Introduction
Impacts of aerosols on clouds and precipitation have been reported as
introducing the largest uncertainty in quantifying the anthropogenic
contribution to climate change (Rosenfeld, 2000; Twomey, 1974; Gryspeerdt et al., 2014; Kaufman et al., 2012). Atmospheric aerosol particles
have been recognized as having two effects on Earth's climate. First, they can
directly alter the energy balance due to scattering and absorption of
incoming solar radiation (e.g. McCormick and Ludwig, 1967). Second, they can
act as cloud condensation nuclei (CCN) and thus modify the cloud
micro-physical properties and lifetime as well as precipitation (Ramanathan
et al., 2001; Krüger and Grassl, 2011). The effects of aerosol-induced
changes of cloud properties on the radiation budget are collectively
referred to as the aerosol indirect effect (AIE). The study presented here
is confined to aerosol–cloud interaction (ACI) using satellite data.
The activation of aerosol particles to CCN, or more specifically the number
concentration of CCN, is a direct link between aerosols and clouds, and the
aerosol activation efficiency is a key aerosol property affecting
ACI. For a given constant cloud liquid-water path
(CWP), an increased aerosol loading is expected to lead to smaller and more
numerous cloud droplets, resulting in an increase of cloud albedo. This
process, termed as the “first AIE” or “Twomey's effect”, may lead to a
net cooling of climate (Twomey, 1974; Feingold et al., 2003). The reduced
cloud droplet effective radius (CDR) also suppresses precipitation and can
consequently increase cloud lifetime, thus maintaining a larger liquid-water
path, with a possible further increase in the cloud optical thickness (COT)
and cloud reflectance. This process, described as the “second AIE”, may
further influence the cloud fraction (CF) (Albrecht, 1989; Feingold et al.,
2001). The interaction mechanisms between aerosols and clouds remain among
the most uncertain processes in the global climate system in spite of a
large number of studies made using both observations (Platnick et al., 2003;
Koren et al., 2005; Krüger et al., 2004) and models (Suzuki et al., 2004; Quaas et al., 2009;
Sena et al., 2016).
In order to better understand aerosol indirect effects, we resorted to
statistical analysis of satellite observations. By virtue of their large
coverage and high spatial and temporal resolution, satellite-borne
instruments have become a promising observational tool in studying
ACIs. Previous studies using a large amount of
satellite data and/or multiple satellite instruments have shown that aerosol
particles can affect cloud properties significantly (Krüger and Grassl,
2002; Menon et al., 2008; Sporre et al., 2014; Rosenfeld et al., 2014;
Saponaro et al., 2017). Satellite measurements suggest that the CDR tends to
decrease with increasing aerosol loading, which is consistent with Twomey's
theory (Matheson et al., 2005; Meskhidze and Nenes, 2010; Koren et al.,
2005). However, positive correlations between CDR and aerosol optical
depth (AOD) have also been found in some study areas, from both observations and
models, especially over land (Feingold et al., 2001; Grandey and Stier,
2010; Yuan et al., 2008). Different behaviours of CDR as a function of AOD for
different AOD regimes (low or high) have been observed by, for example, Tang et al. (2014)
and Wang et al. (2015). Feingold et al. (2001) concluded that there
are three kinds of CDR responses to aerosol enhancement: the CDR decreases
with increasing aerosol loading followed by (1) a saturation of the value of
CDR in response to high AOD, (2) a decrease in the CDR with further
increasing AOD due to suppression of cloud water vapour supersaturation
caused by abundant large particles, or (3) an increase in CDR with further
increases in AOD due to an intense competition for vapour which evaporates
the smallest droplets. Likewise, the aerosol impact on COT is still poorly
quantified. Costantino and Bréon (2013) reported that the relationship
between AOD and COT, which can be either positive or negative, depends on
the balance between the simultaneous CDR increase and CWP decrease when AOD
increases. With regard to the impact of aerosols on the cloud life cycle, it
is of great importance to explore the relationship between the aerosol
loading and cloud fraction, because the cloud fraction is highly associated
with other cloud properties and has a large effect on radiation (Gryspeerdt
et al., 2016). Kaufman and Koren (2006) and Koren et al. (2008) reported an increase in the
cloud cover with an increasing aerosol loading, followed by an inverse
pattern due to the absorption efficiency of aerosol. This brief summary
shows that the aerosol effect on cloud properties and the magnitude of this
effect are still very unclear.
Aerosol and cloud properties may have different vertical distributions and
may actually not physically interact. Costantino and Bréon (2013) and Jones
et al. (2009), using MODIS data, found that the aerosol indirect effect is
stronger for well-mixed clouds than for well-separated clouds (in well-mixed
aerosol and cloud, layers are physically interacting, as further explained in
Sect. 2). These observations show that it is important to consider the
relative altitudes of aerosol and cloud layers when estimating the aerosol
indirect effects. In addition, local differences in aerosol populations and
cloud regimes may have a strong effect on ACI (Sinha et al., 2003; Small et
al., 2011; Kaufman et al., 2005). Yuan et al. (2008) proposed that the
chemical composition of aerosol particles may play a role in determining the
relationship between AOD and CDR. Meteorology can affect the interaction
between aerosol and cloud, which usually further complicates ACI (Koren et
al., 2010; Reutter et al., 2009; Loeb and Schuster, 2008; Su et al., 2010;
Stathopoulos et al., 2017). As a consequence, the widely varying estimates
of the aerosol impact on cloud parameters, either positive or negative,
depend on factors like the aerosol size distribution and chemical
composition, cloud regime, and local meteorological conditions. Therefore,
the dataset used in this study contains not only aerosol and cloud
properties derived from MODIS, CALIOP and CloudSat, but also the
meteorological parameters collected from the daily ERA-Interim reanalysis
data.
The Yangtze River Delta (YRD) is characterized by a variable aerosol
composition and increasing aerosol concentration during the last two decades
(Ding et al., 2013a; Qi et al., 2015). Using multi-sensor retrievals, this
study aims to systematically examine the response of warm cloud parameters
(CDR, CF, COT and CTP) to the increase in the aerosol loading, where AOD is
used as a proxy for CCN number concentration (Andreae, 2009; Kourtidis et
al., 2015). New insights into the changing cloud properties over a wide
range of aerosol loadings, in particular in high AOD conditions, result from
our focus on a systematic understanding of ACI from three perspectives: (1) well-mixed and well-separated clouds, (2) aerosol effects on properties of
well-mixed clouds and (3) well-mixed clouds under different meteorological
conditions.
The paper is organized as follows: Sect. 2 describes the datasets used,
data processing and the main analysis conducted to explore aerosol cloud
interaction. Section 3 starts with a general description of aerosol and
cloud properties and the effect of aerosol loading on the relations between
them, followed by a description of aerosol effects on cloud properties (CDR,
CF, COT and CTP). In the latter we discriminate between well-separated and
well-mixed clouds. The focus will be on well-mixed clouds where ACI takes
place, and aerosol types and meteorological factors are considered to better
understand the possible mechanisms. Overall conclusions and discussions are
presented in Sect. 4.
Methods
Description of the study region
In this study, the YRD, covering the area
27–34∘ N and 115–122∘ E
(Fig. 1), was chosen in order to investigate the aerosol-induced
variability in micro- and macro-physical properties of low warm clouds during
4 consecutive years (2007–2010). The YRD region was chosen because it is
representative of the continental East Asian subtropical climate. The
marine monsoon subtropical climate for YRD is characterized by hot and humid
summers and cool dry winters (Sundström et al., 2012; Zhang et al.,
2010). The mean temperature in summer is about 27–28 ∘C. Mean annual
precipitation ranges from 1000 to 1400 mm and most precipitation occurs in
spring and summer (Zhang et al., 2010; Cao et al., 2016).
The population density in the YRD is very high with intensive human
activities in the region contributing to a very variable and complex aerosol
composition. The YRD has been reported as a major source region of both
black carbon and sulfate (Wang et al., 2014; Andersson et al., 2015). In
addition, other aerosol sources such as dust emissions render the
interactions between aerosols and clouds complicated (Nie et al., 2014). The
continental area of interest is characterized by a high level of
anthropogenic emissions and is well suited for research related to the
indirect effects of aerosols on cloud micro- and macro-physical properties.
Map of annual averaged MODIS/AQUA level 2 AOD for all years during
the period from 2007 to 2010. The black rectangle (27–34∘ N and 115–122∘ E) indicates the
Yangtze River Delta (YRD).
![]()
Data sources
The MODIS sensor, on board the Aqua satellite, has a swath width of
∼ 2300 km and multi-band spectral coverage (King et al.,
2003). The MODIS/Aqua overpass time for the study area is around 13:30 LT (local
time), when continental warm clouds are likely to be well developed.
Therefore MODIS/Aqua was selected as a data source to explore the ACI over
this area. In this work, we used the MODIS Collection 5.1 AOD product
(MOD04) derived from cloud-free pixels (resolution 500 m at nadir) and aggregated to
a resolution of 10 km × 10 km (Remer et al., 2005;
Levy et al., 2010). The AOD over land is retrieved using three MODIS
channels: 0.47, 0.66 and 2.13 µm (Remer et al., 2005). Cloud
properties are retrieved using six spectral channels (King et al., 1997) at
visible and near-infrared wavelengths (i.e., 0.66, 0.86, 1.24, 1.64, 2.12
and 3.75 µm). Here, we used the AOD as a proxy for aerosol burden in
our ACI analysis. The cloud properties used in this
study, CDR, CWP, COT, cloud top pressure (CTP) and cloud phase infrared
(CPI), were obtained from the Level 2 cloud product (MYD06) (King et al.,
2003). Both these products, MOD04 and MYD06, are in good agreement with
ground-based remote sensing data (Levy et al., 2010; Platnick et al., 2003).
More detailed information on algorithms for the retrieval of aerosol and
cloud properties is provided at http://modis-atmos.gsfc.nasa.gov.
Along with the Aqua satellites, CloudSat and CALIPSO (Cloud–Aerosol Lidar
and Infrared Pathfinder Satellite Observations) are flying in the so-called
“A-train” constellation together with other NASA satellites (Stephens et
al., 2002). CloudSat carries the CPR (cloud profiling radar), i.e. the first
satellite-based millimetre-wavelength cloud radar to detect the vertical
information on different-sized cloud droplets (Im et al., 2005). The CPR is
able to penetrate optically thick clouds and detect weak precipitating
particles (Wang et al., 2013). In the present study we utilized the datasets
CloudLayerBase and CloudLayerTop from 2B-CLDCLASS-LIDAR, the latest version
(R04) of the CloudSat standard data products. The data are provided in the
CPR spatial grid with vertical and horizontal resolutions of approximately
480 m and 1.4 × 1.8 km, respectively. CALIOP (Cloud–Aerosol Lidar
with Orthogonal Polarization) on board CALIPSO is the first space-borne
near-nadir polarization lidar optimized for aerosol and cloud measurements
(Winker et al., 2003). It is sensitive to optically thin clouds which could
be missed by CPR (Wang et al., 2013). The datasets Layer_Base_Altitude and Layer_Top_Altitude
retrieved from the CALIOP level-2 aerosol layer product (05kmALay)
were used in the present study. Its footprint is very narrow, with a laser
pulse diameter of 70 m on the ground. The vertical resolution of the CALIOP
layer product varies with altitude: 30 m for h= 0–8.2 km, 60 m for h= 8.2–20.2 km and 180 m for h= 20.2–30.1 km, whereas the horizontal
resolution is 5 km (Liu et al., 2009). Combining CloudSat and CALIPSO
observations has provided new insights into the vertical structure and
micro-physical properties of clouds (Matrosov, 2007).
The daily temperature at the 1000 and 700 hPa levels, relative humidity
at the 950 hPa level and pressure vertical velocity (PVV) at the 750 hPa level were
obtained from ERA-Interim reanalysis data. The daily ERA-Interim reanalysis
contains global meteorological conditions with 0.125∘ × 0.125∘ grids and a 37-level vertical resolution (1000–01 hPa) every
6 h (00:00, 06:00, 12:00, 18:00 UTC)
(http://apps.ecmwf.int/datasets/data/interim-full-daily/). The reanalysis
data were used for the closest collocation with the satellite overpass time
over the study area.
Level 2 MODIS, CALIOP, CALIOP/CPR and ERA-Interim products used to
characterize aerosol and cloud properties.
Product
Dataset
Horizontal resolution
Data source
Aerosol (MYD04 Level 2 Collection 5)
Optical_Depth_Land_And_Ocean
10 km
MODIS
Cloud (MYD06 Level 2 Collection 5)
Cloud_Effective_Radius
1 km
Cloud_Water_Path
1 km
Cloud_Phase_Infrared_Day
5 km
Cloud_TOP_Pressure_Day
5 km
Cloud_Fraction_Day
5 km
Cloud_Optical_Thickness
1 km
Cloud (2B-CLDCLASS-LIDAR)
CloudLayerBase
2.5 km
CALIOP/CPR
CloudLayerTop
2.5 km
Aerosol (05kmALay)
Layer_Top_Altitude
5 km
CALIOP
Layer_Base_Altitude
5 km
Cloud_Aerosol_Discrimination
5 km
Feature_Classification_Flags
5 km
ERA-Interim
Temperature (700, 1000 hPa)
0.125∘
ECMWF
Relative humidity (950 hPa)
0.125∘
Pressure vertical velocity (750 hPa)
0.125∘
Data processing
The MODIS/AQUA, CALIOP/CALIPSO and CPR/CloudSat satellites are part of the
A-Train constellation and observe the same scene on Earth within 1–2 min (Stephens et al., 2002). Therefore, time coincidence of retrievals
is assured when the datasets are extracted for the same date. Meteorological
properties retrieved from the 06:00 UTC ERA-Interim datasets were used here
as the “A-train” satellites constellation overpasses the region of
interest at about 13:30 LT (05:30 UTC). We aggregated CDR, COT and
CWP (1 km × 1 km) to a resolution of 5 km × 5 km to match
the along-track resolution of CALIOP (5 km × 5 km), while CTP, CF
and CPI were directly applied for the analysis since all of them are at a 5 km × 5 km spatial resolution.
Aerosol properties are only retrieved for strictly cloud-free pixels as
determined by the application of a cloud-detection scheme. However, cloud
detection schemes are not perfect and some residual clouds may remain
undetected resulting in high AOD (Kaufman et al., 2005). Another potential
source of error could be the misclassification of high AOD areas, such as in
the presence of desert dust or very high pollution levels, as clouds. To
reduce a possible over-estimation of AOD, cases with AOD greater than 1.5
were excluded from further analysis. In this paper, we focused on warm
clouds with CTP larger than 700 hPa and CWP lower than 200 g m-2, as
most aerosols exist in the lower troposphere (Michibata et al., 2014). In
addition, only cases with CPI = 1 (liquid-water cloud) were included. When
CALIOP detected the presence of aerosol, we averaged the MODIS aerosol
retrievals within a radius of 50 km from the CALIOP target. Likewise, we
averaged the MODIS cloud retrievals within a radius of 5 km from the CALIOP
target. For meteorological properties, we chose the value of the footprint
that is nearest to the CALIOP target. MODIS, CALIOP and CPR datasets are
listed in Table 1.
A quantitative relationship between AOD and cloud
properties has been documented in previous studies (Sporre et al., 2014;
Meskhidze and Nenes, 2010; Koren et al., 2005, Saponaro et al., 2017).
However, the relative vertical positions of aerosol and cloud layers
contribute to the uncertainty in this relationship. Following the method by
Costantino and Bréon (2013), we considered the aerosol and cloud layers to
be physically interacting (well mixed) when the vertical distance between
bottom (top) of the aerosol layer and the top (bottom) of a cloud layer was
smaller than 100 m. Coincident samples with a vertical distance larger than
750 m were assumed to be “well separated”. Coincident samples with a
distance between 100 and 750 m were defined as “uncertain”. The uncertain
cases, as identified using the information from CloudSat, were excluded from
further analysis in this study. Cloud types were identified as single-,
double- and multi-layer clouds using the cloud layer information at each
point. Single-, double- and multi-layer cloud samples accounted for 59,
30 and 11 % of the total samples, respectively. Using the highest
occurrence frequency (OF) of aerosol type below 10 km altitude at each
point, the aerosol type of highest OF was defined following the
Feature_Classification_Flags derived from
CALIOP.
Meteorological and aerosol impacts on cloud macro-physics and micro-physics
are found to be tightly intermingled (Stevens and Feingold, 2009). In an
attempt to isolate aerosol effects, the meteorological effects on clouds
were explored in a statistical sense. Meteorological properties used here
include relative humidity, lower tropospheric stability (LTS) and
PVV. LTS is defined as the difference in
potential temperature between the free troposphere (700 hpa) and the surface,
which is representative of typical thermodynamic conditions (Klein and
Hartmanm, 1993). It has been suggested that relative humidity, LTS and PVV affect aerosol and
cloud interaction (Gryspeerdt et al., 2014; Small et al., 2011). A
positive LTS is associated with a stable atmosphere in which vertical mixing
is prohibited; negative PVV indicates a local upward motion of air parcels.
Results and discussions
Overall aerosol and cloud characteristics
Spatial and time-series analysis of aerosol and cloud
parameters
Spatial distributions of AOD (a), CDR (b), CF (c), COT (d), CWP (e) and CTP (f) averaged
over all years between 2007 and 2010.
![]()
The spatial variations of the aerosol and cloud properties over the study
area, averaged over the years 2007–2010, are shown in Fig. 2. We can see a
decreasing north–south pattern in AOD in Fig. 2a, with the highest values
found in the north-east area. CDR behaves similarly to AOD, except that the
highest values are found in the northernmost area. Contrary to AOD, both COT
and CWP show an increasing north–south pattern. Furthermore, the spatial
distributions of COT and CWP are remarkably similar to each other.
Time series of the monthly-averaged values of AOD (a), CDR (b), CF (c), COT (d),
CWP (e) and CTP (f) for the dataset of MODIS–CALIPSO
coincidences for all months between 2007 and 2010. Month 1 is January.
![]()
Figure 3 shows time series of the monthly-averaged values for the AOD, CDR,
COT, CWP, CF and CTP, calculated for each month during the four years 2007–2010.
Both the monthly-averaged AOD and CDR are highest in June. December
presents the lowest monthly average for the AOD. Overall, the variations of
the monthly-averaged COT and CWP are similar, with the lower values in the
summer and the higher values in the winter. The monthly-averaged CF
approaches its maximum values in January and June, while CTP shows two peaks in
February and September. Note that CTP is plotted along the vertical axis from high to
low. The monthly averages are determined from the numbers of samples
presented in Table 2 for each parameter and each month between 2007 and
2010. Further, the availabilities of data for AOD and cloud properties are
not the same for the whole acquisition period between 2007 and 2010. It
indicates that not every CALIPSO shot has all the corresponding value for
AOD, CDR, COT, CWP, CF or CTP, which will decrease the data sample size to
some extent.
The sample sizes of all months for each parameter.
Parameters
January
February
March
April
May
June
July
August
September
October
November
December
Total
AOD
5428
3332
3892
4704
5598
3638
5944
6630
4306
6728
6110
6400
62 710
CDR
794
669
365
679
714
872
1228
2013
1514
1281
895
582
11 606
COT
886
747
392
732
748
915
1298
2072
1539
1329
967
627
12 232
CWP
1226
1125
620
1310
1226
1245
1490
2187
1929
1715
1261
867
16 201
CF
1398
994
537
955
993
1065
1671
2650
1996
1811
1373
1119
16 562
CTP
1398
994
537
955
993
1065
1671
2650
1996
1811
1373
1119
16 562
Variation of COT and CWP with CDR
Scatter plots of cloud parameters versus CDR in well-mixed
aerosol–cloud layers: (a) COT and (b) CWP, both for all data; (c) COT and (d) CWP,
both for data grouped by moderately polluted (in blue), polluted
(in green) and heavily polluted (in red) atmospheric conditions. Here
moderately polluted refers to AOD < 0.35, polluted refers to
0.35 < = AOD < = 0.8 and heavily polluted refers to
AOD > 0.8. The lines present the least-square fits, and the
resulting relations are presented in each figure. The number of data samples
is also reported in the figure (and following figures).
![]()
Prior to investigating the aerosol impact on warm cloud properties, a
general analysis of cloud properties and the effect of aerosol loading on
the relations between them are discussed below. The overall statistical
relations between the cloud parameters used in this study are derived from
the scatter plots shown in Fig. 4. All CDR, COT, CTP and CWP data shown in Fig. 4 (and later figures) are averaged over AOD bins, from 0.05 to 1.5
with a step of 0.02 on a log–log scale. Student's t test is used to
determine whether two sets of data are significantly different from each
other. The p value is defined as the probability of obtaining a result equal
to or “more extreme” than what was actually observed, when the null
hypothesis is true. The marker ∗ at the top right corner of R value denotes
statistically significant if p < 0.05.
We first explored the response of CDR to the increasing AOD in mixed
aerosol–cloud layers and found that CDR decreases with increasing AOD in
moderately polluted conditions (AOD < 0.35). In polluted and heavily
polluted conditions (AOD > 0.35), however, CDR increases with
increasing AOD. Here we discriminate between moderately polluted (AOD < 0.35), polluted (AOD > = 0.35 and AOD < = 0.8) and
heavily polluted (AOD > 0.8) conditions. The threshold of 0.35
for AOD is chosen based on analysis presented below in Sect. 3.2, where we
compare the relation of cloud parameters and AOD in more detail. Figure 4a
shows a scatter plot of COT versus CDR for well-mixed clouds. The correlation
between these parameters is negative, i.e. COT decreases with CDR, with a
correlation coefficient equal to -0.47. Figure 4c shows the same data but a
distinction is made between data points in moderately polluted,
polluted and heavily polluted conditions. For this dataset, COT increases
with an increasing CDR at moderately polluted conditions. In contrast, for
heavily polluted conditions COT shows a decrease with an increasing CDR.
This may indicate the existence of intense competition between the aerosol
particles for water vapour where the larger droplets are more prone to
condensation of water vapour than smaller ones, and thus grow to larger
sizes. This results in a shift of the droplet spectrum to larger sizes due
to the increase of CDR accompanied by a decrease of COT (Wang et al., 2015).
The data for the three different AOD cases show that the relationship
between CDR and COT is not unique and depends on the aerosol abundance.
Costantino and Bréon (2013) compared the CDR–COT relationship of mixed
and separated aerosol–cloud layers and found an increase in the CDR with
increasing COT, followed by a decrease with higher COT in both cases (mixed
and separated aerosol–cloud layers). Compared to their study, we consider
the effect of aerosol loading on the relationship between CDR and COT in
both cases.
Figure 4b shows a weak correlation between CWP and CDR for well-mixed
cloud layers, with a correlation coefficient equal to -0.15. However, when
different degrees of pollution are considered (Fig. 4d), we see a clear
correlation between both parameters (R= 0.78) in moderately polluted
conditions, where CWP clearly increases with increasing CDR. In polluted and
heavily polluted conditions the variation of CWP with increasing CDR is much
weaker (R= 0.31 for polluted conditions) and in heavily polluted conditions
CWP decreases with increasing CDR (R=-0.33).
Variation of COT and CWP with cloud top height
Scatter plots of cloud parameters in well-mixed aerosol cloud
layers for all data: (a) CTP versus COT, (b) CTP versus CWP, and (c) CWP
versus COT. The lines present the least-square fits, and the resulting relations
are presented in each figure.
![]()
CTP is generally used as a measure of cloud top height (CTH), with higher
CTP implying a lower CTH. Figure 5a shows a positive correlation between
CTP and COT, implying the occurrence of higher clouds with an increasing
COT, which is consistent with the general understanding of ACIs. Note that here and in the following figures, CTP is plotted
along the vertical axis from high to low, i.e. decreasing CTP indicates
increasing CTH, and positive correlations between CTP and other cloud
parameters indicate that an increase in these parameters corresponds to a
higher CTH. Figure 5b shows a positive correlation between CTP and CWP,
which again implies that clouds are higher as CWP increases. An explanation
for this phenomenon is provided by Gao et al. (2014), i.e. clouds grow in
the vertical and more drizzle is produced, so that the CWP becomes larger. Figure 5c shows the relation between CWP and COT. The
CWP increases with the increase of COT, which is in good agreement with the
aerosol second indirect effect hypothesis that the precipitation suppression
can increase CWP and possibly further increase COT. This observation is in
good agreement with those of Costantino and Bréon (2013) that cloud
water amount increases with increasing COT.
Difference between separated and mixed conditions
Scatter plots of cloud parameters versus AOD over YRD on log–log
scale for cases of separated (blue) and mixed (red) aerosol–cloud
layers: (a) CDR versus AOD, (b) CF versus AOD, (c) COT versus AOD and (d) CTP versus
AOD. The lines present the least-square fits, and the resulting relations
are presented in each figure. Error bars represent the confidence level of
the mean cloud parameters' value for each AOD bin, i.e. the statistical
uncertainties, expressed as σ/(n-2), where n is the number of cases
within the AOD bin and σ is the standard deviation of cloud
properties.
![]()
In this section we examine the responses of various cloud properties to the
increasing AOD for well-separated and well-mixed clouds, respectively.
Figure 6 shows relations between cloud parameters (CDR, CF, COT, CTP) and
AOD for both separated and mixed conditions. The strength of the interaction
between cloud properties and AOD is quantified here as the slope of the line
describing the relation between cloud parameters and AOD, on a log–log
scale, as obtained by linear regression. In Fig. 6a, CDR shows a
negative relation with AOD in moderately polluted conditions when aerosol
and cloud layers are mixed, which is in good agreement with Twomey's theory
(Twomey, 1977). We note that, due to the limited number of data points in
the dataset with AOD < 0.35, the present work does not allow
the selection of conditions with a constant CWP. Following, for example, Costantino and
Bréon (2010, 2013) and Wang et al. (2015), we use all available data together. In
polluted and heavily polluted conditions, however, CDR increases with
increasing AOD, suggesting some sort of saturation in ACIs when AOD approaches 0.35. This value for the tipping point
(0.35) is close to the value of 0.4 reported by Feingold et al. (2001). As
discussed earlier, Feingold et al. (2001) proposed three primary responses
of CDR to the aerosol loading. We consider the fact that CDR increases with
an increase in AOD when AOD loading exceeds 0.35 as the “anti-Twomey effect”.
The positive relation between CDR and AOD may be similar to that described
by Feingold et al. (2001), case 3 (see above), i.e. due to intense vapour
competition the smaller droplets evaporate as the number of particles
continues to increase. It may also be that only a subset of aerosol
particles is activated when not enough vapour is available, and once
activated they continue to grow faster, thus preventing water vapour from
condensing onto smaller aerosol particles that are less susceptible to
activation, resulting in the increase of CDR.
Figure 6a also shows that, in well-separated cloud layers, CDR varies much
less with AOD irrespective of whether the AOD is relatively low or high.
Such a weaker variation can be attributed to the fact that no aerosols are
subjected to cloud micro-physical process since there are no physical
interactions between aerosol and cloud layers.
Figure 6b shows that when aerosol and cloud layers physically interact,
the CF shows a decrease with an increasing AOD in moderately polluted
conditions, albeit with a low significance as indicated by the small
correlation coefficient R, followed by an inverse pattern in polluted and
heavily polluted conditions. This outcome is not in agreement with the
findings of Koren et al. (2008) and Small et al. (2011). It could be
explained as follows: here, when aerosol and cloud layers are well-mixed,
the absorption of solar radiation heats the mixed layer and reduces the
cloud cover due to the quite high concentrations of the smoke particles over
the YRD. This feedback would be balanced once the heating of the surface
raises the surface temperature. It destabilizes the atmosphere, resulting in
vertical transport and thus enabling transfer of humidity from the surface
to higher levels in the atmosphere. This effect increases cloudiness (Koren
et al., 2008). Conversely, CF shows an increasing pattern with an increasing
AOD for the whole AOD dataset in well-separated cloud layers. This increase
might be due to absorbing aerosols interacting with incoming solar radiation
above the cloud layer (Costantino and Bréon, 2013). In this process,
absorbing aerosols above cloud tops may heat the aerosol layer and cool the
surface, thereby stabilizing the boundary layer and maintaining a moist
boundary layer. In addition, scattering aerosol reduces the amount of solar
light reaching the surface. This combination of two effects suppresses cloud
vertical development and increases the low cloud cover.
The COT has a negative correlation with AOD in both conditions, as shown in Fig. 6c. There are two effects that may contribute to this negative
relationship. On the one hand, the evaporation of cloud droplets caused by
locally absorbing aerosol makes clouds thinner, which is a radiative effect.
On the other hand, the presence of absorbing aerosol may influence the
satellite-retrieved COT because it can absorb radiation and thus reduce the
cloud reflectance measured by the sensors on the satellite (Meyer et al.,
2013, 2015; Li et al., 2014; Ten Hoeve et al., 2011). Meyer et al. (2013) reported that adjusting for above-cloud aerosol attenuation can
increase the retrieved regional mean COT by roughly 18 % for polluted
marine boundary layer clouds. Li et al. (2014) also found that, due to
absorbing aerosols in the heart of the YRD region, satellite
observations tend to underestimate COT. The radiative effect and retrieval
uncertainty could be the important factors for the decrease of COT with
increasing AOD, as suggested by Ten Hoeve et al. (2011) and Alam et al. (2014).
These authors reported similar results on the decrease of COT with
increasing AOD, which may result from the measured reflectance from a cloud
top at visible wavelengths being smaller than expected due to
absorbing aerosols.
The relationship between CTP and AOD has been plotted in Fig. 6d. There
is a positive correlation between CTP and AOD, which is contradicting the
general understanding that high aerosol loading will result in an increase
of cloud lifetime and higher cloud top. The positive relation between CTP
and AOD has an implication that higher aerosol abundance is not always
accompanied by smaller CTP. This suggests that the primary
effect of aerosol is not always to produce taller and more convective clouds
(Rennóet al., 2013).
Based on the above findings, we conclude that for well-mixed clouds in the
YRD, the CDR shows a decrease with an increasing AOD under
moderately polluted conditions, followed by an increase under polluted and
heavily polluted conditions due to the intense water vapour competition. The
cloud cover behaves qualitatively similar to CDR in response to changing
values of AOD. Meanwhile, cloud optical depth becomes smaller and CTP becomes larger with increasing AOD over the whole range of AOD
values.
Case of mixed aerosol–cloud layers
ACI for single-layer mixed clouds
Scatter plots of cloud parameters versus AOD over YRD on log–log
scale for mixed-aerosol single-layer clouds: (a) CDR, (b) CF, (c) COT and (d) CTP.
The lines present the least-square fits, and the resulting
relations are presented in each figure. The error bars indicate the
statistical uncertainties as in Fig. 6.
![]()
Well-mixed clouds show a stronger relation between aerosol and cloud
properties than separated clouds, as shown above. From here on, we will
focus on potential aerosol indirect effects on well-mixed warm clouds as
defined above. Relations between CDR, CF, COT and CTP with AOD will be
explored in this section. Figure 7 shows the variation of single-layer cloud
properties with AOD when aerosol and cloud layers are mixed. The relation
between CDR and AOD changes from negative for AOD < 0.35 to positive
for AOD > 0.35 (Fig. 7a). As with the CDR, the CF shows similar
variation with the elevated AOD over the whole AOD range. Figure 7c shows
that COT is negatively associated with increasing AOD. In contrast, CTP
decreases with increasing AOD (Fig. 7d), i.e. CTH increases.
In general, the characteristics for cases of mixed-aerosol single-layer warm
clouds (Fig. 7) are quite similar to the case of mixed-aerosol warm clouds (Fig. 6). The slight difference of fits comes from the different types
of clouds that are considered in different conditions. In Fig. 6, the clouds are
not limited to single-layer warm clouds, but also double-layer warm clouds.
Influence of aerosol type on ACI
Scatter plots of cloud parameters versus AOD over YRD on log–log
scale for cases of mixed-dust-aerosol cloud layers (blue) and mixed-smoke-aerosol cloud layers (red): (a) CDR, (b) CF, (c) COT and (d) CTP. The lines
present the least-square fits, and the resulting relations are presented in
each figure. The error bars indicate the statistical uncertainties as in
Fig. 6.
![]()
Eastern China is a region with high concentrations of sulfate, dust, black
carbon and other carbonaceous aerosols. In heavily polluted areas, dust
aerosols become coated with hygroscopic material, making them effective CCN
(Levin et al., 1996; Satheesh et al., 2006). In particular, there are high
emissions of smoke by straw-burning in summertime. ACI is strongly dependent on the aerosol types, their size
distribution and the vertical variation of these, as well as ambient
environmental conditions (Patra et al., 2005; Matsui et al., 2006; Dusek et
al., 2008; Yuan et al., 2008). Thus, aerosol species are indicative of
causal micro-physical and radiative effects. Different aerosol types may
reveal different patterns of ACI. Here, polluted dust (accounting for
34 %) and smoke aerosol (accounting for 38 %), which are the two main
aerosol types occurring in the YRD, are chosen to investigate the variation
of cloud parameters with AOD. Smoke (fine absorbing particles) and polluted
dust (coarse particles) aerosols are identified using the CALIOP
classification. In addition, they have different efficiency for the
absorption of sunlight.
Figure 8 shows the variation of cloud parameters with AOD over the YRD,
where data points for mixed polluted dust-warm clouds and mixed-smoke-aerosol warm clouds are indicated with different colours. Figure 8a shows
that the CDR is, in general, larger in the presence of smoke aerosol than in
the presence of dust. Meanwhile, the cloud fraction is smaller in the
presence of smoke, as shown in Fig. 8b. This can be due to the greater
efficiency of smoke aerosol particles for the absorption of sunlight than
that of dust, resulting in local warming in the presence of smoke aerosol
which in turn leads to evaporation of water and thus an increase in small
droplets or even complete evaporation of cloud droplets and thus a reduction
of cloud cover. Figure 8c shows that the COT decreases
with increasing AOD for both aerosol types albeit with a low significance as
indicated by the small correlation coefficient R. The slope of linear
regression of COT against AOD is much stronger in the
presence of smoke aerosol than in the presence of dust, indicating that the
ACI is stronger for smoke than for polluted dust. In addition to those
mentioned, one factor which probably also contributes to the observed
difference between effects of smoke and polluted dust is that dust does not
absorb sunlight at 0.86 µm (Kaufman et al., 2005). Figure 8d shows
that the slope of linear regression of CTP against AOD is
much stronger for smoke aerosol than that for polluted aerosol, with a
correlation coefficient equal to 0.36. Both these results may be due to the
higher absorption efficiency of smoke (Small et al., 2011).
Influence of relative humidity on ACI
Scatter plots of cloud parameters versus AOD over YRD on log–log
scale for cases of low relative humidity (31 %) condition (blue) and mixed aerosol–cloud
layers under high relative humidity (91 %) condition (red): (a) CDR, (b) CF, (c) COT and (d) CTP.
The lines present the least-square fits, and the resulting
relations are presented in each figure. The error bars indicate the
statistical uncertainties as in Fig. 6.
![]()
Feingold et al. (2001) reported that the aerosol indirect effect depends
highly on the aerosol hygroscopicity and PVV. Wang et al. (2014)
demonstrated that the observed interaction between aerosol and
cloud can be affected by the dynamical and thermodynamical processes in
cloud systems. Therefore, to explore the meteorological impact on the
interaction between aerosol and cloud observed over the YRD, we classify the
data for various meteorological parameters, including relative humidity (this section), LTS
and PVV (Sect. 3.3.4).
Relative humidity is one of the main factors affecting aerosol particle
size and cloud formation. For instance, high relative humidity at cloud base has been
reported to affect the relation between aerosol particles and cloud
properties (Small et al., 2011). Thus, effects of relative humidity need to be accounted
for in ACI studies, as reported in the literature
(Jeong et al., 2007; Loeb and Manalo-Smith, 2005; Quaas et al., 2010).
The cloud properties versus AOD relationships are classified by relative humidity (at
950 hPa) in three equally sized subsets and the mean relative humidity values for each
subset are calculated. In Fig. 9 we show cloud properties as a function of
AOD for only the lowest relative humidity (31 %), representing dry conditions, and the
highest relative humidity (91 %, above the deliquescence point of ambient particles).
Figure 9a shows that the CDR is larger in high relative humidity
conditions than in low relative humidity conditions, irrespective of the
AOD. It is likely that hygroscopic aerosols grow in size caused by
condensation of water vapour (Hanel, 1976; Feingold et al., 2003). The
increasing relative humidity further increases the probability of the cloud droplet
activation and growth of existing cloud droplets as well (Jones et al.,
2009). This indicates that high relative humidity conditions can help the
formation of larger cloud droplets due to a higher water vapour content in
the atmosphere. The cloud fraction is much larger in high relative humidity
conditions than in low relative humidity conditions, as shown in Fig. 9b. Figure 9c shows that the COT decreases with
increasing AOD in both conditions, albeit with a low significance as
indicated by the small correlation coefficient R. However, the COT is larger in high relative humidity conditions than in low
relative humidity conditions for the entire AOD dataset. In contrast, the
CTP is smaller in high relative humidity conditions than in
low relative humidity conditions over the whole range of AOD values (Fig. 9d). This implies that high relative humidity can promote the formation of
thicker and higher clouds.
Influence of boundary layer thermodynamics and dynamics on
ACI
Scatter plots of cloud parameters versus AOD over YRD on log–log
scale for cases of low LTS condition (blue) and mixed aerosol–cloud layers
under high LTS condition (red): (a) CDR, (b) CF, (c) COT and (d) CTP. The
lines present the least-square fits, and the resulting relations are
presented in each figure. The error bars indicate the statistical
uncertainties as in Fig. 6.
![]()
The LTS is an indicator for the mixing state of the atmospheric layer
adjacent to the surface. It describes to some extent the atmosphere's
tendency to promote or suppress vertical motion (Medeiros and Stevens,
2011), which in turn affects cloud properties (Klein and Hartmann, 1993).
Figure 10 shows cloud properties as a function of AOD for two different LTS
conditions: low LTS, with a mean value equal to 10.11 representing an
unstable atmosphere; and high LTS, with a mean value equal to 20.47
representing a stable atmosphere. Figure 10a shows that the CDR is larger
in unstable atmospheric conditions than in stable conditions, irrespective
of the AOD. This indicates that in unstable atmospheric conditions the cloud
droplets are larger, which may be due to stronger interaction between
aerosols and clouds as a result of better vertical mixing of water vapour.
Figure 10b shows that the slope of linear regression of cloud fraction
against AOD is much stronger for stable atmospheric conditions than for
unstable atmospheric conditions in the heavily polluted conditions. This
demonstrates that stable atmospheric conditions can promote the formation of
a cloud (Small et al., 2011). A high LTS indicates a strong inversion,
which prevents vertical mixing and cloud vertical extent, maintaining a
well-mixed and moist boundary layer and providing an environment which
favours the development of a low cloud cover. Figure 10c shows that the
COT is larger in unstable atmospheric conditions than in
stable atmospheric conditions. In contrast, the CTP is
smaller in unstable atmospheric conditions than in stable atmospheric
conditions for the whole range of AOD values (Fig. 9d). This indicates
that unstable atmospheric conditions can promote the formation of thicker
and higher clouds and stable atmospheric conditions can enhance the cloud
cover.
Scatter plots of cloud parameters versus AOD over YRD on log–log
scale for cases of PVV < 0 condition (blue) and mixed aerosol–cloud
layers under high PVV > 0 condition
(red): (a) CDR, (b) CF, (c) COT and (d) CTP. The lines present the least-square fits, and the resulting
relations are presented in each figure. The error bars indicate the
statistical uncertainties as in Fig. 6.
![]()
The PVV, a measure of dynamic convection strength, is very important for
cloud formation. In particular, the vertical velocity can be used to
determine whether a certain region may be susceptible to cloud development
or not. That is, the presence of upward motion, as indicated by negative
PVV, can enhance ACI as it makes the ambient environment favourable for
cloud formation, and vice versa (Jones et al., 2009).
Figure 11a shows that in moderately polluted conditions the CDR is larger
in the presence of upward motion of air parcels than for downward motion.
This observation indicates that the upward motion of air parcels can promote
the formation of larger cloud droplets, thus enhancing ACI. However, the
impact of vertical velocity is weak in polluted and heavily polluted
conditions. Figure 11b shows that the cloud fraction is larger in the
presence of upward motion of air parcels than for downward motion of air
parcels when AOD is greater than 0.35. This indicates that the upward motion
of air parcels can favour cloud development and increase cloud cover in
heavily polluted conditions. The phenomenon is not obvious when AOD is
smaller than 0.35. These results emphasize the importance of vertical
velocity when estimating the potential aerosol effect on cloud droplet
effective radius and cloud fraction. Figure 11c
shows that the COT is larger in the presence of upward
motion of air parcels than for downward motion throughout the range of AOD.
In contrast, the CTP is smaller in the presence of upward
motion of air parcels than for downward motion (Fig. 9d). This implies
that upward motion of air parcels can be helpful for the formation of
thicker and higher clouds.
Error sources and uncertainties
Caution is warranted in investigating the satellite-derived relations
between aerosol and cloud properties. Uncertainties in satellite data may
result from assumptions on the aerosol size distribution used in the
retrieval process, imperfect cloud detection resulting in residual clouds
leading to high AOD values, effects of relative humidity on aerosol
parameters and dynamic effects (Yuan et al., 2008). Below we discuss
several potential factors that may have affected the interaction between
aerosols and clouds in our analysis.
Firstly, the correlation between AOD and cloud parameters may be influenced
by aerosol size distributions (Small et al., 2011). Since the MODIS
retrieval does not provide aerosol size information, it is better to explore
the seasonal differences in the observed ACI due to the difference in
aerosol emissions between the different seasons. However, the relatively low
number of MODIS–CALIPSO coincidences limits the further binning of the data
required to investigate this issue. Secondly, when it comes to the
occurrence of cloud contamination in the AOD dataset, this is a universal
and one of the most difficult problems in aerosol retrieval. Cloud detection
is usually not perfect, so that undetected, or residual, clouds contaminate
the retrieval area, which leads to AOD overestimation and in turn affects the
relation between aerosol and cloud properties (e.g. Sogacheva et al., 2017).
A study by Mei et al. (2016), comparing their MERIS cloud mask with two
independent datasets, shows that of the order of 70–90 % of the cases are
correctly classified as cloud-free. This result is in good agreement with
that from a dedicated study on a consistency between aerosol and cloud
retrievals from the same instrument, which showed that about 20 % of the
pixels may be misclassified (Klueser, 2014). In this study, the samples
with AOD values greater than 1.5 were excluded in a rough attempt to exclude
cloud-contaminated AOD to reduce the uncertainty in the observed ACI.
Thirdly, Feingold et al. (2003) reported that water vapour swelling
increases the AOD. Sheridan et al. (2001) showed an important role of
hygroscopic growth in determining the AOD for sea salt aerosols. The effect
of humidity on the ACI has been discussed in Sect. 3.3.3. Finally, Young (1993) reported that ACI is influenced by dynamics through modifying
radiative and thermodynamic heating. Jones et al. (2009) emphasized the
importance of vertical mixing velocity in cloud formation and ACI as
discussed in Sects. 3.3.4 and 3.3.5. As reported by Yuan et al. (2008),
the potential artefacts mentioned above do not seem to be the primary cause
for the observed relationship between aerosol and cloud parameters. Further
investigations are needed to fully analyse and explain the observed
phenomena.
Conclusions
The high level of anthropogenic emissions in eastern China render this area
an important hotspot for studying how cloud micro-physical properties are
affected by anthropogenic aerosols (Ding et al., 2013b). Based on the
near-simultaneous aerosol and cloud retrievals provided by MODIS, CALIOP and
CloudSat, together with the ERA-Interim reanalysis data, we investigated the
effect of aerosols, with AOD used as a proxy for the aerosol loading, on
micro-physical and macro-physical cloud properties over the YRD for the years 2007 to 2010. In terms of the relative heights of
aerosol and cloud layers, well-mixed and separated clouds were defined. A
statistical analysis was used to examine the aerosol effects on cloud
properties for these two cases. Besides the aerosol impact on CDR, CF, COT
and CTP, the influence of environmental conditions, such as relative humidity, LTS and
PVV, on the relation between cloud properties and AOD was also studied. In
addition, the impact of two different aerosol types, dust and smoke, was
explored.
The analysis of the COT–CDR and CWP–CDR relationships for well-mixed clouds
indicated that they are affected by the aerosol loading. A statistical
analysis of the relation between CWP and COT showed an increase in CWP with
an increasing COT, which is in a good agreement with the findings reported
by Costantino and Bréon (2013).
Consistent with previous findings, we found that the CDR initially decreases
with increasing AOD, followed by an increase after AOD reaches a value of
0.35. This result is consistent with Twomey's hypothesis that increasing
aerosol abundance leads to more numerous but smaller cloud droplets at given
constant cloud water content. The positive relation between CDR and AOD may
be caused by micro-physical processes, which is coupled with intense vapour
competition and evaporation of smaller droplets as a result of a high
abundance of aerosol particles. Also, the analysis of the variation of CF
with increasing AOD showed that CF varies with AOD in a way similar to that
of CDR. This finding differs from those by Koren et al. (2008) and Small et al. (2011) who observed an increase in the cloud cover with an
increasing AOD, followed by a decrease with higher AOD. COT was found to
decrease with an increasing AOD. We argue that the radiative effect and
retrieval artefact due to absorbing aerosol might be important factors in
determining this relationship. This effect can result in increased cloud
evaporation and reduced cloud cover. Meanwhile, CTP tends to increase as
aerosol abundance increases, indicating that the aerosol is prone to
expanding
the horizontal extension. In other words, we found that for well-mixed
clouds over the YRD, the CDR becomes smaller with the increase of AOD in
moderately polluted conditions, which is, in principle, in line with the Twomey
effect, yet the cloud fraction indicates a weak decrease which could be
attributed only to the weak influence of evaporation caused by absorption of
aerosols.
On the other hand, in polluted and heavily polluted conditions, a reduced
cloud coverage can result in more solar radiation reaching the surface,
causing surface heating and thus raising the surface temperature, which then
destabilizes the atmosphere. The resulting advection transports water vapour
from the surface to higher levels in the atmosphere, therefore producing
more cloud. Meanwhile, CDR becomes larger as a result of the stronger water
vapour competition in polluted and heavily polluted conditions. The COT
decreases with the increasing values of AOD throughout the AOD range due to
the radiative effect and possible retrieval artefacts. The behaviour of CTP
is consistent with that of COT, with the cloud getting thinner but with
larger cover, so that CTP becomes larger with an increasing AOD.
Furthermore, joint correlative analysis of different aerosol and cloud
properties revealed that smoke aerosols have a stronger impact on
ACI due to their stronger absorption of solar
radiation compared with polluted dust. Therefore, we can conclude that
absorbing aerosols play an important role in the ACI.
Constrained by relative humidity and boundary thermodynamic and dynamic
conditions, the variation of cloud properties in response to aerosol
abundance was analysed. In general, a high relative humidity can promote the
formation of larger cloud droplets and expand cloud formation, irrespective
of the vertical or horizontal level. With regard to LTS, stable atmospheric
conditions can enhance the cloud cover horizontally. However, unstable
atmospheric conditions can be helpful for the formation of thicker and
higher clouds. Dynamically, an upward motion of air parcels can also
facilitate the formation of thicker and higher clouds. Besides the
meteorological controls mentioned above, other factors may be important in
generating relations between aerosol and cloud properties, such as
temperature advection. These results suggest that effects of ambient
meteorological environments need to be considered when exploring the aerosol
indirect effect. In summary, this study will greatly help us to understand
the mechanisms of ACI and ultimately of indirect aerosol
effects over the YRD.