The relationship between Indian Ocean sea surface temperature and the transition of El Niño events into either La Niña or El Niño–Southern Oscillation neutral conditions is examined in both observations and the retrospective ensemble hindcasts of the National Center for Environmental Prediction Climate Forecast System. The southern Indian Ocean is shown to demonstrate a particularly robust and consistent relationship with the evolution of these transitions. These associations are described, and a physical mechanism involving air-sea interaction in the Indian and western Pacific Oceans is proposed. Observations suggest that easterly surface wind anomalies in the western Pacific Ocean are associated with the emergence of La Niña during boreal summer and fall. Here it is shown that these winds are significantly correlated to southern Indian Ocean sea surface temperature in the preceding spring that is characterized by a large-scale zonal dipole of cool and warm anomalies in the southwestern and southeastern Indian Oceans, respectively. These associations are particularly pronounced for strong El Niño conditions, during the dissipation of which a pronounced wavetrain-like atmospheric pattern accompanies sea surface temperature anomalies in the southern Indian Ocean. Together, the circulation and sea surface temperature anomalies increase the meridional cross-equatorial temperature gradient in the western Indian Ocean and mute intraseasonal variability while strengthening surface equatorial easterly winds in the Indo-Pacific warm pool. Collectively, these anomalies favor subsequent La Niña development. On the basis of these observed associations, a predictive model that demonstrates skill in anticipating the nature of El Niño transitions, involving the southern Indian Ocean, Asian monsoon, and El Niño–Southern Oscillation, is proposed. In the National Center for Environmental Prediction Climate Forecast System, the relationships described above are simulated both consistently and realistically, despite model weaknesses, further bolstering a key role of southern Indian Ocean and predictive relationship. Comparison of fully coupled and sea surface temperature–forced simulations suggests a key role for air-sea interaction in the observed associations. Moreover, it is demonstrated that coupled simulations of El Niño–Southern Oscillation may benefit substantially from improved representation of Indian Ocean variability and Indo-Pacific interaction.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The coupled interaction between the IO and ENSO demonstrates a significant influence on regional [Wu and Kirtman, 2007; Yoo et al., 2006 (hereafter Y06); Yu et al., 2002] and global scales [Annamalai et al., 2007]. In Y06, the role of southern IO (SIO) SST in this influence is addressed specifically, whereby anomalies in the preceding spring are strongly associated with an anomalous circulation in the WPO in summer. Terray et al.  demonstrate that the southeastern IO (SEIO) exhibits a leading association with many modes of Indo-Pacific variability including ENSO, the Indian and Australian summer monsoons, the tropical Indian Ocean dipole, and maritime continent rainfall. In cross-validated hindcasts of linear models, Dominiak and Terray  examine the predictive potential of three ENSO precursors, including the anomalous equatorial Pacific upper-ocean heat content, the zonal wind stress anomaly in the WPO, and SEIO SST anomalies. They show that SEIO SST during late boreal winter is the dominant ENSO precursor and that forecast skill is improved considerably for the models that exploit it as a predictor. While the authors of current study conjecture that the SEIO SST anomalies trigger zonal wind anomalies in the WPO, they do not suggest a possible mechanism whereby their influence persists into late spring and summer. Thus, the large role of the SIO in Indo-Pacific variability and the mechanisms by which its influence is realized remain largely uncertain.
 While studies to date have focused on the potential influence of tropical and northern IO (NIO) SST on the triggering of wind anomalies in the WPO and ENSO, the role of the SIO has received considerably less attention and several questions therefore exist regarding its broader significance. For example, what are the dominant features of SIO SST, precipitation, and large circulation patterns associated with ENSO and, in particular, the decay of El Niño? What is the relative strength of the ENSO association as manifested in northern and southern IO SST? What are plausible mechanisms by which the Indian Ocean may influence ENSO? Does air-sea interaction play an important role in simulating and predicting the relationship between IO SST and ENSO in fully coupled models?
 To address these questions, we focus on the observed seasonal evolution of IO and Pacific SST and their associated atmospheric variability. To quantify the role of air-sea interaction, a comparison is made between the coupled and uncoupled simulations of the National Centers for Environmental Prediction (NCEP) Coupled Forecast System (CFS). In section 2, descriptions of the data, model simulations, and analysis methods are presented. Sections 3 and 4 develop an observational context for our analysis including the identification of the dominant features of El Niño transitions into La Niña and ENSO neutral conditions (section 3) and the contrasting associations between the northern and southern IO (section 4). The possible processes and mechanisms explaining these associations are also explored in this section. Building on this diagnostic effort, the role of air-sea interactions and the reproducibility of these modes in the NCEP coupled and uncoupled models are addressed in section 5. Finally, a summary of the main findings and their broader implications is given in section 6.
2. Observational Data, Model Hindcasts, and Analysis Methods
2.1. Observational Data
 The precipitation and wind fields used in this study are from the Climate Prediction Center Merged Analysis of Precipitation (CMAP) [Xie and Arkin, 1997] and the NCEP/Department of Energy Reanalysis 2 (DOE R2), both at a resolution of 2.5°. We also analyze the NOAA optimally interpolated SST analysis [Reynolds et al., 2002] on a 1° × 1° grid. The period from 1982 to 2006 is assessed for several reasons. First, reliable estimates of precipitation over oceans commenced in 1979 and only slightly thereafter did the data assimilated in the interpolated SST product become approximately homogenous. Second, ENSO precursors and large-scale associations changed significantly following the 1976−1977 climate regime shift [Wang, 1995], and it is therefore questionable whether a single analysis encompassing the pre-1976 and post-1976 period is appropriate. Lastly, the output of the NCEP CFS hindcasts is only available for periods after 1981. Thus, prior to 1982, the issues of data homogeneity and availability and climate stationarity hamper the linear analysis attempted in the current study.
 We consider the Niño3.4 SST averaged in the eastern Pacific (120°W−170°W, 5°S−5°N) as an ENSO index and define warm (El Niño) and cold (La Niña) episodes as exceeding the threshold of ±0.5°C for a minimum of five consecutive overlapping seasons. On the basis of this criterion, during 1982−2006, there were eight El Niño onsets (e.g., 1982, 1986, 1991, 1994, 1997, 2002, 2004, and 2006) and four La Niña onsets (e.g., 1983, 1988, 1995, and 1998). It is therefore evident that La Niña follows only about half of El Niño events, and it is the SIO SST that may play a role in this ENSO evolution. Whether a transition from El Niño to La Niña will occur may also be related to the location of tropical Pacific warming, which has changed between the eastern Pacific and the central Pacific in the past decades [e.g., Larkin and Harrison, 2005; Ashok et al., 2007a; Yu and Kao, 2007; Kao and Yu, 2009; Kim et al., 2009]. The decadal change in Pacific warming has led to claims of a different type of El Niño (dateline El Niño [Larkin and Harrison, 2005], El Niño Modoki [Ashok et al., 2007a], central Pacific El Niño [Kao and Yu, 2009], and warm pool El Niño [Kug et al., 2009]) from the conventional type of El Niño whose SST anomalies are mainly in the eastern Pacific. The mechanisms for the different types of tropical Pacific warming or El Niño have been explained by wind-induced variation in the thermocline [Ashok et al., 2007a], the interaction between ENSO and the Pacific mean state [Sun and Yu, 2009], equatorial zonal ocean advection [Kug et al., 2009], and changes in subtropical forcing and the Hadley cell [Kao and Yu, 2009; Yu et al., 2010]. A detailed discussion of these mechanisms is beyond the scope of this study.
2.2. Model and Hindcasts
 This study applies the NCEP CFS [Saha et al., 2006], in which the NCEP atmospheric Global Forecasting System (GFS) [Perumal et al., 2001] is coupled to the Geophysical Fluid Dynamics Laboratory (GFDL) Modular Ocean Model version 3 (MOM3) [Pacanowski and Griffies, 1998]. The GFS has a T62 lateral resolution, equivalent to approximately a 180 km grid spacing in the tropics, and 64 layers in the vertical, with the uppermost layer at 0.2 hPa. The GFDL MOM3 has a longitudinal resolution of 1° with a finer latitudinal resolution of 1/3° between 10°S and 10°N that increases linearly to 1° by 30°N and 30°S and is fixed at 1° poleward of 30°S and 30°N. There are 40 layers in the vertical with 27 layers in the ocean's upper 400 m and a maximum depth of 4500 m. The model's vertical resolution is 10 m between the surface and 240 m, and it increases linearly from 10 to 511 m between 240 and 4500 m. The land surface model in the CFS is the Oregon State University two-layer land model [Pan and Mahrt, 1987]. The atmospheric and oceanic models are coupled without any flux adjustment, and air-sea interaction occurs using daily mean fluxes and SST between 65°S and 50°N. Sea ice extent is prescribed from the observed climatology. Further details on the modeling framework are described in detail in the works of Wang et al.  and Saha et al. .
 The Atmospheric Model Intercomparison Project (AMIP) [Gates et al., 1999] simulations are also used to contrast the coupled model response with the atmospheric response driven by prescribed SST. We use the ensemble mean of five members derived from perturbed SST forcing from 1982 to 2006 in order to gauge a range of AMIP runs.
 In the hindcast runs, the oceanic and atmospheric initial conditions (ICs) are perturbed from the real-time NCEP Global Ocean Data Assimilation and NCEP/DOE AMIP II R2 [Kanamitsu et al., 2002], respectively. Each hindcast run involves a 9 month integration of 15 members using the ICs of days 9−13, days 19−23, and the last 2 days of the previous month and days 1−3 of the concurrent month. For example, the zero month lead (LM0) June is the forecast for June using the ICs of 9−13, 19−23, and the last 2 days of May and first 3 days of June. The 1 month lead (LM1) July is the month of forecast using the same ICs. We define LM0 MAM as a seasonal mean of LM0 March, LM1 April, and LM2 May, and LM3 JJA as a mean of LM3 June, LM4 July, and LM5 August. Similarly, LM6 SON is computed as a mean of LM6 September, LM7 October, and LM8 November. The integrations cover all months from 1981 to 2006. In this study, we use the ensemble mean of 15 members and the 25 year (1981–2006) average to establish the climatological model means. More about these hindcasts and the performance of the model in simulating and predicting the climate over Asia, IO, and WPO can be found in the works of Yang et al.  and Liang et al. .
2.3. Analysis Methods
 In order to explore linear relationships between the Indian Ocean and ENSO, we apply an empirical orthogonal function (EOF) analysis to identify the leading modes of SST and precipitation variability. We also apply a singular value decomposition (SVD) analysis to depict the relationship between IO SST and large-scale precipitation and show the results in the form of heterogeneous regression patterns of the dominant SVD modes [see Bretherton et al., 1992; Y06]. Linear correlation and regression analyses are also applied and the two-tailed Student's t test is used to determine the statistical significance of the results.
3. Dominant Features of ENSO in Summer
 In order to gain an understanding of the large-scale features related to ENSO during its development, we generate an EOF analysis of precipitation and regress winds and SST against the dominant modes of precipitation for boreal summer. Figure 1 shows the leading EOF modes of precipitation for 30°S–30°N/30°E–160°E as a regression map of precipitation against their principal components (PCs, bottom). The first leading EOF is characterized by positive precipitation loadings along the tropical NIO and the maritime continent and negative loadings over the WPO. This mode explains 20% of the total variation (Figure 1a). Unlike the first mode, large loadings in the second mode exist over the South China Sea (SCS), WPO, the maritime continent, and NIO (Figure 1b). The third mode is likely related to the Indian Ocean dipole, which is a dominant regional ocean-atmosphere coupled mode in the tropical IO [Webster et al., 1999; Saji et al., 1999; Ashok et al., 2001] that drives strong zonal asymmetry (Figure 1c). The PCs reveal strong associations between ENSO and the first EOF, with ENSO onset lagging PC1 but leading PC2. The PCs also highlight the mixed perspectives of the Indian Ocean dipole as being forced by (the first EOF mode) but mechanistically independent of (the third EOF mode) ENSO.
 The lag correlations between the dominant rainfall modes and seasonal mean Niño3.4 SST are calculated to better understand their relationship. The first EOF mode of summer precipitation has a significant and robust correlation (correlation coefficient, r = −0.8) with Niño3.4 SSTs from JJA through the following winter (Figure 2) despite exhibiting a weak correlation with the SSTs of preceding seasons. In contrast to the first mode, the second mode has a significant correlation of about −0.6 prior to JJA followed by a weaker relationship. On the basis of this linear analysis, Niño3.4 SST thus lags the first and leads the second rainfall modes and are thus interpreted as reflecting ENSO-developing and ENSO-decaying conditions, respectively. In contrast, the third mode is not significantly correlated with ENSO, which is consistent with previous studies asserting the independence of the IOD from Pacific influences [e.g., Behera et al., 1999; Saji et al., 1999; Webster et al., 1999; Rao et al., 2002; Yamagata et al., 2002; Ashok et al., 2003; Yu et al., 2003; Shinoda et al., 2004; Yu and Lau, 2004]. So we focus on the first and second modes in the subsequent discussion.
 The regression of 850 hPa winds and SST against the leading modes of precipitation (Figure 1) provides insight into the large-scale circulation and SST patterns associated with the ENSO-developing and ENSO-decaying modes. The ENSO-developing mode (Figure 3a) is characterized by contemporaneous easterly anomalies over the WPO, an anomalous anticyclonic circulation pattern over northwestern Pacific, and anomalous northerly winds across the northeast coast of Australia. These features are canonical precursors to ENSO and are associated with the emergence of La Niña. The ENSO-decaying mode (Figure 3b) is associated with a northward propagation of Rossby waves in response to deep convection in the WPO during the termination of La Niña [Huang, 1984; Kurihara and Tsuyuki, 1987; Nitta, 1987; Tsuyuki and Kurihara, 1989; Lau and Peng, 1992]. Anomalous circulation patterns include an anomalous cyclonic circulation over the SCS and an anticyclonic circulation over East Asia and the East China Sea in conjunction with a strong Southeast Asian monsoon and weak East Asian monsoon. The features are consistent with the result of Yim et al. , who show that the Southeast Asian and East Asian summer monsoons are strongly anticorrelated when ENSO is weak.
4. Observed Association Between Indian Ocean SST and ENSO
 The contrasting annual cycles of the Indian and Pacific basins have been well documented [e.g., Torrence and Webster, 1998; Annamalai et al., 2003, 2005; Wang et al., 2003; Li et al., 2003; Terray et al., 2005; Y06] and a significant contrast exists in their persistence, as the tropical central Pacific is marked by a minimum in spring whereas anomalies in the IO lose their persistence in summer [Yasunari, 1991]. Xie et al.  showed that El Niño–induced IO warming persists through the summer after El Niño dissipated in spring and acts like a capacitor anchoring atmospheric anomalies over the Indo-Western Pacific Oceans. Using a coupled atmosphere-ocean general circulation model, Yu et al.  verified that IO plays a more critical role than Pacific Ocean in influencing tropical climate variability in spring whereas Pacific SSTs are more influential in fall. The summer persistence minimum of the IO can be attributed in part to the interaction between the IO (Y06), ENSO, and the neighboring atmospheric conditions including the convection over the maritime continent. Thus, the summer regional IO climate is more significantly correlated with preceding IO conditions than with those in the Pacific. In contrast, the tropical central Pacific exhibits a stronger relationship with precipitation and atmospheric circulation in boreal fall than does the IO (e.g., Figure 7 in Y06), thus suggesting that the springtime IO may act as a source of persistence for the ENSO system that contributes to overcoming its regional persistence minimum.
4.1. Relative Importance of NIO and SIO SSTs
 To investigate the lagged relationship between IO SST and ENSO modes, the regression of spring SST against the ENSO-developing and ENSO-decaying modes is examined. The ENSO-developing mode (Figure 4a) is more directly related to MAM SIO SST than to Pacific SST (as in Figure 2). On the other hand, the ENSO-decaying mode correlates significantly with ENSO anomalies in boreal spring (Figure 4b), including cold SST anomalies over the IO and eastern central Pacific Ocean and warm SST anomalies in the WPO and midlatitude Pacific Oceans. Salient features of the IO SST patterns associated with ENSO include a strong relationship between the ENSO-developing mode and SIO SST anomalies characterized by zonal dipole pattern and a correspondence between the ENSO-decaying mode and basin-wide variability in the NIO. These features also suggest that ENSO modes are associated strongly with the preceding IO state.
 In Y06, the relative importance of the preceding NIO and SIO SST anomalies for summer regional climate was explored. The precipitation pattern of the leading SVD mode associated with the MAM NIO variability (Figure 9a of Y06) is similar in form to the second EOF of summer precipitation (Figure 1b), exhibiting a pattern correlation of −0.78. However, the pattern correlation with the leading EOF (Figure 1a) is weaker than the correlation with second EOF (r = −0.42). In contrast to NIO SST, the summer precipitation pattern of the leading SVD mode related to spring SIO SST (Figure 9c in Y06) resembles strongly the leading EOF of precipitation (Figure 1a, r = 0.95) and it is not correlated significantly to the second EOF (r = 0.08). In order to explore these relationships further, we have also performed the regression with the leading EOF modes of MAM NIO and SIO SSTs for an extended period and obtained similar results (figure not shown). The regression patterns of 850 hPa winds and precipitation against the leading EOF modes of NIO and SIO SSTs are similar to those of wind and SST generally (Figures 3b and 3a, respectively). The similarity thus supports our hypothesis that springtime NIO (SIO) SST is strongly associated with the ENSO-decaying (developing) mode in summer and the relationships in rainfall modes is representative of variability generally.
 The sliding correlations between the leading EOF modes of MAM IO SST and the seasonal means of Niño3.4 SST further highlight the regional phase relationships (Figure 5). The correlation coefficients for the entire IO, NIO, and SIO SSTs are examined (Figure 5a). Positive SIO is characterized by cold (warm) SST anomalies over the southwestern (southeastern) IO. In general, when the entire IO is considered, strong lag correlations are evident (0.6 < r < 0.8) but lead relationships are weak generally. NIO SST is also highly correlated with the Niño3.4 SST in the preceding seasons. In contrast, the SIO SST is strongly correlated with the Niño3.4 SST of both preceding and subsequent seasons, thus bolstering our earlier speculation that MAM SIO SST may influence both the onset and phase transition of ENSO actively whereas the NIO SST acts largely in response to ENSO. In Figure 5b, the time series of leading EOF PCs of the springtime IO, NIO, and SIO SSTs are shown from 1982 to 2006. For comparison, boreal fall Niño3.4 SST is also shown. There are four La Niña onsets and seven El Niño onsets during this period, marked by blue and red closed circles, respectively. In general, La Niña develops frequently following El Niño, though not all El Niño events transition to La Niña (e.g., during tropical central Pacific warming). Those that do transition to La Niña exhibit a strong relationship to SIO SST with the positive SIO SST pattern yet consistent relationships with the entire IO or NIO warming are lacking. The spatial evolution of SST from the preceding winter throughfall (figure not shown) further confirms that the transition from El Niño to La Niña accompanies positive SIO SST anomalies in boreal spring, particularly in the SEIO.
4.2. Predictive Potential and Possible Mechanism
 To diagnose the large-scale ocean-atmosphere circulation patterns associated with anomalies in the NIO and SIO and to further understand the potential interaction between IO SST and ENSO, we carry out composite analysis of major events for several variables. Events exceeding 0.5 standard deviation of the PCs of NIO and SIO SSTs from Figure 5b are selected for positive NIO (seven) and SIO (six) composites, respectively.
Figure 6 shows anomalies in SST and 850 hPa winds for NIO (left) and SIO (right) warm extremes from the preceding winter to fall. In Figures 6a and 6e, for both the NIO and SIO, positive SST anomalies occur during El Niño in DJF. However, El Niño rapidly decays in MAM and changes its phase to La Niña in JJA for the SIO composites, whereas El Niño gradually decays through MAM and JJA for the NIO composites, with no transition to La Niña.
 For the SIO composites during DJF (Figure 6e), the easterly and northwesterly anomalies associated with El Niño were examined by An , who found the basin-wide warming north of 20°S in spring to be caused primarily by ENSO and to be associated with an increase in heat content associated with surface easterly wind anomalies over the western IO and an increase in solar radiation associated with the suppressed convection via an atmospheric bridge over the eastern IO. Similarly, here it is seen that the perturbed Walker circulation tends to weaken the climatological westerly and anticyclonic circulation over the tropical IO and SIO, causing warm SST anomalies over the SEIO through the wind-evaporation feedback that reduces the latent and sensible heat release from the ocean surface. The perturbed Walker circulation induces cold SST anomalies in the WPO as a result of upwelling induced by strong surface easterlies and related upwelling and turbulent fluxes. In addition, changes in the wind field may induce anomalies in vertical mixing [Meehl, 1997] and in cross-equatorial Ekman ocean heat transport [Loschnigg and Webster, 2000; Loschnigg et al., 2003], all of which introduce long-lasting anomalies into the heat balance and SST distribution in the Indian Ocean. The tropical IO warming during El Niño is associated with southeastward wave propagation from Madagascar to the south of Australia, which is also linked to SST anomalies over the SIO and, eventually, the decay of El Niño itself through the eastward upwelling Kelvin wave [Kug et al., 2006a].
 For the SIO composites during MAM (Figure 6f), easterly anomalies appear in the WPO. These anomalies are known to be an important factor for both the onset and transitions of ENSO through their association with eastward propagating upward Kelvin waves driven by intraseasonal variability. Here we also identify a substantial association between the variance in the 20–70 day band over the warm pool, as diagnosed from outgoing longwave radiation (OLR) and SEIO anomalies in the 6 months (February–July) centered about El Niño transitions. Years of a strong SIO zonal contrast and transition to La Niña conditions exhibit variance of 59.5 ± 25.7 (W m−2)2, while during the termination of events that do not transition to ENSO neutral conditions, the variance is 150.3 ± 27.0 (W m−2)2. These sets are distinct at the 98% confidence limit. Here the timing of events may be suggestive of causality since the SIO anomalies predate the anomalies in intraseasonal variance and are associated with large-scale persistent anomalies formed during the preceding El Niño. Nonetheless, to explore the causal relationships fully, a modeling effort targeting these interactions is required.
 Surface easterly anomalies are related to the weakening and eastward displacement of the Walker circulation in conjunction with the decay of both El Niño and the SST gradient between the IO and WPO [e.g., Diaz et al., 2001]. Along with the eastward movement of Walker circulation, the broader atmospheric circulation, SST anomalies, and wave train patterns over the SIO migrate eastward over time. Although El Niño decays in MAM, the SIO SST anomalies it originally induces persist in conjunction with the wave patterns, which are strongest during MAM and linked to the southwesterly and northeasterly anomalies over cold and warm SST anomalies from DJF to MAM. These midlatitude SST anomaly patterns are strongly influenced by tropical processes, such as ENSO and the Southern Annular Mode fluctuating the surface winds in the circumpolar belt between 40°S and 60°S [Ashok et al., 2007b]. The enhanced midlatitude westerlies contribute to intensification of the cold SST anomalies through enhanced heat loss [Huang and Shukla, 2007, 2008].
 The warming of the SIO from DJF to MAM reduces the temperature gradient between the Northern and Southern hemispheres (whereas NIO warming increases the temperature gradient). For the SIO composites during JJA (Figure 6g), the decreased temperature gradient is associated with a weakening of the cross-equatorial and monsoonal flows, which also extend into the WPO. The southward wind anomalies over NIO converge with the anomalous anticyclonic circulation over SIO and together contribute to moisture convergence and increase the rainfall over the tropical IO (Figure 1a). The warm SST anomalies persist to the northwest of Australia and do not mirror the northward displacement of the ITCZ during summer. The wave patterns over SIO also move eastward and weaken during summer. However, the anomalous anticyclonic circulation associated with northerly flow west of Madagascar and southerly flow west of Australia strengthens the climatological anticyclonic circulation over the SIO and is linked to dissipation of SIO SST anomalies through wind-evaporation-SST feedback. For the SIO composites during SON (Figure 6h), zonal wind anomalies associated with the La Niña perturbed Walker circulation appear over the IO, suggesting an influence of the large-scale adjustment of tropical circulation on the IO during boreal fall.
 The climatology of 850 hPa winds and precipitation in JJA and the SIO composite anomalies are displayed to see the features related to the SIO warming (Figure 7). A strong anticyclonic circulation is the dominant feature of low-level circulation over SIO in DJF and MAM (see Figure 2 in Y06). In JJA (Figure 7a), the strongest anomalies lie in the cross-equatorial flow over the western IO, in conjunction with the anomalous meridional temperature gradient. This flow is connected to the Somali jet and is known to be a key source of moisture and mass to the NIO during the monsoon [e.g., Fasullo and Webster, 2002], and its intensification is associated with monsoon onset in late spring and establishment of monsoonal deep convection [e.g., Fasullo and Webster, 2003]. Along this monsoon flow, the rainfall band moves northward near 20°N during summer. Over the Pacific, a climatological strong anticyclonic circulation is located over the northwest and interacts with the Southeast Asian and East Asian monsoons. It was shown that, during JJA, the NIO warming tends to accompany a strengthened meridional temperature gradient (Figure 6c), which is associated with a strengthened cross-equatorial flow over the western IO. However, persistent warming over the NIO also reduces the temperature gradient between the NIO and the surrounding land region, which may, in turn, weaken the monsoon flow. In contrast to the NIO composites, the SIO JJA composites (Figure 6g) demonstrate a convergence between the northerly anomalies over the NIO and the anomalous anticyclonic circulation in the SIO and enhanced moisture convergence and rainfall over the tropical IO. The anomalous anticyclonic circulation over the northwestern Pacific associated with strong easterly anomalies, is associated with an extension of the North Pacific anticyclone and a decrease in rainfall over the SCS. Thus, in the SIO warming cases, precipitation anomalies relative to the NIO warming cases (Figure 7b) are positive along the tropical IO and the maritime continent and negative over the SCS and WPO. These precipitation features are consistent with the leading mode of precipitation (Figure 1a), which suggests that the ENSO-developing mode is associated primarily with the SIO.
5. Predictability in the NCEP CFS
 The contrasting associations between ENSO and SSTs in the NIO and SIO in boreal spring have been demonstrated through several linear statistical methods and composite analysis using observations. In this section, the features are assessed further in simulations of the NCEP coupled and uncoupled models.
 The coupled CFS has achieved important advances in operational prediction from NOAA's previous dynamical forecast model and demonstrates skill in seasonal forecasts comparable to statistical methods [Saha et al., 2006], and several studies have demonstrated its skill in simulating and predicting ENSO and its teleconnections [Peng and Kumar, 2005; Wang et al., 2005; Saha et al., 2006; Yang et al., 2008], including its coupling with the Asian monsoon [Yang et al., 2008]. As described by Wang et al. , though the frequency of ENSO events is excessive and warm events tend to start too early in the year and last too long, the CFS simulates ENSO reasonably compared to most coupled models and is among the best available for real-time ENSO prediction.
5.1. Dominant Features of Summer Precipitation in the NCEP Model
Figure 8 shows the spatial patterns of the leading EOF modes of summer precipitation as a regression map of precipitation against the PCs of EOF (bottom) for three different runs: (1) 0 month lead (LM0), (2) 3 month lead (LM3) (both from the CFS hindcasts), and (3) an AMIP run. In particular, LM0 JJA precipitation is predicted by May ICs, LM3 JJA precipitation is predicting from February ICs, and AMIP JJA precipitation is the simulated response to observed SST. Overall, the intensity of JJA precipitation is predicted to be weaker than observed (Figure 1a) over the Bay of Bengal and the Arabian Sea and stronger than observed over the SCS. Nevertheless, the spatial patterns of Figures 8a and 8b are qualitatively similar to Figure 1a, with pattern correlations of 0.70 and 0.71 and temporal correlations of 0.85 and 0.62, respectively. The strong similarities demonstrate that JJA precipitation is reasonably predicted by the CFS up to LM3 using February ICs. However, in the AMIP run, the leading EOF mode of precipitation is not reasonably captured and has both a weak pattern (r = 0.44) and temporal correlation (r = −0.20), thus illustrating the importance of air-sea interaction in the prediction of precipitation [Kang et al., 2002; Yoo et al., 2004; Wang et al., 2005; Xie et al., 2009]. Similar simulation deficiencies in the subtropical WPO and rainbands related to the Mei-yu, Baiu, and Changma fronts during summer have been shown as a major problem for atmospheric GCMs [Kang et al., 2002].
 In Figure 9, the lead and lag correlations between the PCs and seasonal means of Niño3.4 SST are shown. For LM0, we use the observed Niño3.4 SSTs of MAM and -DJF and the model predicted Niño3.4 SSTs of LM0 JJA, LM3 SON, and LM6 DJF (all use May ICs). For LM3, we use the predicted Niño3.4 SSTs of LM0 MAM, LM3 JJA, and LM6 SON (all use February ICs). In both LM0 and LM3 cases, the correlation patterns agree generally with observations, although the leading mode of LM3 JJA precipitation has relatively large correlation with LM0 MAM Niño3.4 SST, implying that the leading mode of precipitation from LM0 and LM3 JJA can best be characterized as an ENSO-developing mode. In the AMIP run, summer precipitation is strongly correlated with the Niño 3.4 SST of previous seasons but exhibits a very weak correlation with the simultaneous and subsequent Niño3.4 SSTs. These relationships are inconsistent with observations and thus suggest the importance of air-sea interaction for replicating the observed relationship between SST and precipitation.
 In Figure 10, we show the regression of 850 hPa winds and SST against the leading modes of precipitation for LM0, LM3, and AMIP to understand the large circulation features associated with the ENSO-developing mode in the coupled and uncoupled models. In the NCEP CFS, the easterly anomalies in the WPO in response to ENSO-related SST anomalies are too large and the anomalous anticyclonic circulation over SCS is weaker than observed. Nevertheless, the CFS captures many aspects of canonical La Niña development including the strong easterly anomalies over IO and WPO, the concurrent anticyclonic circulation over SCS, and the southward flow over the northeast coast of Australia. However, the main features related to ENSO onset are not captured in AMIP simulations (Figure 10c). For example, negative anomalies in precipitation over the Bay of Bengal and SCS (Figure 8c) are associated with cold SST anomalies in AMIP runs. This result is also consistent with Wang et al. , who identify a positive correlation between SST and precipitation during summer in AGCM simulations whereas observations suggest it should be weak or negative. These findings underscore the importance of air-sea coupling in simulating and predicting correctly the monsoon climate over Asia, IO, and WPO.
5.2. Relationship Between SIO SST and ENSO in the NCEP Model
 To assess the NCEP model's simulated association between springtime SIO SST and the ENSO-developing mode, the regression of MAM SST against the leading modes of JJA precipitation for LM0, LM3, and AMIP is computed (Figure 11). We use the observed MAM SST for LM0 JJA precipitation and LM0 MAM SST for LM3 JJA precipitation. The observed lag correlation between SIO SST and the ENSO-developing mode (see Figure 4a) is simulated realistically in LM0 (Figure 11a), with a strong zonal dipole pattern in the SIO. In Figure 11b, the ENSO-developing mode for LM3 is associated with both SIO and ENSO-related SST anomalies of LM0 MAM. The association is likely related to the unrealistically persistent ENSO in the model that persists into MAM. Consistent with Wu et al. , an excessive response of the equatorial Pacific zonal wind (Figure 10b) leads to larger persistence of the equatorial Pacific SST anomalies in the CFS. In contrast to the coupled runs, the dominant feature of JJA precipitation in the AMIP runs is strongly related to the ENSO-related SST anomalies of the previous season (Figure 11c), suggesting that fully coupled models considering air-sea interactions are necessary to understand and realistically simulate the relationship between SIO and ENSO.
 To examine the selective interaction between the IO and ENSO in the NCEP model, we compute the regression of JJA 850 mb winds and precipitation against the leading modes of MAM NIO and SIO SSTs for LM0, LM3, and AMIP runs (Figure 12). Here we use the observed MAM SST for LM0 JJA 850 mb winds and precipitation and LM0 MAM SST for LM3. The leading EOF pattern of MAM SIO SST predicted by the CFS using February ICs (LM0 MAM) exhibits negative SST anomalies over the southwestern IO and strong warming over the subtropical IO between 10°S and 30°S, instead of SEIO (not shown). Similar findings are reported by Huang and Shukla  in a different coupled model where differences in the structures of observed and model SST patterns are attributed to excessive strength in their model's South Asian monsoon. Along with the unrealistically simulated MAM SIO SST pattern, related problems appear from the low-level wind and precipitation patterns (Figure 12e), due potentially to an unrealistically persistent ENSO in the CFS. In contrast to the NIO, MAM SIO SST is strongly associated with easterly anomalies in the WPO in LM0 JJA and even in the LM3 JJA predicted by LM0 MAM (Figures 12a–12b and 12d–12e). In Figures 12d and 12e, the interaction between the IO and WPO is more realistically simulated in LM0 than in LM3 JJA. The simulated anomalous anticyclonic and cyclonic patterns over the southwestern IO are displaced southwestward and westward from their observed locations, respectively. As in observations, in the AMIP runs easterly anomalies over the WPO are more strongly correlated with the SIO than with the NIO and the anomalous anticyclonic and cyclonic circulation patterns over the southwestern IO are reasonably captured (Figures 12c and 12f). However, the interaction between the IO and WPO is not well simulated.
 The importance of the SIO for predicting ENSO development in the CFS is further supported by an SVD analysis between LM0 MAM SST and LM3 JJA precipitation, which shows the contrasting regression patterns for NIO and SIO (Figure 13). In Figure 13b (Figure 13a), as in observations, the LM3 JJA precipitation pattern of the first SVD mode associated with the LM0 MAM SIO (NIO) SST predicted by February ICs is highly correlated with the first (second) EOF mode of LM0 JJA precipitation, with a coefficient of 0.84 (−0.91). This feature shows that the interaction between the SIO SST and the ENSO-developing mode in the CFS is captured reasonably well.
 To illustrate the observed association between MAM SIO SST and the transition from El Niño to La Niña, the sliding correlation between EOF PC1 of MAM LM0 IO SST and the simulated seasonal means of Niño3.4 SST is shown in Figure 14, where observed Niño3.4 SST for the SON and DJF of the previous year is used. As in observations, the entire IO and NIO have strongest correlations for the concurrent and preceding season's Niño3.4 SST, although the simultaneous correlation is excessive. The LM0 MAM SIO SST tends to be correlated positively with the Niño3.4 SST of preceding seasons and negatively correlated with subsequent Niño3.4 SST in the CFS, in support of our hypothesized influence of SIO SST on ENSO transitions. However, the latter correlation is weaker than observed, consistent with the finding of Wu et al.  of an excessive sensitivity of tropical zonal wind to ENSO-related SST anomalies and its contribution to a delayed transition in eastern equatorial Pacific SST anomalies in the CFS.
6. Summary and Discussion
 In this study, the influence of the Indian Ocean on the transitions from El Niño events is explored based on observed variability in recent decades. An emphasis is placed on understanding the contrasting roles of the NIO and SIO and their respective associations with the Asian monsoon, surface winds in the WPO, and ENSO. In doing so, this work builds upon earlier studies that identify key roles for the Indian Ocean, both in regards to its broadscale interaction with ENSO and as a source of persistence that aids in overcoming the spring persistence minimum [e.g., Kug and Kang, 2006; Webster and Yang, 1992; Yu et al., 2003; Y06; Kug and Kang, 2006; Xie et al., 2009]. Here the unique role of the southern Indian Ocean is highlighted, and zonal structure in SST anomalies is proposed to be key factor to subsequent interactions with both the Asian monsoon and ENSO systems. The leading modes of interannual variability in summer monsoon precipitation are shown to be characterized by ENSO-developing and decaying modes. The ENSO-developing mode is characterized by strong easterly (westerly) anomalies that extend from the central Pacific Ocean into the IO associated with the emergence of La Niña (El Niño) in a manner consistent with established precursors of ENSO. The ENSO-decaying mode is associated with a northward propagation of Rossby waves from the SCS through East Asia, consistent with their excitation by enhanced deep convection in the WPO.
 While the ENSO-decaying mode is related to basin-wide variability in the NIO, the features associated with ENSO development are strongly correlated with SIO SST and are characterized by a zonal dipole in the preceding spring. The relationships are particularly robust for the emergence of La Niña events subsequent to El Niño, suggesting a predictive capability for SIO SSTs in the biennial tendency of both ENSO and the monsoon. In some respects, this result is substantively inconsistent with Kug and Kang  and Kug et al. [2006a], which suggested that the NIO can influence the surface wind stress in the western Pacific via the Walker circulation and then accelerate transitions in the ENSO cycle. However, the distinct relationships of the northern and southern Indian Ocean are not considered and the zonally asymmetric nature of the Indian Ocean response characterized by the SST gradient in the SIO which accounts for the key aspects of the observed variability is not addressed in these earlier works. Indeed in explaining many of the nuances of the observed variability, we find our explanation to be more complete, while building upon this important earlier work. These details also include addressing the issue as to why some El Niño events transition into La Niña while others do not (e.g., when warming occurs in the tropical central Pacific). Notably, the 1994 El Niño event is a good example as it is an El Niño event unaccompanied by NIO warming, yet La Niña follows it. While the 1994 event accompanies our proposed SIO SST pattern the NIO actually cools and thus the overriding constraint on the transition appears to be the SIO response rather than the “flavor” of ENSO. Nonetheless, 1994 is also a strong IOD year and thus complicates inferences regarding causality. Additional modeling investigations are likely to be instrumental in further quantifying causality among these multiple potential interactions.
 The observed variability and proposed linkages between the SIO and ENSO is summarized in a schematic diagram (Figure 15). The summary includes the interaction of the mean state (solid lines) with anomalous circulation (dashed lines). The thick red (blue) arrows indicate SST warming (cooling) weakening (strenghing) the mean flow through the wind-evaporation feedback over SIO. During El Niño winter, the combined effect of anomalous anticyclonic pattern induced by El Niño and the Southern Annular Mode (strong westerly between 40° and 60°S [L'heureux and Thompson, 2006]) forms zonal SST anomalies, southwest cold and southeast warm SST anomalies, and southeastward wave propagation in the SIO. Y06 proposed that this anomalous anticyclonic pattern developed by a Rossby wave response of the decreased convection over maritime continent during El Niño weakens the mean flow over the tropical and subtropical SIO and further maintains the positive SST anomalies until the mean flow changes its direction in late spring or early summer. When El Niño dissipates in spring, the SIO SST anomalies persist and are strongly associated with an SIO wave pattern, which moves eastward and is more zonally oriented over the SIO than in DJF. Persistent warming over the SEIO tends to decrease the meridional temperature gradient between the northern and southern hemispheres over the tropical IO and increase the zonal temperature gradient between the southwestern IO and SEIO. The decreased meridional temperature gradient may act to weaken the cross-equatorial flow over western IO and thus the monsoonal flow over NIO, favoring an extension of the easterly wind anomalies from the WPO to the NIO during La Niña onset in summer. The increased zonal temperature gradient between the southwestern IO and SEIO may act to strengthen the westerly wind anomalies over the SIO (∼40°S) and thus strengthen the climatological anticyclonic circulation over the SIO. As proposed by Y06, this amplified climatological anticyclone is associated with larger northward moisture transportation and more precipitation over the equatorial IO through the moisture convergence with the weakening cross-equatorial and monsoonal flow over the tropical IO. Persistent SEIO warming also accompanies an increase in precipitation over the maritime continent and a delay in the northward shift of tropical rainband during summer. These persistent SIO anomalies dissipate during the summer monsoon.
 In contrast to the SIO, the NIO exhibits warming that persists in association with El Niño. Consequently, the NIO warm anomalies both increase the meridional temperature gradient between the Northern and Southern hemispheres and decrease the land-sea temperature gradient over the South Asian monsoon region. The anomalies thus suggest a role in strengthening the northward cross-equatorial flow in the western IO and in weakening monsoonal westerlies. The southerly and easterly wind anomalies converge over the Arabian Sea and accompany a northward shift in the summer rainband.
 The observed association between SIO SST and ENSO is not well replicated in AMIP experiments while the coupled NCEP CFS is capable of simulating many major features of both precipitation and its coupled air-sea variability several months in advance. The importance of air-sea coupling in the proposed mechanism is thus suggested. Additionally, as the relationship of SIO SST and WPO variability associated with the onset of La Niña events is reasonably simulated in the CFS, the simulations support the existence of the proposed mechanism. However, the excessive impact of ENSO on the Asian and Indo-Pacific climate, as demonstrated by Yang et al.  and Liang et al. , results in a delayed transition of the eastern equatorial Pacific SST in the model. Further investigation of the role of these model biases on there simulations performed herein and targeted sensitivity tests aimed at further diagnosing causal linkages in the proposed mechanism are likely to improve our understanding of the proposed interaction.
 The authors would like to thank editor Steve Ghan and three anonymous reviewers for many constructive comments. C.-H. Ho was supported by the Korea Meteorological Administration Research and Development Program under grant CATER 2006-4204. Dr. Fasullo's participation is sponsored by NASA Award NNX07AKG82G.