Volume 125, Issue 16 e2020JD032952
Research Article
Free Access

Impact of Interannual Ozone Variations on the Downward Coupling of the 2002 Southern Hemisphere Stratospheric Warming

H. H. Hendon

Corresponding Author

H. H. Hendon

Bureau of Meteorology, Melbourne, Victoria, Australia

Correspondence to:

H. H. Hendon,

[email protected]

Search for more papers by this author
E.-P. Lim

E.-P. Lim

Bureau of Meteorology, Melbourne, Victoria, Australia

Search for more papers by this author
S. Abhik

S. Abhik

Bureau of Meteorology, Melbourne, Victoria, Australia

Search for more papers by this author
First published: 11 August 2020
Citations: 15


The Southern Hemisphere experienced its first recorded major sudden stratospheric warming during September 2002, which subsequently resulted in strong low polarity of the Southern Annular Mode (low SAM) and extreme daily mean maximum temperatures and low rainfall over eastern Australia during October 2002. The warming and weakening of the polar vortex were accompanied by anomalously high values of polar stratospheric ozone, which possibly could have constructively sustained the weakened vortex and subsequent development of low SAM. We explore the impact of this ozone variation by conducting an idealized forecast experiment using the Australian Bureau of Meteorology's operational subseasonal to seasonal prediction system (Australian Community Climate and Earth System Simulator-Seasonal forecast system version 1, ACCESS-S1), whose atmospheric model well resolves the stratosphere. The ACCESS-S1 control forecasts are generated with prescribed climatological monthly mean ozone, whereas the observed monthly mean ozone during 2002 is prescribed during the forecast for the experiment. While the control forecasts initialized on 1 August 2002 demonstrate good skill in predicting the weakening of the polar vortex and the resultant occurrence of low SAM during October, the extremity of the SAM anomaly and associated extreme high temperatures and low rainfall over eastern Australia were significantly underpredicted. Prescribing the observed ozone results in more realistic weakening of the stratospheric vortex and stronger development of low SAM and extreme warm conditions in eastern Australia during October 2002. These results suggest that polar stratospheric ozone variations are a potential source of long lead climate variability, which can be tapped with future ACCESS-S development.

Key Points

  • Extreme heat developed over Australia in October 2002 following the sudden stratospheric warming in September 2002
  • Forecasts initialized on 1 August captured the early breakdown of the polar vortex but underestimated the development of surface extremes during October 2002
  • Prescribing the observed ozone variation during the forecasts strengthened the surface response, making it more realistic

1 Introduction

In late September 2002, the Southern Hemisphere (SH) polar stratosphere experienced the first recorded major sudden stratospheric warming (SSW), whereby the zonal mean zonal wind along 60°S at 10 hPa abruptly reversed from westerly to easterly (e.g., Baldwin, Hirroka, et al., 2003; Newman & Nash, 2005; Thompson et al., 2005). Although the duration of the actual easterly winds was short (~1 week), this warming was noteworthy not only for its initial intensity but also for the long duration of sustained easterly wind anomalies and their downward penetration to the surface especially during October 2002. This SSW is now understood to have occurred during a longer timescale (seasonal) anomalous weakening of the polar vortex that can be traced back to an earlier than normal onset of the vortex in austral winter in the upper stratosphere and a subsequently earlier than normal break down of the vortex in late spring (e.g., Lim et al., 2018; Seviour et al., 2014). This longer timescale, monthly evolution of the polar vortex and its downward coupling to the surface, rather than the SSW itself, is the focus of the present study.

Some aspects of the monthly evolution of the warming and weakening of the vortex during 2002 are summarized in Figures 1a1c, which show height-time sections of monthly mean anomalies of Antarctic polar cap ozone, temperature, and geopotential height and subpolar zonal wind from August 2002 to February 2003. These data are from European Center for Medium Range Weather Forecasts (ECMWF) Reanalysis Interim (ERA-I) (Dee et al., 2011) and are described more fully in section 2. Anomalous warm conditions in the upper stratosphere are already apparent in August (Figures 1a and 1b), which then strengthen markedly and descend to the lower stratosphere during September–December, recalling that the actual SSW occurred in late September. This warming is associated with higher than normal geopotential height (Figure 1b), weaker than normal subpolar zonal wind (i.e., easterly wind anomalies) (Figure 1c), and increased stratospheric ozone (Figure 1a) during September–November. The higher than normal heights and weaker than normal zonal wind peaked in the lower troposphere in October 2002 but persisted in the lower stratosphere through at least February 2003.

Details are in the caption following the image
Time-height sections of monthly mean anomalies of (a) temperature (shading, K) and ozone concentration (contour interval 1 × 10−7 kg kg−1, negative values dashed, and zero contour omitted) and (b) geopotential height (log10(m)) averaged over the polar cap (60–90°S), and (c) zonal wind (ms−1) averaged 55–65°S. (d) The monthly mean anomalies of the SAM indices from CPC and BAS. Time runs from August 2002 to February 2003. Data in (a)–(c) are from ERA-I reanalyses (Dee et al., 2011). Anomalies are formed relative to the climatology of 1990–2012 (not including 2002). Stippling in (a)–(c) indicates the 2002 anomalies that fall in the one-sided 5% tails of the climatological distribution (i.e., magnitude greater than 1.72 standard deviation (σ), given 22 samples not including 2002).

The easterly anomalies and higher than normal heights extending to the surface in Figure 1 are indicative of a shift toward a low-polarity phase of the Southern Annular Mode (low SAM), which strongly peaked in October 2002 (Figure 1d). The October 2002 SAM value was the most negative since 1979 based on the National Oceanic and Atmospheric Administration Climate Prediction Center's Antarctic Oscillation Index (CPC AAO; available from https://www.cpc.ncep.noaa.gov/products/precip/CWlink/daily_ao_index/aao/monthly.aao.index.b79.current.ascii.table;) and the second most negative after 1988 based on the index from the British Antarctic Survey (BAS SAM; http://www.nerc-bas.ac.uk/icd/gjma/sam.html; Marshall, 2003).

The promotion of low SAM following anomalous warming and weakening of the polar stratosphere is well established (e.g., Byrne & Shepherd, 2018; Lim et al., 2018; Seviour et al., 2014; Thompson et al., 2005) and is of interest because of the SAM's impacts on global surface climate. In its low-polarity phase, SAM during austral spring and summer causes rainfall increases in far southern portions of New Zealand and South America (e.g., Gillett et al., 2006; Lim et al., 2016), higher than normal surface temperatures over much of subtropical Australia, and lower than normal rainfall in much of central and eastern Australia (e.g., Hendon et al., 2007; Lim et al., 2019; Lim & Hendon, 2015). Widespread positive temperature and negative rainfall anomalies covered much of Australia during October 2002 (Figure 2). These data are from the Australian Water Availability Project analyses (AWAP; Jones et al., 2009) and are described more fully in section 2. The warm and dry anomalies were particularly strong in eastern Australia where SAM has its strongest impact (Hendon et al., 2007; Lim & Hendon, 2015). The October 2002 Australian mean maximum temperature anomaly was the seventh highest on record during 1910–2019 (http://www.bom.gov.au/climate/change/index.shtml#tabs=Tracker&tracker=timeseries), and large areas of Australia were in the top 20% warmest Octobers for the 1990–2012 climatological period considered in this study (indicated by hatching in Figure 2a). Australian mean rainfall for October 2002 was the lowest on record for the period 1900–2019 (http://www.bom.gov.au/climate/change/#tabs=Tracker&tracker=timeseries); and almost the entire country was in the driest quintile (lowest 20%) based on the same climatological period 1990–2012. Above normal rainfall in western Tasmania and in the far south west corner of the continent is also consistent with the impact of low SAM during austral spring (Hendon et al., 2007).

Details are in the caption following the image
The October 2002 anomalies of (a) Tmax (K) and (b) rainfall (mm day−1) from AWAP analyses (Jones et al., 2009). Hatching in (a) and (b) indicates where the October anomaly is in the top and bottom quintile, respectively, based on data for the period 1990–2012 (without 2002).

The long timescale of the Antarctic polar stratospheric warming and weakening and its downward coupling to the surface potentially provides a source of long lead predictability of SH surface climate, especially for extreme events such as occurred during 2002 (e.g., Baldwin, Stephenson, et al., 2003; Byrne & Shepherd, 2018; Lim et al., 2018). Predictability of downward coupling from the SH stratospheric polar vortex has previously been demonstrated to a lead time of 1–2 months using a forecast model that well resolves the stratosphere (e.g., Seviour et al., 2014). Long lead predictability resulting from downward coupling from the polar stratosphere is of interest because it is in addition to seasonal predictability stemming from boundary forcing such as El Niño and the Indian Ocean dipole, which are normally viewed as the major source of seasonal climate predictability.

The present study is motivated by the possible role that ozone variations played for promoting the extremity and duration, hence predictability of the 2002 vortex weakening and warming and subsequent development of record low SAM and extreme climate variations in Australia during October 2002. The Antarctic polar cap ozone anomalies that developed as a result of the stratospheric warming (Figure 1a) would presumably become radiatively important from late September onward, when the Antarctic polar cap is again in sunlight, and could act to further promote a weakened vortex by providing additional warming (e.g., Son et al., 2013). Studies of the impact of the stratospheric ozone trend (the “ozone hole”) support this possibility. The downward trend in springtime ozone during the last approximately three decades of the twentieth century has been shown to drive an upward trend in the summertime SAM (e.g., Arblaster & Meehl, 2006; Orr et al., 2012; Polvani et al., 2011). In these modeling studies, imposing the observed trend in ozone (net negative anomaly of about 2 ppm or about 3.5 × 10−6 kg kg−1 in the middle polar stratosphere that peaks in October) resulted in a decrease of polar stratospheric temperatures of about 10 K and a subsequent swing to positive SAM at the surface of magnitude about 1 standard deviation with up to a 3-month delay (e.g., Gillett & Thompson, 2003; Polvani et al., 2011).

Interannual variations of polar cap ozone during September and October, as opposed to the long-term trend in ozone, have also been statistically linked to interannual variations in the SAM during October (e.g., Son et al., 2013), but a causal link has yet to be established. Much of the observed interannual variations of the polar stratospheric ozone can be accounted for by variations in stratospheric circulation (e.g., Seviour et al., 2014). However, the question we address here is whether the interannual ozone anomalies observed during the 2002 stratospheric warming, which were of similar magnitude to the net trend during the last part of the twentieth century but of opposite sign, were of sufficient magnitude and phasing to have provided a significant positive feedback onto duration and magnitude of the stratospheric warming and its subsequent promotion of extreme negative SAM at the surface in October 2002.

We address this issue using the Australian Community Climate and Earth System Simulator-Seasonal forecast system version 1, ACCESS-S1, which is the Bureau of Meteorology's current operational coupled model subseasonal to seasonal prediction system (Hudson et al., 2017). The atmospheric model component has a well-resolved stratosphere, so it can depict the dynamics of the downward coupling to the surface (e.g., Roff et al., 2011; Seviour et al., 2014), thus providing the potential for skillful long lead prediction of the development of low SAM and extreme conditions in Australia in October 2002. The default version of the model uses prescribed monthly varying climatological zonal mean ozone concentrations during the forecasts (MacLachlan et al., 2015). We thus conduct an experiment for a hypothetical prediction of conditions in spring 2002 by prescribing the 2002 monthly zonal mean ozone concentration during the forecasts. The details of the model and the experimental setup for prescribing the observed ozone anomalies are provided in section 2. General skill of the model to predict the Antarctic polar vortex variation and its downward coupling is demonstrated in section 3. Analysis of the impact of the observed ozone for the downward coupling and development of extreme low SAM and associated Australian temperature and rainfall anomalies during October 2002 is provided in section 4. Conclusions and suggestions for further experimentation with observed and predicted ozone are provided in section 5.

2 Coupled Prediction Model, Data, and Methods

The ACCESS-S1 prediction system (Hudson et al., 2017) is based on the UK Met Office GC2 coupled model GloSea5 system (MacLachlan et al., 2015). ACCESS-S1 became the operational subseasonal to seasonal prediction system at the Bureau of Meteorology in August 2019, replacing the low-resolution-low-top Predictive Ocean-Atmosphere Model for Australia model. The ACCESS-S1 system has high horizontal resolution (25 km in the ocean and ~60 km in the atmosphere), and its 85 vertical levels in the atmosphere mean that the stratosphere is well resolved to above 1 hPa. Hindcast and real time initial conditions for sea ice and the ocean are provided from the assimilation produced at the UK Met Office (e.g., MacLachlan et al., 2015). The atmospheric initial conditions for the hindcasts are provided by interpolating the ERA-I (Dee et al., 2011) of zonal wind (u), meridional wind (v), temperature, humidity, and surface pressure onto the ACCESS-S1 atmospheric model grid. In real time, the atmospheric initial conditions are provided by the Bureau of Meteorology numerical weather prediction system. Soil moisture is initialized with climatology (MacLachlan et al., 2015), and soil temperatures are interpolated from ERA-I data. Importantly, ozone concentrations for the control hindcasts are updated monthly during the forecasts and are prescribed to be zonally symmetric and to vary climatologically based on averaging the monthly ozone analyses of Cionni et al. (2011) for the years 1994–2005.

The ACCESS-S1 hindcasts consist of an 11-member ensemble, produced by perturbing the atmospheric initial condition (Hudson et al., 2017). Forecasts extend to 7-month lead time. In this study, we analyze 11-member ensemble hindcasts initialized on 1 August during 1990–2012 but focus on 2002. These standard hindcasts, which use climatological ozone, are referred to as the “control” forecasts.

Note that in contrast to the original set up of the ACCESS-S1 model and the version of the GloSea5 model used by Seviour et al. (2014) to assess seasonal predictive skill of the SH stratosphere and the SAM, we change the updating of the climatological ozone concentration that is read in from the ancillary file to occur monthly rather than every 30 days. Updating every 30 days in the GC2 model is based on a 360-day calendar (12 months of exactly 30 days length), which is the calendar used for the free-running climate simulations with this model. Updating every month uses the Gregorian calendar. The monthly mean ozone concentration values in the climatological ancillary file are recovered with the monthly updating but are not recovered with the 30-day updating because of the design of the updating code. As a result, the 30-day updating scheme as implemented in GC2 results in ~10% errors in the polar cap ozone concentration in September and November when the climatological ozone is varying most rapidly (Figure S1 in the supporting information). The net effect is to unrealistically shift the seasonal cycle of polar ozone to later in the spring, relative to the climatological minimum in October when using the 30-day updating. Limited testing suggests higher skill scores to predict the SAM and polar stratospheric variability when using monthly updating that replicates the input monthly ozone climatology (see section 3). We use the monthly updating for both the control hindcasts and the experiment (described below).

To produce the “observed” 2002 ozone concentration, we first calculate the monthly mean zonal mean ozone concentration anomalies for August 2002 to February 2003 using ERA-I reanalyses. The ERA-I ozone analyses have been shown to generally agree well with independent observations (within a few percent) in the stratosphere, but larger discrepancies are found in the polar night region and at the tropical tropopause (Dragani, 2011). We believe that the use of the ERA-I ozone analyses is justified for the current study because we focus on regions outside of the polar night, and the increase in Antarctic polar cap ozone during spring 2002 is large so that we are assessing large impacts. To calculate the anomalies, we subtract the monthly mean climatology based on ERA-I for the years 1980–2017, noting that this is a longer climatological period than that used for the ozone climatology in ACCESS-S1 (i.e., Cionni et al., 2011). We used a longer period to obtain a more stable climatology, but the character of the anomaly during 2002–2003 is not sensitive to using the shorter climatology 1994–2005. We then add the ERA-I anomaly into the ozone climatology from Cionni et al. (2011), which is then read into the model for our experiment during 2002–2003. Our main experiment is to repeat the control hindcast initialized on 1 August 2002 but using 2002 monthly ozone rather than climatological ozone. We refer to this 11-member ensemble with prescribed observed ozone as the “experiment,” noting that it is not a “prediction” per se because we are prescribing the observed ozone variation during the runs.

Note also that in both the control and experiment that zonal mean ozone is prescribed because that is the current capability of the model. During the 2002 SSW, the polar vortex was initially displaced off the pole during September, and there were large zonal wave number 1 and 2 components of the temperature and ozone anomalies. However, inspecting the monthly column ozone anomalies using the multi sensor reanalysis of van der A et al. (2015) (available from http://www.temis.nl/protocols/o3field/o3mean_msr.php) shows that as a result of irreversible mixing and wave breaking the ozone anomalies homogenized and became more zonally symmetric during October and were nearly perfectly symmetric during November (Figure S2). Thus, our use of zonal mean ozone is perhaps not as bad as it first seems for October and November when the ozone anomalies would be radiatively important and which is the period of focus for this study.

The statistical significance of differences between the control and the experiment ensemble mean forecasts for 2002 is assessed by a resampling method that provides a confidence interval for the difference. We resample 4,000 times, with replacement, the anomalies from the 11-member control and experiment, thereby forming 4,000 experiment minus control differences. We sort these differences to provide quantile thresholds. The null hypothesis for this test is that the experiment is not different from the control and we can reject this null hypothesis if more than 90% or 95% of the resampled differences are greater than 0 (or less than 0, depending on the field). We also assess statistical significance of the observed and control forecast anomalies using a standard t test.

For verification and additional analysis, we use the ERA-I monthly reanalyses of mean sea level pressure and pressure level zonal winds and temperature. These data are available on a 1.5° grid. Impacts in Australia are explored using the gridded analyses of rainfall and daily maximum temperature (Tmax) from the AWAP analyses (Jones et al., 2009). We focus on Tmax because its variations show stronger impacts from the SAM than does minimum temperature during austral spring (e.g., Hendon et al., 2007; Lim et al., 2019). The AWAP analyses are an optimum interpolation of the available station observations across Australia and are available monthly on a 0.25° grid. For compatibility with the forecast data, we create anomalies and percentile thresholds based on the 1990–2012 climatology.

To verify the SAM forecasts, we use the BAS SAM index, which is based on the normalized zonal mean pressure difference between 40°S and 65°S using available surface observations (Marshall, 2003; available from http://www.nerc-bas.ac.uk/icd/gjma/sam.html). High-polarity SAM refers to lower than normal surface pressures and a poleward shifted eddy-driven westerly jet and storm track (Thompson & Wallace, 2000). Forecast values of SAM are obtained by a similar calculation as for the BAS SAM index.

We monitor the vertical coupling of the stratosphere and troposphere in the Antarctic polar cap using the approach of Lim et al. (2018) (hereafter referred to as LHT2018). Interannual variations of the polar vortex are strongly phase locked to the seasonal cycle, peaking during spring. This phase-locked interannual variability is captured by the leading mode of a multiple empirical orthogonal function (EOF) analysis of the height-time anomalies (seasonal cycle removed) of monthly mean zonal mean zonal wind averaged over the subpolar region of 55–65°S. For this study, the input data to the multiple EOF are ordered from September–January each year for pressure levels ranging from 1,000 to 1 hPa. We follow the approach of LHT2018 and do not weight the data by level thickness to more effectively extract the stratospheric signal. The returned eigenvector, which we refer to as the stratosphere-troposphere coupled mode (S-T coupled mode), is a function of height (pressure level) and month of the year running from September to January (Figure 3a). The principal component time series (PC1; referred to here as the stratosphere-troposphere coupled mode index) consists of one value each year (Figure 3c). For this calculation, we use ERA-I data during 1990–2012 to be compatible with the hindcast record available from the forecast model. The basic characteristics of the returned eigenvector are, however, insensitive to the years included or the number of months before or after spring used.

Details are in the caption following the image
(a) Leading eigenvector from the height-time domain EOF analysis of subpolar zonal mean zonal winds (averaged over 55–65°S) using ERA-I for the period September–January 1990–2012. The contours are the climatological zonal mean zonal winds of 1990–2012. (b) As in (a) except leading eigenvector using ensemble mean hindcasts from ACCESS-S1 that were initialized on 1 August. The explained variance is indicated above each panel. The eigenvectors are scaled for a 1 standard deviation anomaly. (c) Principal component time series formed by projecting observed (ERA-I; blue bars) and forecasts (initialized on 1 August; yellow bars) onto the observed eigenvector displayed in (a). The correlation between observed and forecast principal components over 1990–2012 is 0.62.

The leading eigenvector displayed in Figure 3a accounts for ~63% of the input variance (this is larger than the explained variance reported in LHT2018 because we use September–January each year, while they used April–March each year). Large positive loadings of PC1 in Figure 3c are indicative of weakening and earlier than normal breakdown of the vortex that is apparent in September in the upper stratosphere (but which LHT2018 show that it begins as early as late winter) and descends into the troposphere over the following 4 months. A positive phase of the index is accompanied by warming of the Antarctic polar cap and development of anomalous circumpolar easterlies (LHT2018, their Figure 3). Negative loadings imply anomalous strengthening of the vortex and a delayed march of the seasonal cycle with a resultant later breakdown of the vortex. As expected, the principal component time series shows the largest loading in 2002. The 2002 SSW can thus be viewed as an extreme occurrence of anomalous seasonal weakening of the vortex, with the breakdown occurring earlier than normal (Table 1) (Black & McDaniel, 2007; Lim et al., 2019, their supporting information Figure S1).

Table 1. Observed, Experiment, Control, and Experiment Minus Control Differences of the October Mean Anomalies for SAM, the S-T Coupled Mode Index Anomalies for September–January, and the Final Breakdown Date of the Polar Stratospheric Vortex at 60°S 50 hPa Based on Ensemble Mean of Forecasts That Was Initialized on 1 August 2002
Index Observed Experiment Control Experiment minus control
October SAM index −3.10 −2.68 −1.38 −1.30 (p = 5%)
S-T mode index 2.56 1.95 1.45 0.5 (p = 15%)
Stratospheric vortex breakdown date 3 December 2002 (cf. 8 December) 1 December 2002 6 December 2002 (cf. 10 December) −5 days (p = 7%)
  • Note. The final vortex breakdown date is defined as the date when the 5-day mean zonal mean zonal winds at 60°S 50-hPa drop below 10 ms−1, following the definition of Black and McDaniel (2007). In the bottom row, the dates in the parentheses are the climatological vortex breakdown dates over 1990–2012 in the observation and in the control hindcasts. For the experiment minus control differences, bold values indicate significant differences at least at the 10% level, with the actual significance threshold indicated in parentheses.

3 Predicting Antarctic Polar Vortex Variations

Prior to exploring the impact of observed ozone variations for the evolution of the downward coupling of the 2002 warming, we first confirm that the ACCESS-S1 model has general skill to predict the seasonally phased locked downward coupling from the stratosphere to the troposphere associated with weakening (and strengthening) of the vortex. Here our focus is on the longer timescale evolution of the vortex and not on the occurrence of the SSW itself. Seviour et al. (2014) previously showed using the GloSea5 system that 1 out of 15 ensemble members produced a SSW when initialized on around 1 August 2002. ACCESS-S1, while producing a sustained weakened vortex when initialized on 1 August 2002 does not produce a SSW from any of its 11 members (Figure S3). Seviour et al. (2014) used more members (15 compared to 11 for ACCESS-S1) and used three lagged start times (25 July and 1 and 9 August), which perhaps provides enhanced spread and so is better able to capture the extremely rare 2002 SSW. The limit of predictability of this SSW has also been shown to only be about 1 week (e.g., Taguchi, 2018). However, our focus here is on the longer timescale evolution of the vortex, and ACCESS-S1 shows good skill to predict this variation as embodied in the S-T coupled mode. The observed and predicted S-T coupled mode index at lead time 1 month (i.e., forecasts initialized on 1 August and verified for September–January) is displayed in Figure 3c using the hindcasts during 1990–2012. The correlation between the predicted and observed S-T coupled mode indices is 0.62.

Using all the control hindcast years (1990–2012), the correlation of observed and predicted SAM index for October at 2-month lead time (i.e., forecast initialized on 1 August and verified for month of October) is 0.57 (not shown). Using the same hindcasts, the skill for predicting the September–November mean SAM at 1 month lead time (i.e., initialized on 1 August) is 0.64 (not shown). Seviour et al. (2014) also reported prediction skill of 0.64 for September–November mean SAM for hindcasts initialized 1 August, but this was based on a shorter hindcast period 1996–2009, and they used an earlier version of the GloSea5 model. Using the same shorter period (1996–2009), ACCESS-S1 produces a correlation skill for the September–November mean SAM at 1-month lead time of 0.75.

For 2002, the predicted value of the SAM index for October with ACCESS-S1 was the most negative, and the S-T coupled index for September–January was the most positive in the hindcast record (1990–2012) when initialized on 1 August. These results confirm that the ACCESS-S1 model skillfully predicts downward coupling from the stratosphere to the troposphere and the surface SAM during spring and specifically for spring 2002.

Some further motivation to prescribe the observed ozone variation during 2002 is provided in Figure 3b, which shows the leading eigenvector from the height-time domain multiple EOF analysis of the Antarctic subpolar zonal mean zonal wind using the control hindcasts initialized on 1 August 1990–2012. Compared to the leading multiple EOF derived from observations (Figure 3a), the leading multiple EOF from the control hindcasts displays realistic downward coupling from the upper stratosphere to the surface from October to January, but the model's zonal wind anomalies are weaker in the middle to lower stratosphere and troposphere. Importantly, the wind anomalies are also less persistent in the upper stratosphere compared to the observed mode (Figure 3a). Consequently, the sustained weakening of the vortex during vortex weakening years does not last as long in the forecast as in the observed. The prescription of climatological ozone could potentially contribute to the reduced persistence of the polar vortex anomalies in the stratosphere and the associated weaker downward coupling to the troposphere during vortex weakening events.

Turning to the forecast for 2002, the control forecasts initialized on 1 August 2002 for polar cap temperature and subpolar zonal wind are displayed in the top panels of Figure 4. Comparing to the observed evolution (Figures 1a and 1c), the overall monthly evolution of the downward coupling is fairly well predicted, confirming that there is predictability of the seasonal weakening of the vortex that extends well beyond the limit of predictability of the SSW itself. But the warming in the middle and lower stratosphere and the extension of the easterly anomalies to the surface are weaker than observed during spring. These deficiencies are consistent with the systematic bias of the model's S-T coupled mode (Figure 3b). Can the inclusion of the interannual ozone anomaly during 2002 improve the simulated strength and persistence of the polar vortex and associated surface climate anomalies?

Details are in the caption following the image
Time-height sections of predicted monthly mean anomalies of (a) polar cap temperature and (b) subpolar zonal wind for August 2002 to January 2003 from ensemble mean control (CTRL) forecasts initialized on 1 August 2002. (c and d) The respective experiment minus control differences. Unit for temperature is kelvin and for zonal wind is meters per second. The ERA-I ozone mixing ratio anomaly is contoured in (c) (interval 1 × 10−7 kg kg−1, negative values dashed, and zero contour is omitted). Stippling and hatching in (a) and (b) indicate the 2002 control forecast anomalies that fall in the one-sided 5% and 10% tails of the hindcast climatological distribution, respectively (i.e., magnitude greater than 1.72σ and 1.32σ, respectively, given 22 samples not including 2002), and stippling and hatching in (c) and (d) indicates significant differences at the 95% and 90% confidence level, respectively.

4 Impact of Ozone on the Evolution of the 2002 Warming

The experiment minus control differences for polar cap temperature and subpolar zonal wind (lower panels of Figure 4) demonstrate that prescribing the observed ozone anomaly strengthens the warming in the middle to lower stratosphere and the easterly anomalies and their downward penetration to the surface especially during October. This additional warming in the middle to lower stratosphere during September–November in the experiment coincides with observed positive ozone anomaly that was prescribed during the forecast. Prescribing the observed ozone, while further weakening the vortex, still does not result in any members producing a SSW although two members now come close (Figure S2). The lack of any members depicting a SSW in late September when initialized on 1 August, while possibly resulting from too few ensemble members or model errors, is consistent with the limit of predictability of the actual SSW being about 1-week lead time (e.g., Taguchi, 2018).

The prescribed 2002 ozone anomalies appear to strengthen the negative SAM response especially in October. The observed, control, and experiment minus control anomalies for zonal mean temperature and zonal wind during October are displayed in Figure 5. The control captures most of the features of the observed, including the peak warm anomaly over the polar cap in the lower to middle stratosphere, which is overlain by cold anomalies in the upper stratosphere. The peak easterly anomalies at ~60°S that extend from the upper stratosphere to the surface and are accompanied by westerly anomalies in the subtropics, indicative of low SAM, are also captured. The experiment minus control difference accentuates these anomalies, resulting in an apparently stronger negative swing in the SAM. The surface pressure anomalies (Figure 6) confirm that prescribing the 2002 ozone anomalies strengthens the low SAM surface response, enhancing the higher pressure over the polar cap and lowering pressures in the midlatitudes.

Details are in the caption following the image
October 2002 anomalies of zonal mean temperature (a, c, and e; K) and zonal wind (b, d, and f; ms−1) from (a, b) ERA-I, (c, d) the ensemble mean of control forecast, and (e, f) the experiment minus control differences. Forecasts were initialized on 1 August 2002. Stippling and hatching in (a)–(d) indicate that the observed and forecast anomalies of temperature and zonal wind are significant as described in Figures 1 and 4. Stippling and hatching in (e)–(f) indicate where the experiment minus control differences for temperature and zonal wind, respectively, are significant at the 95% and 90% confidence level, respectively. The ERA-I ozone anomaly is contoured in (a) and (e) (interval 4 × 10−7 kg kg−1, negative values dashed, and zero contour omitted).
Details are in the caption following the image
October 2002 anomaly of mean sea level pressure from (a) ERA-I, (b) the control forecast initialized on 1 August 2002, and (c) the experiment minus control difference. Unit is hectopascal. Stippling and hatching are the same as in Figure 5.

Prescribing the 2002 ozone anomaly roughly doubles the magnitude of the predicted negative SAM in October and brings the prediction in better agreement with the observed anomaly (Table 1). The S-T coupled mode index is also strengthened by use of the 2002 ozone (Table 1). However, the experiment minus control difference for the S-T coupled mode index is not significant presumably because the index depicts the evolution of the polar cap winds from September to January, so the strong impact of 2002 ozone in October is diluted.

As mentioned in section 3, the date of final breakdown of the vortex during 2002 was substantially earlier than normal (Black & McDaniel, 2007). The final vortex breakdown date from the ensemble mean control forecasts initialized on 1 August 2002 was predicted to be 6 December 2002 (Table 1), which is 4 days earlier than the climatological breakdown date based on all hindcasts initialized on 1 August 1990–2012. In the experiment, the final vortex breakdown was predicted to occur on 1 December 2002, and this difference from the control is statistically significant (p = 0.07). For reference, the observed 2002 and observed climatological breakdown dates using the method of Black and McDaniel (2007) are 3 and 8 December, respectively.

The promotion of more extreme negative SAM by the 2002 ozone variation is reflected in the predictions of Australian surface climate. Figures 7 and 8 show control and experiment predictions initialized on 1 August 2002 for the October mean Tmax and precipitation, respectively. The top panels show the ensemble mean anomaly, and the bottom panels show the probability of being in the upper quintile for Tmax and lower quintile for precipitation. Note that the climatological thresholds for the upper and lower quintiles are based on the control hindcasts 1990–2012 (excluding 2002). Comparing to the observed anomalies (shown in Figure 2), the control ensemble mean forecast well captures the warming in eastern Australia (Figure 7a) and correctly predicts the occurrence of an extreme event in the top quintile category in subtropical eastern Australia and some areas of Western Australia (Figure 7c). However, the ACCESS-S1 control fails to predict the extraordinarily warm conditions extending into northern and western Australia. Likewise, the extreme dry conditions are well depicted by the control forecast in the north and the east of the country, but forecast error is obviously seen in the southern part of Australia. The impact of prescribing observed ozone is to strengthen the predicted warm anomaly in the heart of the region where it is warmest (Figure 7b), so to make the predicted anomaly better agree with the observed (the pattern correlation of the predicted and observed Tmax anomalies increases from 0.11 for the control to 0.34 for the experiment). The predicted probability of a top quintile event (which will occur climatologically 20% of the time) more than doubles over northern and eastern Australia (40–60%) in the experiment forecast (Figure 7d), with the area of high probability matching better where the observed anomaly was extreme (hatched area in Figure 2a). Similar improvements are achieved for the precipitation forecasts (Figure 8), with prescribing the 2002 ozone anomaly acting to expand westward the region of predicted drying (Figure 8b). The pattern correlation for the control and experiment precipitation anomalies (Figures 8a and 8b) with the observed (Figure 2b) increases from 0.46 for the control to 0.64 for the experiment. The probability of a lower quintile event also increases over most of eastern Australia (Figure 8d). For both Tmax and precipitation, the experiment minus control changes in predicted anomalies are consistent with impacts from an enhancement of low SAM.

Details are in the caption following the image
October 2002 Tmax anomaly from the (a) control forecasts and (b) the experiment minus control differences from the forecasts initialized on 1 August 2002. The unit is kelvin. Stippling and hatching in (a) and (b) are the same as in Figure 5. The probability of October 2002 Tmax predicted to be in the top quintile from (c) the control and (d) the experiment forecasts. The climatological occurrence of being in the top quintile is 0.2. The pattern correlation of the control anomaly (a) with the observed anomaly (Figure 2a) is 0.11. The pattern correlation of the experiment anomaly (not shown) with the observed anomaly (Figure 2a) is 0.34.
Details are in the caption following the image
As in Figure 7 except for rainfall and its bottom quintile forecasts. The unit in (a) and (b) is millimeters per day. Stippling and hatching as in Figure 5. The pattern correlation of the control anomaly (a) with the observed anomaly (Figure 2b) is 0.46. The pattern correlation of the experiment anomaly (not) shown with observed anomaly is 0.64.

5 Conclusions

We have shown that prescribing the observed variation of zonal mean ozone resulted in more realistic simulation of the downward coupling of the 2002 stratospheric warming and its subsequent impacts on Australian surface climate. We used the ACCESS-S1 model, which has a well-resolved stratosphere and has good skill to predict vertical coupling in the SH polar vortex when climatological ozone is prescribed during the forecasts (Seviour et al., 2014). Using climatological ozone, the ACCESS-S1 model was able to predict the large-scale monthly evolution of the polar vortex from late winter to spring during the 2002 stratospheric warming when initialized on 1 August 2002. However, it underpredicted the surface response, especially the record-breaking negative SAM and associated rainfall and temperature extremes in Australia that developed during October 2002. Prescribing the observed zonal mean ozone variation during the forecast resulted in more realistic weakening of the stratospheric vortex and more than doubling of the magnitude of the negative SAM during October, bringing the prediction in better alignment with the observations. This enhanced negative SAM, as a result of prescribing the observed ozone variation during the forecast, resulted in more realistic prediction of extreme warm and dry conditions across eastern Australia during October 2002.

The mechanism for intensification of the weakening of the vortex and its downward extension to the surface (i.e., shift to low SAM) because of prescribing the observed interannual ozone variation is not yet determined but presumably is similar to the response to the ozone trend (e.g., Orr et al., 2012). The local response in the stratosphere is an increase in polar cap temperature that peaks in October (Figure 4c), which would act to reduce the equator to pole temperature gradient in the stratosphere. The timing of this response makes sense because incoming insolation over the polar cap increases from September to December, while the interannual ozone anomaly peaked in September but extended into October. However, the mechanism for the subsequent enhancement of the surface response (i.e., increased negative SAM response) is yet to be elucidated for interannual ozone variations but presumably involves some combination of downward control and eddy feedbacks, at both planetary and synoptic scales (e.g., Orr et al., 2012; Song & Robinson, 2004; Yang et al., 2015).

The model's response to the interannual ozone anomaly during 2002 is reminiscent (but with opposite sign) of the response to the negative ozone trend that occurred during the later twentieth century (e.g., Gillett & Thompson, 2003; Orr et al., 2012; Polvani et al., 2011): Depletion of Antarctic stratospheric ozone during austral spring drives a cooling in the middle stratosphere over the pole, which subsequently leads to a promotion of positive surface SAM during early austral summer. However, in the present study the lag of the surface response with respect to the stratospheric ozone anomaly is much shorter (1 month or less). This difference in lags of the SAM response to interannual ozone variations (~1-month lag) and the trend in ozone (2- to 3-month lags) has been previously reported based on observational analyses (Son et al., 2013). It is not clear whether this indicates a different mechanism for the downward coupling in response to interannual versus long timescale variations of stratospheric ozone. This different lag might also reflect that the interannual ozone anomalies such as during the 2002 stratospheric warming are largely a result of the weakening and warming of the vortex (e.g., Seviour et al., 2014), and so their impacts via perturbing the radiative heating are largely confined to when the vortex is already perturbed. In contrast, ozone depletion operates on top of whatever interannual variability of the vortex is present each year, so it will take time for a SAM response to develop in response to the ozone trend. This difference in lag does highlight that a variety of mechanisms might contribute to the downward coupling of polar stratospheric anomalies to the surface (e.g., Hitchcock & Haynes, 2016; Orr et al., 2012; Polvani et al., 2011; Rao et al., 2020; Song & Robinson, 2004; Thompson et al., 2006; Yang et al., 2015), and further study is required to understand the apparently different timescales of the response to interannual and long timescale stratospheric ozone forcing.

The major implication of the present study is that interannual variations of polar ozone may be an additional source of long lead climate predictability that is yet untapped. This predictability can only be realized if accurate prediction of ozone variations and their radiative impacts can be made in the dynamical model. In principle, this is possible because much of the interannual ozone variability is controlled by the dynamics (e.g., Seviour et al., 2014). Ozone is prognostic, for instance, in the ECMWF Integrated Forecasting System model, which uses simplified chemistry and transport, and ozone is initialized via assimilation of available observations (Dethof & Hólm, 2004). However, at the present time the predicted ozone variations in the ECMWF models do not interact with radiation because it introduces forecast errors (Johnson et al., 2019). These errors may not be related to the polar stratospheric ozone variations discussed here. Thus, one possibility to explore is to restrict the impact of predicted ozone variations on radiation in the polar stratospheric regions where the prediction of ozone variations is doable because of their strong tie to the resolved atmospheric dynamics (e.g., Seviour et al., 2014).

Although we showed a pronounced positive impact by prescribing ozone during the major stratospheric warming of 2002, it also remains to be seen the effect of prescribing interannual variations of ozone for all other years especially on forecast skill of the SAM. It could be that 2002 was so anomalous that in other years there will be smaller impacts of prescribing the interannual ozone variations. To this end, we are now conducting experiment forecasts (i.e., prescribing observed zonal mean ozone variations) from 1 August for all hindcast years. These results will be reported in due course.


This project is supported by funding from the Australian Government Department of Agriculture and Water Resources as part of its Rural R&D for Profit program. We thank G. Liu for running the experiments and D. Hudson and M. Wheeler for their constructive feedback on the manuscript. We thank the three anonymous reviewers for their thorough and constructive comments. This research was undertaken on the NCI National Facility in Canberra, Australia, which is supported by the Australian Commonwealth Government. The ERA-Interim reanalyses are described in Dee et al. (2011). The AWAP rainfall and temperature analyses are described in Jones et al. (2009). The BAS SAM index is described in Marshall (2003). The KNMI multisenor column ozone reanalyses are described in van der A et al. (2015). The ACCESS-S1 model used for the experiments is described in Hudson et al. (2017).

    Data Availability Statement

    The NOAA CPC AAO index is updated monthly and available from the website (https://www.cpc.ncep.noaa.gov/products/precip/CWlink/daily_ao_index/aao/aao_index.html).