Available via license: CC BY 3.0
Content may be subject to copyright.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Atmos. Meas. Tech. Discuss., 4, 5631–5688, 2011
www.atmos-meas-tech-discuss.net/4/5631/2011/
doi:10.5194/amtd-4-5631-2011
© Author(s) 2011. CC Attribution 3.0 License.
Atmospheric
Measurement
Techniques
Discussions
This discussion paper is/has been under review for the journal Atmospheric Measurement
Techniques (AMT). Please refer to the corresponding final paper in AMT if available.
Aerosol classification using airborne High
Spectral Resolution Lidar measurements
– methodology and examples
S. P. Burton
1
, R. A. Ferrare
1
, C. A. Hostetler
1
, J. W. Hair
1
, R. R. Rogers
1
,
M. D. Obland
1
, C. F. Butler
2
, A. L. Cook
1
, D. B. Harper
1
, and K. D. Froyd
3
1
NASA Langley Research Center, Hampton VA, 23681, USA
2
Science Systems and Applications, Inc., Hampton VA, 23666, USA
3
Chemical Science Division, ESRL, NOAA, Boulder, CO, USA
Received: 26 August 2011 – Accepted: 31 August 2011 – Published: 7 September 2011
Correspondence to: S. P. Burton (sharon.p.burton@nasa.gov)
Published by Copernicus Publications on behalf of the European Geosciences Union.
5631
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Abstract
The NASA Langley Research Center (LaRC) airborne High Spectral Resolution Li-
dar (HSRL) on the NASA B200 aircraft has acquired extensive datasets of aerosol
extinction (532 nm), aerosol optical thickness (AOT) (532 nm), backscatter (532 and
1064 nm), and depolarization (532 and 1064 nm) profiles dur ing 18 field missions that5
have been conducted over North America since 2006. The lidar measurements of
aerosol intensive parameters (lidar ratio, depolarization, backscatter color ratio, and
spectral depolarization ratio) are shown to vary with location and aerosol type. A
methodology based on observations of known aerosol types is used to qualitatively
classify the extensive set of HSRL aerosol measurements into eight separate types.10
Several examples are presented showing how the aerosol intensive parameters vary
with aerosol type and how these aerosols are classified according to this new method-
ology. The HSRL-based classification reveals vertical variability of aerosol types dur-
ing the NASA ARCTAS field experiment conducted over Alaska and northwest Canada
during 2008. In two examples derived from flights conducted during ARCTAS, the15
HSRL classification of biomass burning smoke is shown to be consistent with aerosol
types derived from coincident airborne in situ measurements of particle size and com-
position. The HSRL retrievals of AOT and inferences of aerosol types are used to
apportion AOT to aerosol type; results of this analysis are shown for several experi-
ments.20
1 Introduction
We introduce an aerosol classification scheme for airborne High Spectral Resolution
Lidar (HSRL) measurements. The ability to accurately characterize and discriminate
aerosol type can improve both measurement retrievals and modeling, on both a re-
gional and global scale. Since 2006, the NASA Langley High Spectral Resolution Lidar25
has routinely participated in chemistry and radiation-focused field missions throughout
5632
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
North America, where its high accuracy, high resolution, vertically resolved measure-
ments of aerosol provide vertical context for ground-based, in situ, and satellite ob-
servations of aerosols and clouds (e.g. Molina et al., 2010; Warneke et al., 2010).
The HSRL also routinely provides validation for the Cloud-Aerosol Lidar with Orthog-
onal Polarization (CALIOP) lidar instrument aboard the Cloud-Aerosol Lidar and In-5
frared Pathfinder Satellite Obser vations (CALIPSO) satellite (Winker et al., 2009). The
aerosol classification introduced here serves to enhance the input provided by HSRL
in both of these roles. Furthermore, the HSRL serves as a test-bed for advanced satel-
lite lidar instruments, and the advanced retrievals required for those measurements
may benefit from aerosol classification like what is described here. For example, ad-10
vanced lidar retrievals of microphysical properties from extinction and backscattering
coefficients and depolarization at multiple wavelengths (M
¨
uller et al., 1999; Veselovskii
et al., 2002), such as might be part of the future Aerosol Clouds and Ecosystems
(ACE) Decadal Sur vey mission (National Research Council, 2007), would benefit from
aerosol type information as a constraint to improve the retrieval efficiency.15
More than two decades ago, Sasano and Browell (1989) characterized different
aerosol types using lidar, specifically a three-wavelength elastic backscatter lidar, with
a retrieval method that optimizes the match between the shape of the backscatter pro-
files at the shorter wavelengths and that at 1064 nm, which is relatively less sensitive
to lidar ratio. Aerosols of five types were identified and classified. Later, the Lidar In-20
space Technology Experiment (LITE) (Winker et al., 1996), the precursor to CALIPSO,
provided the first opportunities to observe vertical distributions of aerosol globally. Kent
et al. (1998) first described the long-range transport of biomass burning aerosols and
characterized the optical properties using a similar lidar retrieval for LITE. Ground- and
ship-based measurements by micropulse lidar (Spinhirne, 1993) provided case studies25
of biomass burning (Campbell et al., 2003), maritime and polluted maritime (Welton et
al., 2002), and dust aerosols (Welton et al., 2000; Powell et al., 2000).
These lidars as well as the CALIOP instrument on the CALIPSO satellite are elastic
backscatter lidars, for which it is not possible to independently measure the aerosol
5633
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
extinction and backscatter coefficients. To retrieve both, it is common to assume that
the ratio of the two, the lidar ratio, is vertically homogeneous throughout the entire
column or layer in question, and the lidar ratio is either prescribed or inferred using
additional measurements as constraints. The need for more accurate lidar ratios to
constrain this type of retrieval continues to provide motivation for aerosol classification5
and characterization studies.
Examples include in situ nephelometer measurements of backscattering plus inte-
grated scattering and absorption measurements, which were used to calculate lidar
ratios for various aerosol types (e.g. Anderson et al., 2000). Cattrall et al. (2005)
moved beyond case studies using Aerosol Robotic Network (AERONET) sun photome-10
ter data sets to estimate lidar intensive parameters for specified aerosol types for use
with spaceborne lidar retrievals. They followed Dubovik et al. (2002) who identified
seasons and locations dominated by four key aerosol types and characterized the in-
dex of refraction and particle size distributions for those types using quality-controlled
AERONET sun photometer data. The types identified by Dubovik et al. (2002) were15
urban-industrial from fossil fuels, biomass burning from forest and grassland fires,
wind-blown desert dust, and marine aerosol. Cattrall et al. (2005) expanded the set
of aerosol types by adding a Southeast Asian type, distinct from urban-industrial pollu-
tion, exhibiting a greater number of large particles relative to fine particles. They also
made this method of aerosol classification useful for lidar retrievals by calculating lidar20
parameters for these five types from retrievals of sky radiance and solar transmittance,
and compared results to an extensive set of Raman lidar measurement case studies
of particular types (Cattrall et al., 2005 and references therein).
The above studies characterize the optical properties of aerosols from samples of
aerosol types that are identified by context. In contrast, an example of using lidar mea-25
surements to automatically classify aerosol types is given by Shimizu et al. (2004),
who used the lidar depolarization measurements to differentiate spherical from non-
spherical aerosol. A more sophisticated automated classification scheme is presented
by Omar et al. (2005) who did a k-means cluster analysis on 26 aerosol intensive
5634
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
variables derived from a comprehensive AERONET dataset, to produce and charac-
terize a set of six aerosol types.
The sun-photometer measurements used in many of these studies pertain to the
entire integrated vertical column. Column-only measurements can cause biased esti-
mates of lidar properties in situations with inhomogeneous aerosols. In contrast, M
¨
uller5
et al. (2007a) use measurements of aerosol extinction profiles provided by multiple
ground-based Raman lidar systems to characterize vertically resolved aerosol optical
properties. The advantage of lidar is the ability to provide vertically resolved measure-
ments, and Raman lidar (Ansmann et al., 1990) and High Spectral Resolution Lidar
(Shipley et al., 1983; Grund and Eloranta, 1991; She et al., 1992) have the addi-10
tional key advantage over the backscatter lidars described above, in that they measure
aerosol extinction and backscatter coefficients independently, without using models or
assumptions about aerosol type. Since extinction coefficients are measured, they also
provide aerosol optical thickness (AOT) measurements comparable to passive satellite-
based (e.g. the Moderate-Resolution Imaging Spectroradiometer (MODIS) (Remer et15
al., 2005) and the Multiangle Imaging Spectroradiometer (MISR) (Kahn et al., 2005))
and ground-based (e.g. Aerosol Robotic Network (AERONET) (Holben et al., 1998))
observations (Burton et al., 2010).
Along with directly measured backscatter and extinction coefficients and AOT, the
NASA Langley airborne High Spectral Resolution Lidar provides vertically resolved20
information about aerosol composition in the form of four aerosol intensive variables
that depend only on aerosol type and not on concentration. This consistent set of four
aerosol intensive parameters: the lidar ratio, aerosol depolarization at two wavelengths,
and the ratio of aerosol backscatter at two wavelengths, provides qualitative informa-
tion about the aerosol physical properties. Two channels of depolarization have not25
been used before for aerosol classification, and we find that this may help to sepa-
rate the optically similar pollution and smoke aerosols. In this report we describe how
these measurements have been used to infer aerosol type. The HSRL has flown on
18 field missions to date, and this has provided an extensive dataset of well-calibrated
5635
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
observations of aerosol types from diverse regions throughout North America. These
observations are not limited to either day or night, unlike sunphotometer measure-
ments. Since the intensive variables do not depend on the amount of aerosol loading,
there is a much smaller effective limitation on the loading that can be used for classifi-
cation than what was required for Dubovik et al. (2002) and Cattrall et al. (2005). The5
airborne HSRL is able to make aerosol extinction measurements down to within 300 m
of the ground, thereby normally sampling a significant portion of the boundary layer
where important aerosol types are located.
This is the first of two companion papers. Here we will describe how measurements
acquired by the NASA Langley Research Center airborne High Spectral Resolution10
Lidar have been used to infer aerosol type and apportion AOT to aerosol type. In
Sect. 2 of this paper, the NASA Langley airborne HSRL system is discussed, followed
in Sect. 3 by a description of the HSRL measurements and how these are used to mea-
sure aerosol intensive parameters. In Sect. 4, the methodology for using these mea-
surements to classify aerosol types is described followed by a discussion of the par-15
ticular aerosol types that are identified from the HSRL data. Examples of the aerosol
classification are presented, followed by a discussion of these results in Sect. 5. Af-
ter discussing the classification methodology and presenting examples of this classi-
fication in this paper, Ferrare et al. (2011) in the companion paper use the results of
this HSRL-based aerosol classification to evaluate aerosol classifications derived from20
CALIPSO measurements and simulated by the GOCART aerosol model (Chin et al.,
2002).
2 NASA Langley airborne High Spectral Resolution Lidar (HSRL)
The LaRC airborne HSRL (Hair et al., 2008) uses the HSRL technique to independently
retrieve aerosol and tenuous cloud extinction and backscatter without a priori assump-25
tions on aerosol type or extinction-to-backscatter ratio. The HSRL technique (Shipley
et al., 1983; Grund and Eloranta, 1991; She et al., 1992) measures aerosol extinction
and backscatter independently, by using a narrow-band iodine vapor filter to separate
5636
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
the broadened spectra of Cabannes scatter ing by molecules from the more narrowly
peaked Mie scattering by aerosols (She, 2001). The observed molecular backscatter-
ing component is attenuated by extinction. Therefore, by comparison with the molecu-
lar backscattering from an atmospheric density profile obtained from the NASA Global
Modeling and Assimilation Office (GMAO) or another source, the aerosol extinction co-5
efficient profile is obtained. The LaRC HSRL employs the HSRL technique at 532 nm
and the standard backscatter technique at 1064 nm. The instrument also measures
depolarization at both wavelengths. The return signal is split into components par-
allel and perpendicular to the polarization of the outgoing beam. The depolarization
ratio here is defined as the ratio of the perpendicular to the parallel component. The10
HSRL instrument is self-calibrating at 532 nm for measurements of aerosol and cloud
backscatter and extinction, in contrast to standard backscatter lidars that are empiri-
cally calibrated by assuming that the aerosol contribution to backscatter is negligible
or known at some altitude. It is self-calibrating at both 532 and 1064 nm for measure-
ments of depolarization. The calibration of the 1064 nm aerosol and cloud backscat-15
ter measurement takes advantage of the inter nally calibrated HSRL measurement at
532 nm. A detailed description of this HSRL system and calibration and data retrieval
techniques is provided by Hair et al. (2008). The vertical resolution of the backscatter
coefficients and depolarization measurements is 60 m, and the horizontal averaging is
10 s (about 1 km) (Rogers et al., 2009). The aerosol extinction profiles have a vertical20
resolution of 300 m, and the horizontal averaging is 60 s (about 6 km) (Rogers et al.,
2009). The vertical and horizontal resolutions can be varied to suit varying measure-
ment needs. The extinction and lidar ratio profiles extend from approximately 300 m
above the surface, as determined by a digital elevation dataset (GLOBE Task Team et
al., 1999), to approximately 2500 m below the aircraft. The 300 m limit at the low end of25
the profile is to avoid ground contamination. The 2500 m near-range limit is to ensure
full overlap between the outgoing laser and the receiver field of view. The backscat-
ter coefficient and depolarization profiles extend further in each direction, from 500 m
below the aircraft to 60 m (2 range bins) above the ground.
5637
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
In summary, the HSRL provides vertically resolved measurements of the following
extensive and intensive aerosol parameters:
– Extensive parameters: backscatter coefficient at 532 and 1064 nm with horizontal
resolution of approximately 1 km and vertical resolution of approximately 60 m; ex-
tinction coefficient at 532 nm with horizontal resolution of approximately 6 km5
and vertical resolution of approximately 300 m; and total column optical depth
at 532 nm derived by integrating the profile of extinction.
– Intensive parameters: S
a
(aerosol lidar ratio) at 532 nm with resolutions matching
those of the extinction coefficient given above, aerosol depolarization at 532 nm
and 1064 nm with horizontal resolution of approximately 1 km and vertical reso-10
lution of approximately 60 m; and aerosol wavelength dependence, which is the
˚
Angstr
¨
om exponent for aerosol backscatter and directly related to the backscatter
color ratio, with resolution matching the backscatter coefficients above.
Rogers et al. (2009) compared the HSRL aerosol extinction measurements with
aerosol extinction derived from simultaneous measurements from the NASA Ames Air-15
borne Sun Photometer (AATS-14) (Redemann et al., 2009) and in situ scattering and
absorption measurements from the Hawaii Group for Environmental Aerosol Research
(HiGEAR) in situ instruments (McNaughton et al., 2009) and found bias differences
between HSRL and these instruments to be less than 3 % (0.01 km
−1
) at 532 nm; root-
mean-square (rms) differences at 532 nm were less than 50 % (0.015 km
−1
). These20
differences are well within the ranges observed by current state-of-the-art instrumen-
tation (Schmid et al., 2006).
3 HSRL measurements
The HSRL data analyzed in this article were acquired between March 2006 and
September 2010. During that time, the airborne LaRC HSRL was deployed on the25
5638
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
NASA Langley B200 King Air aircraft and acquired over 1000 h of data on over 330
science flights during eighteen field campaigns. HSRL continues to participate in field
campaigns, but data beyond 2010 are not included in the analysis presented here. The
campaigns include many process-oriented field projects for NASA, the Department
of Energy (DOE), and the Environmental Protection Agency (EPA), as well as field5
projects devoted to CALIPSO validation. These totals include 101 successful validation
flights for the CALIPSO program. Figure 1 shows the locations of these missions and
Table 1 lists these field missions and the science flight hours associated with them. The
diverse locations of these missions has enabled the HSRL to acquire measurements
of several different aerosol types including smoke during ARCTAS (Jacob et al., 2010;10
Warneke et al., 2010; Knobelspiesse et al., 2011), urban and dust aerosols during
MILAGRO (Molina et al., 2010), and Saharan dust during TexAQS/GoMACCS (Liu et
al., 2008; Parrish et al., 2009; Burton et al., 2010) The HSRL acquired data below the
aircraft, which normally flew at 9 km (MSL); typical flight duration was 3.5–4 h.
Figure 2 shows an example of the suite of HSRL measurements acquired when the15
King Air flew over Mexico City between 17:38 and 17:52 UT on 13 March 2006. These
measurements exhibit variations in aerosol type over Mexico City. The data shown
in Fig. 2 were collected over a distance of about 115 km. The aerosol backscatter
and extinction coefficients are shown along with the four aerosol intensive parame-
ters: aerosol depolarization, extinction-to-backscatter ratio, wavelength ratio of aerosol20
depolarization, and backscatter wavelength dependence. The vertical and horizontal
resolution and lower and upper altitude limits are as described in Sect. 2. These mea-
surements show the variability of the various types of aerosols that were measured
over the region.
The HSRL measurements of aerosol intensive parameters provide information about25
the particle physical properties. For example, backscatter spectral ratios typically are
inversely related to aerosol par ticle sizes (Sasano and Browell, 1989; Sugimoto et al.,
2002). Another intensive parameter, the depolarization ratio, is recognized as a dis-
criminator of dust (Shimizu et al., 2004; Omar et al., 2009). High values of 30 % to
5639
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
40 % depolarization for aerosol are indicative of nearly pure dust (Murayama et al.,
2003; Sugimoto and Lee, 2006; Liu et al., 2008), with smaller values that are still ele-
vated above about 8–10 % usually attributed to a mixture of dust with spherical particles
(Murayama et al., 2003; Sugimoto and Lee, 2006). High depolarization can also indi-
cate ice particles, as in cirrus clouds (e.g. Shimizu et al., 2004). Crystallized sea salt5
(Murayama et al., 1999; Sakai et al., 2010) and aged biomass burning and volcanic
aerosols (Sassen, 2008) can also exhibit some depolarization, but with much smaller
values. The degree of depolarization also varies with relative humidity, since hygro-
scopic swelling increases the sphericity of particles and decreases their depolarization
(Murayama et al., 1996; Sassen, 2000). The spectral dependence of the depolarization10
ratio is dependent on particle size in the case of ice clouds (Somekawa et al., 2008)
and on mixing ratio and spherical and non-spherical particle sizes in mixtures of dust
and non-spherical particles (Sugimoto and Lee, 2006; Somekawa et al., 2008). Finally,
the aerosol extinction-to-backscatter ratio, or lidar ratio, varies with aerosol size, shape,
and composition; tropospheric aerosols typically have low values of approximately 2015
to 50 sr for weakly-absorbing coarse mode particles (i.e. sea salt, dust) and higher val-
ues for small and/or highly absorbing accumulation mode particles (Ackermann, 1998;
Cattrall et al., 2005; M
¨
uller et al., 2007a and references therein).
For the example shown in Fig. 2, over the western part of the city, higher values of
backscatter wavelength dependence and the lidar ratio (S
a
) and lower values of depo-20
larization suggest smaller, more spherically shaped particles (e.g. sulfate drops) more
typically associated with urban/industrial pollution. Lower S
a
and higher depolarization
values over the eastern part of the city suggest higher concentrations of dust. These
measurements are consistent with FLEXPART model simulations which also indicate
urban emissions dominating in the western part of the city, with a mixture of biomass25
burning, urban emissions and dust in the east (de Foy et al., 2011). These HSRL
measurements also clearly show the vertical and horizontal variability of aerosol inten-
sive properties (e.g. depolarization and wavelength dependence) associated with thin
elevated aerosol layers over the western section of Mexico City.
5640
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Figure 3 shows how the aerosol intensive properties measured by the HSRL varied
during the various field experiments from March 2006 through 2010. Note for example
how the aerosols observed over Mexico during the MILAGRO campaign had some-
what different characteristics than those observed over Houston and the eastern US
during GOMACCS and the CALIPSO validation missions respectively. The aerosols5
observed during MILAGRO typically had higher depolar ization ratios, smaller lidar ra-
tios, and smaller backscatter color ratios than the aerosols observed over the eastern
US. This indicates that the aerosols observed during MILAGRO were somewhat larger
and more nonspherical and most likely had higher concentrations of dust; conversely,
the aerosols observed over the eastern and southeastern US typically were smaller,10
more spherical and were urban in nature. The first Caribbean campaign, which made
frequent measurements of maritime aerosol, exhibits much smaller lidar ratios and also
a smaller wavelength ratio of depolarization. During ARCTAS, the lidar ratio was large,
typical of the smoke aerosol frequently seen during that campaign, and the backscat-
ter color ratio was also high, indicating small particles. During the second Caribbean15
campaign, high values of aerosol depolarization reflect the large amount of Saharan
dust observed. These results indicate that the aerosol intensive variables measured by
HSRL vary with location and suggest that this variability can be used as an indicator of
aerosol type. In the next section, we describe our methodology for using these HSRL
measurements to infer aerosol types.20
4 Aerosol classification
4.1 Methodology
As stated above, the four aerosol intensive variables used in the aerosol classification
are the extinction-to-backscatter ratio, S
a
, at 532 nm; the backscatter color ratio, which
is the ratio of the backscattering coefficient at 532 nm to 1064 nm and is related to the25
backscatter
˚
Angstr
¨
om exponent; the aersosol depolarization at 532 nm (actually the
5641
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
natural log of this quantity, since it is more normally distributed); and the spectral depo-
larization ratio, which is the ratio of the particulate depolarization measured in the two
channels, 1064 nm/532 nm. The extensive aerosol parameters: aerosol backscatter-
ing coefficient, scattering ratio, extinction, and optical depth, are not used since these
parameters can vary with aerosol amount as well as type.5
Measurements are prepared for classification by clear ing clouds using a convolution
of the measured signal at 532 nm with a Haar wavelet to enhance edges (Davis et al.,
2000), combined with an algorithm to set a flight-by-flight threshold for separating the
generally sharper cloud edges from the less pronounced aerosol feature boundaries
in each lidar profile. Optional filtering criteria are applied to the HSRL aerosol mea-10
surements at this stage. The classification algorithm itself is not sensitive to outliers
and noise and including or eliminating them has no effect on the classification of the
remainder of observations. The criteria listed in Table 2 were applied for the creation
of the example figures shown in this paper. Generally the points that fail these criteria
have only small contributions to the column optical depth.15
The HSRL aerosol classification uses eight classes, which start with labeled samples
of known aerosol types. Section 4.2 describes the eight classes and how the samples
were chosen. Thirty samples of a few hundred to a few thousand data points each, in
total comprising about 0.30 % of the data, are labeled using a priori knowledge. These
samples are combined to estimate multi-dimensional normal distributions defined by20
the 4-by-4 variance-covariance matrix of the four aerosol intensive variables. Distribu-
tions are generated from the samples for each of the eight classes, after weighting so
that each sample counts equally within a class. Generalized distances are then calcu-
lated for each measurement to each of the class distributions, using the Mahalanobis
distance metric (Mahalanobis, 1936).25
The Mahalanobis distance is appropriate for quantifying the distance between a point
and a distribution, and is therefore a better metric for this application than the Eu-
clidean distance between two points. It assumes the aerosol classes are represented
as multi-normal distributions. When the Euclidean distance is used for classification,
5642
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
as is frequently the case in k-means cluster ing, the measurement points tend to be
forced into roughly spherical clusters unless the classes are widely spaced out, since
each class is identified by only by a single point. The Mahalanobis distance metric, on
the other hand, incorporates more information about the relative shapes and sizes of
the classes, including potentially different widths or variances in each dimension and5
covariance between the variables. The assumption of multi-normal distributions is a
much less limiting assumption, and is consistent with the presentation of results by
Cattrall et al. (2005).
After the class distribution models are calculated, the Mahalanobis distance is used
to classify aerosol measurements from all HSRL aerosol observations. Points with a10
Mahalanobis distance greater than a certain threshold from all the classes are con-
sidered outliers and are not classified. This threshold (Mahalanobis distance = 4.3)
corresponds to the 0.1 % cumulative probability contour of the class distributions, de-
rived by assuming that the Mahalanobis distances belong to a chi-square distribution.
That is, 0.1 % of a random sampling of theoretical points belonging to a class would lie15
at a distance beyond the threshold and would be missed. For points where the Maha-
lanobis distance to one or more classes is within this threshold, the class identification
is inferred from the smallest distance.
Besides providing the most likely class identification for each measurement, the Ma-
halanobis metric also gives an estimate of the probability for each class. The eight20
probabilities are normalized to give an estimate of the relative probability for each class.
We require the normalized probability to be at least 60 % for an observation to be as-
signed to a given class. In cases where none of the eight probabilities exceeds 60 %
because the point is nearly equidistant from two or more of the nearest classes, the
normalized probabilities are recorded, but no class identification is made.25
The classification method described here differs from unsupervised classification
schemes like k-means (MacQueen, 1967) or expectation maximization (EM) cluster-
ing (Dempster et al., 1977) pr imarily in the use of labeled samples. This allows us
to incorporate additional knowledge about aerosol type that may be available only in
5643
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
specific cases. This method also has the benefit that new data can be easily classified
without causing existing classifications to be greatly altered. The classification is not
sensitive to how many outliers, noisy points, or otherwise unreliable measurements are
allowed to be included. Classes can even be added or removed with minimal disruption
to the other classes. This is in sharp contrast to unsupervised clustering methods in5
which any change has global consequences, since those algorithms depend on itera-
tive global minimization of the distance metric.
4.2 Aerosol types
The HSRL aerosol classification has eight types and begins with thirty samples of
labeled data, between two and six samples for each type. Figure 4 shows the charac-10
teristics of the samples in terms of the four intensive variables used for classification.
Also shown are projections of the two-sigma covariances of the model distributions.
The eight particulate classes were chosen to provide a useful separation of the
observations into distinct types. These classes are: ice, pure dust, dusty mix, mar-
itime, polluted maritime, urban, fresh smoke, and smoke. The choice of the number of15
classes is always a somewhat subjective decision; too few classes will cause important
distinctions between different observations to be lost, while too many classes can make
it easy to overlook important similarities and prove difficult to interpret. Our choice of
eight classes was based on extensive inspection of the HSRL data. The types used in
this study reflect the heritage of previous work on classification of lidar measurements20
(Cattrall et al., 2005; Omar et al., 2005), with some additions that are described herein.
The samples are labeled using knowledge about the atmospher ic conditions dur-
ing specific flights. For example, clean air in the Caribbean is labeled “Maritime,”
and samples in regions with elevated AOT near major urban centers such as Mex-
ico City and Washington D.C. are labeled “Urban.” In the case of smoke samples, the25
plume was observed visually from the B200 or was measured by coincident airborne
in situ measurements (Warneke et al., 2010) and/or MODIS images (Saha et al., 2010,
see Fig. S7a). In other cases, coincident observations allow particular samples to be
5644
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
identified. For example, the CALIPSO lidar instrument tracked a plume of Saharan dust
as it was advected from Africa to the Gulf of Mexico, as described by Liu et al. (2008).
CALIPSO and HSRL observed it simultaneously near Houston, Texas ten days later.
Some of the classes, such as pollution, maritime, smoke, and dust, correspond with
aerosol types from previous studies (e.g. Dubovik et al., 2002; Omar et al., 2005) .5
Other categories, pure dust, fresh smoke, and polluted maritime, were added based
on HSRL measurements, simultaneous observations from the aircraft, and preliminary
findings using the fully unsupervised cluster analysis schemes, k-means and Expec-
tation Maximization, and will be discussed below. Particular samples were chosen to
attempt to provide good coverage of the apparent range of lidar intensive observables10
for each type, while simultaneously holding back some known samples to judge the
success of the classification. Removing or replacing individual samples can affect a
small percentage of measurements on the apparent boundaries between classes but
does not affect the results overall. We do not have enough labeled samples to perform
a statistical assessment of the effect of sample selection; however, the experiment de-15
scribed later in Sect. 4.3 gives a quantitative idea of the amount of error due to points
being near the boundaries.
4.2.1 Ice particles
Elevated aerosol depolarization values are usually an indication of dust and/or ice.
During the ARCTAS campaign, many cases were observed of optically thin ice or ice20
crystal haze (Saha et al., 2010). These common ice layers have been previously re-
ported extending above 6.5 km (Greenaway, 1950) and sometimes are precipitating
crystals based on evidence of fall streaks in the HSRL lidar data. Hoff (1988) ob-
serves similar ice crystal precipitation events that are often obvious in ground-based
lidar records although no visible clouds are evident. Curry et al. (1990) point out that25
although Hoff uses the apt term “ground-based cirrus”, these cloud-free ice crystal
hazes are neglected in contemporary cloud classification conventions and in radiation
5645
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
transfer and climate calculations. More recently, model parameterizations are available
for ice fog (Girard and Blanchet, 2001). However, ice crystal hazes observed by HSRL
during ARCTAS were not cleared as clouds in AERONET observations and frequently
contribute significantly to AERONET aerosol optical depth. Nearby Total Sky Imager
camera images also indicated clear conditions, but Millimeter Cloud Radar consistently5
indicated cloudy conditions. The likely explanation for these observations is the pres-
ence of relatively large particles, but in low concentrations. Large (>1)
˚
Angstr
¨
om expo-
nents observed by AERONET suggest that the particles are smaller than typical cirrus
particles. These ice crystal airmasses are not cleared from the HSRL measurements
as clouds either, so we need to be able to separate them from aerosol particles.10
Ice observed by HSRL during the ARCTAS campaign can have particle depolariza-
tion of up to 60 %, which is greater than that associated with pure dust (Murayama et
al., 2003; Sugimoto and Lee, 2006; Liu et al., 2008). Mishchenko and Sassen (1998)
also indicate depolarization values up to 50–70 % are possible for ice crystals with ef-
fective radius on the order of a micrometer. Since the depolarization of ice crystals15
is highly variable (e.g. Sassen and Hsueh, 1998), and can be comparable to that of
dust, it is difficult to use particle depolarization alone to separate ice and dust (Sakai
et al., 2003). Based on our HSRL measurements and from previous HSRL (Eloranta,
2005) and Raman lidar (Whiteman et al., 1992) observations, the lidar ratios for ice are
typically lower than for dust (Sakai et al., 2003).20
Figure 5 shows an example of HSRL measurements acquired over Alaska dur ing
the ARCTAS mission. Aerosol depolarization values were elevated at altitudes above
3 km for much of this period. The very high (0.5–0.6) particulate depolarization values
and low (∼20 sr) lidar ratio values around 23.65 UT and above 5 km are associated
with ice crystals. At other times and altitudes it is difficult to use depolarization alone to25
discriminate ice and dust; however, the higher values of lidar ratio (40–50 sr) strongly
suggest that the particles were much more likely to be dust than ice. The spectral ratio
of depolarization also appears to help distinguish between ice and dust, as illustrated
by the labeled samples in Fig. 4 and by the contrast between low ratio for ice and
5646
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
higher ratio for dust in this example. This case also provides an example indicating the
potential for elevated dust layers to act as ice nuclei (Sassen, 2002).
4.2.2 Dust and dusty mix
Particle depolarization ratios between 30 % and 40 % are characteristic of “pure dust”
from Asia (Shimizu et al., 2004; Sugimoto and Lee, 2006) or the Sahara desert (Liu5
et al., 2008). Figure 6 shows an example of Saharan dust layer observed during an
HSRL flight on 18 August 2010 between Bermuda and St. Croix, Virgin Islands. Here
values of aerosol depolarization are about 35 % and lidar ratio values are about 48–
50 sr. Smaller but still large aerosol depolarization values between about 20 % to 35 %
have been often observed by HSRL particularly in the CHAPS and RACORO cam-10
paigns (in Oklahoma) and in the MILAGRO campaign (in Mexico), as well as near
Houston, Texas during the TexAQS/GoMACCS mission (Liu et al., 2008; Heese et al.,
2009). These values of depolarization are identified by our algorithm as a “dusty mix”.
Motivating and supporting the idea of having two dust categories with different degrees
of depolarization was the observation that the overall distribution of aerosol depolar-15
ization from campaigns excluding ARCTAS (that is, excluding ice) shows a long tail of
large depolarization values which includes samples known to be dust advected from
Africa. Consequently, we also included a second dust category (“Pure dust”) to identify
such cases. Including it has the practical advantage of aiding the separation of ice and
dust. The depolarization values of the optically thin ice observed during the ARCTAS20
campaign can be comparable to that of pure dust, making it is difficult to use particle
depolarization alone to separate ice and dust. As described above, the lidar ratio tends
to be smaller for ice than for dust but the difference is subtle enough that the separation
of types is more reliable when dust is represented as two categories. This is due to
the fact that ice can be more “similar” to pure dust than pure dust is to the remainder25
of the dust observations, where similarity is judged as distance in the four-dimensional
space defined by the four measured variables. Even with two categories for dust, it is
still somewhat difficult to separate ice from dust in certain cases. Misclassified cases
5647
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
are often easy to detect because the ambient temperatures obtained from the NASA
Global Modeling and Assimilation Office (GMAO) are well above 0
◦
C. Consequently, a
simple temperature-based correction is included in the results shown here; any point
categorized as ice but having a temperature above 0
◦
C is reassigned to dust (this is a
very conservative cutoff which potentially can leave some cases incorrectly categorized5
as ice).
4.2.3 Maritime and polluted maritime
Maritime aerosols were observed extensively during HSRL observations over the
Caribbean Sea during several flights in 2007 and 2010. These aerosols were charac-
terized with low lidar ratios (15–25 sr), low par ticulate depolarization (<10 %), and low10
backscatter color and depolarization spectral ratios. The polluted maritime classifica-
tion is generally seen over water or just inland on the Gulf Coast between Houston and
Veracruz during the MILAGRO campaign (March 2006) and over the Atlantic Ocean
east of Virginia during several campaigns. It was also found extensively in the Gulf of
Mexico near the location of the BP Deepwater Horizon oil spill on flights in May and15
July 2010 (see example in Ottaviani et al., 2011, Fig. 7). The lidar ratio for this class
is about 35–45 sr, intermediate between the maritime and pollution classes, consistent
with observations by M
¨
uller et al. (2007a) of polluted marine air over the Maldives dur-
ing the monsoon season. In our polluted maritime class, backscatter color ratio and
spectral ratio of aerosol depolarization are also intermediate between the maritime and20
pollution classes. It is a small category that contains about 3 % of all the HSRL obser-
vations. Most of the fully automated clustering trials did not distinguish between this
type and clean maritime air; however, an experimental run of unsupervised expectation
maximization clustering with eleven classes generated a cluster like this. Supporting
the decision to include the class, it was found that many of these cases would otherwise25
be labeled an incoherent mix of pollution, smoke and maritime.
5648
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
4.2.4 Urban and biomass burning
Urban and biomass burning aerosols typically have relatively small, spherical parti-
cles that produce low depolarization, high backscatter color ratios, and high lidar ratios
(Cattrall et al., 2005; M
¨
uller et al., 2007a). The similarities in the physical characteris-
tics and the resulting optical properties make these types difficult to distinguish. M
¨
uller5
et al. (2007a) have shown that urban and smoke aerosols can be distinguished using
the wavelength dependence of the lidar ratio (355–532 nm) computed from ground-
based Raman lidar measurements, and Russell et al. (2010) indicate that absorption
˚
Angstr
¨
om exponent derived from AERONET shows promise for separating them. The
depolarization spectral ratio measurements acquired by HSRL also appear to confer10
some ability to discriminate among these and other aerosol types. The HSRL mea-
surements from a flight on 2 August 2007 over the Atlantic Ocean east of Virginia
shown in Fig. 7 illustrate the significant differences in the depolarization spectral ratio,
despite the fact that the aerosol depolarization values are small. The aerosols be-
low 3 km are typical of the urban aerosols seen over the eastern US during summer.15
The aerosols in the elevated layer above 5 km are smoke from biomass burning fires
in the northwestern US or southwestern Canada. The elevated layer of smoke has
slightly higher lidar ratio (70–80 sr) than the urban aerosols (50–70 sr) consistent with
previous Raman lidar measurements of smoke (Wandinger et al., 2002). The elevated
smoke layer also has slightly higher particulate depolarization (8–10 %) than the lower20
layer of urban aerosols; this observation of smoke particulate depolarization is consis-
tent with other lidar measurements of long-range smoke transport (Fiebig et al., 2002;
Murayama et al., 2004). Although there have been few multiple-wavelength lidar par-
ticulate depolarization measurements of these aerosols, there have been efforts to use
such measurements to help identify and classify polar stratospheric clouds (Toon et25
al., 2000), examine Saharan dust characteristics (Freudenthaler et al., 2009), and infer
Angstrom exponents for dust (Sugimoto and Lee, 2006). Somekawa et al. (2008) and
Veselovskii et al. (2010) show that multiple wavelength depolarization measurements
may be used to infer some particle proper ties such as size.
5649
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
4.2.5 Fresh smoke
The category “Fresh Smoke” was included based on observations of visible fresh
smoke plumes with very different aerosol intensive parameters measured by HSRL.
For example, samples of fresh smoke in the boundary layer observed on 30 June and
2 July 2008 during ARCTAS had a significantly smaller lidar ratio (30–60 sr) than the ad-5
vected smoke (60–80 sr) from Siberian forest fires seen on other dates during the same
campaign, such as 7–8 July 2008. Figure 8 shows an example of HSRL measurements
for the smoke plume observed over northern Alberta on 30 June 2008. Fresh smoke
plumes observed over fires in North Carolina in March 2008 also had similar smaller
lidar ratios of approximately 50–55 sr. In both of these cases, the smoke was only a few10
hours old and observations were within 10–100 km of the fires. The lower value of lidar
ratios for fresh smoke as compared to aged smoke are consistent with ground-based
Raman lidar measurements over Spain (Alados-Arboledas et al., 2011) and Greece
(Amiridis et al., 2009). Although these studies and M
¨
uller et al. (2007b) indicate that
particle size is likely to increase with age, there is considerable spread in the observed15
backscattering Angstrom exponents. HSRL measurements of backscatter
˚
Angstr
¨
om
exponent indicate larger values (smaller particles) for fresh smoke than aged smoke
on average, but without a clear separation (compare Wandinger et al., 2002; Amiridis
et al., 2009; Alados-Arboledas et al., 2011). The HSRL measurements also showed
that the aerosol depolarization ratio for fresh smoke was typically low (<2–5 %) and20
also typically lower that the depolarization ratio for the more aged smoke (3–8 %). This
result is consistent with the magnitude and variability of previous lidar measurements
of smoke (Sassen, 2000).
4.3 Sensitivity analysis
As described above, the distinctions in the HSRL measurements of these four aerosol25
intensive parameters support the choices of these classes. The Wilks’ overall lambda
statistic (Hill and Lewicki, 2007) gives some indication of how well the data lend them-
selves to separation into classes. Wilks’ lambda varies from 0 to 1 with smaller values
5650
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
indicating significant difference between groups and larger values indicating that the
group means are the same. For the HSRL data classified into eight classes as de-
scribed above, Wilks’ lambda is 0.083. If outliers are also included, the value is 0.137.
Wilks’ partial lambda can be used to indicate the relative discriminatory power of
each intensive parameter. This value is the ratio of Wilks’ lambda calculated with and5
without a given variable. Again, smaller values indicate more importance, allowing the
values to be ranked in order. Wilks’ partial lambda is smallest for the 532 nm depolar-
ization, 0.47, indicating that this variable has the most weight in the classification. This
is followed closely by depolarization spectral ratio and lidar ratio, with par tial lambda
values of 0.54 for each. The backscatter color ratio has the least discriminatory power,10
with a partial lambda of 0.79.
Some classes are easier to distinguish than others. The potential for misclassifica-
tion is illustrated in Fig. 9, which shows the results of a Monte Carlo study wherein
simulated observations are made by perturbing each point 500 times within the mea-
surement uncertainties of the four intensive variables; then these simulated points are15
themselves classified. Table 3 shows the median measurement uncertainties for these
variables. For this test, the uncertainty values for the two spectral ratios are propa-
gated from the single-channel values with an assumption of independence between
the channels, so the uncertainties used here are larger (more conservative) than the
true measurement uncertainty. Figure 9 illustrates the probability that perturbing each20
measurement within the uncertainties will change the inferred classification. This is
one way to understand the relative difficulty in separating various pairs of classes. For
example, the maritime class is quite easy to infer. Even after perturbation, most of the
Monte Carlo points are still classified as maritime, with very small percentages cross-
classified into the other categories. Not surprisingly, smoke and urban are harder to25
separate, and between about 5 % and 15 % of the perturbed points are cross-classified.
Polluted maritime has cross-classification into the related categor ies of maritime and
urban. The dust category and the “fresh smoke” category derived from smoke in the
boundary layer have the most cross-classifications with other categories.
5651
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
The Monte Carlo study quantifies only one possible kind of error in this analysis,
essentially the potential for misclassification due to measurement errors. This experi-
ment does not address potential errors from the choice of classes or training samples or
the assumption that multi-normal distributions are adequate to represent the classes.
These types of potential systematic errors are of course difficult to quantify. Further5
confidence in the results can be gained by comparisons with other data sets, begun in
the next section and carried forward in a future paper (Ferrare et al., 2011).
5 Results of the classification
The ranges of the intensive parameters applicable to each of the aerosol classes are
displayed in Fig. 10, as median, middle 50 % (boxes) and middle 90 % (whiskers).10
Also shown are gray bars representing the number of observations of these various
aerosol types. Figures 11 and 12 also show the results displayed as a series of two-
dimensional histograms. Points are color coded by the aerosol classification derived
from this study and with the color saturation for each hue corresponding to point den-
sity. Less populated bins are not shown; the figures show approximately 50 % of the15
points in each class. Figure 11 also shows the aerosol intensive properties from other
lidar measurements (M
¨
uller et al., 2007a) and derived from ground-based AERONET
observations of aerosol properties (Cattrall et al., 2005; Omar et al., 2005) from exist-
ing literature. There is general qualitative agreement, showing for example that dust
and maritime aerosols typically have lower lidar ratios and backscatter color ratios, and20
smoke and urban type aerosols having higher lidar ratios and backscatter color ratios.
Figure 11 also clearly shows that there can be considerable spread in these observa-
tions for aerosols observed in different locations. This figure shows that using lidar ratio
and backscatter ratio alone would be insufficient to classify all these aerosol types, as
there can be considerable overlap among some of these classes. However, Fig. 12,25
which shows the additional variables used in the current scheme, indicates that aerosol
depolarization and spectral depolarization ratio can be used to distinguish these types.
5652
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
As seen in the figure, the spectral depolarization ratio helps especially in distinguishing
ice from dust and smoke from pollution.
Examples of the results of the classifications for some of the observed aerosol types
described earlier are shown in Figs. 5 (ice and dust), 6 (dust), 7 (smoke and urban) and
8 (fresh smoke). Figure 5 shows the separation of ice and dust in the upper troposphere5
over Alaska, due primarily to the low values of lidar ratio (20–30 sr) and much higher
values of depolarization (>0.4) associated with ice. Figure 6 shows the presence of
pure dust associated with Saharan dust transported over the western Atlantic Ocean.
The identification of pure dust was driven primarily by the aerosol depolarization values
of 0.3–0.35 located near the center of the layer. Around the periphery, where the10
aerosol depolarization was below about 0.3 and the aerosols were likely mixed with
other types, the classification was a dusty mix. Figure 7 shows the classification of
the elevated smoke layer above the urban aerosols for the flight over the eastern US
on 2 August 2007. Here separation between smoke and urban was driven by the
differences in spectral depolarization (lower for smoke) and depolarization (higher for15
smoke). Figure 8 shows the classification of fresh smoke when the B200 flew over fires
in northern Saskatchewan, Canada. Fresh smoke was classified based on the lower
values of the lidar ratio (40–50 sr) combined with low values of aerosol depolarization.
Figure 13 shows the results of the classification for the HSRL measurements acquired
over Mexico City and shown in Fig. 2. The classification indicates urban aerosols when20
the B200 flew over the western part of the city between about 17.75–17.80 UT (17:45–
17:48), and indicates a dusty mix when the B200 flew over the eastern part of the
city between 17.68–17.75 UT (17:40–17:45). As described earlier, the classification
of dusty mix vs. urban was due to variations in the lidar ratio and depolarization over
these locations. Also visible in Fig. 13 is an elevated fresh smoke plume at about 4.5 km25
over the western part of the city. The identification of fresh smoke here is consistent
with WRF-Flexpart (de Foy et al., 2011). Figure 13 also illustrates an example of the
apportionment of AOT among these types.
5653
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Figure 14 illustrates the apportionment of AOT for two cases, discussed previously,
having a significant dust component. These cases are the flight on 13 March 2006
over Mexico City (as in Figs. 2 and 13) and the Caribbean flight of 18 August 2010
(Fig. 6). The black line shown in Fig. 14 is the dust fraction computed using the method
of Sugimoto and Lee (2006), which assumes that the amount of dust in a mixture5
scales linearly with the aerosol depolarization. Figure 14b and d illustrate a comparison
of this computed dust fraction to the total AOT associated with the two types “pure
dust” and “dusty mix” in our classification, for the entire MILAGRO and Caribbean 2010
campaigns. In general, the two estimates are in agreement, but the sum of the AOT
for the two classes exceeds the dust fraction as computed using the Sugimoto and10
Lee (2006) algorithm. This is not surprising since most of the aerosol in the “dusty mix”
type has depolarization less than the assumed depolarization value for pure dust in
that calculation. In the MILAGRO campaign, the dust is mixed with urban and smoke
aerosol, while in the Caribbean campaign it is mixed with maritime aerosol.
The HSRL measurements acquired during the spring and summer ARCTAS cam-15
paigns have been used to apportion the vertical profile of aerosol extinction to aerosol
types. Figure 15 reflects the median aerosol extinction profiles measured during the
spring and summer ARCTAS campaigns, apportioned by aerosol type. B200 flights
were conducted in April 2008 over northern Alaska during the spring ARCTAS cam-
paign (“ARCTAS 1”) and over northern Alberta, northern Saskatchewan, and the south-20
ern Northwest Territories Canada in June and July 2008 during the summer ARCTAS
campaign (“ARCTAS 2”). Figure 15 shows that, during ARCTAS 1, ice was more pro-
nounced in the mid troposphere between 2–5 km, and in the upper troposphere be-
tween 6–7 km during ARCTAS 2. The fraction of aerosol extinction contributed by dust
was relatively constant with altitude during ARCTAS 1 and decreased with altitude25
during ARCTAS 2. During ARCTAS 1, portions of several B200 flights were conducted
over water and the Arctic Ocean which likely explains the significant fraction of maritime
extinction observed near the surface. In contrast, very little maritime aerosol was ob-
served dur ing ARCTAS 2 which is not surprising given that the flights were conducted
5654
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
inland over Canada. Urban aerosols were most prominently observed at the lowest
altitudes, especially during ARCTAS 1. Smoke, contributed by both the fresh and aged
components, was dominant during both ARCTAS 1 and ARCTAS 2. Airborne in situ
measurements acquired during the ARCTAS 1 mission also found smoke as the domi-
nant component. The B200 flights during ARCTAS 2 were designed to sample smoke5
from biomass bur ning fires so it expected that smoke would dominate. Note also that
the lower altitudes had higher concentrations of fresh smoke, especially during ARC-
TAS 2.
The ARCTAS mission also provided an opportunity to compare the classification
measurements with airborne in situ measurements of size and composition. On 1210
and 19 April 2008, the NASA B200 flew patterns that enabled HSRL to acquire coin-
cident data with in situ sensors on the NOAA WP-3D aircraft, which was deployed to
conduct the airborne Aerosol, Radiation, and cloud Processes affecting Arctic Climate
(ARCPAC) field study (Warneke et al., 2010; Brock et al., 2011). The WP-3D deployed
a suite of instruments for measuring gas, aerosol, and radiation properties including15
optical particle counters for measuring the aerosol volume distribution and the Particle
Analysis by Laser Mass Spectrometry (PALMS) instrument for size-resolved single-
particle composition (Froyd et al., 2009). Figure 16 shows the results of the HSRL
aerosol classification and aerosol volume distribution and particle classification distri-
butions from the PALMS. As described by Warneke et al. (2010), the compositional20
resolved volume distributions represent the product of the number fraction of each
aerosol type in a given size bin and the total aerosol volume for that size bin. Figure 16
shows that biomass burning material was the largest component, with other contribu-
tions from sulfate/organic and mineral dust. The HSRL aerosol classification results
are consistent with this, with the majority of the aerosol types classified as smoke and25
a smaller portion classified as urban. Figure 17 shows another example comparing the
HSRL aerosol classification and volume distribution and particle classification distribu-
tions from the PALMS for data acquired on 19 April 2008 during ARCTAS 1. These data
were acquired in the vicinity of Barrow, Alaska when there was an extensive amount
5655
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
of biomass burning smoke over this region (Warneke et al., 2010). This smoke was
produced by fires in Russia. Figure 17 shows that the HSRL aerosol classification indi-
cated that biomass burning aerosols were present in most of the troposphere over the
Barrow region during this flight. The PALMS measurements also show that biomass
burning smoke was dominant during these flights. Additional investigations comparing5
the HSRL aerosol classification results with airborne in situ measurements acquired
during ARCTAS and other field campaigns are ongoing.
The contributions of each type to the total optical depth measured by HSRL dur-
ing each mission are shown in Fig. 18. Some missions were dominated by a single
type; for example, maritime air in the Caribbean campaign, a field mission primarily10
over water in a location chosen for clean conditions. The urban type dominated the
TexAQS/GoMACCS campaign, which occurred near Houston, Texas; the Birmingham
and the San Joaquin Valley (California) (Lewis et al., 2010) campaigns; and CALIPSO
validation flights which have primarily occurred over the East coast of the United States.
The urban type was also seen in large amounts in the MILAGRO campaign near Mex-15
ico City, in that case along with large amounts of dust. CHAPS and RACORO cam-
paigns near Oklahoma City also saw both pollution and dust. The ice classification is
present in significant amounts only in the spring deployment of the ARCTAS campaign
(ARCTAS 1), in which biomass burning smoke (Warneke et al., 2010) was the other
predominant component. Smoke also dominated the summer deployment of ARCTAS20
(ARCTAS 2) (Jacob et al., 2010).
6 Summary
A method to qualitatively classify aerosol types based on airborne HSRL measure-
ments of the aerosol intensive parameters has been presented here. Several exam-
ples show how these aerosol parameters vary with different aerosol types and can25
therefore be used to discriminate among these types. For example, the HSRL mea-
surements show that ice and dust can in many cases be distinguished using the lidar
5656
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
ratio, and to a lesser extent, particle depolarization. Urban and biomass burning smoke
aerosols, which typically have somewhat similar lidar ratios (at 532 nm) and backscatter
color ratios (532/1064 nm), can be difficult to distinguish; however, the HSRL measure-
ments show that urban and biomass burning aerosols can have significant differences
in spectral particle depolarization and that these differences can be used to help dis-5
tinguish these aerosols. Further improvements in distinguishing urban and biomass
burning smoke could be realized through the use of additional backscatter and extinc-
tion measurements at 355 nm (M
¨
uller et al., 2007a). The HSRL measurements also
show differences in the lidar ratio between fresh and aged smoke. This classification
method uses HSRL measurements of the lidar ratio, backscatter color ratio, depolariza-10
tion, and depolarization spectral ratio to infer the appropriate type. The method, which
uses a training set of known aerosol cases to help define the set of lidar parameters
appropriate for each type, was applied to the extensive set of airborne HSRL observa-
tions acquired since 2006. The classification results were used together with the HSRL
measurements of aerosol optical thickness to apportion the aerosol optical thickness15
among the various aerosol types. These results show that the dominant aerosol types
in terms of aerosol optical thickness vary significantly with location. Aerosol classifica-
tion results using HSRL measurements have already been useful in field campaigns,
as evidenced by published examples of the identification of smoke aerosols during
the NASA Arctic Research of the Composition of the Troposphere from Aircraft and20
Satellites (ARCTAS) mission (Warneke et al., 2010) and urban aerosols during the
MILAGRO campaign (Molina et al., 2010).
The HSRL classification results were used to examine the vertical variability of
aerosol types observed during the NASA ARCTAS mission that was conducted dur-
ing the spring and summer 2008. The results show that biomass burning aerosol was25
the dominant aerosol type for both the spring and summer deployments, which is con-
sistent with other measurements (Jacob et al., 2010; Warneke et al., 2010; Brock et
al., 2011). In two cases, the HSRL classification results were shown to be consistent
with aerosol types derived from coincident airborne in situ measurements.
5657
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
As will be discussed in the companion paper, more aerosol type information from
measurements such as HSRL can potentially be used to improve model inputs and
assess the ability of global and regional models to accurately portray aerosol (Ferrare
et al., 2011). Accurate aerosol discrimination can also improve retrievals of aerosol
properties from space. The CALIPSO aerosol algorithm, for example, requires an a5
priori estimate of the lidar ratio in the retrieval of aerosol extinction (Omar et al., 2009).
The HSRL measurements described here show how the lidar ratio varies with these
major aerosol types. A technique for classifying aerosol from lidar measurements such
as the one presented here may be useful as a means of constraining advance multi-
wavelength lidar retrievals such as those using inversion with regularization (M
¨
uller10
et al., 1999; Veselovskii et al., 2002). In such cases, the typing results can essentially
serve as a pre-inversion classifier to more efficiently and rapidly solve for aerosol micro-
physical parameters, potentially allowing these advanced retrievals to become suitable
for operational use from future spaceborne lidars.
Acknowledgements. Funding for this research came from the NASA HQ Science Mission Di-15
rectorate Radiation Sciences Program, the NASA CALIPSO project, and the Department of
Energy Atmospheric Science Program, Interagency Agreement No. DE-AI02-05ER63985. The
authors would like to thank the NASA Langley B200 King Air flight crew for their outstanding
work in support of HSRL measurements.
References20
Ackermann, J.: The extinction-to-backscatter ratio of tropospheric aerosol: A numerical study,
J. Atmos. Ocean. Tech., 15, 1043–1050, 1998.
Alados-Arboledas, L., M
¨
uller, D., Guerrero-Rascado, J. L., Navas-Guzm
´
an, F., P
´
erez-Ram
´
ırez,
D., and Olmo, F. J.: Optical and microphysical properties of fresh biomass burning aerosol
retrieved by Raman lidar, and star-and sun-photometry, Geophys. Res. Lett., 38, L01807,25
doi:10.1029/2010gl045999, 2011.
Amiridis, V., Balis, D. S., Giannakaki, E., Stohl, A., Kazadzis, S., Koukouli, M. E., and Zanis, P.:
Optical characteristics of biomass burning aerosols over Southeastern Europe determined
5658
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
from UV-Raman lidar measurements, Atmos. Chem. Phys., 9, 2431–2440, doi:10.5194/acp-
9-2431-2009, 2009.
Anderson, T. L., Masonis, S. J., Covert, D. S., Charlson, R. J., and Rood, M. J.: In situ measure-
ment of the aerosol extinction-to-backscatter ratio at a polluted continental site, J. Geophys.
Res., 105, 26907–26915, 2000.5
Ansmann, A., Riebesell, M., and Weitkamp, C.: Measurement of Atmospheric Aerosol Extinc-
tion Profiles with a Raman Lidar, Opt. Lett., 15, 746–748, 1990.
Brock, C. A., Cozic, J., Bahreini, R., Froyd, K. D., Middlebrook, A. M., McComiskey, A., Brioude,
J., Cooper, O. R., Stohl, A., Aikin, K. C., de Gouw, J. A., Fahey, D. W., Ferrare, R. A., Gao,
R.-S., Gore, W., Holloway, J. S., Hbler, G., Jefferson, A., Lack, D. A., Lance, S., Moore, R.10
H., Murphy, D. M., Nenes, A., Novelli, P. C., Nowak, J. B., Ogren, J. A., Peischl, J., Pierce, R.
B., Pilewskie, P., Quinn, P. K., Ryerson, T. B., Schmidt, K. S., Schwarz, J. P., Sodemann, H.,
Spackman, J. R., Stark, H., Thomson, D. S., Thornberry, T., Veres, P., Watts, L. A., Warneke,
C., and Wollny, A. G.: Characteristics, sources, and transport of aerosols measured in spring
2008 during the aerosol, radiation, and cloud processes affecting Arctic Climate (ARCPAC)15
Project, Atmos. Chem. Phys., 11, 2423–2453, doi:10.5194/acp-11-2423-2011, 2011.
Burton, S. P., Ferrare, R. A., Hostetler, C. A., Hair, J. W., Kittaka, C., Vaughan, M. A., Obland, M.
D., Rogers, R. R., Cook, A. L., Harper, D. B., and Remer, L. A.: Using airborne high spectral
resolution lidar data to evaluate combined active plus passive retrievals of aerosol extinction
profiles, J. Geophys. Res.-Atmos., 115, D00H15, doi:10.1029/2009jd012130, 2010.20
Campbell, J. R., Welton, E. J., Spinhirne, J. D., Ji, Q., Tsay, S. C., Piketh, S. J., Barenbrug, M.,
and Holben, B. N.: Micropulse lidar observations of tropospheric aerosols over northeastern
South Africa during the ARREX and SAFARI 2000 dry season experiments, J. Geophys.
Res.-Atmos., 108, 8497, doi:10.1029/2002jd002563, 2003.
Cattrall, C., Reagan, J., Thome, K., and Dubovik, O.: Variability of aerosol and spectral lidar and25
backscatter and extinction ratios of key aerosol types derived from selected Aerosol Robotic
Network locations, J. Geophys. Res.-Atmos., 110, D10S11, doi:10.1029/2004jd005124,
2005.
Chin, M., Ginoux, P., Kinne, S., Torres, O., Holben, B. N., Duncan, B. N., Martin, R. V., Logan,
J. A., Higurashi, A., and Nakajima, T.: Tropospheric Aerosol Optical Thickness from the30
GOCART Model and Comparisons with Satellite and Sun Photometer Measurements, J.
Atmos. Sci., 59, 461–483, 2002.
Curr y, J. A., Meyer, F. G., Radke, L. F., Brock, C. A., and Ebert, E. E.: Occurrence and Char-
5659
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
acteristics of Lower Tropospheric Ice Crystals in the Arctic, Int. J. Climatol., 10, 749–764,
1990.
Davis, K. J., Gamage, N., Hagelberg, C. R., Kiemle, C., Lenschow, D. H., and Sullivan, P. P.:
An objective method for deriving atmospheric structure from airborne lidar observations, J.
Atmos. Ocean. Tech., 17, 1455–1468, 2000.5
de Foy, B., Burton, S. P., Ferrare, R. A., Hostetler, C. A., Hair, J. W., Wiedinmyer, C., and
Molina, L. T.: Aerosol plume transport and transformation in high spectral resolution lidar
measurements and WRF-Flexpart simulations during the MILAGRO Field Campaign, Atmos.
Chem. Phys., 11, 3543–3563, doi:10.5194/acp-11-3543-2011, 2011.
Dempster, A. P., Laird, N. M., and Rubin, D. B.: Maximum Likelihood from Incomplete Data via10
the EM Algorithm, J. R. Stat. Soc., B Met., 39, 1–38, 1977.
Dubovik, O., Holben, B., Eck, T. F., Smirnov, A., Kaufman, Y. J., King, M. D., Tanr
´
e, D., and
Slutsker, I.: Variability of absor ption and optical properties of key aerosol types observed in
worldwide locations, J. Atmos. Sci., 59, 590–608, 2002.
Eloranta, E. W.: High spectral resolution lidar, in: Lidar: Range-Resolved Optical Remote15
Sensing of the Atmosphere, edited by: Weitkamp, K., Springer, New York, 143–163, 2005.
Ferrare, R. A., Burton, S. P., Hostetler, C. A., Hair, J. W., Rogers, R. R., Obland, M. D., Butler,
C. F., Cook, A. L., Harper, D. B., Chin, M., Omar, A., and Vaughan, M.: Aerosol classification
of airborne High Spectral Resolution Lidar measurements – comparisons with CALIOP and
GOCART, in preperation, 2011.20
Fiebig, M., Petzold, A., Wandinger, U., Wendisch, M., Kiemle, C., Stifter, A., Ebert, M., Rother,
T., and Leiterer, U.: Optical closure for an aerosol column: Method, accuracy, and inferable
properties applied to a biomass-burning aerosol and its radiative forcing, J. Geophys. Res.,
107, 8130, doi:10.1029/2000jd000192, 2002.
Freudenthaler, V., Esselborn, M., Wiegner, M., Heese, B., Tesche, M., Ansmann, A., Muller,25
D., Althausen, D., Wirth, M., Fix, A., Ehret, G., Knippertz, P., Toledano, C., Gasteiger, J.,
Garhammer, M., and Seefeldner, M.: Depolarization ratio profiling at several wavelengths
in pure Saharan dust during SAMUM 2006, Tellus B, 61, 165–179, doi:10.1111/j.1600-
0889.2008.00396.x, 2009.
Froyd, K. D., Murphy, D. M., Sanford, T. J., Thomson, D. S., Wilson, J. C., Pfister, L., and Lait, L.:30
Aerosol composition of the tropical upper troposphere, Atmos. Chem. Phys., 9, 4363–4385,
doi:10.5194/acp-9-4363-2009, 2009.
Girard, E. and Blanchet, J.-P.: Microphysical Parameterization of Arctic Diamond Dust, Ice
5660
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Fog, and Thin Stratus for Climate Models, J. Atmos. Sci., 58, 1181–1198, doi:10.1175/1520-
0469(2001)058<1181:MPOADD>2.0.CO;2, 2001.
GLOBE Task Team, Hastings, D. A., Dunbar, P. K., Elphingstone, G. M., Bootz, M., Murakami,
H., Maruyama, H., Masaharu, H., Holland, P., Payne, J., Bryant, N. A., Logan, T. L., Muller,
J.-P., Schreier, G., and MacDonald, J. S.: The Global Land One-kilometer Base Elevation5
(GLOBE) Digital Elevation Model, Version 1.0, edited by: National Oceanic and Atmospheric
Administration, N. G. D. C., Boulder, Colorado, 1999.
Greenaway, K. R.: Experiences with arctic flying weather, Publications of the Royal Meteoro-
logical Society, Canadian Branch, 1, 1950.
Grund, C. J. and Eloranta, E. W.: University-of-Wisconsin High Spectral Resolution Lidar, Opt.10
Eng., 30, 6–12, 1991.
Hair, J. W., Hostetler, C. A., Cook, A. L., Harper, D. B., Ferrare, R. A., Mack, T. L., Welch,
W., Izquierdo, L. R., and Hovis, F. E.: Airborne High Spectral Resolution Lidar for profiling
aerosol optical properties, Appl. Opt., 47, 6734–6752, doi:10.1364/AO.47.006734, 2008.
Heese, B., Althausen, D., Dinter, T., Esselborn, M., Muller, E. T., Tesche, M., and Wiegner, M.:15
Vertically resolved dust optical properties during SAMUM: Tinfou compared to Ouarzazate,
Tellus B, 61, 195–205, doi:10.1111/j.1600-0889.2008.00404.x, 2009.
Hill, T. and Lewicki, P.: STATISTICS Methods and Applications, StatSoft, Tulsa, OK, 2007.
Hoff, R. M.: Vertical Structure of Arctic Haze Observed by Lidar, J. Appl. Meteorol., 27, 125–
139, 1988.20
Holben, B. N., Eck, T. F., Slutsker, I., Tanre, D., Buis, J. P., Setzer, A., Vermote, E., Reagan, J.
A., Kaufman, Y. J., Nakajima, T., Lavenu, F., Jankowiak, I., and Smirnov, A.: AERONET – A
federated instrument network and data archive for aerosol characterization, Remote Sens.
Environ., 66, 1–16, 1998.
Jacob, D. J., Crawford, J. H., Maring, H., Clarke, A. D., Dibb, J. E., Emmons, L. K., Ferrare, R.25
A., Hostetler, C. A., Russell, P. B., Singh, H. B., Thompson, A. M., Shaw, G. E., McCauley,
E., Pederson, J. R., and Fisher, J. A.: The Arctic Research of the Composition of the Tropo-
sphere from Aircraft and Satellites (ARCTAS) mission: design, execution, and first results,
Atmos. Chem. Phys., 10, 5191–5212, doi:10.5194/acp-10-5191-2010, 2010.
Kahn, R., Gaitley, B. J., Martonchik, J. V., Diner, D. J., Crean, K. A., and Holben, B.: Multiangle30
Imaging Spectroradiometer (MISR) global aerosol optical depth validation based on 2 years
of coincident Aerosol Robotic Network (AERONET) observations, J. Geophys. Res.-Atmos.,
110, D10S04, doi:10.1029/2004jd004706, 2005.
5661
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Kent, G. S., Trepte, C. R., Skeens, K. M., and Winker, D. M.: LITE and SAGE II measurements
of aerosols in the southern hemisphere upper troposphere, J. Geophys. Res., 103, 19111–
19127, doi:10.1029/98jd00364, 1998.
Knobelspiesse, K., Cairns, B., Ottaviani, M., Ferrare, R., Hair, J., Hostetler, C., Obland, M.,
Rogers, R., Redemann, J., Shinozuka, Y., Clarke, A., Freitag, S., Howell, S., Kapustin, V.,5
and McNaughton, C.: Combined retrievals of boreal forest fire aerosol properties with a
polarimeter and lidar, Atmos. Chem. Phys., 11, 7045–7067, doi:10.5194/acp-11-7045-2011,
2011.
Lewis, J., De Young, R., Ferrare, R., and Allen Chu, D.: Comparison of summer and winter
California central valley aerosol distributions from lidar and MODIS measurements, Atmos.10
Environ., 44, 4510–4520, doi:10.1016/j.atmosenv.2010.07.006, 2010.
Liu, Z. Y., Omar, A., Vaughan, M., Hair, J., Kittaka, C., Hu, Y. X., Powell, K., Trepte, C., Winker,
D., Hostetler, C., Ferrare, R., and Pierce, R.: CALIPSO lidar observations of the optical
properties of Saharan dust: A case study of long-range transport, J. Geophys. Res.-Atmos.,
113, D07207, doi:10.1029/2007jd008878, 2008.15
MacQueen, J.: Some Methods for Classification and Analysis of Multivariate Observations,
Proceedings of 5th Berkeley Symposium on Mathematical Statistics and Probability, 281–
297, 1967.
Mahalanobis, P. C.: On the Generalized Distance in Statistics, Proceedings of the National
Institute of Sciences in India, 2, 49–55, 1936.20
McNaughton, C. S., Clarke, A. D., Kapustin, V., Shinozuka, Y., Howell, S. G., Anderson, B.
E., Winstead, E., Dibb, J., Scheuer, E., Cohen, R. C., Wooldridge, P., Perring, A., Huey, L.
G., Kim, S., Jimenez, J. L., Dunlea, E. J., DeCarlo, P. F., Wennberg, P. O., Crounse, J. D.,
Weinheimer, A. J., and Flocke, F.: Observations of heterogeneous reactions between Asian
pollution and mineral dust over the Eastern North Pacific during INTEX-B, Atmos. Chem.25
Phys., 9, 8283–8308, doi:10.5194/acp-9-8283-2009, 2009.
Mishchenko, M. and Sassen, K.: Depolarization of lidar returns by small ice crystals: An appli-
cation to contrails, Geophys. Res. Lett., 25, 309–312, 1998.
Molina, L. T., Madronich, S., Gaffney, J. S., Apel, E., de Foy, B., Fast, J., Ferrare, R., Herndon,
S., Jimenez, J. L., Lamb, B., Osornio-Vargas, A. R., Russell, P., Schauer, J. J., Stevens, P.30
S., Volkamer, R., and Zavala, M.: An overview of the MILAGRO 2006 Campaign: Mexico
City emissions and their transport and transformation, Atmos. Chem. Phys., 10, 8697–8760,
doi:10.5194/acp-10-8697-2010, 2010.
5662
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
M
¨
uller, D., Wandinger, U., and Ansmann, A.: Microphysical particle parameters from extinction
and backscatter lidar data by inversion with regularization: theory, Appl. Opt., 38, 2346–
2357, 1999.
M
¨
uller, D., Ansmann, A., Mattis, I., Tesche, M., Wandinger, U., Althausen, D., and Pisani, G.:
Aerosol-type-dependent lidar ratios observed with Raman lidar, J. Geophys. Res.-Atmos.,5
112, D16202, doi:10.1029/2006jd008292, 2007a.
M
¨
uller, D., Mattis, I., Ansmann, A., Wandinger, U., Ritter, C., and Kaiser, D.: Multiwave-
length Raman lidar observations of particle growth during long-range transport of forest-fire
smoke in the free troposphere, Geophys. Res. Lett., 34, L05803, doi:10.1029/2006gl027936,
2007b.10
Murayama, T., Furushima, M., Oda, A., and Iwasaka, N.: Depolarization ratio measurements in
the atmospheric boundary layer by lidar in Tokyo, J. Meteorol. Soc. Jpn, 74, 571–578, 1996.
Murayama, T., Okamoto, H., Kaneyasu, N., Kamataki, H., and Miura, K.: Application of lidar
depolarization measurement in the atmospheric boundary layer: Effects of dust and sea-salt
particles, J. Geophys. Res.-Atmos., 104, 31781–31792, 1999.15
Murayama, T., Masonis, S. J., Redemann, J., Anderson, T. L., Schmid, B., Livingston, J. M.,
Russell, P. B., Huebert, B., Howell, S. G., McNaughton, C. S., Clarke, A., Abo, M., Shimizu,
A., Sugimoto, N., Yabuki, M., Kuze, H., Fukagawa, S., Maxwell-Meier, K., Weber, R. J., Orsini,
D. A., Blomquist, B., Bandy, A., and Thornton, D.: An intercomparison of lidar-derived aerosol
optical properties with airborne measurements near Tokyo during ACE-Asia, J. Geophys.20
Res.-Atmos., 108, L05803, doi:10.1029/2002jd003259, 2003.
Murayama, T., M
¨
uller, D., Wada, K., Shimizu, A., Sekiguchi, M., and Tsukamoto, T.: Charac-
terization of Asian dust and Siberian smoke with multiwavelength Raman lidar over Tokyo,
Japan in spring 2003, Geophys. Res. Lett., 31, L23103, doi:10.1029/2004gl021105, 2004.
National Research Council: Earth Science and Applications from Space: National Imperatives25
for the Next Decade and Beyond, The National Academies Press, Washington, D.C., 400 pp.,
2007.
Omar, A. H., Won, J.-G., Winker, D. M., Yoon, S.-C., Dubovik, O., and McCormick, M. P.:
Development of global aerosol models using cluster analysis of Aerosol Robotic Network
(AERONET) measurements, J. Geophys. Res., 110, D10S14, doi:10.1029/2004jd004874,30
2005.
Omar, A. H., Winker, D. M., Kittaka, C., Vaughan, M. A., Liu, Z. Y., Hu, Y. X., Trepte, C. R.,
Rogers, R. R., Ferrare, R. A., Lee, K. P., Kuehn, R. E., and Hostetler, C. A.: The CALIPSO
5663
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Automated Aerosol Classification and Lidar Ratio Selection Algorithm, J. Atmos. Ocean
Tech., 26, 1994–2014, doi:10.1175/2009jtecha1231.1, 2009.
Ottaviani, M., Cairns, B., Chowdhary, J., Van Diedenhoven, B., Knobelspiesse, K., Hostetler,
C., Ferrare, R., Burton, S., Hair, J., Obland, M., and Rogers, R.: Polarimetric retrievals of
surface properties in the region affected by the Deepwater Horizon oil spill, Remote Sens.5
Environ., in preperation, 2011.
Parrish, D. D., Allen, D. T., Bates, T. S., Estes, M., Fehsenfeld, F. C., Feingold, G., Ferrare,
R., Hardesty, R. M., Meagher, J. F., Nielsen-Gammon, J. W., Pierce, R. B., Ryerson, T. B.,
Seinfeld, J. H., and Williams, E. J.: Overview of the Second Texas Air Quality Study (TexAQS
II) and the Gulf of Mexico Atmospheric Composition and Climate Study (GoMACCS), J.10
Geophys. Res.-Atmos., 114, D00F13, doi:10.1029/2009jd011842, 2009.
Powell, D. M., Reagan, J. A., Rubio, M. A., Erxleben, W. H., and Spinhirne, J. D.: ACE-2 multiple
angle micro-pulse lidar observations from Las Galletas, Tenerife, Canary Islands, Tellus B,
52, 652–661, doi:10.1034/j.1600-0889.2000.00059.x, 2000.
Redemann, J., Zhang, Q., Livingston, J., Russell, P., Shinozuka, Y., Clarke, A., Johnson,15
R., and Levy, R.: Testing aerosol properties in MODIS Collection 4 and 5 using airborne
sunphotometer observations in INTEX-B/MILAGRO, Atmos. Chem. Phys., 9, 8159–8172,
doi:10.5194/acp-9-8159-2009, 2009.
Remer, L. A., Kaufman, Y. J., Tanre, D., Mattoo, S., Chu, D. A., Martins, J. V., Li, R. R., Ichoku,
C., Levy, R. C., Kleidman, R. G., Eck, T. F., Vermote, E., and Holben, B. N.: The MODIS20
aerosol algorithm, products, and validation, J. Atmos. Sci., 62, 947–973, 2005.
Rogers, R. R., Hair, J. W., Hostetler, C. A., Ferrare, R. A., Obland, M. D., Cook, A. L., Harper,
D. B., Burton, S. P., Shinozuka, Y., McNaughton, C. S., Clarke, A. D., Redemann, J., Russell,
P. B., Livingston, J. M., and Kleinman, L. I.: NASA LaRC airborne high spectral resolution
lidar aerosol measurements during MILAGRO: observations and validation, Atmos. Chem.25
Phys., 9, 4811–4826, doi:10.5194/acp-9-4811-2009, 2009.
Russell, P. B., Bergstrom, R. W., Shinozuka, Y., Clarke, A. D., DeCar lo, P. F., Jimenez, J.
L., Livingston, J. M., Redemann, J., Dubovik, O., and Strawa, A.: Absorption Angstrom
Exponent in AERONET and related data as an indicator of aerosol composition, Atmos.
Chem. Phys., 10, 1155–1169, doi:10.5194/acp-10-1155-2010, 2010.30
Saha, A., O’Neill, N. T., Eloranta, E., Stone, R. S., Eck, T. F., Zidane, S., Daou, D.,
Lupu, A., Lesins, G., Shiobara, M., and McArthur, L. J. B.: Pan-Arctic sunphotome-
tr y during the ARCTAS-A campaign of April 2008, Geophys. Res. Lett., 37, L05803,
5664
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
doi:10.1029/2009gl041375, 2010.
Sakai, T., Nagai, T., Nakazato, M., Mano, Y., and Matsumura, T.: Ice Clouds and Asian
Dust Studied with Lidar Measurements of Particle Extinction-to-Backscatter Ratio, Particle
Depolarization, and Water-Vapor Mixing Ratio over Tsukuba, Appl. Opt., 42, 7103–7116,
doi:10.1364/AO.42.007103, 2003.5
Sakai, T., Nagai, T., Zaizen, Y., and Mano, Y.: Backscattering linear depolarization ratio mea-
surements of mineral, sea-salt, and ammonium sulfate particles simulated in a laboratory
chamber, Appl. Opt., 49, 4441–4449, 2010.
Sasano, Y. and Browell, E. V.: Light-Scattering Characteristics of Various Aerosol Types De-
rived from Multiple Wavelength Lidar Observations, Appl. Opt., 28, 1670–1679, 1989.10
Sassen, K.: Lidar backscatter depolarization technique for cloud and aerosol research, in: Light
Scattering by Nonspherical Particles: Theory, Measurements, and Applications, edited by:
Mishchenko, M. I., Hovenier, J. W., and Travis, L. D., Academic, San Diego, CA, 2000.
Sassen, K.: Indirect climate forcing over the western US from Asian dust storms, Geophys.
Res. Lett., 29, 1465, doi:10.1029/2001gl014051, 2002.15
Sassen, K.: Identifying Atmospheric Aerosols with Polarization Lidar, in: Advanced Environ-
mental Monitor ing, edited by: Kim, Y. J. and Platt, U., Springer-Verlag, Berlin, 136–142,
2008.
Sassen, K. and Hsueh, C.-y.: Contrail properties derived from high-resolution polarization li-
dar studies during SUCCESS, Geophys. Res. Lett., 25, 1165–1168, doi:10.1029/97gl03503,20
1998.
Schmid, B., Ferrare, R., Flynn, C., Elleman, R., Covert, D., Strawa, A., Welton, E., Turner, D.,
Jonsson, H., Redemann, J., Eilers, J., Ricci, K., Hallar, A. G., Clayton, M., Michalsky, J.,
Smirnov, A., Holben, B., and Barnard, J.: How well do state-of-the-art techniques measur-
ing the vertical profile of tropospheric aerosol extinction compare?, J. Geophys. Res., 111,25
D05S07, doi:10.1029/2005jd005837, 2006.
She, C.-Y.: Spectral Structure of Laser Light Scattering Revisited: Bandwidths of Nonresonant
Scattering Lidars, Appl. Opt., 40, 4875–4884, 2001.
She, C. Y., Alvarez, R. J., Caldwell, L. M., and Kr ueger, D. A.: High-Spectral-Resolution
Rayleigh-Mie Lidar Measurement of Vertical Aerosol and Atmospheric Profiles, Appl. Phys.30
B-Photo, 55, 154–158, 1992.
Shimizu, A., Sugimoto, N., Matsui, I., Arao, K., Uno, I., Murayama, T., Kagawa, N., Aoki, K.,
Uchiyama, A., and Yamazaki, A.: Continuous observations of Asian dust and other aerosols
5665
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
by polarization lidars in China and Japan during ACE-Asia, J. Geophys. Res., 109, D19S17,
doi:10.1029/2002jd003253, 2004.
Shipley, S. T., Tracy, D. H., Eloranta, E. W., Trauger, J. T., Sroga, J. T., Roesler, F. L., and
Weinman, J. A.: High Spectral Resolution Lidar to Measure Optical-Scattering Properties of
Atmospheric Aerosols .1. Theory and Instrumentation, Appl. Opt., 22, 3716–3724, 1983.5
Somekawa, T., Yamanaka, C., Fujita, M., and Galvez, M. C.: A new concept to characterize
nonspherical particles from multi-wavelength depolarization ratios based on T-matrix com-
putation, Part Part Syst Char, 25, 49–53, doi:10.1002/ppsc.200700009, 2008.
Spinhirne, J. D.: Micro pulse lidar, IEEE T. Geosci. Remote Sens., 31, 48–55, 1993.
Sugimoto, N. and Lee, C. H.: Characteristics of dust aerosols inferred from lidar depolarization10
measurements at two wavelengths, Appl. Opt., 45, 7468–7474, 2006.
Sugimoto, N., Matsui, I., Shimizu, A., Uno, I., Asai, K., Endoh, T., and Nakajima, T.: Ob-
ser vation of dust and anthropogenic aerosol plumes in the Northwest Pacific with a two-
wavelength polarization lidar on board the research vessel Mirai, Geophys. Res. Lett., 29,
1901, doi:10.1029/2002gl015112, 2002.15
Toon, O. B., Tabazadeh, A., Browell, E. V., and Jordan, J.: Analysis of lidar observations of Arc-
tic polar stratospheric clouds during January 1989, J. Geophys. Res.-Atmos., 105, 20589–
20615, 2000.
Veselovskii, I., Kolgotin, A., Griaznov, V., Muller, D., Wandinger, U., and Whiteman, D. N.:
Inversion with regularization for the retrieval of tropospheric aerosol parameters from multi-20
wavelength lidar sounding, Appl. Opt., 41, 3685–3699, 2002.
Veselovskii, I., Dubovik, O., Kolgotin, A., Lapyonok, T., Di Girolamo, P., Summa, D., Whiteman,
D. N., and Tanr
´
e, D.: Application of Randomly Oriented Spheroids for Retrieval of Dust
Particle Parameters from Multiwavelength Lidar Measurements, International Laser Radar
Conference, St. Petersburg, Russia, 2010.25
Wandinger, U., Muller, D., Bockmann, C., Althausen, D., Matthias, V., Bosenberg, J., Weiss, V.,
Fiebig, M., Wendisch, M., Stohl, A., and Ansmann, A.: Optical and microphysical character-
ization of biomass-burning and industrial-pollution aerosols from multiwavelength lidar and
aircraft measurements, J. Geophys. Res.-Atmos., 107, 8125, doi:10.1029/2000jd000202,
2002.30
Warneke, C., Froyd, K. D., Brioude, J., Bahreini, R., Brock, C. A., Cozic, J., de Gouw, J.
A., Fahey, D. W., Ferrare, R., Holloway, J. S., Middlebrook, A. M., Miller, L., Montzka, S.,
Schwarz, J. P., Sodemann, H., Spackman, J. R., and Stohl, A.: An important contribution to
5666
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
springtime Arctic aerosol from biomass burning in Russia, Geophys. Res. Lett., 37, L01801,
doi:10.1029/2009gl041816, 2010.
Welton, E. J., Voss, K. J., Gordon, H. R., Mar ing, H., Smirnov, A., Holben, B., Schmid, B.,
Livingston, J. M., Russell, P. B., Durkee, P. A., Formenti, P., and Andreae, M. O.: Ground-
based lidar measurements of aerosols during ACE-2: instrument description, results, and5
comparisons with other ground-based and airborne measurements, Tellus B, 52, 636–651,
2000.
Welton, E. J., Voss, K. J., Quinn, P. K., Flatau, P. J., Markowicz, K., Campbell, J. R., Spinhirne,
J. D., Gordon, H. R., and Johnson, J. E.: Measurements of aerosol vertical profiles and
optical properties during INDOEX 1999 using micropulse lidars, J. Geophys. Res., 107, 8019,10
doi:10.1029/2000jd000038, 2002.
Whiteman, D. N., Melfi, S. H., and Ferrare, R. A.: Raman lidar system for the measurement of
water vapor and aerosols in the Earth’s atmosphere, Appl. Opt., 31, 3068–3082, 1992.
Winker, D. M., Couch, R. H., and McCormick, M. P.: An overview of LITE: NASA’s Lidar In-
space Technology Experiment, P. IEEE, 84, 164–180, 1996.15
Winker, D. M., Vaughan, M. A., Omar, A., Hu, Y. X., Powell, K. A., Liu, Z. Y., Hunt, W. H., and
Young, S. A.: Overview of the CALIPSO Mission and CALIOP Data Processing Algorithms,
J. Atmos. Ocean Tech., 26, 2310–2323, doi:10.1175/2009jtecha1281.1, 2009.
5667
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Table 1. Field campaigns for the NASA Airborne HSRL.
Field Number Flight
mission Location Dates flights hours
MILAGRO Mexico City 3/1–3/31/2006 22 64.4
CALIPSO Validation Eastern USA May–Aug 2006 20 56.7
TexAQS/GOMACCS Texas 8/27–9/28/06 28 89.0
San Joaquin Valley California 2/8–2/20/2007 15 45.0
CHAPS/CLASIC Oklahoma City area 6/03–6/29/2007 22 70.2
CATZ CALIPSO Val. Eastern USA Jan–Aug 2007 20 49.9
CALIPSO Validation Caribbean Jan–Feb 2008 13 42.2
ARCTAS Spring Alaska 3/30–4/22/2008 27 97.9
ARCTAS Summer Canada 6/24–7/13 2008 21 71.5
Birmingham Alabama 9/12–10/15/2008 11 35.1
CALIPSO Validation Eastern USA Jan–Apr 2009 13 39.7
RACORO Oklahoma 5/21–6/27/2009 24 72.9
Ocean Subsurface Atlantic ocean 9/14–9/29/2009 5 18.6
CALIPSO Validation Eastern USA 4/8–4/22/2010 7 15.6
CALIPSO Gulf Oil Spill Gulf of Mexico May, Jul 2010 6 19.7
CalNEX California 5/11–5/24/2010 13 44.5
CARES California 6/3–6/30/2010 25 80.1
CALIPSO Validation Caribbean 8/4–8/27/2010 9 35.9
5668
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Table 2. Criteria for filtering aerosol measurements for certain figures.
Aerosol property Filter criter ia
Depolarization at 532 nm 0 ≤ x ≤ 0.6
Extinction-to-backscatter ratio at 532 nm 0 ≤ x ≤ 100
Backscatter color ratio, 532 nm:1064 nm 0.4 ≤ x ≤ 4.5
Ratio of aerosol depolarization ratios, 1064 nm:532 nm 0 ≤ x ≤ 3.5
5669
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Table 3. Median uncertainties for intensive parameters used in Monte Carlo cross-classification
analysis.
Intensive variable Median uncertainty for HSRL
measurements included in aerosol
classification study
Aerosol depolarization at 532 nm 0.0074
Lidar ratio at 532 nm 12.1 sr
Backscatter spectral ratio (532/1064 nm) 0.128 (propagated from backscatter
uncertainties assuming independence)
Depolarization spectral ratio (1064/532 nm) 0.774 (propagated from aerosol
depolarization uncertainties)
5670
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
CALIPSO/MODIS/CATZ (NASA)
January 17– Aug 11, 2007
Ocean Subsurface (NASA-ODU-NYU)
September 9-29, 2009
TexAQS II/GoMACCS
NOAA-DOE-NASA
Aug 27 – Sep 29, 2006
ARCTAS 1 (NASA-DOE-NOAA)
April 1-20, 2008
Caribbean CALIPSO Val. (NASA)
Jan. 22 – Feb. 3, 2008
ARCTAS 2 (NASA)
June 25 – July 14, 2008
CALIPSO Validation (NASA)
June 14 – Aug 10, 2006
January 22 – April 17, 2009
April 8 – 22, 2010
MAXMex/MILAGRO/INTEX-B
DOE-NSF-NASA-Mexico
March 1-30, 2006
San Joaquin Valley (EPA)
February 8-21, 2007
Field Campaigns:
2006 (3), 2007 (3), 2008 (4), 2009 (3), 2010(5)
CHAPS (DOE-NASA)
June 3-29, 2007
RACORO (DOE-NASA)
June 3-26, 2009
Birmingham (EPA)
Sept 16-Oct 16, 2008
CalNex (NOAA) May 12-25, 2010
CARES (DOE) June 3-28, 2010
Bermuda/Caribbean
August 11-28, 2010
Gulf Oil Spill / CALIPSO validation (NASA)
May 10-11 and July 9-11, 2010
Fig. 1. Location of airborne HSRL flights and field experiments from 2006 through 2010.
5671
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Fig. 2. Airborne HSRL measurements when the NASA B200 King Air flew over Mexico City
between 17:38–17:52 UT on 13 March 2006 during the MILAGRO campaign. The aircraft flew
from from east (right) to west (left). The images cover a horizontal distance of about 115 km.
(a) Aerosol backscatter coefficient (532 nm), (b) aerosol extinction coefficient (532 nm), (c) S
a
,
the lidar ratio (532 nm), (d) aerosol depolarization (532 nm), (e) aerosol depolarization spectral
ratio (1064/532 nm), (f) aerosol backscatter-related
˚
Angstr
¨
om exponent (between 1064 and
532 nm). Variations in the parameters measured by the HSRL reflect variability in aerosol type.
5672
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Aerosol Depolarization
0.0
0.1
0.2
0.3
0.4
MILAGRO
GOMACCS
CALIPSO
SJV
CHAPS
CARIBBEAN
ARCTAS1
ARCTAS2
BIRMINGHAM
RACORO
CALNEX
GULFOIL
CARES
CARIBBEAN
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
Lidar Ratio
0
20
40
60
80
100
MILAGRO
GOMACCS
CALIPSO
SJV
CHAPS
CARIBBEAN
ARCTAS1
ARCTAS2
BIRMINGHAM
RACORO
CALNEX
GULFOIL
CARES
CARIBBEAN
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
Backscatter Color Ratio
1.0
1.5
2.0
2.5
3.0
3.5
MILAGRO
GOMACCS
CALIPSO
SJV
CHAPS
CARIBBEAN
ARCTAS1
ARCTAS2
BIRMINGHAM
RACORO
CALNEX
GULFOIL
CARES
CARIBBEAN
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
Depolarization Spectral Ratio
0
1
2
3
4
MILAGRO
GOMACCS
CALIPSO
SJV
CHAPS
CARIBBEAN
ARCTAS1
ARCTAS2
BIRMINGHAM
RACORO
CALNEX
GULFOIL
CARES
CARIBBEAN
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
Fig. 3. Distributions of aerosol intensive parameters derived from HSRL measurements dur-
ing recent field campaigns. Median values of each of the four intensive variables are shown
for each mission (dots) along with the 25–75th percentiles (bars) and 5–95th percentiles
(whiskers). The number of observations for each field campaign are shown by the grey his-
togram bars. Several CALIPSO validation campaigns in the Eastern US and off the east coast
have been grouped together in the single category “CALIPSO” in this figure.
5673
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Backscatter Color Ratio
0
20
40
60
80
100
Lidar Ratio
0.01 0.10
Aerosol Depolarization
0
20
40
60
80
100
Lidar Ratio
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Spectral Depolarization Ratio
0
20
40
60
80
100
Lidar ratio
Ice
Pure Dust
Dusty Mix
Maritime
Polluted Maritime
Urban
Fresh Smoke
Smoke
Fig. 4. Illustrates the models used in the aerosol classification algorithm in three projections
of a space defined by the four aerosol intensive variables measured by HSRL. Crosshairs
indicate data samples of known type as mean and standard deviation of the four variables.
Ellipses indicate the two-sigma covariance of the aerosol type models that are based on these
samples.
5674
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Fig. 5. 3 April 2008 measurements by the NASA Langley airborne HSRL based out of Barrow,
Alaska during the ARCTAS campaign. (a) Aerosol backscatter coefficient (532 nm), (b) aerosol
type inferred by the method described in this paper, (c) particle depolarization (532 nm), (d) par-
ticle depolarization spectral ratio (1064/532 nm), (e) S
a
, the lidar ratio (532 nm), and (f) aerosol
backscatter-related
˚
Angstr
¨
om exponent (between 1064 and 532 nm), showing an elevated layer
made up of both ice crystals, in the region characterized by the largest particle depolarization
ratios and smallest lidar ratio, and dust, with somewhat smaller particle depolarization and
larger lidar ratio.
5675
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Fig. 6. 18 August 2010 airborne HSRL measurements and aerosol classification for a Sa-
haran dust plume observed on a flight between Bermuda and St. Croix in the US Virgin Is-
lands. (a) Aerosol backscatter coefficient (532 nm), (b) aerosol type inferred by the method
described in this paper, (c) aerosol depolarization (532 nm), (d) aerosol depolarization spec-
tral ratio (1064/532 nm), (e) S
a
, the lidar ratio (532 nm), and (f) aerosol backscatter-related
˚
Angstr
¨
om exponent (between 1064 and 532 nm).
5676
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Fig. 7. A B200 flight on the East coast on 2 August 2007 illustrating a smoke layer advected
from fires in the Northwestern United States and Canada, overlying mostly pollution aerosol
from cities on the eastern seaboard. (a) Aerosol backscatter coefficient (532 nm), (b) aerosol
type inferred by the method described in this paper, (c) aerosol depolarization (532 nm), (d)
aerosol depolarization spectral ratio (1064/532 nm), (e) S
a
, the lidar ratio (532 nm), and (f)
aerosol backscatter-related
˚
Angstr
¨
om exponent (between 1064 and 532 nm). Note the contrast
in spectral depolarization ratio between the two aerosol layers.
5677
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Fig. 8. NASA Langley airborne HSRL observations and aerosol classification are shown for
a portion of a flight on 30 June 2008 over northern Alberta, Canada. Multiple passes over a
fresh smoke plume are evident. (a) Aerosol backscatter coefficient (532 nm), (b) aerosol type
inferred by the method described in this paper, (c) aerosol depolarization (532 nm), (d) aerosol
depolarization spectral ratio (1064/532 nm), (e) S
a
, the lidar ratio (532 nm), and (f) aerosol
backscatter-related
˚
Angstr
¨
om exponent (between 1064 and 532 nm).
5678
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
0.0
0.2
0.4
0.6
0.8
1.0
Probability
Monte Carlo Classification Confidence Test
Ice
9818 cases
Monte Carlo Classification Confidence Test
Pure Dust
1013 cases
Monte Carlo Classification Confidence Test
Dusty Mix
64470 cases
Monte Carlo Classification Confidence Test
Maritime
9748 cases
0.0
0.2
0.4
0.6
0.8
1.0
Probability
Monte Carlo Classification Confidence Test
Polluted Maritime
8768 cases
Monte Carlo Classification Confidence Test
Urban
82648 cases
Monte Carlo Classification Confidence Test
Fresh Smoke
16585 cases
Monte Carlo Classification Confidence Test
Smoke
36805 cases
Ice
Pure
Dust
Dusty
Mix
Maritime
Polluted
Maritime
Urban
Fresh
Smoke
Smoke
Fig. 9. Shows the results of a Monte Carlo experiment in which a cloud of 500 perturbed
measurements for each point is classified and the classification is compared to the classification
of the original unperturbed point. The first panel shows the results for all points that were
originally classified ice; the bins along the x-axis show the statistics of how the perturbed points
were classified, color coded as shown. The second panel is for pure dust, etc., as labeled.
Perturbed ice measurements are still ice; perturbed pure dust are split among pure dust, dusty
mix and ice; dusty mix and maritime are easy to classify. Smoke (especially fresh smoke) is
difficult to separate from pollution.
5679
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Aerosol Depolarization
0.0
0.1
0.2
0.3
0.4
0.5
Ice
Pure Dust
Dusty Mix
Maritime
Polluted
Maritime
Urban
Fresh Smoke
Smoke
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
Lidar Ratio
0
20
40
60
80
100
Ice
Pure Dust
Dusty Mix
Maritime
Polluted
Maritime
Urban
Fresh Smoke
Smoke
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
Backscatter Color Ratio
0
1
2
3
4
Ice
Pure Dust
Dusty Mix
Maritime
Polluted
Maritime
Urban
Fresh Smoke
Smoke
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
Depolarization Spectral Ratio
0
1
2
3
4
Ice
Pure Dust
Dusty Mix
Maritime
Polluted
Maritime
Urban
Fresh Smoke
Smoke
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
0
2•10
4
4•10
4
6•10
4
8•10
4
1•10
5
Number of points
Fig. 10. Colored bars and whiskers show the median (dot), 25–75 percentile (box) and 5–95
percentile (whisker) of the four aerosol intensive parameters, after classifying all HSRL data
from all missions into eight types. The gray bars represent the number of points in each class,
using the right-hand data axis.
5680
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Fig. 11. The results of the classification of HSRL measurements are shown here, projected
onto a two-dimensional subset of the four dimensional space. HSRL measurements are color
coded by inferred aerosol type, with the saturation in each hue indicating relative population
density. Points are shown for the most populous bins such that about half of the population
of each cluster is represented. Also indicated in this figure are the aerosol types identified by
Cattrall et al. (2005), Omar et al. (2005), and M
¨
uller et al. (2007a). (Some of these variables
have been inverted to conform to the axes chosen here.)
5681
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Fig. 12. Like Fig. 11 but showing two-dimensional projections that include the other two aerosol
intensive variables that were used for classification. In this figure, the bins are shown as solid
boxes; individual points within the bins are not displayed.
5682
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
2
3
4
5
6
Altitude (km)
Ice
Pure
Dust
Dusty
Mix
Maritime
Polluted
Maritime
Urban
Fresh
Smoke
Smoke
17.85 17.80 17.75 17.70 17.65
Time (UT)
0.0
0.1
0.2
0.3
0.4
Aerosol Optical Depth (532 nm)
Fig. 13. Results of the aerosol classification for the HSRL measurements shown in Fig. 2. The
top panel shows aerosol type along the flight track as a function of altitude. Urban aerosols
dominate in the western part of Mexico City while dusty aerosol dominates elsewhere. An
elevated smoke plume is also visible around 4.5 km altitude in the west part. The bottom panel
illustrates the apportioning of aerosol optical depth among the types for this flight segment as
stacked histogram bars.
5683
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
17.85 17.80 17.75 17.70 17.65
Time (UT)
0.0
0.1
0.2
0.3
0.4
Aerosol O.D. (532 nm)
Ice Pure Dust
Dusty
Mix
Maritime
Polluted
Maritime
Urban
Fresh
Smoke
Smoke
0.0 0.1 0.2 0.3 0.4 0.5
Dust O.D. after Sugimoto and Lee
0.0
0.1
0.2
0.3
0.4
0.5
Dust O.D. from
HSRL classification
15.015.516.016.517.017.518.018.5
Time (UT)
0.0
0.1
0.2
0.3
0.4
0.5
0.6
Aerosol O.D. (532 nm)
Ice Pure Dust
Dusty
Mix
Maritime
Polluted
Maritime
Urban
Fresh
Smoke
Smoke
0.0 0.1 0.2 0.3 0.4 0.5
Dust O.D. after Sugimoto and Lee
0.0
0.1
0.2
0.3
0.4
0.5
Dust O.D. from
HSRL classification
(a) (b)
(c) (d)
Fig. 14. Time series of AOD apportioned to aerosol types for the flight segments on
13 March 2006 during MILAGRO (a) and for 18 August 2010 during the Caribbean 2010 field
mission (c). The black trace on these panels shows the dust fraction as computed using the
method of Sugimoto and Lee (2006). The Sugimoto and Lee (2006) dust fraction and the to-
tal of the AOD for “Pure Dust” plus “Dusty Mix” are compared for the entire MILAGRO field
campaign (b) and for the 2010 Caribbean campaign (d).
5684
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
ARCTAS 1 (Spring-Alaska)
0 10 20 30 40
Median Aerosol Extinction (532 nm), Mm
-1
0
1
2
3
4
5
6
7
Altitude, km
ARCTAS 2 (Summer-Canada)
0 10 20 30 40 50
Median Aerosol Extinction (532 nm), Mm
-1
0
1
2
3
4
5
6
7
Altitude, km
Ice
Pure Dust
Dusty Mix
Maritime
Polluted
Maritime
Urban
Fresh
Smoke
Smoke
Fig. 15. Aerosol extinction as a function of altitude is shown here apportioned among the eight
aerosol types for the ARCTAS spring and summer campaigns.
5685
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Fig. 16. Aerosol classification from HSRL measurements for 12 April 2008 B-200 flight near
Barrow, Alaska during ARCTAS/ARCPAC (left) and PALMS aerosol composition data from the
NOAA P3 from 22:40–22:57 UT (right). Both instruments indicate mainly biomass burning
aerosol, consistent with known smoke plumes from fires in Russia (see Warneke et al., 2010).
The left panel also shows coincident portions of the flight track profile of the P3 when it was
within 30 km and 1 h of the HSRL flight track.
5686
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Fig. 17. (Left) Aerosol classification from HSRL measurements onboard the B200 for
19 Apr il 2008 B-200 flight near Barrow, Alaska during ARCTAS/ARCPAC, with black trace
showing coincident portions of the flight track profile of the NOAA P3 when the two aircraft
were within 30 km and 1 h of each other. The time axis indicates GMT time on 19 April. Times
beyond 24 h are used to indicate the early hours of 20 April GMT. (Right) PALMS aerosol com-
position data from the NOAA P3 between 00:03–00:19 GMT (20 April). The PALMS instrument
indicates a very dense biomass burning plume between 1.5 and 3 km, and the HSRL classifica-
tion also indicates smoke throughout. These results are consistent with known smoke plumes
from fires in Russia (see Warneke et al., 2010).
5687
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
ARCTAS 1
ARCTAS 2
CalNex
CARES
San Joaquin
Valley
Gulf Oil Spill
TexAQS II/GoMACCS
MILAGRO
Birmingham
2008 Caribbean CAL Val
CALIPSO Validation
2006-2010
RACOROCHAPS
2010 Caribbean CAL Val.
Fig. 18. All HSRL missions through 2010 are shown, along with the partitioning of total optical
depth among the eight aerosol types for each of these missions. Several CALIPSO validation
campaigns in the Eastern US and off the east coast have been grouped together in the single
categor y “CALIPSO Validation” in this figure.
5688
