The discovery of large ozone depletion in the Antarctic spring, now known as the ozone hole , came as a complete surprise, although the effect of chlorine-catalysed ozone depletion in the stratosphere had been discussed since the 1970s . Nevertheless, after its discovery, the processes leading to the formation of the ozone hole were rapidly established and are now understood in great detail e.g. and briefly summarised below.

The decomposition of anthropogenically emitted source gases, such as chlorofluorocarbons (CFCs), releases chlorine in the stratosphere, which is first converted into the passive non-ozone-depleting reservoir compounds HCl and ClONO2.

We present simulations from the three models (CLaMS, WACCM and TOMCAT/SLIMCAT) for the Antarctic winter 2011. This year had a rather large ozone hole with a size within the top 10 of the last 4 decades and was chosen because both MIPAS and MLS data were available.

The setup of the models with respect to initialisation, boundary conditions, transport schemes and PSC formation is different, which also contributes to differences between the model results. The formulation of stratospheric chemistry, especially heterogeneous chemistry of chlorine compounds, is implemented as commonly used and recommended, and is comparable among the three models.

The development of HCl in polar stratospheric winter is strongly influenced by heterogeneous chlorine activation. The heterogeneous reactions that cause this typically occur below a temperature threshold of about 195 K on cold binary aerosol or on PSCs composed of either HNO3 / H2SO4 / H2O STS solution, water ice or crystalline NAT. An important and fast HCl sink is the heterogeneous reaction ClONO2 + HCl → HNO3 + Cl2 . At the beginning of winter, as temperatures drop below the threshold for heterogeneous reactions to occur, this reaction quickly depletes HCl typically until all available ClONO2 has reacted, since HCl is usually available in large excess . This step is therefore referred to as the “titration step”. After that, since no reaction partner is available, HCl mixing ratios would remain constant in the dark in the absence of transport or mixing processes. A further depletion of HCl is only possible if there is a reaction partner. This reaction partner first needs to be formed, and in addition to ClONO2 it could be HOCl which would then react by the heterogeneous reaction HOCl + HCl → H2O + Cl2 . The resulting geographic distribution HCl and its development over the winter is presented in Fig. . The figure shows maps of MLS HCl observations and corresponding CLaMS results on the 500 K potential temperature level for selected days in May, June and July. On 21 May, when the HCl depletion due to heterogeneous processing is first apparent, the model seems to represent the observations well. By 20 June, HCl depletion has mainly occurred at the vortex edge, where it is more distinctly pronounced in the observations. The comparison on 20 July clearly demonstrates the discrepancy between simulations and observations. The model shows HCl mixing ratios well above 1.5 ppbv in the vortex core, while the MLS observations indicate almost zero values of HCl, within the limits of the total estimated uncertainty of the data.

To investigate the HCl discrepancy further, we averaged both the observations and simulation for each day with respect to equivalent latitude and potential temperature using 20 equivalent latitude (Φe) bins of equal area between Φe=50 and 90∘ S. Figure  shows the characteristics of the observed HCl development by displaying cuts through the equivalent latitude/potential temperature space. Figure a shows the time development of the vortex-core average (Φe>75∘ S) as a function of potential temperature. Figure b shows the time development on the 500 K potential temperature level. Grey lines correspond to the position of the cuts or borders displayed in the other panels. Figure c, d show the Φe / θ cross sections for two selected days, 20 June and 20 July.

Figure  shows the corresponding results from the CLaMS simulation. The top panels of Figs.  and agree well with respect to the time at which the HCl depletion starts at the different altitudes and latitudes. However, it appears that after the first “titration” step, HCl depletes much more slowly in the simulations compared to the MLS observations. From the comparison of Figs.  and , it is evident that the model HCl discrepancy is present over a wide altitude range throughout polar winter.

Figure  shows the time series of the similarly averaged HCl mixing ratio for the Arctic winter 2015/2016, both for the MLS observations (Fig. a, b) and the CLaMS simulation (Fig. c, d). This winter was chosen because it was particularly cold in the stratosphere . Here, the HCl discrepancy is also apparent, although to a lesser extent than in the Antarctic. It is most evident at the slightly lower potential temperature of about 475 K in the Arctic vortex core. Compared with the observations, the model requires over 1 month more, until February, to reach the minimum HCl mixing ratios. While the discrepancy is therefore also present in the Arctic, in the following we focus on the Antarctic results where it is much more strongly pronounced.

For further comparison, we now show results for the Antarctic winter 2011 simulated by the Eulerian models SD-WACCM and TOMCAT/SLIMCAT. Figure shows the simulated HCl mixing ratios from these models plotted identically as in Fig. . The HCl discrepancy is clearly also evident in these simulations, as on 20 July, by when HCl had disappeared in the MLS data; both models also show an area of elevated HCl still present in the polar vortex core, although with lower HCl mixing ratios compared to CLaMS. The HCl depletion in the vortex core is stronger in the TOMCAT/SLIMCAT simulation than in SD-WACCM but still not comparable to the MLS observations. For a more detailed comparison, the results of SD-WACCM and TOMCAT/SLIMCAT have been averaged in Φe / θ space in a similar manner to CLaMS and the observations. The WACCM model daily output interpolated to potential temperature levels was averaged using potential vorticity and equivalent latitude calculated from ERA-Interim data. The TOMCAT/SLIMCAT model output was directly averaged within the internal calculated equivalent latitude, also based on ERA-Interim data. A depiction corresponding to Fig.  but for SD-WACCM and TOMCAT/SLIMCAT is shown as Figs. S1–S3 in the Supplement. Figure  shows a time series of the vortex-core-average HCl and ClONO2 simulated by the three models. The green line corresponds to the MLS and MIPAS observations for HCl and ClONO2, respectively. However, only very limited ClONO2 observations are available from June to August, since the presence of PSCs impedes the MIPAS retrieval process.

There are differences between the individual simulations that arise due to different model formulation and initialisation and that are discussed below. Generally, the Eulerian models SD-WACCM and TOMCAT/SLIMCAT show a faster HCl depletion than CLaMS. However, it seems that all models underestimate the rate of HCl depletion after the titration step. The three simulations presented show that the initial titration step is completed by the end of May and no further strong changes of HCl occur until early July. Differences during May are likely caused by different initial chlorine partitioning. In July and August, HCl decreases further, CLaMS being the model that indicates the slowest HCl decrease. There is even an apparent HCl increase in CLaMS that is caused by the descent of air masses with higher inorganic chlorine Cly.

One possible reason for the inter-model difference could be differences in the chemistry formulation of the different models. To investigate this hypothesis, box model versions of TOMCAT/SLIMCAT and CLaMS were integrated along an example trajectory. This trajectory was chosen from an ensemble of 1800 vortex trajectories of 2.5-month duration, along which the ozone and HCl development was closest to the vortex-core average shown in Fig. . This trajectory should therefore be representative for the vortex-core average ozone and HCl development in the 3-D model. For this example trajectory, no significant differences were found between the models. Both models similarly showed no further depletion of HCl after the titration step in the dark unmixed polar night conditions (see Fig. S4 of the Supplement). Thus, the differences among the models are unlikely due to model differences in the formulation of the chemical scheme.

A large part of the inter-model difference may be caused by the fact that the Lagrangian formulation of the model CLaMS minimises numerical diffusion while it is inherent in the Eulerian models SD-WACCM and TOMCAT/SLIMCAT. The numerical diffusion arises in Eulerian models because on each time step the chemical tracers are typically averaged within the extent of a model grid box. The minimisation of the numerical diffusion from this lower-resolution limit depends on the performance of the advection scheme; for example, the scheme used in TOMCAT/SLIMCAT stores subgrid-scale distributions of the tracers. In practice, all Eulerian models likely aim to have tracer advection schemes which limit numerical diffusion, although it will always occur to some extent. The artificial mixing by numerical diffusion could provide additional ClONO2 or HOCl as a reaction partner for HCl or NOx that would form ClONO2. To investigate this hypothesis, a sensitivity simulation by CLaMS was performed, in which a regular Eulerian grid with 2.8∘ × 2.8∘ horizontal spacing and the vertical model grid (with the vertical resolution of 800–1000 m in the lower stratosphere) as vertical levels were defined. Every 24 h of the model run it was checked whether more than one air parcel resided in the same grid box. These air parcels were then set to the average mixing ratio of all air parcels residing in the grid box. Although a Eulerian model that must average over these grid boxes on every time step would be even more diffusive, this study can at least show a lower limit of this effect. The corresponding HCl result (labelled “mix-Euler”) is shown as the orange line in Fig. . Indeed, the additional imposed Eulerian mixing causes a faster HCl depletion until mid-August. Although it is difficult to mimic additional numerical diffusion in a Lagrangian model, this sensitivity supports the assumption that at least part of the difference between CLaMS and the other Eulerian models may be caused by numerical diffusion. Thus, it is likely that the numerical diffusion in the Eulerian models masks part of the effect responsible for the HCl discrepancy investigated in this study. However, it should be noted there is not much difference in the development of HCl between the two resolutions of TOMCAT/SLIMCAT. In the low-resolution simulation, the gradients of ozone and other trace species, especially at the vortex edge, are weaker, but the time series of vortex-core averages are very similar. Potentially, in the high-resolution simulation, the numerical diffusion may already be significant.

In the following, we list potential model deficiencies that could be responsible for the HCl discrepancy.

If the initial ClONO2 / HCl ratio was not set to realistic values, this would have immediate impacts on the first titration step of the reaction HCl + ClONO2. This should have been avoided at least in the CLaMS simulations, as both ClONO2 and HCl have been initialised from observations.

To find out more about the characteristics of the potentially missing process that could explain the difference between simulated and observed HCl, we focus on the comparison between simulations and observations for the vortex-core data points on the 500 K level for 20 June (data from about 300 MLS profiles). This is a time and location when, in both the model and the observations, the HCl depletion after the initial titration step is ongoing. From that, we can get hints about the possible missing process (or processes).

First, we investigate whether the missing process is correlated with temperature. This would be the case if the depletion of HCl is caused by uptake on PSC particles, for example, into the available liquid PSC particles as suggested by the empirical correction in . Similar to HNO3, HCl is soluble in liquid aerosols. The solubility of HCl in the liquid aerosols given by the parameterisation of is included in the CLaMS chemistry module. Although the simulations do not show any significant HCl uptake, it may be that underlying parameters like Henry's law constant are not accurately represented in the parameterisation as suggested by . The uptake of HCl into the liquid aerosol particles should be strongly temperature dependent , with the largest expected effect for the lowest temperatures. To examine such a hypothesis, the chosen subset of MLS data points and corresponding CLaMS points are plotted as function of temperature in Fig. . The temperature for all data points in this subset is below 194 K, suggesting that PSCs should exist at the observation locations shown, which is supported by the MIPAS PSC database . The simulations and observations in Fig.  suggest that the first titration step is completed by 20 June. The temperature dependence of the difference between observations and the simulation does not, however, suggest a missing uptake of HCl into the particles. The largest discrepancies between simulations and observations are found for higher temperatures. Therefore, the uptake of a significant fraction of HCl into the PSC particles seems to be an unlikely explanation for the discrepancy shown between Figs.  and . However, it cannot be excluded that this dependency is caused by a combination of more than one unknown process.

Although the HCl depletion is not directly correlated with temperature, it seems likely that the missing process should require temperatures low enough for heterogeneous chlorine activation on PSCs or cold aerosols. We examine whether the missing process requires sunlight by investigating the history of the same observed air masses using 30-day back trajectories calculated by the CLaMS trajectory module. From the trajectories, we determined how long the air mass experienced sunlight and also how long they were exposed to temperatures below 195 K, a typical value below which PSCs can form and heterogeneous chemistry becomes important. Here, we used solar zenith angles below 95∘ for the definition of sunlight time .

Figure  shows the individual observed HCl mixing ratios as a function of the time spent in sunlight and below 195 K as green symbols. Red symbols show the corresponding simulated HCl mixing ratios. There seems to be a clear anticorrelation between the time spent in sunlight below 195 K and HCl mixing ratios. This suggests that the missing process may both require sunlight and low temperatures in the presence of PSCs.

To further clarify the temporal development of the HCl discrepancy, Fig.  shows also the rate of HCl change dHCl/dt calculated from the time series of HCl within the grid box in the equivalent latitude/potential temperature space for each day (Fig. 9b). It is evident that the first titration step caused by the heterogeneous reaction of HCl with ClONO2 is reproduced correctly in the simulation. After that step, HCl decreases particularly strongly during two periods in mid-June and mid-July. These periods are correlated with a high occurrence fraction of PSCs as shown in Fig. c. Both the HCl depletion rate and PSC occurrence seem to have relative maxima around 11 June (i.e. large rate of HCl depletion), minima around 23 June and again maxima between 5 and 12 July. The PSC occurrence fraction is shown for both MIPAS and CALIOP observations, which are broadly consistent for the overlapping observation times. The corresponding CLaMS PSC occurrence fractions are also shown. The only significant difference between MIPAS and CALIOP is in late May where MIPAS detects a large fraction of potentially small STS particles that could not be detected by CALIOP. Also in the beginning, these STS detections are pole-centred and therefore partly missed by the limited latitudinal coverage of the CALIPSO satellite. The anticorrelation between dHCl/dt and the PSC occurrence fraction is a further indication that PSCs may be relevant here. Possible processes missing in the formulation of the model that could explain the HCl discrepancy are investigated in the following section.

The required missing process (or processes) should clearly improve the characteristic difference between observed and simulated HCl mixing ratios (Figs.  and ). By incorporating different processes into a model, we can investigate the “fingerprint” of the corresponding change to the model results. This method will not provide a proof of whether a hypothetical process occurs in reality. However, it can help to exclude hypothetical processes when they cause different “fingerprints” that can be contrasted with observations. In the following subsections, different hypotheses for the missing processes are investigated by incorporation into the model CLaMS.

We now investigate the impact of HCl processing on ozone, since the amount of active chlorine in the polar stratosphere is essential for the determination of chemical ozone loss. All three participating models do simulate the chemical ozone loss and the formation of the ozone hole sufficiently well. Figure  shows the simulated vortex-average ozone profile on 1 October for the different models in combination with the passive ozone tracer from CLaMS. The difference between the two profiles corresponds to the chemical ozone loss over the winter until this time. The potential process that was included in the NAT decomp simulation increases chlorine activation significantly, especially in the polar winter. This might also affect the simulated ozone depletion. However, the difference in chlorine activation is especially large in the rather dark vortex core, where chemical ozone depletion rates are small.

The impact of the hypothetical NAT decomposition processes on polar ozone depletion would cause an additional 0.2 ppmv vortex-average ozone depletion on the 500 K level as shown in Fig. . Figure  shows the simulated chemical ozone depletion for the three CLaMS simulations, the reference simulation, the simulation including GCR-induced ionisation and the simulation also including the hypothetical NAT decomposition. The ozone depletion is also shown for the simulation for the Arctic winter 2015/2016 employing the same model setup. Although it is clear that an increase in chlorine activation yields more ozone depletion, the overall effect on ozone depletion of this additional hypothetical process is rather low. The increase of column ozone loss in the vortex core due to GCR-induced ionisation ranges from 2 to 3 %. The additional column ozone loss due to the hypothetical NAT decomposition is calculated to be about 1.8 % in the Antarctic winter and only 0.6 % in the Arctic winter.

We have demonstrated that an important polar wintertime process which leads to the removal of HCl is missing in the standard formulation of stratospheric chemistry models. The discrepancy between observed and simulated abundances of HCl is apparent in the dark stratospheric polar vortex cores in winter. In model simulations, there is little change in HCl after the first titration step through the heterogeneous reaction of HCl with ClONO2. In contrast, the satellite observations show a continuous removal of HCl from the gas phase even in the dark core of the polar vortex. We show that this HCl discrepancy is present in three models with different formulations and setups, and it is therefore likely not a problem of a single model code. The discrepancy is most clearly demonstrated in the Lagrangian CLaMS model, while it is somewhat masked in Eulerian global models likely due to the inherent numerical diffusion of Eulerian transport schemes.

A detailed investigation of the characteristics of the HCl discrepancy points to a missing process that likely involves PSCs. Several hypotheses of the cause of the discrepancy have been investigated. The incorporation of GCR-induced ionisation reduces the discrepancy somewhat but is by no means sufficient. The uptake of HCl into PSC particles and photolysis of particulate HNO3 also cannot explain the HCl discrepancy. Simulations with a hypothetical steady decomposition of HNO3 out of NAT PSCs would fit the requirements to resolve the HCl discrepancy. This may be caused indirectly by GCRs. At least the observation of elevated ClONO2 observations in late August supports the presence of an additional NOx source.

However, at present, the specific process responsible for the HCl decomposition remains unclear. Since the discrepancy occurs during the beginning of the chlorine activation period in winter where the ozone loss rates are slow, there is only a minor impact on the overall ozone loss in polar spring.