A connection from Arctic stratospheric ozone to El Niño-Southern oscillation

Antarctic stratospheric ozone depletion is thought to influence the Southern Hemisphere tropospheric climate. Recently, Arctic stratospheric ozone (ASO) variations have been found to affect the middle-high latitude tropospheric climate in the Northern Hemisphere. This paper demonstrates that the impact of ASO can extend to the tropics, with the ASO variations leading El Niño-Southern Oscillation (ENSO) events by about 20 months. Using observations, analysis, and simulations, the connection between ASO and ENSO is established by combining the high-latitude stratosphere to troposphere pathway with the extratropical to tropical climate teleconnection. This shows that the ASO radiative anomalies influence the North Pacific Oscillation (NPO), and the anomalous NPO and induced Victoria Mode anomalies link to the North Pacific circulation that then influences ENSO. Our results imply that incorporating realistic and time-varying ASO into climate system models may help to improve ENSO predictions.


Introduction
Stratospheric ozone is not only vital to protecting life on Earth, as it absorbs harmful solar ultraviolet radiation (Lubin andJensen 2002, Chipperfield et al 2015), but also essential to the control of the stratospheric temperature via atmospheric radiative heating. The latter influences the circulation and chemical composition of the stratosphere (Haigh 1994, Ramaswamy et al 1996, Forster and Shine 1997, and can also affect tropospheric weather and climate (Baldwin and Dunkerton 2001, Graf and Walter 2005, Ineson and Scaife 2009, Cagnazzo and Manzini 2009, Reichler et al 2012, Karpechko et al 2014, Kidston et al 2015, Zhang et al 2016. Antarctic stratospheric ozone has decreased over the past 60 years in association with anthropogenic emissions of ozone depleting substances (Solomon 1990, Ravishankara et al 1994). The Antarctic ozone hole is thought to influence the Southern Hemisphere tropospheric climate (Son et al 2008, Feldstein 2011, Kang et al 2011, Thompson et al 2011, Gerber and Son 2014, Waugh et al 2015. However, the situation in the Arctic is markedly different. The multi-decadal Arctic ozone loss has been much smaller than that of Antarctic ozone (World Meteorological Organization (WMO) 2011) because the winter/spring Arctic polar cap is warmer than the Antarctic polar cap. Thus, the Northern Hemisphere surface climate response to Arctic ozone loss is less evident (e.g., Thompson and Solomon 2005). In contrast, the year-to-year variability of the Arctic stratospheric temperature is much larger than that of the Antarctic Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. stratospheric temperature (Randel 1988), owing to frequent 'sudden warming' events (Charlton and Polvani 2007) in the Arctic stratosphere caused by the abundance of planetary scale waves that propagate into the Arctic stratosphere and perturb the circulation there. The amplitude of the interannual variability in Arctic ozone is comparable with, or even much larger than, that in Antarctic ozone (Manney et al 2011). Cheung et al (2014) and Karpechko et al (2014) used the UK Met Office operational weather forecasting system and the atmospheric circulation model ECHAM5, respectively, to explore the possible surface impacts associated with extreme Northern Hemisphere ozone anomalies. They concluded that stratospheric ozone changes alone did not appear to have a significant effect on surface conditions. More recently, Arctic stratospheric ozone (ASO) variations have been found to cause Northern Hemisphere mid-high latitude tropospheric circulation and sea level pressure (SLP) anomalies. Smith and Polvani (2014) and Calvo et al (2015) performed numerical experiments that showed a statistically significant Northern Hemisphere mid-high latitude surface response to high and low values of synthetic ASO.
However, it is not known whether the impact of ASO extends to the tropics, for example, to influence the El Niño-Southern Oscillation (ENSO), a major climatic mode of tropical variability (e.g., Jin 1996, Timmermann et al 1999, Ashok and Yamagata 2009, Latif and Keenlyside 2009, Yeh et al 2009 that has great global climatic and societal impacts (e.g., Orlove et al 2000, McPhaden et al 2006, Deser et al 2010. It is well known that ENSO can influence tropical ozone through anomalous convection (Camp et al 2003, Xie et al 2014 and the high-latitude stratospheric ozone (Bȑonnimann et al 2004, Eyring et al 2006, Zhang et al 2015 through anomalous propagation and dissipation of ultra-long Rossby waves at mid-latitudes (Gettelman et al 2001, Calvo et al 2004, Manzini et al 2006, Garfinkel and Hartmann 2008, Randel et al 2009, Hurwitz et al 2011, Xie et al 2012. Figure 1 shows the lead-lag correlation coefficients at three-monthly intervals between the ENSO index and zonally averaged ozone, for ENSO variations leading ozone by 3 months to lagging ozone by 24 months. We see that, as expected, ENSO has significant correlations with tropical and high latitude stratospheric ozone when ENSO leads ozone by three months. However, figure 1 also shows an unexpected result: ENSO is significantly correlated with ASO when ENSO lags ozone by 20 months. This implies that changes in ASO may cause a later change in ENSO. The aim of this study is to explain why this happens. Sea surface temperature (SST) and SLP data were obtained from the UK Met Office Hadley Centre for Climate Prediction and Research SST (HadSST) and SLP (HadSLP) field datasets, respectively. Geopotential height and zonal wind were obtained from the National Centers for Environmental Prediction-Department of Energy (NCEP-DOE) dataset (version 2), and temperature data were taken from the Radiosonde Innovation Composite Homogenization (RICH) dataset (Haimberger et al 2008).

Data, methods, and simulations
We calculated the statistical significance of the correlation between two auto-correlated time series using the two-tailed Student's t-test and the effective number (N eff ) of degrees of freedom (Bretherton et al 1999). For this study, N eff was determined by the following approximation (Li et al , 2013: where N is the sample size, and ρ XX and ρ YY are the autocorrelations of two sampled time series, X and Y, respectively, at time lag j. We used the National Center for Atmospheric Research's Community Earth System Model (CESM), version 1.0.6, which is a fully coupled global climate model that incorporates an interactive atmosphere (CAM/WACCM) component, ocean (POP2), land (CLM4), and sea ice (CICE). For the atmospheric component, we used the Whole Atmosphere Community Climate Model (WACCM), version 4 (Marsh et al 2013). WACCM4 is a climate model that has detailed middle-atmosphere chemistry and a finite volume dynamical core, and it extends from the surface to approximately 140 km. For our study, we disabled the interactive chemistry. WACCM4 has 66 vertical levels, with a vertical resolution of about 1 km   , and specified ozone forcing derived from the CMIP5 ensemble mean ozone output. E 1 is a historical simulation covering the period 1955-2005. All forcings and design of E 2 are as E 1 , except that the specified ozone forcing in the region 30-90°N, at 300-30 hPa was replaced by MERRA ozone data for the period 1979-2005. Ozone outside of the region 30-90°N, 300-30 hPa is the same as E 1 . E 3 and E 4 are two ensemble simulations with E 2 using slightly different initial conditions. An overview of all coupled experiments is given in table 1. All of the forcing data used in this study are available from the CESM model input data repository.

A connection between the ASO and ENSO
To probe the relationship between the changes in ASO and ENSO over the past three decades, their lead-lag correlation is shown in figure 3(a). The correlation for ENSO leading the ASO by about three months is statistically significant. This agrees with previous studies showing that ENSO can affect Northern Hemisphere stratospheric ozone after 2-3 months, as the Rossby waves excited by ENSO reach the Northern Hemisphere mid-high latitudes and then move into the stratosphere (Manzini et al 2006, Fischer et al 2008. A significant negative lagged correlation (r=−0.35, lag ca. 20 months) is also observed ( figure 3(a)). This lagged correlation is further confirmed by figure 3(b), which shows time series of the ASO and ENSO indices, but with the time coordinate for ASO shifted forward by 20 months, representative of its observed lag. A spectrum analysis was performed on the changes in the ASO and ENSO indices (figures 3(c), (d)). There are similar lowfrequency spectra in the 1-6 year band in the ASO and ENSO time series. To further establish the relationship between ENSO events and the ASO anomalies, figure 4(a) shows the scatter plot of winter ENSO events against the 20 month leading ASO anomalies. It illustrates that the strong El Niño/La Niña events correspond well to strong negative/positive ASO anomalies. Figure 4(b) pairs the winter NINO3.4 index and 20 month leading ASO index, giving a significant correlation coefficient of r=−0.57. The strong El Niño and strong La Niña events (e.g., 97/98, 15/16) since 1986 follow the strongly decreased and increased ASO anomalies that occurred about 20 months earlier.
To determine the connection between the ASO and the ENSO, we first examine the high-latitude stratosphere to troposphere pathway from the ASO to surface pressure variability. We find that the ASO is significantly correlated with the Arctic Oscillation and North Pacific Oscillation (NPO). A decrease in ASO radiatively cools the Arctic lower stratosphere ( figure 5(a), green contour lines), which enhances the meridional temperature gradient in the lower stratosphere and thus strengthens stratospheric circulation ( figure 5(a), black contour lines) via the thermal wind relationship. The stratospheric circulation anomalies in turn influence tropospheric circulation by the downward control principle (Haynes et al 1991) and tropospheric eddy momentum feedback (Kidston et al 2015). Geopotential height anomalies in the stratosphere propagate down to the surface ( figure 5(a), color shading), resulting in an NPO-like signal over the North Pacific ( figure 5(b)). All forcings and design are as E 1 , except that the specified ozone forcing in the region 30°-90°N, at 300-30 hPa b was replaced by MERRA ozone data for the period 1979-2005. Ozone outside of the region 30°-90°N, 300-30 hPa is the same as E 1 . These are three ensemble simulations using slightly different initial conditions c . . This acts as an extratropical to tropical climate teleconnection via oceanatmosphere dynamic interaction. The NPO-like signal is effectively linked to the VM anomalies (figure 6(a)). Thus, the extratropical to tropical climate teleconnection provides a link from the NPO/VM to ENSO through the following processes: North Pacific westerly anomalies generate SST and associated westerly anomalies through air-sea interaction in the subtropical central eastern North Pacific, central equatorial Pacific, and the western North Pacific. The latter alters the zonal SST gradient anomalies across the western central tropical Pacific ( figure 6(b)), leading to westerly anomalies over the western central Pacific, which drive an El-Niño type event (Jin 1997a, 1997b; figures 6(c), (d)). The dynamic pathway and teleconnection from the ASO to the NPO, and from the NPO/VM to ENSO, together establish the ASO's connection to ENSO. Previous studies show that the NPO/VM lead ENSO by more than a year and a half (Ding et al 2015). Consequently, the ASO leads ENSO by about 20 months.
Note that a 35 month low-pass filter was applied to the ozone and SST data in figure 6. As described above, the lag panels in figure 6 reflect the seasonal footprinting mechanism. However, previous studies have demonstrated that only the low-frequency variations in VM are effective in modulating the development of ENSO through the seasonal footprinting mechanism (Vimont et al 2001, Ding et al 2015. On the other hand, a spectrum analysis was performed on the changes in the ASO and ENSO indices (figures 3(c), (d)). There are similar low-frequency spectra in the 1-6 year band in the ASO and ENSO time series. However, the low-frequency spectra in the 1-3 year band in both ASO and ENSO time series may be related to QBO. The 35 month low pass filter is used to filter out QBO effect on the lead-lag correlation between ASO and ENSO. This is why a 35 month low-pass filter is applied to the SST variations in figure 6. It would highlight the low-frequency variations, which are related to ozone changes, in the SST but remove noise-like highfrequency variations.

Simulated SST responses to ASO variations
North Pacific SST variability can influence the stratospheric polar vortex (Jadin et al 2010, Hurwitz      table 1), when the ASO variations lead SST by (a) 0 months, (b) 3 monthsK and (h) 21 months. Ozone is specified from the CMIP5 ensemble mean ozone output. Simulated SST is as outputs from E 1 . Only regions above the 95% confidence level are colored. Both specified ASO and simulated SST variations were detrended, and a 35 month low-pass filter was applied to ozone and SST.
the North and tropical Pacific, we find that the VMlike pattern appears over the North Pacific following the ASO anomaly (figure 7(a)) and is enhanced in the tropical Pacific after about 6-9 months (figure 7(b)). Finally, an El Niño-like event emerges one year later (figures 7(c) and (d)). This developmental sequence is similar to that observed (figure 6). Note that the ozone forcing is specified in the simulation; therefore, its variation should not be related to North Pacific SST anomalies.
Although the recent CMIP5 models can capture the spatial pattern of ENSO, there are large biases between simulated and observed ENSO variations (see Guilyardi et al 2009); for example, the ENSO index for the period 1979-2005 in E 1 which is a historical experiment simulated by CESM is not significantly correlated with the observed ENSO (figure 8(a); r=0.11). Note that the specified ASO forcing in E 1 is derived from the CMIP5 ensemble mean ozone output, which is not in good agreement with the observed ozone variability ( figure 8(b); r=0.14). According to our results, ASO affects ENSO, so would improving the specified ASO forcing in simulations improve the simulated ENSO variations? The ASO from MERRA (1979MERRA ( -2015 was used as the specified ASO forcing in experiments E 2 -E 4 , because there are no missing values for ASO variations in MERRA and the data are in agreement with the observed ASO variations ( figure 2). Interestingly, the correlation coefficient between the simulated ensemble-mean ENSO and observed ENSO variability is statistically significant (figure 8(c); r=0.42), when MERRA ozone is used as the specified ASO forcing in the simulations. This result not only supports the impact of ASO on ENSO, but also implies that incorporating realistic ASO forcing in the model can improve the simulation of ENSO variability.

Discussion and conclusions
Previous studies have noted that the radiative effect of Antarctic stratospheric ozone (AASO) depletion (Solomon 1990, Ravishankara et al 1994  has a significant impact on the Southern Hemisphere tropospheric climate (Son et al 2008, Feldstein 2011, Kang et al 2011, Thompson et al 2011. In this study, we find that the AASO variations have no significant impact on ENSO (Not shown). This contrast between the influence of AASO and ASO anomalies on ENSO may be due to the lack of a robust teleconnection pathway from the Southern Hemisphere Pacific to the equatorial Pacific. It should be noted that the relatively warm water in the tropical North Pacific and the permanent northern Intertropical Convergence Zone serve as necessary anchors to allow the seasonal footprinting mechanism to operate in a geographically and seasonally favored manner. However, in the Southern Hemisphere Pacific, none of the equivalent anchors are of significance. This may explain why the influence of the AASO on ENSO is not significant. Figure 9 summarizes our findings regarding the influence of the ASO on ENSO using a schematic illustration. A negative ASO anomaly cools the Arctic stratosphere, strengthening the stratospheric circulation. The downward propagation of a negative stratospheric geopotential height anomaly, which reaches the surface as a negative NPO anomaly in about one month, initiates a positive VM phase over the North Pacific. The evolution of the positive VM anomaly reaches the equatorial Pacific and strengthens an El Niño event when other conditions for its occurrence are ripe. The high-latitude stratosphere to troposphere pathway and the extratropical to tropical climate teleconnection take more than a year and a half. A positive ASO anomaly would have the opposite effect, and has the potential to strengthen a La Niña event via the same pathway. This impact of ASO on ENSO makes it a potentially useful predictor of ENSO events. It is well known that ENSO influences stratospheric ozone. This implies that there may be a two-way interaction Figure 9. The two processes that combine to drive the ASO to El Niño connection: the high-latitude stratosphere to troposphere pathway (light blue wavy arrow) and the extratropical to tropical climate teleconnection (yellow curved arrow). ASO − : Arctic stratospheric ozone decrease; NPO − : Negative North Pacific Oscillation phase; VM + : Positive Victoria Mode phase.  1991 1998 1997 1999 2009 2007 2015 2010 or dynamic feedback between the ASO and ENSO. In even broader terms, understanding this kind of connection and potential feedback between the stratospheric tracer gases (such as ozone) and the climate system deserves more attention. Furthermore, the observed ENSO variability can be partly explained by simulated ENSO variability in CESM when an observed ASO forcing is used. This result reinforces the need for climate models to include fully coupled stratospheric dynamical-radiative-chemical processes if they are to more accurately simulate and predict future ENSO variations.