Mountain-wave-induced polar stratospheric clouds and their representation in the global chemistry model ICON-ART

Polar stratospheric clouds (PSCs) are a driver for ozone depletion in the lower polar stratosphere. They provide surface for heterogeneous reactions activating chlorine and bromine reservoir species during the polar night. The large-scale effects of PSCs are represented by means of parameterisations in current global chemistry–climate models, but one process is still a challenge: the representation of PSCs formed locally in conjunction with unresolved mountain waves. In this study, we investigate direct simulations of PSCs formed by mountain waves with the ICOsahedral Nonhydrostatic modelling framework (ICON) with its extension for Aerosols and Reactive Trace gases (ART) including local grid refinements (nesting) with two-way interaction. Here, the nesting is set up around the Antarctic Peninsula, which is a well-known hot spot for the generation of mountain waves in the Southern Hemisphere. We compare our model results with satellite measurements of PSCs from the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) and gravity wave observations of the Atmospheric Infrared Sounder (AIRS). For a mountain wave event from 19 to 29 July 2008 we find similar structures of PSCs as well as a fairly realistic development of the mountain wave between the satellite data and the ICON-ART simulations in the Antarctic Peninsula nest. We compare a global simulation without nesting with the nested configuration to show the benefits of adding the nesting. Although the mountain waves cannot be resolved explicitly at the global resolution used (about 160 km), their effect from the nested regions (about 80 and 40 km) on the global domain is represented. Thus, we show in this study that the ICON-ART model has the potential to bridge the gap between directly resolved mountainwave-induced PSCs and their representation and effect on chemistry at coarse global resolutions.

(nesting) with two-way interaction. Here, the nesting is set up around the Antarctic Peninsula which is a well-known hot spot for the generation of mountain waves in the southern hemisphere. We compare our model results with satellite measurements from the Cloud-Aerosol LIdar with Orthogonal Polarisation (CALIOP) and the Atmospheric InfraRed Sounder (AIRS). We study a mountain wave event that took place from 19 to 29 July 2008 and find similar structures of PSCs as well as a fairly realistic development of the mountain wave in the Antarctic Peninsula nest. We compare a global simulation without nesting with 10 the nested configuration to show the benefit. Although the mountain waves cannot be resolved adequately in the used global resolution (about 160 km), their effect from the nested regions (about 80 and 40 km) on the global domain is represented. Thus, we show in this study that by using the two-way nesting technique the gap between directly resolved mountain-wave induced PSCs and their representation and effect on chemistry in coarse global resolutions can be bridged by the ICON-ART model.

15
Polar stratospheric clouds (PSCs) play a key role in explaining the ozone loss in the polar stratosphere during local spring (e.g., Solomon et al., 1986;Solomon, 1999;Braesicke et al., 2018). Heterogeneous reactions on the surface of PSCs lead to activation of chlorine and bromine species during the polar night, thus enhancing the catalytic ozone depletion cycles as soon as the sun rises (e.g., Solomon, 1999). In addition, PSCs can irreversibly remove nitrogen-containing species by sedimentation, today known as denitrification, thus extending the period of low ozone concentrations during polar spring (e.g., Waibel et al.,20 1999).
One of the processes under investigation is PSCs formed in conjunction with mountain waves. It is known that mountainwave induced PSCs have a significant influence on the ozone depletion over Antarctica and Arctic (e.g., McDonald et al., 2009;Alexander et al., 2011). However, the simulation of mountain-wave induced PSCs with atmospheric chemistry models (ACMs) is still a challenge (e.g., Orr et al., 2015). With horizontal resolutions in the order of a few hundreds of kilometres, ACMs are 25 not able to resolve the orography adequately to directly simulate mountain waves (Morgenstern et al., 2010). Thus, mountain waves and mountain-wave induced PSCs either have to be parametrised (Orr et al., 2015;Zhu et al., 2017;Orr et al., 2020) or have to be calculated as post-processing via Lagrangian models (e.g., Mann et al., 2005) or higher resolved mesoscale models (e.g., Fueglistaler et al., 2003;Eckermann et al., 2006), which have to be limited to a specific region on Earth as discussed e.g.
in Weimer et al. (2016). A method to bridge this gap for interactive calculations is missing so far. 30 Three types of PSCs are known to exist: (1) solid nitric acid trihydrate (NAT) particles, mostly responsible for denitrification (e.g., Fahey et al., 2001), (2) liquid supercooled ternary solution (STS) droplets which grow from the background sulfate aerosol (Junge et al., 1961;Wilka et al., 2018) by taking up nitric acid (HNO 3 ) and on which the major fraction of chlorine and bromine activation takes place (e.g., Peter, 1997;Kirner et al., 2015) and (3) ice clouds forming at the lowest temperatures and leading to dehydration of the lower stratosphere (e.g., Kelly et al., 1989;Khaykin et al., 2013). 35 Various climatologies of PSCs based on satellite measurements were published in the last years Höpfner et al., 2018;Pitts et al., 2018). The application of machine learning for the detection of PSCs from satellite measurements has been discussed recently (Sedona et al., 2020).
Nucleation processes especially of NAT and ice PSCs are still under debate (e.g., Voigt et al., 2018). The question if equilibrium or non-equilibrium processes are needed to describe the growth of STS particles is an issue of research (Zhu et al., 2015). 40 In addition, it is also known that mountain waves can induce PSCs which influence the ozone chemistry (e.g., Hoffmann et al., 2017).
Mountain waves (orographic gravity waves) are stationary waves in the lee of a mountain which can develop in a stably stratified atmosphere (e.g., Fritts and Alexander, 2003). They can propagate upwards into the stratosphere and higher (Wright et al., 2017) and perturb the synoptic temperature field with a local amplitude in the order of up to ±15 K (Meilinger et al., 45 1995;Carslaw et al., 1998;Eckermann et al., 2009).
Apart from heterogenenous nucleation of NAT on meteoric dust (e.g., Hoyle et al., 2013;Tritscher et al., 2019), denitrification in the Arctic region is closely connected to mountain wave activity because large NAT particles can be formed in the low temperatures associated with mountain waves in the northern hemisphere (e.g., Tabazadeh et al., 2000). On the other hand, mountain-wave induced PSCs are also important for the southern hemisphere. Antarctic mountain-wave induced NAT PSCs 50 were detected by Höpfner et al. (2006) for the first time. McDonald et al. (2009) stated that up to 40 % of PSC formation comes from mountain waves in the early Antarctic winter when temperature is close to that of NAT formation. Alexander et al. (2011) concluded that about 30 % of all southern hemispheric PSCs can be related to mountain waves.
In this study, mountain-wave induced PSCs are simulated seamlessly with the ICOsahedral Nonhydrostatic modelling framework (ICON, Zängl et al., 2015) and its extension for Aerosols and Reactive Trace gases (ART, Rieger et al., 2015;Weimer 55 et al., 2017;Schröter et al., 2018). ICON-ART provides the possibility of local grid refinement with two-way interaction so that a coarse global resolution can be complemented by a region with a refined grid (Reinert et al., 2019), such as the Antarctic Peninsula which is one of the hot spots for the generation of southern hemispheric mountain waves (Bacmeister, 1993;Bacmeister et al., 1994;McDonald et al., 2009;Alexander et al., 2011;Hoffmann et al., 2017). Therefore, mountain-wave induced PSCs can be directly simulated with ICON-ART in the refined grid (nest) and their effect can be treated at the global 60 resolution in the same simulation where they cannot be resolved without the nest.
The ICON-ART model is described in Sect. 2, followed by a description of the simulation in Sect. 3 and the satellite data used for comparison with ICON-ART in Sect. 4. The model results are compared with CALIOP and AIRS measurements and mountain-wave induced PSCs are investigated for a typical mountain wave event in Sect. 5. Finally, conclusions and outlook follow in Sect. 6.

2 The ICON-ART model
The ICON model is the operational model for numerical weather prediction at the German Weather Service (DWD, Zängl et al., 2015). In addition, it can be applied to large eddy simulations (Dipankar et al., 2015) and fully coupled with an ocean and a land surface model for climate integrations with the climate physics configuration (Giorgetta et al., 2018). The ART extension has been developed to incorporate aerosols and the atmospheric chemistry into ICON. It can be coupled to ICON in 70 configurations for numerical weather prediction (Rieger et al., 2015) and allows flexible configurations for weather and climate integrations (Schröter et al., 2018).
In this study, the chemistry in ICON-ART is based on the Module Efficiently Calculating the Chemistry of the Atmosphere (MECCA, Sander et al., 2011a) which uses the Kinetic PreProcessor (KPP) to generate Fortran files for solving the specified chemical mechanism (Sandu and Sander, 2006). Photolysis rates are calculated with the CloudJ module (Prather, 2015;Weimer 75 et al., 2017). In this study, a system of 142 chemical reactions including 38 photolytic reactions and 11 heterogeneous reactions on the surface of PSCs is used which is similar to other studies (e.g., Stone et al., 2019;Zambri et al., 2019;Nakajima et al., 2020) and can be found in Appendix A. Trace gas emissions at the Earth's surface are included by a module described in Weimer et al. (2017).
The model equations of ICON-ART are discretised horizontally on an icosahedral-triangular C grid (e.g., Staniforth and 80 Thuburn, 2012;Zängl et al., 2015). The global resolution can be refined by root divisions and bisections of the original icosahedron, resulting in the horizontal resolution description RnBk, as defined by e.g. Zängl et al. (2015). Vertical discretisation is performed on generalised smooth-level coordinates (Leuenberger et al., 2010).
For the purpose of detailed simulations around a specific region, the grid can be refined for the area of interest by further bisections. Here, the parent domain provides boundary conditions for the nested domain. The simulated values in the nested 85 domain are interpolated to the parent grid with a relaxation-based method (Reinert et al., 2019). Thus, the two-way nesting enables the model to simulate e.g. mountain waves directly at hot spots and also to impact the variables at a global resolution that is too coarse to represent these processes.

The PSC scheme
In ICON-ART, the three types of PSCs (ice, NAT and STS) are treated separately. Sensitivity simulations showed that the ICON 90 microphysics for ice clouds, operationally computed up to an altitude of 22.5 km, can be extended to the lower stratosphere up to 30 km (Weimer, 2019). The hydrometeor microphysics in ICON include several nucleation processes interacting with the other hydrometeors as well as sedimentation of the ice particles (Doms et al., 2011). Thus, both nucleation of ice PSCs and dehydration of the lower stratosphere can be captured by the ICON microphysics. A similar approach is used in the Whole Atmosphere Community Climate Model (WACCM, Zhu et al., 2015). Ice particle radius (r ice in m) and particle number 95 concentration (N ice in m −3 ) correspond to the assumed size distributions in the microphysics (Doms et al., 2011): (1) where ρ is the air density in kg m −3 , q ice is the mass mixing ratio in kg kg −1 of water in ice.
Two parametrisations for the microphysics of NAT particles are integrated in ICON-ART. The thermodynamic NAT parametri-100 sation is diagnostic and therefore computes the number of moles of NAT particles in thermodynamic equilibrium, calculates its sedimentation and evaporates the particles again within the same model time step. The volume mixing ratio of HNO 3 condensed in NAT (X HNO3(NAT) in mol mol −1 ) is calculated on the basis of the difference between vapour pressure of HNO 3 (p HNO3 in Pa) and the saturation vapour pressure over NAT (p sat,NAT in Pa): The saturation vapour pressure over NAT is calculated according to Hanson and Mauersberger (1988) and p is the air pressure (in Pa). Particle number concentration (N NAT ) and radius (r NAT ) in the thermodynamic NAT parametrisation are calculated using a threshold in the number concentration of N NAT,max = 2.3 × 10 −4 cm −3 which originates from observations (Fahey et al., 2001). Below this threshold, the radius of the NAT particles is set to 0.1 µm. Above this threshold, the particle number concentration is set to N NAT,max and the radius of the particles is increased accordingly. This method has already been used 110 in a similar way for solid particles by Kirner et al. (2011).
The kinetic NAT parametrisation is a non-equilibrium approach based on prognostic equations for the particle mass (see e.g., Seinfeld and Pandis, 2006, p. 539). Carslaw et al. (2002) applied this approach to NAT particles in a Lagrangian model and van den Broek et al. (2004) extended it to Eulerian models.
In the Lagrangian description by Carslaw et al. (2002), the change in NAT particle radius (r NAT,b ) is calculated prognosti-115 cally by a diffusive growth where G b (in m 2 s −1 ) is a growth factor which depends on the diffusion coefficient of HNO 3 in air (D HNO3 , in m 2 s −1 ), the air temperature (T , in K) and the saturation difference of the HNO 3 vapour pressure (in Pa): In these equations, v HNO3 is the mean thermal velocity of air molecules (in m s −1 ), R * stands for the universal constant of an ideal gas in J mol −1 K −1 , ρ NAT is the crystal density of NAT (1.626 × 10 6 g m −3 , Drdla et al., 1993;van den Broek et al., 2004) and M NAT is the NAT molar mass of 117 g mol −1 . In one of the Eulerian formulations by van den Broek et al. (2004), 125 this radius change is applied to particles in size bins, but directly converted into a change in the particle number concentration ("FixedRad" approach). This approach is used in this study with a size distribution based on van den Broek et al. (2004) as shown in Fig. 1. That is why we added a b subscript to all variables in Eqs. (4), (5) and (6) that depend on the size bin b.
The size distribution of NAT particles can be flexibly specified by the user to investigate its impact on denitrification without any change in the Fortran code but rather changing the respective XML control file (cf. Schröter et al., 2018). Each size bin 130 is defined by radius limits and a maximum particle number concentration, which are kept constant during the simulation. If the calculated particle number concentration in a bin exceeds the maximum, the excess mass is transferred to the next larger size bin. The sum of the maximum particle number concentrations of the bins has to equal the observed value by Fahey et al.
Since the calculation of p sat,NAT by Hanson and Mauersberger (1988) has specific temperature limits, NAT particles are 135 evaporated automatically at temperatures higher than 220 K. For temperatures below 180 K, we calculate p HNO3 with a constant temperature of 180 K. These two NAT parametrisations are also implemented in the ECHAM/MESSy Atmospheric Chemistry model (EMAC, Jöckel et al., 2010;Kirner et al., 2011).
Sedimentation of NAT particles, either formed by thermodynamic or kinetic NAT parametrisation, is calculated by a simple upwind method, using the Stokes velocity of assumed spherical particles (Stokes, 1851).

140
The microphysics of STS particles in the module are calculated by the scheme first published by Carslaw et al. (1995), with one exception: in the original code by Carslaw et al. (1995) the particle number concentration is set to the constant value of 10 cm −3 . We improved this fixed value by applying the mean of all balloon-borne STS measurements by Hervig and Deshler (1998) in order to derive the particle surface area concentration S STS and the radius r STS of STS particles from the internally calculated particle volume concentration V STS : In these equations, V STS has to be given in µm 3 cm −3 to get S STS and r STS in µm 2 cm −3 and µm, respectively. Sedimentation is neglected for STS particles since they are too small to result in relevant redistribution of the major constituents H 2 O, HNO 3 and H 2 SO 4 (Tabazadeh et al., Considine et al., 2000).

150
Since the PSCs in the current version of the model do not interact with each other like in a fully coupled PSC scheme (e.g., Zhu et al., 2015), there are essentially two approaches to calculate PSCs: either they are computed with the total (gaseous + liquid + solid) concentrations of HNO 3 and H 2 O as input for all PSC types (e.g., Kirner et al., 2011) or the PSCs are calculated subsequently with the gaseous fraction that remains after formation of the previously calculated PSC types (operator splitting).
Both approaches have their advantages and disadvantages. We use the second approach because the maximum of HNO 3 and 155 H 2 O taken up by PSCs cannot exceed the gaseous concentrations in this case. First, ice PSCs are calculated, then NAT PSCs and finally STS PSCs.
Particle radius, particle number concentration and particle surface area concentration are used to calculate the heterogeneous reaction rate constants on the surface of PSCs. For NAT and ice, which can grow to relatively large sizes in the order of tens of µm, the following equation is used to calculate the heterogeneous reaction rate constant, assuming spherical particles (Drdla 160 et al., 1993): where γ h,c is the uptake coefficient (i.e. reaction probability) of heterogeneous reaction h and PSC type c, v i(h) is the mean thermal velocity of the gaseous reactant i(h) (in m s −1 ), N c is the PSC number concentration and N j(h) is the number concentration of the reactant j(h) adsorbed on the particle (both in m −3 ). The Knudsen number Kn is calculated by λ mfp /r c 165 with the mean free path λ mfp of air molecules (in m), calculated according to Kennard (1938). In case of the kinetic NAT parametrisation, Eq. (9) is evaluated for each size bin separately and summed up subsequently. STS particles are considerably smaller than ice and NAT particles (e.g., Considine et al., 2000) and hence Kn 1 so that the slip flow correction term in Eq. (9) can be neglected, which yields: The total reaction rate constant is the sum of the reaction rate constants on all three types of PSCs. The uptake coefficients in ICON-ART are used either from Carslaw et al. (1995) or Sander et al. (2011b). A summary of the γ values can be found in Table A2 of Appendix A. They also compare well to other models (e.g., Solomon et al., 2015).

Simulation with nests around the Antarctic Peninsula
In order to compare the results of the PSC scheme in ICON-ART with satellite measurements and to investigate the impact of 175 the nesting technique on mountain-wave induced PSCs, a three-step simulation is conducted, see The chemical trace gases (tracers) are initialised by an EMAC simulation which included tropospheric as well as strato-185 spheric chemistry similar to Jöckel et al. (2016). NAT PSCs are simulated with the kinetic NAT parametrisation. The size distribution is shown in Fig. 1. We prescribe H 2 SO 4 in the lower stratosphere (Thomason et al., 2008;SPARC, 2013).
The emission data sets used in this study are summarised in Table 1. Emissions of CH 4 , CO, CO 2 , N 2 O, SO 2 and CFCl 3 are considered. The GEIA dataset for chlorofluorocarbons (CFCs) provided by the Emissions of atmospheric Compounds and Table 1. Emission datasets used in this study. In the first step, the simulation starts with a global resolution of R2B04 (∆x of about 160 km) as a free-running simulation 195 until 01 May 2008. This period is used as a spin-up period for the chemistry until the southern hemispheric polar vortex begins to form. The resolution of 160 km is comparable to current chemistry climate models (e.g., Morgenstern et al., 2010;Kirner et al., 2015;Zhu et al., 2015;Steiner et al., 2020). Thus, this simulation shows how mountain-wave induced PSCs can be treated in a resolution that is similar to other chemistry climate models.
In the second step, it is of importance that the dynamics of the atmosphere and especially the polar vortex are represented 200 realistically in the model, especially for the evaluation with measurement data. This is why the meteorological variables are reinitialised every second day by ERA-Interim in the period between 01 May and 18 July 2008, but the chemical tracers are free-running. This method was already introduced e.g. in Schröter et al. (2018).
In the third step, to avoid the determination of the simulations by lower resolution models and models with hydrostatic dynamical core, two different free-running simulations are conducted during the mountain wave event from 19 to 29 July

CALIOP
The Cloud-Aerosol LIdar with Orthogonal Polarisation (CALIOP, Pitts et al., 2009;Höpfner et al., 2009;Pitts et al., 2018) onboard the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) was launched on 28 April 2006 (Winker et al., 2007). In 2008, the satellite flew as part of NASA's A-train constellation in an orbit with 98 • inclination at an altitude of 705 km (Stephens et al., 2002;Pitts et al., 2018). Its orbit was sun-synchronous with equator crossings at about 215 01:30 and 13:30 LT. It measured down to 82 • S with a repeat cycle of 16 days. In February 2018, it was moved to a lower orbit.
CALIOP is a light detecting and ranging (Lidar) instrument, set up in nadir geometry. It scans the atmosphere at wavelengths of 532 and 1064 nm with parallel and orthogonal polarisations for the 532-nm channel (Winker et al., 2007). In altitudes from 8.4 to 30 km, the vertical resolution is 180 m or higher. At the Earth's surface, the light beam has a diameter of about 100 m Pitts et al., 2018).
For the detection of PSCs, a combination of two values is used : (1) the ratio of total and molecular backscatter coefficient R 532 and (2) the backscatter coefficient at perpendicular polarisation β 532 , both at a wavelength of 532 nm. Discrimination of the PSC types can be seen in diagrams of R 532 vs. β 532 . Different regions in this diagram refer to the different PSC types (see Pitts et al., 2018).
We use the PSC climatology of Pitts et al. (2018), which is the version 2 level 1B data, averaged with a window of 5 km 225 along the satellite path. It is restricted to night-time southern hemispheric data during the mountain wave event (data is available

AIRS
The Atmospheric InfraRed Sounder (AIRS, Aumann et al., 2003;Chahine et al., 2006) is one of the instruments onboard the Aqua satellite, which was launched in May 2002 and which is part of the A-train constellation right ahead of CALIPSO. Thus, its orbit is the same as CALIPSO in 2008 but about 1 to 2 min ahead of CALIPSO 1 .

235
The AIRS instrument is a nadir sounder with across-track scanning capabilities that scans the atmosphere by 90 footprints per scan with a ground coverage of 1780 km and a size of 13.5 x 13.5 km 2 (nadir) to 41 x 21.4 km 2 (scan edge) per footprint (e.g., Orr et al., 2015;Hoffmann et al., 2017). It measures the spectrally resolved radiances in wavelengths between 3.74 and 15.4 µm. Brightness temperatures in the 15 µm band can be used to derive information about gravity waves in the lower polar stratosphere (Hoffmann et al., 2017). In this study we use a data product averaging over 21 channels around 15 µm to 240 improve the signal-to-noise ratio (Hoffmann et al., 2017). The temperature weighting function in this band peaks at an altitude of around 23 km with a full width at half maximum of 15 km and with information from the altitude range between 17 and 32 km. Therefore, it is well suited to derive information about gravity waves in the altitude region where PSCs are expected to exist.
In this study, this band is used to examine the mountain wave event at the Antarctic Peninsula mentioned above. The same 245 algorithm as in previous studies is used to compare the model data with specific Antarctic Peninsula overpasses of AIRS (Hoffmann and Alexander, 2010;Hoffmann et al., 2016Hoffmann et al., , 2017. In particular, the ICON-ART data are resampled on the AIRS measurement grid and a radiative transfer model is used to simulate AIRS measurements based on the ICON-ART data to allow for a direct comparison with the real observations.  This is reflected in the variables shown in Fig. 4. Temperatures as low as 175 K only occur in the Antarctic Peninsula nest with a 40 km resolution in the lee of the mountain. In addition, it shows the characteristic high and low temperature patterns as calculated by theory (e.g., Queney, 1947) and seen by measurements (e.g., Wright et al., 2017). The temperature perturbation in the order of 10 K is consistent with previous studies of mountain waves (Meilinger et al., 1995;Carslaw et al., 1998;Eckermann 270 et al., 2009).
The horizontal wavelength of the simulated mountain wave is in the order of 240 km, which is a medium-large wavelength compared to other events (see e.g., Alexander and Teitelbaum, 2007;Plougonven et al., 2008;Hoffmann et al., 2014). In addition, the vertical resolution is in the order of 500 m at least in the lower parts of Fig. 4. This is why the mountain wave can be captured by the resolution of 40 km in the Antarctic Peninsula nest.

275
The relatively large temperature gradients in the Antarctic Peninsula nest are also in agreement with gradients in the wind velocities which are shown in the second and third row of Fig. 4. In the lee of the Antarctic Peninsula, the vertical wind velocity changes signs in altitudes where the temperature increases or decreases and has maximum values above 0.3 m s −1 .
Largest horizontal wind velocities in the order of 75 m s −1 occur in altitudes above 27 km.
The mountains also cause perturbations in the potential temperature (forth row in Fig. 4) in the lee. If diabatic processes are 280 negligible, the flow follows the contours of potential temperature which will be shown in the PSC precursors in Sect. 5.4.2.
Up to now, it was only demonstrated that mountain waves are directly simulated in the Antarctic Peninsula nest (right column) and are consistent with the theory of mountain waves. In addition, Figure 4 also shows the impact of the two-way nesting in ICON-ART. The mountain wave at the Antarctic Peninsula cannot be represented adequately in the resolution of 160 km in the simulation without the nests (left column in Fig. 4). The characteristic wave patterns do not occur in this 285 simulation. In contrast to this, the global domain in the simulation with the nests (second column in Fig. 4) shows a decrease in temperature of about 2 K in the lee of the mountains as a result of the two-way nesting and the lower temperature due to the directly simulated mountain wave in the Antarctic Peninsula nest. Of course, the amplitude is lower than it is expected by mountain waves but the effect of the mountain wave is still remarkable also in the global grid of 160 km resolution.
This is also visible in the other variables of Fig. 4. Especially for the vertical wind velocity (second row), one can see that 290 wave-like structures occur in the global domain where they cannot be represented without the nests. Therefore, we can expect that mountain-wave induced PSCs can also be represented in this relatively low global resolution because they are directly simulated in the locally refined regions. Since the nested simulation is free-running we cannot expect the PSCs to occur at exactly the same locations as in the measurements. Therefore, we compare the development of PSCs with respect to the temperature in both datasets, which is a crucial parameter in the formation of PSCs (e.g., Solomon, 1999). We count the occurrence of the different PSC types by the 315 method described above in temperature bins and compare them between ICON-ART and CALIOP.

Comparison of simulated PSCs with CALIOP measurements
The results can be found in Fig. 5. It shows the relative number of grid points (in %) on the CALIOP paths where the different PSC types occur. The "No" PSC category is obviously the dominant category for both ICON-ART and CALIOP, but neglected in this comparison to see only the occurrence of PSCs in the temperature bins.
In Figure 5, the size of the circles correspond to the total number of grid points with PSCs, that is also printed on the right 320 hand side of the panels, and the colour of the circles show the relative number of this PSC type occurrence in the respective As can be seen, ice particles dominate the existence of PSCs in the lowest temperature bins of 181 and 177.5 K for both ICON-ART and CALIOP. In addition, the "NAT-enh" category is negligible for both datasets. The development of pure STS 325 particles is similar between ICON-ART and CALIOP, although the fractions are underestimated at high temperatures. However, the number of measurements in this bin is small compared to the other ones so that this statistic could be non-significant.
Whereas NAT particles in the "NAT Mix" category exist over the whole temperature range with a similar fraction in the measurements, their existence is cut off at temperatures lower than about 183 K, the lower boundary of the 184.5 K bin. This is due to the used operator splitting and will be further investigated in Sect. 5.4.

330
In addition to the analysis of direct temperature as reference, we also analyse the development of PSCs in the same manner relative to T NAT (Fig. 6) and T ice (Fig. 7), calculated with X H2O = 5 ppmv and X HNO3 = 10 ppbv as input. We chose these constant values because the mixing ratios measured by Microwave Limb Sounder (MLS), also provided as part of the CALIOP product, only include the gaseous part of HNO 3 and H 2 O which leads to inconsistencies when knowing that PSCs exist in CALIOP. Nevertheless, this analysis includes the pressure dependence of the PSC existence temperatures so that it provides 335 more realistic thresholds for the existence temperatures than constant values like 195 and 188 K.
Although the number of grid points is relatively low for temperatures around T NAT and higher in Fig. 6, no ice PSCs exist in both simulation (panel a) and measurements (panel b). The major fraction in this temperature region is accounted to the "NAT common TNAT for both y-axes is derived from Hanson and Mauersberger (1988) with input of XH 2 O = 5 ppmv and XHNO 3 = 10 ppbv.
The data are binned in terms of temperature difference with a bin width of 1 K. Apart from that, the structure is the same as in  mix" category in both panels although PSCs usually do not exist. In total, the number of data points in ICON-ART is higher for T > T NAT and lower for T < T NAT in comparison to CALIOP. However, the principal development is similar to CALIOP: 340 at temperatures about 5 K lower than T NAT ice PSCs begin to form and turn out to be the major fraction of all PSC types. As already mentioned in the description of Fig. 5, NAT PSCs are cut off at a certain point, which will be investigated later in this study.
The upper part of Fig 7, which shows the PSC development relative to T ice , corresponds to the lower part of Fig. 6 with a shift of about 8 K. In accordance to the measurements, the fraction of ice PSCs in ICON-ART start to dominate at temperatures 345 lower than T ice with fractions larger than 50 %. The largest fractions occur at temperatures about 6 to 7 K lower than T ice in both ICON-ART and CALIOP.
Although we will show in the next sections that mountain-wave induced PSCs can be directly simulated in the Antarctic Peninsula nest, no "Wave-ice" PSCs are simulated by ICON-ART. As shown in Eq. (1), the ice particle number concentration is essentially set to the constant value of 0.25 cm −3 . The ice number concentration in mountain waves can increase to values of 350 a few cm −3 and then leading to larger backscatter ratios (e.g., Engel et al., 2014). Therefore, the backscatter ratio to determine the PSC types does not get as large as 50 which is needed for the "Wave-ice" category.
In total, we showed in this section that, apart from some differences in NAT at low temperatures and the "Wave-ice" category, the general formation of PSCs with respect to different reference temperatures is similar to the CALIOP measurements.
Therefore, we will show the development of mountain-wave induced PSCs in the next sections qualitatively but a quantitative 355 comparison is difficult because of the too large particle number concentration of ice PSCs.

Direct simulation of mountain waves compared to AIRS
For the comparison with AIRS, temperature and pressure of ICON-ART in the Antarctic Peninsula nest is saved at the time step closest to each of the AIRS overpasses during 20 and 21 July 2008. These data are then convolved with the same temperature weighting functions that apply for the AIRS observations (see e.g., Hoffmann et al., 2017). The resulting brightness temperature   c of Fig. 9). At other overpasses, the phase of the wave is shifted with respect to AIRS, as e.g. in panels c of Fig. 8 and f in Fig.   9. This is most probably a result of the free-running simulation where the wave cannot be expected to be located at exactly the same location as in the measurements.

375
In total, we have shown in this section that, apart from some missing fine structures, the mountain wave event taking place in the end of July 2008 can be represented with the resolution of 40 km in comparison with AIRS.

Impact of mountain-wave induced PSCs on the chemistry
In the previous sections, it was demonstrated that both the PSC formation and the formation of the mountain wave are in relatively good agreement with measurements considering the limits in measurements and simulation setup. In this section, wave on STS is amplified within the nests. The STS particles are assumed to freeze at temperatures 3 K below the frost point Koop et al., 2000). Thus, STS particles are only computed for higher temperatures so that STS particles are formed in the mountain wave where the temperature is higher than this threshold.

390
STS and NAT compete with each other in taking up HNO 3 . This is why in this case study distinct layers exist: NAT PSCs at altitudes higher than 20 km and STS PSCs at lower altitudes where also H 2 SO 4 is enhanced. The NAT volume concentrations are increased in the Antarctic Peninsula nest (second row, right column) which is why they are also increased in the global domain in the simulation with the nests.
In contrast to the literature, the NAT volume concentration decreases when the air masses approach the mountain wave.

395
Since the NAT size bins are advected with the general air masses, the wave-like patterns occur in both nests. As a result of the operator splitting used in ICON-ART, the largest fraction of gaseous H 2 O leads to ice formation at temperatures lower than about 180 K and is not available for NAT and STS PSCs in the mountain wave anymore. Therefore, the largest signal of mountain-wave induced PSCs can be found in ice PSCs in Fig. 10. This is an issue of further investigation in the future. In addition, NAT PSCs are formed by freezing STS particles in the mountain wave, as shown e.g. by Bertram et al. (2000) and 400 Salcedo et al. (2001). This is not integrated in the model so far and should be considered in the future.
The best example of the formation of mountain-wave induced PSCs are ice PSCs (third row in Fig. 10  might be overestimated since the ice number concentration is set to the tropospheric value of 0.25 cm −3 which is too large in comparison to measurements, as discussed previously (see also Buchholz, 2005, for an overview).

405
On the other hand, the ice PSCs are clearly connected to the regions where temperature is decreased and show similar wavelike patterns as the temperature. These increased volume concentrations are also present in the global domain where wave-like patterns can be simulated with the nests in contrast to the simulation without the nests where these structures do not exist.
Therefore, this figure shows that mountain-wave induced PSCs can be directly simulated with ICON-ART and their effect can also be treated in the global domain where mountain-wave induced PSCs cannot be represented without the nests.

Influence of the mountain wave on PSC precursors
As mentioned above, long-lived tracers closely follow the potential temperature. Therefore, the wave perturbations in potential temperature can be seen in all tracers shown in Fig. 11 in the lee of the mountain for the simulation with the nests. The wave-like structures in sulfuric acid (H 2 SO 4 , first row), nitric acid (HNO 3 , second row) and water vapour (H 2 O, third row) correspond to the structures seen in the potential temperature (see Fig. 4).  Sulfuric acid (H 2 SO 4 ) which is prescribed as a climatology in the global domain, but free-running in the nests, only shows a minor signal of the two-way nesting. Due to the missing sink of H 2 SO 4 by sedimentation of aerosols (see Sect. 3), the mixing ratio accumulates in the nested domains (two right columns).
Nitric acid (HNO 3 ), shown in the second row of Fig. 11, is taken up by STS and NAT PSCs so that it is lower in the simulation with nests than in the simulation without the nests. This is shown not only in the nested domains, but can also be 420 returned to the global domain as a result of the two-way nesting.
This property especially occurs for water vapour (H 2 O) where in the lee at altitudes between 22 and 25 km with temperatures lower than about 179 K volume mixing ratios lower than 1 ppmv occur that are connected with uptake in ice PSCs (cf. Fig.   10). It cannot be represented in the simulation without the nests (left column) where the H 2 O volume mixing ratio does not decrease to values lower than 2.5 ppmv. In the global domain of the simulation with the nests, however, values as low as 425 2 ppmv occur in the lee of the mountain as a result of the two-way nesting.

Impact of mountain-wave induced polar stratospheric clouds on chlorine activation
The directly simulated mountain-wave induced PSCs in the simulation with the nests also affect the chemistry due to increased heterogeneous reactions on their surface. The chlorine species are summarised in Fig. 12 in the same way as in the previous figures. The reservoir species HCl (first row) and ClONO 2 (second row) are shown together with the active chlorine species 430 (third row), summarised as ClO x : During July, which is a relatively late stage of the southern polar winter, most of the chlorine species in the altitude range between 15 and 25 km have been already activated. The broad band of volume mixing ratios of ClO x with values up to about 2.2 ppbv that are present in all panels of the third row suggests this. Therefore, additional chlorine activation can only be 435 expected in altitude regions above 25 km. In the previous sections, it was shown that the directly simulated mountain-wave induced ice PSCs reaches altitudes of about 26 km (see Fig. 10). Therefore, we focus on these altitudes in the following analyses.
The mountain-wave induced PSCs lead to increased chlorine activation in the lee of the Antarctic Peninsula that is not present in the simulation without the nests. At altitudes around 26 km, the additional ice PSCs activate both ClONO 2 and HCl 440 in the lee of the mountains. This is shown by values of the reservoir species around zero and values of ClO x up to 3 ppbv in this region that cannot be simulated without the nests.
A summary of the impact of the two-way nesting on mountain-wave induced PSCs and the chemistry can be found in Fig.   13 where differences of the two global domains between with and without the nests are shown for various variables of the Altogether, it was shown so far that with the local grid refinement around the Antarctic Peninsula mountain-wave induced 455 PSCs can be represented in the global domain where they cannot be simulated without the nests. These PSCs lead to increased chlorine activation in the model.

Impact of mountain-wave induced polar stratospheric clouds on ozone
Due the mountain wave event in July 2008 above the Antarctic Peninsula, chlorine activation is increased in the simulation with the nests compared to the simulation without the nests. The impact on ozone cannot be expected to be present at the Antarctic

460
Peninsula itself since the activated chlorine species are transported during the polar night within the polar vortex. Therefore, ozone is affected by the mountain wave event (1) far in the lee of the mountain and (2) later in the year. This is why the time series of the southern hemispheric mean total ozone column for the whole third period of the simulation is shown in Fig. 14.
The ozone columns are averaged for latitudes south of 60 • S. Two lines are shown in panel a: the solid line is the mean total ozone column in the global domain of the simulation with the nests, whereas the dashed line illustrates the simulation without 465 the nests.
As can be seen, differences in the simulations are below 1 DU during the whole day of 19 July (see also panel b). This first day is a spin-up period for the nests that are initialised with the global domain to form the mountain wave. From 20 July until end of the simulation, however, the mean ozone column in the simulation with the nests is smaller than in the simulation without the nests. While ozone is generally decreasing during this period in both simulations, the absolute value of the difference peaks 470 at around 8 DU on 25 July and decreases afterwards. The higher resolution around the Antarctica and the Antarctic Peninsula seems to lead to generally lower ozone columns. This additionally highlights the need of higher resolutions in atmospheric chemistry models. Figure 14 demonstrates that the higher load of active chlorine species due to the directly and interactively formed mountainwave induced PSCs lead to a by up to 8 DU larger decrease of ozone in the southern hemisphere. The prediction skill after 475 10 days is fairly low so that an analysis after the shown period is not possible with a free-running simulation. Future investigations could extend this period with specified dynamics in both simulations to be able to examine the impact of mountain-wave induced PSCs on the ozone hole during September and October.

Conclusions and outlook
Seamless modelling of chemistry-climate interactions is challenging. Not many modelling systems can do this in a consistent 480 way. In the past, it was impossible to directly simulate mountain-wave induced polar stratospheric clouds (PSCs) in global chemistry models due to the coarse resolution in that kind of simulations. In this study, we investigated this problem with the scheme for PSCs in the ICOsahedral Non-hydrostatic modelling framework with its extension for Aerosols and Reactive Trace gases (ICON-ART). The scheme forms ice PSCs based on the microphysics of the meteorological model, STS particles by the analytic expression of Carslaw et al. (1995) with some improvements with respect to the constant particle number concentration 485 and NAT particles by a kinetic non-equilibrium approach with a flexibly selectable size distribution.
We performed a three-step simulation to investigate the impact of mountain-wave induced PSCs in ICON-ART on the chemistry: First, a free-running simulation was conducted from 01 March to 30 April 2008. Second, the dynamics were reinitialised every second day by ERA-Interim data to ensure a realistic development of the polar vortex until 18 July 2008. Third, two free-running simulations followed: a simulation including two-way nesting with nests around the Antarctica (∆x ≈ 80 km) The investigated mountain wave had a horizontal wavelength of about 240 km and is therefore a medium large mountain wave that could be captured by the chosen resolution. For mountain waves with smaller wavelengths, an even higher resolution would 505 be needed to resolve the wave adequately. In addition, even though we had chosen a high resolution, compared to standard climate simulations, we still observed structures that could not be resolved. This underlines the need for high-resolution climate simulations for a suitable physical representation of mountain waves.
By introducing the two-way nesting around the Antarctic Peninsula with resolutions down to 40 km, we were able to directly simulate the main features of the mountain wave and transfer its effect back to the global domain for a typical event on 19 to 29 510 July 2008. Thus, additional mountain-wave induced PSCs, formed also in the resolution of 160 km, lead to enhanced chlorine activation and with that to an up to 8 DU larger ozone depletion above the Antarctic Continent.
Thus, this demonstrates that (1) dynamics, tracers, PSCs and chemistry are interactively and consistently integrated in ICON-ART and that (2) the gap between direct simulations of mountain-wave induced PSCs and their treatment in coarse global resolutions can be bridged by ICON-ART.

515
Future simulations will exploit the nesting technique further and either use it for other known mountain wave hot spots (e.g., Hoffmann et al., 2013Hoffmann et al., , 2017 or longer periods. In addition, the northern hemisphere will be analysed in the future where PSC formation and ozone depletion has been shown to be highly sensitive to the existence of mountain waves (e.g., Tabazadeh et al., 2000;Eckermann et al., 2006;Dörnbrack et al., 2012;Khosrawi et al., 2018). where the reaction rate constants are calculated by the PSC scheme of ICON-ART. Finally, the 38 photolytic reactions are summarised in Table A3. Photolysis rates are calculated by the CloudJ module (Prather, 2015).     HOBr + HCl → BrCl + H2O (a) 0.1 (b) 0.3 (a) Parametrised according to Carslaw et al. (1995).