Two important approaches for satellite studies of polar
mesospheric clouds (PMCs) are nadir measurements adapting phase function
analysis and limb measurements adapting spectroscopic analysis. Combining
both approaches enables new studies of cloud structures and microphysical
processes but is complicated by differences in scattering conditions,
observation geometry and sensitivity. In this study, we compare common
volume PMC observations from the nadir-viewing Cloud Imaging and Particle
Size (CIPS) instrument on the Aeronomy of Ice in the Mesosphere (AIM) satellite and a special set of tomographic
limb observations from the Optical Spectrograph and InfraRed Imager System
(OSIRIS) on the Odin satellite performed over 18 d for the years 2010
and 2011 and the latitude range 78 to 80
We perform the first thorough error characterization of OSIRIS tomographic
cloud brightness and cloud ice water content (IWC). We establish a
consistent method for comparing cloud properties from limb tomography and
nadir observations, accounting for differences in scattering conditions,
resolution and sensitivity. Based on an extensive common volume and a
temporal coincidence criterion of only 5 min, our method enables a
detailed comparison of PMC regions of varying brightness and IWC. However,
since the dataset is limited to 18 d of observations this study does not include a
comparison of cloud frequency. The cloud properties of the OSIRIS tomographic
dataset are vertically resolved, while the cloud properties of the CIPS
dataset is vertically integrated. To make these different quantities
comparable, the OSIRIS tomographic cloud properties cloud scattering
coefficient and ice mass density (IMD) have been integrated over the
vertical extent of the cloud to form cloud albedo and IWC of the same
quantity as CIPS cloud products. We find that the OSIRIS albedo (obtained
from the vertical integration of the primary OSIRIS tomography product,
Polar mesospheric clouds (PMCs) are the highest and coldest clouds of the Earth's atmosphere. During daylight conditions, PMCs are too faint to be observed by the naked eye, but during twilight, when the lower atmosphere lies in Earth's shadow, PMCs become visible against the dark sky when they efficiently scatter light from the sun under the horizon. The striking appearance of a silvery shining veil against the darker summer night sky has also given them the name noctilucent clouds (NLCs). Despite the remarkably dry environment of the summer polar mesosphere of only a few parts per million of water, PMCs can form because the wave-driven residual meridional inter-hemispheric circulation (Lindzen, 1981; Garcia et al., 1985) causes upwelling of air from the stratosphere, and the rising motion over the summer pole leads to strong adiabatic cooling of the summer polar mesospheric region (Fritts and Alexander, 2003). Resulting mesopause temperatures of typically 130–150 K (Lübken, 1999) make it possible for water vapour to condense on cloud nuclei, forming visible PMCs. During the last two decades, extensive research has contributed to knowledge about PMC formation and composition. It is now well established that the clouds consist of water ice that has nucleated onto meteoric smoke material (Hervig et al., 2001, 2009), and that nucleation occurs in bursts at the mesopause (Megner, 2011; Kiliani et al., 2013). The ice particles grow in size by condensation of the surrounding water vapour and sediment when they grow large enough and typically become visible when they reach sizes above 20–30 nm (Rapp et al., 2002; Rapp and Thomas, 2006). Eventually, they descend into a sub-saturated air mass and sublimate.
Ever since the first observations of these night-shining clouds over 130 years ago (Backhouse, 1895; Jesse, 1885; Leslie, 1985 ), the clouds have been studied extensively and continue to be a major subject of middle atmospheric research. This is related to the fact that PMCs are very sensitive to changes in temperature and water vapour and therefore serve both as tracers for long-term changes of the middle atmosphere where direct observations are limited (DeLand et al., 2006; DeLand and Thomas, 2015) and tracers for the dynamics of the polar summer mesosphere (e.g. Fritts et al., 1993). PMCs are frequently used as a tool to study wave activity in the mesosphere (Witt, 1962). Gravity waves excited at lower altitudes that propagate through the PMC layer alter the cloud microphysics and leave behind a detectable footprint, which manifests as a variation in cloud brightness or cloud height that can be observed both using ground-based instruments and satellites. Propagating gravity waves create a downward and upward motion of the PMC layer, and the adiabatic heating and cooling associated with regions of downward and upward motion will cause sublimation and growth of the cloud ice. The resulting structures in the cloud layer can be used to detect and quantify gravity waves that are present in the mesopause region (Thurairajah et al., 2017; Hart et al., 2018). Gravity wave signatures from PMC albedo variations observed by the nadir Cloud Imaging and Particle Size (CIPS) experiment on the Aeronomy of Ice in the Mesosphere (AIM) satellite (Russel et al., 2009) have for example been used to investigate both small-scale gravity waves (Chandran et al., 2009; Taylor et al., 2011) and large-scale gravity waves (Chandran et al., 2010; Zhao et al., 2015). Additionally, CIPS has proven an excellent tool to study various types of cloud structures, such as voids, bands, ice rings and fronts, that are connected to the complex small-scale dynamics present in the summer mesopause region (Thurairajah et al., 2013).
Extensive research has led to an increased knowledge on the climatology of PMCs. A number of recent studies suggest that PMC occurrence frequency has increased in recent decades (e.g. Hervig et al., 2016; Lübken et al., 2018; and references therein).
DeLand and Thomas (2015) show an increase in PMC ice water content (IWC) at
latitudes between 50 and 82
The geographical extent and variation of the PMC layer are monitored
continuously by both ground-based instruments and satellites, which each
possess their advantages and disadvantages. For example, NLC camera networks
and ground-based lidars possess the advantage of being able to observe the
clouds with very high temporal resolution. NLC camera networks can capture
the horizontal evolution of a geographically limited region of the PMC
layer. Numerous ground-based studies from NLC camera networks have led to
remarkable insight into the complexity of spatial structures of the clouds,
planetary wave influence on NLC occurrence frequency, gravity wave effects
on the NLC layer and cloud height measurement (e.g. Dalin et al., 2016,
2015; Kirkwood and Stebel, 2003; Witt, 1962). Ground-based lidar systems on
the other hand possess the unique capacity to monitor the vertical structure
of the cloud with resolutions as fine as 30 s and 50 m (Baumgarten et al., 2012) and give detailed insight into the microphysical processes of a cloud as well as climatology and long-term
changes of PMCs (Baumgarten et al., 2012; Fiedler et al., 2016; Kaifler et
al., 2013; Ridder et al., 2017; Shibuya et al., 2017; Suzuki et al., 2013).
Ground-based instruments alone, however, lack the ability to monitor the
longitudinal and latitudinal variations. Polar-orbiting satellites, on the
other hand, are able to provide information about the climatology and global
distribution of PMCs (e.g. Bailey et al., 2015; Hervig et al., 2016, Liu
et al., 2016) as well as detailed information about horizontal variations of
cloud parameters and wave effects on cloud microphysics (e.g. Rusch et al., 2016; Hart et al., 2018). Each satellite geometry has its limitations: nadir-pointing instruments observe horizontal cloud structures but can provide no
or limited information about vertical structures (Hart et al., 2018). Limb-scanning instruments provide information on vertical cloud structures but
have only coarse horizontal resolution with gaps between observations. Due
to instrument design, limb-viewing instruments observe PMC as a function of
To overcome the limitation from the assumption of a spatially homogeneous cloud layer and to enable the retrieval of clouds as a function of actual height instead of tangent height, the application of a two-dimensional tomographic retrieval for OSIRIS has been presented by Hultgren et al. (2013). The tomographic approach for OSIRIS retrievals (described in Sect. 2) provides a tool to study both vertical and horizontal variations of cloud microphysical properties on a local scale, which is useful for detailed studies of cloud growth and destruction (Christensen et al., 2016; Megner et al., 2016). An early attempt to apply a direct (tomographic) retrieval for a limb-viewing instrument was investigated almost two decades ago. The first approach was carried out by Livesey and Read (2000), who developed a tomographic 2-D retrieval for the Microwave Limb Sounder (MLS) instrument on the Aura satellite. A tomographic reconstruction of atmospheric constituents and gases can be performed using multiple exposures from multiple viewing angles. The tomographic technique has previously been used to study middle atmospheric phenomena other than PMCs, for example gravity wave activity in airglow in the mesopause region (e.g. Nygren et al., 2000; Song et al., 2017), and stratospheric mesoscale gravity waves (Krisch et al., 2017; 2018). More recently, Hart et al. (2018) successfully demonstrated the first application of a tomographic technique to CIPS on AIM. They were able to project a novel 2-D PMC surface map with gravity wave signatures, providing valuable information on wave characteristics such as wave amplitude and dominant horizontal wavelengths. For the OSIRIS instrument, Degenstein et al. (2003, 2004) first showed the possibility to retrieve both vertical and horizontal structures from a series of limb images taken from the OSIRIS infrared imager and were able to map volume emission rates of the oxygen infrared atmospheric band.
In this paper, we perform a detailed common volume (CV) study of PMC cloud brightness and ice mass density (IMD) from the Odin OSIRIS tomographic retrievals and coincident PMC observations from CIPS on AIM. The occasions for Odin's special tomographic scans were chosen to coincide both in time and space with the CIPS instrument. A comparison of the two instruments is therefore ideally suited for instrument validation and the combination of the two datasets will be valuable in future studies of cloud–wave interaction, studies on particle sizes and studies on how the retrieved cloud properties are affected by cloud inhomogeneities. Many scientific questions about the PMC life cycle are connected to the two- or three-dimensional structure of the clouds. Important questions concern, for example, the effect of gravity waves or dynamical instabilities on the growth, sublimation or appearance of the clouds. Combined observations by (horizontally resolved) nadir instruments and (vertically resolved) limb instruments have large potential to address such multi-dimensional questions. This is true in particular if the datasets involve tomographic analysis, as in the case of the OSIRIS data utilized here.
Taking into account that the satellites have different viewing geometry, resolution and sensitivity, we analyse cloud brightness and the cloud ice in the CV and perform a detailed error analysis. One advantage of comparing tomographic OSIRIS observations to CIPS observations is that both instruments measure scattered radiance, although OSIRIS measures with limb-viewing geometry and CIPS uses nadir-viewing geometry. Another advantage is that the same assumption regarding the mathematical shape of the particle size distribution, namely a Gaussian distribution, is used in both the OSIRIS and the operational CIPS v4.2 retrieval.
The specific aims of this satellite comparison study are as follows:
Perform the first thorough error characterization of the Odin OSIRIS tomographic dataset. Validate the tomographic retrieval and error characterization by comparing PMC albedo and IWC from the Odin/OSIRIS retrievals and AIM/CIPS retrievals. Establish a consistent method for comparing cloud properties from a limb sounding tomographic dataset to a nadir-viewing instrument. Produce a combined dataset of Albedo and IWC that will facilitate future studies of the PMC life cycle and PMC particle sizes.
This study focuses on comparing the albedo and IWC between the instruments. A future goal is to produce a combined dataset that can be used to study for example more fundamental issues such as the assumptions of PMC the size distribution of PMCs, an assumption that has been questioned in the past. Each instrument used alone can only provide either very fine horizontal resolution (CIPS) or vertical/coarse horizontal resolution (tomographic OSIRIS). However, when combined in an efficient way, OSIRIS can provide vertical information on cloud structures such as double cloud layers or voids, ice distribution at different altitude levels, and information about the existence of particles of different sizes on different altitude levels that can complement the high horizontal resolution of the clouds from CIPS. Additionally, the combined dataset can be used to investigate how waves (inferred from albedo variations in CIPS) affect the cloud lifetime and how nucleation–sublimation processes affect the vertical distribution of cloud properties (inferred from a vertical cross section from OSIRIS).
The paper is structured as follows. First, in Sect. 2, the OSIRIS tomographic technique is introduced together with a thorough error characterization of the OSIRIS tomography PMC scatter coefficient and IMD. Additionally, this section describes the CIPS PMC dataset and known uncertainties of cloud albedo and IWC. In Sect. 3, the method used for the instrument comparison is described, including a discussion of the challenges in making tomographic limb and nadir observations consistent. In Sect. 4, the results of the comparison are presented and discussed. Section 5 provides the conclusions.
The Swedish-led Odin satellite (Murtagh et al., 2002) was launched on 20 February 2001, into an almost Sun-synchronous polar orbit at 600 km with ascending node near 18:00 local solar time. The Odin mission began as a joint project between aeronomy and astronomy, with the primary focus of the aeronomic part of the mission on coupling processes in the atmosphere, better understanding of ozone variation and processes in the middle atmosphere, and processes that govern PMC formation and mesospheric variability. Odin carries two instruments, the Sub-Millimeter Radiometer (SMR) (Urban et al., 2007) and OSIRIS (Llewellyn et al., 2004). OSIRIS consists of an atmospheric limb-scanning spectrometer and an infrared imager with the ability to measure vertical profiles of atmospheric trace gases and ice layers in the middle atmosphere (Karlsson and Gumbel, 2005). The Odin instruments scan the limb of the atmosphere in the forward direction as the satellite nods up and down while moving in its polar orbit. The OSIRIS spectrometer observes scattered sunlight as limb radiance in the wavelength range from 275 to 810 nm with a spectral resolution of about 1 nm. Odin can be operated in different modes that regulate the vertical resolution and altitude region depending on the species of interest. In a standard stratospheric–mesospheric mode, the satellite typically scans the atmosphere from 7 to 107 km. In 2010, a so-called tomographic mode was developed, scanning only PMC altitudes between typically 70 and 90 km. Since 2010, Odin has continued to perform regular observations in this tomographic mode for selected orbits both in the Northern Hemisphere and in the Southern Hemisphere up until the present. In the tomographic mode, the distance between subsequent scans is shorter, which increases the horizontal sampling compared to the normal mode. The extended number of lines of sight through a cloud volume produces sufficient information to tomographically retrieve two-dimensional distributions of the cloud scattering coefficient as a function of height and horizontal distance. The tomographic algorithm inverts the observed limb radiance into an estimate of the retrieved local scattering coefficient, a measure of the cloud brightness. The tomographic retrieval is based on the multiplicative algebraic reconstruction technique (MART) developed by Lloyd and Llewellyn (1989) and further developed and adapted to OSIRIS data by Degenstein (1999) and Degenstein et al. (2003, 2004). It was described in detail by Hultgren et al. (2013) and Hultgren and Gumbel (2014). MART is based on maximum probability theory that solves the problem on a ray-by-ray basis until the retrieval converges.
The limb radiance is measured by Odin as a function of tangent altitude,
while the tomography retrieves the local scattering coefficient as a
function of actual altitude and angle along orbit (AAO). AAO runs from 0 to
360
The tomographic retrieval uses a grid of
The measured PMC limb radiance contains contributions from molecular Rayleigh scattering from the background atmosphere as well as instrumental effects, e.g. baffle scattering and offset due to dark current. Therefore, a separation of the pure cloud signal from the background is needed before the tomographic retrieval can be performed. As the short limb scans in tomographic mode do not cover tangent altitudes outside the PMC regions, these background signals cannot be measured independently. Rather, the molecular scattering background is estimated by calculations of Rayleigh scattering based on an atmospheric density profile taken from MSIS (Hedin, 1991). The contribution to the signal from the instrumental effects is calculated as the mean value of the background taken during the ordinary limb scans measured the days before and after the tomographic scans.
Spectral analysis of OSIRIS PMC data enables the retrieval of cloud
microphysical properties such as mode radius, number density and IMD
(Karlsson and Gumbel, 2005). Here we provide a brief description of the
spectral analysis on tomographic data. For a detailed description, together
with a discussion about uncertainties, the reader is referred to Hultgren
and Gumbel (2014). The spectral analysis of the microphysical properties
mode radius
It is important to assess the uncertainty of OSIRIS tomography PMC data products. Here we refine the uncertainty discussion provided by Hultgren and Gumbel (2014). The starting point is the uncertainty of the limb radiances measured by OSIRIS. These uncertainties propagate through the tomographic retrieval of the local scattering coefficients, and then through the subsequent spectral analysis of the microphysical cloud properties. As for the accuracy of the OSIRIS limb radiances, there is an absolute error of about 10 %, which is a combination of the uncertainties due to instrument calibration and due to the influence of ozone absorption on the limb radiances in the ultraviolet (Benze et al., 2018). As for the estimated random error of the OSIRIS limb radiances, Hultgren and Gumbel (2014) list as error sources the instrument noise as well as uncertainties in the subtraction of the molecular and instrumental background. For faint PMCs, the random error is directly related to the need to discriminate the cloud signal from the molecular background (and its random error). Another random error is introduced by mesospheric ozone: absorption by ozone along the LOS has a significant effect on limb measurements of PMCs in the ultraviolet. This is a particular concern for tomographic retrievals as these include lines of sight with tangent altitudes well below 80 km, where ozone absorption increases substantially. Lacking direct ozone measurements in conjunction with the PMC measurements, we treat the natural variability of ozone in the upper mesosphere as a contribution to the random error of the PMC limb radiance.
The propagation of these limb radiance uncertainties through the tomographic
retrieval is investigated by a Monte Carlo approach, i.e. by running the
retrieval with limb radiances randomly perturbed within the uncertainty
limits (Hultgren and Gumbel, 2014). The resulting relative error of the PMC
volume scattering coefficient
As described above, the tomographic retrieval of the scattering coefficient
at several wavelengths is the basis for the subsequent spectral analysis of
PMC microphysical properties like mode radius, number density or IMD.
Hultgren and Gumbel (2014) have discussed the corresponding error
propagation. In particular they point out that the IMD is rather independent
of the uncertainties in the retrieval of particle size and particle number
density, and rather independent of the assumptions on the size distribution.
This is due to the fact that the radius dependences of particle volume and of
particle scattering cross section to a large extent cancel each other when
inferring the ice mass. As a result, the local PMC IMD is to a large extent,
albeit not completely, proportional to the local PMC scattering coefficient.
The uncertainty of IMD is subject to the same absolute error of
Summary of OSIRIS PMC uncertainties.
Given the above uncertainties, it is of importance to assess the accuracy of tomographic OSIRIS cloud retrievals by comparison with cloud properties derived from independent measurements and model simulations. As for the latter, Megner et al. (2016) compared the OSIRIS tomographic retrieval of cloud properties to the Community Aerosol and Radiation Model for Atmospheres (CARMA) (Toon et al., 1979; Turco et al., 1979). They investigated which part of the ice particle size distribution OSIRIS captures and evaluated how this affects the retrieved cloud properties. They concluded that the OSIRIS tomographic IWC is within approximately 20 % of the simulated IWC; however, that mean radius and number density are less accurate. Specifically, the tomographic retrieval performs well for retrieving mode radius in the range 50–70 nm, but overestimates mode radius (up to a factor of 3) for small mode radii and underestimates it for large mode radii (80 nm and above). Moreover, the study by Megner et al. (2016) suggested that the tomographic algorithm performs well for retrieving number density for small number densities (which usually occur lower in the cloud) but greatly underestimates it for high number densities (which usually occur higher up in the clouds, where the retrieval misses the smaller particles).
The AIM satellite, launched in 2007, was the first satellite mission fully
dedicated to the study of PMCs, with an overall goal to resolve how PMCs form and
vary (Russell et al., 2009). AIM is moving in a circular, sun-synchronous
orbit near 600 km altitude. AIM currently has two operating instruments, the
Solar Occultation for Ice Experiment (SOFIE) (Gordley et al., 2009) and the
high-resolution UV panoramic imager CIPS. CIPS consists of four
nadir-pointing wide-angle cameras with a combined FOV of 120
The CIPS retrieval algorithms and data products with error analysis and
cloud detection sensitivity have been described in detail by Lumpe et al. (2013). For this study, we utilize CIPS level 2 data, which is the primary
CIPS PMC data product consisting of cloud presence, cloud albedo normalized
to 90
The data contain quality flags that indicate the number of scattering angles that were used to determine the scattering phase function. An observation is most robust when at least six scattering angles are used (marked with quality flag 0), and more uncertain when fewer angles are used (marked with quality flag 1 for four or five angles and quality flag 2 for fewer than four angles). In line with the recommendation for satellite comparisons of microphysical properties, we use data flagged with 0 or 1 in this work.
The random albedo uncertainty in each CIPS
pixel is given as
CIPS IWC is completely determined by albedo and radius. As discussed in
Lumpe et al. (2013), the information content of scattering angle variation
of the albedo observed by CIPS decreases for decreasing particle sizes. By
comparing PMC retrievals to simulated data (their Fig. 21), systematic error
and random error were estimated. For IWC, it was shown that the systematic
error is highly dependent on mode radius. For large particles (
In the present study, we use the Odin OSIRIS dataset from two northern hemispheric seasons, 2010 and 2011. During these seasons, the tomographic orbits were scheduled to coincide in both time and space with CIPS. To date, Odin has continued to be run in the tomographic mode for specific orbits every PMC season for both the Northern Hemisphere and the Southern Hemisphere. The 2010–2011 dataset consists of a total of 180 orbits that have been performed over 12 d in the two mentioned northern hemispheric PMC seasons. In 2010, the Odin/OSIRIS tomographic scans covered tangent altitudes around 74–88 km; in 2011 this range was adjusted to about 76–87 km.
Temporal and spatial coincidence criteria can vary broadly for validation
studies of satellite instruments depending on measuring technique and
comparison quantity. The fact that PMCs are small-scale and variable
phenomena places high demands on the spatial and temporal coincidence
criteria. The time period for this study extends from June to August in 2010
and 2011 (see also Table 1 in Hultgren et al., 2013). CIPS–OSIRIS
coincidences occur between 78
The common volume observations occur between 78
Example orbit showing a CIPS/OSIRIS coincidence on a polar map
plot for CIPS orbit 50777 and OSIRIS orbit 17098. Panel
AIM CIPS occurrence of PMCs in the common volume with OSIRIS
during the observations in 2010–2011. The plot shows the number of CVs
containing a certain fraction of CIPS pixels with identified PMCs. Average
latitude is 80
Recently, Benze et al. (2018) demonstrated a method for comparing PMC
observations from the normal (non-tomographic) OSIRIS limb scans to CIPS.
This method took into account the measurement geometry and instrument
sensitivity. By performing a detailed common volume comparison, their study
showed that the PMC brightness from the normal OSIRIS scans agrees well on
average (
When comparing observations from two instruments with different viewing
geometry and resolution it is necessary to both define the appropriate
common volume and make the observational quantities comparable. The signal
from each satellite needs to be integrated to fill the common volume. The
primary PMC product of OSIRIS is the volume scattering coefficient (m
Overview of parameters.
Since the OSIRIS tomographic PMC products are reported on a
vertical–horizontal plane with the vertical axis as altitude (rather than
tangent altitude as for the normal OSIRIS PMC retrievals),
Albedo at a scattering angle of 90
The two quantities,
As mentioned earlier, OSIRIS observes PMCs and the background atmosphere at
wavelengths between 275 and 810 nm with a resolution of about 1 nm, while
CIPS observes PMCs in a wavelength region centred at 265 nm with a width of
15 nm. The tomographic retrieval provides the local scattering coefficient
for seven wavelength intervals in the UV (centred at 277.3, 283.5, 287.8,
291.2, 294.4, 300.2 and 304.3 nm; also see Table 1 in Karlsson and Gumbel,
2005) and uses these to retrieve the microphysical properties. The reported
CIPS albedo is normalized to a solar scattering angle of 90
Each CIPS pixel is transformed into scattering conditions of OSIRIS by
multiplication of the conversion factors
Besides comparing the cloud albedo between the instruments, we also extend
our study by comparing the cloud ice. While OSIRIS tomography reports cloud
ice as ice mass density (IMD) in units of (ng m
The statistical analysis of albedo is based on 180 coinciding satellite
orbits from NH 2010 and NH 2011. Out of these 180 orbits, 2 OSIRIS orbits
from NH 2010 had to be discarded due to geolocation errors. For each common
volume observation, approximately 14 CV element observations with CIPS are
performed, making a total of approximately 2492 possible CV element
observations available for the statistical analysis for the 2010 and 2011
seasons. The coarser horizontal resolution of OSIRIS requires the
application of appropriate averaging to the CIPS data before a comparison of
retrieved cloud properties is feasible. As discussed in Sect. 2.1, the
OSIRIS averaging kernel can be represented by a Gaussian distribution with
FWHM of 280 km. As a consequence, each comparison between CIPS and OSIRIS
requires the CIPS data to be averaged over several of the common volume elements
of size 55 km
We have performed a statistical analysis of albedo and IWC between OSIRIS and CIPS in the common volume. To gain deeper insight into the relationship between both datasets, we have also compared albedo and IWC along single orbits. The previous section described the coincidence criteria and the method for making the PMC data from OSIRIS and CIPS comparable in the common volume. The cloud properties from each instrument are made comparable in the common volume by integrating the OSIRIS local scattering coefficient and IMD vertically and taking the horizontal mean of the CIPS albedo and IWC. In addition, as described in the previous section, the CIPS albedo is transformed into the scattering conditions of OSIRIS. A coincidence criterion of 5 min has been applied, and a preliminary selection based on the quality flag has been performed to eliminate questionable data pixels. This section is divided into two different parts, starting with the statistical comparison of albedo and IWC and the uncertainties related to the choice of comparison method used in this study in Sect. 4.1. In Sect. 4.2 we present the results from the individual orbits.
Figure 4 shows a scatter plot comparing the OSIRIS albedo and CIPS albedo in
the common volume for the total set of 1292 observations. Each dot
represents the albedo inferred from OSIRIS and the corresponding albedo
inferred from CIPS in the common volume element, and the grey dashed line
denotes the one-to-one line. Figure 4 shows that OSIRIS albedo and CIPS
albedo generally agree well both for faint clouds and bright clouds,
although for most cases, OSIRIS albedo is higher than CIPS. We determine the
offset between the OSIRIS and CIPS albedo results as
The 3-D common volume is defined by the horizontal extent of one
OSIRIS tomographic pixel (
Scatterplot of OSIRIS and CIPS observations of albedo in the common volume. The grey dashed line denotes the one-to-one line. The blue line is the regression line. The average albedo for CIPS and OSIRIS for all the CV in the figure is indicated in the bottom right of the figure. OSIRIS error bars are a combination of systematic and statistical uncertainty. CIPS error bars are a combination of statistical uncertainty and uncertainty due to handling of dim cloud pixels (black), while the extended error bars (red) denote the uncertainty that is introduced by conversion factors that accounts for the difference in wavelength and scattering angle between the instruments. The reader is referred to the error discussion in the text for a more detailed description of CIPS and OSIRIS errors.
The error bars in Fig. 4 represent the error from the basic uncertainty
discussed in detail in section “Discussion of OSIRIS tomography uncertainties” (OSIRIS) and section “Discussion of CIPS PMC uncertainties” (CIPS), combined
with the error that is introduced by transforming the data into a comparable
property in the CV. For OSIRIS, the directional albedo is, as previously
mentioned, obtained as the vertical column integral over the scattering
coefficients. In order to represent the uncertainty of the column albedo
caused by this integration, the error bars of the albedo include the root of
the sum of the squares of both the systematic error (10 % due to
calibration) and the estimated random error of the scattering coefficient
described in Sect. 2. The resulting error of the OSIRIS albedo is
dominated by the uncertainty of the brightest PMC pixels in the column. For
dim or PMC-free areas, the albedo error is essentially determined by the PMC
retrieval threshold at each altitude, which sums up vertically to about
For CIPS, the directional albedo is obtained as the horizontal average of
the pixels contained in the CV element. The corresponding uncertainty is a
combination of the uncertainty due to the conversion factors, a statistical
uncertainty (of
To further analyse the albedo bias between the instruments and to quantify
the contribution from different sources it is useful to again consider how
the dim pixels are treated in the CIPS retrieval. The CIPS sensitivity to
faint clouds has been quantified by Lumpe et al. (2013) and is shown in
their Fig. 18. The sensitivity to faint clouds is strongly dependent on
solar zenith angle and the detection rate is highest for high SZA and
declines for lower SZA, especially for the faint clouds. The SZA during the
date and time for the tomographic scans in 2010 and 2011 is around
60
Relative difference in cloud albedo in CV elements (
Scatterplot of OSIRIS and CIPS mean albedo in the CV. Same as Fig. 4, but using thresholds on CIPS fill factor of 95 % and an OSIRIS scatter
coefficient threshold of
To account for the differences in sensitivity, we additionally apply a
retrieval threshold on the OSIRIS cloud scattering coefficient of
Figure 7 shows a scatterplot comparing OSIRIS and CIPS IWC in the common
volume for a total set of 788 common volume element observations. Each dot
represents the IWC inferred from OSIRIS and the corresponding IWC from CIPS
in the CV element. The systematic uncertainty of CIPS IWC strongly increases
with decreasing particle size. As described in Sect. 2.2.1, we take this
into account by screening out the suspicious IWC detections, only including
CIPS pixels that report a particle radius
The error bars in Fig. 7 represent a total error that is relevant for the
comparison between the instruments by combining systematic and statistical
error from each dataset. Based on the discussion in the section “Discussion of OSIRIS tomography uncertainties”, the
OSIRIS error bars combine the absolute uncertainty of IMD due to the 10 %
measurement accuracy and a statistical uncertainty that is obtained by
propagating the uncertainty of the scattering coefficient through the
derivation of the IMD at each level. The resulting error in IWC in the
common volume element is calculated in the column integration as the
combined error of IMD from all vertical levels. Similarly, the CIPS
error bars (in grey) represent the systematic and statistical uncertainty
propagated from all individual CIPS pixels in the CV element. The individual
uncertainty for each pixel is taken from Fig. 21 in Lumpe et al. (2013),
where both the estimated systematic uncertainty and the statistical
uncertainty of CIPS IWC are given as a function of IWC and particle radius
for different ranges of solar zenith angles. For the range of solar zenith
angles in our study (59 to 71
Unlike for the albedo comparison (Figs. 5 and 6), where correction factors for phase and wavelength are applied, no corrections are needed for the comparison of IWC in the CV element, and hence no additional error bars for this conversion are needed. Generally, CIPS and OSIRIS IWC observations agree well within the common volume. The relative difference is large for the dimmest clouds and decreases for stronger clouds. The correlation coefficient is calculated to be 0.91 between the instruments.
The results from the above statistical comparison show that the instruments agree very well on average, that the choice of method used for the instrument comparison is valid, and that the time constraint and size of the common volume are suitable. In this section, we continue to compare the instruments using the individual strengths of each instrument, i.e. the horizontally resolved CIPS data and the vertically resolved OSIRIS data. As described in Sect. 2.1, the ability of OSIRIS to resolve individual cloud structures varies stochastically. As the number of measurements along the orbit is sparse, the resolution depends on the placement of lines of sight relative to the actual cloud structure. Therefore, when comparing CIPS data to the OSIRIS tomography in this section, we apply a horizontal integration of the CIPS data of both 56 km (size of OSIRIS retrieval pixel) and 280 km (typical OSIRIS averaging kernel) as limiting cases.
We present the observations of cloud brightness from three individual OSIRIS
orbits and the coinciding CIPS orbits in Figs. 8, 9 and 10. These
particular orbits were chosen to illustrate some examples of when the cloud
in the CV show good agreement, and point out one example when the cloud
observations in the CV disagree and thus illustrate for the reader the
range of cloud observations available for this study. Figs. 8a, 9a and 10a show the vertically and horizontally resolved OSIRIS scattering
coefficient for the subset of the orbit that overlaps the coinciding CIPS
orbit strip. The abscissa is given in OSIRIS AAO and
the vertical axis covers the subrange 78–88 km. Figs. 8b, 9b and 10b show CIPS
albedo for the subset of pixels that overlap the OSIRIS field of view along
the OSIRIS line of sight. The horizontal pink lines mark the width of the OSIRIS
field of view, and the CIPS pixels within these lines denote the pixels
within the CV. Note that this plot is not the normal CIPS orbit strip image
that uses polar projection map, but only the subset of pixels in the CIPS CV
level 2 geolocated data that we have plotted on a new grid to facilitate a
comparison using the same abscissa. The region in close proximity
(
Scatterplot of OSIRIS and CIPS common volume ice water content. The average IWC for CIPS and OSIRIS is indicated in the bottom right of the figure. The error bars are a combination of the systematic and statistical uncertainty from each instrument, thus representing a total uncertainty that is relevant when comparing the datasets. The grey dashed line denotes the one-to-one line. The blue line is the regression line.
For OSIRIS orbit 51236 (Fig. 8) observed during 16 July 2010, a bright
continuous cloud layer is visible between approximately 82.5 and 85 AAO, with
the brightest region of the cloud at 84–84.5 AAO and at altitude
Common volume observations of cloud albedo for OSIRIS orbit 51236
and CIPS orbit 17556. The coincidence occurs at latitude 78
For OSIRIS orbit 50796 (Fig. 9) observed on 17 June 2010, a cloud layer is
visible between 81 and 83 AAO, with the brightest cloud layer at an altitude of 85 km at 81.5 AAO. Coinciding CIPS orbit 17117 shows a cloud in the same region that is highly structured in the CV. By comparing the mean albedo in the CV, we note that OSIRIS albedo is 2–
As in Fig. 8, but for OSIRIS orbit 50796 and CIPS orbit 17117. The
coincidence occurs at latitude 79
For OSIRIS orbit 51646 (Fig. 10) taken during 16 July 2010, a bright cloud
layer with a vertical extent of
As in Fig. 8, but for OSIRIS orbit 51646 and CIPS orbit 17965. The
coincidence occurs at latitude 79
In this study, we have compared the PMC cloud properties cloud albedo and
IWC from Odin OSIRIS limb tomography to the nadir-viewing AIM CIPS. The
analysis is performed for 2010 and 2011 in the Northern Hemisphere for a total set
or 180 coinciding orbits at latitudes from 78 to 80
First, we have extended the previous OSIRIS error description by Hultgren et
al. (2013) by performing a detailed error characterization for the local cloud
scattering coefficient and IMD that takes into account absorption of
mesospheric ozone along the LOS. Second, we have compared these cloud
properties to common volume observations from CIPS. To be able to compare
the common volume cloud properties from two different satellite instruments
(OSIRIS using limb geometry and adapting spectroscopy to retrieve cloud
properties, and CIPS using nadir geometry and adapting multi-angle phase
function observations of PMC to retrieve cloud properties), it is necessary
to account for the differences in scattering conditions, observational
volume and sensitivity. In this study, we have averaged the high horizontal
resolution CIPS albedo and IWC to the coarser horizontal resolution of
OSIRIS tomography. Additionally, we have a vertically integrated OSIRIS
scatter coefficient and IMD to obtain albedo and IWC comparable to CIPS. We
have accounted for the differences in scattering conditions by transforming
CIPS albedo into the SSA and wavelengths used by OSIRIS. By adopting a very
narrow spatial and temporal coincidence criterion, we have been able to
capture a large variety of albedo and IWC in the common volume. We have
shown that the OSIRIS error characterization of the volume scattering coefficient
and IMD is valid by demonstrating that the cloud properties within the
common volume largely agree within the specified error for each instrument
analysis. We find that the OSIRIS cloud scattering coefficient shows excellent
agreement with CIPS cloud albedo with a correlation coefficient of 0.96,
although in the common volume OSIRIS observes brighter clouds than CIPS. The
bias between the instruments is found to be
A reason for why OSIRIS IWC is biased low to CIPS IWC is considered to
arise from how IMD is calculated in OSIRIS PMC retrieval.
When calculating IWC from the vertical integration of OSIRIS IMD data, we
currently only take into account retrieval pixels that are bright enough for a spectroscopic size retrieval to be feasible. It is possible that the
OSIRIS analysis misses ice from weak pixels where mean radii smaller than 20 nm are reported. Such weak cloud pixels typically cover an altitude range of 1 km in the upper part of the cloud. A rough estimate of how much ice OSIRIS
misses by ignoring such pixels can be given by considering how much ice can
be produced when converting a typical concentration of water vapour at 86 km
(e.g. 3 ppm water vapour in
Bailey et al. (2015) show that the reported v4.2 CIPS IWC is 50 % smaller
than SOFIE IWC in the common volume, but only
We have also performed a more detailed comparison of cloud properties in individual orbits. The sample of orbits used in this study illustrates the strength of this dataset. The vertically resolved OSIRIS tomography in combination with the horizontally high resolved CIPS data provides tools to study cloud structures and how these, in turn, affect the observed cloud properties in the CV.
This study has validated the OSIRIS tomographic PMC cloud brightness and ice content against nadir-viewing CIPS observations. It has addressed the potential errors from the dataset itself, as well as errors inherent to the comparison of limb tomography and nadir PMC retrievals. Due to the limitation of a total of 18 d of observations during the seasons of the Northern Hemisphere in 2010 and 2011, we have not performed a detailed comparison of cloud frequency. A follow-up paper is planned to discuss how to best compare OSIRIS limb tomography and CIPS column-integrated data when it comes to PMC particle size retrievals.
The Odin OSIRIS tomography PMC dataset used in this article, as well as codes used for the analysis, is available upon request to Lina Broman. AIM CIPS data are available for download from the Laboratory for Atmospheric and Space Physics web page:
JG and SB presented the original idea of this study. LB designed the study, carried out the analysis and wrote the majority of the paper. SB and OMC were responsible for processing the tomographic OSIRIS data. OMC and JG performed tests of the resolution of the OSIRIS tomography. All co-authors contributed to the interpretation of the results, provided feedback and contributed equally to the improvement of the paper.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Layered phenomena in the mesopause region (ACP/AMT inter-journal SI)”. It is a result of the LPMR workshop 2017 (LPMR-2017), Kühlungsborn, Germany, 18–22 September 2017.
We thank Jerry Lumpe for valuable discussion related to the CIPS retrievals. We thank Gerd Baumgarten for providing scattering data from T-Matrix calculations. We thank Matthew DeLand for the helpful discussions regarding Sect. 5. We are grateful to the two anonymous reviewers whose comments have greatly improved this paper. Odin is a Swedish-led satellite funded jointly by Sweden (SNSB), Canada (CSA), France (CNES) and Finland (TEKES). Since 2007, Odin is a third-party mission of the European Space Agency. AIM is supported by NASA Small Explorers contract NAS5-03132. We gratefully acknowledge the efforts of the entire Odin and AIM development, science, and operations teams.
The article processing charges for this open-access publication were covered by Stockholm University.
This paper was edited by Andreas Engel and reviewed by two anonymous referees.