
terized by the product of the equatorial easterly anomalies, which drive
thecurrent,andthedipolemodeindex(DMI)
1
, which measures the
west-minus-eastSST gradient (see Methods). Such nonlinearity also occurs
1
CSIRO Marine and Atmospheric Research, Aspendale, Victoria, 3195 Australia.
2
Physical Oceanography Laboratory, Qingdao Collaborative Innovation Center of Marine Science and Technology, Ocean
University of China, Qingdao, 266003 China.
3
Climate Change Research Centre and ARC Centre of Excellence for Climate System Science, The University of New South Wales, Sydney, 2052 Australia.
4
Centre for Climate Change Research, Indian Institute of Tropical Meteorology Pashan, Pune 411 008, India.
5
Department of Earth and Planetary Science, Graduate School of Science, The University of
Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033, Japan.
6
Climate Variation Predictability and Applicatability Research Program, Japan Agency for Marine-Earth Science and Technology (JAMSTEC),
3173-25 Showa-machi, Kanazawa-ku, Yokohama 236-0001, Japan.
7
Application Laboratory, JAMSTEC, 3173-25 Showa-machi, Kanazawa-ku, Yokohama 236-0001, Japan.
254 | NATURE | VOL 510 | 12 JUNE 2014
Macmillan Publishers Limited. All rights reserved
©2014
in the eastern pole, rendering a negative skewness of SST, in that cool
anomalies off Sumatra grow to greater amplitude than warm anomalies
25
.
The strong nonlinearity along the Equator means that the representa-
tion of extreme pIOD impacts requires more than just the commonly
used DMI. This along-the-Equator nonlinearity can be represented by
two modes of empirical orthogonal function (EOF) of rainfall anomalies.
The pattern of the first EOF (EOF1, 43.4% of the total variance, Fig. 1c)
shows an east–west dipole of reduced convection, featuring anomalously
cold SSTs and a shallow thermocline in the east but anomalies of opposite
polarities in the west (Extended Data Fig. 3). This reflects characteristics
of pIOD events commonly depicted by the DMI. EOF2, which accounts
for 20.7% of the total variance, on the other hand, reflects pronounced
anomalous conditions during extreme pIOD events, as described above.
Both EOFs feature enhanced convection over the western tropical Indian
Ocean and equatorial Africa (Fig. 1d, Extended Data Fig. 3).
EOF1 and EOF2 (or the DMI) display a nonlinear relationship (Fig. 1e,
f). During a moderateevent, the two EOFs are of opposite sign. Thus, the
associated rainfall anomalies tend to offset over the central Indian
Ocean. In contrast, during an extreme pIOD, both EOFs are positive,
rendering large negative rainfall anomalies that extend westward along
0º
30º N
30º S
0.05 N m–2
a 1982 rainfall and wind anomalies b 1997 rainfall and wind anomalies
–6 –4 –2 0 2 4 6
c Observed rst principal pattern d Observed second principal pattern
–3 –2 –1 0 1 2 3
–4 –2 0 2 4
–2
0
2
4
1982
1987
1994
1997
2002
2006
1961
First
p
rinci
p
al com
p
onent
Second principal component
e Observed relationship, 1979–2010
–1 0 1 2
–2
0
2
4
1982
1987
1994
1997
2002
2006
1961
Di
p
ole mode index
(
ºC
)
Second principal component
f Observed relationship, 1979–2010
40º E 80º E 120º E
0.05 N m–2
40º E 80º E 120º E
0º
30º N
30º S
Rainfall anomaly (mm d–1)
0.01 N m–2 0.005 N m–2
0º
30º N
30º S
0º
30º N
30º S
40º E 80º E 120º E 40º E 80º E 120º E
Rainfall anomaly (mm d–1) (s.d.)–1
Figure 1
|
Comparison of moderateand extreme pIOD and identification of
extreme pIOD events. a,b, September–Novemberaverage rainfall (shading, in
units of mm day
21
) and wind stress (vector scale is shown in the top right
corner for each panel) anomalies associated with a moderate (1982) and
extreme (1997) pIOD. c,d, Principal variability patterns of rainfall obtained by
applying a statistical and signal processing method, EOF analysis, to a satellite-
era rainfall anomaly data from the Global Precipitation Climatology Project
version 2 (see Methods), in the equatorial region (10uS–10uN, 40uE–100uE).
The associated rainfall and wind stress vectors from reanalysis data (see
Methods) are presented as linear regression onto the EOF time series. The
colour scale indicates rainfall in mmday
21
per 1 s.d. change; blue or red
contours indicate increased or decreased rain. Note the different vector scales
in cand d. e, Relationship between the two principal component time series.
Values for 1961 are obtained by regressing the rainfall anomaly pattern from a
reanalysis onto the EOF1 and EOF2 pattern (see Methods). An extreme pIOD
event (red dots) is defined as when the first principal component is greater
than 1 s.d. and the second principal component is greater than 0.5s.d. A
moderate pIOD event (green dots) is determined from a detrended DMI
1
when its amplitude is greater than 0.75 s.d. other than the 1994 and 1997
events. Negative IOD and neutral years are indicated with blue dots.
f, Relationship between the second principal component time series and the
DMI.
LETTER RESEARCH
12 JUNE 2014 | VOL 510 | NATURE | 255
Macmillan Publishers Limited. All rights reserved
©2014
–0.06 –0.04 –0.02 0 0.02 0.04
0
20
40
60
80
a Extreme pIOD events
Zonal wind stress anomalies (N m–2)
Number of occurrences
1900–1999
2000–2099
–0.06 –0.04 –0.02 0 0.02 0.04
0
50
100
150
200
250
c All events other than extreme plOD events
Zonal wind stress anomalies (N m–2)
Number of occurrences
–0.2 –0.15 –0.1 –0.05 0
0
10
20
30
40
50
b Extreme pIOD events
Nonlinear zonal advection (N m–2 ºC)
Number of occurrences
–0.2 –0.15 –0.1 –0.05 0
0
100
200
300
400
500
600
700
800
d All events other than extreme plOD events
Nonlinear zonal advection (N m–2 ºC)
Number of occurrences
Figure 3
|
Multi-model statistics associated with
the increase in frequency of extreme pIOD
events. a, Multi-model ensemble histogram of zonal
wind stress t
x
anomalies in the equatorial Indian
Ocean (5uS–5uN, 60uE–100uE), referenced to the
‘control’ period. These are averaged over the July–
October months of Indian Ocean dipole development
phase. Values during extreme pIOD years in each
period are separated into 5 310
23
Nm
22
bins
centred at the tick point for the ‘control’ (blue) and
‘climate change’ (red) periods. The multi-model
median for the ‘control’ (dashed blue line) and the
‘climate change’ (dashed red line) periods are
indicated. b,Thesameasabut for the product of t
x
anomalies shown in amultiplied by the DMI
1
(separated into 0.01 Nm
22
uCbins),approximating
the nonlinear zonal advection (see Methods). c,d,The
same as aand bbut for all years excluding extreme
pIOD events. The histogram for extreme pIOD is
statistically different above the 95% confidence level
from that for non-extreme pIOD events, for both the
‘control’ and the ‘climate change’ periods. On average,
nonlinear advection is greater for extreme pIOD
events than for non-extreme pIOD events.
a Modelled rst principal pattern b Modelled second principal pattern
–3 –2 –1 0 1 2 3
–4 –2 0 2 4
–4
–2
0
2
4
First principal component
Second principal component
c ‘Control’ period, 1900–1999
Extreme = 133
Moderate = 392
–4 –2 0 2 4
–4
–2
0
2
4
First principal component
Second principal component
d ‘Climate change’ period, 2000–2099
Extreme = 367
Moderate = 293
0º
30º N
30º S
0.01 N m–2
0º
30º N
30º S
0.005 N m–2
40º E 80º E 120º E 40º E 80º E 120º E
Rainfall anomaly (mm d–1) (s.d.)–1
Figure 2
|
Multi-model ensemble average of the
principal variability patterns of austral spring
season rainfall and their nonlinear relationship.
a, b, First and second principal variability patterns
of rainfall anomalies referenced to the ‘control’
period (1900–1999), obtained by applying an EOF
analysis to rainfall anomalies in the equatorial region
(10uS–10uN, 40uE–100uE). Note the different
vector scales in aand b. The associated pattern and
wind stress vectors beyond the domain are obtained
by a linear regression onto the EOF time series. The
colour scale indicates rainfall in mm day
21
per 1.0 s.d.
change; blue or red contours indicate increased or
decreased rainfall. c,d, A nonlinear relationship
between the principal component time series for the
‘control’ (1900–1999) and ‘climate change’ (2000–
2099) periods. An extreme pIOD event (red dots) is
defined as when the first principal component is
greater than 1 s.d. and the second principal
component is greater than 0.5 s.d. A moderate pIOD
event (green dots) is determined from a detrended
DMI when its amplitude is greater than 0.75 s.d. but
is not an extreme pIOD event. Negative IOD and
neutral years are indicated with blue dots. The
number of extreme and moderate pIOD events is
indicated.
RESEARCH LETTER
256 | NATURE | VOL 510 | 12 JUNE 2014
Macmillan Publishers Limited. All rights reserved
©2014
the Equator. The pIOD events of 1961, 1994 and 1997 are determined
to be ‘extreme’ when EOF1 is greater than a 1-standard-deviation (s.d.)
value and EOF2 is greater than a positive 0.5-s.d. value. The characteristics
of the pIOD events are only fully captured by the superimposition of these
two EOFs (Extended Data Fig. 4). Without EOF2, the salient feature of the
westward extending equatorial anomalies that characterizes extreme pIOD
would be missed. A similar EOF analysis on verticalvelocity vat 500 mb, a
measure of convection, generates similar patterns (Extended Data Fig. 5).
To assess the influence of greenhouse warming, we use the Coupled
Model Intercomparison Project phase 5 (CMIP5
26
) multi-model database.
The coupled general circulation models (CGCMs) used in this study are
forced with historical anthropogenic andnatural forcings, and future green-
house-gas emission scenarios of Representative Concentration Pathway
(RCP) 8.5, covering the 1900–2099 period. Not all of the 31 CGCMs con-
sidered here are able to simulate the characteristics of observedpIOD events.
We focus on 23 CGCMs that simulate negative skewness of SST off Sumatra
as well as the nonlinear relationship of the tworainfall EOFs (Extended Data
Table 1, Fig. 2). An identical EOF analysis of vat 500 mb in 21 out of the 23
selected CGCMs, in which vis available, produces similar spatial patterns
and their nonlinear relationship (Extended Data Fig. 6). From these 23
CGCMs, we define extreme pIOD events in the same manner as for
the observed events, and compare their frequency in the first (1900–
1999) and second (2000–2099) hundred-year periods. These two adja-
cent periods within a transient scenario are referred to as the ‘control’
and ‘climate change’ periods, respectively.
In aggregation, the frequency of extreme pIOD events based on rainfall
EOFs increases by a factor of 2.7, from about one event every 17.3 years
(133 events in 2,300 years) in the ‘control’ period, to one every 6.3 years
(367 events in 2,300 years) in the ‘climate change’ period (Fig. 2c and d).
This is statistically significant according to a bootstrap test
27
,underscored
by a strong inter-model consensus, with 21 out of 23 models simulating an
increase (Extended Data Table 1). Sensitivity tests to varying definitions of
extreme pIOD further support the robustness of this result (Supplementary
Tables 1 and 2).
Development of pIOD events can interact with an El Nin
˜oevent
28–30
.
The 1997 extreme pIOD developed in conjunction with the strongest El
Nin
˜o of the twentieth century. The 1961 and 1994 extreme pIODs on the
other hand occurred without an El Nin
˜o, supporting the notion that the
generating mechanism behind an extreme pIOD event lies within the
Indian Ocean
2
.WefindnoevidencethattheincreaseinextremepIOD
events in the ‘climate change’ period is induced by a change in the fre-
quency of El Nin
˜oorElNin
˜o Modoki occurrences (see Methods and Ex-
tended Data Fig. 7).
Rather, the increase in extreme pIOD events appears to arise from mean
state changes within the Indian Ocean (Extended Data Fig. 8), consis-
tent with a weakening Walker circulation as projected under greenhouse
warming
19–21
. Relative to the ‘control’ period, the altered mean state is more
conducive to equatorial easterly winds, westward oceanic currents, an en-
hanced west-minus-east SST gradient, and the associated nonlinear zonal
advection. There is a strong link between climatologically stronger easterly
winds along the Equator and more occurrences of a given nonlinear advec-
tion (correlation coefficient r50.9, not shown). These changes lead to
increasing occurrences of extreme pIOD events, because a smaller per-
turbation is required in the ‘climate change’ period to generate the same
sizeofnonlinearzonaladvectionasseenduringextremepIOD events in the
‘control’ period (see Extended Data Fig. 9). Thus, there are increased occur-
rences of extreme pIOD events for a given size of the equatorial easterly
anomaly (Fig. 3a), or a given strength of nonlinear advection (Fig. 3b). On
the other hand, the changes associated with non-extreme pIOD events are
not as apparent (Fig. 3c, d).
The increased frequency in extreme pIODs does not translate to greater
intensityofrainfallanomalies over all regions affected by the pIOD (Fig. 4a–
c). Over northeastern equatorial Africa, the extreme pIOD-induced wet
events do become more intense in the ‘climate change’ period than in the
‘control’ period (Fig. 4d; the means are statistically different above the 95%
confidence level). In contrast, there is no statistically significant difference
betweenthetwoperiodsintheintensityofdryepisodesoverJava(Fig.4f).In
addition,thedifferenceinrainfallintensityoftheextremeeventsisgenerally
a ‘Control’ period b ‘Climate change’ period c ‘Climate change’ minus ‘Control’
–3 –2 –1 0 1 2 3
–3 –2 –1 0 1 2 3 4 5 6
0
20
40
60
80
d
Eastern Africa, extreme pIOD events
Number of
occurrences
1900−1999
2000−2099
–3 –2 –1 0 1 2 3 4 5 6
0
200
400
600
e
Eastern Africa, all other events
–8 –6 –4 –2 0 2 4 6 8 10
0
40
80
120
f
Java, extreme pIOD events
Rainfall anomalies (mm d–1)
–8 –6 –4 –2 0 2 4 6 8 10
0
200
400
600
g
Java, all other events
0º
30º N
30º S
40º E 80º E 120º E
0º
30º N
30º S
0º
30º N
30º S
40º E 80º E 120º E 40º E 80º E 120º E
Rainfall anomaly (mm d–1) (s.d.)–1
Figure 4
|
Multi-model ensemble average of rainfall anomalies (referenced
to the ‘control’ period)during extreme pIOD events and associatedstatistics
in affected regions. a,band c, Ensemble average rainfall anomalies in the
‘control’ and ‘climate change’ periods, and their difference (‘climate change’
minus ‘control’). Stippling in cindicates regions where the differences are
statistically significant at the 95% level as determined by a two-sided Student’s t-
test. d,e, Multi-model ensemble histogram of the rainfall anomalies over northern
equatorial East Africa (0u–5uN, 37.5uE–47uE) for extreme pIOD events and
all events other than extreme pIOD events, respectively. All extreme pIOD events
in each period are separated into 0.5 mm day
21
bins centred at the tick point
for the ‘control’ (blue) and ‘climate change’ (red) periods. The multi-model mean
for the ‘control’ (blue dashed line) and the ‘climate change’ (red dashed line)
periods are indicated. In each period the histograms for extreme pIOD (for
example, red bars in Fig. 4d) and non-extreme pIOD (for example, red bars in
Fig. 4e) are statistically different above the 95% confidence level. f,g,Thesameas
dand e, but for the Java region (8uS–6uS, 105.5uE–108.5uE), separated into
1mmday
21
bins. The two histograms in dand eare statistically different above
the 95% confidence level, but this is not the case for the two histograms in fand g.
LETTER RESEARCH
12 JUNE 2014 | VOL 510 | NATURE | 257
Macmillan Publishers Limited. All rights reserved
©2014
smaller than the difference in the mean rainfall(comparing Fig. 4a and
Extended Data Fig. 8a), despite the far greateranomalies during extreme
pIOD events. This illustrates that in general the impacts of extreme
pIOD events experienced in the ‘control’ period will repeat more fre-
quently in the ‘climate change’ period.
In summary, our finding of a greenhouse-induced increased frequency
of extreme pIOD events is in stark contrast with previous results of no
change in pIOD frequency about the evolving background state. By iden-
tifying nonlinear processes that give rise to extreme pIOD events, we show
that under greenhouse warming, the evolving equatorial Indian Ocean
towards climatologically stronger west-minus-east temperature gradients
and easterly winds is more susceptible to producing more frequent
extremepIODevents.Withtheprojectedlargeincreaseinextreme
pIOD events, we should expect more frequent occurrences of devastating
weather events in affected regions.
METHODS SUMMARY
The extreme pIOD events were diagnosed using a suite of distinct process-based
indicators—such as anomalous equatorial easterlies, low rainfall and atmospheric
subsidence—asinduced by a downstreamextension of the southeasterly trades. For
observations, we focus on historical events in the satellite era (1979–present) monthly
precipitationanalysis, SSTs and othercirculation fields from a globalreanalysis (see
Methods). We focus on austral spring, September–November, in whicha pIOD typ-
ically peaks. A heat budget analysis for the eastern-to-central equatorial Indian Ocean
using the European Centre for Medium-Range Weather Forecasts - Ocean Reanalysis
System 3 revealsthat the strong nonlinear zonal advection of heat sets the observed
1994 and 1997 extremepIOD events apart from other events.The nonlinearity sug-
gests that the traditional DMI, defined as the difference in SST anomalies between
the western (50uE–70uEand10uS–10uN) and eastern (90uE–110uE and 10uS–
0uS) parts of the Indian Ocean
1
is not sufficient to differentiate an extreme pIOD
event. Thus, we propose an identification method for extreme pIOD, in which we
apply EOF analysis to rainfall anomalies and vertical velocity vat 500 mb in the
equatorial Indian Ocean (40uE–100uE, 10uS–10uN). This produces two principal
variability patterns, one depicting an east–westpattern and the other depicting dry
conditions along the central Indian Ocean extending from the east. An extreme
pIOD event is defined as when the first principal time series is greater than 1 s.d.,
and the second greater than 0.5 s.d. This definition exclusively captures the three
observed extreme pIOD events. To select CGCMs, the method is applied to 31
CMIP5 CGCMs, each covering 105 years of a pre-twenty-first-century climate
changesimulation usinghistorical anthropogenic andnatural forcings(1901–2005)
and a further 95 years (2006–2100) under the RCP8.5 forcing scenario
26
.
Online Content Any additional Methods, ExtendedData display items and Source
Data are available in the online version of the paper; references unique to these
sections appear only in the online paper.
Received 28 November 2013; accepted 8 April 2014.
1. Saji, N. H., Goswami, B. N., Vinayachandran, P. N. & Yamagata, T. A dipole in the
tropical Indian Ocean. Nature 401, 360–363 (1999).
2. Webster, P. J., Moore, A. M., Loschnigg, J. P. & Leben, R. R. Coupled oceanic-
atmospheric dynamics in the Indian Ocean during 1997–98. Nature 401,
356–360 (1999).
3. Yu, L. & Rienecker, M. M. Mechanisms for the Indian Ocean warming during the
1997–98 El Nin
˜o. Geophys. Res. Lett. 26, 735–738 (1999).
4. Murtugudde, R., McCreary, J. P. & Busalacchi, A. J. Oceanic processes associated
with anomalous events in the Indian Ocean with relevance to 1997–1998.
J. Geophys. Res. 105, 3295–3306 (2000).
5. Meyers, G. A., McIntosh, P. C., Pigot, L. & Pook, M. J. The years of El Nin
˜o, La Nin
˜a,
and interactions with the tropical Indian Ocean. J. Clim. 20, 2872–2880 (2007).
6. Ummenhofer, C. C. et al. What causes southeast Australia’s worst droughts?
Geophys. Res. Lett. 36, L04706 (2009).
7. Ashok, K., Guan, Z. & Yamagata, T. Influence of the Indian Ocean Dipole on the
Australian winter rainfall. Geophys. Res. Lett. 30, 1821 (2003).
8. Cai, W., Cowan, T. & Raupach, M. Positive Indian OceanDipole events precondition
southeast Australia bushfires. Geophys. Res. Lett. 36, L19710 (2009).
9. Emmanuel, S. Impact to lung health of haze from forest fires: the Singapore
experience. Respirology 5, 175–182 (2000).
10. Frankenberg, E., McKee, D. & Thomas, D. Health consequences of forest fires in
Indonesia. Demography 42, 109–129 (2005).
11. Zubair, L., Rao, S. A. & Yamagata, T. Modulation of Sri Lankan Maha rainfall by the
Indian Ocean dipole. Geophys. Res. Lett. 30, 1063 (2003).
12. Abram, N. J., Gagan, M. K.,McCulloch, M. T., Chappell,J. & Hantoro, W. S. Coral reef
death duringthe 1997 Indian Ocean Dipole linked to Indonesian wildfires. Science
301, 952–955 (2003).
13. Behera, S. K. et al. Paramount impact of the Indian Ocean Dipole on the East
African short rains: a CGCM study. J. Clim. 18, 4514–4530 (2005).
14. Black, E., Slingo, J. & Sperber, K. R. An observational study of the relationship
between excessively strong short rains in coastal East Africa and Indian Ocean
SST. Mon. Weath. Rev. 131, 74–94 (2003).
15. Hashizume, M., Chaves, L. F. & Minakawa, N. Indian Ocean Dipole drives malaria
resurgence in East African highlands. Sci. Rep. 2, 269 (2012).
16. Schott, F. A., Xie, S.-P. & McCreary, J. P. Indian Ocean circulation and climate
variability. Rev. Geophys. 47, RG1002 (2009).
17. Page, S. E. et al. The amount of carbon released from peat and forest fires in
Indonesia in 1997. Nature 420, 61–65 (2002).
18. Udea, H. & Matsumoto, J. A. Possible triggering process of east–west asymmetric
anomaliesover the Indian Ocean in relationto 1997/1998 El Nin
˜o. J. Meteorol.Soc.
Jpn 78, 803–818 (2000).
19. Vecchi, G. A. & Soden, B. J. Global warming and the weakening of the tropical
circulation. J. Clim. 20, 4316–4340 (2007).
20. Zheng, X. T. et al. Indian Ocean Dipole response to global warming in the CMIP5
multimodel ensemble. J. Clim. 26, 6067–6080 (2013).
21. Cai, W. et al. Projected response of the Indian Ocean Dipole to greenhouse
warming. Nature Geosci. 6, 999–1007 (2013).
22. Wijffels, S. & Meyers, G. An intersection of oceanic waveguides: variability in the
Indonesian throughflow region. J. Phys. Oceanogr. 34, 1232–1253 (2004).
23. Feng, M. & Meyers, G. Interannual variability in the tropical Indian Ocean: a two-
year time scale of Indian Ocean Dipole. Deep-Sea Res. 50, 2263–2284 (2003) .
24. Wyrtki, K. An equatorial jet in the Indian Ocean. Science 181, 262–264 (1973).
25. Hong, C. C., Li, T., Lin, H. & Kug,J. S. Asymmetry of the Indian Ocean Dipole. PartI:
Observational analysis. J. Clim. 21, 4834–4848 (2008).
26. Taylor, K. E., Stouffer, R. J. & Meehl, G. A. An overview of CMIP5 and the
experimental design. Bull. Am. Meteorol. Soc. 93, 485–498 (2012).
27. Austin, P. C. Bootstrap methods for developing predictive models. Am. Stat. 58,
131–137 (2004).
28. Fischer, A. et al. Two independent triggers for theIndian Ocean dipole/zonalmode
in a coupled GCM. J. Clim. 18, 3428–3449 (2005).
29. Luo, J. J. et al. Interaction betweenEl Nin
˜o and extreme IndianOcean Dipole. J. Clim.
23, 726–742 (2010).
30. Izumo, T. et al. Influence of the state of the Indian Ocean Dipole on the following
year’s El Nin
˜o. Nature Geosci. 3, 168–172 (2010).
Supplementary Information is available in the online version of the paper.
Acknowledgements W.C. and E.W. are supported by the Australian Climate Change
Science Program,and the Goyder Institute. W.C. is also supported by a CSIRO Office of
Chief Executive Science Leader award. A.S. is supported by the Australian Research
Council.
Author Contributions W.C. conceived the study and directed the analysis. G.W. and
E.W. performed the model output analysis. A.S. conducted the heat budget analysis.
W.C. wrote the initial draft of the paper. All authors contributed to interpreting results,
discussion of the associated dynamics and improvement of this paper.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Readers are welcome to comment on theonline version of the paper. Correspondence
and requests for materials should be addressed to W.C. (wenju.cai@csiro.au).
RESEARCH LETTER
258 | NATURE | VOL 510 | 12 JUNE 2014
Macmillan Publishers Limited. All rights reserved
©2014
METHODS
Data, reanalyses and EOF analysis. We used data in the satellite era (1979–present)
which include Global Precipitation Climatology Project monthly precipitation ana-
lysis
31
, global analyses of SSTs
32
, and circulation fields from the National Center for
Environmental Prediction and National Center for Atmospheric Research global rea-
nalysis
33
. Ocean column data of velocities and temperatures for heat budget analysis
arebasedontheEuropeanCentre for Medium-Range Weather Forecasts - Ocean
Reanalysis System3 (ECMWF ORA-S3)
34
. We use a multivariate signal processing
method referred to as EOF analysis
35
to anomalies of rainfalland vertical velocity v
at 500 mb (ref. 33). The EOF techniquedeconvolves the spatio-temporalvariability
into orthogonal modes, eachdescribed by a principal spatial pattern and an assoc-
iated principal component time series.
Heat budget analysis. We examine the surface heat balance of the tropical Indian
Ocean which can be expressed as:
LTa=Lt~{½(uaLTa=LxzuLTa=LxzuaLT=Lx)
z(vaLTa=LyzvLTa=LyzvaLT=Ly)
z(waLTa=LzzwLTa=LzzwaLT=Lz)zQzresidual
ð1Þ
The variables T,u,vand ware potential temperature, and the zonal, meridional and
vertical ocean current velocities, respectively, averaged over the top 50 m. Differential
operators, x,y,zandt, are along the zonal, meridional and vertical directions, and time,
respectively. All variables are derived from the ECMWF ORA-S3 observational data
assimilation system
34
at a horizontal resolution of 1ulatitude by 1ulongitude, increasing
to0.3uinlatitude towards the Equator.Therate of change of themixedlayer temperature
(dT/dt) is calculated using a centred-difference approximation. Superscript ‘a’ and
overbar denote anomalous and long-term averaged quantities, respectively. Equation
(1) states that the rate of change or tendency of the surface temperature is balanced by
zonal advection of heat by the zonal current (first bracketed terms on the right hand
side), meridional advection (second bracketed terms), vertical advection (third brack-
eted terms), the net surface air–sea heat flux (Q), and all other factors not explicitly
expressed (residual), such as mixing and diffusion.
We use the entire reanalysis period of the ORA-S3, which spans 1959–2006, to
examine processes during the 1961 event. All variables in equation (1) are linearly de-
trended and averaged over the eastern-to-central equatorial region between 5uS–5uN
and 60uE–100uE, over which the 1961, 1994 and 1997 extreme events emerge as the
only pIODs that exhibit large anomalous cooling (Extended Data Fig. 1a). We examine
the heat budget terms averaged over the developing period of IndianOceandipole
events (July–October; Extended Data Fig. 1b). It may be noted that the 1997 event
exhibits exceptionally strong and prolonged cooling compared to the 1961 and 1994
events, which see an earlier start of the cooling at the end of boreal spring.
The nonlinear vertical advective process (waLTa=Lz), that is, the process associated
with anomalous upwelling and anomalous vertical temperature gradients, contributes
substantiallyto the coolingofthe equatorial Pacificduringthese events, especially during
moderate pIOD events (ExtendedData Fig. 1c). During the 1961, 1994 and 1997 events,
however, the nonlinear zonal advection term (uaLTa=Lx) is exceptionally strong,
extending notably farther to the west from the eastern Indian Ocean, as compared to
the other events (Extended Data Fig. 2a–d). Although the nonlinear vertical advection is
more prominent during the 1961 and 1997 events (Extended Data Fig. 1c), which is in
part also driven by the equatorial easterly winds, it is the nonlinear zonal advection that
setsthe1961, 1994 and 1997 eventsapartfromthe rest of the pIODs. Thisstemsfrom the
exceptionally strong westward current and its associated easterly winds (Extended Data
Fig. 2f).
As shown in Extended DataF ig.2f, the significant correlation between the zonal wind
and current (r50.87) means that the nonlinear zonal advection term over the equat-
orial region can be well approximated as a product between the zonal wind (averaged
over 5uS–5uN, 60uE–100uE) and the DMI
1
:
{uaLTa=Lx<txDMIað2Þ
with a~b=L,wherebis the regression coefficient between the zonal wind and zonal
current,andLis the longitudinal width of theequatorialbox(Extended Data Fig. 2a). This
parameterization is used to represent the nonlinear zonal advection term in the CGCMs.
Strikingly similar to the nonlinear zonal advection, the proxy exclusively identifies the
three observed extreme events. It may be noted that the proxy is further from the actual
value for the 1961 and 1997 events than for the 1994 event. This is expected, owing to the
particularlystrongnonlinearverticaladvection of the1961and 1997 events. Using93-year
time series of nine CMIP5 models for which we had access to the required variables, the
robustness of the proxy is signified by the high positive correlation coefficient with the
nonlinear advection, ranging from 0.59 to 0.92, significant above the 99% confidence level.
Characterization of extreme pIOD events. The extreme pIOD events were
characterized using a suite of distinctive process-based indicators, such as anomalous
equatorial easterlies, low rainfall and atmospheric subsidence as induced by a down-
stream extension of the southeasterly trade winds. For observations, we focus on histor-
ical events in the satellite era (1979–present) using Global Precipitation Climatology
Project monthly precipitation analysis
31
from http://www.esrl.noaa.gov/psd/data/
gridded/data.gpcp.html and SSTs
32
and other circulation fields from a global rea-
nalysis
33
. We focus on austral spring, September–November, in which a pIOD
typically peaks, and apply EOF analysis
35
to rainfall anomalies and v(ref. 33) at
500 mb in the equatorial Indian Ocean (40uE–100uE, 10uS–10uN). This produces
two principal variability patterns. The first principal pattern reflects a strong rain-
fall reduction over the eastern pole accompanied by a moderately rainfall increase
over the western equatorialIndian Ocean, and the second principal patternis char-
acterized by a westward extensionof rainfall reduction from the east, accompanied
by a rainfall increase farther west near equatorial east Africa.Note that the wetting
and northwesterly winds off Sumatra in EOF2are oppositeto the drying and south-
easterly anomalies in EOF1. Along the Equator, dry anomalies north of Sumatra
embedded in EOF2 oppose weak wet anomalies in EOF1. The opposing polarity
highlights that during extreme pIOD events, the centre of cold and dry anomalies
is not concentrated in the Sumatra region but shifts northward for a westward
extension along the Equator.
Model selection. We utilize 31 CMIP5 CGCMs forced with historical anthropogenic
and natural forcings, and future greenhouse gas under emission scenario of
Representative Concentration Pathway (RCP) 8.5
26
,coveringa200-yearperiod.Two
features of the nonlinearity associated with extreme pIODs are used to select models.
These are the negative skewness of SST anomalies over the eastern pole, and the non-
linear positive feedback along the Equator involving the west-minus-east SST gradient,
wind and oceanic currents, and nonlinear zonal advection, as indicated by a nonlinear
relationship between the two EOFs. These two features are not mutually inclusive and
are both used in our study.
Although the majority of CGCMs generate variability like that of the Indian Ocean
dipole, only a subgroupof CGCMs simulate the observed nonlinearocean–atmosphere
coupling over the eastern Indian Ocean as depicted by the negative skewness of SST
anomalies over the eastern pole during the austral spring (September–November),
which is 20.85 in observations since 1979.The level of nonlinearity varies vastly among
CGCMs, and we consider negative skewness of any extent. Out of the 31 CGCMs, 23
satisfytheSST skewness criterion. TheselectedCGCMs yield a mean skewnessof20.84,
close to the observed (Extended Data Table 1).
All selected 23 CGCMs reproduce the observed IOD pattern obtained by regressing
September–November SST anomalies onto the DMI, with a pattern correlation greater
than 0.75 (Supplementary Table 3). The same EOF analysis is carried out for each
individual model using rainfall anomalies referenced to the mean over the ‘control’
period. Prior to the analysis, data are interpolated into a common grid of1.5ulatitude by
1.5ulongitude. Our EOF outputs are scaled so that the EOF time series have a standard
deviation of one to facilitate an inter-model comparison and aggregation.Details of the
variance explained by EOF1 and EOF2 are listed in Supplementary Table 3. All 23
models produce the nonlinear relationship between the two leading rainfall EOFs,
indicating their abil ity to generate the nonlinear equatorial positive feedback associated
withtheextreme pIOD. Outputsofvat 500 mb areavailable in 21of the23 CGCMs, and
anonlinearrelationship between thetwoleading vertical velocity EOFsisgenerated in all
the 21 models.
We derive changes in the occurrence of extreme pIOD events by comparing the
frequency of the first 100 years (‘control’ period) to that ofthe second 100 years (‘climate
change’period).OftheeightCGCMswhicharenotabletosimulatethenegativeSST
skewness, only three CGCMs are not able to reproduce the nonlinear relationship
between the two rainfall EOFs, suggesting that the negative skewness of SST anomalies
in the eastern pole is not a prerequisite for the equatorial positive feedback associated
with extreme pIOD events. We also test the sensitivity of our results to varying defini-
tions (Supplementary Tables 1 and 2), including a case in which the criterion of the
negative SST skewness is excluded: that is, including all 31 CGCMs. In all cases, there is a
statistically significant increase (greater than a 130% increase) in the occurrences of
extreme pIOD events from the ‘control’ to the ‘climate change’ period.
Occurrences of extreme pIOD and the El Nin
˜o. The modelled increase in extreme
pIOD events is not induced by a change in the frequency in El Nin
˜o occurrences,
because there is no inter-model consensus between the two periods in the frequency
change of El Nin
˜o defined as when the quadratically detrended Nin
˜o3 (5uS–5uN,
150uW–90uW) SST is greater than 0.5 s.d. (Extended Data Fig. 7a), consistent with
previous studies
36,37
. Nor is there a statistically significa nt relationship between changes
in the number of extreme pIOD events and changes in the number of El Nin
˜oevents
(ExtendedDataFig.7a),extremeElNin
˜odefinedaswithNin
˜o3 rainfall greater than a
threshold value (Extended Data Fig. 7b)
38
,ordetrendedNin
˜o3 SST greater than a
threshold value (for example, 1.5 s.d) (Extended Data Fig. 7c). In addition, there is
no systematic change in the relationship between the Indian Ocean dipole and the El
Nin
˜o/Southern Oscillation (ENSO) (Extended Data Fig. 7d)
21
. Similarly, there is no
inter-model consensus on how Modoki El Nin
˜o, defined as occurring when the index
39
LETTER RESEARCH
Macmillan Publishers Limited. All rights reserved
©2014
is greater than a 0.5 s.d, will change. Nor is there a statistically significant relationship
between changes in the number ofe xtremepIOD events and changes in the number of
Modoki El Nin
˜o events (Extended Data Fig. 7e), and there is little change in the
relationship between the Indian Ocean dipole and the Modoki ENSO (Extended
Data Fig. 7f).
Statistical significance test. We use a bootstrap method
27
to examine whether the
change in frequency of the extreme pIOD events is statistically significant. The 2,300
samples from the 23 CMIP5 CGCMs in the ‘control’ period are re-sampled randomly
to construct another 10,000 realizations of 2,300-year records. In the random re-
samplingprocess,anyextreme pIOD event is allowed to be selected again. The standard
deviation oftheextremepIODfrequencyin the inter-realizationis11.2 events per 2,300
years,farsmallerthanthedifferenceof234eventsper2,300yearsbetweenthe‘control’
and the ‘climate change’ periods (Fig. 2c, d), indicatinga strong statistical significance.
The maximum frequency is 176, far smaller than that in the ‘climate change’ period of
367. Increasing the realizations to 20,000 or 30, 000 yields essentially an identical result.
To further confirm the statistical significance of our result with ample samples of
IOD behaviour across a longer time series without climate change forcing, we use a
Canadian model (CanESM2), in which a pre-industrial simulation of 996 years. We
examinetherarityofextremepIODeventrelativetothatinthe‘climatechange’period
with this same model. In the pre-industrial period the frequency is one per 13 years, bu t
in the ‘climate change’ period there is a 180% increase to one event per 5 years. In the
pre-industrial period, such an extreme pIOD event is far rarer.
Dividing the 996 years into 9 sets of 100 years and a set of 96 years, we find no
frequency in these sets is as high as that in the ‘climate change’ period. The lowest
frequency is one event per 25 years, and the highest frequency, one event per 7.7 years, is
50% lower than the frequency in the ‘climate change’ period. This highlights the
robustness of the greenhouse-warming-induced increase in the extreme pIOD fre-
quency, above that generated by natural variability, whichis represented by the spread
of inter-model differences.
31. Adler, R. F. et al. The version 2 Global Precipitation Climatology Project (GPCP)
monthly precipitation analysis (1979–present). J. Hydrometeorol. 4, 1147–1167
(2003).
32. Rayner, N. A. et al. Global analyses of sea surfacetemperature, sea ice, and night
marine air temperature since the late nineteenth century. J. Geophys. Res. 108,
4407 (2003).
33. Kalnay, E. et al. The NCEP/NCAR 40-Year Reanalysis Project. Bull. Am. Meteorol.
Soc. 77, 437–471 (1996).
34. Balmaseda, M. A., Vidard, A. & Anderson, D. The ECMWF ocean analysis system:
ORA-S3. Mon. Weath. Rev. 136, 3018–3034 (2008).
35. Lorenz, E. N. Empirical Orthogonal Functions and Statistical Weather Prediction.
Statistical Forecast Project Report 1 (MIT Department of Meteorology, 1956).
36. Guilyardi, E. et al. Understanding El Nin
˜o in ocean–atmosphere general
circulation models: progress and challenges. Bull. Am. Meteorol. Soc. 90,
325–340 (2009).
37. Collins,M. et al. The impact of global warmingon the tropical PacificOcean and El
Nin
˜o. Nature Geosci. 3, 391–397 (2010).
38. Cai, W. et al. Increasing frequency of extreme El Nin
˜o events due to greenhouse
warming. Nature Clim. Change 4, 111–116 (2014).
39. Ashok, K., Behera,S. K., Rao, S. A., Weng, H. & Yamagata,T. El Nin
˜o Modoki and its
possible teleconnection. J. Geophys. Res. 112, C11007 (2007).
RESEARCH LETTER
Macmillan Publishers Limited. All rights reserved
©2014
Extended Data Figure 1
|
Heat budget analysis of the extreme pIOD events
based on an ocean reanalysis
34
.a, Temperature anomalies averaged over
5uS–5uN and 60uE–100uE, over the top 50 m, and over September–
November. The filled blue and red circles indicate negative and positive DMI,
with the size of the markers indicating the relative strength of the DMI. b,The
rate of change of the temperatureanomalies as a function of calendar month for
all positive DMI values, with that of 1961 shown in green, 1994 in light red and
1997 in dark red, and all others in grey. c, The heat budget components
averaged over July–Octoberof Indian Ocean dipole development phase, for the
1961, 1994 and 1997 extreme events, and a composite of moderate pIOD events
in the satellite era (1982, 1987, 2002 and 2006). The uncertainty bar on each
composite represents the range of values over the four moderate pIOD events.
The nonlinear zonal advection term (uaLTa=Lx) (dark red in c) is particularly
large during the 1961, 1994 and 1997 events (see Methods for more details).
LETTER RESEARCH
Macmillan Publishers Limited. All rights reserved
©2014
Extended Data Figure 2
|
Nonlinear zonal advection term over the growth
phase of pIOD events. The nonlinear zonal advection term (uaLTa=Lx)
averaged over July to October for: a, a composite of moderate pIOD events,
b, the 1961 pIOD event, c, the 1994 pIOD event and d, the 1997 pIOD event.
The moderate pIOD events taken for the composite in aare the those in the
satellite era: the 1982, 1987, 2002 and 2006 events. Stippled locations in
aindicate composite values that are significant above the 95% confidence level
(P-value ,0.05) according to a Student’s t-test. e, The approximation of the
nonlinear advection term, uaLTa=Lx, averaged over the equatorial boxed
region (5uS–5uN and 60uE–100uE; as shown in a) using the product between
the corresponding zonal wind stress and the DMI (see Methods). The DMI is a
measure of zonal gradient of temperature anomalies averaged over the western
and eastern boxed regions in a.f, The total zonal current versus totalzonal wind
stress averaged over the equatorial box region in a. A particularly strong zonal
current reversal is seen during the 1961, 1994 and 1997 pIOD events (large red
dots in f, see Extended Data Fig. 1a).
RESEARCH LETTER
Macmillan Publishers Limited. All rights reserved
©2014
Extended Data Figure 3
|
Circulation anomalies associated with the
principal variability patterns of austral spring (September–November)
rainfall. a–c, Vertical velocity vat 500 mb (Pa s
21
) from reanalysis data
33
(positive indicating descending motion) (a), SST (uC) (ref. 32) (b) and
thermocline depth (m) (ref. 34)(c) anomalies associated with the first principal
variability pattern (Fig.1c). The patterns are obtained through linear regression
of the corresponding variables onto the principal component time series of
EOF1. d–f, The same as for a–c, but for the second principal variability pattern
(Fig. 1d).
LETTER RESEARCH
Macmillan Publishers Limited. All rights reserved
©2014
Extended Data Figure 4
|
Reconstruction of an extreme pIOD and a
moderate pIOD event using the first two principal rainfall variability
patterns. a–d, Composite of anomalies associated with the 1994 and 1997
extreme pIOD events, showing the observed rainfall and wind stress
anomalies, and anomalies reconstructed from the first principal, the second
principal, and the first and second principal components combined, using
satellite-era rainfall anomaly data from the Global Precipitation
Climatology Project version 2 (ref. 31) and reanalysis wind stress
33
. Note
the different vector scales shown in the top right corner for each panel.
e–h, The same as a–d, but for composites of anomalies associated with the
1982, 1987, 2002 and 2006 moderate pIOD events. The exercise highlights
that the difference between a moderate and an extreme pIOD depends on
the role of the second principal component, and can only be realized with
the use of both of the two principal components.
RESEARCH LETTER
Macmillan Publishers Limited. All rights reserved
©2014
Extended Data Figure 5
|
Principal variabilitypatterns of vertical velocity at
500 mb (
v
), their nonlinear relationship, and the associated wind stress
vectors during austral spring (September–November), based on a
reanalysis
33
. A positive vertical velocity indicates descending, while a negative
vindicates ascending motion. a,b, Spatial patterns obtained by applying a
statistical and signalprocessing method—EOF analysis—to thevertical velocity
anomalies in the equatorial region (10uS–10uN, 40uE–100uE) for data since
1979. The associated pattern and wind stress vectors from reanalysis data are
obtained by linear regression onto the principal component time series of the
EOFs. The first and second principal spatial pattern accounts for 32.6% and
16.8% of the total variance. The colour scale indicates vertical velocityin Pa s
21
per 1 s.d. change; blue or red contours indicate increased or decreased
convection. Note the different vector scales shown in the top right corner in
aand b. c, A nonlinear relationship between the associated principal
component time series. An extreme pIOD event (red dots) is defined as when
the first principal component is greater than 1 s.d., and the second principal
component is greater than 0.5s.d. A moderate pIOD event (green dots) is
determined from a detrended DMI when its amplitude is greater than 0.75 s.d.,
except for the 1994 and 1997 extreme pIOD events. Negative IOD and neutral
years are indicated with blue dots.d, Relationship between the second principal
component time series and rainfall over the eastern equatorial Pacific (Nin
˜o3)
region (5uS–5uN, 150uW–90uW). While the 1997 extreme pIOD was
associated with a large rainfall in the Nin
˜o3 region, the 1961 and 1994 extreme
pIODs were not.
LETTER RESEARCH
Macmillan Publishers Limited. All rights reserved
©2014
Extended Data Figure 6
|
Multi-model ensemble average of the principal
variability patterns of vertical velocity at 500mb (
v
), their nonlinear
relationship, and the associated wind stress vectors during austral spring
(September–November). A positive vertical velocity indicates descending,
while a negative vindicatesascending motion. a,b, Spatial patterns obtainedby
applying a statistical and signal processing method—EOF analysis—to the
vertical velocityanomalies in the equatorial region (10uS–10uN,40uE–100uE).
The associated pattern and wind stressvectors are obtained by linear regression
onto the principal component time series. The colour scale below gives vertical
velocity in m s
21
per 1 s.d. change; blue or red contours indicate increased or
decreased convection. Note the different vector scales shown in the top right
corner in aand b.cand d, A nonlinear relationship between the two principal
component time series for the ‘control’ (1900–1999) and ‘climate change’
(2000–2099) periods. An extreme pIOD event (red dots) is defined as when the
first principal component is greater than 1 s.d. and the second principal
component is greater than 0.5s.d. A moderate pIOD event (green dots) is
determined from a detrended DMI when its amplitudeis greater than 0.75s.d.
but is not an extreme pIOD event. NegativeIOD and neutral years are indicated
with blue dots. Number of extreme pIOD and moderate pIOD events is
indicated in cand d.
RESEARCH LETTER
Macmillan Publishers Limited. All rights reserved
©2014
Extended Data Figure 7
|
Multi-model statistics between El Nin
˜o and pIOD
in selected CGCMs. a, Changes (‘climate change’ minus ‘control’ period) in
the number of occurrences of extreme pIOD events versus changes in the
number of El Nin
˜o events defined as when the amplitude of the detrended
Nin
˜o3 (5uS–5uN, 150uW–90uW) SST index is greater than 0.5 s.d. b, Changes
in the number of extreme pIOD events versus changesin the number of El Nin
˜o
events defined as when the Nin
˜o3 total rainfall is greater than 5 mm day
21
as in
ref. 38. c, The same as b, except an extreme El Nin
˜o is determined from a
detrended Nin
˜o3 (5uS–5uN, 150uW–90uW) SST index when its amplitude is
greater than 1.5 s.d. d, Correlation between a detrended Nin
˜o3 index and a
detrended DMI index
1
for the ‘climate change’ (yaxis) and the ‘control’ periods
(xaxis). e, Changes in the number of occurrences of extreme pIOD events
versus changes in the number of Modoki El Nin
˜o events defined as when the
amplitude of a detrended index
39
(see Methods) is greater than 0.5 s.d.
f, Correlation between a detrended El Nin
˜o index and a detrended DMI index
for the ‘climate change’ (yaxis) and the ‘control’ periods (xaxis). The inter-
model correlation and its statistical significance or otherwise are indicated in
the bottom right corner of each panel, with a P-value less than 0.05, indicating
significance above the 95% confidence level, a condition not met in a,b,cand
e. Models with a stronger relationship between ENSO and the Indian Ocean
dipole in the ‘control’ period tend to have a stronger such relationship in the
‘climate change’ period, and the tendency is statistically significant, although
the relationship weakens slightly in the ‘climate change’ period. The same is
true for the Modoki relationship between ENSO and the Indian Ocean dipole.
LETTER RESEARCH
Macmillan Publishers Limited. All rights reserved
©2014
Extended Data Figure 8
|
Multi-model ensemble average of mean state
changes for selected CGCMs. The changes (‘climate change’ minus ‘control’
period) of the ensemble average mean state for: a, rainfall (mm day
21
), b, SST
(uC), c, wind stress vectors (N m
22
) and d, thermocline depth (m). The result
shows that rainfalloff Sumatra is decreasing, the southern eastern IndianOcean
is warming at a slower rate than the west, there is a trend of easterlies over the
equatorial Indian Ocean, and the thermocline is shallowing in the eastern
equatorial Indian Ocean. Areas where changes are statistically significant at the
95% confidence level are indicated with stipples, in a, b, and d.Inc, vectors in
bold indicate statistical significance at the 95% confidence level.
RESEARCH LETTER
Macmillan Publishers Limited. All rights reserved
©2014
Extended Data Figure 9
|
Schematic of extreme pIOD in response to
greenhouse warming. a, pIOD events are characterized by westward-flowing
wind anomalies (blue arrow at the surface) and the associated westward-
flowing current anomalies (blue arrow at depth) acting against the prevailing
background eastward circulations (black arrows), in association with the
anomalous positive west-minus-east SST gradient. These result in generally
weaker-than-normal eastward atmosphere and ocean circulations (grey
arrows), with anomalously wet condition in the west and dry in the east.
b, During extreme pIOD events, these anomalies are amplified, with
occurrences of strong reversals of the mean eastward winds and currents (grey
arrows). As the mean Walker circulation and the associated eastward-flowing
ocean current weaken (red arrows), wind and current reversals (orange arrows)
can occur more easily in association with pIOD anomalies. Greenhouse
warming thus induces more frequent occurrences of extreme pIOD events.
LETTER RESEARCH
Macmillan Publishers Limited. All rights reserved
©2014
Extended Data Table 1
|
Performance of 23 selected CMIP5 CGCMs forced under climate change emission scenario RCP8.5
These CGCMs are selected in terms of SST skewness in the eastern pole of the Indian Ocean dipole (IODE) and each model’s ability to simulate the nonlinear relationship between rainfall EOF1 and EOF2 (Fig. 2).
The sensitivity of changes in extreme pIOD events from the ‘control’ period to the ‘climate change’ period to different definitions is tested. An extreme pIOD event is defined as when the first principal component
time series is greater than 1 s.d., or 1.5s.d., and the second principal component time series is greater than 0.5 s.d. Numbers in red type indicate a decrease from the ‘control’ period (1900–1999) to ‘climate
change’ period (2000–2099). Multi-model average SST skewness in the eastern pole of the Indian Ocean dipole is 20.84, compared with the observed value of 20.85. The subscripts SST, vand tindicate that the
data of SST, vertical velocity at 500 mb and surface wind stress are available, respectively.
RESEARCH LETTER
Macmillan Publishers Limited. All rights reserved
©2014
