Arctic stratospheric dehydration – Part 2: Microphysical modeling

Large areas of synoptic-scale ice PSCs (Polar Stratospheric Clouds) distinguished the Arctic winter 2009/2010 from other years and revealed unprecedented evidence of water redistribution in the stratosphere. A unique snapshot of water vapor repartitioning into ice particles was observed under extremely cold Arctic conditions with temper- 5 atures around 183 K. Balloon-borne, aircraft and satellite-based measurements suggest that synoptic-scale ice PSCs and concurrent reductions and enhancements in water vapor are tightly linked with the observed de- and rehydration signatures, re-spectively. In a companion paper (Part 1), water vapor and aerosol backscatter measurements from the RECONCILE (Reconciliation of essential process parameters for 10 an enhanced predictability of Arctic stratospheric ozone loss and its climate interactions) and LAPBIAT-II (Lapland Atmosphere-Biosphere Facility) ﬁeld campaigns have been analyzed in detail. This paper uses a column version of the Zurich Optical and Microphysical box Model (ZOMM) including newly developed NAT (Nitric Acid Trihydrate) and ice nucleation parameterizations. Particle sedimentation is calculated in or- 15 der to simulate the vertical redistribution of chemical species such as water and nitric acid. Accounting for small-scale temperature ﬂuctuations along the trajectory is essential to reach agreement between simulated optical cloud properties and observations. Whereas modeling only homogeneous nucleation causes the formation of ice clouds with particle radii too small to explain the measured vertical redistribution of water, we 20 show that the use of recently developed heterogeneous ice nucleation parameteriza-tions allows the model to quantitatively reproduce the observed signatures of de- and rehydration. ﬀ ects of gravitational settling and irreversible dehydration. To corroborate this interpretation we simulated the formation and sedimentation of the ice 20 particles. Simulated water vapor proﬁles agree reasonably with CFH and FLASH-B on-board the balloon sondes, and with MLS satellite measurements. Optical T-Matrix calculations enabled the direct comparison of the simulations with COBALD and CALIOP backscatter measurements. To this end, we examined the e ﬀ ect of small-scale temperature ﬂuctuations and compared homogeneous vs. heterogeneous formation of ice 25 particles.

to the RECONCILE (Reconciliation of essential process parameters for an enhanced predictability of Arctic stratospheric ozone loss and its climate interactions) project and its activities within the same Arctic winter (von Hobe et al., 2013). Whereas the majority of Arctic PSC studies describe wave ice cloud observations above Scandinavia (e.g. Carslaw et al., 1998b;Fueglistaler et al., 2003), we present model simulations based The Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) satellite provides profiling measurements of cloud and aerosol distribution and properties obtained by the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP). The satellite completes 14.5 orbits per day and has a near-global coverage, ranging from 82 • N to 82 • S (Winker et al., 2007). CALIPSO's orbit inclination and its associated ex-25 tensive polar coverage make the satellite an ideal platform for PSC observations. For this purpose, Pitts et al. (2007) introduced a detection algorithm to identify and further classify PSCs. Their composition classification is based on BSR at a wavelength of 27168 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 532 nm and aerosol depolarization (δ aerosol ) obtained from the CALIOP Level 1B Data Product. The data is averaged to a uniform grid with a horizontal resolution of 5 km and a vertical resolution of 180 m. Pitts et al. (2011) distinguish six different PSC composition classes with varying number densities of liquid, NAT, and ice particles. 5 Balloon-borne water vapor measurements are obtained by two different techniques. The Fluorescence Lyman-Alpha Stratospheric Hygrometer for Balloons (FLASH-B) is a Russian water vapor instrument developed at the Central Aerological Observatory (Yushkov et al., 1998). The fluorescence method uses the photodissociation of H 2 O molecules exposed to Lyman-alpha radiation followed by the measurement of the fluo-10 rescence of the resulting excited OH radicals (Kley and Stone, 1978). The intensity of the fluorescent light sensed by the photomultiplier is directly proportional to the water vapor mixing ratio under stratospheric conditions (10 hPa to 150 hPa). The second instrument used within this study is the Cryogenic Frost point Hygrometer (CFH), which was developed at the University of Colorado (Vömel et al., 2007a). The instrument's 15 principle is based on a chilled mirror. The temperature of the mirror can be regulated such that a thin but constant layer of frozen condensate covers the surface. The thickness of the frozen layer is controlled by a photodiode connected to an LED. The photodiode measures variations of the reflected light caused by changes in the thickness of the condensate, which feeds back into the regulation of the mirror temperature. Under 20 these conditions, T frost of the surrounding air is equal to the mirror temperature. The reported overall uncertainty for both instruments in the middle stratosphere is less than 10 % (Vömel et al., 2007b). Water vapor and nitric acid were measured using the Microwave Limb Sounder (MLS) aboard the Aura satellite. MLS provides atmospheric profiles of temperature 25 and composition (including H 2 O and HNO 3 ) via passive measurement of microwave thermal emission from the limb of the Earth's atmosphere (Waters et al., 2006). Vertical scans are performed every 25 s, corresponding to a distance of 165 km along 27169

Water vapor and nitric acid measurements
April 2008 (Lambert et al., 2012). MLS water vapor profiles presented in this study are interpolated to the CALIOP PSC grid using a weighted average of the two nearest MLS profiles . Typical single-profile precisions of the MLS version 3.3 measurements (Livesey et al., 2011) are 4 % to 15 % for H 2 O (Read et al., 2007;Lambert et al., 2007) and 0.7 ppbv for HNO 3 (Santee et al., 2007).

Trajectory calculation
The trajectories used within this study are calculated from six-hourly wind and temperature fields of the ERA-Interim reanalysis produced by the ECMWF (Dee et al., 2011), with a horizontal resolution of 1 • × 1 • . We used the trajectory module of the Chemical Lagrangian Model of the Stratosphere (CLaMS) (McKenna et al., 2002) to calculate tra- 15 jectories for the microphysical study. Vertical velocities are derived from ERA-Interim total diabatic heating rates (Ploeger et al., 2010). Starting at 19:47 UTC on 17 January 2010, a balloon sounding was performed from Sodankylä (hereafter referred to as S1). Individual balloon positions along its pathway at pressure levels < 100 hPa served as start points for the first set of trajectory calculations. With a vertical distance of 20 100 m between two trajectories, two-day backward and three-day forward trajectories with time steps of 15 min were computed. A second sonde was launched on 23 January 2010 at 17:30 UTC (hereafter referred to as S2), for which three-day backward and two-day forward trajectories in the same altitude range as for the first sounding have been calculated. Figure 1 illustrates the pathways of two exemplary trajectories 25 for each sounding within the Arctic vortex. It is apparent that Sodankylä was located in the cold pool with temperatures as low as 183 K as measured by the sondes, while 27170 upstream the air had been by more than 10 K warmer. On 17 January Sodankylä was at the edge of a larger area of synoptic-scale ice clouds seen by CALIOP.
As the sedimentation scheme in the microphysical column model (described in detail in Sect. 2.4) cannot account for wind shear, changes in wind velocity with altitude lead to errors in the location of the sedimentation events. The 440-K and 520-K trajecto-5 ries are about 3 km apart in altitude. At a point 12 h downstream of S1, the distance between these trajectories started at potential temperatures of 520 K (cloud top) and 440 K (cloud base) is about 350 km. This distance is within the area of ice PSC with consistent cold temperatures, and as the modeled sedimentation takes place within these first 12 h, the wind shear has only a moderate effect on the results. Pressure and temperature from the joined backward and forward trajectories constitute the meteorological input for the microphysical model. Additionally, information about total water and nitric acid mixing ratios is needed at the upstream end of the trajectories. For S1, HNO 3 values were taken from MLS, averaged for the corresponding day over cloud-free areas within the vortex and vertically interpolated to the starting 15 pressure of the trajectories. Vertically resolved climatological mean values for January were used as H 2 O input. The H 2 O profile is calculated from ice cloud free measurements conducted above Sodankylä, between 1996 and 2013, using the NOAA frost point hygrometer and the CFH (see Khaykin et al., 2013, for more details). The climatological mean, which excludes the single dehydration measurement from 23 Jan-20 uary 1996 (Vömel et al., 1997), is shown below. Simulations of S2 were initialized with the H 2 O profile from S1 at t = 21 days. Even though a horizontal displacement exists between the individual trajectory end and start points of S1 and S2, respectively, the approach is justified given the temperature distribution on vortex scale: at points, where trajectories are matched, temperatures were above 200 K, ensuring cloud-free air and 25 thus no change in the H 2 O distribution.
Trajectory temperatures were corrected according to Fig. 2, showing temperature deviations between ERA-Interim reanalysis data and measurements, taken by the Vaisala RS-92 on S1. The total measurement uncertainty for the temperature sensor is 0.5 K 27171 cold pool not resolved in the ERA-Interim data, temperatures along the trajectories were changed only within a short time window around the observation. The amplitude of the applied temperature correction is assumed to decrease (using a sine curve) with increasing distance from the observation (vertical red line in Fig. 2c) and equals zero 12 h before and after the observation. Figure 2c illustrates an exemplary trajectory 10 without (black dashed line) and with (red solid line) applied temperature correction. In the absence of better knowledge, the maximum amplitude is assumed to occur at the sonde flight path and is assumed to be altitude dependent as shown in Fig. 2b (red line). 15 Early studies (e.g. Murphy and Gary, 1995;Kärcher and Lohmann, 2003;Hoyle et al., 2005) have investigated the effect of rapid temperature fluctuations and associated high cooling rates on ice cloud formation and properties. Cooling rates of less than one Kelvin per hour favor the growth of preexisting ice particles and total number densities of ice particles remain low. In contrast, high cooling rates of several Kelvin per hour can 20 produce supersaturations high enough to nucleate a major fraction of the stratospheric background aerosol. Gary (2006) found that a significant component of the short-term vertical displacements of isentropic surfaces remains unresolved also by current numerical weather prediction models. Recent microphysical modeling studies confirmed the importance of an adequate representation of cooling rates for cirrus (Brabec et al.,25 2012; Cirisan et al., 2013) and polar stratospheric clouds . We follow the same approach chosen in these studies and make use of the vertical velocity and temperature time series obtained from the SUCCESS (Subsonic Aircraft: trail and Cloud Effects Special Study) data analyzed by Hoyle et al. (2005) to conduct model runs along trajectories with superimposed temperature fluctuations. Only wavelengths < 400 km were considered, which are not resolved in the ERA-Interim wind fields used in our trajectory calculations. The temperature fluctuations are assumed to have a mean amplitude of about ±0.5 K and were superimposed onto the synoptic-5 scale trajectories with random frequencies and a temporal resolution of 1 s as seen in Fig. 2c (black solid line). A more detailed description of the method can be found in .

Microphysical column model
The new column version of the Zurich Optical and Microphysical box Model (ZOMM) 10 with implemented heterogeneous ice and NAT nucleation rates is used to simulate the formation, evolution and sedimentation of ice particles along trajectories. The underlying model, utilized for PSC simulations, has been described by Meilinger et al. (1995) and Luo et al. (2003b) and recently extended by Hoyle et al. (2013) and . The following section provides an overview and the details of the modifications 15 made to ZOMM for the purposes of this study. ZOMM can be initialized with a log-normally distributed population of supercooled binary solution (SBS) droplets, described by a mode radius, number density and distribution width, typical for winter polar stratospheric background conditions (Dye et al., 1992). Driven by temperature and pressure data along trajectories, the uptake and 20 release of nitric acid and water in ternary solution droplets is determined. The total amounts of H 2 O, H 2 SO 4 and HNO 3 contained in the air parcel are set at the beginning of the trajectory. A mixing of air parcels is not possible and therefore the sum of gas and particle phase remains constant unless sedimentation takes place. The mass of sedimenting NAT and ice particles is conserved as described below. Distributed across 26 25 radius bins when the model is initialized, droplets are henceforward allowed to grow and shrink in a fully kinetic treatment and without being restricted to the initial log-normal shape of the distribution (Meilinger et al., 1995) in an initiation of additional size bins. Homogeneous ice nucleation in STS droplets is calculated as well as heterogeneous nucleation of ice on foreign nuclei and NAT surfaces. NAT nucleation is implemented as deposition nucleation on ice particles and as immersion freezing on foreign nuclei (e.g. meteoritic dust). Whereas homogeneous ice nucleation, following Koop et al. (2000), and NAT nucleation on uncoated ice sur-5 faces, described in detail in Luo et al. (2003b), have been accepted pathways of PSC formation for many years now, the possibility of PSC formation via heterogeneous ice and NAT nucleation on foreign nuclei (e.g. Tolbert and Toon, 2001;Drdla et al., 2002;Voigt et al., 2005) had until previously only a narrow observational data basis so that definitive conclusions about nucleation rates were not possible, and also any clear support from laboratory measurements was lacking (see detailed discussion by Peter and Grooß, 2012). The observational impasse has been overcome recently by the wealth of CALIOP PSC observations on NAT and ice PSCs obtained in the winter 2009/2010 (Pitts et al., 2011), which unmistakably suggests that both particle types must have nucleated heterogeneously. A heterogeneous nucleation mechanism, occurring on pre-15 existing particle surfaces, e.g. on meteoritic particles, has been developed to explain CALIOP PSC observations over the Arctic in December 2009 and January 2010. The parameterizations, based on active site theory (Marcolli et al., 2007), for NAT and ice are given in Hoyle et al. (2013) and , respectively. These studies used ZOMM in a pure box model configuration and limitations caused by neglecting 20 sedimentation of NAT and ice particles in the winter polar stratosphere were already pointed out by these authors. For the present study, we developed the stratospheric version of ZOMM further into a column model similar to the existing cirrus column version of ZOMM (e.g. Luo et al., 2003a;Brabec et al., 2012;Cirisan et al., 2013). Sedimentation of ice and NAT parti-25 cles is realized in a Eulerian scheme, allowing particles to sediment within the advected column from one box to the next lower one. For the present study the column consists of a stack of 100 m thick boxes and the timestep for sedimentation is 15 min. Once ice or NAT particles grow to sizes large enough to sediment, the appropriate fraction of Introduction particles is removed from its current box and, according to its size-dependent sedimentation speed, injected into the next lower box. This is done in a way that the number and mass of the particles are conserved. Strictly speaking, such an approach is only possible in situations without horizontal or vertical wind shear. The case investigated in this study is sufficiently close to meeting these criteria, as the column of air parcels 5 rotates with the polar vortex (compare Fig. 1). The optical properties of the simulated PSCs are calculated using Mie and T-Matrix scattering codes (Mishchenko et al., 2010) to compute optical parameters for sizeresolved number densities of STS, NAT and ice. The refractive index is 1.31 for ice and 1.48 for NAT. Following , both crystals are treated as prolate 10 spheroids with aspect ratios of 0.9 (diameter-to-length ratio).

Observations
The Arctic winter 2009/2010 was characterized by a week-long period of unusually cold temperatures in the lower stratosphere. From 15 to 21 January, temperatures below T frost led to widespread synoptic-scale ice PSCs, which were observed by CALIOP 15 (Pitts et al., 2011). The balloon sonde S1 equipped with COBALD and FLASH-B (besides ozone, meteorological parameters and GPS) was launched from Sodanklyä on 17 January 2010 at 19:47 UTC (Khaykin et al., 2013). At about 21:00 UTC the balloon reached its point of burst and the payload began its descent. COBALD and FLASH-B profiles measured during the descent are shown in Figs. 3 and 4, respectively. Parti-20 cle backscatter ratios and the simultaneously captured reduction in water vapor reveal three distinct layers of ice particles. Maximum backscatter ratios at 870 nm reach 200 at a potential temperature of 510 K and mark the clearly defined upper edge of the lowest ice layer. The ice layers are embedded in a cloud of supercooled ternary solution (STS) droplets extending from 440 K upwards and identified by the backscatter 25 increase from the background level below. A vortex wide change in PSC composition occurred on 22 January together with the onset of a major warming, and CALIOP measurements after this time showed predominantly liquid PSCs (Pitts et al., 2011). An unprecedented measurement of vertical redistribution of water followed on 23 January 2010. Sonde S2 with COBALD and CFH was launched at 17: 30 UTC (Khaykin et al., 2013). A stratospheric layer of irreversibly 5 dehydrated air was measured at potential temperatures above 470 K. The reduction in water vapor of 1.6 ppmv was observed in essentially cloud free air with backscatter ratios below 5 at 870 nm. Even though temperatures at this level were as cold as on the 17 January, no ice cloud formed due to the reduced amount of H 2 O and hence a depression of T frost . A clear signal of rehydration was detected below this layer (between 10 450 K and 470 K). The enhancement in water vapor of about 1 ppmv above climatological mean conditions coincided with COBALD backscatter measurements of up to 20. Backscatter values from COBALD suggest the existence of liquid particles and NAT in this layer, which is consistent with the CALIOP observations. 15 To analyze differences in PSC properties resulting from homogeneous or heterogeneous ice nucleation as well as from changes in temperatures and cooling rates, we performed various simulations with different initial conditions. The results with the best agreement between simulated and observed PSC properties are shown in Figs. 3 and 4. This simulation accounted for homogeneous ice nucleation, heterogeneous nucle-20 ation of NAT and ice on foreign nuclei as well as the nucleation of ice on preexisting NAT particles and the nucleation of NAT on preexisting ice particles. Small-scale temperature fluctuations needed to be superimposed onto the trajectories in order to reproduce the observations.

Direct measurement-model comparison
The comparison of measured and simulated optical properties, namely BSR and δ aerosol , is presented in Fig. 3. Here, the modeled temporal evolution of BSR at 532 nm along the trajectories is shown in panels a and d as a function of potential temperature. We started trajectories backward and forward in time as described in Sect. 2.3 along the descent of S1 (FLASH-B and COBALD) and the ascent of S2 (CFH and COBALD). A discussion about the availability of FLASH-B and CFH and about the data quality during ascent and descent can be found in our companion paper (Khaykin et al., 2013). Both vertical profiles have been compared to balloon and satellite-borne measurements. Their approximate observation time is indicated by vertical red lines in 10 the upper panels. Their exact position is equal to the position of S1 and S2, respectively. Panel b shows modeled and observed backscatter for the initial sounding S1 on 17 January and panel e for S2, the sounding which took place on 23 January 2010.
Panel c and f show satellite measurements, a 25 km mean around the profile closest in time and space to the corresponding balloon sounding from Sodankylä. Whereas

15
CALIPSO and Aura were ∼ 500 km far from Sodankylä with a time difference of 3 h between the individual measurements on 17 January, measurements on the 23 January have a horizontal displacement of only ∼ 100 km but a 6 h time lag. All measurements are shown in black, simulations in red. Figure 4 presents the corresponding results for water vapor. Simulated NAT and ice number densities as well as their mean radii are 20 presented in Fig. 5, which shows various scenarios with respect to nucleation mechanisms and will be explained in detail below. (The rightmost column of Fig. 5 depicts the scenario from which Figs. 3 and 4 were derived.) Temperatures above the existence temperature of NAT (T NAT ) ensure sub-saturated conditions (i.e. no PSCs) at the beginning of the simulation. Within the first 48 h after 25 the start of the simulation, a significant enhancement in BSR indicates the formation of a cloud. At t ∼ 17.5 days, temperatures become low enough to permit first NAT nucleation on preexisting particle surfaces. NAT number densities of 10 −3 cm −3 lead to a small but visible increase in BSR. With decreasing temperatures, STS particles grow through uptake of HNO 3 from the gas phase and contribute significantly to a continuous increase of BSR. The onset of reduced water vapor mixing ratios shortly before the point of observation is shown in Fig. 4 and marks the formation of ice particles. The comparison between simulated vertical backscatter profiles (red) with those measured 5 by COBALD (black line in Fig. 3b) shows a reasonable agreement. However, smallscale structures of enhanced BSR seen by COBALD are only partly reflected and BSR values stay below the maximum detected by COBALD. BSR values as large as 200 (at 870 nm) are possible to simulate by choosing a slightly different phase of a fluctuation as seen in the presentation of the ensemble runs below. As Fig. 3c shows, CALIOP 10 measurements (black line) are represented extremely well. Simulated δ aerosol values are enhanced in the same altitude region as in the CALIOP observations, but values of δ aerosol measured by CALIOP fluctuate more strongly producing "jumps" between the ice and STS class (compare dashed lines in Fig. 3c). Even though instrumental noise and the lower resolution might be an explanation for this behavior, the COBALD 15 measurements suggest that fluctuations in the cloud composition profile are possible. A perfect anti-correlation between the profiles of BSR and H 2 O (black lines in Fig. 3b and Fig. 4b) suggests an ice cloud and the corresponding depletion in the vapor phase, both strongly layered. The small color bar next to the vertical profiles denotes the results from the CALIOP particle classification scheme, suggesting clear layers of ice 20 embedded in a broader liquid PSC. Simulated ice number densities in the core of the cloud lie between 10 −3 cm −3 and 10 −2 cm −3 , leading to a maximum BSR at 532 nm of almost 7. In Fig. 3a, downwind of S1, homogeneous ice nucleation sets in and higher ice number densities between 0.1 cm −3 and 1 cm −3 are simulated. Those high number densities cause an increase in BSR by a factor of 3.5. The part of the ice cloud with 25 the highest BSR values is followed by a tail of moderate BSR upon warming, whereas elsewhere BSR values are much smaller. This tail consists of NAT particles (see Fig. 5) that nucleated on the ice cloud, forming a NAT cloud of class "Mix2" and "Mix2-enh", as first described by Carslaw et al. (1998a).
27178 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | The magnitude of the H 2 O reduction in the gas phase observed by FLASH-B is roughly captured by the simulations. Again, fine-scale structures are not reproduced by the model (Fig. 4b). The resolution of MLS is too coarse to capture the local reduction in H 2 O at this time (Fig. 4c). Ice particles evaporate already 12 h after the FLASH-B observation and a clear and permanent redistribution of H 2 O becomes visible at 5 t ∼ 18.5 days in Fig. 4a, which remains visible until the end of the simulation.
Close to the second observation, temperatures drop again and reach values almost as cold as on the 17 January. However, the reduction in H 2 O prevents the formation of ice clouds in the model, except at levels above 520 K. Also the magnitude of BSR from COBALD and CALIOP would not suggest an ice cloud. Values of BSR remain smaller 10 than they were on 17 January and CALIOP observations suggest a predominantly liquid cloud with few embedded NAT particles. The NAT signal is partially obscured by the strong STS signal (Fig. 3f). The comparison between modeled and observed H 2 O reveals that even though the model captures signatures of de-and rehydration, the vertical extent of the dehydrated air remains too small in the simulation. The simulated 15 maximum reduction in water vapor of −1.4 ppm at 514 K is almost as large as observed by CFH. However, the dehydrated region is smaller and ranges only from 485 K to 525 K. A possibility for the underestimated vertical extent of dehydrated air in the simulation might again be the temperature profile. Figure 2b shows that ERA-Interim temperatures are too warm compared to the observation also at the top of the sound- 20 ing, which prevents the formation of ice particles above 525 K. The resulting availability of H 2 O at high altitudes offers the possibility for ice formation on the 23 January 2010 in our simulation, whereas CALIOP observations documented the last ice clouds in the vortex on 21 January (Pitts et al., 2011). Finally, clear signatures of de-and rehydration are also visible in the MLS profile (Fig. 4f), which reveal that the redistribution of water 25 is a large-scale phenomenon. Figure 5 compares simulations using four different scenarios. The top panels show ERA-Interim temperatures, which were corrected according to the measured temperature profile of S1 (compare Sect. 2.3 and Fig. 2) and provide the basis for all four 27179 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | simulations. The first scenario (Fig. 5, 1st column), which accounts only for homogeneous ice nucleation, cannot explain the observations at all. Supersaturations with respect to ice remain too small and since no ice particles form, NAT particles cannot form either. The second scenario (Fig. 5, 2nd column), which includes heterogeneous nucleation of ice and NAT particles according to  and Hoyle et al. (2013), changes the model results completely. Ice supersaturations are sufficient for heterogeneous nucleation of 10 −3 cm −3 ice particles. Within a short time, these particles grow to sizes > 12 µm in radius. With a settling velocity between 100 m h −1 and 200 m h −1 , ice particles can sediment up to 2 km before evaporation. The third scenario ( Fig. 5, 3rd column) shows that the superposition of small-scale temperature fluctu-10 ations on the synoptic trajectories leads to ice formation even in the homogeneous freezing case. However, this result does not agree with the observations, because the nucleation of high ice number densities (between 1 cm −3 and 10 cm −3 ) prevents the growth of ice particles to sizes which could sediment fast enough to achieve a satisfying agreement with the observations (maximum sedimentation distances of only a few 15 hundred meters). The final scenario (Fig. 5, 4th column) includes both, heterogeneous nucleation and superimposed small-scale temperature fluctuations. The inclusion of heterogeneous nucleation in this simulation causes NAT to form prior to ice. Consequently, those NAT particles are assumed to be enclosed by ice and redistributed by the subsequent growth and sedimentation of the ice particles. We will discuss this point 20 further below. Additionally, the nucleation of ice particles starts earlier than in all other cases and the fluctuations enable a larger ice cloud area to be generated. and is principally unknown to us. At best we have an idea of the statistical nature of the small-scale temperature fluctuations, which determine the cooling rates and thus the resulting cloud morphology. Therefore, ensemble calculations applying different sets of small-scale temperature fluctuations need to be performed, in order to retrieve the dependence of PSC properties on the stochastic effects caused by the fluctuations. 5 Cirisan et al. (2013) have performed similar calculations for cirrus clouds. Figure 6 presents a comparison of profiles of BSR and ∆H 2 O for various scenarios with and without heterogeneous nucleation and with 10 different sets of small-scale temperature fluctuations. BSR profiles for 17 January are presented in the upper four panels, ∆H 2 O profiles for 23 January are shown in the lower four panels. The model 10 results, which have already been described above and shown in Figs. 3-5, correspond to the red curves while the other nine members are shown as gray area. The red curves also represent the particular member which produced the best agreement with the measurements in terms of dehydration.

Ensemble calculations for stochastic impact of temperature fluctuations
As turn leads to different backscatter ratios and dehydration strengths. In passing we note, that the spikes in BSR observed during S1 represent onsets of wave ice embedded in a synoptic ice cloud. The spikes contain ice crystals in high number density, which cannot grow to large sizes and thus cannot explain the vertical redistribution of water observed about a week later. Therefore, those ensemble members, which represent 5 the highest BSR measurements best, lead only to shallow denitrification. Rather, it is the synoptic scale ice clouds with moderate ice concentrations which cause the most efficient dehydration. Finally, to demonstrate the need for the temperature correction depicted in Fig. 2b and c, we included model results based on the original ERA-Interim trajectories with 10 and without superimposed small-scale temperature fluctuations as dashed black lines in Fig. 6. The only simulation which produced any dehydration signal is the one combining heterogeneous nucleation with temperature fluctuations. However, the modeled dehydration is much smaller than that observed. 15 Denitrification plays an important role by slowing the conversion of chlorine radicals back into reservoir species. This process may effect an enhancement of ozone destruction and can lead to increased accumulated ozone losses over the course of the winter (e.g. Müller et al., 1994). Denitrification is particularly important in the Arctic with its warmer temperatures (e.g. Chipperfield and Pyle, 1998) and severe denitrification 20 has been a major factor in bringing about the record ozone loss in the Arctic winter 2010/2011 (Manney et al., 2011).

Relevance for denitrification
Recent observations suggest the possibility of heterogeneous ice nucleation on preexisting NAT particles. Pitts et al. (2011) observed an increase in synoptic-scale ice PSCs concomitant with decreasing number densities of NAT mixtures in January 2010.

25
Such a process would imply that the sedimentation of ice particles not only dehydrates but also denitrifies the stratosphere due to the removal of HNO 3 . Khosrawi et al. (2011) investigated this hypothesis and offered ice nucleation on NAT particles as a possible 27182 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | explanation for the low HNO 3 observations by Odin/SMR and Aura/MLS during the same winter. However, this modeling study reveals that dehydration and denitrification are not necessarily related to each other. The observed case does not significantly contribute to the overall denitrification for the following reasons: At the start of the simulations, before ice formation, the nucleation of NAT on foreign nuclei accounts for NAT 5 number densities of 10 −3 cm −3 , but the effective radius of NAT particles formed along the trajectories does not exceed 2 µm. The amount of HNO 3 condensing on these NAT particles is negligible, namely less than 10 % of the total available HNO 3 , while the major fraction of HNO 3 resides in the liquid droplets. This results in a depletion in the gas phase HNO 3 (see denoxification in Fig. 7a) and decelerates the growth of the NAT par-10 ticles (see Fig. 4 in Voigt et al., 2005). Next, ice nucleates, partly on the preexisting NAT, and HNO 3 also co-condenses on the growing ice particles, which sediment and dehydrate efficiently. However, the resulting denitrification is minor, as most of the mass of the falling particles is composed of water molecules. After the ice evaporated, NAT particles are released and would have the chance to denitrifiy the air, if they could survive 15 long enough. However, the air warms rapidly from T < T frost to T > T NAT (see Fig. 1) and does not stay for sufficient time in the interval T NAT − 5 K < T < T NAT − 2 K, which is most efficient for denitrification (Voigt et al., 2005). The resulting denitrification signal in Fig. 7b shows redistributions of HNO 3 over small height differences, but no strong, coherent denitrification. This impression is independent of the nucleation scenario or the 20 phase of the temperature fluctuation and, thus, very robust. Instead, the strong dentrification of the Arctic winter 2009/2010 occurred during the first half of January, i.e. before the onset of synoptic scale ice clouds, and was likely caused by NAT clouds downwind of mountain wave ice PSCs, which can act as mother clouds for so called NAT rocks (Fueglistaler et al., 2002). Evidence of this can be seen in the Odin/SMR and Aura/MLS 25 satellite measurements of HNO 3 shown in Fig. 3 of Khosrawi et al. (2011). In the beginning of January 2010, gas phase mixing ratios of HNO 3 decreased significantly at high altitudes and remained low until the end of February. Below 50 hPa, a concurrent HNO 3 increase is visible in the data, which allows the conclusion that a permanent,

27183
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | vertical redistribution of HNO 3 might took place. Minimum values of gas phase HNO 3 have been observed by Aura/MLS in the second half of January at the same time than CALIOP confirms the existence of synoptic-scale ice clouds. However, the renitrified layer below shows no further increase in HNO 3 . Instead, HNO 3 mixing ratios above 50 hPa increased again with proceeding time and increasing temperatures. Hence, we 5 expect denoxification and no additional de-and renitrification at this time in the winter. Recent CLaMS simulations show a similar picture with HNO 3 fluxes largest in the first half of January . For this reason, we cannot confirm the suggestion by Khosrawi et al. (2011), namely that the observed denitrification was linked to ice particle formation on NAT during the synoptic cooling event in mid-January. This case 10 study indicates that the conditions that led to the observed dehydration in the second half of January 2010 would not support sufficient growth of NAT particles to lead to significant denitrification.

Discussion and conclusions
Unprecedented de-and rehydration has been observed above Sodankylä during the 27184 It was demonstrated that heterogeneous nucleation of ice is essential to reproduce the observed de-and rehydration signatures. Even though ERA-Interim temperatures along the trajectories had to be lowered by up to −1.5 K in order to obtain agreement with the temperatures measured by the sondes, temperatures stayed clearly above T frost − 3 K, which is the temperature required for homogeneous nucleation of ice in 5 ternary solution droplets. Small-scale temperature fluctuations additionally lowered the temperature, caused higher supersaturations and therefore enabled the formation of ice clouds, even when homogeneous ice nucleation was the only allowed ice formation pathway. However, homogeneous nucleation at high supersaturations resulted in ice formation with the characteristics of wave clouds: high number densities of particles 10 remain too small to sediment and cannot explain the observed vertical redistribution of water. In contrast, heterogeneous ice nucleation takes place at lower supersaturations, causing a selective freezing of only a few ice crystals. Those particles can grow to sizes large enough (r 10 µm) to settle fast and reproduce the signatures of de-and rehydration. 15 Even though small-scale temperature fluctuations are not indispensable to achieve de-and renitrification in the present case, the resulting PSC backscatter is too low without small-scale temperature fluctuations and rapid cooling rates help to improve the agreement with the COBALD measurements. However, the two balloon soundings provide only snapshots of the atmosphere, which depend on the precise temperature 20 variations not only at the point of observation, but also upstream along the air parcel trajectories. Discrepancies between ERA-Interim temperature fields and measured temperatures were found, which we needed to correct. Whereas Sodankylä, located at the edge of the cold pool, is characterized by temperatures just below T frost and is therefore very sensitive to smallest temperature changes, the CALIOP measurements 25 show large areas deeper in the vortex with persistent synoptic-scale ice clouds. The observed dehydration above Sodankylä, not only on 23 January but also a few days earlier and later (as shown by Khaykin et al., 2013), is most likely caused by such large-scale fields of persistent ice clouds and not by the observed small-scale struc-27185 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | tures above Sodankylä. The fact that the reduction in H 2 O measured by CFH is not balanced by the rehydration layer below suggests that wind shear and subsequent mixing of air masses have affected the observed profile. An accurate estimate of these effects can only be made using a three-dimensional modeling approach.
Acknowledgements. This work was supported by the European Commission Seventh Frame- on synoptic scales, Atmos. Chem. Phys. Discuss., 13, 8831-8872, doi:10.5194/acpd-13-8831-2013, 2013   . Vertical red line: S1. Top panels: ERA-Interim temperatures (temperature correction applied) shown as absolute temperature (left panel) and relative to the frost point temperature (T frost ), which has been calculated from the climatological mean water vapor profile (right panel). Rows: backscatter ratios at 532 nm, ice and NAT number densities and radii. Columns: four different scenarios. Column 1: only homogeneous nucleation of ice; no superimposed temperature fluctuations. Column 2: same, but in addition allow for heterogeneous nucleation. Column 3: only heterogeneous nucleation of ice, but with superimposed small-scale temperature fluctuations. Column 4: with both. Homogeneous simulations include homogeneous ice nucleation and NAT nucleation on preexisting ice particles only. Heterogeneous model runs include in addition heterogeneous ice and NAT nucleation on foreign nuclei.