Simulation of Record Arctic Stratospheric Ozone Depletion in 2020

In the Arctic winter/spring of 2019/2020, stratospheric temperatures were exceptionally low until early April and the polar vortex was very stable. As a consequence, significant chemical ozone depletion occurred in the Arctic polar vortex in spring 2020. Here, we present simulations using the Chemical Lagrangian Model of the Stratosphere that address the development of chlorine compounds and ozone in the Arctic stratosphere in 2020. The simulation reproduces relevant observations of ozone and chlorine compounds, as shown by comparisons with data from the Microwave Limb Sounder, Atmospheric Chemistry Experiment‐Fourier Transform Spectrometer, balloon‐borne ozone sondes, and the Ozone Monitoring Instrument. Although the concentration of chlorine and bromine compounds in the polar stratosphere has decreased by more than 10% compared to peak values around the year 2000, the meteorological conditions in winter/spring 2019/2020 caused unprecedented ozone depletion. The lowest simulated ozone mixing ratio was about 40 ppbv. Because extremely low ozone mixing ratios were reached in the lower polar stratosphere, chlorine deactivation into HCl occurred as is normally observed in the Antarctic polar vortex. The depletion in partial column ozone in the lower stratosphere (potential temperature from 350 to 600 K, corresponding to about 12–24 km) in the vortex core was calculated to reach 143 Dobson Units, which is more than the ozone loss in 2011 and 2016, the years which —until 2020— had seen the largest Arctic ozone depletion on record.

Due to the greater dynamical activity in the Arctic, which results in a less stable and warmer polar vortex, Arctic ozone depletion is typically much less pronounced and much more variable than in the Antarctic (e.g., Solomon, 1999;WMO, 2018). Stratospheric temperatures in the Arctic in winter and spring are generally much higher than those in the Antarctic, where low temperatures are the key factor in the formation of the ozone hole. However, strong chemical ozone loss has been identified in cold Arctic polar winters and springs exhibiting a persistent polar vortex into spring, for example in March 1997(Douglass & Kawa, 1999Johansson et al., 2019;Kuttippurath et al., 2012;Manney & Lawrence, 2016;Manney et al., 2011;Müller et al., 1997;Newman et al., 1997;Sinnhuber et al., 2011;Tilmes et al., 2004;Weber et al., 2011;WMO, 2018). For Arctic ozone loss to occur, a stable vortex is required as well as temperatures below a threshold of about 195 K to trigger chlorine activation. These conditions exhibit large inter-annual variation. Low temperatures do not need to cover the entire vortex volume. For the Arctic, chlorine activation typically occurs in association with the formation of PSCs over only a small portion of the vortex. Nonetheless, a small cold vortex portion often results in substantial vortex-wide activation (e.g., Wegner et al., 2016). Frequently, an early sudden stratospheric warming (SSW), for example in January, causes the vortex to split or even break down, resulting in very weak chemical ozone loss. However, there are also cases such as winter 2012/2013 that started with a very cold December followed by a SSW in January in which significant ozone loss occurred (Manney et al., 2015).
At the end of the period of catalytic ozone depletion, the Arctic typically follows a pathway involving the predominant formation of ClONO 2 from active chlorine, as opposed to the Antarctic, where active chlorine is mainly converted into HCl. The pathway of chlorine deactivation depends on how low ozone mixing ratios fall (Crutzen et al., 1992;Douglass et al., 1995;Grooß et al., 1997;Müller et al., 1994). In Arctic winters with a cold and stable vortex, for example the winter 1996/1997 or winter 2010/2011, denitrification and ozone depletion cause chlorine deactivation to follow a more Antarctic-like pathway, that is, a faster formation of HCl than in more typical Arctic winters (Douglass & Kawa, 1999;Manney et al., 2011). In winter 2019/2020, which is considered here, this combination of factors also caused a similar evolution of chlorine deactivation .
Here we report on simulations using the model Chemical Lagrangian Model of the Stratosphere (CLaMS) which show unprecedented Arctic ozone depletion in the year 2020 (Dameris et al., 2021;Feng et al., 2021;Manney et al., 2020;Wohltmann et al., 2020) accompanied by a significant increase in surface UV radiation (Bernhard et al., 2020). The reported simulations indicate that the major causes of the severe ozone depletion reported in the Arctic in 2020 are the low stratospheric temperatures and the exceptionally stable polar vortex extending into spring .

Aura-MLS
The observations made by the Microwave Limb Sounder (MLS) on board the Aura satellite are the main data set used in this study. Further details of the MLS measurements in Arctic winter 2019/2020 are reported by Manney et al. (2020). MLS makes its observations in limb-viewing geometry in the A-Train orbit, circling the Earth about 14 times per day and covering latitudes from 82°S to 82°N. We use MLS version 4.2 data . Here observations of O 3 , N 2 O, HCl, H 2 O, ClO, and HNO 3 are used for model initialization, boundary conditions, and comparison with the model results. The vertical resolution of the data in the lower stratosphere ranges from about 3 km (O 3 , HCl, H 2 O, ClO) to about 5 km (N 2 O, HNO 3 ). The reported accuracy or systematic uncertainty in the lower stratosphere (100-22 hPa) is 5%-8% for ozone, 6% at 22 hPa to 44% at 100 hPa for N 2 O, 0.2 ppbv for HCl, 4%-9% for H 2 O, 1.0-2.4 ppbv for HNO 3 , and 0.2-0.25 ppbv for ClO, respectively . The precision of the data is of little consequence here as we mostly use averages over multiple profiles.

ACE-FTS
The Atmospheric Chemistry Experiment-Fourier Transform Spectrometer (ACE-FTS) on the SCISAT satellite is a Fourier transform spectrometer with high spectral resolution (0.02 cm −1 ) operating from a wavelength of 2.2-13.3 μm employing a Michelson interferometer (Bernath et al., 2005). Since 2004, it has GROOSS AND MÜLLER 10.1029/2020JD033339 2 of 17 observed around 30 profiles per day in solar occultation geometry with the majority of the measurements in high latitudes. The vertical resolution is about 3-4 km. Here we use version 3.6 data of O 3 and N 2 O as well as the five main components of total inorganic nitrogen NO y (NO, NO 2 , N 2 O 5 , HNO 3 , ClONO 2 ). The retrieval used in version 3.6 is described by Boone et al. (2013). Ozone measurements are in general agreement with other observations to within ±5% and have a small bias of +2% (Sheese et al., 2017).

Aura-OMI
The Ozone Monitoring Instrument (OMI) onboard the Aura satellite is a nadir-looking, push-broom solar backscatter spectrometer that measures the solar radiation back-scattered by the Earth's atmosphere and surface over the entire UV/VIS wavelength range from 270 to 500 nm (Levelt et al., 2006). OMI is the successor to the Total Ozone Monitoring Spectrometers (TOMS). Since 2004, OMI has provided high spatial resolution measurements on about 14 orbits per day with ground pixels of 13 × 24 km and a swath width of 2,600 km. This results in an almost global coverage of the distribution of the atmospheric ozone column on a daily basis. Ozone columns are derived by means of differential optical absorption spectroscopy (DOAS). The ozone columns show a 0.4% ± 0.6% bias with respect to ground-based stations (McPeters et al., 2008).

Ozone Sondes
We also present data from two balloon-borne ozone sondes that were launched in Ny Ålesund, Spitsbergen (78.9°N, 11.9°E, WMO station number 1004). The balloons were flown with an ECC ozone sonde together with a radiosonde RS41. Using standardized operating procedures the ECC-sondes yield an accuracy of about 5%-10% up to an altitude of 30 km (Smit et al., 2007). These measurements were made as part of an international ozone sonde campaign in 2019/2020 (Wohltmann et al., 2020).

Model Setup
The simulations described here were performed using the CLaMS. Unlike in previous publications, the chemical transport model CLaMS here is driven by operational analyses by the European Centre for Medium Range Weather Forecasts (ECMWF). The vertical velocities were thus derived from heating rates calculated by an offline radiation module based on the Morcrette scheme (Morcrette, 1991;Zhong & Haigh, 1995) as in an earlier model setup (Konopka et al., 2004). The vertical model coordinate is the hybrid potential temperature ζ (Konopka et al., 2004;Pommrich et al., 2014) that is identical to the potential temperature above 100 hPa (above about θ = 400 K) and corresponds to a terrain following σ-coordinate system below 100 hPa with a smooth transition. The model simulation described here starts on November 1, 2019 and runs until mid-April 2020. The model domain consists of the Northern Hemisphere with a horizontal resolution of 100 km. The vertical range spans from the surface to a potential temperature of 900 K divided into 32 levels with a variable resolution in the stratosphere ranging from 0.7 km at 9-12-1 km at the top layer, yielding about 820,000 air parcels in total.
The chemistry scheme used here has been described earlier (Grooß et al., 2011McKenna et al., 2002). The chemical kinetics and absorption cross sections are based on current recommendations (Burkholder et al., 2015). A comprehensive stratospheric chemistry scheme is employed which includes a full set of heterogeneous reactions. The vertical re-distribution of NO y and H 2 O due to sedimenting nitric acid trihydrate (NAT) and ice particles is determined by Lagrangian NAT and ice particle tracking Tritscher et al., 2019). This algorithm allows denitrification and also dehydration to be represented in the model. Heterogeneous chemistry is calculated here only on liquid PSCs and aerosols. This assumption is justified as heterogeneous chlorine activation is dominated by liquid PSCs and aerosols. NAT particles have much lower particle number densities than liquid particles so the heterogeneous reaction rates for liquid particles exceed those for NAT (Portmann et al., 1996;Tritscher et al., 2021;Voigt et al., 2005;Ward et al., 2014;Wegner et al., 2012). NAT particles are important as they cause denitrification through their gravitational settling (e.g., Fahey et al., 2001;Solomon, 1999;Tritscher et al., 2021).

Initialization and Boundary Conditions
The chemical composition at the beginning of the simulation and the boundary conditions at the surface as well as at the top model layer at a potential temperature of 900 K are derived from a variety of data and model results. The procedure is similar to previous work . Data from MLS were used within ±2.5 days of the initial time for O 3 , HNO 3 , H 2 O, HCl, and N 2 O. The observation locations were transferred to the initial time using CLaMS trajectories and averaged to a 2° latitude × 6° longitude grid. In the troposphere below a potential temperature of 350 K, the initialization of these compounds was taken from a multi-annual CLaMS simulation with a simplified chemistry setup (Pommrich et al., 2014). For total inorganic nitrogen (NO y ), chlorine (Cl y ) and bromine (Br y ), tracer correlations with N 2 O were used as listed in the supporting information. NO y was derived from ACE-FTS from the sum of the observations of HNO 3 , NO, NO 2 , N 2 O 5 , and ClONO 2 between October 1 and November 21, 2019. The sum of these NO y compounds correlates well with N 2 O, with different correlations for different latitude ranges (tropics, sub-tropics, mid-latitudes, higher, and polar latitudes). The NO y initialization is based on the correlations in these five latitude ranges, which are listed in the supporting information. Cl y and Br y were initialized using correlations with N 2 O derived from balloon observations, updated from Grooß et al. (2011) as described in the supporting information. The partitioning within these chemical families was taken from a 2-D reference model (Grooß, 1996) with updated boundary conditions (WMO, 2018).
The chemical composition at the upper boundary at the potential temperature level of 900 K was calculated using the same data sources. Twice a month at this level, the data described above are mapped and averaged into bins of equivalent latitude Φ e (Butchart & Remsberg, 1986;Lary et al., 1995). Thus the chemical composition at the upper boundary is determined. Simulations have been performed using this setup for several Arctic winters since 2010.

Results
The CLaMS simulation aims to reproduce the processes involved in Arctic stratospheric ozone depletion during winter 2019/2020. We analyze the observations and the model results in the polar vortex using the concept of equivalent latitude (Butchart & Remsberg, 1986). In the simulation, NAT particles start to form on November 16, 2019. The period of chlorine activation in the stratosphere was exceptionally long . ClO x /Cl y averaged over the polar vortex core (Φ e > 75°N) exceeded 50% at a rather early stage on 14 December at a potential temperature between about 490 and 540 K. Thus, some ozone depletion already occurred in December (compare Figure 4 below). Chlorine activation in 2020 lasted until 22 March, and the polar vortex remained stable into the month of April .

Descent in the Polar Vortex
To evaluate the simulation results, it is essential to demonstrate the ability of the model to reproduce atmospheric observations. One important aspect of the simulation is whether the transport and dynamics are represented correctly. In particular, the diabatic descent of air inside the polar vortex should be realistic. Diabatic descent in any given winter determines not only the ozone mixing ratios at a given potential temperature level in spring, but also the available Cl y (Strahan et al., 2014). We therefore evaluated the simulated descent of vortex air by comparing the simulation with ACE-FTS N 2 O data, which have better vertical resolution and accuracy in the lower stratosphere than the MLS measurements of N 2 O.   correspond to the model initialization and the latest date with a sufficient number of ACE-FTS observations at latitudes within the polar vortex. As only about 20 observed profiles contribute to each of the averages shown, their mean value and scatter may not be representative for the entire vortex, but it is still the best available information for investigating the deviation of the simulation with respect to the observations. The solid lines show the mean mixing ratio profiles and the dotted lines show the standard deviations. The CLaMS simulation shows weaker descent than observed between early November and late March. For example, for 200 ppbv N 2 O, the simulated change is 54 K instead of the observed 64 K. For 100 ppbv N 2 O, the simulated descent is 68 K instead of the observed 90 K. This difference in vertical displacement is on the order of the model vertical resolution. Because of this discrepancy, we introduced a sensitivity simulation, labeled "D25", which is designed to better represent the observed N 2 O vertical profile in the vortex, that is, which better represents the observed diabatic descent. This is achieved by artificially increasing the diabatic cooling rates north of 60°N by 25%. The dashed blue line in Figure 1 shows that the diabatic descent is represented well in the D25 simulation. However, since this artificial modification causes a not self-consistent wind field, the aim of the D25 simulation is only to indicate the sensitivity of the model results with respect to the diabatic descent rate.

Ozone Depletion
The focus of this study is on the simulation of ozone depletion in the Arctic in 2020. To evaluate the simulation results, they are compared with ozone observations. First, the overall development of ozone is compared with MLS data. As the MLS measures about 3,500 profiles per day, the comparison is performed in such a way that similar air-masses are averaged. To do so, both the data and the simulations are averaged on a daily basis into equivalent latitude and potential temperature bins. The width of the equivalent latitude bins is chosen in such a way that they correspond to equal geographical areas. Figure 2 displays different illustrations of these averages for MLS measurements. Panel a shows the temporal development of the vortex core average profiles for equivalent latitudes Φ e > 75°N; panel b shows the temporal development of the mixing ratio at the potential temperature level of 450 K and as a function of equivalent latitude. Panels c and d show equivalent latitude/potential temperature cross sections for two particular days-31 January and 29 March.
These 2 days were selected as representative days in the early phase and toward the end of the ozone depletion period, respectively. Most of the plots below show results for 29 March, as the lowest ozone mixing ratios were found around this time. Figure 3 shows the development of ozone corresponding to Figure 2, but derived from the CLaMS simulation. For this comparison, the model data were not interpolated to MLS locations but were averaged directly into the same bins. This procedure is justified by the high spatial coverage of MLS. Although there are some small differences, the ozone depletion over winter and spring is well reproduced by the simulation. The lowest ozone mixing ratios in the stratosphere are seen at a potential temperature between 400 and 500 K toward the end of March. The gradients of ozone at the edge of the polar vortex are reproduced well. The lowest MLS mean vortex core (Φ e > 75°N) ozone mixing ratio reached is 0.21 ± 0.28 ppmv (mean value ± standard deviation) at 450 K on 27 March. The corresponding ozone mixing ratio from the CLaMS simulation in these air masses is 0.29 ± 0.26 ppmv. Figure 4 shows the simulated ozone mixing ratios in the vortex core (Φ e > 75°N) versus time at four different potential temperature levels together with the corresponding MLS observations. Dashed lines depict the evolution of the passive ozone tracer. The D25 simulation shows the sensitivity with respect to the vertical descent rate. Stronger vertical descent increases Cl y and thus the chemical ozone depletion but at the same time also increases ozone itself since ozone increases with altitude. The latter effect is visible in the increase in passive ozone.
At 450 K, the simulation reproduces the temporal development and especially the observed low ozone mixing ratios, both for the reference simulation and for the D25 simulation with enhanced vertical descent. At 400 K, the ozone mixing ratio in April in the reference simulation is lower than the MLS observations by about 0.2 ppmv. In the D25 simulation at 400 K, the vortex core ozone mixing ratio after late March decreases below the reference simulation in early March reaching about 0.1 ppmv below the reference simulation in April, even though the simulated passive ozone is larger by about 0.2 ppmv. This additional ozone depletion of 0.3 ppmv is due to the Cl y increase caused by the artificially increased descent. At 550 K, the simulated ozone depletion is likely an under-estimation, yielding larger ozone mixing ratios than observed by about 0.3 ppmv by mid-March. This discrepancy is about 0.6 ppmv for the D25 simulation with increased descent. As the simulation D25 is based on inconsistent wind fields, it should only be used to demonstrate the model's sensitivity to descent rate, a critical parameter. The differences between observations and model results arise partly because of the difficulty to accurately reproduce the large vertical gradients of the ozone profiles. However, the shown discrepancy is still small compared to the ozone depletion itself derived as the difference between the simulated ozone and the passive ozone tracer.
The simulated ozone profiles in spring 2020 were compared with a variety of observations. Figure 5a shows the average vortex ozone profile observed by ACE-FTS between 23 and 29 March. The corresponding CLaMS simulations were evaluated at the ACE-FTS tangent point locations. Dotted lines show the standard deviation of the observed mixing ratios. Since only about 20 observed profiles contribute to this vortex core average, their mean value and scatter may not be representative for the entire vortex, but similar to the N 2 O comparison in Figure 1, these data are a good measure of the quality of the simulation results. At a potential temperature of around 450 K, the simulations show lower ozone mixing ratios (by 0.35 ppmv) than those observed by ACE-FTS, but measured and modeled profiles agree within the combined standard deviations. Figure 5b shows the corresponding comparison with MLS ozone data for 29 March for the vortex core Φ e > 75°N. Instead of about 20 profiles as in the case of ACE-FTS, about 230 MLS profiles contribute to this vortex-core average, providing a much better coverage of the vortex core. The dashed blue line corresponds to the model data convolved with the MLS averaging kernel. Here the minimum ozone near 450 K is well reproduced. However, for equivalent latitudes near the vortex edge, the model under-estimates somewhat the minimum ozone mixing ratio (see Figure S9 of the supporting information). The Lagrangian model CLaMS has been shown to better reproduce trace species gradients near the vortex edge than a comparable GROOSS AND MÜLLER 10.1029/2020JD033339 6 of 17    Furthermore, Figures 5c and 5d show the comparison of CLaMS results with measurements by two balloon-borne ozone sondes launched from the Ny Ålesund station on 29 March and 12 April. Between 400 and 500 K, the simulation and the ozone sonde measurements are in good agreement; especially around 450 K, the comparison of the minimum ozone mixing ratio between the simulated and observed ozone mixing ratios is better than for the satellite data.
Although the minimum ozone mixing ratio at a potential temperature around 450 K has been simulated quite well, it is notable from the comparisons with all different data sets that the simulated ozone mixing ratio around 575 K is over-estimated by the simulation, whereas it is somewhat under-estimated below about 400 K. The reasons for these discrepancies remain unclear. A comparison of ClO and HCl data shows that the model might underestimate the vertical extent of the chlorine activation, placing the upper edge of the activated region at a lower altitude than is observed (compare supporting information Figures S1, S2, S4-S6), but the limited vertical resolution of the data prevents a definitive determination.
In the following, we discuss the simulated chemical ozone change. The chemical ozone change ( Figure 6) is calculated as the difference between the simulated ozone and a passive ozone tracer that has identical initialization and boundary conditions and transport as the simulated ozone, but no chemical change.    In the middle stratosphere at mid-latitudes above about 600 K, this simulation also indicates ozone depletion ( Figure 6c). However, the most important ozone-depletion reactions in this region are NO x -induced catalytic cycles. Due to the shorter lifetime of mid-latitude ozone compared to high latitudes, the simulated mid-latitude ozone mixing ratios are mostly determined by chemical production and loss rather than by transport processes (see e.g., Figure 11 of Garcia & Solomon, 1985). Thus, the use of a passive ozone tracer to calculate chemical ozone loss is inappropriate for mid-latitude air masses above about 600 K. These mid-latitude aspects will not be discussed further in this paper as our focus is on chlorine-catalyzed ozone loss in the polar region.
With respect to the variability of surface UV radiation due to polar ozone depletion, the total ozone column is important. To calculate the simulated total ozone column, the ozone column above 900 K potential temperature is added to the column over the vertical model domain considered here. The column above 900 K amounts to about 30 Dobson Units (DU) as determined from a climatology dependent on the equivalent latitude and time of year (Grooß & Russell, 2005). Employing this estimate of the ozone column above 900 K, the geographical distribution of ozone columns is well reproduced by the simulation.   the CLaMS simulation to accurately reproduce the structure and the gradients in the ozone distribution. Inside the vortex, the so-calculated total ozone column is about 10-15 DU larger than the OMI data (see Figure S10 of the supporting information).
The strongest ozone depletion occurs in the vortex core (Φ e > 75°N) in the partial column over the potential temperature range between 350 and 600 K. The average development of the simulated depletion over this region is shown in Figure 8. For this volume, the maximum partial column ozone depletion of 143 DU was reached on 4 April. Toward the end of this time period there is some variability in the deduced ozone depletion values because of the beginning of vortex breakup. This is because the ozone-depleted air masses are no longer congruent with contours of PV (or Φ e ). In the D25 sensitivity simulation with artificially increased polar descent rates, the uncertainty of the partial column ozone loss estimate is also evident. Increased descent causes an increase of the passive ozone tracer but, more importantly, it also leads to an increase in Cl y . Since the ozone mixing ratios at lower altitudes contribute significantly to the column, the derived partial ozone column depletion for the D25 simulation is 160 DU, which is 17 DU more than in the reference simulation. Thus, there is some uncertainty in the derived partial ozone column depletion. However, the fact that the absolute total ozone column distribution is simulated well (Figure 7) strengthens the credibility of the simulation results.
GROOSS AND MÜLLER 10.1029/2020JD033339 9 of 17   Figure 6. Simulated chemical ozone depletion determined by the difference between simulated ozone and the passive ozone tracer averaged in equivalent latitude and potential temperature bins. The results are displayed in the same style as in Figures 2 and 3 Nash et al. (1996) at the potential temperature of 450 K (roughly corresponding to Φ e = 70°N). Equivalent latitude development of the vortex edge is shown in Figure S11 of the supporting information. For 2011, ozone depletion restricted to the potential temperature range 350-550 K in early April was calculated here as 101 DU, somewhat lower than the 112 DU reported by Kuttippurath et al. (2012) for this vertical range. Over the potential temperature range of 380-550 K, ozone loss was calculated as 95 DU, lower than the 120 DU reported by Sinnhuber et al. (2011) for mid-April. About 10 DU of the latter difference is due to the fact that the 2011 CLaMS run stopped at the end of March and thus did not simulate the ozone depletion that occurred in early April 2011. Wohltmann et al. (2020) derive a partial column ozone loss of 133 DU (370-550 K) for 2011, which is more than the ozone loss for 2020. However this value is inconsistent with our simulations (97 DU, 370-550 K) and other publications (95 DU by Isaksen et al., 2012 and 99/108 DU by Adams et al., 2012) besides the studies mentioned above (Kuttippurath et al., 2012;Sinnhuber et al., 2011).
As indicated above, the derived ozone depletion in 2020 exceeds the ozone depletion for previous Arctic winters including 2011. Although the partial column ozone depletion in the Arctic in 2020 reached record or near-record levels, it was still well below typical Antarctic values. For example, in the simulation for Antarctic spring 2011 , the simulated maximum partial ozone column depletion in the lower stratosphere vortex core (Φ e > 75°S, 350-600 K) is 228 DU on 14 October.

Initial Chlorine Activation
Although the main aspects of polar chlorine activation and the resulting catalytic ozone loss are established (WMO, 2018), there are scientific questions on these issues that have not yet been resolved. In particular, HCl depletion has been observed in the cold and dark early winter polar vortex which cannot be reproduced using the processes currently implemented in models . This effect is most pronounced in the Southern Hemisphere, but has also been seen in the 2016 Arctic winter . It is also observed here for the year 2020 (Figure 9), but to a lesser extent than in the simulation for the Arctic winter 2015/2016. At 500 K, the HCl mixing ratios in the simulation for late December and January do not reach the low observed values (Figure 9). The discrepancies on other levels are smaller (compare Figures S1, S2 and S12 in the supporting information). The smaller discrepancy compared to 2016 is likely due to the earlier onset of chlorine activation in late November 2019.

Chemistry During the Termination of Ozone Depletion
In simulations, the development of individual air masses with extremely low ozone mixing ratios cannot be well simulated when mixing is over-estimated in a numerical model, for example when numerical GROOSS AND MÜLLER 10.1029/2020JD033339 11 of 17 diffusion is too strong. The Lagrangian simulations shown here have the advantage of following individual air masses for some time without mixing. Lagrangian simulations allow the low Antarctic ozone mixing ratios observed to be reproduced (Grooß et al., 2011). In the current simulation for the Arctic in 2020, the simulated ozone mixing ratios did not quite reach the lowest levels reported for the Antarctic ozone hole (e.g., Solomon et al., 2005), where chemical ozone depletion can continue until extremely low ozone mixing ratios of the order of 10 ppbv have been reached (Grooß et al., 2011). The lowest simulated ozone mixing ratio in the polar lower stratosphere (38 ppbv) was reached on March 24, 2020 (84°N, 131°E, θ = 439 K). It is very unlikely that the model reproduces the exact location and mixing ratio of the ozone minimum in the real atmosphere. Nevertheless, this individual air parcel is useful here to analyze chlorine deactivation. This parcel was not affected by the CLaMS mixing algorithm for 50 days. The temporal development along this air parcel was reconstructed by exactly repeating the simulation for this air parcel only but saving more output details. Figure 10 shows the temporal development of ozone and chlorine compounds of this air parcel during the termination phase of ozone depletion. For an extended period-from about mid-February to early March-ClO x levels remain strongly elevated in this air parcel, while the mixing ratios of HCl, ClONO 2 , and HOCl remain very low.
For typical Arctic ozone mixing ratios, the dominant fraction of the active chlorine species undergoes deactivation into ClONO 2 through the ClO + NO 2 reaction (Douglass et al., 1995;Müller et al., 1994). For previous cold Arctic winters/springs, a chlorine activation pathway into HCl for air masses with low ozone was reported for Arctic winters/springs in 1997 (Douglass & Kawa, 1999), 2000(Wilmouth et al., 2006(Santee et al., 2008, and 2011 (Arnone et al., 2012;Manney et al., 2011). This is also the case for winter/ spring 2020 . The CLaMS simulation similarly shows more typical Antarctic behavior as low ozone mixing ratios-almost comparable to Antarctic values-are reached, allowing deactivation into HCl to occur (Douglass et al., 1995;Grooß et al., 2011;Müller et al., 2018).
Chlorine activation in the atmosphere is likely dominated by liquid PSCs and aerosol particles (e.g., Portmann et al., 1996;Tritscher et al., 2021), which is also the assumption in the model runs presented here (see Section 3.1). It is well known that the chlorine activation rate has a strong temperature dependence, for example, the dominant chlorine activation reaction ClONO 2 + HCl shows an increase of a factor of 10 for a temperature decrease by 2.3 K (see e.g., Figure 1 of Wegner et al., 2012). As the temperature of the selected air parcel varies by roughly 10 K along its trajectory, Figure 10 illustrates the periods with and without a significant chlorine activation rate: periods corresponding to a first order heterogeneous reaction rate of reactions with HCl larger than 1 d −1 are depicted by gray shading. As temperatures decrease below about 194-195 K the first order HCl depletion rate by the reaction with ClONO 2 or HOCl increases above 1 d −1 GROOSS AND MÜLLER 10.1029/2020JD033339 12 of 17  Grooß et al. (2018). In 2016, this average only corresponds to the vortex core until mid-March (see Figure S11). despite the low mixing ratios of the reactants. However, in late March because of the low ozone mixing ratios this heterogeneous HCl depletion rate is counter-acted by fast chlorine deactivation into HCl. As is the case in the Antarctic, chlorine deactivation into HCl in the Arctic in 2020 already begins at a time (early March) when PSCs and liquid aerosol particles which allow heterogeneous chlorine activation to occur are still present. As the ozone mixing ratio did not decrease to values as low as in the Antarctic (Grooß et al., 2011;Solomon et al., 2005), the deactivation into HCl occurs over a longer time scale of about 2 weeks for this specific air parcel ( Figure 10).
The chlorine deactivation into HCl is also present on a vortex-wide scale. The simulated vortex core averages for HCl and ClONO 2 (Φ e > 75°N) at a potential temperature of 500 K are shown in Figure 9a together with the corresponding MLS HCl data. ACE-FTS observations of ClONO 2 are also shown as individual points and not as an average, since the sparse coverage of ACE-FTS observations is not likely to be representative for the entire vortex. The rate of increase of vortex-averaged HCl mixing ratios in late March 2020 is lower than for the example given of the air parcel with the lowest ozone mixing ratio found in the simulation ( Figure 10) since the HCl increase rate is largest for the lowest ozone mixing ratio (Grooß et al., 2011). Furthermore, in the vortex average for the potential temperature of 500 K, an HCl increase up to about 2.5 ppbv in mid-April is simulated, consistent with the HCl observations. The formation of ClONO 2 is also simulated in the model in February and March, but the vortex average is under-estimated compared to ACE-FTS ClONO 2 observations. This is likely in part due to the ACE-FTS sampling with a better coverage toward the vortex edge, where ClONO 2 abundances are larger, as noted for example by Santee et al. (2008) and evident in Figures of the supporting information. Figure S12   O 3 different pathways that are mainly determined by the value of the ozone mixing ratio. The formation of both ClONO 2 and HCl at different locations has been simulated (see Figure S13 of supporting information). As small differences in the simulated ozone mixing ratio have a large impact on the deactivation pathways, some deficiencies might arise in the simulation of chlorine deactivation.

Relation of Arctic Ozone Loss to Vortex Dynamics and Temperatures
In accordance with observational studies (e.g., Manney et al., 2020;Wohltmann et al., 2020), it is shown here through our CLaMS simulations that Arctic ozone depletion in 2020 was unprecedented despite the fact that halogen levels in the stratosphere have been decreasing since about the year 2000 (Engel et al., 2018;Newman et al., 2007), meaning that a decrease in chlorine-catalyzed ozone depletion would be expected. From our simulation, it is evident that in particular the low temperatures, their extension into spring, and the stable stratospheric vortex  are responsible for the strong ozone depletion observed. Wohltmann et al. (2020) indicate that the low dynamical resupply of ozone-rich air due to the strong transport barrier at the vortex edge may in addition have contributed to the observed low ozone mixing ratios. Feng et al. (2021) noted that one reason for the stable and cold vortex in this winter was an extremely weak planetary wave driving in the Northern Hemisphere indicated by an extremely low extra-tropical winter mean eddy heat flux. The increase in greenhouse gas concentrations over the last few decades has resulted in more dynamic activity and could therefore cause a less stable polar vortex (Kretschmer et al., 2018). However, the increase of greenhouse gas concentrations causes an increase in infrared radiative emissions and thus a temperature decrease in the stratosphere (e.g., Maycock et al., 2018). Rex et al. (2006) suggested that as the coldest Arctic stratospheric winters have been getting colder, this has resulted in an increasingly strong ozone loss in these winters. There is a debate about such projections based on studies using extreme value statistics (Rieder & Polvani, 2013;Rieder et al., 2014) or sunlit vortex volumes (Pommereau et al., 2013). Rieder et al. (2014) argue that the temperature decrease in the lower stratosphere is dominated by ozone depleting substances as opposed to well-mixed greenhouse gases, which dominate the temperature trend in the upper stratosphere.
As it is expected that in the upcoming decades ozone-depleting substances decline and greenhouse gases increase, the importance of dynamical processes for determining Arctic springtime ozone will likely grow (Butler et al., 2016;Portmann et al., 2012). Chemistry Climate Model (CCM) simulations show that even after 2040, when EESC has declined substantially, early springtime Arctic total column ozone in certain years can drop by about 50-100 DU below the long-term average (Bednarz et al., 2016;Langematz et al., 2014). Such losses are substantial, but are less than the loss in winter 2019/2020 reported here. Nevertheless, although stratospheric chlorine levels are decreasing, it is possible that Arctic ozone losses in the range observed in 2020 may continue to occur in the coming decades.

Conclusions
Arctic winter/spring 2019/2020 was exceptionally cold in the polar stratosphere and the polar vortex was stable for an unusually long period until early April, which led to significant chemical ozone depletion through the end of March and the conservation of this depleted ozone in the polar vortex until early April 2020. The extent of ozone loss has been simulated by CLaMS, where these simulations correspond well with satellite observations. Although stratospheric halogen loading (EESC) has decreased by about 11%-12% since its peak in 2000, the particular meteorological conditions in 2020 allowed enhanced levels of active chlorine to persist over a long period of time, causing unprecedented Arctic ozone depletion. Chlorine activation already started in late November 2019 and lasted until early March. Because of the low ozone values reached, a substantial production of HCl occurred in late March and April, resembling deactivation patterns seen in the Antarctic. In the vortex core average, the strongest simulated ozone depletion was 2.74 ppmv at 470 K on 30 March. The lowest simulated ozone mixing ratio in the polar lower stratosphere was about 40 ppbv. The calculated vortex core average ozone partial column depletion between the potential temperatures of 350 and 600 K was 143 DU in early April. This number is very sensitive to the representation of diabatic descent in the polar vortex. An additional ozone partial column loss of 17 DU was calculated for a simulation with enhanced descent, which might be considered as more realistic in the sense that it better fits the observed N 2 O mixing ratios as a measure of diabatic descent. The partial ozone column loss in the lower stratosphere of the reference simulation exceeds that in the simulations for the Arctic winters in 2016 (126 DU) and 2011 (106 DU), which showed the strongest ozone depletion before 2020. However, Arctic ozone depletion values in 2020 still remained below the typical values of Antarctic ozone depletion. If Arctic winters with such low temperatures and with an extension of low temperatures into spring recur in the coming decades, similarly strong springtime ozone depletion can be expected despite decreasing stratospheric chlorine loading.