From 1950 to 2008, the linear trend of tropical sea surface temperature (SST) indicates the largest warming across the Indian and western Pacific Oceans, while the eastern Pacific trend is slightly negative. The interannual SST variations due to the El Niño- Southern Oscillation (ENSO) are superimposed onto these longer-term SST trends, which can potentially influence the structure and amplitude of the resulting atmospheric teleconnections. In this study, the cold phase of ENSO, or La Niña, is examined using composite differences based on the observations and the model-simulated response to SSTs. The analyses show that during the recent period (1980–2008), the global teleconnection pattern associated with La Niña has been associated with higher heights from the tropics to the mid-latitudes. The model simulations attribute these apparent changes in the teleconnections, across the tropics and the Northern Hemisphere, to warmer SSTs in the Indian Ocean and warm pool region. The implications of these results are discussed within the context of seasonal predictions in an evolving climate.
 Since 1950, tropical sea surface temperatures (SST) have trended upwards over most of the global oceans, with the most abrupt change occurring around 1976 [Kumar et al., 2004]. The upward linear trend has not been uniform, with the largest warming in the tropical Indian Ocean and the warm pool region of the western Pacific, while the trend is weakly downward in the tropical eastern Pacific (Figure 1). Using various datasets and statistical techniques, these general SST trends have been extensively documented by Cane et al.  and Compo and Sardeshmukh , among many others.
 Of particular relevance to this study is that the interannual variations of SSTs related to the El Niño–Southern Oscillation (ENSO) phenomenon are superimposed onto the tropical SST trends. SST anomalies associated with ENSO affect the amplitude and structure of the global atmospheric circulation [Hoerling and Kumar, 2002]. For instance, during the cold phase of ENSO, La Niña, a westward retraction of the warmest equatorial SSTs contributes to suppressed convection in the central and eastern Pacific. The redistribution of tropical heating is conducive to the formation of teleconnections, or wavetrains, which are quasi-fixed nodes and antinodes embedded within the atmospheric zonal flow that span the globe [Trenberth et al., 1998].
 It is the global teleconnection response associated with La Niña, in the presence of SST trends, that is the focus of this study; the importance of which has been highlighted in a series of studies related to prolonged droughts in the mid-latitudes [Hoerling and Kumar, 2003; Schubert et al., 2004]. In particular, we seek to understand how the atmospheric teleconnection during La Niña is influenced by the SST trends across the global tropics, and what may be the implications for seasonal predictions that key on tropical ENSO SST variability. The displacement of SSTs during La Niña, along with the tropical SST trends, can combine to influence the configuration and intensity of the resulting teleconnections. Skillful seasonal climate predictions are partially dependent on accurately resolving the ENSO-forced component of the extratropical circulation [Barnston et al., 1999].
2. Data and Model Simulations
 The gridded SST analysis is based on monthly ERSSTv3b from 1950–2008 [Smith et al., 2008]. For the same period, the atmospheric teleconnections are identified using gridded 200-hPa geopotential height data from the NCEP/NCAR Reanalysis [Kalnay et al., 1996]. The 1950–2000 base period is removed from the total fields to form anomalies. La Niña composites are formed based on the December–January–February (DJF) seasonal mean, and the La Niña events are selected based on the DJF Niño 3.4 SST index values less than the negative one standard deviation. Based on this definition, the La Niña events for pre-1980 period include 1951, 1956, 1971, 1974, and 1976, and for the post-1980 era are 1989, 1999, 2000, 2008 (with the year based on January and February in the DJF seasonal mean).
 Because of the small number of La Niña events in the post- and pre- 1980 periods, the observed 200-hPa height anomalies are supplemented with a large ensemble of simulations from three different atmospheric general circulation models (AGCMs): 24-members from the European Center/Hamburg model (ECHAM4.5); 10-members from the Experimental Climate Prediction Center's model; and 9-members from the Seasonal-to-Interannual Prediction Project model. The AGCMs are forced by the observed SST pattern using AMIP-style integrations. The anomalous 200-hPa geopotential height composites are based on the average of all 43 simulations (computed also from the DJF average from 1950 to 2000).
 Another set of model integrations were created by forcing an AGCM with the SST difference between the La Niña composites for the post-1980 and pre-1980 periods (Figure 2, top right). The SST difference over various spatial domains was added to a climatological seasonal cycle of SSTs, and then the model was run for 35 years.
3.1. Sea Surface Temperature: Trends and La Niña
 We start our analysis of the changes in tropical SST associated with La Niña by examining La Niña composites before and after 1980. Figure 2 shows the global SST anomaly composite for La Niña episodes prior to 1980 (Figure 2, top left), the composite based on La Niña after 1980 (Figure 2, bottom left), along with the difference between the two periods (Figure 2, top right). The difference field does not depend on the choice for climatology.
 A comparison between Figures 2 (top left) and 2 (bottom left) indicates that, during the post-1980 epoch, the “horseshoe pattern” of above-average anomalies in the Pacific Ocean has strengthened and its mid-latitude extensions have shifted equatorward. In the difference map (Figure 2, top right), this shift is manifest as increased warming near 30° latitude in the North Pacific and South Pacific, although the South Pacific warming signal also has a contribution from the diminished cooling that is evident just south of the equator (Figure 2, bottom left). While the La Niña SST pattern tends to be nearly symmetric about the equator, it is interesting to note that in the composite difference (Figure 2, top right), the trend is nearly asymmetric in the eastern half of the subtropical Pacific basin. Also notable is the increased cooling that is confined to a narrow equatorial region, which indicates that La Niña in the post-1980 period has a stronger below-average equatorial SST signal.
 Most remarkable in Figure 2 is the warming that is evident across nearly the entire 30°S–30°N span, and particularly in the Indian Ocean and in the warm pool region, just north of Australia. As Figure 1 illustrates, warming of tropical ocean basins accompanying the La Niña SST composites is consistent with the SST trend of tropical warming, especially over the Indian and the warm pool region of the Pacific.
 To further summarize the SST trends, Figure 2 (bottom right) shows the longitudinal profile (averaged 30°S to 30°N) of the SST anomalies associated with the La Niña events before and after 1980. For the La Niña events after 1980, anomalous SST is warmer across all of the tropical oceans. However, some of the largest warming is evident in the Indian Ocean and warm pool region (∼50°E to ∼150°E), and is a consequence of the pronounced warming trend (Figures 1 and 2).
 The SST trends documented in this study likely arise from a combination of low-frequency natural variations and anthropogenic influences [Hoerling et al., 2008; Compo and Sardeshmukh, 2009]. How much the character of the La Niña intrinsically has changed is undetermined and remains a topic of study. Some studies have indicated that low-frequency changes in the tropical Pacific have also resulted in altered characteristics for the ENSO variability [Ashok et al., 2007]. Irrespective of whether characteristics of the La Niña episodes have changed or not, we are interested in the consequences of SST trends and how they may potentially influence the overall teleconnection response associated with La Niña.
3.2. Global Teleconnection Response to La Niña
 By influencing the distribution of rainfall (and corresponding heating) anomalies, changes in SSTs can result in an altered ENSO teleconnection pattern. It would be ideal to first quantify the difference in observed rainfall composites for La Niña during the two periods, but oceanic rainfall estimates are not sufficiently reliable prior to the satellite era in 1979 [Lau and Wu, 2007]. On the other hand, the possible influence of changes in rainfall on atmospheric teleconnections can be gleaned from the analysis of large-scale circulation fields. Because the tropical-to-extratropical teleconnections are most apparent in the upper-level circulation, the anomalous 200-hPa geopotential height field is also examined.
 Shown in Figure 3 (left) are the observed 200-hPa height anomaly composites for La Niña before 1980 (Figure 3, top) and after 1980 (Figure 3, middle), and the difference between the two composites is shown in Figure 3 (bottom). The 200-hPa geopotential height response has considerable differences between the two periods with lower (higher) height anomalies during the La Niña before (after) 1980s, particularly across the subtropical-to-middle latitudes of both hemispheres. The banding structure of higher heights in these latitudes is also reminiscent of the zonally symmetric global response to SST anomalies reported during the extended La Niña event from 1998–2001 [Hoerling and Kumar, 2003; Lau et al., 2008; Seager et al., 2003]. The tropical latitudes also reflect increasing heights during the later period, especially over the warm pool region (Figure 3, bottom). The analysis also indicates that changes in teleconnection during La Niña are similar to the change in the mean state of the 200-hPa height over the two periods (see Figure S1 in the auxiliary material).
 Because the comparison of observed composites is based on a limited number of cases, they could be subject to sampling errors. To substantiate the observed composites of Figure 3 further, composites based on the AMIP model simulations are shown in Figure 3 (right). Since model runs are based on a large ensemble, the results are highly robust. The structure and scale of the observed composites are largely reproduced in the model runs, with increased height anomalies across much of the globe in the post-1980 composite. In the tropics, the largest modeled height response also occurs over the warm pool region, but, when compared to the observations, extends further into the Indian Ocean (Figure 3, bottom).
Figure 3 also shows that the models capture the above-average height banding in the middle latitudes of the Northern Hemisphere well, although there are some differences in the location of the strongest height anomalies. In addition, the increase (decrease) in observed heights across the Southern Hemisphere extratropics (polar latitudes) (Figure 3, bottom) are not well replicated in the model simulations. This disparity in the Southern Hemisphere is largely accounted for by the opposite signed anomalies in the pre-1980 model and observations composites (Figure 3, top). Further, although there is a general agreement between the model and observations in the post-1980 composites, large negative anomalies over Antarctica are not well simulated. Work by L'Heureux and Thompson  also help to confirm the Southern Hemisphere results shown in the post-1980 La Niña composites, which reflects the relationship between La Niña and the positive polarity of the Southern Annular Mode (SAM). Another notable difference is over the North Atlantic, a feature that in observations is due to recent trend in the North Atlantic Oscillation (NAO), but cannot be replicated in the AMIP simulations from the knowledge of SST alone.
 Regardless of these differences, a substantial change is apparent in the global 200-hPa teleconnection response during La Niña in the pre versus post 1980 period, and it is supported by both the observations and the AMIP model simulations. To help diagnose which SST changes account for the atmospheric circulation changes, a set of additional idealized AGCM simulations forced with differences in the La Niña SST composites (Figure 2, top right) is created. For each experiment, specified SST anomalies were superimposed on a climatological SST used in a control simulation. All experiments, including the control simulation, are integrated for 35-years and the results are presented as an average anomaly relative to the control simulation.
 Four experiments are created and the 200-hPa height responses are shown in Figure 4. Figure 4 (top left) is forced by the global SST pattern (“global”), Figure 4 (top right) is forced by SST between 30°S and 30°N (“tropics-alone”), Figure 4 (bottom left) is forced by SST in the Indian Ocean and warm pool region (“Indian Ocean/warm pool-alone”), and Figure 4 (bottom right) is forced by the tropical Pacific/Atlantic between 30°S and 30°N (“tropical Pacific/Atlantic-alone”). Rectangular boxes in Figure 2 indicate the spatial extent of the various SSTs domains.
 When the Figure 4 simulations are compared to Figure 3 (bottom) it becomes evident that nearly all of the simulations (except for the tropical-Pacific/Atlantic-alone run) provide a good match to the observed and modeled teleconnection changes. In particular, these three experiments capture the prevalence of above-average heights over the tropics and in the mid-latitudes of the Northern Hemisphere. The greatest positive height anomaly in the tropics is evident over the warm pool and Indian Ocean, and which appears both in the AMIP runs and in the observations (Figure 3, bottom). In contrast, the only simulation that does not match the differences in the observations well is the tropical Pacific/Atlantic-alone simulation. The tropical Pacific/Atlantic-alone pattern (the pattern with negative SST trends confined to the equator) leads to an atmospheric circulation response that is most reflective of a La Niña height pattern in pre-1980 era (Figure 3, top), but still indicating higher height anomalies over the Northern Hemisphere.
 More interestingly, the Indian Ocean/warm pool-alone simulation closely replicates the tropics and extratropical Northern Hemisphere height patterns that are also apparent in the global and tropics-alone experiments. The only clear difference between the three runs is that the amplitude of the higher heights is stronger over the Atlantic sector in the global and tropics-alone simulations. Regardless, running a model forced just by warming trend in the Indian Ocean/warm pool region, captures nearly all of the above-average height anomalies across the tropics and middle latitudes of the Northern Hemisphere.
 It is also interesting to note that most of the model simulations forced with the SST differences in Figure 2 (top right) lead to Southern Hemisphere circulation anomalies that generally oppose the observed relationship between La Niña and the positive phase of the SAM. This result is consistent with the analysis of Li et al.  who, based on model simulations, also noted a similar tendency for the warm Indian Ocean SST anomaly to oppose recent trend in the SAM.
 In conclusion, these results show that the pre-1980 and post-1980 changes in the La Niña teleconnection pattern, particularly across the tropics and Northern Hemisphere, are primarily in response to the SST trend in the Indian Ocean and warm pool region.
 The analysis of potential changes in the ENSO teleconnections in a changing climate has also been reported in recent studies [Meehl and Teng, 2007; Müller and Roeckner, 2008]. The analysis presented in this paper further confirms that changes in the atmospheric mean state, which can be attributed to SST trends during the last 50-years, coincide with changes in teleconnections during La Niña. From the perspective of seasonal prediction, where the temporal and regional character of SST is important to ascertain global influence of ENSO, the results in this study indicate that tropical SST trends since 1950 (particularly in the Indian Ocean and warm pool region) contributed to changes in the global teleconnection response to La Niña. Further, if the La Niña teleconnection is accompanied with systematic circulation changes in the recent period, as this study indicates, then the seasonal climate outlooks need to be supplemented and adjusted using model simulations that best capture the observed changes in atmospheric teleconnection due to the recent SST trend.
 We would like to thank Amy Butler and Zeng-Zhen Hu, and two anonymous reviewers for their constructive comments.