Numerical simulations of the three-dimensional distribution of polar mesospheric clouds and comparisons with Cloud Imaging and Particle Size (CIPS) experiment and the Solar Occultation For Ice Experiment (SOFIE) observations



[1] Polar mesospheric clouds (PMC) routinely form in the cold summer mesopause region when water vapor condenses to form ice. We use a three-dimensional chemistry-climate model based on the Whole-Atmosphere Community Climate Model (WACCM) with sectional microphysics from the Community Aerosol and Radiation Model for Atmospheres (CARMA) to study the distribution and characteristics of PMCs formed by heterogeneous nucleation of water vapor onto meteoric smoke particles. We find good agreement between these simulations and cloud properties for the Northern Hemisphere in 2007 retrieved from the Solar Occultation for Ice Experiment (SOFIE) and the Cloud Imaging and Particle Size (CIPS) experiment from the Aeronomy of Ice in the Mesosphere (AIM) mission. The main discrepancy is that simulated ice number densities are less than those retrieved by SOFIE. This discrepancy may indicate an underprediction of nucleation rates in the model, the lack of small-scale gravity waves in the model, or a bias in the SOFIE results. The WACCM/CARMA simulations are not very sensitive to large changes in the barrier to heterogeneous nucleation, which suggests that large supersaturations in the model nucleate smaller meteoric smoke particles than are traditionally assumed. Our simulations are very sensitive to the temperature structure of the summer mesopause, which in the model is largely dependent upon vertically propagating gravity waves that reach the mesopause region, break, and deposit momentum. We find that cloud radiative heating is important, with heating rates of up to 8 K/d.

1. Introduction

[2] Polar mesospheric clouds (PMC), also known as noctilucent clouds, were first observed in 1885 a couple of years after the eruption of Krakatoa [Schröder, 2001]. They are ice clouds [Hervig et al., 2001] that exist at ∼83 km, below the polar summer mesopause, where extremely cold temperatures of 130 K or less exist [Höffner and Lübken, 2007; Lübken, 1999; Lübken et al., 2008]. At moderately high latitudes (>50°), noctilucent clouds are visible from the ground at twilight during the summer months because they scatter sunlight in an otherwise dark sky. Thomas [1996, 2003] and Olivero and Thomas [2001] have suggested that PMCs could be sensitive indicators of climate change in the middle atmosphere. This idea is controversial [von Zahn, 2003], but recent satellite studies do support a long-term trend in PMC brightness [DeLand et al., 2006, 2007]. To determine whether PMCs are responding to climate change, any anthropogenic trends must be separated from natural variation, which may be influenced by the solar cycle [DeLand et al., 2006], volcanoes [Mills et al., 2005], and other human impacts such as water vapor and metal particles from the space shuttle [Stevens et al., 2003]. This work presents a three-dimensional model with detailed microphysics to help understand these issues.

[3] Because of the remote and extreme conditions of the polar mesosphere, it has been difficult to make measurements or perform laboratory investigations to determine the details of the cloud microphysics that are responsible for the formation of PMCs. It has been assumed that the ice particles form via heterogeneous nucleation; however, the actual nucleation process is not known. Rapp and Thomas [2006] survey the possibilities, with the leading candidates being meteoric smoke, ion clusters, and sulfate aerosols. Many simulations of PMCs [Jensen, 1989; Lübken and Berger, 2007; Rapp et al., 2002; Rapp and Thomas, 2006; Turco et al., 1982; von Zahn and Berger, 2003] have assumed heterogeneous nucleation upon meteoric smoke and use the smoke distribution of Hunten et al. [1980]. Recent simulations of the formation of meteoric smoke [Bardeen et al., 2008; Megner et al., 2006; Megner et al., 2008] indicate that because of particle depletion in the summer polar mesopause region by the meridional circulation, the concentration of smoke particles traditionally thought to be large enough to function as ice nuclei for PMCs is much less than indicated by Hunten et al. [1980]. Laboratory studies of the heterogeneous nucleation of ice at temperatures down to 153 K by Trainer et al. [2009] indicate that the contact angle decreases with decreasing temperature, reducing the nucleation rate at the low temperatures present in the summer polar mesopause. These issues have cast doubt on the assumption that heterogeneous nucleation on meteoric smoke is a viable mechanism for the production of PMC ice crystals; however, the spatial and temporal variability of temperature, water vapor, or meteoric smoke distributions may still allow sufficient nucleation for PMC formation. Recent observations by Hervig et al. [2009a] have detected the presence of meteor smoke in the polar mesosphere, and its seasonal variation matches that predicted by the model.

[4] We use a three-dimensional chemical climate model with sectional microphysics to investigate the formation of polar mesospheric clouds nucleated by meteoric smoke. Sectional microphysics means that the size distributions of the cloud and smoke particles are resolved into several discrete size bins. These simulations are compared to observations from the 2007 Northern Hemisphere PMC season from the Aeronomy of Ice in the Mesosphere (AIM) mission [Russell et al., 2009]. These are the first simulations of PMCs that include the impact of the meridional circulation on the distribution of meteoric smoke particles and are among the first efforts to compare PMC simulations with the AIM data. Section 2 contains a description of the model we used, results are presented in section 3, and section 4 contains a summary of these results along with a discussion of their implications.

2. Model Description

[5] We have combined the Whole Atmosphere Community Climate Model (WACCM) [Garcia et al., 2007], a chemistry-climate model, and a sectional microphysics model based on the Community Aerosol and Radiation Model for Atmospheres (CARMA) [Jacobson et al., 1994; Toon et al., 1988; Turco et al., 1979] to form the WACCM/CARMA model [Bardeen et al., 2008] used in this study. We have extended our WACCM/CARMA meteoric smoke model [Bardeen et al., 2008] by adding additional microphysical processes for Brownian diffusion of particles at high altitudes as well as the nucleation, condensational growth, and sublimation of ice particles in the mesosphere. As before, advection, eddy diffusion and wet deposition of meteor smoke particles are handled by WACCM, while sedimentation and coagulation are the responsibility of CARMA. For all cases, we use a horizontal resolution of 4° × 5° and a vertical grid with 125 levels and ∼0.35 km spacing near the mesopause. WACCM incorporates a gravity wave parameterization [Garcia et al., 2007; Sassi et al., 2002] to include the effect of vertically propagating gravity waves generated in the troposphere upon the model's atmospheric circulation. This parameterization has two components, one to represent orographic waves and another that provides a specified background spectrum to represent waves generated by other sources such as convection and frontal systems. The source of the orographic waves, the launch direction of the background waves, and the filtering of the waves as they propagate vertically are all dependent on the meteorology in the model; however, the background spectrum is prescribed and does not vary with changes in climate. The default tuning of the gravity wave algorithm is a compromise between generating a good climatology in the both stratosphere and the mesosphere. We conducted several simulations with different choices for the source strength and efficiency of the background gravity waves to better simulate the temperature structure of the summer polar mesopause and to investigate the sensitivity of the model to changes in the gravity wave tuning. We discuss these issues in more detail in section 3.2.

[6] CARMA tracks meteoric smoke particles in 28 size bins with radii from 0.2 to 100 nm with the volume ratio doubling between bins, and ice particles in 28 size bins with radii from 0.2 nm to 1 μm with a volume ratio of 2.6 between bins. Because CARMA fully interacts with the water vapor from the WACCM model, for each model time step the water vapor from WACCM is passed into CARMA and any changes to water vapor from deposition or sublimation are then passed back to WACCM. The total amount of water vapor in WACCM is then adjusted accordingly. For some simulations, we have coupled the microphysics with the WACCM radiation code to account for the effect of radiative heating of the ice particles on the particle growth rate and the atmospheric thermal structure. The initial condition for all of these simulations uses a steady state meteoric smoke distribution from the fourth year of a WACCM/CARMA meteoric smoke simulation [Bardeen et al., 2008]. The supply of smoke particles is refreshed with a horizontally and temporally uniform meteoric influx of ∼16 kt/yr [Hughes, 1978], using an ablation profile by Kalashnikova et al. [2000], which has a peak production rate at ∼85 km.

2.1. Brownian Diffusion

[7] For small particles in the upper atmosphere, Brownian diffusion plays a minor, but nonnegligible role in the distribution of particles [Bardeen et al., 2008]. In this study, in addition to sedimentation, we have included the solution of the diffusion equation

equation image

in CARMA's vertical transport calculation, where D is the Brownian diffusivity, ρa is the density of the atmosphere, and N is the number of particles. The Brownian diffusivity is

equation image

where k is the Boltzman constant, T is the temperature, Cc is the slip correction factor, μ is air viscosity, and r is the particle radius. The slip correction factor compensates for noncontinuum effects upon Stokes' Law and is given by

equation image

where Kn is the Knudsen number, the ratio between the mean free path (λ) and the particle radius. The coefficients for the slip correction factor are by Kasten [1968] as suggested by Toon et al. [1989]. For small particles in the mesosphere, the Knudsen number will be very large and Cc will be ∼1.246 Kn, resulting in a diffusivity of

equation image

where υt is the mean thermal velocity of an air molecule. The strong radius dependence of the diffusivity shows that Brownian diffusion will be much more important for the smaller particles than for the larger ones.

2.2. Heterogeneous Nucleation

[8] In the presence of ice nuclei (in this case meteoric smoke particles), heterogeneous nucleation will be preferred over homogeneous nucleation. We use the classical theory for heterogeneous ice nucleation, assuming that surface diffusion is the dominant process as formulated by Keesee [1989]. The nucleation theory assumes that a germ of ice forms upon a spherical substrate particle. For the particle to grow, the germ must reach a critical size to overcome the energy barrier imposed by the Kelvin effect. The contact angle, θ, describes how well the nascent ice crystal interacts with the surface of the substrate. It is defined by Young's relation, where

equation image

is the ratio of the difference between surface tensions of the air/substrate, condensate/substrate interfaces, and the surface tension of the condensate/air interface. For heterogeneous nucleation, the critical radius and critical free energy are given by

equation image

where M is the molecular weight of water, R is the gas constant, T is temperature, ρ is ice density, and S is saturation ratio. The nucleation rate, J, is given by

equation image

where rN is the radius of the ice nuclei, A is the absorbed water vapor concentration, d is the mean jump distance of the absorbed molecule, v is the vibrational frequency of the absorbed molecule against the surface, ΔGdes is the free energy of activation for desorption, and ΔGsd is the free energy for surface diffusion. For mesospheric conditions, Keesee [1989] assumes values of d = 0.1 nm, v = 1013 s−1, ΔGdes = 2.9 × 10−13 erg, and ΔGsd = 2.9 × 10−14 erg. The Zeldovich factor, Z, adjusts for nonequilibrium effects on the i-mer concentrations and is given by

equation image

The factor f, which scales the free energy to adjust for the cap geometry, is defined by

equation image

Given the lack of sufficient measurements relevant to the materials, temperatures, pressures, and water vapor concentrations in the summer polar mesosphere, there are large uncertainties in the nucleation rate calculation. We conduct sensitivity tests with different contact angles to explore the response of these simulations to different nucleation rates.

2.3. Condensational Growth and Sublimation

[9] Depending on the saturation of water vapor in the atmosphere with respect to ice, water vapor may deposit itself on existing ice particles, or ice particles may sublimate, releasing water vapor to the atmosphere. We follow the formulation of Toon et al. [1989] for calculating the particle growth rate in the presence of radiative heating. The equation for the growth rate of ice particle radius is

equation image

where nsat is the number density of water vapor at saturation, S is the saturation ratio (equation image/nsat), equation image is the number density of water vapor in the atmosphere, Ak is the Kelvin correction, Qrad is the particle radiative heating rate, Le is the latent heat of sublimation, and NA is Avagadro's number. D is the diffusion coefficient, K is the thermal conductivity, and Fv and Ft are ventilation factors to adjust the diffusion coefficient and thermal conductivity for the effects of sedimentation of the particles as defined by Toon et al. [1989]. For small particles in the mesosphere, the Knudsen number is very large, and g0 and g1 simplify to

equation image

where υt is the mean thermal velocity of an air molecule, υw is the mean thermal velocity of a water molecule, αc is the sticking coefficient, and αt is the thermal accommodation coefficient. We assume that αc = 0.93 and αt = 1.0. In the kinetic limit, the growth rate is independent of the diffusion coefficient and the thermal conductivity. The growth rate in terms of the mass of ice is given by

equation image

2.4. Vapor Pressure of Water

[10] For psat, the saturation vapor pressure of water over ice, we use the expression of Murphy and Koop [2005],

equation image

to determine the saturation vapor pressure at the low temperatures found in the polar summer mesosphere. This expression does not consider cubic ice, which may occur at typical PMC temperatures. The vapor pressure over cubic ice would be ∼10% higher than for hexagonal ice [Murphy and Koop, 2005; Shilling et al., 2006].

2.5. Radiative Heating

[11] The absorption and emission of radiant energy by the ice particles can change their temperature and affect both the particle's growth rate and the atmospheric temperature.

[12] The net radiative heating rate for the particle, Qrad, is calculated as

equation image

where kabs is the absorption coefficient, Jλ is the mean intensity of the incident radiation, and Bλ is the Planck function at wavelength λ. The integral is performed over wavelengths from 0.2 to 330 μm to cover both incoming solar and upwelling longwave radiation. The particle temperature is determined by solving the steady state energy balance equation for the particle, which includes terms for the net radiative heating, latent heating, and heat conduction with the atmosphere,

equation image

where Tp is the temperature of the particle and Ta is the temperature of the atmosphere. Solving for Tp results in

equation image

The heating of the atmosphere caused by the flow of heat between the particles and the atmosphere is then

equation image

where Cp is the specific heat of the atmosphere, ρa is the density of the atmosphere, and np is the number of particles. To solve equation (16), we first assume that Tp = Ta, and then iterate over equations (10), (12), (14), and (16) until Tp converges.

[13] It is difficult to extend the WACCM radiation code to add other aerosols into the shortwave and longwave radiation calculations; however, PMC optical depths are small and do not affect the overall radiative transfer calculation. While the WACCM radiation code is a band model, it reports only the broadband fluxes. To solve for the mean intensity, we assume the spectral pattern of the energy incident upon the particle is in the form of a Planck blackbody with the same broadband flux as reported by WACCM. For the short wave, we assume a blackbody temperature of the sun (5780 K) and scale the mean intensity to conserve the total flux. For the longwave, we select a blackbody temperature that has the same total flux as reported by WACCM. Gaps in the spectrum caused by absorption of energy within the atmosphere, particularly in the upwelling longwave radiation, will cause errors in the estimating Jλ, the mean intensity incident on the particle. The particle's absorption coefficients are calculated by using Mie theory with refractive indices for water ice compiled by Warren [1984] and by Toon et al. [1994]. Uncertainties in the particle shape and the refractive indices of ice at the cold temperatures in the summer mesopause could lead to errors in the determination of the particle temperature. In future versions of the model, we plan to fully integrate the ice particles into the WACCM radiation calculation so that more accurate estimates of the impact of radiative heating can be done.

3. Results

[14] We conducted a control run, a simulation that is similar to 2007 conditions, and four studies to investigate the sensitivity of the simulations to the gravity wave parameterization, the contact angle, latent and radiative heating, and the phase of the solar cycle. The specifications for these runs are summarized in Table 1. In section 3.1, these results are compared to observations from the CIPS experiment [McClintock et al., 2009] and SOFIE [Gordley et al., 2009] for the 2007 Northern Hemisphere PMC season. In the control simulation, the WACCM gravity wave parameterization was tuned to produce a temperature profile at 70°N that is similar to the climatology of Lübken [1999]. We discuss the sensitivity of the model to different gravity wave tunings in section 3.2. For the control run, the cosine of the contact angle is assumed to be 0.95; however, in section 3.3 the sensitivity of the model to a range of contact angles is explored. Latent heating of the atmosphere from the formation of PMC particles is included in the control run, but because of the large uncertainties in the mean intensity and the optical properties of the ice particles, radiative heating of the cloud particles is not included. In section 3.4, we look at the sensitivity of the model to different assumptions about latent and radiative heating. The control run is configured for solar minimum conditions, which last began in 2007. Section 3.5 contains results for a similar simulation using solar maximum conditions.

Table 1. Specification of the Different WACCM/CARMA Simulations Used in This Studya
RunSolar CycleGravity Wavecos(θ)Heating
  • a

    Includes the control run and sensitivity studies to examine the response to changes in the gravity wave parameterization, contact angle, particle heating physics, and phase of the solar cycle. The gravity wave parameterization is controlled by the source strength (τb*) and the efficiency of the wave breaking to drive the circulation (effgw). The contact angle, equation image, influences the nucleation rate.

Gravity WaveMinimumVariousVarious0.95Latent
Contact AngleMinimum0.00150.0875VariousLatent
Particle HeatingMinimum0.00150.08750.95Various
Solar MaximumMaximum0.00150.08750.95Latent

3.1. Control Run

[15] Comparisons to satellite observations require taking into account the local time of the satellite overpass. Figure 1 shows the local time response of the peak extinctions of polar mesospheric clouds (PMCs) in the control run compared to the peak extinctions in lidar observations at the Arctic Lidar Observatory for Middle Atmosphere Research (ALOMAR) by Fiedler et al. [2005]. The peak backscatter (βmax) for a particular lidar profile is the largest value of the lidar backscatter detected within that vertical profile. The model shows less variation in the mean of the peak backscatter than is seen in the observations; however, the observation and its variability are within the one sigma variation of the model. The model appears to be missing peaks at 0800 and 2000 LT that may that may be due to errors in the diurnal variation in temperature in the model. However, the observed and modeled mean backscatter at 1200 and 2400 LT, which are used for satellite comparisons with CIPS and SOFIE, are similar to each other. WACCM is a free running climate model, so we don't expect a particular year of a climate run to exactly match a specific year of observations. Instead, we compare our simulations to seasonal averages of CIPS data and to probability distributions and seasonal averages of SOFIE data.

Figure 1.

The variation of polar mesospheric clouds peak backscatter from a 532 nm lidar with local time from a 7 year climatology (1997–2003), measured from 1 June to 15 August at the Arctic Lidar Observatory for Middle Atmosphere Research (ALOMAR) (69°N, 16°E) by Fiedler et al. [2005] and as calculated from the control simulation using data that are at the ALOMAR latitude and within 15° in longitude. The shading indicates one standard deviation from the mean of the model.

3.1.1. Comparison With CIPS

[16] CIPS is an ultraviolet imager aboard the AIM satellite, observing at a wavelength centered at 265 nm and using four nadir-oriented cameras (PX, PY, MX, and MY) [McClintock et al., 2009]. It acquires images at approximately seven viewing angles, covering over a 90° range of scattering angles, so that it can retrieve both the cloud albedo and the scattering phase function. For the model, we assume spherical particles, a scattering angle of 90°, and a nadir view and use Mie theory to calculate the cloud albedo at a wavelength of 265 nm. The CIPS retrieval uses multiple scattering angles to separate the cloud albedo from the background, and the resulting cloud albedo has been referenced to this same viewing geometry. To approximate the overpass time of CIPS, we use simulated cloud properties from 1200 LT (±1.5 h). Equatorward of latitude ∼78°N, this corresponds to the local time of the CIPS descending node; however, for this analysis we have also used data from the ascending node, which corresponds to a local time of midnight. The seasonally averaged albedo (from 30 d before solstice to 70 d after solstice) for clouds with an albedo greater than 1 × 10−6 and 7 × 10−6 sr−1 from the control run and CIPS level 4, version 3.12 data, is shown in Figure 2. The albedo values and the latitudinal pattern are very similar between CIPS and the model; however, the model has more dim clouds and higher variability at low latitude, where the CIPS retrieval is not as sensitive. The longitudinal variation is different between the model and CIPS. A difference in the spatial pattern is expected, since the model is free running and does not have the same wind and temperature patterns as occurred in 2007. There should be similar amounts of spatial variability between CIPS and the model; however, the model appears to have more variability. There is more similarity between the model and CIPS when the higher cloud threshold is used, where CIPS should be able to detect most of the clouds and therefore the comparison is more robust than when using the lower cloud threshold. As will be shown in section 3.1.2, SOFIE observations are more sensitive to PMCs than CIPS, and the excellent agreement of the probability distributions for ice water content (IWC), extinction, and mass density between the model and SOFIE suggest that differences between the model and CIPS at low albedos are more likely caused by a lack of sensitivity in CIPS measurements than a bias in the model toward the production of dim clouds.

Figure 2.

Polar maps of the average albedo at 265 nm corresponding to a nadir viewing angle and a scattering angle of 90° for clouds with albedos larger than (top) 1 × 10−6 sr−1 and (bottom) 7 × 10−6 sr−1, for (left) the WACCM/CARMA simulation, and (middle) CIPS experiment data from the 2007 Northern Hemisphere season. (right) Differences between the simulation and measurements. Locations of latitudes 75°N, 60°N, and 45°N are indicated by innermost, middle, and outermost dashed circles, respectively.

[17] The CIPS cloud detection sensitivity varies with solar zenith angle (SZA). This occurs for two reasons: as the SZA increases, the background Rayleigh scattering signal decreases, while at the same time the CIPS sunward (PX) camera samples more forward-peaked scattering angles. Both of these factors enhance the discrimination between cloud and background contributions to the measured scattering profile, and hence increase detection sensitivity. Sensitivity analyses using simulated CIPS data indicate that a 2 × 10−6 sr−1 cloud will be detected with 50% probability at 85° SZA, while an equal detection rate at a 35° SZA requires a 7 × 10−6 sr−1 cloud. Similarly the cloud brightness threshold for 100% detection at 85° and 35° SZA is 6 × 10−6 and 25 × 10−6 sr−1, respectively. The relationship between SZA and latitude changes from orbit to orbit; however, a 35° SZA corresponds to about 55° latitude, and the highest latitudinal extent for CIPS of ∼82° latitude occurs near a 70° SZA. Therefore, robust cloud detections are obtained poleward of ∼65° latitude with nearly 100% sensitivity for clouds brighter than 10 × 10−6 sr−1, and 50% sensitivity for clouds brighter than ∼5 × 10−6 sr−1. Clouds are detected at lower latitudes, but with higher uncertainty. Because of these sensitivity and uncertainty issues, it is difficult to draw conclusions from the CIPS data alone about the validity of the dimmer clouds that often form at lower latitudes in the model.

[18] The operational CIPS level 4 retrieval requires that the cloud be detectable at multiple scattering angles so that the phase function can also be retrieved. Benze et al. [2009] have an alternate CIPS retrieval that just uses the most forward scattering angle, which is more sensitive to the dim clouds than the other scattering angles are, particularly at lower latitudes. However, since this retrieval does not get the phase function, assumptions must be made about the effective radius and size distribution to reference the albedo back to a nadir view at a 90° scattering angle. Figure 3 shows the frequency of clouds with a threshold greater than 1 × 10−6 and 7 × 10−6 sr−1 from the control run and the two CIPS retrievals as a function of time and latitude. The patterns are similar, but the overall magnitude of the cloud frequency in the simulation is generally higher and has more coverage at lower latitudes than either CIPS retrieval. However, the model matches better with the Benze et al. retrieval than the operational retrieval. The season appears to start and end a little earlier in the model; however, there are no data from CIPS until 27 d before solstice, that is, −27 d relative to solstice (DRS). In the observational data, there is a reduction in cloud frequency in the middle of the season (∼5–25 DRS) which has been identified by Merkel et al. [2009] and associated with a warming near the mesopause. A shorter reduction in cloud frequency with a later onset can be seen in the model around 30–40 DRS. Figure 4 shows a map of the seasonally averaged cloud frequency for clouds with an albedo greater than 1 × 10−6 and 7 × 10−6 sr−1. Again, there are generally similar latitudinal patterns between the model and the observations with better agreement in magnitude and at lower latitudes between the model and the Benze et al. [2009] retrieval. There is more spatial variability in both the Benze et al. retrieval and the control run than in the operational retrieval.

Figure 3.

Times series of the average frequency of clouds with an albedo at 265 nm that, when normalized to a nadir viewing angle and a scattering angle of 90°, is larger than (top) 1 × 10−6 sr−1 and (bottom) 7 × 10−6 sr−1, from (left) the control run, (middle) the CIPS 2007 Northern Hemisphere season from the CIPS operational retrieval, and (right) CIPS retrieval of Benze et al. [2009].

Figure 4.

Polar maps of the average frequency of clouds with an albedo at 265 nm that, when normalized to a nadir viewing angle and a scattering angle of 90°, is larger than (top) 1 × 10−6 sr−1 and (bottom) 7 × 10−6 sr−1, (left) for the control run, (middle) for the 2007 Northern Hemisphere season from the CIPS operational retrieval, and (right) for the CIPS retrieval of Benze et al. [2009].

[19] The latitude dependence of the seasonally averaged albedo and effective radius, Re (Re = ∫r3n(r)dr/∫r2n(r)dr, where n(r) is the radius-dependent concentration), for clouds with an albedo greater than 1 × 10−6 sr−1 is shown in Figure 5. In the model and in the data, there is an increase in albedo and effective radius as latitude increases. There is good agreement between the CIPS and the modeled albedo; however, the effective radius is higher at high latitude and increases with latitude more rapidly in the model than in the CIPS data. SOFIE is more sensitive to small particles than CIPS and the effective radius observed by SOFIE is smaller than that detected by CIPS or predicted by the model. The CIPS data may have an observational bias in the effective radius related both to the use of multiple scattering angles by the CIPS retrieval, causing a reduced sensitivity at low latitudes, and to assumptions about the shape of the size distribution in the CIPS retrieval. As particle size decreases, CIPS will have increasing difficulty sensing the smallest particles, which could cause a high bias in the effective radius. In the simulations, the largest variability of number density with latitude is with the smaller particles, with a larger amount of small particles at lower latitudes. Most of these particles are too small to be detected by CIPS, which would tend to make the latitudinal dependence of effective radius flatter in CIPS than in the model.

Figure 5.

(left) The Northern Hemisphere season average albedo at 265 nm that, when normalized to a nadir viewing angle and a scattering angle of 90°, is larger than 1 × 10−6 sr−1 and (right) effective radius (Re) for CIPS for 2007 (black) and the control run (blue). The shaded area is one standard deviation from the mean. S indicates the season average Re for SOFIE for 2007; W indicates the season average Re from the control run sampled at the SOFIE local times and locations, using the SOFIE threshold for ice detection. CIPS data from latitudes less than 60°N are suspect and have been ignored.

3.1.2. Comparison With SOFIE

[20] SOFIE, a solar occultation instrument aboard the AIM satellite, uses 16 spectral bands in the visible and infrared (0.330 to 5.006 μm) [Gordley et al., 2009]. Since SOFIE is an occultation instrument, it only makes 15 observations each day per hemisphere on a circle of latitude that varies slowly between 65°N and 85°N during the PMC season. It has a long path length of about 280 km; however, it has an excellent vertical field of view of 1.5 km and is sensitive to an ice extinction of ∼1 × 10−7 km−1 in the infrared. In addition to ice particles, it has bands configured to retrieve temperature and H2O, CO2, CH4, O3, and NO. To compare with SOFIE, we sample the model in the grid column containing a SOFIE observation, using model fields at a time that is within 1.5 h of the SOFIE sample time. An extinction threshold of 1 × 10−7 km−1 at 3.064 μm is used to define the existence and vertical extent of a cloud in the column. In 2007, SOFIE version 1.022b had 1470 observations in the Northern Hemisphere from 20 May to 10 September. SOFIE identifies some PMCs extending below 79 km; however, these measurements may be contaminated by clouds that are actually higher than the tangent point altitude but in the line of sight. Therefore, we exclude points whose altitude of maximum extinction (Zmax) is less than 79 km, as suggested by Hervig et al. [2009b]. We do not exclude observations where the lowest extent of the cloud (Zbot) is below 79 km, but there may also be some contamination in these measurements.

[21] Table 2 summarizes various cloud parameters retrieved from SOFIE (version 1.022b) for the 2007 Northern Hemisphere season, and gives corresponding model results from the control run. Figure 6 shows probability distributions of the cloud parameters from SOFIE and from the model control run. The model finds clouds at SOFIE locations and local times 66.9% of the time, while SOFIE detected clouds 77.1% of the time, 28.6% of these having a cloud height (Zmax) below 79 km. The means and probability distributions of the cloud height and cloud top (Ztop) are very similar between SOFIE and the control run. The SOFIE cloud base (Zbot) has events below 79 km, which are not found in the control run and may be from line-of-sight contaminations and are responsible for large cloud thicknesses (dZ) that are also not seen in the control run. The means and probability distributions for the extinction (β) and mass density (M) at Zmax are also very similar, although the model has a few more very faint clouds than SOFIE. The probability distribution for the column IWC is similar between the control run and SOFIE, the mean of the simulated IWC being 19.3% less than SOFIE. The mean effective radius, Re of particles in the control run is 18.0% greater than that in SOFIE. The simulation has a flatter probability distribution for effective radius than SOFIE and has a greater probability of particles larger than the mean. The biggest difference between the model and the control run is the number density (N), the model having a mean that is 62.8% less than that observed. The probability distributions of number density are also very different, with SOFIE seeing many more events with number densities greater than 100 cm−3; however, the mode of the distribution is near 100 cm−3 for both.

Figure 6.

Probability distributions for several cloud properties from the SOFIE 2007 Northern Hemisphere season compared to the control and solar maximum runs sampled at SOFIE local times and locations. The vertical lines indicate the seasonal means.

Table 2. Numbers of Clouds and Mean Values for a Number of Cloud Properties From SOFIE for the Northern Hemisphere (NH) 2007 Season, From the Control Run, and From the Solar Maximum Run
 SOFIE v1.022b 2007 NHSolar Minimum (SMIN)Solar Maximum (SMAX)
Clouds1134959 725 
Zmax < 79 km4200 0 
Zmax (km)83.783.2−0.5%83.60.4%
Zbot (km)80.480.80.5%81.20.5%
Ztop (km)−0.6%
dZ (km)−13.2%
IWC (μg m−2)37.630.3−19.3%21.0−30.9%
B(3.064) at Zmax (km−1)4.37E-054.54E-053.9%3.26E-05−28.3%
Re at Zmax (nm)35.942.418.0%40.5−4.6%
M at Zmax (ng m−3)13.613.70.36%9.8−28.4%
N at Zmax (cm−3)204.075.9−62.8%57.5−24.3%

[22] Figure 7 shows scatterplots of number density versus effective radius for SOFIE and the control run. It shows that the high number density events that SOFIE detects at Zmax are all near the smallest effective radii, ∼15 nm, it detects at Zmax. Figure 7 also shows that the largest PMC mass densities are at ∼80 ng m−3, independent of the effective radius, while the model shows smaller mass densities for the smallest particle sizes. To determine number density from the SOFIE data, the retrieval uses the extinction in channels where scattering is important, along with those where the extinction is mostly due to absorption, and tries to determine fit parameters for the particle size distribution to match these extinctions, assuming that the distribution is Gaussian [Hervig et al., 2009b]. To get a successful retrieval, the extinction at 1037 nm needs to be greater than ∼1 × 10−8 km−1. When this extinction limit is applied to the model control run results, the correlation between number density and effective radius, shown in Figure 7, is more similar to that of SOFIE, eliminating events with an effective radius less than ∼15 nm and concentration less than 1 cm−3. However, this criterion also means that SOFIE does not retrieve number density when the radius is small and/or the number density is low. Figure 8 shows that the number density is only retrieved by SOFIE for clouds with larger mass densities, although in version 1.022b the number density is retrieved for more clouds than in version 1.01. Figure 8 also shows a scatterplot of the extinction at 1037 and 867 nm for version 1.01 and version 1.022b. As can be seen from the control run, Mie theory predicts a ratio of 2 between the extinction at these wavelengths; however, at the lower extinctions there appears to be a significant amount of scatter between the SOFIE bands, and the larger number densities tend to be retrieved from the area with more scatter. The noise characteristics have been improved in version 1.022b, and the average number density at Zmax has been reduced from 407 cm−3 in version 1.01 to 204 cm−3 in version 1.022b. Changes in the seasonal average SOFIE number densities are also due to the retrieval of results from more of the events with smaller number densities. If a threshold for the extinction at 867 nm of 3 × 10−7 km−1 is used to eliminate the area with the most scatter between the 1037 and 867 nm bands, then the average number density at Zmax is 120 cm−3. This suggests that noise could still be causing a bias in the SOFIE number density retrieval. The number density from the control run of 76 cm−3 is below even this lowest value, so the model is still likely to be underestimating the number density.

Figure 7.

The correlation of number density with effective radius (Re) for clouds (left) with an extinction at 3.064 μm greater than 1 × 10−7 km−1 from SOFIE for the 2007 Northern Hemisphere season and for the control run and (right) with an extinction at 1037 nm greater than 1 × 10−8 km−1. The thick black line indicates the number density and effective radius relationship for a constant mass density of 80 ng m−3; and the thin black lines represent mass density values of 40, 20, 10, 5, 2.5 and 1 ng m−3.

Figure 8.

(left) A histogram of mass density for all SOFIE events and for only those events where a size distribution was retrieved for version 1.01 (solid lines) and version 1.022b (dashed lines). Scatterplots of the extinction at 1.037 um and 867 nm are shown for (center) SOFIE version 1.01 and (right) SOFIE version 1.022b. SOFIE points with number densities greater than 250 cm−3 are green, and lesser values are in red. Results from the control run are shown in blue.

[23] Figure 9 shows the seasonal average vertical profiles of temperature, water vapor, and ice extinction retrieved from SOFIE and from the model control run. Figure 9 also has temperatures from the Lübken climatology and from Sounding of the Atmosphere using Broadband Emission Radiometry (SABER) [Remsberg et al., 2008] for the year 2007. The temperatures in the model and the observations are similar, except near the mesopause. Here the model replicates the Lübken climatology to which it was tuned but is significantly colder than SOFIE and peaks at a higher altitude than SOFIE and SABER do. The SOFIE version 1.01 data is known to have a warm bias near the mesopause because the effects of nonlocal thermodynamic equilibrium (non-LTE) were not fully treated. The mesopause is colder in version 1.022b, which includes non-LTE effects; however, it is still not as cold as the other observations or the model. If the model were tuned to SOFIE version 1.01 temperatures, it would likely be too warm to make PMCs. Use of version 1.022b temperatures would be cold enough to support PMCs, but would likely be too warm near the mesopause for effective nucleation to occur. The simulated water vapor in the control run is ∼1.15 ppmv lower than SOFIE throughout the stratosphere and mesosphere; however, the mean and variation are similar near the altitude where PMCs sublimate. The lower water vapor is likely due to tropical tropopause temperatures that are too cold in the WACCM simulation [Garcia et al., 2007]. The initial condition for the model was based on 1995, and thus has lower values of methane, a source of water vapor in the mesosphere, than presently occur. The simulation also has less water vapor in the lower thermosphere than is observed by SOFIE and a more pronounced peak at ∼82 km that is caused by the sublimation of cloud particles. The features are probably more exaggerated in the model because of the water vapor bias and because the model lacks temperature variability caused by subgrid-scale gravity waves. Increased temperature variability should cause sublimation to occur over a broader range of altitudes, which should smooth out the peak. The average simulated extinction at 3.064 μm is remarkably similar to the SOFIE observation in the regions where ice exists.

Figure 9.

The average vertical profiles for temperature, water vapor, and ice extinction at 3.064 μm from SOFIE for the 2007 Northern Hemisphere season and for the control run, including season average temperature profiles from SABER and the climatology of Lübken [1999]. The shaded areas indicate one standard deviation from the average. SOFIE extinction at 3.064 μm from altitudes below 76 km has been ignored, since that extinction is not from ice particles. Horizontal gray lines indicate the 82 km level, which is approximately where sublimation of PMCs occurs, and that at ∼87.5 km, which is the mesopause height. A vertical gray line is drawn at 150 K, approximately the temperature at which sublimation occurs.

[24] Figure 10 shows the seasonal average vertical profiles of mass density, effective radius, number density, and cloud frequency from SOFIE and the control run. There is good agreement throughout the altitude range in mass density, effective radius, and cloud frequency; however, SOFIE detects ice mass below 79 km that is not in the simulation and may be from line-of-sight contamination. We also see that the difference in number density between the model and SOFIE is consistent throughout most of the altitude rage. It is interesting that the vertical profiles of mass density and effective radius match so well, while there is such a large bias in the number density, again suggesting missing high number density events in the simulation and/or a high bias in the SOFIE data. Both mass density and effective radius are dominated by the relatively large particles in the size distribution, which are the easiest to detect optically. It is also these larger particles that are important to the higher albedo clouds detected by CIPS.

Figure 10.

The average vertical profiles for mass density, effective radius, number density, and the frequency of clouds with an extinction greater than 1 × 10−7 km−1 from SOFIE for the 2007 Northern Hemisphere season and for the control run. The shaded areas indicate one standard deviation from the average. The dashed vertical lines indicate the frequency of a cloud occurring anywhere in the column.

[25] Mass density and effective radius are more robust retrievals from SOFIE than number density, since number density requires assuming the shape of the size distribution and making fits to match cloud extinction in multiple bands [Hervig et al., 2009b]. The uncertainty in the SOFIE number density is estimated to be ∼50%, and comparisons with the ALOMAR lidar show that the SOFIE version 1.01 estimates (407 cm−3) are roughly twice that of the lidar (228 cm−3) [Hervig et al., 2009b], although SOFIE version 1.022b results (204 cm−3) are slightly less than the lidar. It should be noted that the ALOMAR lidar results for 2007 are the largest reported by Baumgarten et al. [2008] and for the years 1998 to 2008 (G. Baumgarten, personal communication, 2009), with the lowest values being approximately half of the 2007 levels. This makes it unclear whether the higher values retrieved by SOFIE and the ALOMAR lidar in 2007 are an aspect of the 2007 conditions that are not captured by the model, natural variation, a low bias in the model, or a high bias in the observations. Increasing the number density in the model would likely require increasing the nucleation rate. There are several things that could affect the nucleation rate in the simulation including the low bias in water vapor, a potential bias in temperature, missing subgrid-scale gravity waves that are not resolved by the model but can cause large temperature fluctuations near the mesopause [Philbrick et al., 1984], condensation nuclei that are either too few in number or too small in size, and the inclusion of other nucleation processes.

3.1.3. Comparison of Total Ice Mass With SNOE and SBUV

[26] Figure 11 shows the temporal evolution of the daily average ice mass from the control run and compares the latitudinal distribution of ice mass with estimates based on data from the Solar Backscatter Ultraviolet instrument (SBUV) and Student Nitric Oxide Explorer (SNOE) by Stevens et al. [2007] and with an estimate from SOFIE at SOFIE latitudes. The limb-viewing geometry of SNOE is much more sensitive to ice particles than the nadir-viewing geometry of SBUV, so it is expected that the total ice retrieved by SNOE would be higher than that from SBUV. The shape of the latitudinal ice distribution from the model is similar to the estimates based on SBUV and SNOE; however, the magnitude is closer to SBUV than to SNOE. The average IWC in the simulation is 19.3% lower than in SOFIE, but using the SOFIE average IWC scaled for midsolar cycle and local time still yields a value below the SNOE estimate. The local time response of Fiedler et al. [2005] (Figure 1) indicates that the ice mass might be as much as 30% higher than estimated by the model in portions of the 0900–1200 LT time period; however, these increases would still put the estimated ice mass from the simulation below the SNOE estimate. The extinction measured by SOFIE is proportional to the mass and the imaginary refractive index of ice at all particle sizes. For the SNOE estimate, using only the backscattered data, the SNOE radiance is roughly proportional to the mass for particle radii larger than ∼30 nm; however, the radiance is proportional to the sixth moment of the radius for smaller particles [Englert and Stevens, 2007]. Therefore, the retrieval of IWC from SOFIE should be slightly more accurate than is possible with the SNOE data; however, the SOFIE retrieval does depend on the refractive index of ice being well known. However, even with the refractive index uncertainties included, the SOFIE ice mass densities have uncertainties of less than 10%. Variations in the total ice mass with local time beyond those observed by Fiedler et al. [2005] that are not captured by the model and are not measured by SOFIE could be responsible for the differences between the SNOE ice mass retrieval and those from SOFIE and the model. Further observations and modeling studies are needed to better understand the variability of PMCs with local time throughout the polar region.

Figure 11.

The evolution of the total ice mass during the PMC season for the control run and for a similar run under solar maximum conditions (left) and the average latitude dependence of the ice mass (right) calculated in a manner comparable to the estimates of Stevens et al. [2007] by averaging for midsolar cycle and 0900–1200 local time in 5° latitude bins. The shading for the model is one standard deviation from the midsolar cycle mean. S indicates an estimate of the total ice mass determined by using the SOFIE ice water content at the SOFIE latitudes scaled for solar cycle and local time. W indicates a similar estimate of the total ice mass but using the ice water content from the control run sampled at the SOFIE local times and locations scaled in the same way.

3.2. Importance of Temperature and Gravity Waves

[27] Changes to the WACCM parameterization controlling the gravity wave spectrum can cause significant changes in the summer mesopause temperature profile and meridional winds, which can then have a major impact on PMC formation and the distribution of meteoric smoke particles. Figure 12 shows the temperature at 82 km, near the altitude where PMCs sublimate [Lübken, 1999], the mesopause temperature, and the mesopause height for several of the simulations, along with temperatures from the Lübken climatology and SABER [Remsberg et al., 2008] instrument. We conduct sensitivity tests for the gravity wave parameterization by varying the source strength of the gravity wave spectrum (τb*) and the efficiency with which breaking waves drive the circulation (effgw). The default values for these parameters are 0.006 Pa and 0.125, respectively [Garcia et al., 2007]. We first lowered the source strength to get the temperature at 82 km to approximate the Lübken climatology and then lowered the efficiency to keep the mesopause temperature from being too cold. Lowering the source strength causes the waves to break at a higher altitude, which results in colder temperatures. Lowering the gravity wave efficiency causes the waves to have less of an effect on the circulation, resulting in warmer mesopause temperatures. The end result of tuning the gravity waves for the control run is a mesopause comparable to that in the Lübken climatology, but still results in the mesopause height being slightly higher than in the Lübken climatology. The model also fails to accurately represent the temporal evolution of the mesopause height. This suggests that there are still errors in dynamical representation of the mesopause by WACCM; however, the mesopause height in the model and the Lübken data are within ∼1 km for most of the PMC season, and the a weaker vertical gradient of temperature near the mesopause at the beginning and end of the season than during the season makes the location of the mesopause harder to define. As shown in Figure 12, changes in the source strength have the largest impact on temperature, and lowering the source strength and efficiency causes the evolution of temperature at 82 km to have a bit of a w shape rather than a u shape with a slight warming in the middle of the season. This midseason warming is seen more strongly in the SABER data for 2007. Because of the large water vapor gradient near the mesopause caused by photolysis, changes in the altitude where the ice particles sublimate can have a large impact on the size of ice particles and the total ice mass. Table 3 and Figure 13 show the results for several different combinations of τb* and effgw.. While there is variability in the seasonal evolution of the total ice mass, Figure 13 shows that higher source strengths and/or higher efficiencies tend to produce more total ice mass throughout the season than do lower source strengths and/or lower efficiencies. The extinction and effective radius increase with source strength and efficiency, but the number density and cloud height decrease. A τb* of 0.002 Pa creates a temperature profile that more closely resembles the SABER data and a total ice mass that is roughly twice that of the control run; however, this also creates a cloud height that is 2 km too low compared to SOFIE data and an extinction at Zmax that is 78% larger than SOFIE.

Figure 12.

Mesospheric temperatures at 70°N from SABER, the climatology of Lübken [1999], and several simulations with different gravity wave tunings and particle heating for the PMC season. (left) The temperature at 82 km, the approximate altitude where PMC ice particles will sublimate; (middle) the minimum temperature, which occurs at the mesopause; (right) the height of the mesopause. The SABER data is from 2007 but is available for only part of the season because of the satellite's periodic yawing away from high latitudes. The control run uses a background source strength (τb*) of 0.0015 Pa and latent heating only.

Figure 13.

The evolution of the total ice mass during the PMC season for runs with different tunings of the gravity wave parameterization. Several different combinations of the source strength (τb*) and gravity wave efficiency (effgw) are shown. The solid black line is from the control run.

Table 3. Numbers of Clouds and Mean Values for a Number of Cloud Properties From SOFIE for the Northern Hemisphere 2007 Season and From Several Runs With Different Tunings for the Gravity Wave Parameterization
 SOFIE 2007 NHWACCM/CARMA at Solar Minimum
τb* (Pa) 0.001250.00150.00150.00150.001750.002
effgw 0.08750.08000.08750.09500.08750.0875
Zmax < km420000345
Zmax (km)83.784.083.883.382.582.281.5
Zbot (km)80.480.281.480.879.979.678.9
Ztop (km)87.188.587.687.787.486.986.4
dZ (km)
IWC (μg m−2)37.626.919.230.344.143.450.9
B(3.064) at Zmax (km−1)4.37E-054.31E-052.80E-054.54E-056.85E-056.62E-057.78E-05
Re at Zmax (nm)35.937.238.842.449.451.957.8
M at Zmax (ng m−3)13.613.08.413.720.619.923.4
N at Zmax (cm−3)204.088.582.175.977.359.248.9

[28] Figure 14 shows the distribution of meteoric smoke using the gravity wave tuning used by Bardeen et al. [2008] and by the control run. Using the tuning from the control run results in a reduction in the meridional winds, which increases the number density of meteoric smoke particles larger than 1 nm near the polar summer mesopause. The nucleation of ice particles on meteoric smoke, along with the subsequent growth, sedimentation, and sublimation of the ice particles, causes a redistribution of the meteoric smoke particles with higher number densities near the sublimation area. Having an accurate temperature profile is vital to having realistic simulations of PMCs; however, there is a significant discrepancy between the SABER and the Lübken data, with simulations using temperatures near the Lübken climatology producing PMCs that are a better match to the SOFIE and CIPS data.

Figure 14.

Zonal average concentration of meteoric smoke particles with a radius greater than 1 nm for the month of July for simulations based on (top) the gravity wave tuning of Bardeen et al. [2008] and (bottom) the control run. The simulations on the left include only the meteoric smoke aerosol; those on the right also include PMCs. The black lines are temperature contours, and the green lines are contours of the meridional wind.

3.3. Importance of Contact Angle

[29] Recent laboratory measurements of ice nucleation at low temperatures [Trainer et al., 2009] indicate that the cosine of the contact angle, m, decreases with decreasing temperature. Using classical nucleation theory to interpret the data, they calculated m decreasing from 0.91 at 195 K to 0.65 at 153 K for formation of ice over a flat silicon substrate. Extrapolation of their fit of the data to lower temperatures suggests that m might reach 0.30 at 140 K and could go to 0.00 near 135 K. While the data do not replicate the temperatures, materials, or particle sizes of heterogeneous nucleation of ice on meteoric smoke at the mesopause, they are suggestive of an additional barrier to ice nucleation at the low temperatures of the polar summer mesopause. Table 4 and Figure 15 show the results for several simulations with m ranging from 0.3 to 0.95 to investigate the sensitivity of the simulations to changes in nucleation rate caused by changes in the contact angle. The average cloud height, extinction, mass density, and column IWC are surprisingly insensitive to changes in the contact angle; however, the effective radius increases and the number density decreases with decreased m. The biggest change in the simulations is seen in Figure 15, where lower values of m result in less ice formation at the beginning of the season and more ice formation later in the season. The increased ice later in the season corresponds with an increase in supersaturation later in the season in the runs with lower m values. These results suggest that because of the large supersaturations at the simulated mesopause, the details of ice nucleation may not be critical to creating a reasonable simulation of PMCs. Although this may make it more difficult to determine what nucleation mechanisms are responsible for PMC formation, simulations forced by assimilated meteorology to recreate specific observational periods covering an entire season might reveal information about the required nucleation rate by comparing the evolution of the total ice mass during the season.

Figure 15.

Evolution of the total ice mass during the PMC season for runs with different contact angles, where the cosine of the contact angle varies between 0.3 and 0.95. The black line is from the control run.

Table 4. Numbers of Clouds and Mean Values for a Number of Cloud Properties From SOFIE for the Northern Hemisphere 2007 Season and From Several Runs With Different Values for the Contact Anglea
 SOFIE 2007 NHWACCM/CARMA at Solar Minimum
  • a

    m = the cosine of the contact angle.

m 0.950.
Zmax < 7942000000
Zmax (km)83.783.382.983.182.982.9
Zbot (km)80.480.880.480.680.380.3
Ztop (km)87.187.787.887.987.487.3
dZ (km)
IWC (μg m−2)37.630.338.636.6030.829.2
B(3.064) at Zmax km−14.37E-054.54E-055.83E-055.60E-054.61E-054.32E-05
Re at Zmax (nm)35.942.445.348.253.755.5
M at Zmax (ng m−3)13.613.717.616.913.913.0
N at Zmax (cm−3)204.075.986.582.750.142.9

3.4. Importance of Particle Heating

[30] Simulations by Espy and Jutt [2002] and by Siskind et al. [2007] have shown that radiative heating of the cloud particles can have a significant impact on the particle growth rate and on the atmospheric temperature. Our control run includes the effect of latent heat released by condensing ice particles, but it does not include the effect of radiative heating of the ice particles. To explore the sensitivity of the model to these heating assumptions, we conduct tests with (1) neither latent nor radiative heating and (2) both latent and radiative heating, using two different sets of optical properties based on refractive indices for ice by Warren [1984] and Toon et al. [1994]. The imaginary refractive indices in Toon et al., which were measured at 163 K, tend to be slightly lower than those from Warren, which were intended for temperatures near 215 K. The differences can be as large as a factor of 5 but typically are less than 50%. A new set of optical constants and a discussion of how they compare with the work of Toon et al. [1994] and Warren [1984] can be found in the work of Warren and Brandt [2008].

[31] Table 5 and Figure 16 show results from these sensitivity tests, and Figure 12 shows the temperature changes near the mesopause for these runs. Not including latent heating in the simulation has only a small impact on the simulation, slightly decreasing the temperature at 82 km, increasing the total ice mass, lowering the cloud heights, and increasing the mass density, IWC, effective radius, and number density. However, decreases in total ice mass about 15 and 35 d after solstice in the control run are absent or not as prominent in the run without latent heating. Particle radiative heating causes a larger change in temperature than the latent heating case, particularly for the Warren optical constants. With radiative heating, the total ice mass is half or less of that seen in the control run, and the mass builds more slowly during the beginning of the season and shrinks more gradually at the end of the season than in either the no-heating or latent-heating runs. The cloud heights are increased with radiative heating, and the extinction, mass density, effective radius, and IWC are significantly reduced. Both latent and radiative heating increase the midseason heating over that of the no-heating run, and there tends to be a warming at the beginning of the season and cooling at the end of the season relative to the no-heating run. Because our simplistic coupling with the WACCM radiation code does not preserve the spectral features of the upwelling radiation, our estimates of radiative heating are only approximate, and the gravity wave parameterization may also need to be retuned to keep the overall temperature in the range of the observations; however, these results are suggestive of the importance of radiative heating. Figure 17 shows the July averaged ice mixing ratio, cloud heating rate, temperature perturbation, and vertical wind perturbation from the run using Toon et al. [1994] refractive indices. This shows a heating rate of up to 8 K/d, a warming of 4 K near cloud level, a cooling of 8 K above the clouds, and an increase in vertical wind near the clouds. These magnitudes are similar but somewhat reduced from the values shown by Siskind et al. [2007], and our ice mixing ratio is only a third of theirs. The patterns of the temperature and vertical wind perturbations near the pole are similar to those of Siskind et al., but in our simulation the cooling and increased updraft extend to lower latitudes, which may be caused by an interaction with the meridional circulation. Particle radiative heating looks to be important in reducing the particle size and controlling the temporal evolution of the ice mass. It may also enhance the meridional circulation. Radiative heating appears to have a significant impact on PMCs, and we intend to reevaluate this in future versions of the model that are more tightly integrated with the WACCM radiation code.

Figure 16.

Evolution of the total ice mass during the PMC season for runs with no heating, latent heating, and both radiative and latent heating. The solid black line is from the control run.

Figure 17.

(top) Zonally averaged ice mixing ratio and cloud heating rate from the radiative heating run using the optical properties of Toon et al. [1994]; (bottom) temperature and vertical wind differences between the run with latent and radiative heating and the control run, which contains only latent heating.

Table 5. Numbers of Clouds and Mean Values for Several Cloud Properties From SOFIE for the Northern Hemisphere 2007 Season and From Runs With Different Assumptions for the Heating of the Particles and the Feedback on Atmospheric Temperatures
 SOFIE 2007 NHWACCM/CARMA at Solar Minimum
Heating Regime NoneLatentLatent and radiative [Warren, 1984]Latent and radiative [Toon et al., 1994]
Refractive Indices N/AN/AWarren [1984]Toon et al. [1994]
Zmax < 794200010
Zmax (km)83.783.183.384.283.8
Zbot (km)80.480.680.882.081.4
Ztop (km)87.187.687.787.687.9
dZ (km)
IWC (μg m−2)37.632.730.312.018.1
B(3.064) at Zmax (km−1)4.37E-054.81E-054.54E-051.62E-052.40E-05
Re at Zmax (nm)35.944.542.431.833.9
M at Zmax (ng m−3)13.615.713.74.97.2
N at Zmax (cm−3)

3.5. Effects of Solar Cycle

[32] PMCs are known to vary with the solar cycle [DeLand et al., 2006, 2007], because of changes in photolysis and radiative heating. There is reduced PMC brightness during solar maximum conditions as compared to solar minimum conditions. The solar cycle repeats about every 11 years, with the latest solar minimum starting in 2007 and the latest solar maximum in 2001. Solar maximum conditions have increased solar radiation, which will cause increased photolysis of water vapor, decreasing the water vapor concentration and increasing the temperature near the mesopause. To investigate the response of the model to the solar cycle, Table 2 and Figures 3, 6, and 10 show the results of a simulation using solar maximum conditions. The cloud bottom height and the height of the maximum extinction have increased slightly relative to solar minimum conditions, while the cloud top height has decreased. The total ice mass in the solar maximum run is about two thirds of the solar minimum control run. The beginning of the PMC season starts 15 d later and ends 5 d earlier in the solar maximum simulation than in the solar minimum simulation. As indicated in Table 2, the column IWC and the extinction, mass density, number density, and effective radius at the height of the maximum extinction are all reduced at solar maximum relative to the control run. The mean simulated CIPS albedo during solar maximum is reduced by 39.6% from 50°N to 82°N compared to solar minimum, while DeLand et al. [2007] observed an average reduction in albedo from SBUV of 16.9% between solar minimum and solar maximum for the same latitude range. DeLand et al. [2007] also detected a smaller decrease in albedo at low latitudes compared to high latitudes, while the simulation shows a greater reduction in albedo at low latitudes. The possible dry bias in the simulation may cause a greater reduction in albedo during solar maximum than is seen in the observation. Removal of the dry bias, inclusion of the radiative heating of ice particles, additional simulations covering the entire solar cycle, and further observations from CIPS and SOFIE would help improve the understanding of changes in PMCs during the solar cycle.

4. Summary and Discussion

[33] We have developed a three-dimensional model of PMCs that includes detailed treatment of microphysical processes. The control run compares very well with new PMC observations from the SOFIE and CIPS instruments on the AIM satellite. We find that the formation of PMCs is very sensitive to the atmospheric temperatures near the polar summer mesopause. In the model, these temperatures are largely controlled by the gravity wave parameterization and to a lesser extent by the latent and radiative heating of the atmosphere by the clouds themselves. The gravity wave parameterization also affects the mesospheric meridional circulation, which has a strong influence on the distribution of meteoric smoke particles. This sensitivity of the model to the gravity wave parameterization suggests that an accurate simulation of the impact of climate change upon PMCs will require a model whose gravity wave sources are also free to respond to changes in the climate and are not controlled by arbitrary tunings. It also indicates that PMCs may be a very sensitive indicator to changes in mesospheric temperatures caused by changes in the sources and filtering of vertically propagating gravity waves.

[34] Our simulations show that despite conditions that are depleted in large meteoric smoke particles relative to previous estimates, heterogeneous nucleation on meteoric smoke alone allows the model to create realistic cloud fields. We find that because of the large supersaturations generated within the model, the simulations are not very sensitive to the contact angle or the actual nucleation rate; however, laboratory data for ice nucleation on silicates extrapolated to PMC temperatures suggest that the barrier to nucleation could be even greater than we tested. There are large uncertainties in the mass influx and coagulation coefficient of meteoric smoke, the role of particle charging of the smoke in coagulation and nucleation, and the contact angle and free energies associated with the classic nucleation theory. Further laboratory and field studies are needed to constrain these and other parameters in order to compare different candidates for PMC ice nucleation processes. In particular, measurements of heterogeneous nucleation of ice need to be taken at temperatures relevant to the summer polar mesopause; substrates that are analogous to meteoric material should also be considered. Thus, while heterogeneous nucleation of meteoric smoke may be sufficient to explain the observed cloud fields, these simulations do not rule out other candidates such as ion nucleation, sulfate aerosols, or homogeneous nucleation as important sources of ice nuclei.

[35] The largest discrepancy between the model and the SOFIE observations is the inability of the model to generate ice particle number densities as high as those seen in the observations. High number densities are important, because they are thought to be necessary for the generation of polar mesospheric summer echoes (PMSE) [Rapp and Lübken, 2004]. The mean of the simulated number densities (76 cm−3) is well below that for the ALOMAR lidar (228 cm−3) and SOFIE 1.022b (204 cm−3) for 2007 [Hervig et al., 2009b]. More years of observation and more simulations are needed to help determine whether the higher number densities for 2007 could be caused by the solar cycle or other natural variation or whether biases may exist in either the model or the observations.

[36] The model is missing some physics that may be important to the generation of higher ice number densities. The temporal and spatial scales of the model limit the gravity waves that can be included in the simulation. Waves with high spatial and temporal frequencies are known to generate large localized temperature fluctuations that could result in a temporary increase of the number of ice particles. Also, the water vapor in the model is ∼1.15 ppmv lower than that observed by SOFIE throughout the stratosphere and mesosphere, which is caused by simulated tropopause temperatures that are too low and by less methane being present in the model than is observed for 2007 conditions. Higher water vapor concentrations should generate higher supersaturations that would increase the nucleation rate and thus generate higher particle number densities. It is also possible that other nucleation processes are required to generate higher number densities in the model. The SOFIE retrieval for number density requires making assumptions about the shape of the size distribution, and the uncertainty in this retrieval can be as large as 50%. Reduction of the noise in the 1067 and 867 nm bands and an increase in sensitivity in the SOFIE version 1.022b retrieval has lowered the average number density at Zmax from 407 to 204 cm−3. A comparison of these two bands indicates that even in the 1.022b retrieval, noise may still exist that could reduce the value further, perhaps to as low as 120 cm−3. However, this value is still higher than the number density of 76 cm−3 from the model control run and is lower than the value of 228 cm−3 from the ALOMAR lidar for 2007.

[37] While these simulations are promising, it would be beneficial to incorporate the microphysics used here into a model that can be driven by observed conditions. This would allow a direct rather than just a statistical comparison between the model and specific years observed by the AIM satellite. However, the limited set of meteorological observations in the mesosphere and the feedback of heating by the PMCs upon the atmospheric temperature will make this difficult. Also, further work is needed to reduce the discrepancies between the different types of temperature observations made in this region. While this paper has emphasized the differences between the model and the observations as well as possible reasons for these differences, it is noteworthy that the model does so well at predicting most of the observed properties. The model generally reproduces the SOFIE probability distributions and the CIPS albedo. The AIM data set is a significant advancement in the observation of these clouds. The agreement between the model and these observations is an important step forward in our understanding of these fascinating clouds.


[38] The authors wish to thank Scott Bailey, Jerry Lumpe, Uwe Berger, Steve Massie, and Doug Kinnison for their assistance. We are grateful to the National Center for Atmospheric Research (NCAR) and the National Aeronautics and Space Administration (NASA) for the use of computer time to help run these simulations. We would also like to thank the SABER, CIPS, and SOFIE science teams for the use of their data and the NCAR for supporting our use of the WACCM model. The NCAR is operated by the University Corporation for Atmospheric Research under the sponsorship of the National Science Foundation (NSF). Partial support for Charles Bardeen was provided by a NASA Earth System Science Fellowship grant NNG05GQ75J and an NCAR Advanced Study Program Postdoctoral Fellowship. Additional support for this project came from NASA heliophysics guest investigator grant NNX08AK45G, NASA Aura grant NNG06GE80G, NSF grant ATM0435713, and NASA AIM grant NAS5-03132.