Extra-tropical atmospheric response to ENSO in the CMIP5 models
The seasonal mean extra-tropical atmospheric response to El Niño/Southern Oscillation (ENSO) is assessed in the historical and pre-industrial control CMIP5 simulations. This analysis considers two types of El Niño events, characterized by positive sea surface temperature (SST) anomalies in either the central equatorial Pacific (CP) or eastern equatorial Pacific (EP), as well as EP and CP La Niña events, characterized by negative SST anomalies in the same two regions. Seasonal mean geopotential height anomalies in key regions typify the magnitude and structure of the disruption of the Walker circulation cell in the tropical Pacific, upper tropospheric ENSO teleconnections and the polar stratospheric response. In the CMIP5 ensembles, the magnitude of the Walker cell disruption is correlated with the strength of the mid-latitude responses in the upper troposphere i.e., the North Pacific and South Pacific lows strengthen during El Niño events. The simulated responses to El Niño and La Niña have opposite sign. The seasonal mean extra-tropical, upper tropospheric responses to EP and CP events are indistinguishable. The ENSO responses in the MERRA reanalysis lie within the model scatter of the historical simulations. Similar responses are simulated in the pre-industrial and historical CMIP5 simulations. Overall, there is a weak correlation between the strength of the tropical response to ENSO and the strength of the polar stratospheric response. ENSO-related polar stratospheric variability is best simulated in the “high-top” subset of models with a well-resolved stratosphere.
El Niño/Southern Oscillation (ENSO) events contribute to extra-tropical atmospheric variability. Recent literature recognizes two types of El Niño events: eastern Pacific (EP) El Niño and central Pacific (CP) El Niño. These events are characterized by positive sea surface temperature (SST) anomalies in the eastern and central equatorial Pacific, respectively, in boreal autumn and winter (Ashok et al. 2007; Kug et al. 2009).
Meteorological reanalyses and atmospheric simulations forced by observed SSTs show that both types of El Niño modulate the extra-tropical atmosphere. Whereas EP El Niño events do not impact the extra-tropical Southern Hemisphere (SH) stratosphere (e.g., Hurwitz et al. 2011a), CP El Niño events enhance convective activity in the South Pacific Convergence Zone in austral spring, forcing a tropospheric planetary wave that propagates toward SH high latitudes and upward into the Antarctic stratosphere. This wave enhancement affects Antarctic surface temperatures (Schneider et al. 2012) and sea ice concentrations (Song et al. 2011), and leads to anomalously high polar stratospheric temperatures during austral summer (Hurwitz et al. 2011a and Hurwitz et al. 2011b; Zubiaurre and Calvo 2012).
While the NH impacts of EP El Niño events are well established, fewer studies have examined the impacts of CP El Niño events. Manzini et al. (2006), García-Herrera et al. (2006) and Randel et al. (2009), among other studies, have shown that EP El Niño events deepen the North Pacific low and enhance planetary wave driving, leading to a weakening of the Arctic vortex in boreal winter. Garfinkel et al. (2012a) found deepening of the North Pacific low and weakening of the Arctic vortex in late boreal winter, in response to both EP and CP El Niño, in long model simulations. Other analyses of the NH response to CP El Niño (Hegyi and Deng 2011; Xie et al. 2012; Graf and Zanchettin 2012) have found contradictory results, possibly resulting from the dependence of the NH response on the precise definition of CP El Niño and/or the small number of observed events included in these analyses (Garfinkel et al. 2012a). The present multi-model study will support efforts to better understand the NH response to CP El Niño events.
The atmospheric response to La Niña events, characterized by negative SST anomalies in the central and/or eastern equatorial Pacific, is less constrained. La Niña events in the satellite era have had a breadth of locations and magnitudes, and thus it is harder to assess their extra-tropical teleconnections. The available observations suggest that, in the seasonal mean, La Niña events tend to have the opposite atmospheric impacts as El Niño events in the NH, namely a weakening of the North Pacific low (DeWeaver and Nigam, 2002) and a relative strengthening of the Arctic stratospheric vortex (Garfinkel et al. 2012b). Simulations by Zubiaurre and Calvo (2012) indicate that the simulated atmospheric response to CP La Niña is roughly opposite that of CP El Niño. However, Hoerling et al. (1997) and Mitchell et al. (2011) find that, in the NH winter, EP El Niño and La Niña teleconnections are nonlinear. Manzini et al. (2006) find no significant response to La Niña.
In this paper, recent developments in the understanding of the extra-tropical atmospheric response to ENSO, as outlined above, will serve to evaluate interannual atmospheric variability in the Coupled Model Intercomparison, Phase 5 (CMIP5) models (Taylor et al. 2012). The CMIP5 models include an interactive ocean, and are thus capable of simulating ENSO-like variability in the tropical Pacific. The CMIP5 models simulate interannual variability in the central and eastern equatorial Pacific with realistic amplitude (Bellenger et al. 2013), though few models are able to capture strong EP ENSO events (Kim and Yu 2012). More CMIP5 models show a realistic range of ENSO frequencies in the 2–7 year band, in the eastern equatorial Pacific, than for the CMIP3 group of models (Bellenger et al. 2013). SST anomalies peak in November through January, consistent with observations, in approximately half of the CMIP5 models (Bellenger et al. 2013). This is the first multi-model study to consider the responses to two flavors (EP and CP) and both phases (El Niño and La Niña) of ENSO, in both hemispheres. Section 2 introduces the methods, ENSO atmospheric response diagnostics and the meteorological reanalysis dataset used to compare with the CMIP5 models. Section 3 diagnoses the boreal winter response to ENSO in the CMIP5 models. Section 4 summarizes the conclusions and presents a brief discussion.
2.1 Geopotential height diagnostics
Geopotential height diagnostics
Walker circulation disruption (WC)
Z′(200–230°E, 10°S–10°N, 250 hPa)—Z′(110–130°E, 10°S–10°N, 250 hPa), NDJF
North Pacific low minimum (NPLM)
min[Z′(170–230°E, 30–60°N, 250 hPa)] for El Niño events, DJF
max[Z′(170–230°E, 30–60°N, 250 hPa)] for La Niña events, DJF
South Pacific low min (SPLM)
min[Z′(170–220°E, 45–30°S, 250 hPa)] for El Niño events, OND
max[Z′(170–220°E, 45–30°S, 250 hPa)] for La Niña events, OND
Arctic vortex weakening (NPVW)
Z(north of 59°N, 50 hPa), DJF
Antarctic vortex weakening (SPVW)
Z(south of 59°S, 50 hPa), OND
WC diagnoses the ENSO response to the difference between Z′ in the equatorial eastern Pacific and Z′ in the equatorial western Pacific. Positive values suggest weakened upwelling (i.e., weakened convection and relative downward motion) in the eastern Pacific and enhanced upwelling in the Southeast Asian region. That is, disruption of the zonal Walker Circulation, as expected during El Niño events. Negative values suggest the reverse: a strengthening of the Walker Circulation, as expected during La Niña events. WC measures the change in upper tropospheric divergence associated with El Niño and La Niña events, and thus is a good indicator of the strength of the atmospheric teleconnections. This diagnostic is computed for the NDJF (boreal winter) seasonal mean, coincident with the observed peak in tropical SST anomalies. WC is strongly anti-correlated (r = −0.81, for the 1979–2011 period) with an outgoing longwave radiation (OLR) diagnostic based on the NOAA interpolated OLR dataset (Liebmann and Smith 1996), and constructed from differences between the same two regions as for WC. Thus, WC serves as a proxy for the physical tropical atmospheric response to ENSO.
NPLM diagnoses changes in the strength of the North Pacific low. The NPLM is designed such that negative values indicate a deepening of the North Pacific low, as expected during El Niño events (Garfinkel and Hartmann 2008), while positive values indicate a weakening of the North Pacific low, as expected during La Niña events (Garfinkel et al. 2012b). The minimum (maximum) Z’ in the Aleutian low region (170–230°E, 30–60°N), at 250 hPa, is calculated for each El Niño (La Niña) event. Using the minimum (maximum) values of Z’ allows the precise location of the North Pacific low to vary between the CMIP5 models. DJF geopotential height anomalies are used to compute the NPLM, since previous studies (e.g., Garfinkel and Hartmann 2008) have shown that the extra-tropical upper tropospheric response to ENSO is strongest in this season. Note that NPLM yields stronger correlations with WC than does a diagnostic of the average geopotential height anomaly in the Aleutian low region; this result suggests that the precise geographic position of the low varies between models.
SPLM is equivalent to NPLM, but diagnoses changes in the strength of the South Pacific low for the OND seasonal mean. The minimum (maximum) Z’ in the region 170–220°E, 45–30°S, at 250 hPa, is calculated for each El Niño (La Niña) event. Atmospheric reanalyses and simulations have shown that the extra-tropical upper tropospheric and stratospheric response to CP ENSO is strongest in austral spring and early summer (Hurwitz et al. 2011a, 2013).
NPVW and SPVW diagnose the polar cap responses to ENSO at 50 hPa. Zonal mean geopotential height at 50 hPa is calculated poleward of 59° latitude in the Arctic (NPVW) and Antarctic (SPVW). Positive values of these diagnostics indicate relative vortex weakening, as expected during El Niño, while negative values indicate vortex strengthening, as expected during La Niña. NPVW is computed for the DJF season, when the observed NH response to ENSO peaks at 50 hPa (Calvo Fernandez et al. 2004; Manzini et al. 2006). Similarly, SPVW is computed for the OND season (Hurwitz et al. 2011a, b, 2013).
2.2 CMIP5 simulations
Lists of CMIP5 models included in the analysis
Niño index anomalies, numbers of composited events and mean composite size in MERRA and the CMIP5 simulations
EP El Niño
EP La Niña
CP El Niño
CP La Niña
1991/1992, 1994/1995, 2002/2003, 2004/2005, 2009/2010
1988/1989, 1998/1999, 1999/2000
NDJF Niño anomaly
Mean composite size
NDJF Niño anomaly
Mean composite size
NDJF Niño anomaly
Four types of ENSO events are examined: EP El Niño, characterized by positive sea surface temperature (SST) anomalies in the Niño-3 region (5°S–5°N, 210°–270°E), and CP El Niño, characterized by positive SST anomalies in the Niño-4 region (5°S–5°N, 160°–210°E), as well as EP and CP La Niña events, characterized by negative SST anomalies in the same two regions. ENSO events are identified based on NDJF seasonal mean SST anomalies, constructed from the monthly mean SST fields provided by the CMIP5 archive. The NDJF season generally corresponds with the observed boreal winter peak in equatorial SST anomalies. SST timeseries are de-trended and computed with respect to the 1979–2000 period (for the historical simulations), and SST anomalies are computed with respect to each 100-year subset (for the piControl simulations). EP El Niño events are identified when the Niño-3 anomaly is both greater than 1 standard deviation above the appropriate mean and 0.1 K larger than the corresponding Niño-4 anomaly. Similarly, EP La Niña events are identified when the Niño–3 anomaly is both less than 1 SD below the climatological mean and 0.1 K less than the Niño-4 anomaly. CP El Niño and CP La Niña events are identified analogously. While Kug and Ham (2011) determined that two distinct types of La Niña events could not be identified from the observational record, the present analysis uses two types of La Niña events for ease of comparison with the response to El Niño events and to highlight similarities between the EP and CP responses. Note that similar results were found when the models with fewer than three events of a particular type in the historical simulations (five events in the piControl simulations) were excluded from the analysis.
2.3 Observational datasets
The Extended Reconstructed Sea Surface Temperature (ERSST) Version 3b dataset (Xue et al. 2003) is used to identify observed ENSO events between 1979 and 2012. Table 3 documents the observed ENSO events included in this analysis and their associated Niño indices; note the small number of events as compared with the CMIP5 multi-model means. No observed events meet the criteria for EP La Niña.
Monthly mean geopotential height fields from the Modern-Era Retrospective Analysis for Research and Applications (MERRA) reanalysis (Rienecker et al. 2011) diagnose the observed atmospheric response to ENSO. Data are available from 1979 through 2012. Z anomalies are computed with respect to the 1979–2000 climatological mean.
3.1 Upper tropospheric seasonal mean response
In the extra-tropics, the upper tropospheric responses to El Niño exemplify the Rossby wave responses known as the Pacific-North America (PNA) pattern in the NH and the Pacific-South America–1 (PSA–1) pattern in the SH (Wallace and Gutzler 1981; Mo and Paegle 2001; Calvo Fernandez et al. 2004). These responses are characterized by patterns of upper tropospheric divergence and convergence anomalies that originate in the tropical Pacific, and can be seen in Fig. 1 as alternating patches of positive and negative geopotential height anomalies in the meridional direction, in the Pacific sector. The North Pacific and South Pacific lows strengthen in response to both types of El Niño: mid-latitude height anomalies are negative in the central and western Pacific region (Fig. 1). Correspondingly, both NPLM and SPLM are negative in response to EP and CP El Niño, in MERRA and in both sets of CMIP5 simulations (Figs. 2b, c). NPLM and SPLM are positive in response to both types of La Niña events.
Correlations between geopotential height diagnostics in the CMIP5 historical simulations
EP El Niño
EP La Niña
CP El Niño
CP La Niña
WC and NPLM
WC and SPLM
NPLM and NPVW
WC and NPVW
SPLM and SPVW
WC and SPVW
WC and NPLM
WC and SPLM
NPLM and NPVW
WC and NPVW
SPLM and SPVW
WC and SPVW
3.2 Polar stratospheric seasonal mean response
As discussed in Sect. 1, previous observational and single-model studies have shown that CP El Niño events tend to weaken the polar vortices while CP La Niña events strengthen the polar vortices. EP El Niño events tend to weaken the Arctic stratospheric vortex and have a negligible impact on the Antarctic vortex. While the MERRA reanalysis demonstrates the sign of these expected responses, likely the responses are not statistically robust because of the small composite size (Table 3). The present CMIP5 study will test the robustness of these previous results.
The simulated Antarctic stratospheric responses to ENSO are weaker than in MERRA and only some have the expected sign (Fig. 2e). Figure 4b shows that the magnitude of the Walker Circulation anomaly is not related to the simulated Antarctic vortex weakening.
4 Conclusions and discussion
The CMIP5 models simulate the expected upper tropospheric responses to ENSO in boreal autumn and winter. El Niño events disrupt the zonal Walker circulation cell, leading to increased upper tropospheric geopotential heights in the eastern tropical Pacific but decreased heights in the western tropical Pacific. The CMIP5 models generate upper tropospheric wavetrains in response to ENSO. In the historical simulations, the structure of these wavetrains is consistent with the MERRA reanalysis composites of ENSO events in the satellite era. On average, El Niño and La Niña events generate the opposite seasonal mean, extra-tropical atmospheric responses. For example, the mid-latitude North Pacific and South Pacific lows deepen during El Niño events, but weaken during La Niña events. Consistent with the realistic simulation of the North Pacific low response to ENSO, Polade et al. (2013) find the North Pacific teleconnections that link an ENSO–PDO mode with North American winter precipitation are well captured in the CMIP5 models.
In the Arctic stratosphere, observations suggest that the wintertime polar vortex should weaken in response to El Niño events and strengthen in response to La Niña events. The CMIP5 historical and piControl multi-model means do show a vortex weakening during both types of El Niño events and vortex strengthening during both types of La Niña events, in boreal winter, but generally with weaker magnitudes than for the MERRA reanalysis. Furthermore, in the CMIP5 models, there are more strong correlations between the upper tropospheric responses to ENSO and the polar stratospheric responses in the high-top models (i.e., those with a well-resolved stratosphere) than in the low-top models, consistent with Charlton-Perez et al. (2013).
In the Antarctic stratosphere, most CMIP5 models fail to capture the observed weakening of the polar vortex in response to CP El Niño events. This result reflects a weak poleward planetary wave response in the CMIP5 models, which in turn reflects a weak SPCZ convective response to CP El Niño. Brown et al. (2013) found that the southeastern “diagonal” portion of the SPCZ is poorly represented in the CMIP5 models: the SPCZ precipitation signature is both too zonal and confined to the deep tropics. The GEOSCCM captures the expected stratospheric response to CP El Niño when forced by observed SSTs (Hurwitz et al. 2011b), but fails to capture this response when forced by SSTs from CCSM3, one of the CMIP3 models with a tropically confined SPCZ (Hurwitz et al. 2013). Thus, it is expected that better simulation of the SPCZ, and its full response to ENSO, would improve the simulation of ENSO teleconnections in the SH.
The CMIP5 atmospheric responses to EP and CP ENSO flavors are indistinguishable. In the NH, this result is consistent with the single model study by Garfinkel et al. (2012a). However, in the SH, a strong poleward wavetrain and vortex weakening have been shown to occur only during CP-type El Niño events, both in observations and in model studies (Hurwitz et al. 2011a, b; Zubiaurre and Calvo 2012). The CMIP5 models’ failure to capture two distinct boreal winter mean responses to El Niño in the SH may occur because these models (1) fail to capture the convective response to CP El Niño in the SPCZ region, as discussed above, and (2) are not able simulate two distinct ENSO modes. Only 9 of the 20 CMIP5 models investigated by Kim and Yu (2012) were able to simulate both the EP and CP ENSO modes with realistic amplitudes. Kug et al. (2012) found that only 5 of 21 CMIP5 models simulated the observed negative correlation between SST variability in the Niño-3 and Niño-4 regions.
The magnitudes of the equatorial SST anomalies associated with ENSO events are similar in the CMIP5 historical and pre-industrial simulations, as are the resulting extra-tropical atmospheric responses to these events. The structure and magnitude of the seasonal mean, upper tropospheric teleconnections in the piControl simulations (Fig. 5) are similar to those in the historical simulations (Figs. 1, 3, 4). That is, the large difference in climate forcings between the two sets of simulations does not affect the essence of the ENSO teleconnections. This conclusion is consistent with Hurwitz et al. (2013), who compared the atmospheric response to CP El Niño under contemporary and late twenty-first century climate conditions.
Detecting robust stratospheric responses to ENSO requires composites of at least 20 El Niño or La Niña events (Garfinkel et al. 2012a). The small number of events observed during the satellite era limits confidence in the atmospheric responses in MERRA; however, the signs of the responses in this analysis agree with studies with larger sample sizes (e.g., modeling studies by Hurwitz et al. 2011b; Zubiaurre and Calvo 2012; Garfinkel et al. 2012a). While individual historical simulations had composites as small as 2 events (1 event in the piControl simulations; see Table 3), the multi-model means included many dozens of events (i.e., 4–10 events per model × 30 models), and thus represent a large enough sample to detect robust stratospheric signals.
Margaret M. Hurwitz thanks the NASA Atmospheric Composition, Modeling and Analysis Program (ACMAP) and Modeling, Analysis and Prediction (MAP) program for funding, and the World Climate Research Programme (WCRP) and Stratospheric Processes and their Role in Climate (SPARC) DynVar for travel support. Sarah Ineson was supported by the Joint DECC/Defra Met Office Hadley Centre Climate Programme (GA01101). The authors thank two anonymous reviewers for their helpful feedback.