Volume 119, Issue 24 p. 14,129-14,143
Research Article
Free Access

Tomographic and spectral views on the lifecycle of polar mesospheric clouds from Odin/OSIRIS

Kristoffer Hultgren

Corresponding Author

Kristoffer Hultgren

Department of Meteorology, Stockholm University, Stockholm, Sweden

Correspondence to: K. Hultgren,

[email protected]

Search for more papers by this author
Jörg Gumbel

Jörg Gumbel

Department of Meteorology, Stockholm University, Stockholm, Sweden

Search for more papers by this author
First published: 26 November 2014
Citations: 15

Abstract

Vertical and horizontal structures of Polar Mesospheric Clouds (PMC) have been recovered by tomographic retrieval from the OSIRIS instrument aboard the Odin satellite. The tomographic algorithm has been used to return local scattering coefficients at seven wavelengths in the ultraviolet. This spectral information is used to retrieve PMC particle sizes, number density, and ice mass density. While substantial horizontal variations are found, local vertical structures are overall consistent with the idea of a growth-sedimentation process leading to a visible cloud. Large numbers of small particles are present near the top of the observed cloud layer. Toward lower altitudes, particle sizes increase while particle number densities decrease. A close relationship is found between the distribution of local PMC scattering coefficient and ice mass density. The bottom of the cloud often features large particles with mode radii exceeding 70 nm that rain out of the cloud before sublimating. The number density of these large particles is small, and they do not contribute significantly to the overall cloud brightness. As a consequence, the presence of these large particles can be difficult to identify for remote sensing techniques that integrate over the entire cloud column. When it comes to deriving absolute values of particle mode radius and number density, there is a strong sensitivity to assumptions on the mathematical form of the particle size distribution. We see a continued strong need to resolve this issue by co-analysis of various remote sensing techniques and observation geometries.

Key Points

  • Tomographic/spectroscopic analysis of vertical and horizontal PMC structures
  • Local structures in cloud brightness follow the distribution of ice mass density
  • Large ice particles exist well below the cloud brightness peak

1 Introduction

Polar Mesospheric Clouds (PMCs), also known as Noctilucent Clouds when viewed from the ground, are a summertime phenomenon existing mainly at locations poleward of ±50° and at typical altitudes of 80–85 km. The wave-driven general circulation leads to temperatures falling to 130 K and below in the upper mesosphere [Lübken et al., 2009], which allows for the formation of water-ice particles of sizes typically smaller than 100 nm. PMC particle sizes have been retrieved from the ground using lidars [e.g., Baumgarten et al., 2010], from rockets [e.g., Gumbel et al., 2001], and from satellites [e.g., von Savigny et al., 2004; von Savigny et al., 2005; Rusch et al., 2009]. However, none of the previous investigations has addressed at the same time vertical and horizontal variations of particle sizes and other microphysical parameters on a local scale. In this paper, results from tomographic methods applied to limb-scanning measurements by the Optical Spectrograph and Infrared Imager System (OSIRIS) on Odin are presented. The tomographic retrieval technique for PMCs has been presented by Hultgren et al. [2013] and is based on limb observations of PMC scattering by the Odin satellite. Based on multiple lines of sight in the orbit plane from consecutive positions along the satellite orbit, the retrieval inverts the observed radiances into an estimate of the local volume scattering coefficient along both the vertical and horizontal dimensions. As the tomographic retrieval is done at seven different wavelengths in the ultraviolet, the spectral information can be used to derive PMC particle sizes, particle number densities, and ice mass densities.

The life cycle of PMCs and the underlying microphysical processes have been investigated numerous times [e.g., Rapp and Thomas, 2006, and references therein]. Results from infrared measurements of specific water-ice band features in solar occultation have confirmed the basic idea that the particles consist of water ice [Hervig et al., 2001; Eremenko et al., 2005], with possible addition of minor fractions of meteoric material [Hervig et al., 2012].

PMC formation is usually described as a growth-sedimentation process [Rapp and Thomas, 2006]. Basic processes can conveniently be summarized by the one-dimensional scheme in Figure 1. PMC formation is controlled by the degree of saturation, as defined by the background temperature field and the distribution of water vapor. Nucleation primarily takes place at altitudes with the largest saturation ratio, near the mesopause temperature minimum typically at 86–90 km. Once the ice particles have formed, they grow by direct deposition of the surrounding water vapor. At the same time, gravity leads to sedimentation of the growing particles. The resulting downward increase of particle size has been confirmed by various observational techniques [e.g., Gumbel and Witt, 1998; von Savigny et al., 2005; Baumgarten and Fiedler, 2008]. Growth can continue as long as the particles are in the supersaturated region. Time in the supersaturated region is prolonged as sedimentation is counteracted by upward vertical wind that usually prevails in the polar summer mesopause region. While sedimenting, particles can consume a substantial fraction of the available water vapor. A cloud visible from the ground or space can be generated if a sufficient number of particles grow to sufficiently large sizes, as determined by the scattering properties of the particles. At the end of the growth-sedimentation process, particles sediment out of the supersaturated region and sublimate.

Details are in the caption following the image
Illustration of the vertical structure of the PMC particle evolution. To the right, three example particles represent three basic particle categories that reach different final stages in the PMC growth/sedimentation process. See text for details.

In reality, this simple picture is modified, e.g., by local action of turbulence and horizontal transport processes [e.g., Rapp et al., 2002; Berger and von Zahn, 2007]. The statistical nature of nucleation, turbulent transport, and growth has important consequences for cloud evolution and particle population [Megner, 2011; Kiliani et al., 2013]. This is illustrated in Figure 1 by three representative particles that are thought to nucleate simultaneously in the upper part of the supersaturated region. Particle 1 follows an optimal path and grows to large sizes by efficiently taking up water vapor while moving through the supersaturated region. It reaches maximum size at the bottom of the supersaturated region. Due to the large fall speed at this stage, it can sediment deep into the unsaturated region before completely sublimating. As compared to particle 1, particle 2 falls behind due to the statistical processes. Once particle 1 has a head start, the differences between both particles easily amplify as particle 2 finds itself in an environment that has been partially depleted of water vapor. Particle 2 grows and sediments but remains relatively small. When it enters the less supersaturated region at lower altitudes, the Kelvin effect becomes critical and often leads to sublimation of the particle before it reaches the actual bottom of the supersaturated region. In that case, the water from particle 2 is deposited within the supersaturated region and further contributes to an efficient growth of particles of category 1. It is thus in this less supersaturated region closer to the bottom of the cloud where the fate of the two particles entirely diverges. Particle 1, which is larger, will grow very fast, whereas particle 2 will sublimate. Particle 3 in Figure 1 is left behind even more than particle 2. It never grows large enough for a fall speed that would allow it to sediment against the prevailing vertical wind. In an instructive study, Berger and von Zahn [2007] described these statistical influences, with particular focus on the history of those PMC particles in the cloud that grow to “visible” sizes. Megner [2011] points out that these statistical processes lead to a growth of visible PMC with surprisingly little sensitivity to the detailed conditions in the nucleation region.

An aim of this paper is to analyze the above growth-sedimentation description by studying the vertical and horizontal variations of basic PMC properties. While the individual OSIRIS measurements result in snapshots of instantaneous clouds structures, the analysis of the entire tomographic data set provides insight in all stages of the PMC lifecycle. The paper is organized as follows. Section 2 describes the methodology of retrieving PMC parameters by applying tomography and spectral analysis to data from the OSIRIS instrument. Section 3 presents results from using these methods, before sections 4 and 5 discuss implications for both PMC microphysics and PMC remote sensing.

2 Methodology

2.1 Odin/OSIRIS Tomography

On 20 February 2001, the Odin satellite was launched into an almost circular Sun-synchronous orbit at 600 km. The orbit has an ascending node around 18:00 local time and a period of 96 min [Murtagh et al., 2002; Llewellyn et al., 2004]. The satellite carries two instruments: the Optical Spectrograph and InfraRed Imager System (OSIRIS) and the Sub-Millimeter Radiometer (SMR). By nodding the entire satellite up and down, the co-aligned optical axes of OSIRIS and SMR are performing limb-scanning measurements in the forward direction. The Optical Spectrograph (OS) of OSIRIS (hereafter simply referred to as OSIRIS) obtains spectra of scattered sunlight over the range 275–810 nm with a spectral resolution of about 1 nm. For the PMC spectral analysis presented in this paper, ultraviolet wavelengths below 310 nm are chosen, thus utilizing absorption by ozone in the stratosphere to avoid complications due to upwelling radiation from the lower atmosphere. Seven ultraviolet wavelength bands (Table 1) that are not perturbed by airglow and auroral emissions are chosen [Karlsson and Gumbel, 2005].

Table 1. Ultraviolet Wavelength Intervals (in nm) Used for the Spectral PMC Analysis
Interval UV1 UV2 UV3 UV4 UV5 UV6 UV7
Mean 277.3 283.5 287.8 291.2 294.4 300.2 304.3
Limits 276.3–278.3 282.9–284.1 286.4–289.1 289.9–292.6 292.6–296.1 298.1–302.4 302.8–305.9

Odin usually operates in one of two primary scan modes covering tangent altitudes between 7 and 107 km. However, during 180 orbits of the northern summers of 2010 and 2011, Odin was programmed with a special “tomographic mode,” only concentrating on the altitudes 73–88 km in 2010 and 77–88 km in 2011 [Hultgren et al., 2013]. By operating in this mode, the horizontal distance between subsequent scans was significantly reduced, which strongly increased the number of lines of sight through a given volume. This allows for the tomographic retrieval to provide two-dimensional distributions of volume emission rate as a function of radial distance from the center of the Earth (equivalently: altitude) and angle along the satellite track (equivalently: horizontal distance).

As discussed by Hultgren et al. [2013], to extract PMC properties, the pure cloud radiance needs to be separated from the background radiation due to molecular Rayleigh scattering and instrumental effects. In short, Odin's ordinary PMC analysis uses limb scans ranging from the troposphere to the lower thermosphere, which provides limb radiance profiles covering extensive height ranges below and above the PMC layer. Molecular and instrumental background can then readily be discriminated by fitting the data in these cloud-free altitude ranges [Karlsson and Gumbel, 2005]. This procedure is not possible here since the tomographic scans are restricted to the narrow altitude region of the PMCs and background fits to cloud-free altitude ranges above and below the PMCs are not available. Instead, the molecular scattering background is calculated in terms of Rayleigh scattering based on an atmospheric density profile from the MSIS climatology. The instrumental background is taken as the mean value of the background obtained from ordinary limb scans during the days before and after the tomography scans. These backgrounds are then subtracted from the measured total limb radiance in order to obtain the pure PMC signal as input to the tomographic retrieval. As discussed by Hultgren et al. [2013], this approximate background subtraction scheme contributes to the overall uncertainty of the PMC tomography. As compared to ordinary limb scans, it degrades the detection threshold for the weakest PMCs.

2.2 Tomographic Retrieval of Scattering Coefficient

The radiances along the individual lines of sight serve as input to the tomographic retrieval algorithm. The aim of the tomographic retrieval is to convert integrated line-of-sight radiances into local information about the scattering process, in terms of the volume scattering coefficient β as discussed by Hultgren et al. [2013]. Our algorithm is based on an extended Maximum Probability method, the Multiplicative Algebraic Reconstruction Technique (MART), developed by Lloyd and Llewellyn [1989] and modified to its current state by Degenstein et al. [2003, 2004]. The algorithm solves the problem in an iterative manner with iterations performed on a ray by ray basis until the retrieval converges.

In Figure 2, the first panel shows the input radiance field at wavelength 300 nm from orbit 50799 (17 June 2010). Retrieved β are shown in the second panel. Solar scattering angles during the tomography scans vary between 70° and 100° along the orbit. In Figure 2, β is plotted normalized to a scattering angle of 90°. This normalization is achieved by applying T-matrix calculations of the phase function based on the particle size retrieval described in the next section. The local time of the data varies approximately between 17:30 at 70°N on ascent (northbound) through noon at the highest latitudes and 8:15 at 70°N on descent (southbound).

Details are in the caption following the image
Example of PMC data obtained during part of orbit 50799 on 17 June 2010. The first panel shows the input PMC limb radiance field after subtraction of molecular and instrumental background. The black crosses show the actual position of the tangent points of the individual lines of sight. The radiance field shown in the panel is interpolated between the radiances measured from these tangent locations. The second panel shows the tomographically retrieved volume scattering coefficient at wavelength 300 nm. The third, fourth, and fifth panels show, respectively, PMC particle size (mode radius), particle number density, and ice mass density derived from the spectral analysis of the volume scattering coefficients at 270–305 nm. White color indicates that the PMC is too weak for spectral analysis of particles sizes etc. The dashed line defines a vertical cut through the data corresponding to the profiles shown in Figure 3.

Several factors contribute to the uncertainty of the PMC limb radiance that is the input to the tomographic retrieval. A systematic error is introduced by an accuracy of approximately 5% of the OSIRIS calibration in the ultraviolet. The statistical error is governed by instrument noise as well as uncertainties in the subtraction of molecular and instrumental background. Based on uncertainty in the input radiances, Hultgren et al. [2013] estimated a typical statistical error in β of 10−11 m−1 str−1, which is less than 1% of typical PMC peak brightness. A more rigorous error analysis can be achieved by propagating the errors of the individual radiances through the tomographic retrieval algorithms. In a Monte Carlo approach, individual input radiances are chosen randomly in accordance with the normal-distributed error ΔLPMC about the PMC radiance LPMC. Running the retrieval for a large ensemble of randomly perturbed radiances results in a distribution of β with a standard deviation that represents the retrieval error. Resulting uncertainties of β are shown in the first panel of Figure 3 for the vertical cut through the data defined in Figure 2.

Details are in the caption following the image
Four panels with profiles and error bars for scattering coefficient β, mode radius rm, number density N, and ice mass density Mice corresponding to the vertical cut through the data in Figure 2. The gray-shaded area represents the error due to the 5% accuracy of the OSIRIS calibration in the ultraviolet. The error bars represent statistical uncertainties and their propagation through the spectral analysis.
For the further discussions, it is useful to define the relationship between the scattering coefficient and the microphysical properties of the PMC particle size distribution. The local scattering coefficient β can be expressed as a function of total number density N, normalized size distribution f(r), and differential scattering cross section ∂σ/∂Ω for the direction in question:
(1)

The differential scattering cross section is in turn a function of particle size and shape. Equation 1 provides a basis for retrieving microphysical PMC properties from the optical data.

2.3 Particle Size Retrievals

Karlsson and Gumbel [2005] described the basic idea of the spectral analysis of the OSIRIS PMC data. The spectral dependence of the scattering coefficient at the seven wavelengths in Table 1 is the basis for fitting an effective particle size to the measured spectra. In general, a large number of free parameters can be invoked in describing a PMC particle population and defining the details of size distribution and particle shapes. However, the information contained in the scattering spectrum at 275–310 nm is limited and allows in practice to determine only one independent parameter of the particle size distribution f(r). In other words, assumptions are needed about properties of the particle population that cannot be retrieved explicitly. In particular, a mathematical form of f(r) as well as the shape of the particles must be assumed. It is also important to note that f(r) will in general depend on the PMC volume considered. In particular, local, nadir-integrated, or limb-integrated conditions may be different. In line with earlier work [Rapp et al., 2007; Lumpe et al., 2013], we use a normal distribution
(2)
with the mode radius rm and the width Δr. (Note that the above normalization factor in equation 2 strictly only applies for combinations of rm and Δr that do not lead to significant contributions to f(r) at r < 0. In other cases, an explicit truncation for r < 0 and a numerical normalization need to be introduced.) Using the mode radius as parameter to be retrieved, a further assumption is needed on the width. There is strong evidence that the distribution width varies with mode radius [Baumgarten et al., 2010]. Column retrievals from the Cloud Imaging and Particle Size instrument (CIPS) onboard the AIM satellite use a width that varies as Δr ≈ 0.39 × rm up to rm = 32 nm and that is approximately constant above [Lumpe et al., 2013]. The work presented in the current paper is based on the same assumption. The dependence of retrieval results on the assumed size distribution will be discussed in section 2.6. As for the shape of the particles, there is strong evidence that PMC particles are not spherical [Baumgarten et al., 2002; Rapp et al., 2007; Hervig and Gordley, 2010]. In line with the operational retrieval used for CIPS [Lumpe et al., 2013], our retrieval is based on ice spheroids with an axial ratio 2.

Using the above assumptions, PMC mode radii are determined by comparing the OSIRIS PMC spectra to theoretical scattering spectra from numerical T-matrix simulations [Mishchenko and Travis, 1998; Baumgarten et al., 2008]. The fitting of the experimental scattering data uses an exponential dependence on wavelength as expressed by the Ångström exponent [e.g., von Savigny et al., 2005]. A fit in the seven wavelength intervals is performed in each horizontal/vertical pixel of the tomographic output. By applying this fit to an Ångström exponent to each pixel of the tomographic analysis, the vertical and horizontal distributions of PMC particle sizes can be obtained. An example is the third panel of Figure 2. Typical radius uncertainties are shown in the second panel of Figure 3. These uncertainties are analyzed in a similar way to the β-uncertainty in the previous section. Based on random perturbation of input limb radiances, a Monte Carlo approach is used to retrieve perturbed scattering coefficients at all seven wavelengths. Subsequent Ångström fits to the perturbed scattering coefficients result in an uncertainty distribution of the derived mode radii.

2.4 Retrieval of Particle Number Density

Equation 1 provides a means to retrieve total particle number densities N from the tomographic results. With β and rm retrieved in the previous sections and ∂σ/∂Ω calculated by the T-matrix approach, this relationship allows a determination of N. An example result is shown in Figure 2 (fourth panel), while uncertainties are included in Figure 3 (third panel). The uncertainty of N is relatively large. The reason is the strong dependence of ∂σ/∂Ω on the retrieved rm, typically as a function of rmn with n ~ 3–4 in the transition regime between Rayleigh and Mie scattering of interest here. The inference of the particle number density is thus very sensitive to uncertainty in the preceding rm retrieval. In addition, there is a substantial sensitivity of the N retrieval to the necessary assumption on the form of the particle size distribution. In particular, there may be an unidentified large amount of small particles to which the optical measurement is not even sensitive. Connected to this is the question of what particles should actually be counted as part of the ice particle population. For example, should small water cluster ions be included in N, and in that case down to what size? In summary, the PMC particle number density is not a well-defined quantity. It should be handled very carefully when discussing and comparing results of different PMC measurements.

2.5 Retrieval of Ice Mass Density

The ice mass density (in units ng m−3) can be written as
(3)
With rm and N retrieved in the previous sections, an inference of Mice is thus straightforward. Results are shown in Figure 2 (fifth panel), while uncertainties are included in Figure 3 (fourth panel). One might expect that Mice has a large uncertainty since it is sensitive to the uncertainties of both rm and N, and in addition to the assumption on the form of the size distribution. However, these uncertainties cancel partially, as can be seen by rewriting equation 3 with N from equation 1:
(4)

Since ∂σ/∂Ω has a dependence on particle size with typically rn with n ~ 3–4, much of the r-dependencies in the numerator and denominator of equation 4 cancel out. The result is that Mice is rather insensitive to the retrieval uncertainties of rm and N. Rather, as equation 4 indicates, the local ice mass density is rather directly connected to the local scattering coefficient β. This is evident from Mice plotted in the fourth panel of Figure 2. Its spatial structure strongly reflects the spatial structure of β in the first panel of Figure 2. Based on these ideas, it is feasible to establish an (empirical) relationship between cloud brightness and cloud ice. This has been utilized, e.g., for inferring ice water content from the cloud albedo in earlier versions of the PMC retrieval for the AIM/CIPS instrument [Lumpe et al., 2013].

2.6 Dependence on Particle Size Distribution

As discussed in section 2.3, the information content of the optical measurement is limited, and assumptions on the form of the size distribution are necessary in order to retrieve a mode radius. The results of the retrieval will generally be sensitive to this assumption. This is illustrated by comparing the retrievals in this paper to retrievals based on a broader size distribution. As compared to assumption described in section 2.3, this alternative assumption is based on a normal size distribution with Δr = 0.5 × rm up to rm = 60 nm and constant above. Redoing our retrievals with this assumption results in Figure 4, which can be directly compared to Figure 2.

Details are in the caption following the image
Results of rm, N, and Mice for the same data set as in Figure 2. In the current figure, the retrieval is based on the somewhat broader assumption on the particle size distribution discussed in the text.

As compared to the narrower assumption used in Figure 2, retrieved mode radii in Figure 4 are 10–20 nm smaller, retrieved number densities are about a factor 2.5 larger, and retrieved ice mass densities are about a factor 1.2–1.3 larger. These are substantial differences. In general, the differences due to different assumptions on the particle size distribution are larger than our typical retrieval uncertainties (Figure 3). When communicating retrieval results on PMC microphysical parameters, it is of obvious importance to always include information about the assumptions underlying the specific study. While this sensitivity to the assumed size distribution is strongest for the retrieval of rm and N, the retrieval of the ice mass density is less affected. This is consistent with the statement in section 2.5 that Mice is a rather robust retrieval quantity. Mice is rather insensitive to uncertainties in the spectral particle size analysis and, correspondingly, Mice is rather insensitive to assumptions on the particle size distribution. This makes PMC ice content potentially a very attractive quantity for the co-analysis of different data sets.

3 Results

In the current work, we have analyzed 180 orbits of tomographic OSIRIS PMC observations during the 2010 and 2011 Northern Hemisphere PMC seasons. Examples from four orbits are shown in Figures 5-8. In each figure, the first panel shows the volume scattering coefficient that is retrieved by the tomographic routines and that serves as input to the subsequent retrievals of PMC parameters. The second, third, and fourth panels show retrieved mode radius, particle number density, and ice mass density, respectively. These figures reveal both distinct vertical features and substantial horizontal structures of the clouds. Throughout the analyzed orbits, we find consistent vertical structures with large numbers of small particles near the top of the observed cloud layer, and increasing particle sizes and decreasing particle number densities toward lower altitudes. As discussed in the next section, this general structure is consistent with the growth-sedimentation scenario described in the introduction and illustrated in Figure 1.

Details are in the caption following the image
Tomographic results from Odin orbit 50789 on 16 June 2010. The first panel shows the PMC volume scattering coefficient tomographically retrieved at wavelength 300 nm. The second, third, and fourth panels show, respectively, PMC particle size (mode radius), particle number density, and ice mass density derived from the spectral analysis of the volume scattering coefficients at 275–310 nm. The circles indicate the locations of size distributions discussed further in Figure 10.
Details are in the caption following the image
Tomographic PMC results from Odin orbit 50790 on 16 June 2010. See Figure 5 for details.
Details are in the caption following the image
Tomographic PMC results from Odin orbit 51228 on 15 July 2010. See Figure 5 for details.
Details are in the caption following the image
Tomographic PMC results from Odin orbit 51653 on 12 August 2010. See Figure 5 for details.

Figures 5-7 represent results that are typical for the middle of the PMC season (June, July) with an extended cloud cover at latitudes above 70°N. Figure 8 from 13 August is typical for results toward the end of the season. Here, only geographically limited cloud patches are found and particle mode radii remain small, typically below 70 nm.

4 Discussion

The tomographic and spectral analysis of Odin/OSIRIS PMC data reveals vertical and horizontal cloud information in terms of scattering coefficient, particle size, number density, and ice mass density. This provides important new insight in PMC structures. The data presented in the previous section are in line with current ideas of PMC evolution, but they also reveal some surprises that impact our understanding of the clouds and our interpretation of observations.

As for results in line with current ideas, our two-dimensional cloud structures are consistent with the overall idea of a PMC growth-sedimentation process. At high altitudes, near the temperature minimum, large numbers of small ice particles are nucleated. The particles then grow by a direct surface deposition of the surrounding water vapor and settle downward by gravity. As can be seen in the third panels of Figures 5-8, particle number densities consistently decrease with decreasing altitude over the observable altitude range of the cloud. Based on microphysical particle studies, we know that coagulation of ice particles is not a significant process during the PMC lifetime [Rapp and Thomas, 2006]. As discussed in section 1, the downward decreasing number density is rather connected to the statistical nature of the particle growth. Particles that happen to become larger during the fall through the surroundings subsequently fall faster than their smaller siblings. During their descent, they enter altitudes containing more water vapor and subsequently grow even faster. These particles, the ones that got a head start, are the ones that are most likely to eventually become visible as the clouds we see at lower altitudes. A contributing factor to the vertical gradient of the number density distribution is also the continuity equation: when the particles grow large and fall faster, they spend less time in any given volume. Finally, the particles leave the supersaturated region and sublimate near the bottom of the cloud. In Figures 6-9, this is most clearly seen in distinct cloud regions where large particles “rain out” of the cloud down to altitudes around 80–81 km (e.g., Figure 5 around 2200 km, Figure 6 around 1800 km, and Figure 7 around 1300 km).

Details are in the caption following the image
The range of PMC parameters retrieved from the Odin/OSIRIS tomographic measurements. Shown are mode radii and number densities distributed in bins of different ice mass densities Mice (in units of ng m−3).

Somewhere during this growth-sedimentation process, the combination of particle number density and particle size becomes “optimal” for maximum scattering, and hence, the peak of the scattering coefficient (cloud brightness) is found. The same argument applies to ice mass density. As described in section 2.5, the ice mass density largely follows the distribution of the scattering coefficient, and therefore, the second, third, and fourth panels in Figures 5-8 very much reflect the structures of the first panels.

While the above picture is consistent with a one-dimensional conventional growth-sedimentation description, three-dimensional reality is in general far more complicated. Indeed, we find substantial horizontal variation of the overall vertical PMC structure. On the one hand, cloud areas are found without pronounced layers of larger particles near the bottom. On the other hand, distinct layers of large particles can be found without pronounced layers of smaller particles above. These varying features can be understood in terms of the variable dynamical history (wind shears, wave activity, local turbulence) of individual cloud parcels as they undergo the growth-sedimentation process. The range of particle populations generated by this complex cloud evolution can be illustrated by plotting the relationship of the basic parameters particle size, particle number density, and ice mass density, as in Figure 9. All retrieval pixels with identified PMCs are included in this plot, regardless brightness or altitude. This results in a large parameter space with mode radii covering 1 order of magnitude and number densities covering 4 orders of magnitude. It becomes clear how areas with large particles are connected to areas with low densities, and vice versa.

Among the more surprising findings of this work is the presence of small concentrations of large particles well below the PMC brightness peak. Figures 5-7 show many regions with fairly large particles (>80 nm) in the lower part of PMCs. However, these regions do not contribute much to the overall cloud brightness. Rather, the bright regions are dominated by particle sizes around 50–70 nm. This has important consequences for remote sensing techniques that integrate along a line of sight. Regions dominated by the very large particles coincide with parts of the cloud that hold only a very few particles and are rather dim. Most often, volumes with larger brightness and smaller particles are located above these volumes with larger particles. Column-integrated measurements are dominated by the brightest cloud volume along the line of sight, and hence, the smaller particles at the brighter altitudes dominate the size retrieval. Therefore, instruments that integrate along lines of sight (nadir or limb) through the clouds will usually retrieve these smaller radii. A nadir-viewing technique can only “see” the larger particles when observing cloud areas where no bright cloud layers are situated above the larger particles. In the orbits presented in this paper, such conditions are only found in Figure 6 around 2700 km and in Figure 7 around 1400 km. However, as can be seen at several locations in the figures, this does not mean that large particles are rare (e.g., Figure 5 around 2200 km; Figure 6 around 1200 km, 1800 km, and 3300 km; and Figure 7 around 1000 km and 2100 km). Rather, due to the reasons discussed above, they are often present but usually not observable for a column-integrating instrument.

An important question concerns the nature of this population of few large particles near the bottom of the cloud. More insight in the vertical variation of the particle population can be gained by plotting the actual size distribution at various altitudes. This is done in Figure 10. Shown are the absolute size distributions N × f(r,rmr) for the locations and altitudes marked in Figures 5 and 6. These size distributions follow equation 2 with the retrieved mode radius rm and the assumed width Δr as described in section 2.3 and the retrieved N described in section 2.4. Figure 10 suggests that the large particles near the bottom of the cloud do not constitute a “mysterious” layer that occurs unexpectedly. Rather, they represent the tail of the size distribution that exists higher up at the cloud's brightness peak. A likely scenario is that the lower large particle layer is located around the transition from slight supersaturation to subsaturation. In this region, smaller particles start sublimating quickly, first due to the Kelvin effect and then due the overall unsaturated conditions. Still, the largest particles can be found here as they fall fast through these altitudes and as their sublimation takes significantly longer than the sublimation of the smaller particles.

Details are in the caption following the image
Particle size distributions at the three altitudes in the PMC at the locations defined in Figures 5 and 6. In each plot, the three curves represent the highest observable altitude in the cloud, the altitude of maximum scattering, and an altitude near the cloud bottom, respectively.

As described above, we find these large particles most often at the bottom of vertically extended cloud layers, below volumes with larger cloud brightness and smaller particles. Nevertheless, we also find numerous occasions with larger particle in the absence of smaller particle above. This can in principle be understood in a one-dimensional picture when considering an “old,” well-processed cloud region where upper cloud layers have been dehydrated and can no longer sustain the presence of small particles. However, we suggest that three-dimensional dynamics provides a more likely explanation for these structures, as wind shears are a common feature near the summer mesopause [Gadsden and Schröder, 1989], thus providing a mechanism that can efficiently decouple lower PMC layers from the nucleation and growth processes occurring at higher altitudes.

When discussing vertical profiles of particle sizes, it is important to distinguish between individual particles and the mode radius of the particle size distribution. In general, these two quantities will peak at different altitudes. An individual PMC particle typically reaches its maximum size at or near the bottom the supersaturated region when it encounters a supersaturation low enough for the Kelvin effect to start its sublimating: it grows while falling to this altitude and it sublimates once it has fallen below this altitude (Figure 1). The mode radius of the particle size distribution, on the other hand, can continue to grow with decreasing altitude below this transition region. This is due to the disappearance of the smaller particles in the distribution at these altitudes, leaving the large particles to dominate the distribution (Figure 10).

When discussing Figures 5-10, it is important to note that our analysis does not provide direct information about the smallest particles in the PMC and, thus, about the cloud's nucleation region. This is a common limitation of optical PMC studies that are hardly sensitive to the small Rayleigh scatterers of sizes below about 20 nm. For the present study, analyzable PMC radiances are mostly restricted to altitudes below 86 km. Megner [2011] argues that only limited conclusions about the nucleation processes can be drawn from this kind of “visible” PMC data. This is related to the major conclusion by Megner [2011] that the properties of the visible PMC are rather insensitive to the detailed nucleation conditions. Rather, it is the subsequent growth conditions that primarily determine the observable cloud properties.

The substantial horizontal structure that we find in the data is in accordance with expectations; we know that PMCs are highly structured, influenced by gravity waves, Kelvin-Helmholtz instabilities, and other inhomogeneous effects [e.g., Witt, 1962; Baumgarten et al., 2012; Thurairajah et al., 2013]. Horizontal structures are even expected to exist on shorter scales than shown in Figures 5-8. However, these are not observable with our current tomographic possibilities and our resolution which is of the order of 100 km. It must be pointed out that this currently limits our ability to quantify how usual or how rare certain observed features are. This applies, e.g., to the likelihood of cloud volumes with large particles to occur in the presence or in the absence of volumes with smaller particles located above. Such conclusions may very well turn out different when based on finer horizontal resolutions. For detailed studies of horizontal structures, nadir-imaging instruments like CIPS are much better suited. Our analysis is complementary to this. Its benefit is that it adds vertical information to the horizontal studies. First comparisons of PMC structures between the OSIRIS tomographic analysis and AIM/CIPS have been presented by Hultgren et al. [2013]. Upcoming comparisons will focus on particle sizes, number densities and ice water content.

The vertical/horizontal information presented in this paper can also be compared to the vertical/time information on NLCs provided by ground-based lidar measurements. As cloud structures drift through the lidar measurement volume, the time coordinate of lidar data becomes a proxy for horizontal coordinate in our plots. Assuming a typical horizontal wind of, e.g., 50 m/s, the horizontal resolution of 100 km corresponds to an integration time of 0.5 h of the lidar as the structure passes the field of view. State-of-the-art lidar systems can obtain NLC integration times substantially shorter than this [Baumgarten et al., 2012]. When aiming at direct comparisons, a discussion is needed to what extent local NLC properties can vary over the time scales that characterize the comparison process.

5 Conclusions

In this paper we have analyzed vertical and horizontal structures of Polar Mesospheric Clouds observed by the OSIRIS instrument aboard the Odin satellite by using a tomographic retrieval algorithm. The algorithm was used to return scattering coefficients at seven different wavelengths in the ultraviolet. This provided two-dimensional PMC data with a vertical resolution of typically 1 km and a horizontal resolution of typically 100 km. The subsequent spectral analysis provides new insight into PMC processes by revealing particle sizes, number density, and ice mass density in both the vertical and horizontal dimensions. Our findings are in line with the established growth-sedimentation description of PMC particle evolution. Figures 5-8 are thus consistent with the idea of cloud particles nucleating in large number at the top of the cloud region, growing by direct deposition of surrounding water vapor while sedimenting through the cloud region, and eventually leaving the supersaturated cloud region and sublimating. At the same time, Figures 5-8 indicate how much this simple one-dimensional picture can be modified by the three-dimensional reality where wave activity and wind shears cause substantial horizontal variation of PMC particle growth.

The most interesting finding of this study is the frequent occurrence of large particles at the cloud bottom with mode radii exceeding 70 nm. These particles can be considered as raining out of the cloud at the very end of their life cycle. Due to their size and fall speed they can penetrate well into the unsaturated altitude range below the cloud before finally sublimating. However, the number density of these large particles is small and they do not contribute significantly to the overall cloud brightness. This has important consequences for the remote sensing of PMC particle populations. As column-integrating measurements (nadir or limb) are dominated by the brighter cloud regions along the line of sight, the presence of these large particles will in general be difficult to identify.

As the information content in the spectroscopic data is limited, the analysis of the cloud parameters requires reasonable assumptions on the particle size distribution and particle shape. Like earlier studies, we find a strong sensitivity of the particle size retrieval (and subsequent number density retrieval) on these assumptions. Progress in this area has been made based on detailed lidar studies [Baumgarten et al., 2010]. Nevertheless, we see a continued strong need to address this issue, e.g., in terms of comparing size retrievals of different remote sensing techniques and observation geometries. Ice mass density, on the other hand, turns out to be rather insensitive to particle size assumptions. Rather, there is a close relationship between local ice mass density and PMC scattering coefficient. This gives additional support to the idea of using ice content as a geophysical parameter suitable for the intercomparison of different remote sensing studies.

The tomographic PMC study described in this paper will continue. Distributions of cloud brightness, particle sizes, number density, and ice content like in Figures 5-8 need to be compared to microphysical model descriptions. Important questions concern the thermal, humidity, and dynamical structure of the mesopause region that can give rise to the observed cloud patterns. Important complementary information is available from the Sub-Millimetre Receiver (SMR) onboard Odin. Using the short PMC limb scans from 2010 and 2011 described in section 2.1, tomographic analysis is possible even for the SMR data on water vapor and temperature [Christensen et al., 2013].

While the current analysis has provided the first tomographic retrieval of PMC properties, a limb-scanning instrument like Odin/OSIRIS is not really optimal for tomography. Decisive for tomographic retrieval capabilities is the number of lines of sight that can be obtained through a given volume. Even though Odin was operated with dedicated short limb scans for the analysis presented here, the use of limb scans with one measurement at a time puts a basic limit on this number. More suitable for tomography are instruments based on limb imagers that can provide many lines of sight simultaneously. It this way, the current horizontal resolutions of about 100 km could be improved by up to an order of magnitude. Limb imagers also open the possibility to go from two-dimensional to three-dimensional PMC retrievals by introducing a second horizontal data dimension perpendicular to the satellite track. Based on these concepts, a new satellite mission MATS (Mesosphere Airglow/Aerosol Tomography and Spectroscopy) is currently being prepared by a consortium of research groups and industrial partners with funding from the Swedish National Space Board.

Acknowledgments

We thank OHB-Sweden and SSC Space for making tomographic scans with Odin/OSIRIS possible. Odin is a Swedish-led satellite funded jointly by Sweden (SNSB), Canada (CSA), France (CNES), and Finland (TEKES). Since April 2007, Odin is a third-party mission of the European Space Agency. We thank Linda Megner for valuable discussions concerning PMC microphysics. We thank Gerd Baumgarten for supporting us with numerical scattering data from the T-matrix algorithm. All data used to achieve the results presented in this paper can be requested from the corresponding author at [email protected].