The relationship between polar mesospheric clouds and their background atmosphere as observed by Odin-SMR and Odin-OSIRIS

In this study the properties of polar mesospheric clouds (PMCs) and the background atmosphere in which they exist are studied using measurements from two instruments, OSIRIS and SMR, on board the Odin satellite. The data comes from a set of tomographic measurements conducted by the satellite during 2010 and 2011. The expected ice mass density and cloud frequency for conditions of thermodynamic equilibrium, calculated using the temperature and water vapour as measured by SMR, are compared to the ice mass density and cloud frequency as measured by OSIRIS. We find that assuming thermodynamic equilibrium reproduces the seasonal, latitudinal and vertical variations in ice mass density and cloud frequency, but with a high bias of a factor of 2 in ice mass density. To investigate this bias, we use a simple ice particle growth model to estimate the time it would take for the observed clouds to sublimate completely and the time it takes for these clouds to reform. We find a difference in the median sublimation time (1.8 h) and the reformation time (3.2 h) at peak cloud altitudes (82–84 km). This difference implies that temperature variations on these timescales have a tendency to reduce the ice content of the clouds, possibly explaining the high bias of the equilibrium model. Finally, we detect and are, for the first time, able to positively identify cloud features with horizontal scales of 100 to 300 km extending far below the region of supersaturation (>2 km). Using the growth model, we conclude these features cannot be explained by sedimentation alone and suggest that these events may be an indication of strong vertical transport.

The PMC microphysical retrieval and resulting uncertainties in cloud brightness and microphysical products are described in detail by Hultgren and Gumbel (2014). To summarise, the ice mass density is retrieved with an accuracy of around 5 ng/m 3 and the mode radius is retrieved with an accuracy of 10 nm, these estimates do not include errors resulting from the assumptions in the particle size distribution. 130 For the first time, the current study shows OSIRIS tomographic results from the northern hemisphere PMC season of 2011. This season was not included by Hultgren et al. (2013) and Hultgren and Gumbel (2014) due to retrieval stability issues for this season, which resulted in unphysical variations of the cloud scattering coefficient with wavelength. While this rendered about half of the orbits recorded during the northern hemisphere 2011 season unusable for this study, a filtering method was 135 developed to ensure the quality of the remaining half of the data. The method consisted in removing each pixel for which the difference in the retrieved scattering coefficient varied unphysically (by visual inspection) across the measured wavelengths. If a large number of pixels are flagged the entire orbit is excluded from the analysis with the result that 36 of the 89 orbits from 2011 are available for use in the current study. 140 In total, there are 35 tomographic orbits available that provide both SMR data in frequency mode 13 and usable OSIRIS data. Of these, 11 orbits are from July 2010, 12 from August 2010, 4 from June 2011 and 8 from July 2011. Due to the few orbits available from June, only data results from July and August will be shown in this paper when discussing seasonal differences in the data.

The thermodynamic model 145
In order to quantitatively compare the background atmosphere measured by SMR to the cloud properties measured by OSIRIS, we use a simple model which predicts the expected ice mass density from the observed background atmosphere by simply assuming thermodynamic equilibrium. Similar models have successfully been used previous studies (e.g. Russell et al. (2010) and Rong et al. (2012)) to investigate the relationship between PMCs and the background atmosphere. For the study 150 presented in this paper, the main advantage of such a simple model is that is directly maps the atmosphere observed by SMR into ice mass, without requiring any further assumptions. Furthermore, to estimate the time it would take the observed cloud pixels to reach thermodynamic equilibrium a growth model is used. Both these models are introduced in this section.

Growth model 155
For clouds to form in the atmosphere, the partial pressure of water vapour needs to exceed the saturation vapour pressure. Several expressions have been used to calculate the saturation pressure for water vapour over ice under mesospheric conditions (see e.g. Rapp and Thomas (2006)), and in this study we will use the formula from Murphy and Koop (2005), which is derived from a numerical 5 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-268, 2016 Manuscript under review for journal Atmos. Chem. Phys.
where T is the ambient temperature. If the mixing ratio of water vapour is given by Q gas and the mixing ratio of ice (defined as the ratio of water molecules in ice phase to the total number of molecules in the atmosphere) is given by Q ice , the total water pressure is

165
where p is the ambient pressure. In this study T and Q gas is taken from the SMR measurements and Q ice from OSIRIS. The pressure is taken from the MSISE-90 model (Hedin, 1991).
In general clouds will grow if the supersaturation ratio S ≡

Pgas
Psat is grater than one, and sublimate if the opposite is the case. Hesstvedt (1961) estimated the growth rate of single ice particles as

170
where r is the radius of the ice particle, ρ ice the density of ice, m h2o the molar mass of water, R the molecular gas constant, and f is a sticking parameter which we set to 0.83 following Gadsden (1998). Finally, since the growth/sublimation rate of a particle is proportional to its surface area a factor, Φ, defined as the ratio between the surface area of a non-spherical particle to that of a volume equivalent sphere, is included (Turco et al., 1982). For spheroids with and axial ratio of 2, 175 Φ = 1.095444. If P gas and P sat is known, Eq. 3 can be numerically integrated for each ice particle to determine r(t). Assuming that the total water content is conserved, the time to reach equilibrium size and the time to sublimate (r(t) = 0) can be estimated for single ice particles.
PMCs consist not of a single particle, but an ensemble of particles. However, the exact size distribution for PMC particles is highly uncertain. Thus, for simplicity, the particles are assumed all to 180 be spheroids with the same radius and with an axial ratio of 2. The radius used is the corresponding mode radius retrieved from OSIRIS, and the number of particles is then determined by ensuring that the ice mass density was equal to the ice mass density retrieved from OSIRIS. This single radius particle size distribution differs from the Gaussian size distribution used in the OSIRIS retrievals.
However, since Eq. 3 is independent of radius, the particle size distribution will not change with time 185 since all particles grow with the same speed. This means that the time it takes for a single radius size distribution to grow to equilibrium is equal to that of a Gaussian. This does not necessarily hold for sublimating clouds, as the smallest particles in the Gaussian distribution sublimates completely.
However, once this stage is reached, the total ice remaining cloud parcel is negligible, and thus this effect will not significantly affect the results presented in this paper.

190
Furthermore, the growth model presented does not take into account the fact that forming ice on a spherical surface (i.e. small particles) requires more energy than on flat surfaces. This is know as the Kelvin effect and can be accounted for by adjusting the saturation pressure according to: where P sat (∞) is the vapour pressure above a flat surface (i.e. Eq. 1), m the molecular weight of 195 water, σ the surface free energy (0.122 J/m 2 ), ρ the density of the ice particle, k the Boltzmann constant and r n the radius of the nucleation kernel. Since OSIRIS cannot measure particles of the size where this term becomes important, the Kelvin effect will generally not be used in this study.
We will only use it in the discussion of reformation of clouds in Sec. 4.

Equilibrium model 200
With enough time, the amount of ice in an air parcel will reach thermodynamic equilibrium, in this case the ice mass density is given by This value will be referred to as the ice mass density in equilibrium or simply "the equilibrium model". When this level is reached, all the excess water has been converted into ice, and the satura-205 tion ratio of the atmosphere is 1. This is rarely the case in the real atmosphere, and assuming thermodynamic equilibrium has been shown to overestimate the ice mass density. Hervig et al. (2009b) found a better agreement between measurements and the equilibrium model by replacing P tot with P gas in equation 5. Rong et al. (2014) found that the overestimation changes depending on the season, and introduced an additional scaling factor F to further reduce the ice mass, in particular late 210 in the season. In this paper we will not discuss any of these adjustments in depth, but only include results using the model from Hervig et al. (2009b) in Sec. 3.2 to contextualise the results presented.

OSIRIS clouds and background atmosphere
A first step in comparing the clouds measured by OSIRIS with their immediate background atmo-215 sphere is to look at the deviation from the mean background atmosphere in areas where clouds are detected. This is done by identifying all pixels in the OSIRIS measurements where an ice mass density > 0 ng/m 3 is retrieved. The mean background atmosphere, with and without clouds, at that latitude and altitude is then subtracted from each pixel. This method removes the effect of zonal differences in cloud and cloud-free pixels, and hence the anomalies in temperature and water vapour 220 are determined. Figure 1 shows mean and standard deviation of these anomalies for cloud and cloudfree pixels. The mean water vapour and temperature profiles across all latitudes are added to these anomaly values to provide context for the comparison.
In general the atmosphere is about 3-4 K cooler when clouds are observed, with a smaller difference in temperature at lower altitudes. During July a clear depletion of water vapour (1-2 ppmv) can clouds) and the total water content for pixels with clouds (ice + vapour) are equal between 80 and 84 km in July, while at higher altitudes, as well as all altitudes in the August data, the pixels with clouds show an elevated total water content compared to the pixels without clouds. The fact that 230 total water content is preserved in the core cloud altitude region (82-84 km) in July indicates that the growth from sub-visible to visible clouds occurs locally, and thus water vapour is transformed into ice when the conditions allow it. On the contrary, in August and at the highest/lowest altitudes, where enhanced total water content is observed, the water in the clouds is not simply depleted from the local surroundings of the clouds, but rather from other areas of the atmosphere. This can for example 235 be due to long formation times at the highest altitudes, or an indication that the clouds/background atmosphere have undergone transport separating the two.
Below PMCs a general enhancement of water vapour is expected, as the clouds sublimate and release available water into the atmosphere. However, Fig. 1 shows no sign of direct water vapour enhancement under the areas where clouds are detected. This was further investigated by comparing 240 the difference between the mean water vapour concentration and the concentration 0-1.5 km under the clouds, but no measurable enhancement was found. The reason for this is that while for some orbits, enhancements are found under the clouds, this enhancement is highly variable and can exist even when no cloud is observed. Thus, since the deposition of water vapour occurs at the end of the 8 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-268, 2016 Manuscript under review for journal Atmos. Chem. Phys. Published: 28 April 2016 c Author(s) 2016. CC-BY 3.0 License. life cycle of a clouds, there is no direct correlation between individual cloud observations and water 245 vapour enhancements below the cloud.

Water vapour and ice budget
Polar mesospheric clouds are highly sensitive to their background atmosphere, and small errors in the measured temperature or water vapour concentrations used in the calculations described in Sec.
2.2 can give large errors in the equilibrium ice content as well as growth and sublimation times.

250
Thus, before using the data to look at individual measurements, we investigate the ability of the data to reproduce the expected large-scale properties of the ice distribution.
The amount of ice expected in thermodynamic equilibrium can be compared to the ice retrieved from OSIRIS measurements. To take into account the sensitivity of the OSIRIS measurements, the ice mass density in pixels with ice mass density below a certain threshold is set to 0. The threshold is 255 determined by estimating the average ice mass density needed to reach a cloud brightness of 2·10 −9 m −1 str −1 at 83 km. This results in a value of 10.08 ng/m 3 so for simplicity the threshold is set to 10 ng/m 3 . The OSIRIS data are also filtered using the same method to ensure that the two datasets are consistent. Figure 2 depicts the overall ice distribution at different altitude. The left figure shows the mean retrieved ice mass density while the right shows cloud presence. In agreement with previous studies we find that assuming thermodynamic equilibrium overestimates the ice mass roughly by a factor of two across the entire region. This indicates that the whole measured region is, on average, highly supersaturated, with supersaturation ratios greater than 100 often occurring above 84 km. Even in 265 areas where clouds are measured, a considerable supersaturation (S > 10) is observed.

Vertical Comparison 260
Replacing P tot with P gas in Eq. 5, as in Hervig et al. (2009b), reduces the difference between the retrieved ice mass density and the one predicted by the equilibrium model, especially around the peak ice mass density. However, the discrepancy still remains, in particular at the highest altitudes.
Looking at the cloud presence, which we define as the frequency of the number of pixels with an 270 ice mass density above the aforementioned detection threshold, the agreement between the model and the measurements is better than for ice mass density, at least up to 83 km. This indicates that it is the strength of the clouds, rather than the frequency, that leads to the excess mass in the equilibrium model. Below 82 km there are actually more clouds in the OSIRIS data than the equilibrium model predicts, despite a lower total ice mass density measured. The reason for this is that at the lowest and ±20 % water vapour (solid).
In the OSIRIS data a large difference in cloud presence is seen between July and August, with a presence at the peak of 40% in July and 18% in August. This difference is significantly less in the 280 equilibrium model where the cloud presence in July is 60% and 50% in August. However, in both the measurements and the model the peak presence altitude is about 1 km higher in August than in July.

Comparison of Zonal Means
Cloud cover varies with latitude, with higher cloud coverage at higher latitudes than at lower. In Overall the results from the equilibrium model are in agreement with previous studies. It overes-295 timates ice mass by a factor of ∼2, and the overestimation is larger late in the season than in the middle, in agreement with Rong et al. (2014). The position of the peak ice mass density is reproduced at the same altitude as the measurement and the seasonal variation is seen with a higher ice mass density peak in August compared to July. The latitudinal variation of the integrated water content is also seen in both the equilibrium model and the measurements, while the cloud frequency in 300 the model shows less latitudinal variation than seen in the measurements, in particular in July.

Sensitivity analysis
The ice content predicted by the equilibrium model is very sensitive to changes in temperature and water vapour, and in order to evaluate how any errors in the SMR measurements influence this analysis, Fig. 2 includes the ice mass density calculated from all SMR measurements with an assumed 305 systematic error of ±5 K and ±20% water vapour concentration. The results from these perturbed calculations are shown by the thin gray lines. In particular, the strong dependency on temperature is seen as e.g. cloud presence at the cloud peak falls from 60% to 20% if the atmospheric temperature is 5 K warmer than the temperature retrieved from SMR. This radical change underlines how sensitive PMCs are to small changes in temperature.

310
Considering this sensitivity, and the size of the estimated systematic uncertainties in the retrieved water vapour (> 2 ppmv) and temperature (> 5 K) from SMR, the results presented in this study and 11 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-268, 2016 Manuscript under review for journal Atmos. Chem. Phys.  previous studies agree remarkably well. In fact, this consistency is a strong indication that the tomographic SMR data is free from substantial biases in temperature and water vapour concentration.

315
As the measurements capture clouds both in growing and sublimating phases, each cloud pixel observed by OSIRIS can be classified based on the state of the background atmosphere. If there is excess water available (P gas > P sat ), the cloud is growing and thus classified as such. If the amount of ice exceeds that expected from thermodynamic equilibrium, the cloud is classified as shrinking (P gas < P sat , but P tot > P sat ). Finally, if the cloud is outside the region of where any ice should 320 exist in thermodynamic equilibrium (P tot < P sat ) it is classified as forbidden. For this classification all detected OSIRIS clouds are included (not only those with an estimated ice mass density above the aforementioned threshold). Figure 4 shows the result from the classification. A majority of the clouds are in a growing phase above 82.5 km, while below this, a majority of the clouds are sublimating, i.e. either classified as 325 shrinking or forbidden. Due to the general downward motion of the ice particles, this altitude, where the number of growing clouds is equal to the number of shrinking clouds, is where the maximum ice mass density is found (see Fig. 2). If only the August data is examined the altitude where more than 50% of the clouds are sublimating is located 1 km higher, at 83.5 km. these clouds in the August data compared to July. This might be due to the limited data available in the tomographic dataset. However, it could also depend on the fact that the altitudes where these 340 clouds are found moves upwards with time during a season, and thus fewer of these clouds would be filtered out as near-or far-field clouds in the SOFIE measurements.

Investigation of individual clouds
To look further at how individual measurements of clouds relate to the measured background atmosphere, the equilibrium ice mass density is compared to the measured ice mass density for each orbit 345 separately. Figure 5 shows the measured temperature, water vapour, and calculated ice mass density from the model, together with the measured ice mass density from the OSIRIS data. The three orbits are meant to illustrate some typical features seen across all the orbits measured. The abscissa is given in Angle-Along-Orbit (AAO) which is the effective latitude along the orbit plane, with AAO = 0 • at the equator, and AAO = 90 • at the northernmost position of the satellite.

350
The top panels show an orbit recorded on July 15, 2010. The black curve indicates the area where the supersaturation is greater than 1, and if follows to a large degree 150 K contour line, which previously has been used as a proxy for supersaturation (e.g. López-Puertas et al. (2009)). The clouds measured by OSIRIS (thin contours) are mostly contained within this area. The right panel shows this in more detail with ice mass density from the equilibrium model as coloured contours and the 355 OSIRIS values as black contours. In general there is a good agreement between the two. However, some of the internal variability seen in the measured clouds is not reproduced by the equilibrium model.
While the top panels show clouds mainly confined within the area of supersaturation, the second row shows an example where clouds are present outside the region of supersaturation with one 360 strong cloud located at 85 • AAO and a thin patch at 98 • AAO. These clouds are (assuming that the retrieved water and temperature is correct) undergoing sublimation. Using Eq. 3, the sublimation time for these clouds is estimated to be less than 20 minutes for the strong cloud, and 10 minutes for the thin cloud. Thus the conditions around these clouds must have undergone a rapid change to allow a detection outside the saturated region. 14 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-268, 2016 Manuscript under review for journal Atmos. Chem. Phys. Published: 28 April 2016 c Author(s) 2016. CC-BY 3.0 License.
The lower panels show another case of clouds outside the saturated region, with cloud cover down to 80 km at 87 • AAO. These clouds were identified in Hultgren and Gumbel (2014) as regions below the typical cloud brightness peak, with large particles (>80 nm mean radius) "raining out" of the saturated region. To refine this hypothesis, the sublimation time was calculated at different altitudes of the clouds. The resulting sublimation time is around 2 h at 82.5 km, but is rapidly reduced to less 370 than 10 minutes below 81.5 km. Typical sedimentation speeds at 81.5 km for a 100 nm particle is on the order of 0.1 m/s (Turco et al., 1982) thus falling from 82 to 80 km would take more than 5 hours.
This indicate that these vertical structures in the clouds cannot come from sedimentation alone.
Horizontal transport (zonal) of the clouds could explain these areas far below the region of supersaturation. However, due to their small sizes, ice particles tend to be transported with the air, hence 375 staying within the same air parcel. There is however a possibility for such vertical features to arise due to wind shear. If there is a strong horizontal gradient in the temperature the supersaturated region can extend to significantly lower altitudes just outside the orbital plane. If the horizontal winds are significantly smaller inside this region of supersaturation than below it, clouds sedimenting out of this region cloud be blown into the orbit plane. The result would be observed as an apparent vertical 380 separation between the S > 1 line and the cloud bottom, which would be larger than the true vertical separation. However, investigating orbits preceding and following these events, we do not detect any such strong horizontal gradients in temperature.
Another explanation for these structures is strong vertical winds. Indeed similar vertical cloud features have been measured by LIDARs (Kaifler et al., 2013) and linked to gravity waves prop-385 agating through the clouds. Kiliani et al. (2013) also note that ice particles can experience strong downdrafts at the lower edge of the clouds. Using the sublimation time, we estimate that the vertical transport speed needed is on the order of 1-3 m/s, which is larger than what is suggested in Kaifler et al. (2013) and Kiliani et al. (2013) where the vertical transport speeds reported are ∼0.1 m/s. Thus, if the observed cloud is due to vertical winds, it is a particularly case with particularly strong 390 downdraft. Such vertical windspeeds (1-5 m/s) have been observed lasting over 1 h in the mesopause region by ground based VHF radar (Rapp and Hoppe, 2006), and thus we believe such winds to be a plausible cause of the observed features.
Finally, it should be mentioned that since the horizontal field of view of the two instruments differ, we cannot rule out that these low lying clouds are caused by dynamical features resolved by SMR sizes unless these favourable conditions persist. Furthermore, as the largest ice particles sediment downward, only small, and hence sub-visible, particles will remain at the highest altitudes. Both these effects can lead to areas were cloud presence is predicted by thermodynamic equilibrium, but 405 no cloud is observed. This discrepancy is also reflected in the differences in cloud presence and frequency between the equilibrium model and measurements, shown in Fig. 2 and Fig. 3. 16 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-268, 2016 Manuscript under review for journal Atmos. Chem. Phys. Published: 28 April 2016 c Author(s) 2016. CC-BY 3.0 License.

Discussion on the lifetime of clouds
Since the background atmosphere of the clouds is constantly changing, the observed cloud pixels are always growing or sublimating. To determine to what degree the clouds are in equilibrium (or not) 410 with their environment, we use the growth model described in Sec. 2.2. For each detected cloud, Eq.
3 is numerically integrated to give an indication of the temporal evolution of the detected cloud. As the initial condition all ice particles have a radius of 5 nm and Q gas ( is the amount of water in ice phase if the cloud consisted of 5 nm particles. This is scaled such that when the particles have grown to the mode radii measured by OSIRIS, the modelled Q ice 415 is equal to the measured Q ice . Figure 6 shows the hypothetical evolution of three cloud pixels from orbit 56735. The example is taken from the cloud located at 92 • AAO at three different altitudes. The figure shows two growth regimes, a fast growth in the beginning due to the large amount of excess water, and a slower growth as the cloud particles asymptotically grow towards their equilibrium sizes.

420
The measured mode radii of the clouds are shown by the black crosses in the figure, and thus give an indication where along the growth curve each cloud pixel is. From the figure it is clear that the equilibrium radii decrease with altitude, with equilibrium radii above 100 nm at the lower edge of the cloud. At the highest altitude, the time it takes for a particle to reach equilibrium is longer than in the middle of the cloud. This is consistent with the discussion in Sec. 3.1, where no local depletion 425 of water vapour was seen at the highest altitudes indicating slower cloud growth.
In order to quantify how far a given cloud pixel is from equilibrium, the time from the current radius to reaching 90% of equilibrium radius is calculated. Figure 7 shows the histograms for all detected cloud within the core cloud region (82-84 km) in July. It can be seen that a large portion of the cloud pixels are less than 15 minutes from equilibrium. In total, over 50 % of the clouds are less 430 than 1 hour from equilibrium, and the mean time to equilibrium is around 2 hours. In terms of radius, most of the detected cloud pixels are between 0.6-1 times their equilibrium radius, with a mean value of 0.86. Since the volume of an ice particle scales as r 3 , while the growth is approximately linear in time, this rather modest distance from equilibrium (in terms of both time and radius) can lead to large differences in the observed ice mass density.

435
While Fig. 7 only considers the pixels where clouds are expected according to thermodynamic equilibrium, Fig. 8 shows the time to reach equilibrium for pixels where no cloud is expected, i.e. the sublimation time. The histogram for all these "forbidden" pixels is shown in the left plot in Fig. 8.
It can be seen that a large number of the clouds sublimate fast, and more than 50% are gone after 2.1 hours (dashed vertical line). Thus, as discussed in Sec. 3.4 these clouds indicate areas where a rapid 440 change of the background atmosphere has occurred, either through transport or temporal changes.
As the clouds seen in a sublimating phase can disappear due to atmospheric variability within hours, it is of interest to estimate the time it would take to reform these clouds. To estimate this, we look at the clouds observed in a growing phase, and calculate the time it takes for a 5 nm particle to 17 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-268, 2016 Manuscript under review for journal Atmos. Chem. Phys.  grow to the measured radii. The right plot in Fig. 8 shows the time to reach the measured state for all 445 detected growing clouds, and with a median of 3.2 hours this is significantly longer that the median sublimation time of 2.1 hours in Fig. 8. This asymmetry in cloud destruction and reformation might indeed be one of the reasons why assuming thermodynamic equilibrium overestimates the ice mass density by a factor of two as discussed in Sec. 3.2.

Discussion and conclusions 450
In this paper we have compared measurements of cloud ice mass density of polar mesospheric clouds from Odin-OSIRIS with simultaneous measurements of water vapour and temperature from Odin-SMR. The comparison was done on a set of special tomographic measurements performed by Odin, and data from July 2010 and 2011 as well as August 2010 were analysed. We compared the measurements using a model assuming thermodynamic equilibrium. Consistent with previous measurements 455 we find that many general features of the clouds such as the altitude of maximum ice mass density, as well as the latitudinal and seasonal variation of the ice column, are reproduced by the equilibrium model.
Though many large scale features are well represented by the equilibrium model, the model produces too much ice compared to the OSIRIS observations. This discrepancy has been reported in 460 previous studies. In this study we suggest a possible explanation for this discrepancy. By applying a simple growth model to the clouds observed by OSIRIS we are able to estimate the time to sublimate for the clouds in a shrinking phase, and compare this to the time it would take to grow to the clouds observed in a growing phase. The median time for sublimation in the core cloud region, at 82-84 km in July, was only 2.1 hours, while the expected time of regrowth was 3.2 hours. This means that 465 18 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-268, 2016 Manuscript under review for journal Atmos. Chem. Phys. Published: 28 April 2016 c Author(s) 2016. CC-BY 3.0 License. temperature fluctuations on these time scales have a tendency to reduce the total ice compared to an equilibrium situation. The long reformation time also explains why many areas with high amounts of excess water are seen without the presence of PMCs.
One important factor not considered in the discussion above is that although the median reformation time from 5 nm to the observed radii is 3.2 hours, this reformation time would be longer if the 470 Kelvin effect was included. In fact, rerunning the growth model with the Kelvin effect included we conclude that 20% (433) of the clouds included in the right plot of Fig. 8 would not even be able to reform from 5 nm. This would further increase the depletion of ice expected from short temporal variations in the temperature.
It should be noted that these results do not imply that atmospheric variability in general leads to 475 fewer PMCs. On the contrary, if the mean measured atmosphere is used as an input to the equilibrium model, the amount of ice predicted is significantly less than if the measured atmospheric variability is included. Our results only suggest that atmospheric variability occurring on smaller time scales than ∼ 3 hours in the core cloud region (82-84 km in July) has a tendency to reduce the visible ice.
This statement is in agreement with modelling studies. Jensen and Thomas (1994), for example, 480 show that temperature fluctuations with periods of 1 h reduce the observed albedo in the modelled PMCs, and Rapp et al. (2002) conclude that short term variations in temperature reduces visible ice, while variations on time scales larger than 6 hours enhance PMC production.
In the dataset presented we also detect "forbidden" clouds, far below the region of supersaturation in 5 of the 35 analysed orbits. These have been seen in previous studies using limb sounding tech-485 niques, however due to the tomographic nature of the current measurements we are for the first time able to distinguish the low lying clouds from far-and near-field clouds. These cloud regions tend to have small number densities and consist of large particles. The estimated sublimation time for these cloud pixels are on the order of 10-30 min, and since they are found up to 2 km below the region of supersaturation it is clear that sedimentation alone cannot cause this phenomenon. We suggest that 490 these clouds are linked to strong downdrafts. However, further investigation will be needed to verify this.
Indeed the tomographic measurements by Odin were planned such that they, as much as possible, would coincide with measurements taken by the Cloud Imager and Particle Size Experiment (CIPS) on board the AIM satellite. Thus, comparing the results of this study, both in terms of cloud 495 lifetimes and cloud classification, with CIPS images captured a short time before or after the Odin measurement would help refine and verify the conclusions drawn in this paper.