Introduction
The Amazon Basin can serve as a natural laboratory to study anthropogenic
effects on cloud microphysical and radiative properties. In its remote
parts, an absence of pollution similar to the preindustrial era still
prevails, while in other regions, cities and biomass burning emit high
numbers of aerosol particles into the atmosphere. This is especially
important during the dry season, when rainout is less frequent (Artaxo et
al., 2002; Kuhn et al., 2010; Martin et al., 2010). Under background
conditions, cloud condensation nuclei (CCN) consist mostly of secondary
organic aerosol (SOA) particles formed by the oxidation of volatile organic
compounds (VOCs), which condense and grow sufficiently to form CCN
(Pöschl et al., 2010). Anthropogenic emissions may enhance the oxidation
process, leading to increased SOA and CCN concentrations (Kanakidou et al.,
2000; Hallquist et al., 2009). Even though aerosol particles can be
scavenged by precipitation, nanoparticles produced in the upper troposphere
can be transported downwards by deep convective systems, approximately
reestablishing the surface aerosol concentration (Wang et al., 2016).
These processes illustrate the complex feedbacks between the vegetation, the
aerosols serving as CCN, and the clouds providing water to the vegetation.
There are, however, still plenty of open questions. The main difficulty in
this regard is the quantitative comparison of the aerosol effect to other
processes, given that the anthropogenic influences alter more than just
aerosol particle concentrations. Human activities associated with
urbanization and agriculture change the local landscape and the Earth's
surface properties, also altering the energy budget (Fisch et al., 2004) and
consequently the thermodynamic conditions for cloud formation. According to
Fisch et al. (2004), the convective boundary layer is deeper over pasture
during the dry season because of the increased sensible heat fluxes. This
effect results in greater cloud base heights with potentially stronger
updrafts, which should also be considered when analyzing the aerosol effect.
One possible way to compare different effects on cloud microphysical
properties is through a sensitivity calculation. It can provide specific
quantifications of aerosol and thermodynamic effects on cloud microphysical
quantities. One such sensitivity study,
in which the author calculates cloud droplet number concentration (Nd)
sensitivities to several aerosol and thermodynamic drivers, such as total
aerosol particle concentration (Na), updraft speed (w), and liquid water
content (LWC), was proposed by Feingold (2003). However, this analysis was limited to adiabatic stratocumulus
clouds for which collision–coalescence was not considered. Another modeling
study, in which they identified three
regimes that modulate the Nd sensitivity, was proposed by Reutter et al. (2009). The regimes are
aerosol-limited, updraft-limited, and the transition between them. The
authors highlight that the Nd dependence on Na and w may vary given
their relative magnitudes. This study is limited to cloud base, therefore
not addressing cloud evolution. The Reutter et al. (2009) study was extended
by Chang et al. (2015), who took into account the evolution of the systems
by considering the sensitivities on precipitation and ice phase, but was
relatively limited in terms of representativeness because of the use of a 2-D
model. Satellite studies (e.g., Bréon et al., 2002; Quaas et al., 2004; Bulgin et
al., 2008) have an intrinsic limitation given the characteristics of the
remote sensors. This kind of study usually deals with vertically integrated
quantities and frequently focuses on oceanic regions because of the
favorable surface contrast.
Locations where cloud profiles have been collected for different
HALO flights. Clouds formed over southern Amazonia and in the Manaus region
are subject to higher aerosol loadings due to the presence of the
deforestation arc and urban emissions. Clouds formed over the northern and
northwestern Amazon are driven by background conditions with low aerosol
concentration. During the GoAmazon2014/5 IOP2, maritime clouds were also
profiled on the Atlantic coast.
The main goal of this study is to expand the sensitivity calculations
usually found in the literature to include (1) aerosol and thermodynamic
effects on cloud droplet number concentration, size, and shape of the size
distribution; (2) comparison with the effect of cloud evolution, i.e.,
in-cloud processing; and (3) in situ observations of the less frequently
studied convective clouds over tropical continental regions. For this
purpose, we report on recent measurements over the Amazon rainforest during
the ACRIDICON-CHUVA campaign (Wendisch et al., 2016), in which a wide variety
(in terms of aerosol concentrations and thermodynamic conditions) of cloud
types were probed. We quantify the aerosol-induced changes in cloud
microphysical properties and compare them to the effects of updraft
intensity, which are related to thermodynamic properties, over different
regions in the Amazon. Both processes are analyzed with a focus on cloud
evolution. Our methodology should prove useful for better understanding
aerosol–cloud interactions over the Amazon, which is a region, as are the
tropics as a whole, with poor forecasting skill (Kidd et al., 2013).
Section 2 describes the experiment, its data, and the methods used for the analysis.
Results are presented in Sect. 3, followed by the conclusions in Sect. 4.
Methodology
Campaign and methodology
During the years 2014 and 2015, the GoAmazon2014/5 campaign took place in
the Amazon to improve our understanding regarding aerosol particles,
atmospheric chemistry, clouds, radiation, and their interactions (Martin et
al., 2016). In conjunction with the second Intensive Operations Period
(IOP2) of this experiment, the ACRIDICON-CHUVA campaign took place during
September–October 2014 (Wendisch et al., 2016). It included 14 research
flights with the German HALO (High Altitude and Long Range Research
Aircraft). A previous campaign dedicated to study aerosol–cloud interactions
took place in the Amazon in 2002 (LBA-SMOCC; Andreae et al., 2004), but it
had been relatively limited in terms of range and ceiling of the aircraft
measurements. The high endurance of the HALO aircraft, which carried
sophisticated microphysical, aerosol, and solar radiation instrumentation,
allowed for long-range flights from remote areas in the northern Amazon to the
deforestation arc in the south and to the Atlantic coast in the east
(Fig. 1). The flights were planned to cover five different mission types
focusing on different cloud, aerosol, chemistry, and radiation processes (see
Wendisch et al., 2016, for details). The flights were numbered
chronologically as ACXX, in which XX varies from 07 to 20. For this study, the
cloud profiling missions are of particular interest and their respective
locations are shown in Fig. 1. In this study, we take advantage of HALO's
capabilities of comparing different types of clouds formed over different
Amazonian regions, focusing on their warm microphysics. In addition to the
HALO measurements, ground-based equipment was also operated in and near
Manaus city (Machado et al., 2014; Martin et al., 2016).
Definition of the symbols and abbreviations discussed in this study.
Symbol/abbreviation
Definition
DSD
Droplet size distribution
Nd
Cloud droplet number concentration (cm-3)
Deff
Effective diameter (µm)
LWC
Liquid water content (gm-3)
ε
Relative dispersion parameter (dimensionless)
Λ
Gamma DSD curvature parameter (µm-1)
μ
Gamma DSD shape parameter (dimensionless)
N0
Gamma DSD intercept parameter (cm-3 µm-1-µ)
Mp
DSD moment of order p
Da
Mean diameter obtained as M2/M1 (µm)
γ
Parameter associated with DSD shape, given by Λε2Da (µm-2)
Na
Aerosol number concentration (cm-3)
w
Updraft speed (ms-1)
H
Altitude above cloud base (m)
CCN
Cloud condensation nuclei number concentration (cm-3)
The results shown here were obtained from the measurements of four different
instruments (for a list of all HALO instruments, see Wendisch et al., 2016),
covering aerosol, cloud, and meteorological properties. We will focus on
aerosol and CCN number concentrations, cloud droplet size distributions
(DSDs), and updraft speed. The instruments are briefly described below.
CCP
For the cloud droplet size distribution measurements, a modified cloud
combination probe (CCP, manufactured by Droplet Measurement Technologies,
Inc., Boulder, CO, USA) was adopted on HALO covering an overall size diameter
range from 3 to 950 µm. The probe consists of two separate
instruments, the CDP (cloud droplet probe; Lance et al., 2010; Molleker et
al., 2014) and a grayscale optical array imaging probe (CIPgs, cloud imaging
probe; Korolev, 2007). By means of a two-dimensional shadow cast technique,
the CIPgs detects cloud particles with size diameters ranging from 15 to
2000 µm. The in-house-developed analysis algorithm from the Max
Planck Institute for Chemistry and the Institute for Atmospheric
Physics
in Mainz sizes and sorts the recorded images into bins of roughly
15 µm width depending on particle shapes and dimensions. The CDP
is an optical particle counter (OPC) that detects scattered laser light (in
forward direction) arising from individual particles passing through the
illuminated optical sample area (Lance et al., 2010; Molleker et al., 2014).
The optical sample area has a cross section of 0.2 mm2 (±15 %)
perpendicular to the flight direction. The CDP detects particles with sizes
from 3 to 50 µm and classifies these into size histograms of bin
widths between 1 and 2 µm. In addition to size histograms recorded
at 1 Hz frequency, the CDP stores single-particle data (signal amplitude and
microsecond-resolved detection time) of continuous intervals with up to 256
particles every second. This feature can be used to assess the spatial
distribution of the droplets in case of multimodal size distributions
(Klingebiel et al., 2015). The main uncertainties for the CCP size
distributions are due to the uncertainty of the sample area (and thus the
scanned air volume), as well as counting statistics. We applied a filter to
eliminate DSDs with concentrations lower than 1 cm-3 for D<50 µm or lower than 0.1 cm-3 for D>50 µm.
AMETYST-CPC
The aerosol concentrations used in this study refer to the total
concentration of particles measured with a butanol-based condensation
particle counter (CPC). Four CPCs were deployed on HALO as part of the new
basic aerosol instrument package for HALO named AMETYST (Aerosol MeasuremenT
sYSTem) and which also includes two Grimm 1.129 OPCs, a two-channel thermal denuder operated at 250 ∘C, a
Radiance Research three-wavelength PSAP (particle soot absorption photometer),
and optionally two DMAs (differential mobility analyzers). AMETYST is
operated behind the HALO sub-micrometer aerosol inlet (HASI). The CPCs are
Grimm 5.410 models, operating at two different flow rates. The CPC internal
butanol saturation setting is user-selectable to vary minimum detectable
particle sizes. Data used in this study were obtained from 0.6 L min-1 5.410
CPC set to a nominal lower cutoff size of 10 nm. Concentrations reported
are normalized to standard temperature and pressure conditions. Original
data are recorded at 1 Hz temporal resolution. In-cloud data at altitudes
below 9 km were removed from the dataset based on cloud probe data (here
CAS-DPOL instrument of DLR) to exclude apparent sampling artifacts of the
inlet in the presence of liquid droplets in clouds.
CCNC
A cloud condensation nuclei counter (CCNC) was used to obtain CCN number
concentrations. The instrument has two columns with a continuous-flow
longitudinal thermal gradient in which the aerosol particles are subject to
controlled supersaturation (S) conditions. When particles travel
longitudinally in the center of each column, they grow by water condensation
(depending on their physical and chemical compositions) and are counted as
CCN if they reach 1 µm in size (1 Hz sampling rate). The CCNC is manufactured
by Droplet Measurement Technologies (DMT) – Roberts and Nenes (2005).
Calibrations were performed between flights following Rose et al. (2008). At
one column, S was set to be relatively constant at S≈0.55 %,
while the other was subject to 100 s stepping variations between 0.2 and
0.55 %.
BAHAMAS
Vertical wind speeds were obtained from the Basic Halo Measurement and Sensor
System (BAHAMAS) sensor installed at the nose of the aircraft (Wendisch et
al., 2016). The 3-D wind measurements were calibrated following Mallaun et
al. (2015), resulting in an uncertainty of 0.3 m s-1 for the
horizontal components and 0.2 m s-1 for the vertical components.
Sensitivity calculation
Several earlier studies calculated cloud sensitivity to aerosols and/or
updrafts (Feingold, 2003; McFiggans et al., 2006; Kay and Wood, 2008; Reutter
et al., 2009; Sorooshian et al., 2009; Kardys et al., 2012; Chang et al., 2015),
but they were usually limited in scope by not individually considering the factors that
contribute to the cloud microphysics. This study aims to expand
the sensitivity methodology by concurrently considering cloud evolution,
updraft speed, and aerosol effects on clouds and by taking advantage of the
comprehensive ACRIDICON-CHUVA dataset to represent different kinds of clouds
and thermodynamic conditions. As pointed out by Seinfeld et al. (2016),
major field campaigns provide a key opportunity for improving our knowledge
of the aerosol–cloud–climate interactions, further motivating the results to
be presented here.
Three factors will be considered as the main drivers of cloud microphysical
properties, each representative at least partially of thermodynamic and
aerosol conditions and cloud evolution. Those factors, and other parameters
discussed in this study, are defined in Table 1. For the aerosol
characterization, we will use averaged concentrations measured by the
AMETYST CPC (referred here as Na – see Tables 1 and 2) at the cloud base
level. This level was obtained from the CCP–CDP measurements as the lowest
level at which the LWC is higher than 0.01 g m-1. As the profiles always
started with cloud base penetrations, this ensures a precise estimation of
cloud base altitude. Table 2 also shows that CCN
concentrations were proportional to Na for the chosen instrument
supersaturation. A linear fit between Na (as the dependent variable) and
CCN (as the independent variable) results in R2=0.96, with angular
and linear coefficients equal to 1.57 and 243 cm-3, respectively. For
the purposes of the sensitivity calculations, we will use Na instead of
CCN concentrations because they are not dependent on instrument or cloud
supersaturations. It is known that polluted clouds tend to have lower
supersaturations given the enhanced condensation. Therefore, the use of
constant-supersaturation CCN concentrations does not provide a common
benchmark between the clouds probed here. Conversely, it is difficult to
obtain the supersaturation within the clouds and the consequent CCN
concentration modulation. In that regard, Na proved to be most adequate
for providing a framework to compare polluted and clean clouds. The
sensitivity calculation (see below) uses derivatives of the concentrations; thus, the choice of Na or CCN should have no significant impact on the
results to be presented here (because of the linear correlation between CCN
and Na). The most pristine clouds are observed near the coast (AC19),
followed by the ones measured over the forest. The flights AC7, AC12, and
AC13 each showed increasing aerosol concentrations as the flights moved
towards the southern Amazon. For the flights closer to Manaus city, the
aerosol loading of the clouds depends on localized aspects such as
small-scale biomass burning and the pollution plume from urban and/or industrial
activities (Cecchini et al., 2016). In this study, we will focus on aerosol
number concentrations, and their chemical composition will not be addressed.
Na and CCN at cloud base for each flight considered in
this study. *CCN concentrations for flight AC20 showed pronounced scaling
with S. The value shown is for the measurements in which S>0.52 %. This value is closer to the maximum droplet concentration measured
at the base of the clouds (= 1422 cm-3).
Flight
Na (cm-3)
CCN (cm-3)
S (%)
AC19
465
119
0.52
AC18
744
408
0.50
AC9
821
372
0.51
AC20
2331
1155*
0.55
AC7
2498
1579
0.50
AC11
2691
1297
0.49
AC12
3057
2017
0.44
AC13
4093
2263
0.44
The second factor that affects cloud microphysics is the updraft intensity
(w). It, along with the aerosol population, defines the supersaturation
inside the clouds and thus affects the droplets' condensational growth. The
intensity of the updrafts depends both on meteorological conditions (e.g.,
temperature and humidity profiles) and on the latent heat release of
condensing water. Aerosols may indirectly affect the amount of latent heat
released (smaller droplets in polluted clouds have a favorable area-to-volume
ratio), but the speed of the ascending air can be understood as a response
to the thermodynamic conditions in the clouds. Therefore, w can be used as a
benchmark to compare different clouds subject to similar
thermodynamic conditions relevant to cloud microphysics.
Lastly, it is important to have an estimate of how cloud microphysical
properties evolve throughout the system evolution, but more importantly, how to
detect similar cloud stages over the different flights for comparison. The
HALO cloud profiling missions were planned to capture growing convective
clouds in the different Amazonian regions. The aircraft penetrated the
systems first at cloud base and then at ascending altitudes in the cloud
tops of the growing convective elements. This strategy allows the use of
altitude above cloud base (herein referred as H, in meters, also known as
cloud depth) as a proxy for cloud evolution. Measurements at higher altitudes
reflect later stages of the cloud life cycle as the systems develop upward.
We use the derivatives of the microphysical properties with respect to H,
which can be understood as variations during the cloud evolution. This will
put the sensitivities to Na and w into perspective, highlighting the
importance of detecting cloud stage. It could be argued that the ratio H/w would
be a more direct estimate of the cloud lifetime, given that it is the time
that it took for the cloud to reach H. However, this approach would need
prescribed w profiles below each measurement, which is not feasible in this
study given that different clouds can be measured in the same profiling
mission. Additionally, there is high w variability horizontally between the
cloud edges and cores, adding extra complexity. Therefore, we will use H as
the proxy for cloud evolution even though it does not represent cloud
lifetime directly (i.e., does not have units of time). The profiling strategy
of measuring growing convective clouds favors this interpretation.
The sensitivities are calculated as partial derivatives on a natural log
scale. In this way, they are normalized for quantitative comparison. Based
on the terminology in the literature (e.g., Feingold, 2003; Chang et al.,
2015), we consider the sensitivities as follows:
SYXi=∂lnY∂lnXiXj,Xk,
in which X is the independent variable, i.e., w, Na, and H, and Y is the cloud
microphysical property of interest. For the sensitivity calculation, we will
focus firstly (Sect. 3.2) on cloud droplet number concentration
(Nd) and effective droplet diameter (Deff) of cloud DSD with
D<50 µm. In Sect. 3.3 we also consider the sensitivities in
LWC, relative dispersion (ε), and curvature parameter (Λ; see respective text for details). The three factors chosen for X in this
study are not necessarily independent; therefore, in order to follow the
partial derivative formalism, we include the criteria expressed by the
vertical line (Eq. 1). The subscript in X identifies the different
independent variables considered. This notation means that two independent
variables remain constant while the sensitivity to the third is being
calculated. As an example, the sensitivity of Nd to Na, w, and H can
be expressed as
SNdNa=∂lnNd∂lnNaw,H,SNdw=∂lnNd∂lnwNa,H,SNdH=∂lnNd∂lnHw,Na.
Equation (2) recognizes that several parameters can affect Nd, and they
should be analyzed individually. Other sensitivities, such as
SDeffNa or SDeff(w), are
obtained analogously.
As it is not feasible to analyze the sensitivities under exactly constant
conditions as in Eq. (2), we decided to use Na, w, and H intervals
instead. These quantities were binned into {0, 500, 1000, 3000, 4500 cm-3},
{0, 0.5, 1, 2, 4, 8 m s-1}, and {0, 200,
500, 950, 1625, 2637, 4156 m}, respectively. In this way,
there are 4, 5, and 6 Na, w, and H intervals, respectively. The values of
the bins were chosen in order to maximize the number of data in each
interval, which required growing spacing in w and H. We use constant Na
values for each profile and the respective bins effectively group different
flights according to the pollution level. Note that flight AC19 falls in the
first interval; flights AC9 and AC18 in the second; AC7, AC11, and AC20 in
the third; and AC12 and AC13 in the fourth (see Table 2). We then produce
4-by-5-by-6 matrices containing averaged Nd and Deff values for the
combined intervals, covering all variations possible. The sensitivities are calculated as follows: 1) firstly, we choose one independent variable for the sensitivity calculation, which corresponds to one dimension of the matrix; 2) the other two dimensions are then fixed and we obtain the individual curve representing the relation between Nd or Deff and the independent variable of choice; 3) the sensitivity is calculated as the linear fit of this curve on the natural logarithm scale (in order to be normalized); 4) the process is repeated for every combination of dependent and independent variables possible.
Different bin configurations were tested and the results proved to be
relatively insensitive to the bin number and width.
Results
Cloud droplet size distributions related to different aerosol and
thermodynamic conditions
The first qualitative indication of the effect of Na, w, and H on cloud
microphysical properties can be seen in Fig. 2. This figure shows DSDs
(dN/ dlogD in the vertical axis) grouped into four categories according to
the aerosol concentration (Na) at cloud base: (1) maritime clouds, with
Na≤500 cm-3; (2) clouds under Amazonian background
conditions, with 500 cm-3 < Na≤1000 cm-3;
(3) moderately polluted clouds, with 1000 cm-3<Na≤3000 cm-3;
and (4) polluted clouds, with Na>3000 cm-3.
Solid lines in Fig. 2 represent DSDs for neutral vertical speed (-1 m s-1≤W≤ 1 m s-1), while the DSDs with dashed and
dotted–dashed lines indicate the updraft and downdraft regions, respectively
(W>1 m s-1; note that we use W to
differentiate from w, which refers only to the updraft portion). They
represent averages for all profiles matching the aerosol intervals chosen
(one maritime, two Amazonian background conditions, three moderately
polluted, and two polluted). Individual profiles can be found in the
Supplement (Figs. S1–S4). In general, all profiles show droplet
growth with altitude as they continually go through the condensational and
collision–coalescence processes. The enhanced DSD widening with altitude
presented in Fig. 2a, b suggests relative predominance of
collision–coalescence. Those observations also support the choice of H as
a proxy for cloud lifetime.
Droplet size distributions as functions of altitude above cloud base,
aerosol particle number concentration, and vertical wind speed (W). Four
1000 m thick layers are considered in the vertical, and the legends in the
graphs show the respective upper limit of each one. Solid lines represent
averaged DSDs for -1 m s-1≤W≤1 m s-1, i.e., for
relatively neutral vertical movements. Dashed lines represent averaged DSDs
for the updraft regions where W > 1 m s-1, and
dotted–dashed lines represent the downdrafts
(W < -1 m s-1).
From Fig. 2, it is evident that aerosols and updrafts affect the droplet
size distribution and its evolution in different ways and magnitudes. Clouds
that develop under similar aerosol conditions tend to have similar DSDs not
only at cloud base but also higher in the warm layer. The individual
profiles shown in Figs. S1–S4 confirm the pattern that is evident in
Fig. 2. Conversely, the updraft effect is limited to modulations of
the DSDs, especially in the D<10 µm range. Note that DSDs
subject to similar w values can be widely different depending on the respective
pollution. The resulting vertical evolution of the clouds is dependent on
the Na value, being more pronounced the cleaner the clouds are. We only
observed significant concentrations of precipitation-sized droplets (e.g.,
> 100 µm) for Na<3000 cm-3.
The main motivation for calculating sensitivities is to quantify and compare
the role of Na, w, and H in the formation and evolution of the DSDs as seen
in Fig. 2. In this way, it will be possible to check the magnitudes of the
effects of aerosols, thermodynamics (as seen from the updrafts), and cloud
evolution in the determination of the warm-phase characteristics. Note,
however, that we are focusing on only one portion of the updraft effects,
i.e., the condensation and collision–coalescence effects. For instance,
Heymsfield et al. (2009) showed that small droplets carried up by updrafts
can significantly participate in the cold processes of the clouds, which are
not addressed here. This study considers the first stage of the cumulus
clouds just before the formation of ice particles. Regardless, Fig. 2
shows evidence that all three chosen independent variables have specific roles in
determining cloud DSD characteristics. Together they explain most of the
warm-phase properties.
Comparing the main drivers of bulk microphysical properties of Amazonian
clouds
For quantitative comparisons, it is interesting to consider bulk DSD
properties such as Nd and Deff instead of the whole DSD as in
Fig. 2. We will quantify the influence of Na, w, and H in these properties as a
means of understanding the effects on the overall DSD. This analysis will be
complemented by the study of the DSD shape in the next section. By comparing the
sensitivities of cloud droplet concentration and size to Na and w, it is
possible to make a comparison that represents, at least partially, the
contrasts between the importance of aerosols and thermodynamics in cloud
characteristics. A significant portion of the previous work in this field
was dedicated to understanding the processes that lead to the observed
Nd. Twomey (1959) provides theoretical considerations of Na and w
effects on the supersaturation, which ultimately defines Nd for a given
CCN spectra. More recent studies report on observations and modeling efforts
to portray these processes in different regions of the world, calculating
cloud sensitivities to both updraft speed and aerosol conditions. By
analyzing aerosol and updraft conditions around the globe, Sullivan et al. (2016)
note that w can be the primary driver of Nd in some regions.
Reutter et al. (2009), using an adiabatic cloud model, argue that Nd
sensitivities to aerosol concentrations and w can vary depending on their
relative magnitudes. Adiabatic clouds are not highly sensitive to w (at cloud
base) when CCN concentrations are low and vice versa. Some studies also
consider sensitivities in droplet size, such as Feingold (2003). However,
cloud evolution is rarely put into perspective representing a limitation of
previous studies. In the following, we will show our extended calculations
of the sensitivities, in which we consider the effects of aerosols, updraft
speed, and H on Nd and Deff.
Based on Eq. (2), it is evident that there exist several values for each
sensitivity. As an example, SNdNa has different
values depending on the chosen pair of {w, H}.
However, given the nature of in situ measurements, individual
SNdNa values are associated with reduced sample
sizes and, therefore, compromise the statistical confidence. In this case,
we present averaged values and the respective standard deviation for all
{w, H} pairs considered, applying the same
calculation to the other sensitivities as well. The intervals chosen for
Na, w, and H imply that those averages are representative of the lower
∼ 4 km of the clouds, with updrafts up to 8 m s-1 and aerosol
concentrations ranging from 500 to 4500 cm-3.
The results of the Nd and Deff averaged sensitivities (Table 3)
reflect the patterns observed in Fig. 2. The effective diameter shows
strong association with the aerosol concentration and H while being almost
independent of w. Specifically regarding Na, the sensitivities calculated
represent the first step in the parameterization of the aerosol indirect
effect for climate models, i.e., its relation to cloud microphysical
properties. Multiple studies have focused on this issue from several
observational setups such as satellite or surface remote sensing and in situ
measurements. Pandithurai et al. (2012) provide a compilation of this type
of calculation (see their Table 2), showing a high variability among the
sources. According to Schmidt et al. (2015), the differences are due to not
only the measurement setup but also to the region (ocean or land), the types of
clouds, and the differences in the methodologies. Remote sensing techniques
often retrieve vertically integrated quantities at relatively rough
horizontal resolution, which can smooth the results, meaning lower
sensitivities. Conversely, in situ airborne measurements are closer
to the process scale and may result in more accurate estimates of the
aerosol effect (Werner et al., 2014). However, the studies reviewed in
Pandithurai et al. (2012) and Schmidt et al. (2015) are mostly for stratus
or cumulus clouds over ocean. Additionally, measurements of w were either not
available or were not differentiated in most of the previous analyses, while
the results are often integrated in altitude or limited to a specific cloud
layer (e.g., cloud top in satellite retrievals). Our study focuses on
tropical convection over the Amazon and takes into account both the updraft
speed and altitude of the measurements.
The values of the sensitivities with regard to Na presented here are
among the highest reported in literature. They are not far from the
theoretical limit of SNdNa=1
(Nd≤Na) and SDeffNa=-0.33, which is quite common for in situ airborne studies
(Werner et al., 2014). The limit for Deff is an approximation and stems
from the relation (if LWC is held constant) Deff∝LWCNd1/3 (e.g., Martin et al., 1994). Given
the precautions taken here to isolate the aerosol effects, these values show
that Amazonian clouds are highly sensitive to pollution. Human-emitted
particles affect not only the DSDs close to cloud base but also over at
least the lower 4 km of the warm-phase domain. The meteorological and cloud
morphology conditions in the Amazon also seem to enable the high sensitivity
values found. A previous study by Vogelmann et al. (2012) found relatively
invariant Nd as a function of Na. Beyond instrumental and
methodological differences, this study also focused on shallow (200 to 500 m thickness)
broken clouds with weak updrafts over Oklahoma. This type of
cloud favors the entrainment mixing feedback, in which polluted clouds tend to
have lower LWC because of enhanced droplet evaporation. The differences between
the results shown here and the study of Vogelmann et al. (2012) suggest that
the entrainment mixing process is not dominant over the Amazon. Possible
reasons include abundant water vapor, thicker clouds, stronger convection
and updrafts, and low vertical wind shear. High humidity of the surrounding
air induces weaker LWC and Nd depletion by the entrainment mixing process
(see, for instance, Korolev et al., 2016) because of slower evaporation.
Stronger convection induces deeper clouds that have a relatively low
area-to-volume ratio as compared to the clouds reported in Vogelmann et al. (2012).
Therefore, the entrainment at cloud edges are not as dominant. Low
area-to-volume ratios are also favored by the weak vertical wind shear
typical of tropical regions. This mechanism was studied in Fan et al. (2009), who concluded that convection invigoration is favored under low vertical
wind shear conditions, while the opposite happens with high vertical wind
shear.
The sensitivities to the updraft speed have a distinct behavior when
compared to the aerosol effect. Not only does it show lower values overall
but it also shows different behaviors for Nd and Deff. It shows that even
strong updrafts are not able to significantly increase the effective droplet
size by enhancing condensation. In fact, this sensitivity oscillates around
zero with slightly negative and positive values (see Table S2) and with
relatively low R2 values. This finding is similar to what Berg et al. (2011)
observed in Oklahoma City. Close to cloud base, they found a significant
relation between Nd and w and a low correlation between Deff and
w. Here we show that this feature is not limited to cloud base but persists
with altitude on average. Feingold (2003), using an adiabatic cloud
parcel model, found a negative value for SDeffw, with a higher absolute value for polluted clouds. The result could
be explained by activation of smaller aerosol particles with increasing
updraft speed, leading to higher concentrations of small droplets that
skewed the mean diameter to lower values. Although we observed slightly
negative sensitivity for highly polluted clouds at their base (Table S2),
our measurements show that the overall averaged Deff is nearly
independent of w for the Amazonian clouds.
Freud et al. (2011) and Freud and Rosenfeld (2012) showed similar
observations in the Amazon, India, California, and Israel. They provide
theoretical formulations that support some of those observations. These
authors showed that the vertical evolution of Deff behaves almost
adiabatically because of the predominance of inhomogeneous mixing in
convective clouds. In this way, droplet effective size can be obtained from
cloud base Nd, pressure, and temperature. In fact, this is the
framework for a new technique developed to obtain CCN retrievals from
satellites (Rosenfeld et al., 2016). Our study provides a new look at those
observations and theoretical considerations by specifically quantifying,
without any adiabatic assumption, each process with our formulation of
sensitivity.
Nd and Deff averaged sensitivities to
Na, w, and H. Standard deviations are also shown. R2
values are averages of the individual fits. The total variations for
Na, w, and H are 500 to 4500 cm-3, 0 to 8 m s-1,
and 0 to 4156 m, respectively. Intervals grow logarithmically (or close to)
for w and H.
SNd‾
SDeff‾
Na
0.84 ± 0.21
-0.25 ± 0.074
R2=0.91
R2=0.89
w
0.43 ± 0.28
0.028 ± 0.058
R2=0.81
R2=0.46
H
-0.13 ± 0.16
0.28 ± 0.058
R2=0.38
R2=0.93
Comparisons of the sensitivities to w and Na can be used to infer the
roles of the aerosols and thermodynamic conditions in the DSD
characteristics. Not only do the aerosols primarily determine the size of
the droplets but they also have the biggest impact on the number
concentration, high variability in SNdw
notwithstanding. This result shows that in terms of the warm layer, aerosols
play a primary role in determining DSD characteristics.
The sensitivities to H are calculated in order to put the aerosol and
updraft effects into perspective regarding cloud evolution. This calculation
shows that, on average, droplet growth with cloud evolution is comparable in
absolute value and is opposite to the aerosol effect. For this reason, studies
should take into account the altitude of the measurements. Polluted
Amazonian clouds show slower droplet growth with altitude (Cecchini et al.,
2016) and SDeffH values may vary with Na. With lower
SDeffH, SDeffNa values possibly
increase with altitude. The most important factor evident in Table 3 for
Deff is that it shows strong relations with Na and H, while being
independent of w. This result is of great value for parameterizations or
other analyses of cloud droplet size.
Whereas Deff shows a clear relation to Na and H, being relatively
constant at fixed altitude, Nd displays a highly variable behavior. The
averaged SNdH has a slightly negative value with high
standard deviation. There can be either droplet depletion or production with
altitude, but the former prevails on average. New droplet activation should
be expected in polluted clouds, where not all aerosols are activated at
cloud base. In fact, Table S6 shows that SNdH is
positive for most polluted clouds probed when updraft speeds are
> 0.5 m s-1, although R2 values are quite low. Droplet
depletion with altitude can be a result of evaporation and/or collection
growth. Cecchini et al. (2016) showed that Amazonian background clouds
present rather effective collision–coalescence growth, which would suggest a
negative SNdH for those clouds. This mechanism is
difficult to observe in the present study, with relatively low R2 in
the individual SNdH (Table S6). Overall, the highly
variable relation between Nd and H suggests that droplet concentration is
not closely tied to altitude above cloud base, as it is the case for
Deff. Conversely, droplet concentration has significant
horizontal variation given different mixture and w conditions, while the
effective diameter remains similar at constant altitude levels.
Effects on DSD shape and relation between sensitivities
The use of a parametric function to represent the DSDs can be of interest in
order to understand the sensitivities in the overall shape of the DSDs. One
function widely adopted in many applications and especially in models (Khain
et al., 2015) is the gamma function. One of the forms of the gamma function
represents the DSDs as
ND=N0Dμexp(-ΛD),
in which N0, μ, and Λ are intercept, shape, and curvature
parameters, respectively. The advantage of using this function is that it
can be analytically integrated, providing relatively simple equations for
the DSD parameters. Nd, Deff, and LWC can be calculated from the
moments of the gamma DSD (units are cm-3, µm, and g m-3,
respectively):
Nd=M0Deff=M3M2LWC=10-9ρwπ6M3,
in which ρw is the density of liquid water (considered as 1000 kg m-3 here)
and Mp is the pth moment of the DSD, given by
Mp=∫0∞DpNDdD=N0Γ(μ+p+1)Λμ+p+1.
By substituting Eq. (7) into (5) it is possible to write Deff as a function
of Nd and LWC:
Deff=1096πρwγLWCNd,
in which γ is a parameter that depends on the DSD shape and droplet
size. It can be written as a function of ε, defined as the
ratio between the DSD standard deviation and its average, which is much more
common in the literature (e.g., Liu and Daum, 2002; Tas et al., 2015):
γ=Λ2(μ+2)(μ+1)=Λε2Da.
Da is the mean diameter resulting from the ratio between the second- and
first-order moments. By substituting Eq. (9) into (8), applying the natural
logarithm and the partial derivative to lnXi (as in Eq. 1), it is
possible to write
∂lnNd∂lnXi=∂lnΛ∂lnXi+2∂lnε∂lnXi+∂lnLWC∂lnXi-2∂lnDeff∂lnXi,
which is an explicit representation of the relation between the
sensitivities. Note that ∂lnDeff∂lnXi=∂lnDa∂lnXi because of the
similarities in the equations of both diameters. The first two terms on the
right-hand side in Eq. (10) represent the DSD shape, in which Λ is
related to the curvature of the gamma curve and ε is the
relative dispersion around the DSD mean geometric diameter. Lower (higher)
Λ and higher (lower) ε values are associated with broader
(narrower) DSDs. Equation (10) shows that, in order to compare the
sensitivities in Nd, Deff, and LWC, the DSD shape has to be taken into
account.
Several aspects of the aerosol–cloud–interaction physics can be illustrated
by Eq. (10). The Twomey effect states that clouds subject to high aerosol
concentrations have enhanced albedo because of the more numerous droplets
with increasing aerosol loading (Twomey, 1974). This effect is defined when
comparing clouds with the same LWC. Translating it into Eq. (10) (with
Xi=Na), it means that the LWC derivative is neglected, which defines a
relation between droplet concentration, effective diameter, and DSD shape.
By comparing to the expression
SDeff(Na)‾=-13SNd(Na)‾
often found in the literature, we can conclude that the value of the
sensitivity of Nd to Na is offset by some effect on DSD shape. In
other words, two-thirds of the Nd sensitivity is allocated into DSD
narrowing or broadening, while the remainder effectively alters
Deff.
The effects of enhanced aerosol concentrations on the DSD shape are of great
interest to the climate change community, given that they contribute to the
aerosol indirect effect. Liu and Daum (2002) report that pollution, in
addition to
lowering droplet size, tends to broaden the DSDs, which would result in
weaker cooling forcing compared to previous calculations. They show that the
previous estimations of the aerosol indirect effect considered a fixed
ε, possibly overestimating the cooling forcing. Recently, Xie
et al. (2017) reports improved model comparisons with satellites when better
estimating the relative dispersion. Therefore, it is important to understand
the relation between ε (and Λ) and not only aerosols
but also updraft speed and height above cloud base. The overall averages
presented in Table 4 show that the DSD curvature (Λ) is sensitive
to Na and H, but the values are rather small for ε. This
results from the not-so-simple relation between DSD shape and Na, w, and
H. Figure 3 shows the variations in the sensitivities of Λ and ε with Na and H (no significant variations were found for w), in which it is
clear that the overall averages in Table 4 must be analyzed with caution for
DSD shape. The ε sensitivities have a significantly different
behavior for clean and polluted clouds and also change sign along H. Both
features result in a low overall average as presented in Table 4, but this
does not mean that the ε sensitivity is negligible. Instead, a
more detailed analysis should be considered.
Same as Table 3, but for the sensitivities in Λ,
ε, and LWC.
SΛ‾
Sε‾
SLWC‾
Na
0.23 ± 0.34
-0.015 ± 0.16
0.078 ± 0.34
R2=0.64
R2=0.54
R2=0.34
w
0.046 ± 0.17
0.039 ± 0.094
0.49 ± 0.34
R2=0.49
R2=0.46
R2=0.77
H
-0.43 ± 0.32
0.094 ± 0.16
0.67 ± 0.21
R2=0.64
R2=0.42
R2=0.76
Variations in the sensitivities of Λ and ε
with (a) Na and (b) H. Note that the
sensitivities of ε are multiplied by 2 in order to be consistent
with Eq. (10). The curves are averaged over all values of the third dependent
variable. For instance, the curve SΛ(w) in
(a) is averaged over all H values. Blue curves represent the sum
of the sensitivities of Λ and ε, equivalent to the
first two terms on the right-hand side in Eq. (10).
The sensitivities in Λ and ε usually have opposite
signs, given their relation to DSD shape – broader DSDs tend to have higher
ε values but lower Λ values. Nevertheless, their
sensitivities to
Na and H are conceptually similar and illustrates interesting processes.
Figure 3a shows that the DSD shape variation with altitude is much more
pronounced in cleaner clouds, which is a result of a strong
collision–coalescence process. The higher the aerosol concentration, the
lower the sensitivity of ε to H. For the most polluted clouds
measured by HALO, the relative dispersion parameter is almost insensitive to
H, meaning that it does not change much as the cloud grows. There is,
however, still some effect on the DSD curvature, making the summation of the
first two terms on the right-hand side in Eq. (10) nonnegative in this
case (see solid blue line in Fig. 3a). For the sensitivities of Λ
and ε to w, the same summation (dashed line in Fig. 3a) is
basically null, meaning that these two terms have no contribution in
Eq. (10). Nevertheless, stronger updrafts tend to produce narrower DSDs
in the maritime clouds in which the aerosol population is limited in terms of
number concentration and particle type–chemistry.
The patterns along H of the DSD shape sensitivities (Fig. 3b) pose an
interesting question for the parameterization of the aerosol indirect effect
in Amazonian clouds. There are significant changes in ε
tendencies as the clouds evolve. Note that aerosols induce broader DSDs up
to H∼500 m, but the opposite happens above that point. In fact, for
our higher-altitude bin (2637 m < H≤4156 m), the average
ε is lowest for the most polluted clouds (=0.28, while
clouds over the forest and Atlantic Ocean show values of 0.32 and 0.42,
respectively). In other words, the effect of broader DSDs under polluted
conditions may not directly apply for convective clouds over the Amazon,
where growth processes in the cloud can significantly change this pattern.
This highlights the need to take cloud evolution into account and there is
no direct relation between aerosols and cloud relative dispersion in the
warm phase of Amazonian clouds. For satellite retrievals, in which integrated
quantities are of likely interest, the relative dispersion will depend not
only on the aerosol concentration but also on cloud depth and life cycle
stage.
Regarding the sensitivities to w, Fig. 3 and Tables 3 and 4 show that
updraft speed has little impact on DSD shape or droplet size. The result in
terms of Eq. (10) is the equality between the sensitivities in Nd and
LWC, which is generally the case when we compare the averages shown in Tables 3
and 4. In other words, the updraft effect is limited to increases in the
droplet concentration and water content, modulating both Nd and LWC in the
same proportion. Overall, the observations shown here should help understand
which cloud properties are affected by aerosols, cloud evolution, and
thermodynamic conditions. The latter was found to be associated with bulk
water contents in the clouds, while the overall shape of the DSDs is
determined by the aerosol condition during cloud formation and the
subsequent evolution.
Concluding remarks
The ACRIDICON-CHUVA campaign and the capabilities of the HALO aircraft
allowed the analysis of the sensitivities of Amazon tropical convective
clouds to aerosol number concentrations and updraft speed while also
considering cloud evolution. The sensitivity formulation identified that
aerosol number concentrations play a primary role in the formation of the
warm phase of convective clouds, determining not only droplet concentration
but also diameter and overall DSD shape. Conversely, the
thermodynamic conditions, as represented by the updraft intensity, affect
primarily DSD bulk properties such as water content and droplet
concentration. We have shown that the altitude above cloud base is critical
when analyzing aerosol and updraft impacts on clouds, given that the DSD
properties evolve with further processing in the system.
We showed that an increase of 100 % in aerosol concentration results in an
84 % increase in droplet number concentration on average, while the same
relative increase in updraft wind speed results in only 43 % change.
Regarding mean droplet size, we found it to be effectively independent of
the updraft speed. Roughly, the effective droplet diameter decreases 25 %
when aerosol concentration doubles. The comparison between the aerosol and
the thermodynamic effects shows that the aerosol concentration is the
primary driver for DSD, whereas the updrafts mainly affect droplet number
concentration and liquid water content. During cloud evolution, droplet
number concentration is depleted while the diameter sensitivity to the
growth processes is quantitatively similar to the aerosol effect.
Additionally, the aerosol effect on DSD shape inverts in sign with altitude,
favoring broader droplet distributions close to cloud base but narrower
droplet distributions
higher in the clouds. This highlights the importance of differentiating the
analysis by altitude above cloud base, which is an appropriate proxy for DSD
lifetime for our measurements.
The results presented here can potentially be used to validate and derive
new parameterizations in numerical models, which usually fail to correctly
represent Amazonian convective clouds. One common issue of the models is the
representation of the precipitation daily cycle, in which the modeled rainfall
tends to occur earlier than in the observations. One possible reason for
that is the misrepresentation of the cloud DSDs that can lead to
artificially high efficiency in rain formation. Therefore, model runs can be
performed in order to assess the factors that control DSD formation, and
comparisons can be made with our results as a benchmark. The analysis of the
ε and Λ parameters can be especially useful in that
regard. The results presented here detail several aspects of the Amazonian
clouds and their relation to aerosol and thermodynamic conditions. For
instance, it was shown that aerosols can induce DSD broadening only close to
cloud base, preferably under high w conditions. Higher in the clouds,
increased aerosol loading leads to DSD narrowing. Additionally, DSD
broadening with altitude is pronounced only in clean clouds, in which the
collection processes are efficient. The result is growing ε
with altitude, while this parameter remains relatively constant with H in
polluted clouds. Good models should be able to reproduce such details in
order to generate better forecasts. Therefore, we believe the results
presented here can be of use for that purpose, by providing specificities
of Amazonian clouds that models should aim to reproduce.