Evaluation of polar stratospheric clouds in the global chemistry-climate model SOCOLv3.1 by comparison with CALIPSO spaceborne lidar measurements

. Polar Stratospheric Clouds (PSCs) contribute to catalytic ozone destruction by providing surfaces for the conversion of inert chlorine species into active forms and by denitriﬁcation. The latter describes the removal of HNO 3 from the strato-sphere by sedimenting PSC particles, which hinders chlorine deactivation by the formation of reservoir species. Therefore, an accurate representation of PSCs in chemistry-climate models (CCMs) is of great importance to correctly simulate polar ozone concentrations. Here, we evaluate PSCs as simulated by the CCM SOCOLv3.1 for the Antarctic winters 2006, 2007 and 5 2010 by comparison with backscatter measurements by CALIOP onboard the CALIPSO satellite. The year 2007 represents a typical Antarctic winter, while 2006 and 2010 are characterised by above-and below-average PSC occurrence. The model considers supercooled ternary solution (STS) droplets, nitric acid trihydrate (NAT) particles, water ice particles, and mixtures thereof. PSCs are parameterized in terms of temperature and partial pressures of HNO 3 and H 2 O , assuming equilibrium be-tween gas and particulate phase. The PSC scheme involves a set of prescribed microphysical parameters, namely ice number 10 density, NAT particle radius and maximum NAT number density. In this study, we test and optimize the parameter settings by several sensitivity simulations. The choice of the value for the ice number density affects simulated optical properties and dehydration, while modifying the NAT parameters impacts stratospheric composition via HNO 3 -uptake and denitriﬁcation. Depending on the NAT-parameters, reasonable denitriﬁcation can be modeled. However, its impact on ozone loss is minor.


Introduction
Although the occurrence of clouds in the wintertime polar stratosphere has been observed for a long time, their importance for 25 stratospheric ozone depletion was only recognized after the discovery of the Antarctic ozone hole in the mid 1980s (Farman et al., 1985). Stratospheric clouds composed of supercooled ternary solutions (STS, H 2 SO 4 -HNO 3 -H 2 O mixtures), crystalline nitric acid trihydrate (NAT) and water ice provide surfaces, on which inert reservoir species like HCl and ClONO 2 are transformed into active forms (Solomon et al., 1986). The activated species then are responsible for springtime ozone depletion induced by catalytic cycles (Molina and Molina, 1987). While STS droplets are responsible for most of the chlorine activation temperature bias. WACCM-CCMI (Garcia et al., 2017), where the cold bias was reduced by introducing additional mechanical forcing of the circulation via parametrized gravity waves, compared best with observations.
In this study, we compare a simple equilibrium scheme of STS, NAT, ice and mixtures thereof with state-of-the-art PSC satellite data, aiming to optimize the scheme for economic and efficient use in a chemistry-climate model (CCM). To this end, 95 we evaluate the representation of PSCs in the CCM SOCOLv3.1 for the Antarctic winter 2007. We convert the simulated PSCs into an optical signal to mimick a satellite measurement and compare the results with CALIPSO observations. We further evaluate the impacts of the simulated PSCs on the chemical composition of the stratosphere by comparison with satellite observations of HNO 3 , H 2 O and O 3 . A more detailed description of our methodology and the datasets utilized is given in Sect. 2. In Sect. 3 we present the results of the comparison, and Sect. 4 provides conclusions. The state-of-the-art chemistry-climate model SOCOLv3.1 (Stenke et al., 2013;Revell et al., 2015) is based on the middleatmosphere general circulation model (GCM) MA-ECHAM5 (European Centre/HAMburg climate model; Roeckner et al., 2006), coupled to the chemistry module MEZON (Model for Evaluation of oZONe trends; Egorova et al., 2003). MEZON 105 contains 57 chemical species, 56 photolysis reactions, 184 gas-phase reactions and 16 heterogeneous reactions in and on aqueous sulfuric acid aerosols as well as three types of PSCs, namely STS droplets, NAT and water ice. Heterogeneous hydrolysis of N 2 O 5 on tropospheric aerosols is as well taken into account. The chemistry module MEZON covers stratospheric ozone chemistry in detail as well as the tropospheric background chemistry, including the oxidation of isoprene (Pöschl et al., 2000).
The coupling between the GCM and the chemistry module takes place through simulated winds and temperatures, as well as 110 through the radiative forcing caused by ozone, methane, nitrous oxide, water vapor and CFCs. The dynamical time step is 15 min, whereas the radiation and chemistry schemes are called every 2 h.
The parametrization of PSCs in MEZON includes the three PSC types water ice, NAT and STS. STS droplets form upon the uptake of gas-phase HNO 3 and H 2 O by aqueous sulfuric acid aerosols (supercooled binary solutions, SBS), following the expression by Carslaw et al. (1995). In SOCOLv3.1, the binary aerosols are prescribed as a time series of observed monthly 115 mean sulfate aerosol surface area density, mainly based on SAGE (Stratospheric Aerosol and Gas Experiment) observations (Stenke et al., 2013). NAT is formed if the HNO 3 partial pressure exceeds its saturation pressure (Hanson and Mauersberger, 1988). For NAT particles, a mean radius of 5 µm is assumed, and the maximum number density is set to 5·10 −4 cm −3 . This limitation accounts for the observational evidence that NAT clouds are often strongly supersaturated and prevents condensation of all available gas-phase HNO 3 onto NAT particles. The assumptions of n N AT,max =5·10 −4 cm −3 and r N AT =5 µm allow for 120 ∼10% of the HNO 3 at beginning of winter to be taken up into NAT particles (0.77 ppbv at 50 hPa and 195 K, assuming 5 ppmv H 2 O). For water ice, a particle number density of 0.01 cm −3 is prescribed. This represents the background ice number density but not ice formed in mountain waves, where very high nucleation rates result in much higher ice number densities of ∼ 5-10 cm −3 (Hu et al., 2002) and particle sizes of <3 µm (Höpfner et al., 2006). For water ice particles as well as for is included. The fall velocities of NAT and ice particles are based on Stokes theory (described in Pruppacher and Klett, 1997).
Advection of PSC particles is not explicitly calculated in SOCOL, but at the end of each chemical time step all condensed HNO 3 and H 2 O evaporates back to the gas phase. To prevent spurious PSC formation caused by potential model temperature, HNO 3 and/or H 2 O biases in regions where PSCs are usually not observed, and to avoid overlap with the regular cloud scheme of the GCM, the occurrence of PSCs is spatially restricted. Water ice particles are allowed to occur between 130 hPa and 130 11 hPa and polewards of 50 • N/S. NAT particles are allowed between the tropopause and 11 hPa. STS and NAT particles may form at all latitudes.
For the present study SOCOLv3.1 was run with T42 horizontal resolution (about 2.8 • x 2.8 • in latitude and longitude) and 39 vertical levels between the surface and the model top centered at 0.01 hPa (∼80 km). In order to allow for a direct comparison with observations, the model was run in specified dynamics mode, i.e. the prognostic variables temperature, vorticity, 135 divergence and the logarithm of the surface pressure are relaxed towards ERA-Interim reanalysis data (Dee et al., 2011). We applied a uniform nudging strength throughout the whole model domain, with a relaxation timescale of 24 h for temperature and logarithm of the surface pressure, 48 h for divergence and 6 h for vorticity. The boundary conditions follow the specifications of the reference simulation REF-C1 of phase 1 of the Chemistry Climate Model Initiative (CCMI-1; Morgenstern et al., 2017). All simulations for this study were run between 01 May 2007 and 31 October 2007 with a 12-hourly output time step. 140 We chose 2007 for our evaluation, which represents an average winter in terms of PSC occurrence, while data coverage for CALIPSO was rather high.

CALIPSO PSC observations
The simulated PSCs in SOCOL are compared to measurements from the CALIOP instrument onboard CALIPSO, an Earth observation satellite in the A-train constellation in operation since 2006 (Winker and Pelon, 2003;Winker et al., 2007Winker et al., , 2009).

145
The A-train of satellites orbits the Earth 14-15 times per day, covering the latitudes between 82 • S and 82 • N on each orbit.
CALIOP is a dual-wavelength lidar with three receiver channels, one measuring the 1064 nm backscatter intensity, the two others measuring the parallel and perpendicular polarized components (β and β ⊥ ) of the 532 nm backscattered signal. The frequency of the lidar pulse is 20.25 Hz, corresponding to one measurement every 333 m along the flight track. From the measured backscatter coefficients (e.g. β 532 ) the total (sum of particulate and molecular) to molecular backscatter ratio can be calculated, with β m being the molecular backscatter coefficient. β m is calculated as described in Hostetler et al. (2006) using molecular number density profiles provided by the MERRA-2 (Modern-Era Retrospective analysis for Research and Applications, version 2) reanalysis products (Gelaro et al., 2017). With the separation of the 532 nm backscatter signal into parallel and perpendicular polarized components, the depolarization ratio (δ aerosol , i.e. the perpendicular to parallel component) 155 of the 532 nm signal can be derived, which is an indicator of the particle shape and hence phase (liquid/solid).
In this study we use the Lidar Level 2 Polar Stratospheric Cloud Mask Product (available via Michael C. Pitts), which was derived with version 2 (v2) of the PSC detection algorithm (Pitts et al., 2018) from the CALIOP v4.10 Lidar Level 1B data products. This CALIOP PSC dataset contains profiles of PSCs with classification and optical properties, also providing temperature, pressure and tropopause height derived from MERRA-2 reanalyses. The spatial resolution of PSC data is 5 km in high number densities and ice as well as to wave ice PSCs.

MLS observations
In this study, modeled HNO 3 , H 2 O and O 3 mixing ratios are compared to satellite measurements of the instrument Microwave  The colors indicate the number of PSC measurements in one bin. Dotted lines denote dynamical classification boundaries or thresholds and solid lines denote fixed classification boundaries.

Model-measurement comparison
While CALIOP measures backscatter signals and depolarization ratios, the SOCOL model provides surface area densities (SAD) for STS, NAT and water ice as function of pressure, latitude and longitude. From the simulated SADs and the assumed 190 microphysical parameters, we calculate the number density and/or radius for each particle type. This information is used in Mie and T-matrix scattering codes (Mishchenko et al., 1996) to compute optical parameters of the simulated PSCs, i.e. R 532 , δ aerosol and β ⊥ , for comparison with CALIOP observations. For NAT and ice particles, circular symmetric spheroids with an aspect ratio of 0.9 are assumed. Refractive indices of 1.31 for ice and 1.48 for NAT were chosen. The CALIOP PSC data product includes both detection threshold values, R 532,thresh and β ⊥,thresh , for each measurement. To achieve a better comparability 195 between model and observations, these daily threshold values are also applied on the calculated optical properties of the PSCs simulated by SOCOL. For this purpose, we calculated the daily mean thresholds from all observations for each pressure level.
This procedure is essential for a fair comparison between model and satellite data, as the geographical PSC extent strongly depends on these detection limits.
To ensure best possible comparability between model and measurements, observational uncertainties have to be applied 200 to the calculated optical properties of the modeled PSCs. We followed the approach by Engel et al. (2013). The uncertainty scales inversely to the square root of the horizontal averaging distance along a flight path, which we set to 135 km. This value corresponds to the best case for detection, which maximizes the comparability with the model (which obviously does not have a detection threshold). An example for the added measurement noise is shown in Fig. 2. When looking at the individual PSC types (Fig. 2a), STS and NAT, due to their spherical shape and fixed radius, appear at constant δ aerosol -values of 0 and 0.167, respectively. The variable radius of ice particles results in a variable δ aerosol -value. Applying the uncertainties to the parallel and the perpendicular backscatter coefficients primarily causes a large spread in depolarization ratio (Fig. 2b). When considering all PSC particles to be mixed within a grid box (Fig. 2c), their points are located mainly at the lower and left side of the composite histogram.
3 Results and discussion 210 3.1 Comparison along an orbit As a first example we compare SOCOL with CALIPSO along a single flight track. Figure 3 shows a curtain of observed backscatter ratios R 532 along orbit 2 on 01 July 2007 (Fig. 3a) and the corresponding PSC compositions (Fig. 3g) boxes overflown by CALIPSO. Figures 3c and 3f show the same, but before detection thresholds and instrument uncertainty had been added. The model output also reveals a large PSC over the Antarctic Peninsula. However, the spatial extent of the simulated PSC is larger. The simulated backscatter ratio R 532 peaks around 6, which is substantially lower than observed. Due to the coarse resolution and orography, SOCOL is not able to capture high ice particle number densities associated with mountain wave events. Applying the CALIPSO classification scheme on the model output results in a layer of ice PSCs located 225 around ∼20 km, which is slightly higher than in the observations. The ice cloud is surrounded by NAT mixtures, while the observations indicate STS. Below those NAT mixtures, pure STS clouds occur in the model, most of which are tenuous enough such that they fully disappear after applying the optical thresholds (Fig. 3e).
The actual modeled composition (see Appendix, Fig. A1) shows a similar pattern than the CALIPSO classification scheme, but with more ice Mix and less STS. These differences can also be seen in Fig. 2c, where most of the ice mixtures (blue) are   The modeled month-to-month variability of R 532 values and areal extent agrees well with CALIPSO observations. In July, the center of the PSC area is also tilted towards East Antarctica and slightly towards the Peninsula in August. However, peak values of R 532 are clearly lower for SOCOL. In comparison to the observations, the spatial distribution of SOCOL PSCs is more homogeneous. As mentioned above, this results mainly from a poor representation of mountain waves in the model.  The observed PSC area is calculated from the daily fraction of PSC measurements within ten equal-sized latitude bands, while the modeled PSC area is determined for every grid box based on the PSC occurrence (above the detection thresholds) for two until end of October, which is longer than observed. However, SOCOL simulates a substantially larger PSC area, in particular between 13 and 23 km altitude, where 1.5·10 7 km 2 are almost continuously exceeded.

Spatial distribution
It is most likely that the different methods for calculating PSC areal coverage contributes to this overestimation. For each output time step, we considered the entire grid box to be covered by PSCs as soon as PSCs (above the detection thresholds) occur in the model. Further, also a cold-temperature bias in the model contributes to the larger PSC area.

265
The modeled PSC area calculated without the optical thresholds applied (Fig. 5c) is significantly larger, especially below 13 km altitude, where large areas with STS clouds occur in the model (see also Fig. 3f). Those large-scale STS clouds are very tenuous since they are filtered out by the conservative PSC detection threshold and hence do not play an important role in ozone chemistry. However, it highlights the crucial role of the detection thresholds when comparing PSC areas. Due to this sensitivity to the applied methods, quantitative comparisons of the areal coverage must be interpreted with caution.
270 Table 1. Overview over the SOCOL simulations and the microphysical parameter settings.

Sensitivity to microphysical parameters
As described in Sect. 2.1 SOCOL's PSC scheme includes some prescribed microphysical parameters such as the ice particle number density, n ice , or the NAT radius, r NAT . These values had once been chosen based on what was known about PSCs back then. However, the current parameter setting might not be optimal. For example, the rather low value for n ice of 0.01 cm −3 prevents the formation of ice PSCs with high number densities as observed in mountain wave events. To investigate the sensi-275 tivity of the simulated PSCs to the microphysical parameters in the PSC scheme, we performed additional simulations for the Antarctic winter 2007 with increased n ice and/or increased n NAT,max (Table 1). Figure 6 shows the composite histograms for the various SOCOL simulations. There are considerable differences to the observations (Fig. 1), but also between the simulations. PSCs in the REF simulation show a strong relative maximum located in the STS domain with 1/R 532 values between 0.4 and 0.2 (Fig. 6a). Only very few PSCs are classified as ice, i.e. the relative 280 maximum towards the upper right, as observed by CALIPSO, is missing. That the PSC mixtures in the simulations are located more at the lower and left side of the histogram can also be seen in Fig. 2c. There are several reasons for this difference. First, SOCOL does not resolve mountain waves due to the coarse model resolution and orography. Furthermore, the modeled PSCs are representative for large grid box (2.8 • x2.8 • horizontally and approximately 2 km vertically), while the observations resolve much smaller scale structures (starting from 5 km horizontally along a track and 180 m vertically). Finally, the fixed ice number 285 density of 0.01 cm −3 does not allow for large ice particle cross sections, even if mountain waves would be resolved. Based on these findings we performed one sensitivity simulation with a tenfold ice number density, S n(ice) . As shown Fig. 6b the tenfold increase in n ice results in a strong maximum to the upper right, mainly within the enhanced NAT mixture domain. The higher number density of ice particles increases the cross section of ice, leading to enhanced backscatter in ice-containing grid cells.
Due to its solid state, backscatter from ice has δ aerosol >0. This results in a shift towards higher R 532 and higher δ aerosol values in 290 the histogram. Overall, modifying n ice leads to a better agreement with CALIPSO.
While ice PSCs are less important for stratospheric ozone chemistry, NAT formation and subsequent denitrification of the stratosphere play a crucial role. NAT formation in SOCOL depends on two parameters, n NAT,max and r NAT . To test the model's sensitivity to those parameters, we ran further simulations with the upper boundary for NAT number densities increased by a latter is not presented here.
The simulation with four times higher n NAT,max (Fig. 6c) shows a maximum shifted towards lower R 532 values compared to the REF simulation, which is located around the optical thresholds at the lower left corner. As long as temperatures are below T NAT and enough HNO 3 is available for NAT formation, an increase in n NAT,max or r NAT results in more HNO 3 -uptake by NAT particles. This reduces the available gas-phase HNO 3 for STS growth. Also, more HNO 3 through sedimentation of the solid NAT particles is removed. With larger r NAT this removal occurs even faster due to the higher sedimentation velocity.
The reduction in surface area density of STS results in less backscatter and subsequently a shift towards lower R 532 values in the composite histogram. This shift towards lower R 532 values worsens agreement with observations.
In a final simulation (S n(ice),n(N AT,max) , Fig. 6d) we set n ice to 0.05 cm −3 and n NAT,max to 10 −3 cm −3 . This simulation shows a superposition of the two effects described above, resulting in two distinct relative maxima in the composite histogram.

305
One maxima is located to the upper right, similar to S n(ice) . The second maximum at low R 532 and low δ aerosol values is similar to the pattern in S n(N AT,max) . The shift towards lower R 532 values is again a result of less STS formation due to the reduced availability of HNO 3 . Although the composition histograms of all sensitivity simulations differ substantially from observations, we find the best agreement for the simulation S n(ice),n(N AT,max) .
To investigate the impact of the applied modifications on the simulated chemical composition of the polar stratosphere (60- Prior to the decline, an increase in HNO 3 is observed at 68 hPa. It results from the evaporation of sedimenting NAT particles formed at higher altitudes (renitrification) and is an indication of denitrification of the upper levels. During July/August the absolute HNO 3 values from the reference run agree well with the observations. However, in late winter SOCOL again underestimates MLS. All simulations show a decline due to HNO 3 -uptake into NAT particles and STS droplets. However, S REF

320
(black) and S n(ice) (cyan) show a weaker and delayed HNO 3 decline with a plateau in July/August. In S n(N AT,max) (green) the decline at both levels is considerably stronger than in S REF as well as in MLS. This is due to the enhanced uptake of HNO 3 into NAT particles and the subsequent removal by sedimentation. As a consequence also the renitrification at lower levels is clearly enhanced. Both indicates a more efficient denitrification than in S REF .
The simulation S n(ice),n(N AT,max) (red), in which n NAT,max is twice as large as in S REF , but only half of S n(N AT,max) , falls 325 in between the other simulations. The denitrification starts about half a month later than in S n(N AT,max) . The HNO 3 -uptake is reduced and subsequently HNO 3 stays longer in the gas-phase. However, in August HNO 3 concentrations reach about the same level as in S n(N AT,max) . Simulations with enhanced r NAT have similar effects (not shown).     Increasing the parameter n ice affects the modeled stratospheric composition only very little by reducing dehydration. But the increased SAD of ice leads to slightly lower O 3 in S n(ice) compared to S REF . Increasing the upper NAT boundary overall reduces SAD of PSC due to reducing the abundance of HNO 3 . However, due to enhanced denitrification, S n(N AT,max) and S n(ice),n(N AT,max) show even slightly lower O 3 concentrations.

345
We have presented an evaluation of PSCs as simulated by the CCM SOCOLv3.1 in specified dynamics mode for the Antarctic winter 2007. SOCOL considers STS droplets as well as water ice and NAT particles. PSCs are parametrized in terms of tem- Overall, the spatial agreement with CALIOP observations is good and the observed month-to-month variability is represented. However, due to the coarse model, mean orography, but also the fixed ice number densities, mountain wave events and associated wave ice clouds with high backscatter ratios over the Antarctic Peninsula are not resolved in SOCOL. The temporal 355 and spatial evolution of PSCs inside the polar vortex as expressed by the areal coverage indicates an overestimation of PSCs in SOCOL. This is partly explained by a cold temperature bias, but also by the coarse model resolution: even a small amount of PSCs within a grid cell adds a large contribution to the areal coverage. This is reflected by the sensitivity of this quantity towards the applied detection thresholds.
Furthermore, we have tested the assumptions about the maximum NAT number density, NAT radius and ice number density 360 by various sensitivity simulations. The parameter n ice determines primarily the optical signal through its impact on the particle cross section and also dehydration due to changing settling velocities with changing particle radius. While increasing n ice from 0.01 cm −3 to 0.1 cm −3 improves the agreement of the optical signal with CALIOP, the simulated dehydration is more realistic for smaller n ice and therefore larger ice particles.
The upper boundary for NAT number densities determines the HNO 3 -uptake and subsequently the magnitude of STS for-365 mation, which is crucial for halogen activation. We have shown that for an increased max. NAT number densities the temporal agreement of de-and renitrification with MLS measurements is improved. However, SOCOL in general clearly underestimates observed HNO 3 in the polar stratosphere, which makes a solid conclusion about the best set of microphysical parameters respectively. Further work would be required to extend our findings to simulated PSCs in the Arctic or to other years. Nevertheless, this study demonstrates that also a simplified PSC scheme based on equilibrium assumptions may achieve good approximations of fundamental properties of polar stratospheric clouds needed in chemistry-climate models.
Code and data availability. Since the full SOCOLv3.1 code is based on ECHAM5, users must first sign the ECHAM5 license agreement before accessing the SOCOLv3.1 code (http://www.mpimet.mpg.de/en/science/models/license/, last access: 2020). The SOCOLv3.1 code is then freely available upon request from Andrea Stenke (andrea.stenke@env.ethz.ch