Tropical circulation and precipitation response to ozone depletion and recovery

Among the few well established changes in atmospheric circulation in recent decades are those caused by stratospheric ozone depletion. They include a strengthening and poleward contraction of the westerly atmospheric circulation over the Southern extratropics, i.e. a strengthening Southern Annular Mode (SAM), in austral spring and summer. Associated effects on extratropical temperature and precipitation and more recently subtropical precipitation have been documented and are understood in a zonal mean framework. We present zonally asymmetric effects of ozone depletion that reach into the tropics and affect atmospheric circulation and precipitation, including the South Pacific Convergence Zone (SPCZ), the most important rainband of the Southern Hemisphere. Using observation-based analyses and model simulations we show that over the 1961–1996 period, ozone depletion led to increased precipitation at the northern flank of the SPCZ and to decreased precipitation to the south. The effects originate from a flow pattern over the southwestern Pacific that extends equatorward and alters the propagation of synoptic waves and thus the position of the SPCZ. Model simulations suggest that anticipated stratospheric ozone recovery over the next decades will reverse these effects.


Introduction
The Intertropical Convergence Zone (ITCZ) consists of a zonally oriented part as well as of three diagonally oriented branches: the Southern Hemisphere Convergence Zones. Of these, the South Pacific Convergence Zone (SPCZ) is the most pronounced (Vincent 1994). It is an important component of the global hydrological cycle and the ocean freshwater budget (contributing to a pronounced sea-surface salinity front, see Linsley et al 2006) and is responsible for a substantial fraction of Southern Hemisphere precipitation variability (Cai et al 2012). The SPCZ is characterized by low-level convergence to the west of the South Pacific High. It is formed by transient synoptic waves originating south of that propagate eastward along the extratropical jet and are refracted equatorward. Upon reaching the SPCZ region, wave activity converges due to the decrease in zonal wind speed (Widlansky et al 2011, Matthews 2012, van der Wiel et al 2015. While intraseasonal variability of the SPCZ is related to the Madden-Julian Oscillation, its interannual-to-decadal variability depends on El Niño/Southern Oscillation (ENSO) and the Pacific Decadal Oscillation (Kidwell et al 2016). Large decadal variations in the SPCZ position since 1600 have been suggested based on ocean salinity proxies (Linsley et al 2006). The SPCZ is expected to change due to 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. greenhouse warming (Cai et al 2012, Widlansky et al 2013. Effects of other forcing factors such as stratospheric ozone depletion have received little attention. Filling this gap is the aim of this paper. Stratospheric ozone depletion is known to affect atmospheric circulation and precipitation in the extratropics. Thompson and Solomon (2002) and Gillett and Thompson (2003) showed that ozone depletion over Antarctica in spring induces a strengthening and poleward contraction of the westerly wind over the Southern extratropics, which can be understood as a strengthening of the Southern Annular Mode (SAM). These changes lead to an increase in precipitation at subpolar latitudes, a decrease at southern midlatitudes and an increase over (Thompson et al 2011). Warming in South Africa was found as a consequence of ozone depletion (Manatsa et al 2013). Kang et al (2011) found an increase in precipitation across the subtropics. In particular, Gonzalez et al (2014), exploring six climate models, attributed the 1960-1999 wetting trend over Southeastern South America to zonally asymmetric responses to stratospheric ozone depletion. A recent study (Manatsa et al 2016) hinted at an effect on East African precipitation, but little is still known about ozone depletion effects in the tropics.
Most previous studies on ozone depletion effects on atmospheric circulation and climate adopted a zonal mean perspective. In this paper we will take a zonally non-symmetric perspective to address the effect of ozone depletion on the SPCZ. For this purpose we use a set of chemistry-climate model simulations from a range of models, which allow us to directly address the effect in targeted simulations. In addition, we analyse several observation-based data sets to compare empirical evidence with our model simulation results.

Model simulations
Our main data set to study the effect of ozone depletion are simulations by the chemistry-climate model SOCOL (Schraner et al 2008). SOCOL is a combination of the Middle Atmosphere version of the European Centre/HAMburg Model 4 (MA-ECHAM4) spectral AGCM and the chemistry-transport model MEZON. It is run with T30 horizontal truncation and 39 vertical levels on a hybrid sigma-pressure coordinate system. The model spans the atmosphere from the surface to 0.01 hPa. An ensemble of nine simulations was performed in an 'all forcings' set-up (Fischer et al 2008, termed ALL) for the entire 20th century. The arguably most important boundary condition is sea-surface temperatures (SSTs) and sea ice, which are taken from HadISST1.1 (Rayner et al 2003). Ozone depleting substances were specified according to the CCMVal protocol (see Schraner et al 2008 for more detail). Furthermore, an additional ensemble of three simulations, branched off the ALL simulations, was performed with ozone depleting substances held fixed at their 1951 levels (termed noODS, see table 1).
In addition to these simulations, we used all available model simulations from the Chemistry-Climate Model Validation Activity for SPARC (CCMVal-2, Morgenstern et al 2010) and the Chemistry-Climate Model Initiative (CCMI, www. met.reading.ac.uk/ccmi/, Morgenstern et al 2017) that provide precipitation fields. In total, 17 simulations from four models were available (see table 1), covering the period 1960 to 2050. Note that the SOCOL simulations described above (Fischer et al 2008) were neither part of CCMVal-2 nor of CCMI, but SOCOL participated in CCMVal-2 with two additional simulations with a slightly different set-up (Morgenstern et al 2010). Note also that some models are coupled while others use prescribed SSTs and sea ice. Uncoupled simulations have the caveat that the possible role of SSTs in the climate system response is not captured, while they have the advantage of being more directly comparable with reanalyses and observations. We use the REF-B2 (in CCMVal-2, corresponding to IPCC's A1B scenarios in terms of greenhouse gases and WMO's A1 scenario for halogens, see SPARC 2010) or REF-C2 (in CCMI, corresponding to IPCC's RCP6.0 scenario and WMO's A1 scenario for halogens, see www.pa.op.dlr.de/ CCMI/) scenario simulations as equivalent of ALL and the SEN-B2b or SEN-C2fODS, respectively, for the noODS scenario.
In addition to analyses of individual models, we also calculated a multi-model mean. It is based on the ALL minus noODS difference of each model, weighed with (n −0.5 þ m −0.5 ) −1 , where n and m are the number of members of ALL and noODS, to account for the dependence of the standard error on the ensemble size. We defined an ensemble comprising all CCMVal-2 and CCMI simulations as well as a multi-model ensemble that also includes the 9þ3 SOCOL simulations. For precipitation, the ensemble means are expressed relative to a similarly weighted climatology , i.e. as percentage anomalies.

Observation-based data sets
The South and Central Pacific region is not well observed, particularly when it comes to precipitation. We therefore used several observation based data sets. In order to assess the time when ozone depletion started, i.e. around 1960, we used some sparse station as well as reanalysis data for precipitation and atmospheric circulation. Precipitation station data for the region 55°S to 5°S and 170°E to 230°E were taken from the Global Historical Climatology Network (GHCN-Monthly, Peterson and Vose 1997) and were required to have at least 40 years of data in the 1958-2015 period and at least 30 years of data in the 1961-1996 period, which will later be used for most of the analyses. However, the data quality is largely unknown. For the period after 1979, we also used Global Precipitation Climatology Project (GPCP) combined satellite-based precipitation data Version 2.3 (Adler et al 2003).
For the atmospheric analyses we used the reanalyses JRA-55 (Kobayashi et al 2015) as well as NCEP/NCAR (Kalnay et al 1996) and 20CRv2c (Compo et al 2011). Results were similar, but because of large uncertainties in the tropical circulation in the latter two data sets (e.g. Pomposi et al 2015, Compo et al 2011, the main focus is on JRA-55.

Diagnostics and analysis procedure
The variables analysed in this paper are precipitation, 450 hPa geopotential height (a level common to both SOCOL and JRA-55; the variable is termed Z450 in the following), as well as 950 hPa wind. To address wave propagation across the South Pacific, we analysed du/dx (zonal stretching deformation, Widlansky et al 2011) at 200 hPa. Negative values indicate where the group velocity of propagating waves decreases and thus wave activity converges, representing a 'graveyard' of propagating disturbances (van der Wiel et al 2015).
The analysis sequence was as follows. After analysing trends in model simulations and observation-based data, we addressed the ozone-induced change in the simulations as the linear trend in the difference ALL-noODS. The ozone-induced change in the observation-based data was extracted using a multiple regression model (see below). After finding a consistent response in the SPCZ region, we supplemented this 'forward' approach with a 'backward' analysis. We defined a precipitation index for the identified SPCZ signature and regressed the global fields onto that index (both in simulations and observation-based data) to confirm the high-latitude origin and to elucidate the possible mechanism. This led to the identification of a likely dynamical origin of the response in southern midlatitude geopotential height. We finally defined an index for that feature and again regressed the global fields onto this index (in simulations and observation-based data), confirming its relation foreward to the SPCZ and backward to high-latitude geopotential height changes.
We focus our analysis on October to December, which is the period immediately following the ozone loss season and the start of the summer rainfall season in the SPCZ. Analyses for December-February or October-February (shown in the supplement) gave similar results.
Linear trends were calculated using a least-squares regression. Significance was tested using the Mann-Kendall test with p < 0.05. Following Thompson and Solomon (2002) all trends in this paper are multiplied by the period length (i.e. they are expressed as changes over the period of trend calculation).

Total column ozone gradient
As a measure for the ozone forcing, which is understood to result from an intensified latitudinal heating gradient in the lower stratosphere, we used the difference in total column ozone between the highand mid-latitudes with a one-month lead time (i.e. September-November). As no total column ozone observations are available poleward of 80°S in September, we chose the regions 60°S-80°S and 30°S-50°S. The difference between the two zonal bands is termed dTOZ. Prior to 1979, the data were taken from the HISTOZ data set (Brönnimann et al 2013), which is a nine member ensemble assimilation of historical ground-based and satellite (BUV) data into the ALL simulations. For each year we chose the assimilated member that was closest to all observations south of 30°S in terms of Eucledian distance. From 1979 onward we used total column ozone data from JRA-55, which agrees very well with SBUV data but is spatio-temporally complete.
The time series of dTOZ from observation-based data, ALL and noODS (figure 1), reveal that ozone depletion began already around 1960 (see also Schneider et al 2015). dTOZ declined until the mid 1990s, stronger in ALL than in observation-based data. While the observed dTOZ remains at a low level for another 10 years, the simulated recovery starts after around 1994. Based on this, we chose the 35 year period of 1961À96 as our analysis period, which captures the entire ozone depletion period and avoids the very large 1997/98 El Niño event (analyses over different time periods gave very similar results). In addition, we also performed analyses over the 1979À96 period, which enables using the satellitebased precipitation data and facilitates the comparison with other studies. It is important to check possible multicollinearity of dTOZ. The correlation between dTOZ and CO 2 concentrations is À0.729, while it is close zero with all other variables. The Variance Inflation Factor is 2.33, which indicates that multicollinearity should not have a large effect. We tested various model set-ups, using different combinations of variables or detrending all variables while omitting greenhouse gases. Results (figure S1 available at stacks.iop.org/ERL/12/064011/ mmedia for Z450) are almost identical for all regression models.

Results and discussion
3.1. Model evaluation and precipitation trends For addressing ozone depletion effects on tropical precipitation, it is important that the model realistically simulates the precipitation climatology as well as the known effects on the high-latitude atmospheric circulation. In fact, 1961À90 precipitation climatologies for the October-December season (figure 2) from JRA-55 and ALL agree very well with each other. The same is true for climatologies of 200 hPa du/dx. Thus, the model simulates the SPCZ and the main processes responsible for its climatology well.
A previous study  also found that the ALL simulations realistically reproduce shifts in the ITCZ in the second half of the 20th century. Furthermore, we reproduced the results by Thompson and Solomon (2002), who analysed the change in atmospheric circulation and temperature over the Southern extratropics over the 1969À99 period for December-February. The change in ALL fits well with the observations shown in that paper (figure S2, from Brönnimann 2015), although magnitudes are smaller. The trend (expressed as change over the considered time period) completely disappears in the noODS simulations. The model thus produces a consistent response of the SAM to stratospheric ozone depletion.
Over the 1961À96 period, October-December precipitation in JRA-55 increased by up to 50% (of its 1961À90 average) in the northeastern flank of the SPCZ (figure 2). This number is confirmed with station data from Héréhérétué, Atuona and Rikitea (French Polynesia; dots in figure 2). Precipitation decreased in the southwestern part and further to the south (global maps in figure S3). The general pattern of the trend around the SPCZ as well as the trend at subpolar latitudes is well reproduced in the ALL simulations. Although there are differences in the subtropical eastern Pacific and around New Zealand and in the southeastern Pacific, the model simulations appear suitable for analysing precipitation trends in the SPCZ region.

Ozone depletion-induced trends
The trend in the atmospheric circulation differences between ALL and noODS over the 1961-96 period Environ. Res. Lett. 12 (2017) 064011 illustrates the effect of ozone depletion. The patterns for Z450 and 950 hPa wind ( figure 3, top) show a strengthening of the westerlies near 60°S and decreasing (increasing) Z450 poleward (equatorward). In du/dx, a northwest-to-southeast tilted pattern appears over the South Pacific (red ellipses in figure 3), with significantly negative values at the SPCZ elongation.
The circulation response to ozone depletion in JRA-55, as obtained by the regression approach   figure 3, bottom), shows close similarities with the SOCOL results. Changes in GPH are similar over the southern mid-and high-latitudes, i.e. a decrease over Antarctica and an increase in three areas southeast of the continents. Good agreement is also found for 950 hPa wind. All main features are highly significant. A very similar pattern as in the model results is also found for du/dx, although the position slightly differs. Model and the observation-based approaches thus both suggest that ozone depletion altered the atmospheric flow over the southern mid-and highlatitudes and into the tropical Pacific region.
The corresponding trend patterns in precipitation (figure 4) are more variable. In the SOCOL simulations, zonal signatures appear at southern mid and high latitudes (global maps in figure S4). Over the Pacific, we find an elongation of the SPCZ into the South Pacific and a decrease in precipitation to the south; both features are statistically significant (p<0.05). An analysis of the individual ALL members (expressed relative to the noODS ensemble mean) shows that this response is robust. This is confirmed by studying the individual CCMVAl-2/CCMI models (figure S4), although the SPCZ itself differs between models. The CCMVal-2/CCMI ensemble (figure 4) again shows an elongation of the SPCZ and a significant drying to the south. Finally, the multimodel ensemble (encompassing SOCOL and the CCMVal-2/CCMI ensemble) clearly shows that the SPCZ is altered.
The ozone-induced precipitation change in observation-based data (figures 4, S5) agrees with the model results despite variability. The regression results indicate a wetting at subpolar latitudes, a drying south of the SPCZ, and a precipitation increase in or just north of the tip of the SPCZ, although the latter is not significant. The precipitation station observations only partly show the signature found in JRA-55, but they also do not contradict the result. The NCEP/NCAR and 20CRv2c show similar results as JRA-55 (figure S5). The same analysis (but only post-1979) was also performed with GPCP data (figure 4) and confirmed the main patterns, although the significance level is not reached for that short period.
Taking the circulation and the precipitation response together, both in the models and in observation-based analysis, we find that the response of the Pacific is not zonally oriented, but shows a clear northwest-to-southeast tilt. In the models, the pattern is slightly shifted (larger response near the tip of the SPCZ) relative to the regression approach (larger response in the northern flank of the tip).

Relevance for past and future precipitation trends
According to our results, ozone depletion induced precipitation changes in the SPCZ region on the order of À25 to 40% over the 1961-96 period. How important is this for precipitation trends? In JRA-55, SOCOL and CCMVal-2/CCMI, the total (figures 2, S6) and ozone-induced precipitation changes (figure 4) have similar patterns. In the SPCZ region, ozone depletion explains about half of the total trend. The precipitation increase in the northeastern part and the tip of the SPCZ and the decrease in the southwestern part 1961À96 can therefore, to a considerable extent, be explained by ozone depletion. . Also shown is the simulated change due to future ozone recovery (2015À50) in the CCMVal-2/CCMI ensemble and the regression result for GPCP satellite-based precipitation . All trends are expressed as % changes over the trend period. The black line shows the 4 mm day À1 isoline from the corresponding climatology.
Orange lines indicate p-values of 0.05. Dots in the middle left panel indicate observations. Green dots indicate the points whose difference defines DR SPCZ .
Environ. Res. Lett. 12 (2017) 064011 During the past 20 years with stable but low ozone levels (1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015), precipitation trends in the SPCZ region were different than in the earlier period (although this period is very short; figure S3). With ozone recovery now underway, we expect the effects of ozone depletion to reverse. In particular, we expect a decreased precipitation in the northern tip of the SPCZ and increased precipitation to the south, which might have societal impacts in the island countries affected by the SPCZ variability (e.g. Cai et al 2012). Over the 2015À50 period in the CCMVal-2/CCMI ensemble (figures 4, S6), the ozone-induced change is almost a mirror image of that over the 1961À96 period, confirming the robustness of our results. The changes in the SPCZ are statistically significant.

Mechanism of the teleconnection
In order to analyse the mechanism of this response, we defined an index DR SPCZ as the difference in precipitation between the northeastern part of the SPCZ and the area south of the SPCZ (19°S, 135°W minus 37.5°S, 150°W). This index best expresses the common signature in all analyses (e.g. it is positive for each of the 9 members in SOCOL, statistically significant for 3 members). We then regressed atmospheric fields onto DR SPCZ both in ALL and in JRA-55, i.e. we analysed in a 'backward' manner, which flow features are likely to cause changes in the tip of the SPCZ (figure 5).
The patterns are similar in ALL and in JRA-55. They are also similar to those attributed to ozone depletion, but the positive node in Z450 over the South Pacific as well as the wave pattern in du/dx over the South Pacific are more pronounced in this case. Thus the forward analysis (studying the consequences of ozone depletion) and the backward analyses (exploring the causes for changes in the SPCZ) yield very similar results, both in models and in observation-based analyses.
The figure implies that the pattern originates in the southern South Pacifc region, from where a wave pattern emerges. Specifically, a trough-ridge pattern near New Zealand appears in the 6400 gpm isoline of Z450 (figure 3). We therefore defined a longitudinal index of Z450 along 40°S (DZ SP ¼ Z450 [40°S, 110°E þ 40°S, 190°E À 40°S, 150°E þ 40°S, 230°E] in September-November) and regressed atmospheric fields onto that index. Results (figure 6) reproduce the flow pattern over Antarctica as well as in the SPCZ region, both in JRA-55 and in ALL (but again with a slight shift in the SPCZ region between the two). Those ALL members (relative to noODS) that produce a strong trend in DZ SP also produce strongest changes in DR SPCZ .
For comparison, we also regressed October-December averaged fields from JRA-55 onto a September-November averaged SAM index of Marshall (2003), (figure S7). Results show pronounced, zonally symmetric changes in wind and Z450 (though with a clear node southeast of New Zealand). We also find an imprint in du/dx and a slight precipitation increase in a narrow zone in the tip of the SPCZ. However, compared to the pattern for DZ SP (figure 6), the SAM imprint is much more zonally symmetric and confined to high-and mid-latitudes.

Discussion
In the extratropics and subtropics, our results agree well with previous studies (e.g. Thompson and Solomon 2002, Gillett and Thompson 2003, Thompson et al 2011, Kang et al 2011, Purich and Son 2012, Manatsa et al 2013, Schneider et al 2015. This mechanism is understood in a zonal mean framework: A positive SAM induces a poleward momentum flux, which in turn induces a meridional circulation cell that causes the drying near 45°S and the wetting near 60°S (Kang et al 2011, Hendon et al 2014. Kang et al (2011) found an adjacent meridional cell that leads, in a zonal mean sense, to uplift and increased precipitation in the subtropics. This is confirmed in our study. However, our signature over the SPCZ region differs from the latter study, which analysed the 1979-2000 period in the CMAM model. This requires further analyses as CMAM simulations are also in our analysis (3 ALL, 2 noODS) and agree with the SOCOL results in that respect, even when analysed only for the 1979-2000 period ( figure S4).
The signature in the SPCZ cannot be explained in a zonally averaged framework. The two main areas of interest are the Indian Ocean south of Australia and the western Pacific. The former is the origin of synoptic waves which then propagate along the jet across Australia and are refracted equatorward. The latter is the region where wave activity converges in the SPCZ region. The positive Z450 signatures east of the southern continents suggest that ozone depletion weakened, and even reversed, the troughs behind the Andes and New Zealand. Interaction of the weakened (southward shifted) westerlies at these latitudes with orography remains to be studied. The change in DZ SP can be understood as a change towards a trough (ridge) west (east) of New Zealand (pink and black lines in figure 6 indicate the 6400 gpm isoline for DZ SP ¼ 100 gpm and DZ SP ¼ À100 gpm, respectively). This configuration might have changed the wave propagation across the South Pacific, as evidenced in du/dx. Synoptic-scale studies are necessary to further elaborate this link.
Are these changes part of the SAM? Our analysis shows the classical SAM features, which are described in the literature (Gillett et al 2006, Villalba et al 2012. The response in du/dx and precipitation is similar to the ozone-induced pattern, but much weaker and thus can only partly explain it. Recent work has also emphasized zonally asymmetric aspects of the SAM and its relation to tropical circulation (Ding et al 2012, Fogt et al 2012, Clem and Renwick 2015. Our results further demonstrate the importance of zonally asymmetric flow patterns over the southwestern Pacific. The same might hold for the South Atlantic and South Indian Ocean, which were not the focus of our study.

Conclusions
The effects of ozone depletion on atmospheric circulation include a zonally asymmetric pattern over the South Pacific that originates from a ridge east of New Zealand and propagates into the tropics. It alters the shape and position of the South Pacific Conver- gence Zone (SPCZ). Using observation-based analyses and model simulations, we show that, over the 1961À96 period, ozone depletion led to increased precipitation at the northern flank of the SPCZ and decreased precipitation to the south. This trend is expected to reverse over the next decades. Our study demonstrates that ozone depletion was an important driver of climatic changes in the tropical South Pacific in the past, and so will ozone recovery be in the future. The study also presents a new teleconnection mechanism between the stratospheric polar vortex over Antarctica and tropical circulation. While the presented case is an 'anthropogenic teleconnection' , the mechanism might also be relevant for other forcing mechanisms known to affect the polar vortex such as solar or volcanic forcing.