Use of in situ cloud condensation nuclei, extinction, and aerosol size
distribution measurements to test a method for retrieving cloud
condensation nuclei profiles from surface measurements
Steven J. Ghan,
1
Tracey A. Rissman,
2
Robert Elleman,
3
Richard A. Ferrare,
4
David Turner,
1
Connor Flynn,
1
Jian Wang,
5
John Ogren,
6
James Hudson,
7
Haflidi H. Jonsson,
8
Timothy VanReken,
2
Richard C. Flagan,
2
and John H. Seinfeld
2
Received 28 December 2004; revised 21 April 2005; accepted 19 July 2005; published 19 January 2006.
[
1
]
If the aerosol composition and size distribution below cloud are uniform, the vertical
profile of cloud condensation nuclei concentration can be retrieved entirely from surface
measurements of CCN concentration and particle humidification function and surface-
based retrievals of relative humidity and aerosol extinction or backscatter. This provides
the potential for long-term measurements of CCN concentrations near cloud base. We
have used a combination of aircraft, surface in situ, and surface remote sensing
measurements to test various aspects of the retrieval scheme. Our analysis leads us to the
following conclusions. The retrieval works better for supersaturations of 0.1% than for
1% because CCN concentrations at 0.1% are controlled by the same particles that control
extinction and backscatter. If in situ measurements of extinction are used, the retrieval
explains a majority of the CCN variance at high supersaturation for at least two and
perhaps five of the eight flights examined. The retrieval of the vertical profile of the
humidification factor is not the major limitation of the CCN retrieval scheme. Vertical
structure in the aerosol size distribution and composition is the dominant source of error in
the CCN retrieval, but this vertical structure is difficult to measure from remote sensing at
visible wavelengths.
Citation:
Ghan, S. J., et al. (2006), Use of in situ cloud condensation nuclei, extinction, and aerosol size distribution measurements to
test a method for retrieving cloud condensation nuclei profiles from surface measurements,
J. Geophys. Res.
,
111
,D05S10,
doi:10.1029/2004JD005752.
1. Introduction
[
2
] One of the greatest sources of uncertainty in estimates
of global climate change by climate models is in the
treatment of indirect effects of aerosols on cloud optical
depth through the role of aerosols as cloud condensation
nuclei (CCN). All cloud droplets form on aerosol particles,
so the CCN concentration has a powerful influence on
droplet number concentration. However, the maximum
supersaturation (which largely determines the number of
CCN activated) in updrafts depends on the updraft velocity,
which is highly variable within the droplet nucleation zone
of clouds. Furthermore, droplet number is reduced by
evaporation, by droplet collision and coalescence with other
droplets and with precipitating drops, and the precipitation
process (which reduces the liquid water path of the cloud)
which depends on both the mean and the dispersion of the
droplet number size distribution [
Liu and Daum
,2002].
[
3
] These complicating factors make it very difficult to
represent aerosol indirect effects in climate models, to
evaluate that representation, and to isolate the aerosol
indirect effect from field measurements. Aircraft measure-
ments have been used to evaluate droplet nucleation models
[
Lin and Leaitch
,1997;
Gultepe et al.
,1998;
Yum and
Hudson
,2002;
Hudson and Yum
,2002;
Snider et al.
,2003;
Conant et al.
,2004;
Peng et al.
,2005],butsuchhigh-
quality measurements are too costly to permit the collection
of the thousands of independent samples needed to isolate
the indirect effect in models and observations. Moreover,
they do not permit the simultaneous measurement of cloud
base properties (updraft velocity and CCN concentration)
and column integrated properties (liquid water path and
optical depth). Satellite retrievals provide a large sample
size of measurements of column integrated properties [
Han
et al.
,1998],butcannotprovideestimatesofupdraft
velocity and CCN concentration at cloud base. Surface in
situ measurements on mountaintops [
Hallberg et al.
,1997;
Menon and Saxena
,1998;
Menon et al.
,2002]providean
JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 111, D05S10, doi:10.1029/2004JD005752, 2006
1
Pacific Northwest National Laboratory, Richland, Washington, USA.
2
California Institute of Technology, Pasadena, California, USA.
3
Department of Atmospheric Science, University of Washington,
Seattle, Washington, USA.
4
NASA Langley Research Center, Hampton, Virginia, USA.
5
Brookhaven National Laboratory, Upton, New York, USA.
6
NOAA Climate Monitoring and Diagnostics Laboratory, Boulder,
Colorado, USA.
7
Desert Research Institute, Reno, Nevada, USA.
8
Naval Postgraduate School, Monterey, California, USA.
Copyright 2006 by the American Geophysical Union.
0
1
4
8
-
0
2
2
7
/
0
6
/
2
0
0
4
J
D
0
0
5
7
5
2
D05S10
1of15
economical source of measurements but are only useful
when cloud base is near the elevation of the site. Conse-
quently, there have been few attempts to use measured CCN
concentration to evaluate the treatment of indirect effects in
climate models [
Menon et al.
,2003;
Ovtchinnikov and
Ghan
,2005].
[
4
]Surface-basedremotesensingofferssomeappealing
advantages to these other measurement strategies. By look-
ing upward from the surface, profilers can provide useful
information about the aerosol up to cloud base, about
updrafts within the cloud, and about column-integrated
cloud properties such as liquid water path and cloud optical
depth. This permits long-term collection of data that can be
used to isolate the aerosol indirect effect and evaluate the
treatment of it in single column versions of global climate
models.
[
5
]
Kim et al.
[2003] and
Penner et al.
[2004] used
surface-based remote sensing of cloud optical depth and
liquid water path to demonstrate how the dependence of
optical depth on liquid water path (i.e., the droplet effective
radius) varies from day to day, but only used a surface
measure of the aerosol to relate to this dependence.
Feingold et al.
[2003] extended this method by relating
the droplet effective radius to the aerosol extinction near
cloud base.
[
6
]Althoughaerosolextinctionmightserveasafirst
approximation to CCN concentration, further improvements
are possible by (1) accounting for the influence of water
uptake on extinction and (2) using the resulting dry extinc-
tion to scale surface measurements of CCN concentration.
This provides the ability to estimate the full CCN spectrum
at cloud base, if the spectrum is measured at the surface.
[
7
]ThismethodforestimatingCCNconcentrationnear
cloud base was suggested by
Ghan and Collins
[2004,
hereinafter referred to as GC]. In this retrieval, surface
measurements of the CCN concentration
CCN
(S,
z
0
)at
supersaturation
S
are scaled by the ratio of the dry extinction
(or 180
°
backscatter) profile
s
de
(
z
) to the dry extinction (or
180
°
backscatter) at or near the surface,
s
de
(
z
0
):
CCN S
;
z
ðÞ¼
CCN S
;
z
0
ðÞ
s
de
z
ðÞ
=
s
de
z
0
ðÞð
1
Þ
The dry extinction (or 180
°
backscatter) profile
s
de
(
z
) is
determined from the extinction (or 180
°
backscatter) profile
at ambient humidity
s
e
(
z
)andthedependenceofextinction
(or 180
°
backscatter) on relative humidity,
f
(
RH
(
z
)):
s
de
z
ðÞ¼
s
e
z
ðÞ
=
fRHz
ðÞ
ðÞð
2
Þ
The aerosol particle humidification factor
f
(
RH
) is measured
at the surface and is assumed to apply at all levels up to
cloud base using the retrieved relative humidity profile. GC
describe the instruments that can be used to provide the
necessary measurements for this retrieval.
Anderson et al.
[2000] and GC show that for
RH
up to 80%,
f
(
RH
) for
extinction is indistinguishable from
f
(
RH
) for 180
°
back-
scatter. We will therefore use the same expression for both.
[
8
] The method assumes the humidification factor mea-
sured at the surface is representative of the humidification
factor at altitude, and it assumes that the vertical structure
of CCN concentration is identical to the vertical structure
of dry extinction or backscatter. Since both extinction/
backscatter and CCN concentration are determined entirely
by the size distribution of aerosol number, composition,
and geometric shape, both of these assumptions are valid if
(1) the aerosol size distribution (but not necessarily the total
aerosol number) is independent of altitude, and (2) the
aerosol composition and particle shape are independent of
altitude. GC used in situ aerosol size distribution measure-
ments, Mie theory, and Ko
̈hler theory to examine the vertical
variability of the size distribution, but did not have the CCN
or aerosol composition measurements needed to investigate
the vertical variability of aerosol composition and shape.
Clearly the impact of this variability on the retrieval also
needs to be tested.
[
9
] In May 2003 the Atmospheric Radiation Measure-
ment (ARM) program conducted an aerosol intensive
observation period (IOP) that provides the data needed to
test assumptions A and B. The goal of this study is to
evaluate the GC CCN retrieval and to understand what is
limiting its performance. In section 2 we describe the design
of the ARM experiment, and in section 3 we describe the
use of the measurements to evaluate the performance of the
retrieval scheme. Conclusions are summarized in section 4.
2. Experiment Design
2.1. Instruments and Measurements
[
10
] To distinguish between different sources of error in
the retrieval scheme, a variety of measurements were
collected. These include both in situ and remote sensing
measurements. In situ measurements were collected both
from aircraft and at the surface.
2.1.1. Measurements From Aircraft
[
11
] In situ measurements include (1) CCN concentration,
(2) aerosol size distribution, (3) relative humidity, (4) aerosol
scattering and absorption, and (5) aerosol particle humidi-
fication factor. Although in situ measurements of aerosol
composition and shape are not available (except for
composition at the ground), the measurements of CCN
concentration, aerosol scattering and absorption, and hu-
midification provide the opportunity to test the influence of
variability in aerosol composition and shape on the CCN
retrieval because each of these fields depend on aerosol
composition and shape.
[
12
]TheCCNconcentrationsweremeasuredfrom
the Center for Interdisciplinary Remotely Piloted Aircraft
Studies (CIRPAS) Twin Otter aircraft every second by the
California Institute of Technology (Caltech) CCN counter.
The CCN counter has three columns, each operating with a
linear axial temperature gradient, allowing each column to
achieve one supersaturation. Only two of the columns
operated during the IOP. Because of undetected problems
with the detector on column 2, the supersaturation for
column 2 could not be determined for any of the flights,
so the CCN concentrations for column 2 will not be
considered here. The operating supersaturation of column
1wasdeterminedfromthecriticalsupersaturationof
(NH
4
)
2
SO
4
particles with dry size such that 50% of a
controlled size are able to activate in the CCN counter.
The Kohler theory [
Brechtel and Kreidenweis
,2000a,
2000b] is used to determine the critical supersaturation as
afunctionofdrysize(activationdiameter
d
pc
), and a
differential mobility analyzer (DMA) is used to select a
D05S10
GHAN ET AL.: CCN RETRIEVAL EVALUATION
2of15
D05S10
variety of dry diameters which are then split to the CCN
counter and a TSI Model 3010 condensation particle counter.
Droplet density is calculated from Tang’s polynomials [
Tang
and Munkelwitz
, 1994]; the full Pitzer model [
Pitzer
,1973;
Pitzer and Mayorga
, 1973] is used to calculate the osmotic
coefficient; surface tension values from
Pruppacher and
Klett
[1997] are used for surface tension. The calibrated
activation diameters and the operating supersaturations for
Column 1 are given in the legend in Table 1.
[
13
] The supersaturations listed in Table 1 are quite high,
all above 2%. Such high supersaturations are typically
expected only for strong updrafts and clean conditions.
Moreover, unless particles have high insoluble contents,
most of the particles that can be activated at such high
supersaturations are usually quite small, with radius between
20 and 100 nm (although larger particles are also activated,
their number concentrations are usually much smaller than
those of particles smaller than 100 nm radius). These
particles have little impact on extinction or backscatter,
which are most sensitive to particles with radius between
100 and 600 nm [
Ghan and Collins
,2004].Thusextinction
and backscatter will be well correlated with CCN concen-
tration at such supersaturations only if the particles have high
insoluble contents or if the aerosol size distribution varies
little with altitude so that the concentration of particles
with radii between 20 and 100 nm varies in concert with
the concentration of particles with radii between 100 and
600 nm. These in situ CCN measurements therefore
provide a difficult test of the CCN retrieval scheme.
CCN concentration at lower supersaturations, which is
dominated by larger particles that produce stronger extinc-
tion and backscatter signatures, should be more accurately
retrieved by the scheme.
[
14
]Theaerosolsizedistributionwasmeasuredat72.5s
intervals at ambient relative humidity by a Caltech DMA
[
Wang et al.
, 2003]. Particles were dried to below 25% RH
prior to measurements. The sizes are centered at 23 diam-
eters ranging from 19 to 620 nm.
[
15
] Relative humidity is calculated from the ambient
temperature (calculated from Rosemount total temperature
and true airspeed) and dew point temperature (measured by
Edgetech EG&G chilled mirror).
[
16
]Aerosolscatteringatwavelengthsof450,550,
and 700 nm was measured every 8 s by a TSI model
3563 nephelometer for dry conditions. The data have been
corrected for nonidealities and corrected to ambient tem-
perature and pressure [
Anderson and Ogren
,1998].Aerosol
absorption at wavelengths of 467, 530, and 660 nm is
measured by a Particle Soot Absorption Photometer
(PSAP). The scattering data have been adjusted to the
PSAP wavelengths using the A
̊
ngstro
̈m exponent. Unreal-
istic data points due to instrument malfunction, adjustment
in flight, and data acquisition problems have been removed
from all data sets.
[
17
] The humidification factor at a wavelength of 540 nm
is approximated by
fRH
ðÞ¼
1
$
RH
lo
1
$
RH
g
ð
3
Þ
where
g
is determined from a fit to humidograph scattering
measurements at two different humidities:
g
¼
ln
s
hi
=
s
lo
ðÞ
ln 1
$
RH
lo
ðÞ
=
1
$
RH
hi
ðÞ
½&
ð
4
Þ
where
RH
lo
and
RH
hi
are typically 30% and 80%,
respectively.
2.1.2. Surface Measurements
[
18
]Atthesurface,bothinsituandremotesensing
measurements were collected at the ARM Climate Research
Facility (CRF) central site near Lamont Oklahoma. Remote
sensing measurements were provided by the CRF Raman
lidar (CARL) and the micropulse lidar (MPL). CARL
provides retrievals of both aerosol extinction and 180
°
backscatter at a wavelength of 355 nm [
Ferrare et al.
,
2001;
Turner et al.
,2002],andrelativehumidityisestimated
from the Raman lidar retrieval of absolute humidity and from
a retrieval of temperature from an Atmospheric Emitted
Radiance Interferometer (AERI). The Raman lidar retrievals
are performed every 10 min and interpolated to a vertical
resolution of 39 m. Comparisons of the CARL aerosol and
water vapor profiles with these additional data sets acquired
during the IOP as well as trends derived from long-term
CARL measurements revealed several issues with the
CARL data that adversely impacted retrievals of both
aerosol and water vapor profiles. The sensitivity of the
CARL had significantly declined since the end of 2001.
This loss of sensitivity has greatly impacted the quality of
the CARL aerosol backscattering and extinction profiles
derived since this time and during the Aerosol IOP. There-
fore the automated algorithms used to derive aerosol and
water vapor profiles from the CARL data were modified in
an attempt to reduce or remove these adverse effects. The
extensive modifications made to the CARL automated
Table 1.
Flight Summary With Operating Conditions for CCNC3 Column 1
Flight
Number
Date
Flight Begin Time,
UTC
Flight End Time,
UTC
Flight Length,
hours
Activation
Diameter, nm
Operating
Supersaturation, %
614May 1553
2019
4.4 15±0.8 2.8±0.2
714May 2124
2248
1.4 15±0.8 2.8±0.2
815May 1634
1909
2.6 15±0.8 2.8±0.2
917May 1402
1805
4.0 13±0.6 3.6±0.4
10
18 May
1543
1745
2.0
15 ± 0.8
2.8 ± 0.2
12
21 May
1551
1847
2.9
18 ± 0.9
2.1 ± 0.2
13
22 May
1325
1813
4.8
18 ± 0.9
2.1 ± 0.2
14
25 May
1852
2212
3.3
18 ± 0.9
2.1 ± 0.2
15
27 May
1420
1929
5.2
18 ± 0.9
2.1 ± 0.2
16
28 May
1824
2205
3.7
18 ± 0.9
2.1 ± 0.2
17
29 May
1411
1751
3.7
18 ± 0.9
2.1 ± 0.2
D05S10
GHAN ET AL.: CCN RETRIEVAL EVALUATION
3of15
D05S10
algorithms reduced but could not eliminate these adverse
effects [
Ferrare et al.
,2004,2006].Modifications
and upgrades performed during 2004 have dramatically
enhanced the sensitivity of CARL to surpass all previous
performance levels [
Turner and Goldsmith
,2005].
[
19
]TheMPLprovidesverticalprofilesofattenuated
180
°
backscatter every 30 s with 30 m vertical resolution.
Current processing yields 10-min averaged profiles of
aerosol extinction and 180
°
backscatter. However, in con-
trast to the Raman lidar technique, the MPL retrievals of
extinction and backscatter are not truly independent, but are
in fact related through an assumed constant extinction to
backscatter ratio. This assumption will not always be valid,
particularly in the case of separated aerosol layers. How-
ever, under well-mixed conditions the assumption typically
has reasonable local validity.
[
20
]
Schmid et al.
[2006] compare in detail the Raman
lidar and MPL retrievals of extinction with the in situ
measurements collected during this IOP. We therefore will
not compare the estimates here.
[
21
]Surfaceinsitumeasurementsconsistedofaerosol
humidification and CCN spectra. Surface humidification
measurements were provided by the Aerosol Observing
System (AOS) humidograph system at the ARM Climate
Research Facility [
Sheridan et al.
,2001].Thesame
parametric representation for
f
(
RH
), given by equation
(3), is used. For consistency with the aircraft measurements,
the humidification at 550 nm wavelength is used.
[
22
]CCNspectralmeasurementsatthesurfacewere
provided by two Desert Research Institute CCN spectrom-
eters [
Hudson
,1989],whichwereoperatedovertwo
different but overlapping supersaturation ranges. Concen-
trations at supersaturations between 0.03% and 1% are
considered most accurate. The CCN concentrations were
averaged over the period spanning the aircraft overflights.
The time means will be used to scale the vertical distribu-
tion profile provided by dry extinction.
2.2. Platforms and Flight Patterns
[
23
]Allairborneinsitumeasurementsusedinthisstudy
were collected from the CIRPAS Twin Otter, which has a
cruising speed of about 50 m s
$
1
.
[
24
]The2003ARMaerosolIOPhadavarietyofobjec-
tives, but most required coincident in situ and remote
sensing measurements of vertical profiles of aerosol and
its microphysical and radiative properties. Thus, although a
variety of aircraft flight patterns were employed on different
days, useful data for testing the CCN retrieval scheme were
collected on most flight days. Two flight patterns were most
common: the spiral and the level legs. Spirals were typically
performed with a 1 km diameter centered over the central
site, with ascent/descent speeds of 2–3 m s
$
1
.Levellegs
were typically 15–30 km in length crossing over the central
site, spaced every 500–1000 ft in altitude, with 180
°
turns
between legs. All flight patterns were designed to prevent
sampling of the aircraft’s own exhaust.
2.3. Sampling
[
25
]Criticaltothesuccessofthisstudyisthecollocation
of the aircraft and remote sensing measurements, both in
space and time. To ensure this, samples were discarded
unless all of the following conditions were met: (1) Aircraft
is within 30 km of SGP CF (36
°
N37
0
97
°
W30
0
), (2) lidar
samples at the same altitude as the aircraft and within 60 min
of aircraft flyover, (3) cloud-free (number concentration of
particles with diameter larger than 2.5
m
m<10cm
$
3
),
(4) relative humidity <95%, and (5) estimated error in
extinction retrieved from Raman lidar <50% of extinction.
The 30 km and 60 min proximity criterion were determined
from a compromise between the need to accumulate a
sufficient number of samples and the need for collocation
of in situ and remote sensing samples. We have found
results to be insensitive to the spatial and temporal range of
the sampling filter for distances between 5 and 30 km and
time differences between 15 and 60 min. To permit com-
parison on a point-by-point basis, for each day the aircraft
data were averaged over all the resulting samples within the
40 m thick lidar layers. This produces a single vertical
profile of all fields for each day. However, values for many
layers may not be defined, particularly for days without
spiral flight patterns.
[
26
]Toensureacomparableevaluationofdifferent
retrievals, all quantities were sampled only when all sam-
pling criteria were met. Although reliable in situ data were
discarded, we felt it was more important to ensure a
comparable evaluation than to have the most extensive
sampling for each retrieval.
3. Analysis
[
27
] To evaluate the performance of the CCN retrieval
scheme, we look at the data in three different ways. First we
look at vertical profiles of normalized quantities to identify
the vertical structure in the data and to see the relationships
between different quantities. Then we use the full scheme to
retrieve vertical profiles of CCN concentration. Finally, we
look at vertical profiles of quantities that are sensitive to the
size distribution and composition and hence can be used to
determine whether the assumptions of the retrieval are valid.
3.1. Vertical Structure and Relationships
[
28
] The objective of this study is to determine how well
the CCN retrieval scheme can determine the vertical profile
of CCN concentration below cloud, and to understand what
is limiting its performance. The measure of performance
will be the agreement with in situ measurements of CCN
concentration. To isolate errors due to differences between
the CCN instruments on the ground and in the aircraft, we
will compare vertical profiles of CCN concentration and dry
extinction normalized by values at the lowest altitude
available for all profiles. This still tests the validity of
equation (1), but removes errors due to the very different
designs and calibration procedures for the CCN instruments
[
Nenes et al.
,2001].Errorsinthemeasuredvariabilityof
CCN (the gain) are not removed by normalization.
[
29
] Given the anchor point of the retrieval scheme at the
surface, it is likely to perform well at altitudes near the
surface. Such agreement is neither useful nor indicative of
the performance of the retrieval scheme, because surface
measurements without the scaling by dry extinction should
provide close approximations to the CCN concentrations
near the surface. We therefore have extended our evaluation
up to 5 km above the surface. Although the performance of
the scheme is likely to be worse far from the surface, such
D05S10
GHAN ET AL.: CCN RETRIEVAL EVALUATION
4of15
D05S10