Observational evidence of EPP–NOx interaction with chlorine curbing Antarctic ozone loss

Abstract. We investigate the impact of the so-called energetic particle precipitation (EPP) indirect effect on lower stratospheric ozone, ClO and ClONO2 in the Antarctic springtime. We use observations from Microwave Limb Sounder (MLS) and Ozone Monitoring Instrument (OMI) on Aura, Atmospheric Chemistry Experiment – Fourier Transform Spectrometer (ACE-FTS) on SciSat, and Michelson Interferometer for Passive Atmospheric Sound (MIPAS) on Envisat, covering the overall period of 2005–2017. Using the Ap index to proxy EPP, we find consistent ozone increases with elevated EPP during years with easterly phase of the quasi biennial oscillation (QBO) in both OMI and MLS observations. While these increases are opposite to what has been previously reported at higher altitudes, the pattern in the MLS O3 follows the typical descent patterns of EPP–NOx. The ozone enhancements are also present in the OMI total O3 column observations. Analogous to the descent patterns found in O3, we also found consistent decreases in springtime MLS ClO following winters of elevated EPP. To verify if this is due to a previously proposed mechanism of conversion of ClO to the reservoir species ClONO2 in reaction with NO2, we used ClONO2 observations from ACE-FTS and MIPAS. As ClO and NO2 are both catalysts in ozone destruction, the conversion into ClONO2 would result in ozone increase. We find a positive correlation between EPP and ClONO2 in the upper stratosphere in the early spring, and the lower stratosphere in late spring, providing the first observational evidence supporting the previously proposed mechanism relating to EPP–NOx modulating Clx driven ozone loss. Our findings suggest that EPP has played an important role in modulating ozone depletion in the last 15 years. As chlorine loading in the polar stratosphere continues to decrease in the future, this buffering mechanism will become less effective and catalytic ozone destruction by EPP–NOx will likely become a major contributor to Antarctic ozone loss.



MLS
We use ozone and ClO profiles (v4.2) from the Microwave Limb Sounder (MLS), on the Aura satellite (Schwartz et al., 2015;Santee et al., 2015). The data has been sorted according to Livesey et al. (2017), i.e removing data that do not meet the 115 recommended quality standards. The O 3 profiles have been validated by Froidevaux et al. (2008), with further comparison to ground-based and other satellite measurements by Hubert et al. (2016). Here, we use stratospheric O 3 observations (15 km to 50 km) with vertical resolution around 3 km, and uncertainty of no more than 4%.
MLS ClO is valid throughout the stratosphere although the lower-most altitudes (15-18 km) suffer from a negative bias.
The bias, which has been uniform throughout the MLS period, is least significant in the polar region and is also systematic: 120 Each latitude is affected the same way. We mitigate the effect of the bias by looking at anomalies as any systematic bias will not affect the overall gradient of the trend. Since anomalies are differences from a mean, any shift is cancelled in subtraction.
The vertical resolution of stratospheric ClO is around 3 km and the error on individual profiles is around ±0.1 ppbv (Livesey et al., 2017). We do not use ClO from dusk until dawn (i.e. nighttime) due to rapid conversion of ClO to the Cl 2 O 2 dimer at nighttime (Brasseur and Solomon, 2005). Excluding these measurements avoids the change in partitioning between day and 125 night. We sort for day by only using profiles with solar zenith angle < 90 • . MLS ClO profiles have been validated by Santee et al. (2008).

OMI
We analyse ozone total column data from the Dutch-Finnish built Ozone Monitoring Instrument (OMI), also on Aura (Bhartia, 2012  den stratospheric warming that occurred in the SH that spring, disrupting the polar stratosphere therefore any NO x descent. We also exclude November 2003 due to the extremely large SPEs, known as the Halloween event, that occurred throughout late 150 October and early November of that year. These events caused large amounts of particle precipitation resulting in in situ NO x increases in the Antarctic stratosphere (López-Puertas et al., 2005). These increases in November would likely mask any EPP effects from the previous winter. The 2004 and 2005 springs are not included due to the aforementioned instrumental anomaly.

EPP Proxy
Analogous to Gordon et al. (2020), we use the geomagnetic activity index A p as a proxy for the overall winter EPP levels. We take the mean A p index from May to August of each individual year (consistent with previous studies of e.g. Siskind et al. (2000); Seppälä et al. (2007)) and denote this 4-month mean A p asÂ p . The averageÂ p for the study period was 8.3 and thê A p values for each individual year are given in Table 1.

QBO
To account for the influence of the QBO in our analysis (see Gordon et al., 2020), we bin the years according to the phase of the QBO in May. To determine the phase of the QBO, we use the equatorial zonal mean zonal wind at the 25 hPa level (see Baldwin and Dunkerton, 1998, for explanation of use of this level in the SH). Years where the zonal mean zonal wind is easterly are designated easterly QBO (eQBO), while westerly winds are designated westerly QBO (wQBO). The QBO phase 165 for each year of the study is listed in Table 1.

Methods: Anomalies and Correlation
We analyse correlation betweenÂ p and various trace gases in the atmosphere. For this purpose, we use the Spearman rank correlation coefficient ρ, which correlates two non-normally distributed datasets (von Storch and Zwiers, 1999). For significance testing purposes, the correlation is characterised as significant if the p-value is less than 0.05, that is, the correlation is 170 significant at 95% or higher. Correlation studies can be misleading in their results as they view data through a purely statistical lens and do not account for underlying physics. Here, significance of a correlation is tested if we have a reason to speculate on a connection based on known physical or chemical properties or analysis of observational data. Thus, we first check for evidence in anomalies of observational data. As discussed in the Introduction, work by Gordon et al. (2020) has shown evidence that EPP (as proxied by 175Â p ) and QBO affect trace gases in the stratosphere. Here, we will examine the composite anomalies for different combinations of QBO phase andÂ p level for each trace gas analysed. Years withÂ p > 8.3 are designated as highÂ p (h-Â p ) and those witĥ A p < 8.3 are designated as lowÂ p (l-Â p ). 8.3 is chosen as it is the meanÂ p for the study period. These are indicated in Table 1. In the time period under investigation there has been a reduction in equivalent effective stratospheric chlorine (EESC).
This reduction in chlorine and the following gradual recovery of stratospheric ozone has been mitigated in the analysis by 180 de-trending the observations for all correlation calculations. Here, detrending was performed by calculating the gradient of the yearly trend with a linear least squares fit, then subtracting this from the data. This was not applied to the results presenting composite anomalies, which are shown here as an indication of the overall variability in the volume mixing ratios.
We note that other factors can also pay a role in Antarctic stratospheric ozone levels, most notably solar spectral irradiance (SSI) varying with the 11-year solar cycle, and the El Niño-Southern Oscillation (ENSO). Due to the limited time series of 185 observations, it is not possible to robustly control for all. However, we note that the effect of SSI has limited influence on springtime Antarctic ozone variability and the effects are mainly limited to above 10 hPa level (Matthes et al., 2017). Some studies have suggested that ENSO can both influence stratospheric ozone variability (Lin and Qian, 2019), and potentially be influenced by Antarctic ozone variability (Manatsa and Mukwada, 2017). But, as with solar irradiance, the ENSO influence on Antarctic ozone variability appears to be limited to the upper stratosphere, above the 10 hPa level (Lin and Qian, 2019).  Table 1 Years with eQBO (b and c) display a positive anomaly (∼+0.1 ppmv or <10 % reduction from the mean) in the middle stratosphere in October, while wQBO years (d and e) show the opposite. This suggest the anomaly is likely related to the QBO phase and could be linked to the effect noted by Garcia and Solomon (1987) and Lait et al. (1989): more ozone is present in the 205 Southern polar stratosphere in years with eQBO. In the lower stratosphere in November, positive (negative) anomaly occurs in high (low)Â p years. This indicates that these changes are linked to EPP: highÂ p results in ozone increases in November.
In December, in the middle stratosphere (∼20 hPa) highÂ p appears to results in negative ozone anomaly (∼ −0.1 ppmv or < 10 % reduction from the mean).
The above analysis indicates increases in ozone associated with highÂ p , and thus high EPP, while also finding ozone 210 decreases associated with the westerly phase of the QBO. We now look to see if the ozone increases linked toÂ p are correlated withÂ p levels, and how this is modulated by the QBO phase. This is presented in Figure 2 for a) all years, b) eQBO years, and c) wQBO years. As in Figure 1, ozone is cos(latitude) weighted zonal mean average over 60 • S to 82 • S. Note that for all correlation analyses presented here, the data has been linearly detrended to avoid misattribution from linear increases or decreases from reduced EESC since 2005. There is significant anti-correlation (ρ ∼ −0.4 to −0.6) in the upper stratosphere 215 around 2 hPa in panels a) and b). This suggests that increases inÂ p indeed result in ozone loss in this area, particularly during eQBO. These ozone reductions are consistent with O 3 loss due to the EPP-NO x descending in the polar vortex. However, in panel b), for eQBO conditions, the negative correlation pattern, which descends in time, is accompanied by a strong positive correlation (ρ > 0.6) below ∼10 hPa in November. This indicates that EPP in eQBO years also contributes to ozone increases.
At this time both panels a) and b) show positive correlation in the middle and lower stratosphere, though this is only statistically 220 significant during eQBO years. These results seem to suggest that increasedÂ p results in ozone enhancement in November, Contours show the correlation coefficient with 0.2 interval (black contour for zero) and stippling indicates statistical significance (p < 0.05).
10 https://doi.org/10.5194/acp-2020-847 Preprint. Discussion started: 29 September 2020 c Author(s) 2020. CC BY 4.0 License. and that eQBO strengthens this relationship. There is little consistent correlation present in panel c) -there is no clear relation between polar springtime ozone profile variability andÂ p in wQBO years.

OMI column observations
We now repeat the analysis for daily OMI total ozone column, instead of profile measurements. This is to verify whether the 225 changes in ozone associated with EPP and the QBO are detectable in the ozone total column. While we lose the information contained in vertical profiles, we gain higher horizontal resolution. Note that here the OMI data with 0.25 • gridding has been averaged over 1 • latitude bins.
The composite 3 day running mean O 3 column from 2005 to 2017 is presented in Figure 3a). The figure shows zonal mean ozone for each latitude poleward of 50 • S in 1 • bins as spring progresses from early August to the end of the December. Note

245
We now examine the correlation between ozone column (detrended) andÂ p level. This is shown in Figure 4, with the panels from top to bottom presenting: a) all years, b) eQBO, and c) wQBO. Figure 4a) displays the correlation for all years of the study. Overall, the correlation |ρ| < 0.6 everywhere, with little statistical significance, when all years are taken into account and no QBO based binning is done. In panel b), for eQBO years, correlation is positive poleward of 60 • S for almost all of spring.
Areas of significant positive correlation (ρ ≥ 0.6) occur throughout August to October, and early November shows consistent 250 significant positive correlation. This agrees with Figure 3: elevatedÂ p results in ozone increases at high Southern latitudes and this is more prevalent in eQBO years. At lower latitudes, between 50 • S and 60 • S there are patches of significant negative correlation. For wQBO years, shown in panel c), the correlation is highly variable, with |ρ| < 0.4, and not significant. Any influence of EPP on the ozone column is generally weaker during wQBO years. This is consistent with Gordon et al. (2020) who reported significant correlation between stratospheric NO 2 column and EPP (as proxied by A p ) during eQBO years.

MLS ClO observations
First, we investigate the composite mean ClO, and ClO anomaly, from MLS observations. This is done for years with different combinations ofÂ p and QBO phase as before and is presented in Figure 6. This supports the above hypothesis that in years with highÂ p , and therefore more EPP-NO x , we should find reduced ClO, as enhanced NO 2 drives ClO to its ClONO 2 reservoir. The downward propagating signal closely resembles the typical descent does not have the same significant anti-correlation descending in the stratosphere, but does show a weak negative correlation following approximately the same descent pattern. We note there is also a significant anti-correlation in the upper stratosphere in November to December, also present in a) and b). This may be related to the EPP-NO y that remains in the upper stratosphere (see Figure 11 of Funke et al., 2014a) while the bulk descends to lower stratosphere.  Panel a) appears to support the hypothesis that ClO decreases are due to reactions forming ClONO 2 . This is further supported by panel b) which also shows that eQBO amplifies the signal. Panel c) shows little consistent statistically significant correlation at this time.
Due to the limited coverage in the spring, it is difficult to draw conclusive statements from ACE-FTS observations alone.

345
Thus we also analyse MIPAS ClONO 2 observations. Figure 10a) (Figure 2), although below ∼20 km, the anomaly is negative. Similarly we find descending negative anomaly in lowÂ p years (up to −0.06 ppbv). These results support the hypothesis that the O 3 increases in highÂ p years result from enhanced NO 2 driving ClO to its ClONO 2 reservoir. The altitude correlation betweenÂ p and MIPAS ClONO 2 is shown in Figure 11. We again see a descending feature similar to those in Figures 2 and 7. As this feature shows positive (often significant) correlation (ρ > 0.6) it is likely that this again is 360 due to descending EPP-NO x . Note also that as the ClONO 2 increases appear to coincide with ClO decreases, it is unlikely that this correlation is due to the decrease in EESC over this time period as that would result in each correlation being the same sign. This figure shows that more ClONO 2 forms in highÂ p years, and in the same area as ClO decreases (Figure 7), implying that the ClO depletion found earlier is due to ClONO 2 formation.

365
We have presented observational evidence that Antarctic springtime stratospheric ozone increases are associated with higher than average EPP during the preceding winter. Ozone increases due to the so called EPP indirect effect had been previously suggested (Funke et al., 2014a), but, to our knowledge, this is the first time this has been shown in observations. Following the results of Gordon et al. (2020), we propose that this is due to EPP-NO x which remains the lower stratosphere at least until November, having originally been transported from the mesosphere within the polar vortex. Jackman et al. (2000) and Funke 370 et al. (2014a) further proposed that should this NO x reach the lower stratosphere (as shown by Gordon et al., 2020), it would react with ClO to form ClONO 2 , preventing some of the NO x and Cl x driven catalytic ozone destruction. We examined polar ClO and ClONO 2 during the Antarctic spring and found decreases in ClO with consistent increases in ClONO 2 associated with above average EPP. Thus, this provides direct observational evidence supporting the hypothesis of Jackman et al. (2000); Funke et al. (2014a).

375
Throughout the analysis (where possible) we have controlled for the phase of the QBO. Gordon et al. (2020) suggested that the QBO affects NO x in the stratosphere via its influence on both transport of trace gases from the equatorial region, and on wave forcing in the polar region (i.e. the Holton-Tan effect). Here, we have again seen the importance of the QBO: correlations of ozone withÂ p are higher (≥ 0.6 in OMI total ozone) and with more occurrences of statistical significance in eQBO years. This is in agreement with the higher correlation found in eQBO years between NO 2 andÂ p by Gordon et al. (2020). Our 380 results further underline the appreciable effect of the QBO on the lower polar springtime stratosphere, and that the QBO phase should be accounted for in long-term studies of this region.
Our results have shown that the EPP indirect effect has indeed affected ozone over the period 2005-2017, likely due to the interference of EPP-NO x in Cl x catalysed ozone destruction. This period has also been marked by the continuing formation of the ozone hole every spring, although following the Montreal Protocol, the size of the ozone hole is generally decreasing 385 with time (Solomon et al., 2016). The mechanism suggested in this paper (NO 2 buffering ClO) requires chlorine activation in the spring, but as chlorine loading in the polar stratosphere continues to decrease with the ban in CFC emissions, EPP-NO 2 will no longer hinder ozone depletion, likely instead becoming a major contributor. As ozone itself plays a vital role in both atmospheric chemistry and dynamics, this reinforces the importance of accounting for EPP in predicting the future of the polar middle atmosphere.