The El Niño–Southern Oscillation (ENSO) influences the most extensive tropospheric circulation cells on our planet, known as Hadley and Walker circulations. Previous studies have largely focused on the effect of ENSO on the strength of these cells. However, what has remained uncertain is whether interannual sea surface temperature anomalies can also cause synchronized spatial shifts of these circulations. Here, by examining the spatiotemporal relationship between Hadley and Walker cells in observations and climate model experiments, we demonstrate that the seasonally evolving warm-pool sea surface temperature (SST) anomalies in the decay phase of an El Niño event generate a meridionally asymmetric Walker circulation response, which couples the zonal and meridional atmospheric overturning circulations. This process, which can be characterized as a phase-synchronized spatial shift in Walker and Hadley cells, is accompanied by cross-equatorial northwesterly low-level flow that diverges from an area of anomalous drying in the western North Pacific and converges towards a region with anomalous moistening in the southern central Pacific. Our results show that the SST-induced concurrent spatial shifts of the two circulations are climatically relevant as they can further amplify extratropical precipitation variability on interannual timescales.
Changes in the zonal equatorial Walker cell (WC; tropical-mean zonal cell) and the meridional Hadley Cell (HC; zonal-mean meridional cell) are known to cause major climate disruptions across our planet. Because of their considerable impacts on various regional climates and extreme events, such as heat waves (Garcia-Herrera et al., 2010), tropical cyclones (Wu et al., 2018), sea level rise in the western Pacific (Timmermann et al., 2010), droughts (Dai, 2011; Lau and Kim, 2015) and regional monsoon variability (Kumar et al., 1999; Bollasina et al., 2011), the variations in the strength and position of WC and HC have been examined extensively across a wide range of timescales. It has been shown that changes in the strength of the WC are connected to those of the HC, in part due to the shared ascending branch of zonal and meridional overturning cells in the warm-pool region (Vecchi and Soden, 2007; England et al., 2014; Liu and Zhou, 2017; Ma et al., 2018; Klein et al., 1999; Karnauskas and Ummenhofer, 2014). On interannual timescales, the co-variability between WC and HC strengths is tied to the El Niño–Southern Oscillation (ENSO)-related sea surface temperature (SST) gradients along the Equator, associated with uneven spatial distribution of tropical convection (Oort and Yienger, 1996; Klein et al., 1999; Minobe, 2004). However, observational analyses suggest more complicated relationships, which cannot be fully explained by peak ENSO dynamics alone (Clarke and Lebedev, 1996; Mitas and Clement, 2005; Tanaka et al., 2005; Ma and Zhou, 2016).
In general, these two large-scale circulations can change independently from each, either in terms of their strength or their geographical position, but during strong El Niño and La Niña winters these circulations can change in unison. More specifically during peak El Niño events in boreal winter, the WC weakens and the rising branch shifts eastward. This is accompanied by a strengthening of the HC and an enhanced upward motion in the equatorial region (Ma and Li, 2008; Bayr et al., 2014; Guo and Tan, 2018; Minobe, 2004; Klein et al., 1999) (also see Fig. S1 and Table S1 in the Supplement). For peak La Niña conditions in winter the atmospheric response is approximately the opposite. Previous studies have mostly focused on the response of the HC and WC strength to the peak phase of ENSO (i.e., a weakening of WC and strengthening of HC). Here we address a different question: Under what circumstances do the WC and HC show concurrent shifts in their geographic position? We further investigate whether these two dominant atmospheric circulation cells are coupled even in the absence of tropical ENSO-related SST anomalies. To address these questions, we conduct a comprehensive analysis of the dynamical coupling between HC and WC using observational data and a series of SST-forced atmospheric general circulation model (AGCM) simulations. We will focus on 2 important degrees of freedom that characterize variations in these circulations: their strength and spatial position.
We used the monthly reanalysis circulation dataset from the European Centre for
Medium-Range Weather Forecasts (ECMWF) atmospheric interim (ERAI) data from
1979 to 2017 (Dee et
al., 2011). In addition, we obtained the monthly SST and precipitation data
from the extended reconstruction of global SST (ERSSTv5) since 1854 (Smith et al., 2008) and Global Precipitation Climatology project
(GPCP) version 2.3 from 1979 to 2017 (Adler et al., 2003), as well as the
Climatic Research Unit (CRU) TS4.03 land precipitation from 1901 to 2017 (Harris et al., 2014), respectively. The monthly anomalies for
1979–2017 were calculated by removing the seasonal cycle and linear trend
coefficient during the analysis period. The ENSO variability was
characterized by a spatial average of SST anomalies over the tropical
eastern Pacific Niño3 region (5
We used 40 AGCM simulations from the forced AGCM Intercomparison Projection
(AMIP): 19 model runs are part of the Coupled Model Inter-comparison Project
(CMIP) Phase 5 (Taylor et al., 2012) and 21 models are part of
Phase 6 (Eyring et al., 2016) (i.e., AMIP5 and AMIP6; Tables S2 and S3). The AMIP simulations were forced by the observed SST and
sea ice concentrations from 1979–2008 for AMIP5 and from 1979–2014 for
AMIP6. The AMIP runs use the observed SST boundary forcing, which allows us
to directly compare SST-forced atmospheric responses in simulations and
observations. The AMIP multi-model ensemble (MME) represents the SST-forced
signal, whereas a mixed signal of SST-forced variability and unforced
atmospheric noise is present in the observations and individual AMIP
simulations. Additionally, to examine the impact of SST forcing amplitude on
the WC–HC synchronization, we analyzed a set of 40 multi-model CMIP5
historical simulations covering the industrial period from 1900–2005 (see Taylor et al., 2012 for details). Only one ensemble member
(r1i1p1) was used for each model. All observations and MME data were
interpolated to a regular
We calculated the monthly mass stream function (MSF) anomalies on the
three-dimensional atmospheric circulation to describe the WC and HC
circulations:
To examine the phase synchronization of the large-scale circulation, we
calculated the complex analytical signal (
Phase-synchronized spatial shifts of the Walker
and Hadley circulations.
First, we conduct an EOF analysis of the monthly MSF anomalies, obtained
from atmospheric reanalysis data covering the period 1979–2017. In addition
to their co-varying strength (characterized by EOF1 for WC (WC1) and EOF2
for HC (HC2) – shown in Fig. S1) that has been discussed
extensively in previous studies (Ma and Li, 2008; Bayr et al., 2014; Guo
and Tan, 2018; Minobe, 2004), we find that the EOF2 of the WC (WC2) and the
EOF1 of the HC (HC1) are related to each other (correlation coefficient CC
Previous studies showed that the ENSO-related SST variability leads to the
co-varying strength of WC and HC (i.e., Walker and Hadley strength modes)
(e.g., Minobe, 2004; Guo and Tan, 2018). Here, we explore the relationship
between ENSO and the co-varying spatial shifts of the WC and HC (Fig. 1c and
e). No statistically significant linear correlation between the PCs of HC1 or WC2 and Niño3 anomalies can be found on
interannual timescales (CC
This discrepancy calls for two important aspects of ENSO to be considered:
one is the peak-phase ENSO signal (which occurs in boreal winter); the other
are the seasonally modulated characteristics of ENSO, i.e., the combination
mode between ENSO and the Indo-Pacific warm-pool SST annual cycle (C mode;
with a time evolution that peaks in boreal spring) (Stuecker et al.,
2013, 2015). The C mode arises from an amplitude modulation
of the warm-pool annual cycle (with a frequency of 1 year
SST-forced phase-synchronized spatial shifts of
the HC and WC.
In the previous section, we showed that the “Walker and Hadley shift
modes” are coupled to each other, and we hypothesized an important role of
the seasonally modulated dynamics of ENSO. The previously applied linear
correlation analysis is well suited to analyze the amplitude relationships;
however, it is not sensitive to the seasonally modulated nonlinear coupling.
Therefore, before focusing on the physical linkage between these shift
modes, we further examine the co-varying WC–HC shifts in the framework of
phase synchronization and nonlinear coupling. Phase
synchronization, which requires nonlinear dynamics, is
characterized by cooperative and organized oscillatory behavior between two
fluctuating systems. To study this process, we first calculate the
generalized phase difference between WC2 and HC1 variabilities (i.e.,
We next examine the question of whether the synchronization of Walker and Hadley shift modes can in principle be driven by random atmospheric stochastic variability, or if SST-forced variability is a prerequisite. The SST-forced AMIP MME shows more frequent phase-synchronized (PSYN) months than the observations (Fig. 2a) and most CMIP5 models (Fig. S4), exhibiting the effects of both SST forcing and atmospheric noise: there are four long-lasting (greater than 6 months) PSYN periods in the observations but eight such events in the AMIP MME. The WC–HC phase synchronization is particularly prominent during the two extreme El Niño events (1982–1893 and 1997–1998) that have much longer PSYN periods (exceeding 1 year) in both observations and the AMIP MME. An interesting point is that the WC–HC phase difference shows a decadal change after 2000 in both observations and the AMIP MME. This may be a consequence of the Pacific WC intensification which can be linked to the concurrent tropical Atlantic warming and tropical eastern Pacific cooling (England et al., 2014; McGregor et al., 2014; associated with multi-decadal climate variability in both the Atlantic and Pacific) and/or zonal shifts in the dominant ENSO SST pattern (Sohn et al., 2013). Further exploration of the underlying physical mechanisms is, however, beyond the scope of this study.
Seasonal dependency of phase-synchronized spatial
shifts of the HC and WC.
Global pattern of phase-synchronized spatial
shifts of HC and WC. Composite anomalies during PSYN months (i.e., the
absolute tendency of phase difference is less than 0.3) with
February–March–April (FMA[1]) extreme El Niño (FMA[1] Niño3
To further explore the relative effects of random atmospheric noise and
SST-forced variability on the WC–HC synchronization, we generate pseudo-PCs
by randomly shuffling the PCs for both the observations and the AMIP MME and
repeating this 10 times. A comparison of the phase difference tendencies
using the original PCs (pink dots in Fig. 2b, c) and of the pseudo-PCs (sky-blue dots) reveals that during the evolution of El Niño events
(Niño3
Impact of phase-synchronized spatial shifts of
the HC and WC on global precipitation.
Our analysis has revealed that the Walker and Hadley shift modes are
more connected to the seasonally modulated dynamics of ENSO (i.e., the
C mode) rather than the peak-phase ENSO amplitude signal. Figure 3a clearly
shows the statistically significant co-variability between observed time
series of WC–HC PCs and the C mode (which is defined here as
Schematic diagram showing the strength and
spatial shift of Walker and Hadley circulations.
Our results suggest that the spring SSTgrad anomalies that occur after the
El Niño winter peak phase (Stuecker et al., 2015; Zhang et al., 2016)
play an important role for the phase synchronization of WC and HC shift
modes. We further examine the physical mechanism for the WC–HC phase-synchronized spatial shifts by exploring the global climate patterns that
occur during springtime phase-synchronized periods in WC and HC variability
(Fig. 4; composite when
We emphasize that the simulated ENSO SST anomaly amplitudes in coupled models of the CMIP5 are correlated with the probability of phase synchronization between the WC and HC shift modes (Fig. S4). The models exhibiting large ENSO variability can initiate strong teleconnections through the atmospheric bridge process, thus likely leading to the phase-synchronized zonal and meridional shifts of the WC and HC, respectively (see Fig. S8). An underlying cause could be the strong meridional asymmetry of the warm-pool climatological SST and zonal winds (Fig. S9) and C-mode-associated extreme shifts of the South Pacific Convergence Zone (SPCZ) shifts (McGregor et al., 2012). We further hypothesize that stronger phase synchronization of the WC–HC shift modes can generate larger global precipitation responses, regardless of the ENSO amplitude. To illustrate this, we use the individual 40 AMIP simulations that have identical ENSO SST amplitude prescribed as their boundary forcing but exhibit different strengths of WC2–HC1 phase synchronization. The strong PSYN models and weak PSYN models are thus classified as a function of their strength of WC2–HC1 phase synchronization, where this strength is measured by the (interannual) correlation coefficient between WC2 and HC1 PC variations in the individual AMIPs (see Tables S2 and S3 for chosen model groups).
By comparing the strong El Niño (i.e., 1982/1983 and 1997/1998) composite averaged for eight strong PSYN models with that averaged for eight weak PSYN models (Fig. 5), we see that the models with a strong coupling between the HC and WC shift modes generate a prominent springtime (FMA) asymmetry between WNP divergence and southern central Pacific convergence. In particular, the geographical positions of minimum low-level divergence and maximum low-level convergence exhibit a clear contrast between the strong PSYN models and weak PSYN models (see pink and cyan star symbols in Fig. 5): the NW-SE oriented pattern is predominant for the strong PSYN models, but it is less pronounced for the weak models. Accordingly, the increased precipitation anomalies in the extratropical regions (e.g., East Asia and South America) are larger for the strong PSYN models compared to the weak PSYN models (Fig. 5b and d). These differences in anomalous precipitation and circulation centers are statistically significant during boreal spring but not during boreal winter. We also emphasize that these differences in precipitation are statistically insignificant for the tropical regions, indicating that the extratropical precipitation response is not primarily a result from an enhanced tropical convective activity. Differences in model physics may generate a strengthened WC–HC synchronization (i.e., NW-SE skewed asymmetric circulation pattern changes), further intensifying the post-ENSO impact on extratropical precipitation variability. This reflects the pronounced impact of WC–HC phase-synchronized spatial shifts on extratropical precipitation variability, regardless of the ENSO amplitude.
Focusing on the zonal and meridional displacements of two major atmospheric circulation cells, using observational datasets and the AMIP MME simulations, we examined the intricate coupling between the WC and HC. In addition to a well-known ENSO-driven coupling of WC and HC variability amplitude, our study shows that these two circulations can also shift their positions in a synchronized manner, albeit in a more subtle manner. Figure 6 illustrates the important difference between the strength modes (WC1&HC2) and the shift modes (WC2&HC1). The ENSO-driven strength mode in its winter maximum is characterized by an equatorially symmetric Pacific cell pattern between western Pacific divergence and central Pacific convergence, leading to weakening WC and strengthening HC strengths (Fig. 6a). In contrast, our analysis revealed that the seasonally evolving springtime warm-pool SSTs in combination with post-peak-phase El Niño SST anomalies can produce a meridionally asymmetric WC and anomalous cross-equatorial flow (from off-equatorial WNP divergence zone toward the off-equatorial southern Ocean of central Pacific and Indian Ocean convergence zone), thus connecting the zonal WC and meridional HC (Fig. 6b). This feature is reminiscent of the atmospheric ENSO combination mode. Coherent WC–HC shifts have pronounced influences on precipitation, as well as wind and sea level patterns in both tropical and extratropical regions.
An important finding of our study is that the phase synchronization is only present for extreme El Niño events and not for La Niña, indicating an important nonlinearity of the tropical climate system. In addition, potential asymmetric heat capacitor effects of both the Indian and Atlantic oceans, which tend to be more pronounced during El Niño compared to La Niña (Ohba and Watanabe, 2012; An and Kim, 2018) may further affect the phase synchronization properties discussed here. We found that the meridionally asymmetric response for phase synchronization of the WC–HC shift mode is quite different from the meridionally symmetric response associated with the strength-related WC–HC modes (Figs. S10 and S11). Our analysis also highlights the fact that the analysis of phase relationships (Rosenblum and Pikovsky, 2003) between modes of climate variability can reveal important new insights into their physical coupling mechanisms, which are less apparent in amplitude space (as described for instance by temporal correlation coefficients).
Observation data for this research are available at (1) the ECMWF website
(
The supplement related to this article is available online at:
AT and KSY designed the study. KSY wrote the initial paper draft and produced all figures. KSY, AT and MFS contributed to the interpretation of the results and to the improvement of the paper.
The authors declare that they have no conflict of interest.
This study was supported by the Institute for Basic Science (project code IBS-R028-D1). This is IPRC publication 1496 and SOEST contribution 11219.
This research has been supported by the Institute for Basic Science (grant no. IBS-R028-D1).
This paper was edited by Ben Kravitz and reviewed by two anonymous referees.