Revisiting the Influence of ENSO on the Arctic Stratosphere in CMIP5 and CMIP6 Models

Abstract

The Arctic stratospheric response to El Niño–Southern Oscillation (ENSO) is assessed using the historical simulations provided by the Coupled Model Intercomparison Project Phases 5 and 6 (CMIP5 and CMIP6, respectively). CMIP6 models can well reproduce the ENSO signals in the Arctic stratosphere and have an ameliorated performance compared to CMIP5 models. Specifically, El Niño is associated with an intensified Pacific–North American pattern that leads to a considerable enhancement of planetary wavenumber 1 but a small reduction of planetary wavenumber 2, and thus, a warm and weakened stratospheric polar vortex. The case for La Niña is nearly the opposite, with a cool and strengthened stratospheric polar vortex. In CMIP6, the ENSO-related stratospheric signal matures in the February–March–April season and increases with ENSO magnitude, regardless of the ENSO phase. However, the stratospheric response to strong El Niño (La Niña) is weaker (stronger) than that which should be achieved if the response changes linearly with the amplitude of El Niño (La Niña). An asymmetric time evolution of stratospheric signals exists between strong El Niño and La Niña events. The stratospheric response caused by strong El Niño is weaker from late winter to early spring but stronger in middle and late spring compared to that caused by strong La Niña. By contrast, the Arctic stratospheric signal in moderate El Niño events is larger than that in moderate La Niña. Compared to ENSO-neutral winters, stratospheric sudden warming occurs more (less) frequently in El Niño (La Niña), as simulated by CMIP6 high-top models.

1. Introduction

El Niño–Southern Oscillation (ENSO), which occurs in the tropical Pacific Ocean, is one of the most important ocean–atmosphere coupled phenomena. It affects oceanic and atmospheric conditions in both the tropics and extratropics. Previous studies have pointed out that ENSO remarkably impacts the stratosphere in both hemispheres. In the Northern Hemisphere, the warm ENSO phase (i.e., El Niño) acts to warm and weaken the winter stratospheric polar vortex, as revealed by both reanalysis data and model simulations [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17].

El Niño can modulate the Arctic stratosphere mainly through its mid-tropospheric teleconnection in the North Pacific. Specifically, it can induce a positive Pacific–North American (PNA) pattern and deepen the Aleutian low that would, in turn, constructively interfere with the climatological stationary troughs [6,7,18]. As a result, during El Niño, the planetary wavenumber 1 enhances in the troposphere and propagates upward into the extratropical stratosphere. When the planetary wave breaks, it decelerates the stratospheric polar jet, warms the polar region, and favors more occurrences of stratospheric sudden warming (SSW) events [19]. Although planetary wavenumber 2 decreases simultaneously, its effect is much smaller than that of planetary wavenumber 1. This is also the case for La Niña (the cold ENSO phase) but with the opposite sign; that is, the planetary wavenumber 1 reduces, but the planetary wavenumber 2 increases in the North Hemisphere, and thereby the stratospheric polar vortex strengthens during La Niña winters. However, the intensity of the impact of La Niña on the Arctic stratosphere is far less than that of El Niño [2,4,20].

Several controversial issues remain regarding ENSO’s impact on the Arctic stratosphere. For example, the stratospheric responses to ENSO are asymmetric in the Northern Hemisphere, as commonly thought [2,9,10,13,21,22]; that is, the response to El Niño is stronger than that to La Niña. However, Weinberger et al. [23] argued that no asymmetry is found in the Arctic stratospheric responses to El Niño as compared to La Niña. Besides, Rao and Ren [9,10] pointed out that the stratospheric responses to strong El Niño and moderate La Niña are weaker than those to moderate El Niño and strong La Niña. In contrast, some studies argued that the influence of strong El Niño is stronger than that of moderate El Niño [23,24,25,26].

In addition, the time evolution of the stratospheric signals induced by ENSO is suggested to be different on sub-seasonal time scales. For instance, Calvo et al. [27] detected obvious signals of eastern Pacific El Niño in the boreal stratospheric polar region from November to March. Trascasa-Castro et al. [25] also found robust stratospheric responses to extreme El Niño as early as November. However, it is also argued that El Niño-related signals may not occur in the Arctic stratosphere earlier than January [2,22,24]. There are also controversies in the relationship between ENSO and SSW. It is revealed that El Niño tends to cause more SSW events while La Niña leads to fewer ones, compared to ENSO-neutral winters [19,23]. On the contrary, Butler and Polvani [28] and Butler et al. [29] reported that the occurrence frequency of SSW in La Niña winters is equal to that in El Niño winters in observations. Song and Son [30] indicated that the relationship between ENSO and SSW depends on the SSW identification.

It is suggested that sufficient composite samples are necessary to detect robust stratospheric response to ENSO [8,23,31]. The model simulations from the Coupled Model Intercomparison Project (CMIP) can provide large samples of ENSO events to examine their signals. For example, Hurwitz et al. [21] analyzed ENSO signals in the extratropical atmosphere using 13 high-top models from the CMIP Phases 5 (CMIP5). Calvo et al. [27] also assessed the Northern Hemisphere stratospheric response to different El Niño flavors based on 11 CMIP5 high-top models.

This study aims to revisit the influence of ENSO on the Arctic stratosphere in CMIP Phases 6 (CMIP6) models and present a comparison with CMIP5. We focus on the following two aspects: (1) the responses of the Arctic stratosphere to ENSO in CMIP6 models, such as the relationship between ENSO and SSW events, the asymmetry and nonlinearity of ENSO signals in the stratospheric polar region, and (2) comparisons of the model performance in simulating the ENSO’s impact on the boreal winter stratospheric polar vortex between CMIP6 and CMIP5 models and between CMIP6 high-top and low-top models.

2. Data and Methods

2.1. Data

In this study, we select 21 CMIP5 (

Table 1. CMIP5 models used in this study.

No.	CMIP5 Models	Resolution for Atmospheric Variables (Lon × Lat)	Resolution for SST (Lon × Lat)	Time Period
1	ACCESS1-0	192 × 144	360 × 300	1950–2005
2	ACCESS1-3	192 × 144	360 × 300	1950–2005
3	CanESM2	128 × 64	256 × 192	1950–2005
4	CMCC-CESM	96 × 48	182 × 149	1950–2005
5	CMCC-CMS	192 × 96	182 × 149	1950–2005
6	CNRM-CM5	256 × 128	362 × 292	1950–2005
7	FGOALS-s2	128 × 108	360 × 196	1950–2005
8	GFDL-CM3	144 × 90	360 × 200	1860–2005
9	GFDL-ESM2G	144 × 90	360 × 210	1861–2005
10	IPSL-CM5A-LR	96 × 96	182 × 149	1850–2005
11	IPSL-CM5A-MR	144 × 143	182 × 149	1850–2005
12	IPSL-CM5B-LR	96 × 96	182 × 149	1850–2005
13	MIROC5	256 × 128	256 × 224	1850–2005
14	MIROC-ESM	128 × 64	256 × 192	1950–2005
15	MIROC-ESM-CHEM	128 × 64	256 × 192	1950–2005
16	MPI-ESM-LR	192 × 96	256 × 220	1950–2005
17	MPI-ESM-MR	192 × 96	802 × 404	1950–2005
18	MPI-ESM-P	192 × 96	256 × 220	1950–2005
19	MRI-CGCM3	320 × 160	360 × 368	1950–2005
20	MRI-ESM1	320 × 160	360 × 368	1950–2005
21	NorESM1-M	144 × 96	320 × 384	1950–2005

) and 15 CMIP6 models (

Table 2. CMIP6 models used in this study.

No.	CMIP6 Models	Resolution for Atmospheric Variables (Lon × Lat)	Model Top	Resolution for SST (Lon × Lat)	Time Period
1	ACCESS-CM2	192 × 144	85 km (High-top)	360 × 300	1950–2014
2	BCC-ESM1	128 × 64	2.19 hPa (Low-top)	360 × 232	1950–2014
3	CanESM5	128 × 64	1 hPa (High-top)	361 × 290	1850–2014
4	CESM2	288 × 192	2.25 hPa (Low-top)	320 × 384	1850–2014
5	CESM2-FV2	144 × 96	2.25 hPa (Low-top)	320 × 384	1850–2014
6	CESM2-WACCM	288 × 192	4.5 × 10–6 hPa (High-top)	320 × 384	1850–2014
7	EC-Earth3	512 × 256	0.01 hPa (High-top)	362 × 292	1850–2014
8	GFDL-CM4	360 × 180	10 hPa (Low-top)	1440 × 1080	1850–2014
9	IPSL-CM6A-LR	144 × 143	0.01 hPa (High-top)	362 × 332	1850–2014
10	MPI-ESM-1-2-HAM	192 × 96	0.01 hPa (High-top)	256 × 220	1850–2014
11	MPI-ESM1-2-HR	384 × 192	0.01 hPa (High-top)	802 × 404	1850–2014
12	MPI-ESM1-2-LR	192 × 96	0.01 hPa (High-top)	256 × 220	1850–2014
13	MRI-ESM2-0	320 × 160	0.01 hPa (High-top)	360 × 364	1950–2014
14	NorESM2-LM	144 × 96	3 hPa (Low-top)	360 × 384	1950–2014
15	NorESM2-MM	288 × 192	3 hPa (Low-top)	360 × 384	1950–2014

) with daily geopotential height, air temperature, zonal and meridional wind, and sea surface temperature (SST) fields available to us. The first member (r1i1p1 and r1i1p1f1) of the historical experiments for each model has been analyzed. The horizontal resolution of SST and atmospheric variables are interpolated to 1° × 1° and 2.5° × 2.5°, respectively, based on the bilinear interpolation method. Atmospheric variables include eight pressure levels (1000, 850, 700, 500, 250, 100, 50, and 10 hPa) in the vertical direction. The time period covered by each model is shown in Table 1 and Table 2.

2.2. Methods

To identify ENSO events, we first calculate the standardized detrended winter (December–January–February; DJF)-mean Niño3.4 index by averaging the SST anomalies in the Niño3.4 region (5° S–5° N, 170°–120° W). We focus on strong and moderate ENSO events. In each model, strong El Niño (La Niña) events are defined as the Niño3.4 index above 1.5 (below –1.5) standard deviations, and moderate El Niño (La Niña) events as the Niño3.4 index between 0.75 and 1.5 (–0.75 and –1.5) standard deviations. Other winters are classified as ENSO-neutral winters. The total number of strong and moderate ENSO events identified in the CMIP5 and CMIP6 models is shown in

Table 3. Numbers of ENSO events and SSW events in different ENSO groups in CMIP5 and CMIP6 models.

Models	ENSO Magnitude	Winter SST Anomalies in Niño3.4 Region	Winter Numbers	SSW Numbers	SSW Frequency
CMIP6	Strong El Niño	2.32	158	97	0.61
	Moderate El Niño	1.19	299	179	0.60
	Strong La Niña	–2.01	108	61	0.56
	Moderate La Niña	–1.26	356	187	0.53
	Neutral	—	1039	618	0.59
	All	—	1960	1142	0.58
CMIP5	Strong El Niño	1.61	155	52	0.34
	Moderate El Niño	0.80	273	94	0.34
	Strong La Niña	–1.54	74	25	0.34
	Moderate La Niña	–1.01	271	80	0.30
	Neutral	—	961	256	0.27
	All	—	1734	507	0.29

.

Following Charlton and Polvani [32], the SSW events are identified when the zonal-mean zonal wind at 10 hPa and 60° N reverses from a westerly to an easterly wind during the period from November to March. However, we exclude the stratospheric final warming events with a final direction reversal of circumpolar zonal-mean zonal wind. In addition, similar to Charlton and Polvani [32], there should be 20 consecutive days of westerly wind between two individual SSW events. Given this, the SSW events in this study represent the major SSW events, according to their definition from the World Meteorological Organization. However, for simplicity, the word major is omitted in the context.

We use two methods to diagnose planetary wave activities. One is a Fourier decomposition of the waves in geopotential height anomalies in the zonal direction. The other is the Eliassen–Palm (EP) flux. Under the quasi-geostrophic approximation, equations of EP flux in spherical coordinates are as follows [33]:

$$F_{\varphi }=\rho a\cos \varphi (-\overline{v’u’})$$

(1)

$$F_z=\rho af\cos \varphi \overline{v’\theta ’}{(\frac{\partial \overline{\theta }}{\partial z})}^{-1}$$

(2)

$$\nabla \cdot \overrightarrow{F}={(a\cos \varphi )}^{-1}\frac{\partial }{\partial \varphi }(F_{\varphi }\cos \varphi )+\frac{\partial F_z}{\partial z}$$

(3)

where “ ¯ ”and “ ′ ” represent the zonal mean and the zonal deviation from the zonal mean, respectively; ρ, a and φ are the air density, the Earth’s radius and the latitude, respectively; θ and f represent the potential temperature and geostrophic parameters, respectively; F_φ is the horizontal component of EP flux, representing the eddy momentum flux; F_z is the vertical component of EP (EP_z) flux, representing the eddy heat transport; and $\nabla \cdot \overrightarrow{F}$ is the divergence of EP flux.

We define the PNA index for each month in each CMIP6 model using the standardized geopotential height anomalies (represented by Z in the Equation (4)) at four points [34]:

PNA = 1/4 [Z (20° N, 160° W) − Z (45° N, 165° W) + Z (55° N, 115° W) − Z (30° N, 85° W)].

(4)

In Section 3.4, the Aleutian low strength and the stratospheric polar vortex strength are diagnosed. The strength of the Aleutian low is defined as 500-hPa geopotential height anomalies averaged over 30°–55° N, 170° E–140° W, and the strength of the stratospheric polar vortex is defined as 10-hPa geopotential height anomalies averaged poleward of 70° N.

For a given variable in each model, both long-term climatology and linear trend are removed to obtain anomalies before composite analyses. The climatology of an atmospheric variable is averaged over all years in all models; thus, there are 1734 and 1960 years for CMIP5 and CMIP6 models, respectively. The two-tailed Student’s t-test is used to test the significance of the composites.

3. Results

3.1. SST Anomalies and Arctic Stratospheric Response Associated with ENSO Events

Figure 1 shows the composite DJF-mean SST anomalies (SSTa) associated with ENSO events with different magnitudes in CMIP6 and CMIP5 models. The spatial pattern of SSTa for the El Niño and La Niña events in both CMIP6 and CMIP5 models are similar to the observations. However, the magnitude of SSTa in CMIP6 is relatively larger than those in CMIP5 (Figure 1a,b,d,e,j,k,m,n) with the discrepancies of 0.71, 0.39, 0.47, and 0.25 K in the Niño3.4 region for strong El Niño, moderate El Niño, strong La Niña, and moderate La Niña, respectively (Table 3). This implies that the CMIP6 models’ capability is improved in simulating different ENSO events, given the fact that the ENSO intensity is underestimated by CMIP5 models [35].

Figure 2 shows the temporal evolution of composite geopotential height anomalies area-averaged over the polar cap (70°–90° N, 0–360°) for different ENSO groups in CMIP6 and CMIP5 models. Clearly, the CMIP6 models can well reproduce the response of the Arctic stratosphere to ENSO; namely, El Niño greatly weakens the stratospheric polar vortex, while La Niña strengthens it (Figure 2a–d). We also see that the stratospheric responses are larger in strong ENSO events (Figure 2a,c) than those in moderate ENSO events (Figure 2b,d), indicating a monotonic increase of ENSO impact with its amplitude. Significant ENSO signals mainly appear from February to April, although the signals can extend to early May in strong El Niño events (Figure 2a), and appear relatively early, in mid-January, in strong La Niña events (Figure 2c). Generally, the CMIP5 models can also reproduce the response of the Arctic stratosphere to ENSO (Figure 2e–h) but with a less significant and much smaller magnitude as compared to CMIP6 models.

Previous studies suggested that only CMIP5 high-top models can better simulate the Arctic stratospheric response to ENSO [21,27]. Following Charlton-Perez et al. [36] and Wilcox and Charlton-Perez [37], we also divide the CMIP6 models into high-top (with model top pressure ≤ 1 hPa) and low-top models (with the model top pressure > 1 hPa) (see Table 2). Figure 3 shows the temporal evolution of composite polar-cap mean geopotential height anomalies for different ENSO groups in CMIP6 high-top and low-top models. In high-top models, the stratospheric responses to strong and moderate El Niño and strong La Niña events resemble those averaged over all the CMIP6 models; however, little signals can be found in moderate La Niña events (Figure 3d). The reason for this remains unknown. Iza et al. [20] found that while strong La Niña has a robust signature in the winter Arctic stratosphere in reanalysis data, the signals from relatively weaker La Niña might be obscured by the impacts from other factors such as the quasi-biennial oscillation and SSW events. Therefore, it might be the relatively high SSW frequency (~0.62 yr^–1,

Table 4. Numbers of ENSO events and SSW events in different ENSO groups in CMIP6 high-top and low-top models.

Models	ENSO Magnitude	Winter Numbers	SSW Numbers	SSW Frequency
High-top	Strong El Niño	101	69	0.68
	Moderate El Niño	188	137	0.73
	Strong La Niña	66	42	0.64
	Moderate La Niña	234	144	0.62
	Neutral	687	457	0.67
	All	1276	849	0.67
Low-top	Strong El Niño	57	28	0.49
	Moderate El Niño	111	42	0.38
	Strong La Niña	42	19	0.45
	Moderate La Niña	122	43	0.35
	Neutral	352	161	0.46
	All	684	293	0.43

, see Section 3.2 for details) that would lead to the undistinguished stratospheric signals in moderate La Niña events. This needs further investigation. In low-top models, the stratospheric polar vortex is also significantly weak in El Niño and strong in La Niña events from February to April. For moderate El Niño and strong La Niña events (Figure 3f,g), the stratospheric signals in early winter (from mid-November to mid-January) are opposite to those in late winter and early spring. Similar opposite signals between early winter and late winter during El Niño are also emphasized in previous studies [2,22]. Thus, such sub-seasonal signals cannot be sufficiently captured by seasonal mean responses, highlighting the importance of winter period selection for investigating the stratospheric responses to ENSO. These results indicate that the capability of low-top models is significantly improved from CMIP5 to CMIP6 in simulating the response of the Northern Hemisphere stratospheric polar vortex to ENSO since the stratospheric dynamical variability is poorly reproduced in CMIP5 low-top models [36].

3.2. ENSO-SSW Relationship

Previous studies reported different relationships between ENSO and SSW [19,23,28,29,30]. Here, we revisit such relationships in CMIP5 and CMIP6 models. Table 3 provides the number of SSW events with their frequency in CMIP5 and CMIP6 models for ENSO and ENSO-neutral conditions. The SSW frequencies are 0.58 and 0.59 yr^–1, respectively, for all years and the ENSO-neutral years in CMIP6 models, which is consistent with the results from Charlton and Polvani [32] with a frequency of about six events per decade in the reanalysis data. The SSW frequency is slightly enhanced under El Niño conditions and reduced under La Niña conditions compared to ENSO-neutral conditions. Overall, the SSW events occur a little more frequently in strong ENSO events than in moderate ENSO events. By contrast, the frequency of SSW in CMIP5 models is quite low, with only ~0.3 events per year. This agrees well with the results of Rao and Garfinkel [38] that most of the CMIP5 models underestimate the SSW frequency. This may be partly due to the weaker planetary wave activities in CMIP5 compared to CMIP6 models (see next subsection). Additionally, it may be because the winter stratospheric polar vortex simulated by CMIP6 models is closer to observations than CMIP5 models [39,40].

Table 4 lists the numbers of ENSO and SSW events in CMIP6 high-top and low-top models. In high-top models, the SSW frequencies under different ENSO conditions and ENSO-neutral winters are around 0.6–0.7 events per year, which are closer to the observations and much higher than those in the low-top models. This indicates a remarkable improvement of the CMIP6 high-top models in simulating the stratospheric polar variability. Nevertheless, CMIP6 low-top models still outperform CMIP5 models in simulating the boreal winter SSW events, with an evident increase in SSW frequency. In addition, compared to those in ENSO-neutral winters, the SSW frequency is slightly enhanced during El Niño and reduced during La Niña in CMIP6 high-top models.

3.3. Tropospheric Teleconnection and Planetary Wave Activities

As noted previously, the main pathway through which ENSO affects the stratospheric polar vortex is its tropospheric teleconnection in the North Pacific, which interferes with the climatological stationary waves and thus modulates the propagation of planetary waves from the troposphere into the extratropical stratosphere. Figure 4 and Figure 5 show the composite 500-hPa geopotential height anomalies averaged from January to March and their wavenumber 1 and 2 components for different ENSO groups in CMIP6 and CMIP5 models, respectively. In CMIP6 models, a canonical pattern can be clearly seen over the PNA region, with an anomalous low over the Northeastern Pacific and eastern United States, respectively, and an anomalous high over Canada during El Niño (Figure 4a,d). In contrast, La Niña induces a nearly opposite PNA pattern (Figure 4g,j). The intensity of ENSO-related signals in the middle troposphere increases with ENSO amplitude. In a total of 921 strong and moderate ENSO events, the correlation coefficient between the winter Niño3.4 SSTa and JFM PNA index reaches up to 0.63. As seen in the middle and right panels in Figure 4, this pattern has a constructive (destructive) interference with the climatological stationary wavenumber 1 and destructive (constructive) interference with wavenumber 2 during El Niño (La Niña), leading to enhanced (suppressed) planetary wave activities entering into the extratropical stratosphere [6]. The composite geopotential height anomalies in the upper troposphere and low stratosphere exhibit similar results (figures not shown). Overall, the CMIP5 models can also reproduce the ENSO teleconnection in the PNA region (Figure 5). However, such teleconnection exhibits a more typical spatial pattern and larger amplitude in CMIP6 models.

Next, we diagnose the EP_z flux at 100 hPa, which roughly represents the upward propagation of planetary wave activities from the troposphere into the extratropical stratosphere. Figure 6 shows the temporal evolution of composite EP_z flux anomalies of planetary waves (wavenumbers 1–3), and the wavenumber 1 and wavenumber 2 components averaged in 55°–75° N for different ENSO groups in CMIP6 and CMIP5 models. In CMIP6 models (Figure 6a–d), the wavenumber 1 (wavenumber 2) of EP_z flux anomalies intensifies, and wavenumber 2 (wavenumber 1) reduces during El Niño (La Niña) mainly from January to March, matching the results in Figure 4. We further calculate the correlation coefficients of the PNA index with the wavenumber 1 and wavenumber 2 EP_z flux during the ENSO events in CMIP6 models, which are 0.37 and –0.32 during January–February–March season, respectively, exceeding the 95% confidence level. In Figure 6a–d, the increase (decrease) of wavenumber 1 overwhelms the decrease (increase) of wavenumber 2 during El Niño (La Niña), resulting in active (dormant) upward propagation of planetary waves and thus a weakened (strengthened) stratospheric polar vortex. Also, the ENSO-related planetary wave activities enhance with ENSO amplitude. In CMIP5 models (Figure 6e–h), the temporal evolution of planetary wave activities resembles that in CMIP6 models. Nevertheless, the CMIP6 models can better characterize the planetary wave activities, at least in strong ENSO events.

3.4. Asymmetry and Nonlinearity

ENSO signals exhibit obvious asymmetry and nonlinearity in both the troposphere and stratosphere. We have shown that the Arctic stratospheric and mid-tropospheric responses to El Niño and La Niña are symmetric in CMIP6 in terms of the sign. Now we turn to the amplitude of the response. Figure 7 shows the ratios of the Aleutian low strength and the stratospheric polar vortex strength averaged in two adjacent months between two different ENSO groups in CMIP6 models. The ratios of winter SSTa in the Niño3.4 region in different ENSO groups are also shown. It is found that the asymmetries in terms of the SSTa amplitude are relatively weak between strong El Niño and strong La Niña (~1.15) and between moderate El Niño and moderate La Niña (~0.94). We note that large SSTa can lead to even stronger responses if the ratio of Aleutian low strength or stratospheric polar vortex strength between two ENSO groups exceeds that of SSTa. Otherwise, large SSTa can only result in relatively weak responses.

It is seen that strong El Niño has a smaller winter response in the middle troposphere and Arctic stratosphere as compared to strong La Niña (purple bars in Figure 7), confirming the results of Rao and Ren [9,10]. In the stratosphere, the average strength of stratospheric polar vortex in strong El Niño is only 50–60% of that in strong La Niña from January to March. However, in spring, the response to strong El Niño is much larger than that of strong La Niña. This is because significant La Niña signals in the Arctic stratosphere can be found as early as mid-January and peak in February and March, while strong El Niño signals peak in March and April (Figure 2a,c). By contrast, for moderate ENSO events (green bars in Figure 7), the responses to El Niño in both the troposphere and stratosphere are larger than those to La Niña: on average, 1.65 times those to La Niña in the Arctic stratosphere from February to April, which is also consistent with that in Rao and Ren [9,10]. However, for all the moderate and strong ENSO events, the responses to El Niño and La Niña are almost equal in magnitude (the ratio is around ~1 for both the DJFM-mean Aleutian low strength and JFMA-mean stratospheric polar vortex strength). This indicates fairly symmetric responses between El Niño and La Niña, which is identical to the results in Weinberger et al. [23]. In addition, although strong El Niño leads to a stronger Aleutian low and stratospheric polar vortex than moderate El Niño (red bars in Figure 7), such an increase is not linear. That is, nearly double SSTa during strong El Niño do not generate double responses in the middle troposphere and Arctic stratosphere, verifying the findings of Weinberger et al. [23]. For La Niña, such nonlinearity seems to be a little bit complex. Less large Aleutian low in strong La Niña is associated with a much larger response in the stratosphere as compared to moderate La Niña (blue bars in Figure 7). The stratospheric polar vortex strength in strong La Niña is around a factor of 2.3 larger than that in moderate La Niña on average from January to April, agreeing well with the findings of Iza et al. [20] that robust polar stratospheric responses can only be found in strong La Niña events.

4. Summary

The amplitude and zonal location of the maximum SSTa are usually different among ENSO events, leading to conspicuous diversity or so-called asymmetry or nonlinearity of ENSO signals. In this study, we reexamine the ENSO’s impact on the Arctic stratosphere based on large samples of ENSO events provided by historical simulations from CMIP6 and CMIP5 models. The main conclusions are as follows.

CMIP6 models can well reproduce the Arctic stratospheric responses to ENSO. Specifically, El Niño leads to a deepened low in the North Pacific and an enhanced high over Canada in the middle troposphere, which constructively interferes with the climatological stationary wave. Although the planetary wavenumber 2 is relatively reduced, the wavenumber 1 is considerably reinforced, resulting in a net effect of increased vertically-propagating planetary waves from the troposphere into the extratropical stratosphere and thus a warm and weakened stratospheric polar vortex in the Northern Hemisphere. The case for La Niña is nearly the opposite, with a cool and strengthened stratospheric polar vortex. Compared to the CMIP5 models, the CMIP6 models’ capability is remarkably improved in simulating ENSO and its Arctic stratospheric response, manifesting as the larger amplitude of SSTa in Niño3.4 region, stronger and more typical PNA pattern, more active planetary waves, and intensified response of the stratospheric polar vortex.

The occurrence of SSW is 0.58 events per year in CMIP6 models, which is quite close to the observations. The SSW frequency is slightly increased during El Niño and decreased during La Niña relative to ENSO-neutral winters in CMIP6 high-top models. However, this is not true for the CMIP5 models and CMIP6 low-top models.

In CMIP6 models, ENSO signals in the Arctic stratosphere mature in late winter and early spring, mainly from February to April, irrespective of the ENSO phase. Overall, the responses of the 10-hPa stratospheric polar vortex to El Niño and La Niña are symmetric, but not for those induced by strong and moderate ENSO events. The asymmetry of the response to strong El Niño and La Niña is characterized by a different temporal evolution in the polar stratosphere. Significant signals associated with strong La Niña appear from early January and peak in February and early March, while strong El Niño-related signals peak in March and April. For moderate ENSO, the stratospheric response to El Niño is much larger than that to equivalent magnitude La Niña.

The Arctic stratospheric responses increase monotonically with ENSO magnitude in CMIP6 models, whereas the increase is not proportional to that of Niño3.4 SSTa. The response to strong El Niño events is weaker than that which should be achieved if the response changes linearly with the amplitude of El Niño. In contrast, strong La Niña leads to a much larger Arctic stratospheric response, around 2.3 times those to moderate La Niña.

The analyses in this study are based on a large number of ENSO samples provided by CMIP6 and CMIP5 models. This overcomes some uncertainties using limited samples in the reanalysis data and meanwhile helps to minimize the contamination of ENSO signals by other factors. For example, the main results in this study remain unchanged if we remove the linear effect of the stratospheric quasi-biennial oscillation on the stratospheric circulation (figures not shown). As CMIP6 models outperform CMIP5 models in simulating the polar stratospheric response to ENSO, similar approaches could also be used to examine the stratospheric response to eastern and central Pacific El Niño, respectively, in both hemispheres. Relevant studies are ongoing and will be reported in the future.