The Unusual Stratospheric Arctic Winter 2019/20: Chemical Ozone Loss From Satellite Observations and TOMCAT Chemical Transport Model

Satellite observations of relevant trace gases and meteorological data from ERA5 were used to describe the dynamics and chemistry of the spectacular Arctic 2019/20 winter/spring season. Exceptionally low total ozone values of slightly less than 220 DU were observed in mid‐March within an unusually large stratospheric polar vortex. Very high OClO and very low NO2 column amounts observed by GOME‐2A were indicative of unusually large active chlorine levels and significant denitrification, which likely contributed to large chemical ozone loss. Using results from the TOMCAT chemical transport model (CTM) and ozone observations from S5P/TROPOMI, GOME‐2 (total column), SCIAMACHY, and OMPS‐LP (vertical profiles), chemical ozone loss was evaluated and compared with the previous record Arctic winter 2010/11. The polar‐vortex‐averaged total column ozone loss in 2019/20 reached 88 DU (23%) and 106 DU (28%) based upon observations and model, respectively, by the end of March, which was similar to that derived for 2010/11. The maximum ozone loss (∼80%) observed by OMPS‐LP was near the 450 K potential temperature level (∼18 km altitude). Because of the larger polar vortex area in March 2020 compared to March 2011 (about 25% at 450 K), ozone mass loss was larger in Arctic winter 2019/20. It is shown that Arctic cap temperatures were at a record 40‐year low in winter, suggesting that Arctic cold winters are getting colder. Tropical UTLS temperatures show an apparent upward trend during the last two decades that may indicate a weakening of the Brewer‐Dobson circulation and strengthening of the polar vortex.

In this study, we report on chemical ozone loss in Arctic winter 2019/20 derived from total ozone data from TROPOspheric Monitoring Instrument (TROPOMI) and ozone profiles from Ozone Mapping and Profiler Suite-Limb Profiler (OMPS-LP) satellite data in combination with results from the 3D chemical transport model (CTM) TOMCAT (Chipperfield, 2006). A particular focus in this study is on the comparison between 2019/20 and the previous record winter 2010/11 for which ozone column data from Global Ozone and Monitoring Experiment-Metop A (GOME-2A) and limb data from SCanning Imaging Absorption SpectroMeter for Atmospheric CHartographY (SCIAMACHY) are also used.
In many regards, this Arctic winter set a new record in terms of ozone loss and atmospheric circulation induced by a record high positive Arctic oscillation (AO) . There has been a controversy over whether cold Arctic winters may be getting colder, possibly related to overall stratospheric cooling in a changing climate (Rex et al., 2004(Rex et al., , 2006Rieder & Polvani, 2013). In this study, we also briefly address how the very recent Arctic winter may support this conclusion.
The structure of this study is as follows. Section 2 describes the observational data and CTM used here. Section 3 gives a brief description of the polar meteorology in the Arctic winter/spring 2019/2020, including a comparison to the record winter/spring in 2010/11. Section 4 shows results from other trace gas observations (NO 2 and OClO) along with ozone followed, in Section 5, by chemical ozone loss calculations using the combination of model and observational data. Our summary and concluding remarks on how this record winter may be interpreted in the context of past Arctic winters and expected long-term changes in ozone and atmospheric dynamics are provided in Section 6.

Merged WFDOAS Total Ozone
The merged GOME, SCIAMACHY, GOME-2, and TROPOMI (GSG) total ozone timeseries consists of total ozone data retrieved using an advanced version of the University of Bremen Weighting Function DOAS (WFDOAS) algorithm (Coldewey-Egbers et al., 2005). The merging of the various instruments has been briefly described in Weber et al. (2018). A monthly mean latitude-dependent bias correction, used to successively adjust SCIAMACHY (2002SCIAMACHY ( -2012 and GOME-2A (2007-present) to the initial GOME (1995-2011) data record, has been applied here to daily gridded data. Recently bias-adjusted WFDOAS data from GOME-2B (with a better global coverage than GOME-2A) starting in 2015 and TROPOMI (Veefkind et al., 2012) starting in 2018 have been added into the merged daily WFDOAS total ozone timeseries, available at a spatial resolution of 1.25° × 1° (longitude × latitude).

SCIAMACHY and OMPS-LP Ozone Profiles
SCIAMACHY aboard Envisat (2002Envisat ( -2012 and OMPS-LP aboard SUOMI-NPP (2012-present) observe the atmosphere in limb geometry from a sun-synchronous orbit and collect radiances in the UV/visible spectral region. Ozone concentrations are retrieved from 60 down to 10 km (or cloud top height), by using for both instruments the SCIATRAN radiative transfer model and retrieval software package . The typical vertical resolution of the retrieved profiles is about 2.5 km (OMPS-LP) and 3.7 km (SCIAMACHY). Details of the retrieval algorithm and a validation of the ozone profiles can be found in Jia et al. (2015) and Arosio et al. (2018).

GOME-2A OClO and NO 2 Columns
Stratospheric OClO and NO 2 columns retrieved from UV/visible observations of instrument such as GOME and SCIAMACHY have already been used in previous studies (Richter et al., 2005;Wagner et al., 2001;Weber et al., 2003). Here, the data analysis follows Richter et al. (2005) but using GOME-2A data instead of GOME observations. Since OClO is rapidly photolyzed, substantial amounts can only be measured at very large solar zenith angles (SZA). In order to remove the effect of changing illumination during the time series, only observations at 90° SZA are used. As rapid photolysis also changes the OClO concentration along the light path, no attempt is made to convert OClO slant columns into vertical columns. As the geometry of the light path remains the same for all measurements at 90° SZA, the results are still comparable from day to day and between years.
The variations in local equator crossing times (and twilight zones at a given day of the year) of the various satellites complicate the comparison between satellites. For this reason, we limit our comparisons to results from the GOME-2A instrument (launched in 2006) that covers both cold Arctic winters studied here. As a result of the sun-synchronous orbit of GOME-2A, the latitude probed at 90° SZA varies from 65° to 85° over the winter/spring period, and does not reflect vortex or polar-cap averages as the ozone data used here.

TOMCAT Chemical Transport Model
TOMCAT/SLIMCAT (hereafter TOMCAT) is a 3-D chemical transport model (CTM), which has been described in Dhomse et al. (2019). The model contains a detailed description of stratospheric chemistry, including heterogeneous reactions on sulfate aerosols and PSCs. Here model simulations are performed at a horizontal resolution of 2.8° × 2.8° with 32 altitude levels from the surface to ∼60 km. The model is forced with the fifth generation ECMWF reanalysis (Hersbach et al., 2020) also known as ERA5 meteorological fields.
The model setup used here is similar to the control simulation used in Dhomse et al. (2019). Briefly, the model includes updated surface mixing ratio scenarios for ozone depleting substances from Carpenter et al. (2018). The model also includes an additional time-dependent chlorine contribution from Very Short Lived Substances (VSLS), as described in Hossaini et al. (2019). We use the Coupled Model Intercomparison Project Phase 6 (CMIP6) (Eyring et al., 2016) recommended stratospheric aerosol surface area density (SAD) which are available for the 2004-2016 time period (see ftp://iacftp.ethz.ch/pub_read/luo/CMIP6/). After 2016, we use climatological monthly mean SAD values calculated for the 2006-2016 time period (Dhomse et al., 2015). To diagnose chemical ozone loss in the Arctic stratosphere, the model has an additional passive ozone tracer that is initialized from the model chemical ozone tracer on December 1 of each year.
In recent model updates Feng et al. (2021), the supersaturation of HNO 3 for type I PSC formation was implemented according to Grooß et al. (2018) and the Cl 2 O 2 absorption cross sections are from Burkholder et al. (2015) with an assumed quantum yield of 1. Solar flux variations  are taken from the NRLSSI2 empirical model (Coddington et al., 2016) that are recommended for CMIP6 simulations as implemented in Dhomse et al. (2016). For the year 2020, solar fluxes are held constant at December 2019 values.

ERA5 Reanalysis
Wind and temperature data from the ERA5 reanalysis are used here for determining dynamical properties (vortex and PSC volume) and driving the TOMCAT CTM. For the polar vortex diagnostics, 6-hourly data at a spatial resolution of 0.75° × 0.75° were used.

Meteorology
The Arctic winter/spring 2019/20 exhibited a strong polar vortex with persistent PSCs observed from mid-November until April, as shown in Figure 2. Here the PSC volume was calculated using ERA5 data at potential temperature levels in order to identify grid boxes with temperatures below the PSC formation threshold as described in Feng et al. (2007). PSC volume in units of million square km was estimated from PSC areas calculated from 400 to 750 K altitudes in steps of 25 K potential temperature. For the vertical extent, it was assumed that 25 K roughly corresponds to 1 km altitude (Knox, 1998).
In Arctic winter 2019/20, the PSC volume was at a record high since 1979 in the second half of November. It remained high throughout December and January and was again at a record high in the second half of February through to nearly the end of March. The early PSC formation (chlorine activation) and very high PSC volumes in March, as shown in Figure 2a, favored strong depletion in ozone. The previous record winter 2010/11 showed a similar evolution in the volume of type I and type II PSCs as this year, with the exception that the volumes were generally smaller in 2011 and PSCs started to form later (end of November). Temporary lows in PSC volume had a very similar timing in both winters with local minima observed in the middle of January and a sharp short-term drop in early February, most likely related to minor stratospheric warming events perturbing the polar vortex. Maximum PSC volumes were reached at the end of January in both winters. At that time type II (ice) PSCs were also maximum in both winters. Ice PSCs can lead to strong dehydration and removal of water vapor that may result in a delay in deactivating active chlorine into HCl in early spring . Figure 2b shows northern hemisphere (NH) monthly mean eddy heat fluxes at 100 hPa (area-weighted average from 45°N to 75°N) for the two cold Arctic winters considered here, along with data from their preceding winters 2009/10 and 2018/19, respectively. The eddy heat flux is a measure of the planetary wave activity which drives the Brewer-Dobson circulation and associated ozone transport as well as stratospheric meteorology. From January to March, the eddy heat flux was persistently below the long-term mean in both 2010/11 and 2019/20, resulting in reduced ozone transport as well as lower stratospheric temperatures (Newman et al., 2001;Randel et al., 2002;Weber et al., 2011). In February 2020, the 100 hPa eddy heat flux reached the lowest value since 1980, which may have been responsible for setting record high PSC volumes starting by the end of February 2020. Both the continuous low dynamical activity, also linked to the positive WEBER ET AL. anomaly in the Arctic Oscillation , and very low stratospheric temperatures contributed to the very low ozone in Arctic winter 2019/20. Further details on the stratospheric meteorology in this particular winter and comparisons to past winters can be found in this journal's special issue and other studies (Dameris et al., 2021;Inness et al., 2020;Lawrence et al., 2020;Manney et al., 2020).

Trace Gas Observations
The evolution of ozone, NO 2 , and OClO, above the Arctic in 2019/20 are displayed in Figure 3. Corresponding timeseries for the year 2010/11 (previous record winter) as well as the years preceding both cold winters, 2009/10 and 2018/19, with more typical conditions are also shown to demonstrate the large variability from year-to-year. Panel (a) shows the evolution of the polar cap mean total ozone (50°N-90°N). Due to the Brewer-Dobson circulation, total ozone normally increases over the winter reaching, on average, an annual maximum in March (thick gray curve). Starting in mid-February, however, polar ozone strongly declined in 2011 and 2020. In mid-March, polar-cap mean were the lowest in both winters since the mid-1990s (since start of the WFDOAS merged total ozone timeseries), a time when stratospheric halogens originating from man-made ozone depleting substances (ODSs) were maximum (Newman et al., 2007). In 2020, the polar cap mean remained very low from March until May.
WEBER ET AL.  The polar minimum total ozone evolved in a very similar way to the mean (Figure 3b). A first record minimum was observed in March 2011 and this record was broken again in March 2020 with total ozone being slightly below 220 DU for a brief period (Inness et al., 2020;Wohltmann et al., 2020), a value which commonly defines the boundary of the Antarctic ozone hole (e.g., NASA Ozone Watch, 2020). The steady decline in minimum ozone in polar winter is considered a good proxy for continued polar chemical ozone loss (Müller et al., 2008). The rapid declines and rises in minimum ozone observed in early November and late January 2020 are, in contrast, purely dynamical in nature. They are caused by subtropical streamers intruding into polar latitudes producing so-called ozone mini holes by fast horizontal advection of ozone away from a region with a strongly elevated tropopause (Dameris et al., 2021;James et al., 2000;Weber et al., 2002).
In both Arctic winters 2010/11 and 2019/20, the mean NO 2 and OClO slant columns were record minima and maxima, respectively, by late February and early March (Figures 3c and 3d). In particular, OClO levels reached a new record (since 2007) in March 2020, which indicates substantial chlorine activation up to the last measurements mid of March. After that, the sun is too high for GOME-2A measurement at 90° SZA and the time series ends before the end of chlorine activation is reached.
Stratospheric NO 2 levels are generally small in the winter polar regions as most NO 2 is converted into its night-time reservoirs (N 2 O 5 and HNO 3 ) during polar night (e.g., Burrows et al., 1999). As the sun returns to the Arctic, NO 2 levels usually increase through mixing of polar air masses with mid-latitude air and destruction of HNO 3 by photolysis and reaction with OH. A stable vortex and denitrification by subsidence of condensed HNO 3 in PSCs delay this process as well as subsequent deactivation of active chlorine into their reservoir species (ClONO 2 ), explaining both the low NO 2 columns and the extended chlorine activation observed in 2011 and 2020.

Chemical Ozone Loss
The chemical ozone loss in both cold Arctic winters 2010/11 and 2019/20 is estimated using the vortex-average approach (Harris et al., 2002). In this approach, total ozone and ozone profiles are averaged daily within the confines of the polar vortex and the difference to a passive ozone tracer, here from the TOMCAT CTM, is considered as the accumulated ozone loss. The vortex edge is here defined by the combination of maximum wind speed and potential vorticity (PV) gradient (Nash et al., 1996) and was determined at the 450 K potential temperature level. The timeseries of vortex-averaged total ozone and chemical ozone loss from TROPOMI and TOMCAT are shown in Figure 4 for the winter 2019/20. The same quantities for the winter 2010/11 are displayed in Figure 5 using observations from GOME-2A and SCIAMACHY. While panel (a) shows the various timeseries of total ozone over the course of the winter, corresponding polar ozone losses are displayed in panel (b). The TOMCAT vortex-averaged ozone has evidently a negative bias with respect to TROPOMI, which is particularly large in the middle of the winter. Based on comparisons with Microwave Limb Sounder (MLS) ozone (not shown) Feng et al. (2021), this may be due to an underestimate of descent in the very low stratosphere (around 420 K) and too much horizontal mixing with mid-latitude air at this level.
TROPOMI, OMPS-LP (and SCIAMACHY) measure in the optical range and thus do not observe the polar night region. As a consequence, only 10%-20% of the polar vortex area is usually covered by satellite observations, depending on the exact location, in late December and early January. The agreement between observations and model improves significantly when using model averages from the sunlit part of the polar vortex only (green and dotted line in Figure 4a). In the beginning of December observations agree well with the model in the sunlit part of the vortex, but this bias increases with time until the end of March (when the polar vortex is completely illuminated), indicating that the modeled ozone loss is slightly larger than that observed (by ∼15 DU).
The modeled ozone loss, the dark blue curve in Figure 4b, is calculated for the full vortex average. The observed total ozone loss (light blue curve) is only an estimate of the full-vortex averaged loss. It is assumed that the observed model-observation bias (difference between dotted and green line) in the sunlit part of the polar vortex is representative for the full vortex. If this bias is then added to the full-vortex modeled ozone loss we obtain an estimate of the observed full-vortex ozone loss in the entire vortex. Due to this assumption, there are some remaining uncertainties in the evolution of the observed full-vortex ozone loss over the winter, however, by mid-March the entire polar region is sunlit.
A proper estimate of the accumulated ozone loss obtained at the end of March (day 90) requires that in early December observation and model in the sunlit part of the vortex agree well when ozone depletion is minimal (see difference between dotted and green line on day −30). This was the case in both Arctic winters considered here. As the modeled vortex ozone is lower than observations (see difference between black and WEBER ET AL.  Alternatively, one could also have taken the difference between the sunlit part of modeled passive ozone and observations (not shown here). One important assumption for both alternatives of the observed ozone loss calculation here is the negligible difference between passive ozone and satellite observations in the sunlit part of the polar vortex ozone in early December.
By the end of March 2020, the TROPOMI accumulated total ozone loss amounts to 88 DU (23%) and for the TOMCAT CTM to 106 DU (28%) with respect to TOMCAT passive ozone (red curve in Figure 4a). These losses are quite similar to the results from the previous record winter 2010/11 as indicated by the triangles in Figure 4b (see also Figure 5). On March 15, whereas the observed minimum total ozone is near to its lowest value (Figure 3b), the mean total ozone loss is even slightly higher in 2011 than 2020, for both observations and CTM. Figure 6 shows the time series of March daily mean vortex-averaged ozone profiles from SCIAMACHY (2011) and OMPS-LP (2020). Only profiles with a PV value higher than 38 PVU (at 475 K) were averaged to obtain the daily Arctic vortex mean. The rapid decline of ozone near 450 K potential temperature levels throughout March is clearly evident. It appears that the largest decline in 2020 was slightly below 450 K, a bit lower than in 2011 apparently in agreement with Microwave Limb Sounder (MLS) observations reported in Manney et al. (2020). However, one needs to be cautious here as the vertical sampling of SCIAMACHY (3.3 km ≈ 75 K) is too coarse to clearly support this. Figure 7 shows a time-altitude cross section of the accumulated ozone loss from SCIAMACHY (2010/11) and OMPS-LP (2019/20). Similar to total ozone, the ozone loss here is calculated from the difference of the daily mean observed ozone profiles to the passive ozone from TOMCAT. These differences are calculated only from TOMCAT passive ozone profiles collocated with SCIAMACHY/OMPS-LP observations mainly in the sunlit part of the polar vortex. Passive ozone is initiated in the model on December 1 each Arctic winter. A comparison between observations and passive ozone in early December, a period where chemical ozone loss is still very small, revealed differences of about 12 DU (2010) and 21 DU (2019) in the 350-550 K column, which are accounted for in the ozone loss shown in Figure 7. By the end of March 2020, a maximum ozone loss of 2.1 ppmv near 450 K was observed. In 2011, the accumulated loss reached a maximum of 2.2 ppmv which is comparable to the value from 2020. These values are slightly smaller than, but in good agreement with, Manney et al. (2011Manney et al. ( , 2020, confirming that the ozone losses observed in both winters were at a record low. with the total ozone data in Figures 4b and 5b. The OMPS-LP subcolumn ozone loss is 105 DU, about 20 DU larger than the TROPOMI total column ozone loss at the end of March 2020. The modeled subcolumn loss on the other hand (95 DU) is closer to the observation-derived total column loss of 88 DU. In 2011 the observation-derived and modeled subcolumn ozone loss at the end of March 2011 differ slightly by 5 DU. The largest error in the estimated ozone loss comes from uncertainties in establishing a proper initial Arctic vortex-mean ozone value from the UV/visible observations as well as uncertainties in the CTM, for example, vertical transport in the polar vortex and uncertainties in photochemical data propagating into uncertainties in the model chemistry. The overall uncertainty in the established polar ozone loss is estimated to be about 15% (about 15 DU) as given by the difference between observed and modeled ozone column loss. Nevertheless, both CTM and observations agree that Arctic vortex-averaged ozone losses in both Arctic winters 2011 and 2020 were very similar.
The product of mean column ozone loss and average vortex area provides an estimate of the total number of ozone molecules lost (which is proportional to the ozone mass loss). At 450 K, the vortex area was, on average, about 20 million square km, about 4 million square km larger than in 2011 (see also Figure 10a in Lawrence et al., 2020). As a consequence, the mass loss in ozone was about 25% larger in 2020 compared to 2011 assuming a similar vortex-averaged column ozone loss in both winters (see Figures 4 and 5). In terms of ozone mass loss, the Arctic winter 2019/20 therefore sets a new record high.

Summary and Conclusions
The Arctic winter/spring 2019/20 is one of the coldest on record (Dameris et al., 2021;Inness et al., 2020;Lawrence et al., 2020) with temperatures sufficient for PSC occurrence from November until April. The PSC volume was at a record high (based on available observations) in November and from late February throughout March, which indicates substantial polar ozone loss. OClO slant columns were very high and NO 2 slant columns very low in February and March, indicating large chlorine activation and extensive de-WEBER ET AL.  nitrification, both consistent with continued chemical ozone losses well into early spring. In February 2020, the planetary wave activity was at a record low resulting in very low temperatures in February and March and record high PSC volume.
Using total ozone observations from TROPOMI and TOMCAT/SLIMCAT model simulations, the total column ozone loss was estimated to be 88 DU and 106 DU, respectively, by the end of March (23% and 28% loss, respectively). From OMPS-LP observations ozone profile losses were found to reach 2.1 ppmv at 450 K (∼18 km) in good agreement with MLS observations . The combined uncertainty of derived polar ozone loss (from model and observations) is estimated to be on the order of 15%. The vortex-averaged ozone loss was very similar in both Arctic winters 2010/11 and 2019/20, but the ozone mass loss was significantly higher (about 25%) in 2020 than 2011 due to the larger area of the Arctic vortex in March 2020.
An important question concerns the reason behind such a large ozone loss observed in winter 2019/20, given the fact that ODSs are mostly declining as a consequence of the Montreal Protocol and Amendments. While in the Antarctic first signs of ozone recovery have been detected (de Laat et al., 2017;Solomon et al., 2016;Weber et al., 2018), the variability in stratospheric meteorology and ozone in the Arctic is still too large to uniquely identify ozone recovery (e.g., Chipperfield et al., 2017;Dhomse et al., 2018). CTM calculations, however, show that the ozone loss would have been larger in 2019/20 with ODS at levels of the mid-1990s (Feng et al., 2021).
The Arctic winter 2019/20 exhibited record-low polar-cap temperature at 100 hPa based upon ERA5 reanalysis data as shown in Figure 8. Although most Arctic winters were rather warm (above 1σ variability) after the mid-1990s, about two winters in a decade were extremely cold. The hypothesis discussed for some time that cold Arctic winters are getting colder in a changing climate (Rex et al., 2004(Rex et al., , 2006Rieder & Polvani, 2013) gains new relevance after the record temperatures observed in 2019/20. Also shown in Figure 8 are tropical 100 hPa temperatures during boreal winter, displaying an anti-correlation with polar-cap temperatures (Yulaeva et al., 1994). This correlation is a result of the inter-annual variability in the Brewer-Dobson circulation (weak circulation associated with higher tropical upper troposphere-lower stratosphere (UTLS) temperatures due to weaker vertical ascent in the tropics).
The two unusual cold Arctic winters of the past decade fall in a period where a weak positive tropical UTLS temperature trend is apparent. This suggests that a slight weakening of the Brewer-Dobson circulation, opposite to the expected long-term trend from climate change (e.g., Aschmann et al., 2014;Garfinkel et al., 2017), may have contributed to these two recent extreme Arctic winters. A strengthening of the Arctic polar vortex in recent decades has been linked to a warming in the North Pacific sea surface temperature (Hu et al., 2018, and references therein). Acknowledgments ERA5 reanalysis were made available to U. Bremen by ECMWF via a data agreement with the German Weather Service (DWD). Wuhu Feng, Sandip S. Dhomse, and Martyn P. Chipperfield are grateful for access to the ERA5 data. Part of the data processing was done at the German HLRN (High Performance Computer Center North). The model simulations were performed on the UK national Archer and Leeds Arc4 HPC systems. The financial support of part of this work from the State of Bremen, DAAD grant (CA), and ESA SOLVE Living Planet Fellowship (CA) is gratefully acknowledged. The UB contribution also benefited from the DFG VolImpact and BMBF SynopSys-Ozon Projects. Open access funding enabled and organized by Projekt DEAL.