1. Introduction
Variability in the Indian Ocean, important for understanding climate on the interannual time scale for many surrounding countries, has become an active topic of research in recent decades (Schott et al. 2009). In austral spring [September–November (SON)], the major mode of interannual sea surface temperature (SST) variability in the Indian Ocean is the tropical Indian Ocean dipole (IOD) zonal mode (Saji et al. 1999; Webster et al. 1999; Murtugudde et al. 2000), although other modes exist, including the Indian Ocean Basin (IOB) mode (Klein et al. 1999; Liu and Alexander 2007) and the Indian Ocean subtropical dipole (IOSD) mode (Behera and Yamagata 2001; Morioka et al. 2010, 2013). It is recognized that the development and variability of these modes is caused by processes both internal and external to the Indian Ocean. For example, results from both observations and coupled models suggest that the IOD is an intrinsic mode of the Indian Ocean coupled system, which either can be externally triggered, by El Niño–Southern Oscillation (ENSO), or can self-generate, provided the thermocline off Sumatra is shallow enough to support Bjerknes feedback (Schott et al. 2009). Therefore, the ability of state-of-the-art coupled general circulation models to realistically simulate the present-day variability and future evolution of the IOD relies heavily upon accurately modeling the complex interplay of numerous ocean and atmosphere processes.
Previous studies focusing on the performance of models in phase 3 of the Coupled Model Intercomparison Project (CMIP3) have shown that large diversity exists in the IOD strength (Saji et al. 2006; Cai et al. 2011b), dynamical and thermodynamical feedbacks (Liu et al. 2011), coherence with ENSO (Saji et al. 2006), and its local and remote rainfall teleconnections (Cai et al. 2009a, 2011b). Cai et al. (2011b) further demonstrated that projected intermodel differences in future changes of temperature and rainfall depend on how well models simulate historical and present-day IOD properties; models with a stronger present-day IOD amplitude systematically produce a weaker eastern tropical Indian Ocean warming rate with greater future rainfall changes in IOD-influenced regions. Further, although the models simulate the teleconnection pathway—that is, the impact on the subtropics (i.e., southern Australia, South Africa, and South America) is conducted through equivalent-barotropic Rossby wave trains emanating from the tropical Indian Ocean (Saji and Yamagata 2003; Liu et al. 2007; Chan et al. 2008; Cai et al. 2011c)—this extratropical teleconnection is weaker in the majority of models in CMIP3, relative to the observed (Cai et al. 2009a).
It is yet to be seen if the newly available coupled models partaking in phase 5 (CMIP5) show any improvement in simulating the IOD and its properties. The objective of the present study is to assess basic IOD properties in historical simulations of CMIP5 in a comparison with CMIP3 and to examine the sensitivity of future climate changes to the present-day IOD strength and rainfall teleconnection incorporating CMIP5 outputs.
2. Models, data, and IOD definition
We analyze the historical (CMIP5) and twentieth-century (20C3M; CMIP3) experiments, utilizing SST, thermocline (Z20), precipitation, and zonal wind outputs from available coupled models. In total, one ensemble member (i.e., run 1) from 20 models in CMIP5 and all 24 models in CMIP3 (see Table 1 for all model information) is used in this study. We take outputs from a common 50-yr period of the twentieth century (i.e., 1950–99), stratified into four seasons, but restrict our analysis to SON, the peak season of the IOD. SST from the Hadley Centre Global Sea Ice and SST (HadISST1; Rayner et al. 2003), Z20 from Simple Ocean Data Assimilation–Parallel Ocean Program, version 2.2.4 (SODA–POP V2.2.4; Carton and Giese 2008), and zonal wind from the National Centers for Environmental Prediction (NCEP)–NCAR reanalysis (Kalnay et al. 1996) are utilized to provide an observed reference for the coupled models.
CMIP5 and CMIP3 modeling centers (or group) and model names.
For future climates, we use outputs from twenty-first century experiments: representative concentration pathway (RCP) 8.5 and Special Report on Emissions Scenarios (SRES) A2 for CMIP5 and CMIP3, respectively. As the experimental design and greenhouse gas scenarios are not identical for CMIP5 and CMIP3, future rainfall changes are expressed in terms of percentage change in climatology per degree Celsius of global warming [GW; % (°C of GW)−1]—this allows for comparison between different future scenario experiments with the assumption that the global warming response is linear. Likewise, future temperature changes are expressed in terms of degree Celsius per degree Celsius of global warming [°C (°C of GW)−1]. Outputs of only 15 of the 20 models in CMIP5 available to us contain precipitation from both historical and RCP experiments. These 15 CMIP5 and all CMIP3 models are used to test the relationship between the present-day simulation of the IOD properties and future rainfall changes.
In each model and the observations, the IOD is described through an empirical orthogonal function (EOF) analysis on detrended SST anomalies in the tropical Indian Ocean domain (20°S–20°N, 40°–120°E). The IOD index is taken as the time series associated with the EOF spatial pattern (principal component), standardized to have a standard deviation of 1. Similar to other studies, we employ EOF analysis as opposed to standard indices such as the dipole mode index (DMI; Saji et al. 1999) as it allows each model to exhibit its own dominant pattern of variability, as opposed to an imposed structure (Saji et al. 2006; Liu et al. 2011). Similarly, the ENSO index is taken as the time series associated with the EOF spatial pattern of SST anomalies in the tropical Pacific Ocean domain (25°S–25°N, 120°E–80°W).
3. Simulated IOD and its impact on future rainfall changes
a. Model present-day climate IOD
Figure 1a shows the SON multimodel ensemble–mean (MMEM) EOF pattern from the models in CMIP5. Overall, it resembles the observed (Fig. 1c); however, the anomaly is too strong in both IOD poles, particularly in the eastern tropical Indian Ocean. A comparison with models in CMIP3 indicates that no MMEM improvement is evident in the simulation of the spatial structure of the IOD (Figs. 1a,b). This is reinforced in the MMEM statistics between CMIP3 and CMIP5 (large squares in Fig. 1d). The MMEM pattern correlation coefficient between the simulated and observed IOD for the models in CMIP5 is 0.82, which is comparable to that for the models in CMIP3 (0.83). However, a few models have improved statistics in regard to the IOD, so that several outlying models in CMIP3 are in better agreement with observations in their CMIP5 versions [e.g., GFDL CM2.1 (ID = 7), INM-CM3.0 (ID = 11), and NCAR-PCM1 (ID = 17); Fig. 1d]. [Note that all identification (ID) numbers are given in the inset in Fig. 1.] That is, the intermodel deviation of the IOD pattern correlation coefficients is reduced from 0.18 in CMIP3 to 0.14 in CMIP5.
Spatial patterns of the first EOF mode of detrended SST anomalies over the tropical Indian Ocean during 1950–99, calculated from (a) 20 models in CMIP5, (b) 24 models in CMIP3, and (c) observations (HadISST). Also shown are the ranges in percentage of explained variance for the models in CMIP as well as the observed variance. (d) Taylor diagram highlighting intermodel deviations in amplitude (std dev), root-mean-square differences (RMSD), and pattern correlation coefficients of the CMIP3 (blue) and CMIP5 (red) IOD patterns vs observations (ID = 0, black). Large squares indicate the MMEM for the two CMIP groups.
Citation: Journal of Climate 26, 17; 10.1175/JCLI-D-12-00807.1
We calculate the standard deviation of the spatial patterns as a measure of the IOD amplitude (radial distance in Fig. 1d). The majority of models in CMIP5 simulate an IOD amplitude larger than the observed, except for three models [GISS-E2H (ID = 8), GISS-E2-R (ID = 8), and MRI-CGCM3 (ID = 16)]. However, both the CMIP5 and CMIP3 MMEM have amplitudes that are 1.7 times as large as observed. Again, there are fewer outliers in the models in CMIP5 (red dots in Fig. 1d) with the intermodel deviation reducing from 0.095 in CMIP3 to 0.086 in CMIP5. Cai and Cowan (2013) demonstrate that this reduction in intermodel deviation also occurs when calculating the amplitude using the DMI in the models.
b. Relevance of the IOD to future rainfall changes
Because of the biases in modeled present-day IOD amplitude, one must consider how to interpret future changes over IOD-influenced regions. For example, a robust relationship exists whereby models in CMIP5 with a larger present-day IOD amplitude produce a smaller future warming in the eastern tropical Indian Ocean (Figs. 2a,b). Such relevance also applies whereby models in CMIP5 with a larger present-day IOD amplitude produce a larger rainfall reduction over IOD-influenced regions in the twenty-first century, where a positive IOD leads to a dry condition (Figs. 2c,d). Figure 2c shows the intermodel variations of the present-day IOD amplitude versus projected rainfall changes averaged over the eastern tropical Indian Ocean (0°–10°S, 100°–110°E). Larger (smaller) symbols of the same type represent the models from the same modeling group in CMIP5 (CMIP3). Comparing the line of best fit for the models in CMIP5 and CMIP3, we find that the IOD–rainfall relationship is similar for both model generations (r = 0.55 for CMIP5, r = 0.59 for CMIP3, and r = 0.58 for combined CMIP5 and CMIP3, all significant at 95% assuming independent models). The significant correlations suggest that the simulation of the present-day IOD amplitude has a direct implication for the response and rainfall changes of the eastern Indian Ocean.
Intermodel variations in IOD amplitude of the present-day climate vs variations in future SON (a) surface temperature changes [°C (°C of GW)−1] and (c) rainfall changes [% (°C of GW)−1], over the eastern tropical Indian Ocean. The observed amplitude is indicated by a red vertical line. Additionally shown is a correlation map, with respect to all models, between (b) IOD amplitude of the present-day climate and gridpoint future SON surface temperature changes, and (d) IOD amplitude of the present-day climate and gridpoint future SON rainfall changes. Statistically significant correlations at the 95% confidence level are shown within the black contours. Also shown are MMEM SON trends in rainfall [% (°C of GW)−1] averaged over models with (e) small [blue circled group in (c)] and (f) large [red circled group in (c)] IOD amplitudes.
Citation: Journal of Climate 26, 17; 10.1175/JCLI-D-12-00807.1
This analysis for the eastern tropical Indian Ocean region can be applied at each grid point to assess regions where the present-day IOD amplitude is relevant to projections of future rainfall changes. For this test, models in CMIP5 and CMIP3 are combined to produce a larger sample size (N = 39). A systematic well-defined pattern emerges showing that the relationship over the eastern Indian Ocean extends to the subtropics over northern and southern Australia (Fig. 2d), somewhat similar to the IOD–rainfall teleconnection on interannual time scales. This result, with the inclusion of models in CMIP5 and increased confidence levels, reinforces the notion that over IOD-influenced regions, where present-day climate IOD properties are relevant to such future climate changes (Fig. 2d), model selection can make a marked difference to projections.
Figures 2e and 2f compare the MMEM rainfall changes over the subgroup of models (within the blue circle in Fig. 2c) that have a small IOD amplitude relative to observations and another that have the largest IOD amplitude (within the red circle in Fig. 2c). The rainfall change in the group with smaller amplitudes displays a very modest decline (Fig. 2e), but that in the group with greater amplitudes shows a 15%–20% reduction over the eastern Indian Ocean, extending into the subtropics and over Australia. Thus, the projected rainfall changes over the eastern tropical Indian Ocean and Australia are sensitive to model simulation of present-day IOD amplitude. By contrast, the average increase in rainfall over the eastern region of the African continent does not show a similar sensitivity to IOD amplitude. Figure 2f also highlights that the well-defined systematic pattern in Fig. 2d can be seen as being embedded in the map of future rainfall changes. However, we note that models with an IOD amplitude close to the observed (Figs. 2a,c) tend to be those that cannot simulate the observed pattern, suggesting that the seemingly realistic amplitude is achieved through unrealistic processes (e.g., Cai et al. 2009b; Liu et al. 2011). For example, the subgroup of models that have an IOD amplitude comparable to observations (within the blue circle in Fig. 2c) on average display an amplitude of 0.13, compared to an observed amplitude of 0.11. However, their averaged pattern correlation coefficient with the observed is 0.54, substantially less than the MMEM value of approximately 0.8.
4. CMIP5 IOD–rainfall teleconnection and positive feedback strength
The relevance of simulated IOD amplitude to rainfall projections in IOD-influenced regions is achieved through the IOD–rainfall teleconnection. That is, models with a greater IOD amplitude systematically produce a greater rainfall change in IOD-affected regions because the positive feedbacks project onto a stronger IOD–rainfall teleconnection that already operates in the modeled present-day climate. Is there any difference in the IOD–rainfall teleconnection between models in CMIP5 and CMIP3? Figure 3a examines the sensitivity to the IOD index of rainfall anomalies over the eastern Indian Ocean. During positive IODs (pIODs) and negative IOD (nIODs), rainfall decreases and increases respectively, and the sensitivity in CMIP5 and CMIP3 is comparable. It has been shown that the sensitivity in CMIP3, although spatially resembling the observed teleconnection, is slightly weaker than the observed (Cai et al. 2009a). There is a slightly more sensitive response in CMIP5 with a stronger slope (thick line in Fig. 3a), and this can be considered as an improvement.
(a) Linearly detrended SON rainfall anomalies averaged over the tropical eastern Indian Ocean (0°–10°S, 100°–110°E) vs the IOD index (taken as the time series associated with the EOF1 of tropical Indian Ocean SST anomalies) for models in CMIP5 and CMIP3. (b) CMIP5 MMEM one std dev anomaly pattern of gridpoint detrended rainfall associated with the IOD index [mm day−1 (unit of IOD index)−1]. Green contours show the statistically significant correlation coefficients at the 95% confidence level. (c) As in (b), but for CMIP3 MMEM.
Citation: Journal of Climate 26, 17; 10.1175/JCLI-D-12-00807.1
A map of the IOD–rainfall teleconnection is constructed in a similar way by regressing linearly detrended gridpoint rainfall anomalies onto the IOD index in the historical experiments. The CMIP5 MMEM pattern (Fig. 3b) depicts the IOD-induced impacts on rainfall during pIODs. A reduction is seen over the eastern tropical Indian Ocean and Australia (Saji et al. 1999; Ashok et al. 2003; Weller and Cai 2013), as well as an increase over eastern Africa (Black et al. 2003). This interannual variability pattern resembles that of the intermodel variations (Fig. 2d), which further highlights that it is through this rainfall teleconnection already operating in the present-day climate that the IOD amplitude is relevant to rainfall projections. The CMIP5 and CMIP3 MMEM spatial patterns (Figs. 3b,c) show a negligible difference. Large unrealistic anomalies still exist over the equatorial western Pacific (Figs. 3b,c) instead of the central and equatorial Pacific region associated with the covarying ENSO (Cai et al. 2009a). This feature has been linked to the Pacific SST bias, in which the SST anomalies associated with ENSO extend too far west in the coupled models (Cai et al. 2009a; Zheng et al. 2012). This bias appears to still exist in models in CMIP5 (Kim and Yu 2012).
To examine the possibility that ENSO and its interaction with the IOD has an influence on future projections over the Indian Ocean and the Australian region during SON, we examine the IOD–ENSO interactions (i.e., Behera et al. 2006; Saji et al. 2006; Cai et al. 2011a; Luo et al. 2010) as simulated by the models. Similar to Fig. 2d, Fig. 4a highlights regions where the present-day climate ENSO amplitude is relevant to projections of future SON rainfall changes. It can be seen that apart from significant regions in the tropics (i.e., the eastern tropical Indian Ocean and the western tropical Pacific), unlike the IOD, ENSO has little influence on projections in the extratropics during this season. This is predominantly because the majority of models display a weaker IOD–ENSO interaction in SON relative to observations (Fig. 4b). Despite this, Fig. 4b reveals there is a tendency for models with a greater ENSO strength to display a greater correlation between the IOD and ENSO. Using a smaller number of models in CMIP3, Saji et al. (2006) suggested that there exists no significant relation between the two quantities. However, by increasing the number of models in CMIP3 it is found that a statistically significant relationship exists (Cai et al. 2011a). This tendency is even more robust in models in CMIP5. Thus, the relevance of the IOD to future projections over regions such as the western Pacific may simply be through its interaction with ENSO and mainly in models that have a greater ENSO amplitude and a greater correlation between the two. This can be tested by examining the relevance of the IOD–ENSO correlation to future rainfall changes (Fig. 4c). Regions where this may be true are mainly confined to the Pacific Ocean, especially along the equatorial band, with only significant regions over the eastern Indian Ocean in the Northern Hemisphere (Fig. 4c). Therefore, over IOD-affected regions, ENSO and its related biases (e.g., the western tropical Pacific bias) appear to have little influence on the teleconnection mechanism relevant to the link between present-day climate and future projections.
(a) Correlation map, with respect to all models, between ENSO amplitude of the present-day climate and gridpoint future SON rainfall changes [% (°C of GW)−1]. (b) Intermodel variations in ENSO amplitude of the present-day climate vs variations in the correlation between the IOD and ENSO. (c) Correlation map, with respect to all models, between IOD–ENSO correlation of the present-day climate and gridpoint future SON rainfall changes [% (°C of GW)−1]. In (a) and (c), statistically significant correlations at the 95% confidence level are shown within the black contours.
Citation: Journal of Climate 26, 17; 10.1175/JCLI-D-12-00807.1
For the eastern tropical Indian Ocean, there is a tendency for models with a stronger IOD–rainfall teleconnection to be associated with a greater projected rainfall change (Fig. 5a) in both CMIP5 (black line) and CMIP3 (gray line). However, models in CMIP5 seem to produce a weaker tendency, with a smaller slope. Using all 39 models, point-to-point correlation between the IOD–rainfall teleconnection and future rainfall changes shows that the systematic pattern extends to southeastern Australia and that the pattern resembles that associated with the interannual IOD–rainfall teleconnection pattern (Fig. 5b).
(a) Intermodel variations in regression of present-day climate rainfall onto the IOD index [mm (SON)−1 (unit of IOD index)−1] vs future SON rainfall changes [% (°C of GW)−1] over the eastern tropical Indian Ocean. The observed regression is indicated by a red vertical line. (b) Correlation map, with respect to models, between IOD–rainfall teleconnections and gridpoint future SON rainfall changes. (c) Intermodel variations in IOD amplitude vs the regression of gridpoint zonal wind stress onto the IOD index [N m−2 (unit of IOD index)−1] over the eastern tropical Indian Ocean. (d) Map of regression coefficients [N m−2 (unit of IOD index)−1] obtained by regressing gridpoint zonal wind-to-IOD sensitivity onto IOD amplitude. Statistically significant correlations at the 95% confidence level are shown within the black contours in (b) and (d).
Citation: Journal of Climate 26, 17; 10.1175/JCLI-D-12-00807.1
The linkage between intermodel variations of IOD properties and future rainfall changes relies upon the fact that coupled models with a greater IOD amplitude and rainfall teleconnection invariably possess stronger Bjerknes-like positive feedbacks. Here, we provide an example of one of the feedbacks using the sensitivity of anomalies of surface zonal wind stress over the eastern tropical Indian Ocean to the IOD index through a linear regression analysis (Fig. 5c). Intermodel variations of the IOD amplitude versus the zonal wind stress sensitivity for CMIP5 and CMIP3 show a statistically significant correlation at the 95% confidence level. This reinforces the robust relationship whereby models with larger IOD amplitudes produce a greater zonal wind–IOD sensitivity (positive feedback). The majority of models in CMIP5 and CMIP3 appear to overestimate this feedback [observed value of −7.1 × 10−3 N m−2 (unit of IOD index)−1], producing too large an amplitude of SST anomalies over the eastern tropical Indian Ocean. The spatial pattern of the intermodel variations calculated using zonal wind stress at each grid point (Fig. 5d) indicates that models with greater IOD amplitudes tend to have a greater wind response to the SST gradient. The pattern again resembles that associated with interannual variability. Cai et al. (2011b) suggest that future climate changes in the form of easterly wind trends in the equatorial Indian Ocean provides a perturbation that induces a greater response in the coupled models with a greater IOD amplitude (hence a stronger positive feedback).
Cai and Cowan (2013) have examined the cause of the IOD amplitude and feedback biases in detail with respect to the mean state of the models. They show that the majority of models produce too strong a Bjerknes feedback in the equatorial Indian Ocean, involving winds, SST, and thermocline, leading to the bias. The thermocline–SST feedback was found to exert the strongest influence on the simulated IOD amplitude; models with a stronger feedback systematically generate a greater amplitude (Cai and Cowan 2013), as seen in the relationship on interannual time scales (Zheng et al. 2010; Cai and Qiu 2013). The strength of the thermocline–SST feedback in most models is predominantly controlled by the climatological west–east slope of the equatorial thermocline, which features an unrealistic mean slope tilting upward toward the eastern Indian Ocean. The unrealistic thermocline structure is accompanied by too strong a mean easterly wind, and an overly large west-minus-east SST gradient. Their analysis was based upon multimodel statistics using models in CMIP3 and CMIP5 as one sample set.
Here, we carry out a similar analysis but examine the three positive feedback components in terms of the two modeling phases (i.e., CMIP3 and CMIP5), with respect to the observed feedback strength (Fig. 6), in an attempt to ascertain the cause for the reduction in intermodel deviation of the CMIP5 IOD amplitude, relative to that in CMIP3. In addition, within each model phase, models are divided into three subgroups according to their IOD amplitude. For example, the blue squares and circles in Fig. 6 represent the wind–SST feedback for models in CMIP3 and CMIP5, respectively, in terms of percentage of the observed strength, for models with an IOD amplitude 1) less than the observed, 2) less than 1.5 times or equal to the observed, and 3) 1.5 times greater than the observed. Also shown is the all model mean for the two phases associated with each positive feedback. Superimposed is an intermodel spread calculated as the standard deviation of the subgroup samples. Similar to Cai and Cowan (2013), models with a stronger feedback systematically generate a greater IOD amplitude for all three components, with the rate of change between subgroups associated with the SST–thermocline feedback being the greatest. Although the mean SST–thermocline feedback is greater in models in CMIP5, the intermodel deviation is reduced relative to CMIP3. Together with a reduction in intermodel deviation of the wind–SST feedback, this could be the cause of the reduction in intermodel deviation of the IOD amplitude from CMIP5 to CMIP3. Thus, reducing biases in the equatorial Indian Ocean mean state climate, which affect these feedbacks involved in IOD development and its strength, may reduce the overestimation of the positive feedbacks, leading to more realistic model IOD amplitudes and reliable rainfall projection in IOD-affected regions.
(a) Relationship between the amplitude of the IOD and components of the Bjerknes feedback, including the sensitivity of winds to SST (blue), thermocline depth to winds (green), and SST to thermocline depth (red). Model feedback is represented as a percentage of the observed feedback, with squares (circles) denoting CMIP3 (CMIP5), and vertical dashed lines the std dev within the samples. Solid lines represent the multimodel mean for all CMIP3 and CMIP5 combined.
Citation: Journal of Climate 26, 17; 10.1175/JCLI-D-12-00807.1
5. Conclusions
We have analyzed the performance of models in CMIP5 simulating the IOD. Most models in CMIP5 and CMIP3 generate an IOD variability that is too strong relative to observations. This bias has important implications for projected SON rainfall trends under enhanced greenhouse warming over IOD-influenced regions, because these trends are sensitive to the simulation of IOD amplitude and the IOD–rainfall teleconnection in the modeled present-day climate. The average SON rainfall trend pattern is similar to the correlation pattern between intermodel variations of IOD amplitude and gridpoint future rainfall changes. These patterns in turn resemble the pattern of interannual variability of rainfall associated with the IOD. Underpinning the overly large amplitude of the IOD are modeled Bjerknes-like feedbacks that are too strong. With a stronger Bjerknes-like feedback strength, the magnitude of the eastern Indian Ocean response to climate change perturbations is greater, leading to a greater rainfall reduction in IOD-influenced regions, where a positive IOD leads to reduced rainfall. We show that large differences in future rainfall trends are obtained between models with a small and a large IOD amplitude. Given that the present-day IOD properties are realistically simulated, caution needs to be exercised in interpreting climate projections in IOD-affected regions.
Acknowledgments
We acknowledge the WCRP Working Group on Coupled Modelling, responsible for CMIP, and thank the climate modeling groups for producing and making available their model output. This study is supported by the Goyder Research Institute and the Australian Climate Change Science Programme. We thank Tim Cowan, Arnold Sullivan, and Ben Ng for their comments before submission, and three anonymous reviewers for their helpful comments, which improved the paper.
REFERENCES
Ashok, K., Z. Guan, and T. Yamagata, 2003: Influence of the Indian Ocean dipole on the Australian winter rainfall. Geophys. Res. Lett., 30, 1821, doi:10.1029/2003GL017926.
Behera, S. K., and T. Yamagata, 2001: Subtropical SST dipole events in the southern Indian Ocean. Geophys. Res. Lett., 28, 327–330.
Behera, S. K., J. J. Luo, S. Masson, S. A. Rao, H. Sakuma, and T. Yamagata, 2006: A CGCM study on the interaction between IOD and ENSO. J. Climate, 19, 1688–1705.
Black, E., J. M. Slingo, and K. R. Sperber, 2003: An observational study of the relationship between excessively strong short rains in coastal East Africa and Indian Ocean SST. Mon. Wea. Rev., 131, 74–94.
Cai, W., and T. Cowan, 2013: Why is the amplitude of the Indian Ocean dipole overly large in CMIP3 and CMIP5 climate models? Geophys. Res. Lett., 40, 1200–1205, doi:10.1002/grl.50208.
Cai, W., and Y. Qiu, 2013: An observation-based assessment of nonlinear feedback processes associated with the Indian Ocean dipole. J. Climate, 26, 2880–2890.
Cai, W., A. Sullivan, and T. Cowan, 2009a: Rainfall teleconnections with Indo-Pacific variability in the WCRP CMIP3 models. J. Climate, 22, 5046–5071.
Cai, W., A. Sullivan, and T. Cowan, 2009b: Climate change contributes to more frequent consecutive positive Indian Ocean dipole events. Geophys. Res. Lett., 36, L23704, doi:10.1029/2009GL040163.
Cai, W., A. Sullivan, and T. Cowan, 2011a: Interactions of ENSO, the IOD, and the SAM in CMIP3 models. J. Climate, 24, 1688–1704.
Cai, W., A. Sullivan, T. Cowan, J. Ribbe, and G. Shi, 2011b: Simulation of the Indian Ocean dipole: A relevant criterion for selecting models for climate projections. Geophys. Res. Lett.,38, L03704, doi:10.1029/2010GL046242.
Cai, W., P. van Rensch, T. Cowan, and H. Hendon, 2011c: Teleconnection pathways of ENSO and the IOD and the mechanisms for impacts on Australian rainfall. J. Climate, 24, 3910–3923.
Carton, J. A., and B. S. Giese, 2008: A reanalysis of ocean climate using Simple Ocean Data Assimilation (SODA). Mon. Wea. Rev., 136, 2999–3017.
Chan, S. C., S. K. Behera, and T. Yamagata, 2008: Indian Ocean dipole influence on South American rainfall. Geophys. Res. Lett., 35, L14S12, doi:10.1029/2008GL034204.
Kalnay, E., and Coauthors, 1996: The NCEP-NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc.,77, 437–471.
Kim, S. T., and J.-Y. Yu, 2012: The two types of ENSO in CMIP5 models. Geophys. Res. Lett., 39, L11704, doi:10.1029/2012GL052006.
Klein, S. A., B. J. Soden, and N.-C. Lau, 1999: Remote sea surface temperature variations during ENSO: Evidence for a tropical atmospheric bridge. J. Climate, 12, 917–932.
Liu, L., W. Yu, and T. Li, 2011: Dynamic and thermodynamic air–sea coupling associated with the Indian Ocean dipole diagnosed from 23 WCRP CMIP3 models. J. Climate, 24, 4941–4958.
Liu, N., H.-X. Chen, and L.-G. Lü, 2007: Teleconnection of IOD signal in the upper troposphere over southern high latitudes. J. Oceanogr., 63, 155–157.
Liu, Z., and M. Alexander, 2007: Atmospheric bridge, oceanic tunnel, and global climatic teleconnections. Rev. Geophys., 45, RG2005, doi:10.1029/2005RG000172.
Luo, J.-J., R. Zhang, S. K. Behera, Y. Masumoto, F.-F. Jin, R. Lukas, and T. Yamagata, 2010: Interaction between El Niño and extreme Indian Ocean dipole. J. Climate, 23, 726–742.
Morioka, Y., T. Tozuka, and T. Yamagata, 2010: Climate variability in the southern Indian Ocean as revealed by self-organizing maps. Climate Dyn., 35, 1059–1072.
Morioka, Y., T. Tozuka, and T. Yamagata, 2013: How is the Indian Ocean subtropical dipole excited? Climate Dyn., doi:10.1007/s00382-012-1584-9, in press.
Murtugudde, R., J. P. McCreary, and A. J. Busalacchi, 2000: Oceanic processes associated with anomalous events in the Indian Ocean with relevance to 1997–1998. J. Geophys. Res., 105, 3295–3306.
Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland, L. V. Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan, 2003: Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. J. Geophys. Res., 108, 4407, doi:10.1029/2002JD002670.
Saji, N. H., and T. Yamagata, 2003: Possible impacts of Indian Ocean dipole mode events on global climate. Climate Res., 25, 151–169.
Saji, N. H., B. N. Goswani, P. N. Vinayachandran, and T. Yamagata, 1999: A dipole mode in the tropical Indian Ocean. Nature, 401, 360–363.
Saji, N. H., S.-P. Xie, and T. Yamagata, 2006: Tropical Indian Ocean variability in the IPCC twentieth-century climate simulations. J. Climate, 19, 4397–4417.
Schott, F. A., S.-P. Xie, and J. P. McCreary Jr., 2009: Indian Ocean circulation and climate variability. Rev. Geophys.,47, RG1002, doi:10.1029/2007RG000245 .
Webster, P. J., A. M. Moore, J. P. Loschnigg, and R. R. Leben, 1999: Coupled ocean–atmosphere dynamics in the Indian Ocean during 1997–98. Nature, 401, 356–360.
Weller, E., and W. Cai, 2013: Asymmetry in the IOD and ENSO teleconnection in a CMIP5 model ensemble and its relevance to regional rainfall. J. Climate, 26, 5139–5149.
Zheng, X.-T., S.-P. Xie, G. A. Vecchi, Q. Liu, and J. Hafner, 2010: Indian Ocean dipole response to global warming: Analysis of ocean–atmospheric feedbacks in a coupled model. J. Climate, 23, 1240–1253.
Zheng, Y., J.-L. Lin, and T. Shinoda, 2012: The equatorial Pacific cold tongue simulated by IPCC AR4 coupled GCMs: Upper ocean heat budget and feedback analysis. J. Geophys. Res., 117, C05024, doi:10.1029/2011JC007746.
