The relative impacts of El Niño Modoki, canonical El Niño, and QBO on tropical ozone changes since the 1980s

Some studies showed that since the 1980s Modoki activity—a different sea surface temperature anomaly pattern from canonical El Niño–Southern Oscillation (ENSO) in the tropics—has been increasing in frequency. In the light of an analysis of the observations and simulations, we found that Modoki, as a new driver of global climate change, can modulate the tropical upwelling that significantly affects mid-lower stratospheric ozone. As a result, it has an important impact on the variations of tropical total column ozone (TCO), alongside quasi-biennial oscillation or canonical ENSO. Our results suggest that, in the context of future global warming, Modoki activity may continue to be a primary driver of tropical TCO changes. Besides, it is possible can serve as a predictor of tropical TCO variations since Modoki events precede tropical ozone changes.


Introduction
Since tropical ozone levels can affect the global distribution of ozone through dynamic transport processes (Hood et al 1997), studying changes in tropical ozone is an important area of research. It is well-known that the quasi-biennial oscillation (QBO) and the canonical El Niño-Southern Oscillation (ENSO) are the main processes controlling the interannual variations of tropical ozone.
ENSO is the most important mode of interannual variability of Pacific sea surface temperature (SST) (e.g., Rasmusson and Carpenter 1982, Philander 1990, Trenberth 1997. It can influence stratospheric circulation through modulating the anomalous propagation and dissipation of stratospheric ultra-long Rossby waves (e.g., Sassi et al 2004, Manzini et al 2006, Garfinkel and Hartmann 2007, Ineson and Scaife 2009. In turn, these circulation changes influence the global distribution of ozone. Randel et al (2009) and Calvo et al (2010) showed that ENSO events are tied to fluctuations in tropical upwelling. Upwelling can adjust the low-ozone tropospheric air entering the lower stratosphere, which influences the level of mid-lower stratospheric ozone. Thus, ENSO events are linked to coherent anomalies in zonal mean ozone in the tropical mid-lower stratosphere (e.g., Brönnimann et al 2006, Fischer et al 2008, Cagnazzo et al 2009, Hood et al 2010.
Equatorial stratospheric winds and temperatures exhibit a downward-propagating oscillating pattern with an average period of just over two years, known as the QBO. Reed (1964) proposed that QBO-induced meridional circulation could drive a QBO signal in total column ozone (TCO). During a descending easterly phase (i.e., winds from the east) there is an induced upwelling at the equator in the lower stratosphere which results in a negative anomaly in column ozone at the equator. During a descending westerly QBO phase the circulation is reversed, producing a positive column ozone anomaly at the equator. Subsequently, based on observations and simulations, the characteristics of the QBO signal in stratospheric ozone, as well as the mechanism by which the QBO affects ozone, have been widely studied (e.g., Randel and Wu 1996, Butchart et al 2003, Tian et al 2006, Lee et al 2010.
In an empirical orthogonal function (EOF) analysis of tropical Pacific ocean SST from 1980 to 2010, the second leading mode of tropical SST variability showed two cold centers in the eastern and western Pacific and a warm center in the central Pacific, referred to as El Niño Modoki (the negative phase of El Niño Modoki, known as La Niña Modoki, shows two warm centers in the eastern and western Pacific and a cold center in the central Pacific), as compared with a cold center in the eastern Pacific and a warm center in the western Pacific in canonical El Niño events (Ashok et al 2007, Zhang et al 2009. A detailed analysis in Ashok et al (2007Ashok et al ( , 2009 showed that El Niño Modoki is independent from canonical El Niño. They also pointed out that El Niño Modoki and canonical El Niño Modoki are different SST modes which lead to different climate effects. The distributions and gradient patterns of SST anomalies during Modoki events, which are associated with abnormal convection and patterns of ultra-long Rossby wave propagation, lead to significant and profound effects on temperature and circulation in the troposphere (Feng and Li 2011, Karori et al 2013 and stratosphere (Hurwitz et al 2011, Graf and Zanchettin 2012, Lin et al 2012, Zubiaurre and Calvo 2012. Since the distributions and gradient patterns of SST anomalies of Modoki events are different from those of canonical ENSO, the response of the atmosphere to the two types of ENSO is also different (Trenberth and Smith 2006, Xie et al 2012, Zubiaurre and Calvo 2012. Modoki activity has been increasing in frequency, especially since the 1980s Yamagata 2009, Yeh et al 2009). The response of ozone to canonical ENSO events has been carefully investigated in previous research (e.g., Gettelman et al 2001, García-Herrera et al 2006, Oman et al 2012, Ziemke et al 2010. However, the relationship between Modoki and ozone, i.e., the effect that Modoki has on ozone interannual variability compared with canonical ENSO and QBO have not yet received sufficient attention. Therefore, the present study focuses on the effects of El Niño Modoki events on ozone variation; in particular, tropical ozone changes.

Data, methods and simulations
All data used in the analyses are described in the supplementary information.
We defined a tropical total column ozone (TCO) index (TTOI) as the averaged TCO anomaly between 30°S and 30°N to represent the variation of tropical TCO. The TCO anomalies were obtained by removing the seasonal cycle of TCO.
The monthly Nino 3 index (5°N-5°S, 150°-90°W), hereafter N3I, and the ENSO Modoki index, hereafter EMI, were used to identify monthly occurrences of canonical El Niño events and El Niño Modoki events, respectively. N3I was defined as the area-mean SSTA over the region (5°S-5°N , 150°-90°W). Following Ashok et al (2007), the EMI was defined as follows: QBO1 and QBO2 indices at 10 hPa and 30 hPa, respectively, are defined to discuss the effects of QBO on ozone based on NCEP Reanalysis 2 (NCEP2). The QBO1 index represents the dynamical transport of ozone caused by the zonal wind anomalies in QBO phase in the upper-middle stratosphere (10 hPa); QBO2 index is a feedback of NOx distribution anomalies on ozone as a result of QBO phase by chemical process in the middle stratosphere (30 hPa) (Wallace et al 1993, Randel et al 1996. The effective number ( ) N eff of degrees of freedom (DOF) (Bretherton et al 1999), which used in this study, can be determined by the following approximation : where N is the sample size, and ρ XX and ρ YY are the autocorrelations of two sampled time series, X and Y, at time lag j, respectively. An expression to calculate the vertical component of Eliassen-Palm (E-P) Flux was given by Andrew et al (1987).
where ρ o is the density of background air, θ is the potential temperature, a is the radius of the Earth, ν is the meridional wind, ϕ is the Earth's latitude, and f is Coriolis parameter. The subscript z denotes derivatives with respect to height z. The overbar denotes deviations from the zonal mean.
The formulae to calculate the vertical component of Brewer-Dobson (BD) circulation in a pressure coordinate Environ. Res. Lett. 9 (2014) 064020 F Xie et al system were given by Edmon et al (1980).
where θ is the potential temperature, a is the radius of the Earth, ω is mean vertical velocity in pressure coordinates. Subscripts p and ϕ denote derivatives with pressure p and latitude ϕ, respectively. The overbar denotes the zonal mean and the accent denotes the deviations from the zonal mean value. Three time-slice (R1-R3) and one free run (R4) simulations, which are forced by observed SST from the United Kingdom Meteorological Office Hadley Centre for Climate Prediction and Research, are integrated by the Whole Atmosphere Community Climate Model version 4 (WACCM4) (Marsh et al (2013)). The SST used in the control experiment (R1) is the monthly mean climatology for the period 1980-2010. The SST used in the sensitivity experiment (R2) is the same as that used in R1, except that the tropical Pacific SST is modified to represent the composite tropical Pacific SST associated with El Niño Modoki events during the period 1980-2010. For instance, the El Niño Modoki SST anomalies in January used for forcing the model are obtained by compositing all the observed SST anomalies during El Niño Modoki events that only occurred in January from 1980 to 2010. The monthly SST forcing is an averaged situation of SST anomalies in each month during El Niño events from 1980 to 2010. The SST used in the sensitivity experiment (R3) is the same as that used in R2, but for canonical El Niño events. WACCM4 has 66 vertical levels extending from the ground to 4.5 × 10 −6 hPa (∼145 km geometric altitude), and the model's vertical resolution is 1.1-1.4 km in the tropical tropopause layer and the lower stratosphere (<30 km). The simulations presented in this paper are performed at a resolution of 1.9°× 2.5°, with interactive tropospheric and stratospheric chemistry . The monthly mean climatologies of surface emissions used in the model were obtained from the A1B Environ. Res. Lett. 9 (2014) 064020 F Xie et al

Tropical TCO EOFs from 1979 to 2011
By performing an EOF analysis of the normalized total ozone mapping spectrometer/solar backscatter ultraviolet (TOMS/ SBUV) tropical TCO from 1979 to 2011, the leading principal component (PC1, figure 1(a)), which accounted for 28.1% of the variance, is found to be strongly correlated with TTOI (linear correlation coefficient: 0.88), implying the leading EOF mode may well represent the temporal variability of tropical TCO. Using merged ozone data, a previous study (Camp et al 2003) showed that the QBO is associated with the first mode of EOF analysis for tropical TCO time series from 1978 to 2002 (accounting for nearly 50% of the variance). However, it is interesting that PC1 is not related to the QBO in this study. The PC1 and EMI (Ashok et al 2007) are correlated with a statistically significant linear correlation coefficient of 0.27 ( figure 1(a)). In fact, the correlation coefficient is the largest when PC1 lagged the EMI by 3-4 months, as discussed further below. The negative loadings pattern of EOF1 ( figure 1(b)) implies that El Niño Modoki can cause negative anomalies in tropical TCO. However, the second EOF and PC (figures 1(c) and (d)), which accounted  Ozone regressions are shown as local percentage variations, with contour intervals of 1.5%/EMI index. Panel (b) is the same as (a), but with ozone regressed onto the N3I. The EMI and N3I lead the ozone time series by three months in the regression. Panels (c) and (d) are the same as (a) and (b), but based on RICH temperature data from 1980 to 2012. Contour intervals are 0.1%/EMI index. The El Niño signals in zonalaveraged ozone/temperature anomalies were derived from linear regression analysis. El Niño variability lagged behind ozone/temperature variations by three months when performing the linear regression analysis. The ozone/temperature anomalies were obtained by removing the seasonal cycle. Anomalies that are significant at the 95% confidence level (two-tailed Student's t-test using the N eff of DOF) are covered with + symbols.
for 24.0% of the variance, is associated with the QBO1. The PC2 and QBO1 index are correlated with a linear correlation coefficient of 0.89 ( figure 1(c)). The third mode of EOF analysis for TCO related to the QBO2 (figures 1(e) and (f)), and the fourth mode is associated with the N3I (figures 1(g) and (h)), which accounted for only ∼3.3% of the variance. Figure 1 illustrates that, since El Niño Modoki become active after the 1980s (Ashok and Yamagata 2009, Yeh et al 2009), its effect has become an important controlling factor for tropical TCO variations in the recent three decades.
It is important to note that, if the time scale of the EOF analysis of TCO variability based on TOMS/SBUV data is changed to 1979-2002, the first mode is also associated with the QBO, which is consistent with Camp et al (2003) (not shown). However, when performing the EOF analysis of TCO variability from 1979 to 2011, we found that the first mode relates to Modoki events. This implies that the difference of the first mode of TCO variability between the two periods may be caused by some recent shift in the atmospheric or oceanic state, i.e., the presence of Modoki events. Figure 2 shows the lead-lag correlation (LLC) between PC1 of tropical TCO variability and EMI. A remarkable feature shown by figure 2 is that El Niño Modoki lead PC1 changes by 3-4 months, implying El Niño Modoki can be considered as a predictor of significant ozone variation.
Note that above results based on NIWA (not shown) data are consistent with those based on TOMS/SBUV (figures 1 and 2).

The mechanism by which Modoki affects stratospheric ozone
Why does El Niño Modoki lead to a decrease in tropical TCO, and why does it make a greater contribution to tropical TCO variability than canonical El Niño (figures 1(a) and (g))?
To answer this question, it is necessary to analyze the vertical distributions of ozone anomalies and various relevant processes during El Niño Modoki events. The influence of the spatial structure of Modoki and canonical ENSO events on SAGEII zonal mean ozone (derived from regression analysis) is illustrated in figures 3(a) and (b). According to these data, El Niño Modoki leads to a significant decrease of ozone in the tropical mid-lower stratosphere. However, these decreases are smaller in canonical El Niño events. It apparent that El Niño Modoki has a more important impact on mid-lower stratospheric ozone than canonical El Niño, and thus the decreases in tropical TCO are likely to be mainly related to the significant decrease of mid-lower stratospheric ozone caused by El Niño Modoki events. It is well-known that ozone is an important radiative gas in the stratosphere. A decrease in stratospheric ozone weakens absorption of ultraviolet radiation, which cools the stratosphere due to radiation processes. Thus, the implication is that the significant decrease of midlower stratospheric ozone caused by El Niño Modoki events is also related to temperature changes in the stratosphere. The influence of the spatial structure of Modoki and canonical El Niño events on radiosonde innovation composite homogenization (RICH) zonal mean temperature (derived from regression analysis) is illustrated in figures 3(c) and (d). The results show that the El Niño Modoki signal can also be found in the stratospheric temperature field, indicating that El Niño Environ. Res. Lett. 9 (2014) 064020 F Xie et al , but for zonal mean stratospheric temperature anomalies based on WACCM4 outputs. Anomalies that are significant at the 95% confidence level (two-tailed Student's t-test using the N eff of DOF) are covered with + symbols.
Modoki leads to a significant decrease in temperature in the tropical mid-lower stratosphere, and has a more important impact on mid-lower stratospheric temperature than canonical El Niño. The above results from regression analysis are supported by the WACCM4 simulations. Simulated ozone and temperature anomalies during the two types of El Niño events (R2-R1 and R3-R1) are consistent with the regressed ozone and temperature anomalies (figures 4 and 3). Zubiaurre and Calvo (2012) indicated that in the tropics the anomalous warming of the tropical troposphere and cooling of the lower stratosphere present during canonical El Niño events weakens during El Niño Modoki episodes, which is different from our result. Their temperature anomalies during two types of El Niño obtained through a composite analysis of the simulated temperature. Our temperature anomalies obtained by two ways: first, the regressions to observation according to two types of El Niño indices; second, the simulations forcing by composite SST anomalies. These may be the main reasons causing the difference between the two studies.
From the EOF analysis of tropical TCO variability from 1979 to 2011 (figure 1), it is found that PC1, related to Modoki events, accounted for ∼28% of the variance. However, PC4, associated with canonical ENSO, accounted for only 3.3% of the variance. The explained variance of PC1 is ∼8 times larger than PC4. According to figures 4(a) and (b), the average percentage change of tropical stratospheric ozone (30°N-30°S; 200-30 hPa) caused by Modoki events (∼8.3%) is also ∼6 times larger than that caused by canonical ENSO events (∼1.4%). Therefore, it is understandable, to a certain extent, why the explained variance of PC1 is ∼8 times larger than PC4.
Although both canonical El Niño and El Niño Modoki are related to SST anomalies over the tropical Pacific, the results in this study suggest that El Niño Modoki has had a more significant influence on stratospheric ozone than canonical El Niño since the 1980s. The SST anomalies can affect convection which further influences transport (Dessler and Sherwood 2004), and it also affects planetary wave activities in the tropics which further influences tropical upwelling (Ortland and Alexander 2014). Thus, the different SST anomalies in two types of El Niño events would lead to different transport processes between troposphere and stratosphere. Figure 5(a) shows the LLC between the vertical component of 150 hPa middle-latitude E-P flux variations and the EMI and the N3I from 1979-2011. E-P flux represents the planetary wave propagation from troposphere to stratosphere. It is found that since the EMI has better correlation with 150 hPa middle-latitude E-P flux variations than the N3I, the correlation coefficient between the EMI and the 70 hPa tropical BD circulation changes is larger than that between the N3I and the BD circulation ( figure 5(b)). It implies that El Niño Modoki activity may stronger modulate the tropical upwelling changes than canonical El Niño by influencing the planetary wave propagation. Stronger upwelling in El Niño Modoki events in the tropics lead to more lowozone tropospheric air entering the lower stratosphere. This may be the reason why El Niño Modoki results in a more evident decrease in stratospheric ozone than canonical El Niño. Figure 5 shows that the E-P flux and BD circulation changes response to the EMI about lagging behind for 3-4 months. It explains why tropical TCO changes lag behind the EMI ∼3-4 months (figure 2).

Conclusion and discussion
The results presented in this paper show that El Niño Modoki has had a significant effect on tropical ozone since the 1980s, alongside that of the QBO, and canonical El Niño. Based on EOF analysis, we found that the leading mode of TCO variability, accounting for ∼28% of the variance, is associated with El Niño Modoki events (figures 1(a) and (b)), and the second and third modes are related to the QBO, accounting for 24% and 9.2% of the variance, respectively (figures 1(c) and (d)). The fourth mode is caused by canonical El Niño (figures 1(g) and (h)), accounting for only 3% of the variance. Modoki activity can affects the tropical TCO is likely to cause by directly transports ozone-poor air in the tropics from the troposphere to the stratosphere, which leads to significant Environ. Res. Lett. 9 (2014) 064020 F Xie et al ozone decreases in the lower-middle stratosphere (figures 3 and 4), since El Niño Modoki can significantly enhance the BD circulation (figure 5). As a result, it tends to decrease tropical TCO. As is known, in the context of future global warming, there may be more El Niño Modoki events Yamagata 2009, Yeh et al 2009). The results presented in this paper suggest that El Niño Modoki will remain a primary driver of changes in tropical ozone. Since El Niño Modoki activity leads tropical ozone changes by 3-4 months, it could serve as a predictor for tropical ozone variations.
An 11-year period is found in the PC1 of EOF analysis of tropical TCO variations ( figure 1(a)), which is deemed related to the solar cycle (Shindell et al 1999). Solar activity controls the level of ultraviolet (UV) radiation entering the Earth's atmosphere, and thus can affect ozone production and destruction in the middle and upper stratosphere (Gray et al 2010). Variation in solar activity tends to take place over an 11-year cycle, and hence the detected 11-year periodicity of ozone variability is basically considered the result of these corresponding solar variations (e.g., Austin et al 2008, Dhomse et al 2011, Gray et al 2009. However, figure 6(a) shows that the PC1 of tropical TCO variability is ahead of the solar variations from 1979 to 2011. This feature is confirmed in figure 6(b), which shows the LLC between PC1 and solar activity, indicating that PC1 variations lead the solar variations by ∼1 year. The result is supported by longer data (55 years) from WACCM4's free run simulation R4 (figures 6(c) and (d)). This is a very interesting phenomenon. If the solar activity modulates the 11-year cycle of ozone why it changes lag behind the tropical TCO variations. However, this phenomenon needs further certification and may deserve further analysis. For another, as El Niño Modoki has an 11-year cycle since the 1980s (Ashok et al 2007) and the variations of EMI lead the tropical TCO changes, whether the 11-year cycle of topical TCO also relatives to El Niño Modoki activity which also deserve further investigation by simulations. Figure 6. (a) The low-pass filtered series PC1 × (−1) of EOF analysis of tropical TCO variability (black line) and solar activity (blue line) after filtering out the signals whose periods are shorter than 84 months; the TCO from TOMS/SBUV data, the solar activity from F 107 data. (b) Lead-lag correlation (LLC) between the low-pass filtered PC1 × (−1) and solar activity. Negative lags refer to the PC1 leading solar activity. The dotted lines denote the 95% confidence level. (c) And (d) same as (a) and (b), but for the TCO based on 55 years time series from a free run simulation R4 by WACCM4 from 1951 to 2005. The vertical dashed lines in (b) and (d) represent the lead-lag month where are the largest LLC coefficients.