Introduction
The impact of cirrus on the atmospheric radiative energy budget and the
Earth's climate system is uncertain , which is partly due to
the limited knowledge about the formation and development of cirrus
. Until now the fraction that homogeneous or heterogeneous
ice nucleation contributes to the cirrus formation has not been sufficiently
quantified . As a result, the evolution of the cirrus
microphysical properties during its life-cycle is insufficiently represented
in climate models . Furthermore, the influence of cirrus on
the Earth's radiation budget is highly variable because it strongly depends
on their
microphysical properties such as ice crystal number, size and shape
. In
particular, optically thin cirrus (τ≤0.03), so called sub-visible
cirrus (SVC), is difficult to observe and not well represented in general
circulation models . Sub-visible cirrus may extend over
large areas . Therefore, their influence on the energy
budget of the Earth can probably not be neglected.
estimated the annually and globally averaged radiative forcing of SVC with
+1 W m-2 (warming effect), while the local forcing might be
significantly higher. Especially the location and time where SVC occurs
determine their radiative effects. Whether SVC heats or cools the atmosphere
depends on surface albedo α, solar zenith angle θ0 and cirrus
optical thickness τ . In general SVC and cirrus have a
heating effect at the top-of-atmosphere (TOA) since the reduction of outgoing
infrared radiation usually dominates the cooling effect due to reflection of
solar radiation .
In order to quantify the microphysical and optical properties of SVC, which
are needed to determine their radiative effects, more observations of this
cloud type are required. As a consequence, several satellite missions and
field studies were performed in the past, e.g., by ,
, , and , to
establish a reliable database on SVC. Airborne in situ measurements by
, , , and
were utilized to determine ice crystal size and
ice crystal shape of SVC. Optical and
microphysical parameters derived from these measurements are used in
radiative transfer simulations (RTS) and numerical weather prediction and
climate modeling . Despite these efforts, in situ observations
of SVC are still scarce and partly accidental due to the challenge of
locating SVC. sampled an Arctic SVC after it was
detected by an airborne lidar. Airborne campaigns dedicated to visible
cirrus, e.g., the Contrail,
volcano and
Cirrus Experiment (CONCERT, ), Mid-Latitude Cirrus
(ML-CIRRUS, ) and tropical cirrus sampled during the
Airborne Tropical TRopopause EXperiment (ATTREX) are more frequent
and occasionally
include observations of SVC. Further international airborne missions like the
Tropical Composition, Cloud and Climate Coupling (TC4) and
the Cirrus Regional Study of Tropical Anvils and Cirrus Layers – Florida
Area Cirrus Experiment (CRYSTAL-FACE) mission were conducted, trying to fill
the knowledge gap about the formation process and physical properties of
tropical cirrus .
While satellite observations are suited to study the global coverage of
cirrus, their spatial and temporal resolution is still limited and can not
resolve the high spatial variability of cirrus. As a consequence the
three-dimensional (3-D) radiative effects of different cirrus properties,
e.g., τ, ice crystal size and shape, can not be studied using the coarse
resolution of satellite remote sensing. Ground-based lidar and radar remote
sensing can provide a high temporal resolution but are limited to a fixed
location. In situ airborne measurements can provide cirrus properties with
both.
For passive remote sensing of cirrus, nadir and sideward viewing observations
are available. For nadir measurements τ and the effective radius
reff of liquid water droplets can be retrieved by the
bi-spectral reflectivity method following and
. , , and
adapted this method for ice clouds by introducing some
modifications with regard to the thermodynamic phase and crystal shape of the
ice particles. Especially due to the crystal shape and low values of τ,
cirrus retrievals lead to additional uncertainties compared to liquid water
clouds .
For low τ, the reflected radiation is dominated by the surface
reflection below the cirrus. This may introduce a bias in the retrieval of
τ of up to 30 % when α is not accurately known or
inhomogeneous . Over dark ocean surfaces the radiance I
reflected by the cirrus might be weak and can be on the same order of
magnitude as Rayleigh scattering in the atmosphere. In addition,
inhomogeneities of cirrus lead to three-dimensional (3-D) radiative effects, which may cause a bias in the
one-dimensional (1-D) radiative transfer simulations .
Incorrectly assumed ice crystal shapes also contribute to the retrieval
uncertainty. investigated the influence of ice crystal
shape on derived τ and reff. Evaluating a case study, they
concluded that different shapes can lead to relative differences in τ of
up to 70 %. In a worst-case scenario, all these effects render retrievals
of τ rather inaccurate. However, observations in the sideward or limb
viewing direction and improvements of retrieval techniques may overcome these
limitations.
Limb measurements of SVC and cirrus were first introduced and utilized for
satellite measurements by . Since then, several
applications based on this method have been developed and are routinely used,
e.g., for trace gas measurements
.
Many trace gas retrievals from aircraft, balloons and satellites are based on
ultraviolet (UV)/visible (VIS)/near infrared (IR) sideward viewing
measurements in combination with differential optical absorption spectroscopy
(DOAS), e.g., performed by . Compared to nadir
observations, radiance measurements in limb or sideward viewing geometry are
supposed to be more sensitive to optically thin clouds due to their
observation geometry. One recent study was accomplished by
, who used satellite limb measurements, especially for SVC
investigation in the tropical tropopause layer. This data source improved SVC
observations with respect to cloud climatology and microphysics.
In the present study, retrievals of τ are based on simultaneous airborne
nadir and sideward viewing observations of cirrus and are compared to
elaborate the potential of sideward viewing measurements to derive optical
parameters of SVC and optically thin cirrus. This includes a sensitivity
study using RTS presented in Sect. 2 and measurements collected on board the
High Altitude and LOng range research aircraft (HALO) of the German Aerospace
Center (DLR). With a maximum ceiling altitude of around 15 km HALO is
capable of operating in and above SVC and cirrus at mid-latitudes and in
polar regions for in situ measurements. The airborne observations are
obtained with the Spectral Modular Airborne Radiation measurement sysTem
(SMART) and the Differential Optical Absorption
Spectrometer (mini-DOAS) both assembled on HALO. The
instrumentation is introduced in Sect. 3. Observations from four campaigns,
the Mid-Latitude Cirrus experiment (ML-CIRRUS), the Next-generation Aircraft
Remote sensing for Validation Studies (NARVAL North and South), and the
Aerosol, Cloud, Precipitation, and Radiation Interactions and Dynamics of
Convective Cloud Systems (ACRIDICON-CHUVA) are used to
cross-calibrate the two individual instruments in terms of absolute radiance
I as presented in Sect. 4. In Sect. 5 an iterative retrieval of τ is
introduced. Utilizing the cross-calibrations together with nadir and sideward
viewing measurements of upward I, the retrieved results are presented and
compared to reference measurements of τ to emphasize the advantages of
sideward viewing observations. Section 6 concludes the study.
Sensitivity of upward radiance measurements in the nadir and sideward viewing directions
Radiative transfer simulations are performed to investigate the sensitivity
of solar radiance measurements in nadir and sideward viewing geometry for SVC
and thin cirrus. In this way the potential of sideward viewing versus nadir
observations for cirrus cloud parameter detection is examined.
Illustration of the measurement geometry. (a) shows the
side view with solar zenith angle θ0 and the viewing angle
θV. The opening angle of the nadir looking radiance sensor
of SMART is indicated by Δ. The top view (b) shows the
definition of the relative solar azimuth angle ϕ between the
line-of-sight (LOS) and the Sun.
Figure illustrates the measurement geometry. The solar
zenith angle θ0 is the angle between zenith and the Sun. The viewing
angle θV represents the angle of the sensor viewing
direction which is measured between the line-of-sight (LOS) and the nadir
direction. For a sensor measuring in nadir θV is
0∘ and a sensor orientation close to the horizon is around
θV≈90∘. The relative solar azimuth angle
ϕ represents the angle between the LOS and the Sun's direction. It is
calculated from the difference in the azimuth angle of the Sun and the
azimuth angle of the observation geometry of the optical inlets. For ϕ=0∘ the LOS is pointing directly in the direction of the Sun and with
ϕ=180∘ the LOS is looking away from the Sun.
For the RTS a typical mid-latitude cirrus with a cloud base height of
10 km and a cloud top height of 12 km is assumed. This closely
represents the cloud situation which is investigated in Sect. 4. Calculations
are performed for θ0=25, 50 and 75∘, representing three
different scenarios. The relative solar azimuth angle is set to ϕ=0,
90 and 180∘.
The simulations are carried out with the libRadtran 2.0 radiative transfer
package . The Fortran 77 discrete ordinate radiative
transfer solver version 2.0 (FDISORT 2) following is
selected to run the simulations. The incoming extraterrestrial solar flux
density given by is applied and molecular absorption is
calculated using LOWTRAN . A marine aerosol profile is
chosen , and for vertical profiles of temperature,
humidity, and pressure, a mid-latitude summer atmosphere profile is assumed.
A spectral α typically for oceans is chosen according to
. To represent ice crystals, a mixture of different
particle shapes is used when not otherwise specified. The ice crystal
scattering phase function is parameterized according to .
Wavelength sensitivity
Using solar spectral radiation for passive remote sensing purposes,
measurements at wavelengths sensitive to scattering and absorption by liquid
water droplets and ice crystals are selected. Wavelengths less than λ= 900 nm are applied to retrieve τ from nadir radiance
measurements. Figure a presents simulated
upward radiances IRTS reflected by an optically thin cirrus
with τ=0.03 and reff= 10 µm, as well as
clear sky radiance as a function of the sensor viewing angle. Radiative
transfer simulations for two wavelengths, λ= 532 nm and
λ= 1180 nm, are carried out. To easily distinguish the different
geometries, simulated I in nadir geometry is denoted with
IRTSN, while all geometries deviating from nadir
are referred to sideward viewing geometry and are indicated by
IRTSV. The sensitivity ετ is
defined by
ετ=dIdτ.
Simulated upward radiance IRTS at λ= 532 nm and λ= 1180 nm
for the cloudy (solid line) and clear sky (dashed line) cases as a function
of the viewing angle θV. The left plot shows simulations
for a SVC with τ=0.03 (a) and the right plot presents the
simulations for a thick cirrus with τ=2.0 (b). In the
corresponding lower plots the relative difference between cloud and clear sky
atmosphere with respect to the cloudy atmosphere is
shown.
In general, IRTSV increases with increasing
θV due to the longer LOS. For a wavelength of λ= 532 nm, no difference between cloudy and clear sky conditions is
discernible for all θV, because Rayleigh scattering by
molecules dominates and exceeds the scattering by thin cirrus. Therefore, at
λ= 532 nm SVC with τ=0.03, which is presented in the
simulations, can not be detected. Conversely, for λ= 1180 nm
separation between the simulations with and without cirrus at large viewing
angles for θV>70∘ is present because the
reflected IRTSV is increased due to a larger LOS.
At λ= 1180 nm wavelength Rayleigh scattering is comparably weak
and does not significantly contribute to the reflected radiation. In the
nadir direction, a detection of SVC is not possible due to low τ and the
overwhelming backscattering from the ground.
For comparison, simulations of a thicker cirrus with τ= 2.0 are
presented in Fig. b. Here, the influence of
the Rayleigh scattering at λ= 532 nm is reduced and a distinction
between cloudy and clear-sky becomes possible. However, the relative
difference between cloudy and clear-sky conditions is still more pronounced
at λ= 1180 nm.
The RTS suggest that sideward viewing observations at near-IR wavelengths
(λ>900 nm) are more suitable for the detection of SVC and
cirrus. As a result the retrieval in Sect. 4 is performed at 1180 and
1600 nm wavelengths in the IR region, which are sensitive to τ and
reff and not disturbed by Rayleigh scattering.
Optical thickness and viewing angle
In general, back-scattered radiation by clouds increases with increasing
τ. This sensitivity (see Eq. ) is the basis of most
retrieval algorithms of cloud optical properties. To quantify how
ετ is effected by θV of the sensor, RTS
are performed for a set of different θV ranging between
θV=0∘ (nadir) and θV=90∘ (sideward viewing). Cirrus optical thickness is varied in the
range of τ= 0.03–4 covering various kinds of cirrus clouds.
Simulated radiance IRT,1180 for three different sensor
orientations as a function of cirrus optical thickness τ. Results for
solar zenith angles of θ0=25∘ (a) and θ0=75∘ (b) are displayed. The sensitivity ετ
is given in the lower panels.
The first simulations presented in Fig. display
simulated IRTS,1180 at λ= 1180 nm wavelength for
two different θ0=25∘ (a) and θ0=75∘ (b)
as a function of τ. For each scenario, ετ is
calculated and given in the lower panels of Fig. .
Simulations for nadir geometry are represented by solid black lines. Results
for sideward viewing sensor orientations are shown by dashed
(θV=53∘) and gray (θV=78∘) lines. All scenarios show an increase in
IRTSV for increasing τ, which results from
enhanced reflection.
Due to the apparently longer LOS for both θ0, sideward viewing sensor
orientations yield larger ετ of simulated
IRTSV as compared to the nadir geometry for cirrus
clouds with τ≤1 which includes SVC. This indicates that sideward
measurements are most suited to retrieve τ below 1 and for the detection
of SVC. The almost linear increase in the nadir radiance
IRTSN indicates a constant ετ for
the investigated range of τ and θ0. For τ≥1 the
sensitivity of sideward viewing observations is in the same range compared to
nadir measurements or slightly lower depending on the combination of
θ0 and θV.
For low τ and a high Sun, the highest ετ is given for
the sideward viewing geometry (θV=78∘) for τ≤1. A similar pattern emerges for low Sun (θ0=75∘),
resulting in larger ετ and a steep decrease for increasing
τ. It shows that ετ decreases with τ and for
τ<2 drops below ετ of nadir measurements. The
sensitivity of I with respect to τ can also be interpreted in terms of
the uncertainty of retrieved τ related to an initial uncertainty in
measured I. The higher the ετ, the weaker the impact of
uncertainties in the measurements on the uncertainties of the retrieved
τ. As shown in Fig. b, a high
ετ is calculated for IRTS,1180 for τ≤1 and indicates a lower measurement uncertainty. Therefore, sideward viewing
observations at λ= 1180 nm allow a more accurate determination
of τ compared to nadir observations for optically thin clouds with τ≤1.
In a second step, the influence of ϕ is investigated on
IRTSV in respective simulations.
Sensitivity ετ at 1180 nm in units of
mW m-2 sr-1 as a function of viewing angle θV
and relative solar azimuth angle ϕ for cirrus optical thickness τ
and solar zenith angle θ0. Panel (a) for τ=0.1,
θ0=25∘, Panel (b) for τ=0.1, θ0=75∘, and Panel (c) for τ=2, θ0=25∘
in (c) and (d) for τ=2, θ0=75∘.
Different scales of the plots have to be considered.
Figure shows ετ for a
wide range of θV between 0∘ and 90∘
and ϕ between 0∘ and 180∘ for two clouds with τ=0.1 and τ=2 and two different solar zenith angles (SZA) of θ0=25∘ and θ0=75∘.
The graphs represent ετ in units of mW m-2 nm-1 sr-1 for different ϕ as a function of
θV.
For τ=0.1 and θ0=25∘
(Fig. a), ετ ranges
between 5 and 66 mW m-2 nm-1 sr-1. For larger
θV (sideward viewing observations), ετ
increases significantly, reaching the maximum for
θV=90∘ and ϕ=0∘. Observations under
these angles are better suited in comparison to other angle combinations as
they enable us to achieve the largest possible ετ and
reduced relative measurement errors, which results in increased retrieval
accuracy.
A similar pattern is derived for simulations assuming a lower Sun (θ0=75∘) as shown in Fig. b.
Compared to θ0=25∘ the increase in ετ for
θV=90∘ and ϕ=0∘ is stronger,
reaching values of 377 mW m-2 nm-1 sr-1, while for all
other geometries ετ almost remains constant at the same
magnitude, reaching 80 mW m-2 nm-1 sr-1. Additionally, the
maximum ετ is more concentrated on a single combination of
θV and ϕ represented by the high peak for ϕ=0
compared to all other ϕ. Therefore, measurements in the range of these
angles are recommended to achieve high values of ετ for
reasonable retrievals of τ.
Figure c shows the simulated
ετ for clouds of τ=2, θ0=25∘ and a
wide range of geometries. Compared to the optically thin cirrus, the maximum
of ετ is reduced for optical thick cirrus, not exceeding a
value of 15 mW m-2 nm-1 sr-1 and shifted to smaller θ0.
While sideward viewing measurements are predicted to become saturated for thick clouds, for low τ
the optimal θV is about θV=60∘, with the largest
ετ occurring for ϕ between 0 and 60∘. Respective
simulations for τ=2, θ0=75∘ (low Sun) are presented in
Fig. d. Here, the maximum of ετ
is small, with 5 mW m-2 nm-1 sr-1 at θV
and ϕ=0 compared to all other simulations varying τ and θ0.
The RTS show that the choice of the best viewing geometry (nadir or sideward
viewing observations) strongly depends on τ and ϕ. In order to
probe a large range of cirrus with sufficiently large retrieval sensitivity,
measurements in different viewing directions, at least in the nadir and
sideward viewing directions depending on τ and θ0, are
recommended. Measurements in sideward viewing geometry are strongly dependent
on θV, especially around θV=90∘. In order to avoid spurious results by mispointing with the
sensor, a careful alignment of the optical sensor and an accurate
determination is required. Considering these findings, the retrieval of
τ in Sect. 4 is performed for θV≤60∘ only.
Influence of surface albedo
The influence of α on the retrieval of cloud optical properties
derived by passive remote sensing using the Moderate Resolution Imaging
Spectroradiometer (MODIS) was investigated by . They
showed that retrievals of clouds with τ<0.5 are strongly influenced by
variations in α. Based on RTS, concluded that
IN measured in the nadir direction strongly depends on the
underlying surface reflectivity and that uncertainties in assumed α
may cause errors of up to 50 % in the retrieval of τ.
In order to quantify and compare the influence of α on I measured in
different θV and nadir directions, RTS are performed. To
cover the natural variability of surfaces ranging from ocean surface to
ice-covered regions, α is varied between α=0.1 and α=0.9. Figure shows simulated
IRTS,1180V at λ= 1180 nm wavelength
for two clouds with τ=0.1 and τ=2 and both observation
geometries.
Influence of the surface albedo α on the measured upward
radiance IRTS,1180V at λ= 1180 nm as a
function of cirrus optical thickness τ and sensor orientation
θV.
In general, the reflected I increases with increasing α. The
stronger the increase, the more strongly the measurements are effected by
α. For both observation geometries, the steepest derivative,
γ=dIdα,
is obtained for the thin cirrus with τ=0.1. In general for increasing
τ of thick clouds, α becomes less important for I compared to
cirrus clouds with lower τ. To quantify the impact of changes in
α, the relative difference between IRTS simulated for
α=0.1 and α=0.9 is calculated for each case and presented
in Table . Maximum differences of up to
84 % are noticeable in nadir geometry for clouds of τ=0.1.
Optically thick clouds show lower dependencies on α due to the
increased contribution of radiation reflected by the cirrus. Comparing nadir
and sideward viewing geometries, the simulations show a smaller γ for
sideward viewing observations independent of α. The relative
difference in IRTSV for τ=2 between α=0.1 and α=0.9 is reduced to 14 %. This indicates that I
measured in sideward viewing geometry is less influenced by changes in
α (e.g., ). This difference in I is most
pronounced for optically thin clouds where the surface contribution to
measured I is relatively large. Under unknown or variable surface albedo
conditions, observations in the sideward viewing direction are
favored over
those in the nadir direction when retrieving the optical properties of thin
cirrus.
Relative difference in IRTS,1180nm for surface
albedo α=0.1 and α=0.9 for different viewing angles
θV and optical thickness τ.
cirrus optical thickness
viewing angle
τ=0.1
τ=0.5
τ=2
θV=0∘
84 %
69 %
44 %
θV=78∘
58 %
29 %
14 %
Crystal shape sensitivity
By changing the ice crystal shape in the RTS (similar cloud to that described
above), the sensitivity of I with respect to the ice crystal scattering
phase function is investigated and compared for different viewing geometries.
Ice crystals with shapes of columns, droxtals and plates are chosen and
implemented in the simulations to cover the natural variability of cirrus
based on the ice crystal single scattering properties provided by
. Most cirrus are composed of a mixture of ice crystal
shapes . Particle shape dependent scattering effects
are lower due to smoothing over different crystal shapes. Therefore, an ice
crystal mixture as given by is included in the simulations
and serves as a reference. This is denoted with the acronym “GHM” further
on. The simulated IRTS,1180V as a function of
θV is presented in Fig. .
Simulated radiance IRTS,1180V at λ= 1180 nm wavelength for different ice crystal shapes as a function of
the viewing angle θV of the sensor (a). In
Panel (b) the relative differences in simulated radiance with
respect to the reference shape “Ghm” are presented for the other three ice
crystal shapes.
The increase in IRTS,1180V with increasing
θV is significantly influenced by the ice crystal shape.
In the simulated cases, droxtals and the GHM ice crystal mixture show a
larger increase in IRTS,1180V with increasing
θV than columns and plates. While in nadir geometry
(θV=0∘), columns and plates have a higher
IRTS,1180V than droxtals and GHM,
IRTS,1180V measured at viewing angles
θV>50∘ is higher for droxtals and the GHM
crystal mixture. The spatial distribution obtained for droxtals results from
the enhanced forward and reduced sideways scattering compared to other
crystal shapes.
For simulations in the nadir direction the relative difference between the
lowest (droxtals) and highest (columns) IRTS,1180N
differs by up to 41.5 % of the absolute radiance of
6.1 mW m-2 nm-1 sr-1 obtained by the “GHM” crystal
mixture.
For sideward viewing observations the relative and absolute changes in
IRTS,1180V are even larger between
θV=60∘ and θV=90∘.
With increasing θV the differences in
IRTS,1180V increase up to a maximum of 43.5 %
at θV=78∘ between droxtals and plates with
respect to the absolute value of 33.8 mW m-2 nm-1 sr-1 for
GHM.
The simulations show that the relative change in simulated
IRTS,1180V due to ice crystal shape effects
increases with θV. Therefore, for cirrus of low τ the
interpretation of sideward viewing observations relies even more strongly on
a correct assumption of ice crystal shape than nadir observations.
Multiangular observations covering the angular pattern
(Fig. ) may provide sufficient information to retrieve
ice crystal shape as proposed by .
Airborne measurements
Simultaneous airborne measurements of I in nadir and sideward viewing
geometry were conducted during four campaigns using HALO. During NARVAL
shallow convection in the North Atlantic trade-wind region of the northern
Atlantic (NARVAL South, December 2013) and cloud systems associated with the
North Atlantic mid-latitude stormtrack (NARVAL North, January 2014) were
probed . During the ML-CIRRUS campaign natural and contrail
cirrus in the mid-latitudes were investigated in March and April 2014
. Deep convective clouds were observed during the Aerosol,
Cloud, Precipitation, and Radiation Interactions and Dynamics of Convective
Cloud Systems (ACRIDICON-CHUVA)
mission in September 2015 .
During these missions, a suite of different active and passive remote sensing
instruments was operated on board HALO, including passive solar radiance
measurements by SMART and the mini-DOAS
. While SMART measured radiometrically calibrated
radiance ISN in the nadir direction, the mini-DOAS
instrument simultaneously measures in the nadir and varying sideward viewing
directions in UV/VIS/IR wavelength ranges. The mini-DOAS measurements are
traditionally analyzed by applying the DOAS technique. DOAS relies on an
analysis of intensity ratios of two spectroscopic observations made under
largely different atmospheric conditions. By exploiting ratios of I, DOAS
measurements are inherently radiometrically calibrated in a relative but not
absolute sense. Therefore no absolute radiometric calibration for I for the
mini-DOAS is available. In addition to the two passive sensors, active lidar
measurements with the Water Vapor Lidar Experiment in Space (WALES) were
performed during NARVAL and ML-CIRRUS.
In Fig. a the position of the apertures at the aircraft
fuselage is indicated. The optical inlets of mini-DOAS and SMART for upward
radiation are shown in Fig. b and c, respectively.
(a) Optical
inlets of mini-DOAS (b) and SMART (c) mounted at the lower
aircraft fuselage.
The SMART instrument
Depending on the configuration, SMART measures spectral upward
FS,λ↑ and downward irradiance
FS,λ↓, as well as spectral upward radiance
ISN. The system is extensively described in
and . In this paper the focus is
on ISN measurements which are available for the
four HALO missions introduced above.
To cover almost the entire solar spectral range, SMART measures
ISN with two separate spectrometers, one for the
VIS range from λ= 300 nm to λ= 1000 nm and a second
one for sampling the IR range from λ= 900 nm to λ= 2200 nm. By merging the spectra, about 97% of the solar spectrum
is covered . The spectral resolution
defined by the full width at half maximum (FWHM) is 8–10 nm for the IR
spectrometer and 2–3 nm for the VIS spectrometer.
The radiance optical inlet of SMART has an opening angle of Δ=2∘ and a sampling time of 0.5 s. Considering aircraft groundspeed
and the distance of 500 m between the cloud and the aircraft the resulting
footprint is about 18 × 110 m for an individual
ISN measurement. For a distance of 1000 m
between sensor and cloud the footprint increases to 35 × 220 m.
Prior to each campaign SMART was radiometrically calibrated in the laboratory
using certified calibration standards traceable to NIST and by secondary
calibration using a travelling standard during the operation on HALO. The
total measurement uncertainty of ISN is about
5.4 % for the VIS and 14.5 % for the IR range which consist of
individual errors due to the spectral calibration, the spectrometer noise and
dark current, the radiometric calibration and the transfer calibration
. In Table the
contributions of each individual source of uncertainty is given for
measurements at λ= 1180 nm wavelength. The main uncertainty
results from the signal-to-noise ratio (SNR) and the calibration standard,
while spectral and transfer calibration errors are almost negligible.
Averaging a time series of measurements will reduce the contribution of
sensor noise to the signal.
Individual sources of uncertainty and total uncertainties for the
upward radiance IS,1180N at a wavelength of
λ= 1180 nm.
Source of uncertainty
λ= 1180 nm
IS,1180N
Spectral calibration
< 1 %
Radiometric calibration
8.5 %
Signal-to-noise ratio
11.6 %
Transfer calibration
< 1.1 %
Total
14.5 %
The mini-DOAS instrument
The mini-DOAS is a passive airborne remote sensing system originally designed
to retrieve vertical profiles of trace gases, aerosol and cloud particles
. The analysis is based on the DOAS technique that
applies least square retrievals to the spectral shape of the observed upward
radiance ImDV by the mini-DOAS in sideward viewing
channels . Spectral absorption bands of molecules and
particles are measured at moderate spectral resolution (FWHM = 0.47, 1.2
and 10 nm for the UV, VIS and
IR, respectively) to quantify the absorption of solar radiation by trace
gases along the light path. DOAS measurements are primarily used to infer
trace gas concentrations and associated photochemistry in the atmosphere.
Here, measured ImDV are employed for the remote
sensing of clouds.
The mini-DOAS is designed as a compact, lightweight and robust system to be
operated aboard HALO. The instrument consists of six telescopes which are
connected via fiber bundles to six optical spectrometers. One set of the
optical inlets is fixed in the nadir configuration, while the other
telescopes can be tilted between θV=0∘ and
θV=90∘. Two sets of three different
spectrometers are applied to cover the UV spectral range from 310 to
440 nm (FWHM 0.5 nm), the VIS range from 420 to 650 nm (FWHM 1 nm)
and the IR range from 1100 to 1680 nm (FWHM 10 nm). In the UV and VIS
range, charged–coupled device (CCD) sensors are used as detectors. The
detection in the IR range is performed by photo diode arrays (PDAs).
The telescopes are mounted on an aperture plate at the lower side of the
aircraft fuselage. The scanning telescopes have rectangular fields-of-view of
about 0.6∘ in vertical direction and 3∘ in horizontal
direction. During scanning measurements the telescopes are directed to the
starboard side of the aircraft. Changes in aircraft roll angles are
compensated within 0.2∘. The orientation of the nadir telescope is
kept fix with respect to the aircraft major axis. Therefore no compensation
of the aircraft roll angle is performed.
The evacuated spectrometer housing is immersed into an isolated water/ice
tank to ensure a constant temperature and pressure of the spectrometers
independent of changing outside conditions. Evacuation of the housing and
temperature stabilization is necessary to guarantee a stable optical imaging,
which is indispensable for DOAS applications. A spectral calibration of the
spectrometers ensures that wavelength shifts are less than 0.05 nm.
The WALES instrument
The Water Vapor Lidar Experiment in Space Demonstrator (WALES) is an airborne
Differential Absorption Lidar (DIAL) with additional aerosol and cloud
detection capabilities operated on the German research aircraft Falcon and
HALO .
For particle detection WALES has two backscatter and depolarization channels
at λ= 532 nm and λ= 1064 nm wavelength and an
additional high spectral resolution lidar (HSRL) channel at λ= 532 nm . The HSRL channel allows the retrieval of
the backscatter coefficient of clouds at λ= 532 nm without
assumptions about the phase function of the cloud particles. Unfortunately,
larger cirrus particles usually show a pronounced forward scattering peak,
which may contain a significant fraction of the scattered energy. This may
lead to an underestimation of τ calculated from the individual particle
extinction cross sections (see, e.g., ). The optical
thickness data presented in this paper are corrected for the forward
scattering effect following the algorithm proposed by .
To apply this correction scheme, an reff is assumed, which
determines the width of the forward scattering peak. The best compensation of
the multiple scattering decay below the cloud is found for reff= 35 ± 5 µm, in good agreement with the climatological
values proposed by . The mean correction factor for the
data set shown in this paper was 7 %.
Panels (a) and (c) show comparisons of SMART
radiance ISN and mini-DOAS raw signal for nadir
channels at λ= 1180 nm and λ= 1600 nm wavelength.
Panels (b) and (d) show time series of measured SMART
radiance IS,λN and calibrated mini-DOAS
radiance ImD,λN for the ML-CIRRUS flight on
26 March 2014. The shaded areas indicate the measurement uncertainties.
Cross-calibration
Since no radiometric calibration is available for mini-DOAS, simultaneous
measurements of SMART and mini-DOAS are used to cross-calibrate the mini-DOAS
with SMART. The cross-calibration relies on the radiometric calibration of
SMART and allows us to derive calibrated ImD from mini-DOAS
measurements. Flight sections with inhomogeneous α and various cloud
conditions are selected to obtain a calibration valid for a wide range of
different I. Such conditions were present during the ML-CIRRUS flight on 26
March 2014, including measurements over southern Germany, Belgium, the United
Kingdom, Ireland and the northern Atlantic Ocean west of Ireland. The
cross-calibration is performed for the nadir and sideward viewing scanning
telescopes of the mini-DOAS when aligned to the same cloud area as SMART. The
results are presented for two wavelengths at λ= 1180 nm and
λ= 1600 nm which are frequently used in cloud retrievals and
show the best discrimination potential for small τ as presented in the
sensitivity study. Different FWHMs of both spectrometer systems are
considered by convoluting the spectrally higher resolved measurements of the
mini-DOAS with the corresponding FWHM of the SMART spectrometer (8–19 nm).
Nadir radiance
The nadir sensors of the mini-DOAS operate in a fixed position, thus
providing a large data set of simultaneous measurements with SMART. The time
stamps of both instruments are corrected for temporal offsets in the data
acquisition. Scatter plots of IS,λN and
mini-DOAS raw data are shown in Fig. a and c for
both wavelengths. For each data point a linear regression following
and is performed. Using the method of
Theil and Sen, the influence of outliers on the regression is reduced and the
linear calibration equation IS,λN=a0⋅NmD,λN+a1 for the mini-DOAS radiances is
determined. IS,λN is the radiance measured by
SMART, NmD,λN the raw signal of mini-DOAS and
a0 and a1 the calibration coefficients. The linear regressions are
indicated by the gray lines in Fig. a and c.
For the ML-CIRRUS flight on 26 March 2014 the nadir geometry calibration
coefficients are determined as a0=0.31 mW m-2 sr-1 and a1=0.55 mW m-2 sr-1 for λ= 1180 nm with an
uncertainty of ±0.24 mW m-2 sr-1. Similar calibrations are
performed for flights during the NARVAL and ACRIDICON-CHUVA campaigns. All
calibration coefficients are summarized in
Table . The coefficients depend on
various environmental conditions where the temperature dependence of the
mini-DOAS spectrometers is the most influential parameter.
The uncertainty is mostly related to differences in the
field-of-view (FOV) and the related
difference in the observed scene and possible minor mismatches of the nadir
orientation of both sensors. This means that both sensors do not always
observe the exact same cloud area. For the λ= 1600 nm
wavelength, a0 is higher compared to λ= 1180 nm in all
analyzed flights, indicating the different spectral sensitivities of both
sensors with SMART in comparison with mini-DOAS, being relatively more
sensitive at λ= 1600 nm than at λ= 1180 nm
wavelength.
Calibration coefficients a0 and a1 in units of
mW m-2 nm-1 sr-1 for mini-DOAS nadir and scanning channel
radiance obtained for NARVAL (19 December 2013), ML-CIRRUS (26 March 2014)
and ACRIDICON-CHUVA (9, 12 and 23 September 2014).
1180 nm
1600 nm
Nadir
sideward viewing
Nadir
sideward viewing
(mW m-2 nm-1 sr-1)
a0
a1
a0
a1
a0
a1
a0
a1
NARVAL (19.12.)
0.26
5.40
0.23
0.90
0.28
1.32
0.26
0.10
ML-CIRRUS (26.03.)
0.31
0.55
0.31
0.00
0.43
0.25
0.47
0.02
ACRIDICON-CHUVA (09.09)
0.24
5.28
0.37
2.80
ACRIDICON-CHUVA (12.09.)
0.34
0.94
0.51
0.77
ACRIDICON-CHUVA (23.09.)
0.31
3.43
0.40
0.59
The derived cross-calibrations of mini-DOAS are applied to all mini-DOAS
measurements. A measurement example of a time series of calibrated mini-DOAS
radiances ImD,λN is shown in
Fig. b and d for an 18 min flight section
measured on 26 March 2014.
The radiance time series for λ= 1180 nm of both sensors agree
within the SMART error range for most data points, except for some radiance
peaks. These differences likely result from the different FOV of both
instruments and the presence of patches of low cumulus with high
reflectivity. A similar result is obtained for λ= 1600 nm. The
differences in the mean radiance between both instruments for the time period
presented in Fig. are
0.75 mW m-2 nm-1 sr-1 at λ= 1180 nm and
0.5 mW m-2 nm-1 sr-1 at λ= 1600 nm, which
results in relative differences of 5.4 % at λ= 1180 nm and
1.9 % at λ= 1600 nm compared to the SMART absolute values.
Sideward viewing radiance
The scanning telescopes of the mini-DOAS typically run in a sequential mode
scanning different θV. During selected flight segments the
scanning sequences are configured to include nadir measurements. Due to this
sequential mode less measurements from the sideward viewing channels are
available for cross-calibration with SMART because only measurements in nadir
sensor orientation are applicable for the cross-calibration. To ensure a
statistically sufficient number of samples, the entire flight of 26 March
2014 is analyzed applying the same methods used for the calibration of the
nadir channels. Figure a and c show the
cross-calibration of SMART radiances IS,λN
and mini-DOAS raw data NmD,λV and the linear
fit (gray line) used for calibration.
Panels (a) and (c) show a comparison of SMART
radiance IS,λN and mini-DOAS raw signal
NmD,λV for the scanning channels at λ= 1180 nm and λ= 1600 nm wavelength. Panels (b) and
(d) show time series of measured SMART radiance
IS,λN and calibrated mini-DOAS radiance
ImD,λV for the ML-CIRRUS flight on
26 March 2014. The shaded areas indicate the measurement errors.
For the IR scanning channels the calibration coefficients are determined as
a0= 0.31 mW m-2 sr-1 with no offset a1 for λ= 1180 nm and an uncertainty of ±0.2 mW m-2 sr-1.
Similar to the nadir channels, the calibration coefficients for the the
sideward viewing channel at λ= 1600 nm wavelength with a0=0.47 are higher compared to the λ= 1180 nm wavelength.
The calibration of the sideward viewing channels is repeated for the NARVAL
flights, while for all ACRIDICON-CHUVA flights no nadir observations of the
sideward viewing channels are available.
Table provides a summary of all
calibration coefficients derived for the sideward viewing channels.
Similarly to Fig. b and d, Fig. b and d show
time series of SMART radiance IS,λN and
calibrated mini-DOAS nadir observations of
ImD,λV with the sideward viewing channels for
an 18 min flight segment of ML-CIRRUS on 26 March 2014. In general, the
radiance pattern observed by SMART is represented by the calibrated mini-DOAS
radiance. However, individual data points differ due to differences in FOV,
resulting in mean differences of 0.78 mW m-2 nm-1 sr-1 at
1180 nm and 0.38 mW m-2 nm-1 sr-1 at λ= 1600 nm, which results in relative differences of 3.7 % at
λ= 1180 nm and 2.4 % at λ= 1600 nm compared to
the SMART absolute values. This ranges below the uncertainty range of SMART.
Temporal stability of cross-calibration
The mini-DOAS instrument is not explicitly designed to maintain a stable
radiometric calibration but more for a stable wavelength calibration. For
DOAS measurements absolute values of I are not needed as only relative
intensities are used. More important is the wavelength accuracy to determine
absorption and emission bands of gasses precisely. As a result the
radiometric calibration of the mini-DOAS can change from campaign to campaign
and even between several flights. Therefore, cross-calibration coefficients
for different campaigns and flights are derived to consider these changes in
radiometric calibration and the optical setup, for example when changing the
optical fibers. Using different calibration factors for the mini-DOAS
instrument as inferred for the different campaigns,
Fig. shows a comparison of measured I at
λ= 1180 nm wavelength from a 4 min long flight segment over the
Amazon region on 12 September 2014. The comparison clearly indicates that the
measurements of I of both sensors are not systematically biased and agree
within the errors of each sensor except when differences at small spatial
scales appear resulting from the different FOV.
Time series of the nadir radiance of SMART
IS,1180N and of the mini-DOAS
ImD,1180N nadir channel at λ= 1180 nm
using different calibrations as indicated in the legend. The uncertainty
range of SMART radiance is shaded gray.
The deviation of the different calibrations is below
2.9 mW m-2 nm-1 sr-1 which is inside the measurement
uncertainties of SMART and indicates a reasonable stability of the
calibrations.
Retrieval of cirrus optical thickness
Iterative algorithm
By using all three calibrated radiance data sets obtained from SMART
ISN, mini-DOAS nadir channels
ImDN, and sideward viewing channels
ImDV, an iterative retrieval algorithm of τ is
developed and applied. It is based on the bi-spectral reflectance method
described by and . Here, the
retrieval is adapted for ice clouds with respect to ice crystal shape and
used wavelengths, e.g., by and . To retrieve
τ, rough aggregates are assumed using pre-calculated ice crystal
parameterizations following
. The iterative algorithm utilizes the spectral reflectivity
Rλ, which is defined as the ratio of spectral upward
Iλ to spectral downward Fλ↓,
Rλ=IλπFλ↓.
For the ML-CIRRUS data, Fλ↓ is taken from the actual
SMART measurements on HALO. Measured Fλ↓ allows
us to identify and eliminate any
influence of the radiation field above the aircraft, for example by cirrus.
As an alternative to pre-calculating look-up-tables (LUTs) by extensive
forward simulations, an iterative algorithm is applied that runs RTS adjusted
to each single measurement. This allows us to set up simulations by actual
input parameters for each measurement, e.g., θ0, ϕ, longitude,
latitude and flight altitude. In that way, uncertainties caused by inaccurate
assumptions in the RTS input are minimized. Additionally, the iterative
method is not limited to a specific pre-calculated grid of τ and
reff as used in LUTs where a certain interval of preselected
τ and reff is given. The iterative algorithm automatically
adjusts the range of τ and reff without interpolation
until it reaches the final result.
Scheme of the iterative algorithm. For
every single measurement i an iteration loop is started with an initial
guess τ0 until measured Rmeas and simulated
Rsim reflectivity converge within 5 % difference or
a maximum of 100 iteration steps is reached. At the end of the process the
result is saved.
Figure shows a scheme of the retrieval algorithm,
which starts with an initial guess of τ0. Using the initial guess of
τ and of any other cloud parameters, the cloud reflectivity
Rsim is simulated and compared to the measurements
Rmeas of SMART and mini-DOAS, respectively. The ratio
between Rsim,n and Rmeas derived
for each iteration step n is used to scale the particular guess
τn by
τn+1=τn⋅RsimRmeas.
The adjusted τn+1 is used in the RTS for the new iteration step
n+1. The iteration of τ is repeated until the change in τn
between two iteration steps is smaller than 5 % or a limit of n>100
iteration steps is reached. These stop criteria determine the accuracy of the
iterative retrieval. If a lower relative stop criterion (change in τn
smaller than 5 % between two iteration steps or more then 100 iteration
steps) is used, the iteration may come closer to the true searched value and
the retrieval accuracy increases as well as the necessary iteration steps and
the computational time. To limit the computational time, the second stop
criteria is used to limit the maximum number of iteration steps. For a
typical cirrus observed during ML-CIRRUS with an average τ of 0.32, the
cirrus optical thickness can be retrieved with a accuracy of about τ ± 0.03. The retrieval of τ by SMART and mini-DOAS bases on the
measurements at λ=1180nm and is scaled to λ=532nm to consider the wavelength dependence of τ and to be
able to compare it with WALES measurement at λ=532nm.
Therefore, the retrieval considers RTS at both wavelengths. In the RTS τ
is defined and changed at λ=532nm while the measurements
are compared to simulations at λ=1180nm to determine the
correct solution.
In case of measurements of optically thin cirrus, the retrieval can be
applied for τ only. For these situations
IRTS,1600N at λ= 1600 nm wavelength
(ice absorption band) is too low and only measured with high uncertainty to
retrieve reff. For a cirrus cloud with τ=0.03, the
simulated upward nadir radiance IRTS,1600N and the
sideward viewing radiance IRTS,1600V in the range
of 0.2 mW m-2 sr-1. Such low I are in the range of the
electronic noise of the spectrometers, leading to a low signal-to-noise ratio
and high retrieval uncertainties. Especially for cirrus with low τ the
variation of IRTS,1600N and
IRTS,1600V with respect to changes in
reff is low.
Simulations show that for τ=0.5 the difference in
IRTS,1600N in the nadir direction is only
0.1mW when changing reff from 10 µm to
20 µm, indicating the low sensitivity of reff
retrievals at this wavelength. Therefore, a reliable retrieval of
reff with reasonable accuracy is not feasible. For
IRTS,1600V the difference is
1.4 mW m-2 sr-1 and about a magnitude larger, indicating that a
retrieval of reff might be reasonable. However, in order to be
consistent between both nadir and sideward viewing retrieval,
reff has been fixed. A value of reff= 30 µm was chosen, a typical value of ice crystals observed by
in situ measurements during ML-CIRRUS . Therefore, the
influence of an invalid assumption of reff on the iterative
retrieval is analyzed. For this purpose the retrieval is tested for a typical
cirrus of τ=0.3 and is run with three different assumptions of
reff of 20, 30 and 40 µm, representing the uncertainty of reff.
These simulations imply that the retrieved τ changes only by ±0.02
between smallest and largest reff, resulting in a relative
error in τ of 6.7 %. The uncertainty in measured
IS,1600N and ImD,1600V
causes a retrieval uncertainty of less than τ ± 0.2. This
justifies the fixed choice of reff in this specific cloud case.
However, the dependence of retrieved τ and the
assumption of reff may vary with α, ice crystal size,
τ and λ used in the retrieval.
ML-CIRRUS case study
The iterative retrieval is applied for a selected leg of the ML-CIRRUS flight
on 26 March 2014. For this day the Terra MODIS image (overpass time 10:40
UTC) indicates clouds, with a west to east gradient in τ ranging from
5.8 to 0.38 (Fig. ) including small cloud
free regions. For large areas, cirrus with τ≤1 is indicated by
MODIS providing provides a well suited test case to compare sideward viewing
and nadir observations even when τ ranges above the SVC level. As
discussed in Sect. 2, for low τ ranging up to 1, ετ of
sideward viewing observations is higher than for nadir observations. An
advantage of using a test case with τ higher than SVC is the
insensitivity of the retrieval uncertainty with respect to the radiance
measurement uncertainty. The reflected I is still sufficiently large and
exceeds the noise level of the nadir looking instruments to make a comparison
between nadir and sideward viewing instruments possible.
In Figure the flight track of HALO is
indicated by the blue line. The cloud retrieval is applied to the HALO flight
segment for the leg between 08:15 and 08:36 UTC (highlighted in red) when
HALO did fly above the cirrus. During this period the aircraft flew
constantly at 12.6 km height from south to north along 14∘ W.
Due to low horizontal advection and hence slow cloud formation it can be
expected that the Terra MODIS image (Fig. )
will actually reflect the cloud cover investigated by HALO. The cirrus
developed along a warm conveyor belt and contained embedded contrails as
indicated by the lidar backscatter profiles at λ= 1064 nm and
λ= 532 nm of WALES (see Fig. ).
Investigated cloud field observed by MODIS-Terra on 26 March 2014.
The flight track of HALO is indicated by the blue line. The flight leg
between 08:15 and 08:36 UTC for which the cirrus retrieval is performed is
indicated by the red line.
Vertical profiles of backscatter ratios at λ= 1064 nm
(upper panel) and λ= 532 nm (lower panel) measured by WALES
between 07:50 and 08:50 UTC. The time period for which τ is retrieved
is marked by the black rectangle.
Time slices of the investigated flight segment on 26 March 2014
(a) and zoom (b) of τ at λ= 532 nm
retrieved from SMART (black line), WALES (gray line), mini-DOAS sideward
viewing (diamonds) and nadir spectrometers (crosses) along the flight track
of ML-CIRRUS flight on 26 March 2014. Periods with the second cloud layer are
marked by the black lines at the top of (a).
The time period for which τ is retrieved is marked by the black frame.
The selected flight segment is characterized by a constant cloud top height
and a slightly increasing cloud bottom height towards northern flight
direction. While the upper most cloud top is relatively homogeneous, there is
significant variability in the layer below which is visible in the
backscatter profile of WALES. The beginning of the black marked area shows
high backscatter ratios of up to 500 indicating high reflectivity of a dense
cirrus. At the end of the selected time period the backscatter decreases. The
lower part of the cirrus shows small-scale variability mainly connected to
sedimentation of ice crystals.
Time series of cirrus optical thickness
Figure a shows a 20 min long flight segment of
retrieved
τ at λ= 532 nm calculated from SMART, mini-DOAS nadir and
sideward viewing spectrometers. WALES measurements are included for comparison.
Along the analyzed cirrus, the retrieved τ ranges between 0.1 and 1.3,
indicating the horizontal variability of the cirrus. The general decrease in
τ towards higher latitudes (increasing time) matches with the cloud
pattern observed by WALES. While SMART and mini-DOAS nadir channels resolve
the cirrus variability observed by WALES, the sideward viewing channel
retrieval does not cover these fluctuations due to the reduced time
resolution of the scanning mode and the large spatial scale
(tens of kilometers) over which sideward viewing
measurements average. At some locations, e.g., 08:21 UTC, τ
retrieved by SMART and mini-DOAS significantly exceed the measurements of
WALES. Most likely
both instruments retrieve larger τ than WALES since ice crystals were falling out of
the cirrus obscured to the lidar measurements. A second segment with higher
retrieved τ
is likely due to an underlying cirrus between 8.5 and 9.5 km altitude that
is also obscured to the detection by WALES. Therefore, a positive systematic offset of the
retrieved τ occurs for SMART and mini-DOAS. These data points are excluded from the following
analysis. Nevertheless, there is a slight chance that a few cloud fragments
of these second cloud layers are still affecting the SMART and mini-DOAS
retrievals.
Both passive sensors have a larger FOV compared to WALES and, therefore, are more likely sensitive to cloud layers located below the cirrus.
Average τ are calculated for the filtered time period (indicated by the
gray box in Fig. )
for each instrument. Due to different sampling intervals, a different
resolution and number of observations are included in the averaging
calculations. The retrieved averages of τ at 532 nm are
0.54 ± 0.2 (SMART), 0.49 ± 0.2 (mini-DOAS nadir
spectrometer), 0.27 ± 0.2 (mini-DOAS sideward viewing spectrometer)
and 0.32 ± 0.02 (WALES). The results indicate a reasonable agreement
of τ retrieved by SMART and the mini-DOAS nadir channel, while lower
τ are inferred from mini-DOAS sideward viewing and WALES measurements.
Taking the WALES measurements as a reference, the measurements of SMART and
mini-DOAS overestimate τ. However, by estimating the uncertainty of the
mini-DOAS and SMART based on RTS, the measurement
error of IS,1180N (14.5 %) by SMART results in
an uncertainty range of retrieved τ of ±0.2, which covers the
values of τ obtained by WALES. The uncertainty range of τ is
determined by running the retrieval twice with a bias of measured
IS,1180N with ±14.5 % uncertainty at
1180 nm wavelength as the upper and lower borders. The resulting upper and
lower retrieved τ represent the retrieval uncertainty. The mean τ
inferred from the mini-DOAS sideward viewing observations is significantly
lower than measured by SMART and mini-DOAS nadir measurements. Differences in
τ range up to ±0.73 between SMART and mini-DOAS sideward viewing
observations. This may result from the different FOV of the sideward viewing
geometry that does not observe the exact same clouds as SMART and nadir
channels did. With the scanning sensors orientated to starboard, the sideward
viewing retrieval corresponds to cirrus 8 km east of the flight track. As
the MODIS satellite image in Fig.
indicates, the cirrus becomes slightly thinner towards the east, which
possibly is due to the lower values of τ. Other potential reasons are
the assumed ice crystal shapes for the RTS and different fields-of-view of
the passive and active remote sensing instruments. On the other hand, the
agreement between mini-DOAS sideward observations and WALES is significantly
better. The maximum difference in τ between mini-DOAS sideward channels
and WALES is ±0.25, while the difference between the mean values is
±0.05 (15.6 %). With WALES and mini-DOAS measuring in different
viewing geometries but showing better agreement, the differences in τ
retrieved by SMART are most likely caused by uncertainties in α. As
discussed in Sect. 2.3, nadir observations are more strongly affected by
α than by sideward observations. This is confirmed by the smaller
differences between WALES and mini-DOAS sideward observations and indicates
the advantage of the sideward viewing retrieval due to a reduced surface
influence and lower retrieval uncertainty.
Figure b displays a zoom of the time series
between 08:20 to 08:24 UTC. During this flight segment, τ inferred by
WALES is characterized by systematic oscillations varying between 0.2 and 1.1
also visible in the backscatter profile of WALES in
Fig. . The lag time between two maxima is approximately
between 20 and 25 s flight time, which corresponds to a horizontal
distance between 4.4 and 5.5 km. This pattern is present in the
measurements of SMART, WALES and the mini-DOAS nadir channels even though
partly obscured in the latter measurements due to its reduced time and space
resolution.
Figure a to d show scatter plots of retrieved
τ for the different instrument combinations. A linear regression through
the origin is performed and displayed in all cases. Data where a second cloud
layer was present below the cirrus (gray points) are excluded. The comparison
between SMART and WALES in Fig. a shows that the
majority of the data is below the 1 : 1 line (gray). The linear regression
results in f(x)=0.6621×x. The regression confirms that SMART
systematically retrieves higher τ compared to WALES.
Compared to SMART, mini-DOAS nadir observations of τ depart less from
WALES (Fig. b. Similar to SMART, the slope of the
linear fit f(x)=0.6943×x indicates that mini-DOAS systematically
overestimates τ compared to WALES. This similarity between SMART and
mini-DOAS is obvious as SMART and mini-DOAS rely on the same radiometric
calibration and retrieval. As indicated in Fig. b
retrieved τ from WALES and the mini-DOAS sideward viewing channels agree
well confirmed by the linear regression in Fig. c
that gives a slope of f(x)=1.0328×x close to unity. The
overestimation of retrieved τ by the mini-DOAS nadir channels compared
to the sideward channels is visible in Fig. d
which results in a linear fit of f(x)=1.642×x.
(a) Comparison of the retrieved cirrus optical thickness
τ by WALES and SMART at λ= 532 nm wavelength.
(b) Comparison of the retrieved cirrus optical thickness τ by
WALES and the mini-DOAS nadir channel at λ= 532 nm wavelength.
Measurements when a second cirrus layer was present are displayed in gray and
are discarded in the regression. (c) Comparison of the retrieved
cirrus optical thickness τ by WALES and mini-DOAS sideward viewing
channels at λ= 532 nm wavelength. No data are discarded.
(d) Comparison of the retrieved cirrus optical thickness τ by
mini-DOAS nadir and sideward viewing channels at λ= 532 nm
wavelength.
Overall the comparison provides evidence that the inferred τ agrees
between the different sensors.
Having nadir and sideward viewing observations at the same time allows us to
select the appropriate measurement geometry depending on the cloud situation,
e.g., τ and α. The sensitivity studies in Sect. 2.4 suggest that
a combination of nadir and sideward viewing measurements allows a retrieval
of τ for a wide range of cirrus clouds depending on the observation
conditions. For thin clouds the sideward viewing geometry would be preferred.
In case the cloud becomes optically too thick, leading to high upward
IS,1180V and a saturation of ετ,
no retrievals of τ are possible. Then, switching to nadir observations
of IS,1180N still enables us to determine the
amount of reflected radiation and to retrieve τ.
Probability distribution of cirrus optical thickness
For further comparison the probability density functions (PDFs) of τ
retrieved by SMART, mini-DOAS nadir spectrometers and WALES were
investigated. A PDF of mini-DOAS sideward viewing spectrometers is not
included because of the limited number of data points, making a statistically
meaningful PDF impossible. The PDFs are shown in Fig. .
Corresponding mean and median values of the distributions are given in
Table . SMART (black solid line) and mini-DOAS (red
solid line), which are based on the same radiometric calibration and
retrieval method, show a comparable PDF, indicating that both instruments
measured the same cloud area. In both cases observed τ range from 0.15
to 1.25 for SMART and mini-DOAS and from 0.15 to 0.7 for WALES (black dashed
line). The PDF maxima for SMART and mini-DOAS are around τ=0.4, and
are slightly more pronounced for mini-DOAS. For SMART and mini-DOAS, the PDFs
are skewed to small τ, with a median of 0.47 for SMART and 0.48 for
mini-DOAS. This is slightly smaller than the mean values of 0.5 for SMART and
0.51 for mini-DOAS. Both PDFs are long-tailed towards large τ, slowly
decreasing to higher values of τ. In contrast, τ measured by WALES
(black dashed line) show a stronger shift to low τ, as the mean value of
τ is significantly lower. The most frequent τ is around 0.2. The
WALES measurements do not show τ larger than 0.7. This results in a
stronger decrease in the WALES PDF to higher τ compared to SMART and
mini-DOAS. The difference may be explained by a different FOV and therefore
measuring of different horizontal parts of the clouds. It is assumed that
SMART and mini-DOAS, e.g., due to a similarly large FOV, average over larger
areas and are influenced by 3-D radiative effects caused by clouds,
atmosphere and surface, which are not considered in the presented 1-D RTS and
the iterative retrieval . In contrast, WALES has a more
narrow FOV resulting from an opening angle of the telescope of 0.08∘.
Because of the smaller FOV of WALES, the spot of the laser at the cloud top
covers a smaller area compared to SMART and mini-DOAS, which have a spatial
resolution in the range of tens of meters, depending on the distance between
aircraft and cloud top. Therefore, WALES resolves finer cloud structures that
may exhibit lower τ (cloud gaps) or larger τ. In case of the most
unfortunate situation, WALES would measure a cloud free region but SMART and
mini-DOAS would receive IN from a much larger area including
clouds with various τ. This better spatial resolution of WALES to SMART
and mini-DOAS may explain the shift of WALES to lower τ, but does not
give reasons for the lower amount of high τ.
Mean and median of the PDFs of cirrus optical thickness τ
derived from WALES, SMART and mini-DOAS.
mean
median
WALES
0.35
0.33
SMART
0.56
0.52
mini-DOAS
0.52
0.47
Differences in the PDF of τ may also result from the measurement
methodologies. While WALES uses a laser with small FOV for active remote
sensing, SMART and mini-DOAS are passive remote instruments relaying on
scattered sunlight. Therefore, SMART and mini-DOAS are influenced by the RTS
of the whole atmosphere, while WALES is only sensitive to scattering within
its narrow LOS. Additionally, the different wavelengths of the measurements
may introduce biases in the retrieved τ due to different penetration
depth of the reflected radiation into the cloud .
Therefore, the wavelength selection defines the layer in the cloud which is
probed. While WALES uses backscatter measurements at λ= 532 nm
and λ= 1064 nm the measurements of IS,1180 by
SMART and mini-DOAS are performed at λ= 1180 nm. Although the
retrieval accounts for the wavelength dependence of scattering, absorption
and refraction on ice crystals by scaling the
retrieved τ at λ= 1180 nm to λ= 532 nm to make
it comparable between the different instruments.
Referring to the sensitivity studies from Sect. 2 the influence of α
and the ice crystal shape effects on the upward I measured in nadir
geometry is larger compared to the sideward viewing measurements. While nadir
observations, especially of optical thin clouds, are strongly influenced by
α, sideward viewing observations are less effected. This is
demonstrated in this case study where the sea surface albedo may vary due to
different surface wind speeds and indicates the advantage of
sideward viewing measurements. An other possible reason for the differences
in the PDF and the mean values between mini-DOAS nadir and sideward
retrievals of τ are the varying angular dependencies of measured I for
different ice crystal shapes. For the RTS in the retrieval an assumption for
the ice crystal shape has to be made which slightly influences the result for
the nadir retrieval. This is more pronounced for the retrieval using the
sideward channels of the mini-DOAS which is presented in the sensitivity
study in Sect. 2.2. The WALES measurements are less effect by different ice
crystal shapes but more on the ice crystal size assumption which is a general
difference between the active and remote sensing instruments presented here.
PDFs of cirrus
optical thickness τ at λ= 532 nm retrieved from SMART,
mini-DOAS and WALES measurements. The bin size is 0.05 units of τ.
Conclusions
The potential of airborne spectral radiance measurements in the sideward
viewing direction for cirrus remote sensing is investigated. For this purpose
radiative transfer simulations (RTS) are performed and airborne measurements
of the Spectral Modular Airborne Radiation measurement sysTem (SMART) and the
Differential Optical Absorption Spectrometer (mini-DOAS) are compared. A
sensitivity study based on RTS showed that sideward viewing measurements are
generally more suited for detecting and investigating optically thin cirrus
than observations in the nadir direction. Using sideward viewing
observations, the sensitivity ετ of measured radiance
IV is larger than for nadir measurements up to a factor of 10,
depending on the selected observation geometry and cloud properties. For
cirrus optical thickness τ≤1 and all simulated sideward viewing
geometries ετ is larger compared to nadir observations.
This results in a higher retrieval accuracy due to a reduced influence of
measurement uncertainties. The RTS indicate that large observation angles
θV (close to the horizon) and small relative solar azimuth
angles ϕ (observations in the direction of the Sun) result in the
highest ετ.
For retrievals of τ using sideward viewing measurements, the wavelength
selection is crucial. Simulations indicate that wavelengths larger than
λ= 900 nm are best suited. Reflected IV of smaller
wavelengths is significantly contaminated by scattering and absorption due to
the reducing interference from Rayleigh scattering. Furthermore, the sideward
viewing orientation reduces the influence of the surface albedo α on
reflected IV. As a result, a precise assumption of α in
the retrieval algorithm is less crucial. This substantially improves the
uncertainties of passive solar remote sensing especially in locations of
highly variable α, where an exact assumption of α is
impossible.
Contrarily, for sideward observations, a reasonably good assumption of the
ice crystal shape used in the RTS is important. The RTS showed that in
sideward viewing geometry the shape effects on reflected IV are
more pronounced than for nadir measurements. An incorrect assumption would
bias the retrieval of τ significantly. On the other hand, the
sensitivity for different ice crystal shapes may offer the possibility of
retrieving shape information when measuring at different viewing angles.
Nevertheless, smoothing of horizontal variability of optical thickness fields
by sideward viewing observations has to be taken into account.
Using the SMART, mini-DOAS nadir and sideward measurements in conjunction
with an iterative retrieval, τ is derived for a case study of ML-CIRRUS.
The inferred τ from SMART, mini-DOAS and the additional lidar
measurement by the Water Vapour Lidar Experiment in Space (WALES) show a
reasonable good agreement in τ for the nadir channels, with absolute
differences of ±0.22 (66.6 %) between SMART and WALES and ±0.17
(52.3 %) between mini-DOAS and WALES observations, respectively. The
retrieval using mini-DOAS sideward channels is also successful demonstrated
for a reduced set of observations limited to θV between 85
and 90∘. Differences in τ range up to ±0.73 between SMART
and mini-DOAS sideward viewing observations and are partly caused by the
different viewing geometries. First, the sideward telescopes view in the
starboard direction, probing the cirrus cloud top at approximately 8000 m
aside the flight track. Second, the nadir observations may suffer from
uncertainties in α while the sideward observations are less effected
by changes in α. Even for sea surfaces as presented here, α may
change due to different wind speeds. Other potential reasons are the assumed
ice crystal shapes in the RTS and different fields-of-view of the passive and
active remote sensing instruments. This conclusion is apparent from different
probability distributions. While SMART and mini-DOAS show a median around
τ=0.4, the median for WALES is shifted to lower τ around 0.2,
indicating that WALES observed small τ more frequently. The difference
in mean values of τ between mini-DOAS sideward channels and WALES is
smaller with ±0.05 (15.6 %). This shows the advantage of the sideward
viewing retrieval due to a reduced surface influence and lower retrieval
uncertainty, because of high ετ compared to the nadir
measurements. For future dedicated cloud observations it is recommended to
adjust θV to the most sensitive direction between 60 and
90∘ to reduce the uncertainty in the sideward viewing retrieval.
Additional sideward viewing scans in homogeneous cloud conditions might be
used to estimate the cirrus ice crystal shape and minimize the retrieval
uncertainties. The case study shows that cirrus retrievals using airborne
sideward viewing observations with mini-DOAS are possible and can increase
the potential of remote sensing on HALO significantly. Therefore, we suggest
sideward viewing measurements for passive remote sensing of optically thin
cirrus clouds.