Retrieved from the Moderate Resolution Imaging Spectroradiometer (MODIS) on-board the Aqua satellite, 12 years (2003–2014) of aerosol and cloud properties were used to statistically quantify aerosol–cloud interaction (ACI) over the Baltic Sea region, including the relatively clean Fennoscandia and the more polluted central–eastern Europe. These areas allowed us to study the effects of different aerosol types and concentrations on macro- and microphysical properties of clouds: cloud effective radius (CER), cloud fraction (CF), cloud optical thickness (COT), cloud liquid water path (LWP) and cloud-top height (CTH). Aerosol properties used are aerosol optical depth (AOD), Ångström exponent (AE) and aerosol index (AI). The study was limited to low-level water clouds in the summer.
The vertical distributions of the relationships between cloud properties and aerosols show an effect of aerosols on low-level water clouds. CF, COT, LWP and CTH tend to increase with aerosol loading, indicating changes in the cloud structure, while the effective radius of cloud droplets decreases. The ACI is larger at relatively low cloud-top levels, between 900 and 700 hPa. Most of the studied cloud variables were unaffected by the lower-tropospheric stability (LTS), except for the cloud fraction.
The spatial distribution of aerosol and cloud parameters and ACI, here defined as the change in CER as a function of aerosol concentration for a fixed LWP, shows positive and statistically significant ACI over the Baltic Sea and Fennoscandia, with the former having the largest values. Small negative ACI values are observed in central–eastern Europe, suggesting that large aerosol concentrations saturate the ACI.
Aerosols and especially their effect on the microphysical properties of clouds are among the key components that influence the Earth's climate. As the magnitude and sign of such effects are not well known, understanding and quantifying the influence of aerosols on cloud properties constitutes a fundamental step towards understanding the mechanisms of anthropogenic climate change (Boucher, 2013).
As aerosols may act as cloud condensation nuclei (CCN), an increase in their number concentration can lead to an increase in the number of cloud droplets in super saturation conditions and a decrease in the cloud droplet radius. The decrease in the droplet effective radius resulting in an increase of the cloud albedo, under the assumption of a constant liquid water path, is known as the Twomey effect (Twomey, 1977). The decrease in droplet size can also impact the precipitation cycle, as the smaller droplets require longer time to grow into precipitating droplet sizes. Additionally, a possible decrease in the precipitation frequency of liquid clouds increases the lifetime of clouds (Albrecht, 1989). These impacts of aerosols are called the first and second indirect effects.
A quantitative evaluation of the effects of aerosols on clouds may be possible mainly in a statistical sense because of the local interactions between meteorological conditions and aerosols (Tao et al., 2012). Satellite-based remote sensing instruments can provide a large data set for statistical analysis from long-term observations of the aerosol indirect effect on a large spatial scale with daily global coverage, complementing localized ground measurements and providing necessary parameters for climate models.
A common approach in the satellite-based investigation of the first aerosol indirect effect (AIE) is the concept of the aerosol–cloud interaction (ACI), which relates the cloud optical thickness (COT), cloud effective radius (CER) or cloud droplet number concentration (CDNC) to the aerosol loading. The aerosol loading is usually expressed by the aerosol optical depth (AOD) or aerosol index (AI, defined in Sect. 3), which are used as proxies for the CCN concentration.
Many studies describe the interaction between aerosols and clouds through the correlation of the satellite-retrieved aerosol concentration and cloud droplet size on a global or regional scale. Inverse correlations on a global (Bréon et al., 2002; Myhre et al., 2007; Nakajima et al., 2001) and a regional scale (Costantino and Bréon, 2010; Ou et al., 2013) have been found, while Sekiguchi et al. (2003) and Grandey and Stier (2010), applying satellite data on a global scale, found either positive, negative or negligible correlations between the CER and AOD depending on the location of the observations. Jones et al. (2009) emphasized that the ACI should be inferred in aerosols or cloud regimes determined on a regional scale, as the relevance of aerosol type, aerosol concentration and meteorological conditions differs around the world.
Areas located at high latitudes are excluded from most of the studies due to a seasonal limitation of the satellite coverage and a smaller number of observations when compared to the global averages over the year. Lihavainen et al. (2010) compared in situ and satellite measurements to quantify the ACI on low-level clouds over Pallas (Finland), a northern high-latitude site, and concluded that the ACI values derived from ground-based measurements were higher than those obtained from satellite observations. Unlike the in situ instruments, the wavelengths used in the satellite retrievals constrain the detection of fine particles to those larger than about 100 nm, thus making it impossible to account for all CCN. Sporre et al. (2014a, b) combined aerosol measurements from two clean northern high-latitude sites with satellite cloud retrievals and observed that the aerosol number concentration affects the CER, while no impact on the COT was observed. As both studies focused on specific locations, no information was thus provided on a larger scale in the Baltic region. This work investigates whether the first indirect effect can also be observed by means of satellite-derived observations over the Baltic Sea region, a region that offers a northern clean atmospheric background (Fennoscandia) contrasted by a more polluted one (central–eastern Europe).
To determine whether it is possible to observe the response of the properties of low-level liquid clouds to different aerosol loadings in different atmospheric conditions, 12 years of aerosol and cloud properties available from the Moderate Resolution Imaging Spectroradiometer (MODIS) retrievals were investigated on a regional scale.
The satellite retrieval products are introduced in Sect. 2, the approach adopted for the ACI analysis is described in Sect. 3 and the results of the analyses are presented in Sect. 4.
The area covered in this study is situated at high latitudes
(50
The MODIS Collection 06 Level 3 (C6 L3) product provides cloud and aerosol
parameters at daily time resolution and at a regular
1
The MODIS C6 L3 product includes cloud microphysical parameters (CER, COT,
cloud liquid water path (LWP)) with statistics (mean, minimum, maximum, standard deviation) determined
at three different wavelengths (1.6, 2.1 and 3.7
We filtered the MODIS cloud data according to the following criteria:
Cloud parameters were only considered in the liquid phase. To eliminate possible outliers, retrievals with a standard deviation higher
than the mean values were discarded. Observations with a mean cloud-top temperature less than 273 K were
eliminated to ensure only warm liquid cloud regimes. The multilayer flag was applied to select only single-layer clouds. Transparent-cloudy pixels (COT < 5) were discarded to limit
uncertainties (Zhang et al., 2012). The CER derived from the 3.7 To exclude precipitating cases, observations were discarded when the
difference between CER at 3.7
The science data sets (SDSs) for the atmospheric aerosol information in the
MODIS C6 L3 provide the AOD retrieved at several wavelengths and as a
product from the application of either the “Deep Blue” or “Dark Target”
algorithm, or a combination of both retrievals (Levy et al., 2013; Sayer et
al., 2014). The SDS “Aerosol_Optical_Depth_Land_Ocean_Mean” is the
sole product providing the AOD at 0.55
To assess the effect of meteorological conditions on cloud properties, the ECMWF ERA-Interim reanalysis data were applied to derive the lower-tropospheric stability (LTS). Although not a ready-to-use product, the LTS is computed as the difference between the potential temperature at 700 hPa and at the surface (Klein and Hartmann, 1993), describing the magnitude of the inversion strength for the lower troposphere.
After selecting the cloud parameters as listed in the previous section, the number of observations were binned for both aerosol and cloud products. From the obtained histograms, 95 % of the most frequent ranges were selected from the total data set by filtering out 2.5 % of data from the extremes. These statistically more robust data sets were used in further analysis.
The product of the AOD, representing the column-integrated optical
extinction of aerosol at a given wavelength, and the derived AE, describing
the spectral dependency of the AOD, results in a third aerosol property of
interest, the aerosol index (AI). The AI is used as a proxy for the fine-mode aerosol particles, which have a larger contribution to the CCN than the
coarse-mode particles (Nakajima et al., 2001). MODIS Collection 6 provides
the AE only over land. To homogeneously estimate the AI over the Baltic Sea
and the surrounding land areas, the AE is evaluated by applying equation
Aiming to observe how the variation in aerosol conditions influences cloud properties, we adopted the approach of Koren et al. (2005) to analyse the average vertical distribution of the relationships between aerosols and cloud properties. The AOD and AI data sets were firstly sorted in ascending order and successively divided into five equally sampled classes that represent the averages of aerosol conditions for each of the classes. The cloud properties were then divided according to these AI and AOD classes and plotted as functions of cloud-top pressure.
The response of the cloud properties to clean-versus-polluted aerosol
conditions was studied spatially. The 25th and 75th percentiles of
the AI and AOD (AI–AOD) were computed for each spatial grid point, the
former constituting the upper limit for the AI–AOD values representing low
aerosol loadings and the latter representing the lower limit for the AI–AOD values for
heavy aerosol loadings. These percentile values were then used to divide
cloud parameters for clean and polluted aerosol conditions. The difference
between a cloud parameter value in low- and high-aerosol conditions is
Matsui et al. (2006) found that aerosols impact the CER stronger in an unstable environment (low LTS) than in a stable environment (high LTS) where the intensity of the ACI is reduced due to the dynamical suppression of the growth of cloud droplets. Following this result, we also compared cloud microphysical properties with both the AI–AOD and the LTS.
The area of this study was divided into three subregions as presented in Fig. 1: Area 1 covers the Baltic Sea, while Area 2 and Area 3 include only land pixels over Fennoscandia and central–eastern Europe respectively.
The area covered in this study and its division into three subregions: Area 1, the Baltic Sea, is represented by the colour blue; Area 2, covering the land areas of Fennoscandia, is represented by the colour green and Area 3, in red, includes the land areas of central–eastern Europe.
The ACI related to the CER was computed using the formulation from
McCominsky and Feingold (2008):
Figure 2 presents the time series of AI and AOD averages during the summer
months from 2003 to 2014 for each subregion. It is easy to see in Fig. 2
that these three areas have generally different aerosol conditions: within
the land subregions, the lower AI and AOD averages occur over Area 2, while
over Area 3 these values are higher during the entire period. Area 1, the
Baltic Sea, is considered as a third subregion per se due to the dominance
of maritime aerosol conditions. The AI is highest over Area 3
(central–eastern Europe), with an overall AI mean value of 0.29
Time series of summer (JJA) averages for AOD (circles) and AI (squares) for the three subregions. The three subregions are colour-coded as in Fig. 1.
Spatial distributions of AOD
The spatial variations of the aerosol and cloud properties are shown in Fig. 3. A decreasing south–north gradient of AOD is observed in Fig. 3a where the
highest values are found over Area 3 (northern Germany and Poland), and the
lowest are found over Area 2 (the Atlantic coast of Norway and northern Sweden). While
no discontinuities can be seen for the AOD distribution over Area 1 and Area 2, a clear distinction is evident in the AE (Fig. 3b). Indicating the
dominance of fine particles, high values of the AE are found over the entire
Area 1, over the eastern part of Area 3 and over the north-western part of
Area 2. Low values (AE < 1) are only partially found over the land
of Areas 2 and 3. The validity of the MODIS AE over land is generally
considered unrealistic. Nonetheless, in the case of the dominance of fine-mode
aerosols, the MODIS AE agrees with AERONET (Levy et al., 2010), while
disagreements occur in coarse aerosol cases (Jethva et al., 2007; Mielonen
et al., 2011). Over ocean, a good agreement between MODIS AE and AERONET is
found globally with the limitation of AOD > 0.2 (Levy et al.,
2015), a restriction that cannot be applied in our study area where the
regional AOD is about 0.2. As the sensitivity of AE to AOD errors is
especially critical for low AOD values, pixels with AOD < 0.2 are
expected to have a less-qualitatively accurate AE. Nevertheless, the AE over
Area 1 (Fig. 3b) matches the median range of 1.46–1.49 obtained from a
validation study that compares the AE retrieved by SeaWiFS and MODIS
Aqua or Terra with the three AERONET stations over the Baltic Sea (Melin et
al., 2013). Comparable high AE values were collected by Rodriguez et al. (2012) from 2002 to 2011 at the subarctic ALOMAR Observatory (Andøya,
Norway): the AE peaks during the summer season with a multi-annual mean and
standard deviation of 1.3
Figure 4 presents the 10-year average of the cloud properties, divided into five classes of the AI (Fig. 4a–d) and AOD (Fig. 4e–h) plotted as a function of cloud-top pressure. It can be observed that the lowest values of CTP correspond to the higher classes of AI–AOD. Assuming the CTP to be an indicator of the cloud-top height, this may suggest an enhancement of the cloud vertical structure. This result was also found by Koren et al. (2005), where convective clouds over the North Atlantic showed a strong correlation between the aerosol loading and the vertical development of the clouds. Furthermore, the cloud droplet effective radius (Fig. 4a, e) has smaller values in higher AI–AOD classes. The opposite behaviour, lower average values corresponding to the lower classes of the AI–AOD can be seen for the COT (Fig. 4c, g) and LWP (Fig. 4d, h), while the CF (Fig. 4b, f) shows a weaker signal for both AI and AOD cases. Overall, Fig. 4 reveals that the cloud parameters are clearly affected by the AI–AOD segregation at lower levels of CTP. For this reason, we limit our data set to cloudy pixels where the CTP is between 700 and 900 hPa.
Cloud properties, 10-year averaged, as a function of cloud-top pressure:
CER
CER as a function of AI, stratified for subranges of CTP and LWP, for the three subregions. The legend on the right of the figure lists the LWP bins.
In Fig. 5 the CER is plotted as a function of AI for fixed values of the LWP (five intervals as above) and the CTP (between 700 and 950 hPa, in 50 hPa bins). The highest AI in Area 1 (the Baltic Sea) is around 0.35 for the lowest clouds (CTP 900–950 hPa), decreasing to 0.3 for the highest clouds (CTP 700–750 hPa). Over Area 2 (Fennoscandia) the aerosol loading is not clearly connected to the cloud height, showing a constant AI average of approximately 0.25. As expected, Area 3 has the highest average of AI out of the three subregions, with values as high as 0.6 for the lowest clouds and a small decrement for the highest clouds. The cloud droplet size in Area 1 (the Baltic Sea) and Area 2 (Fennoscandia) shows a strong negative correlation with the AI, while a weak correlation is observed over Area 3 (central–eastern Europe). Area 1 has no results for the high LWP bins: during summer months few or no convective clouds form over the Baltic Sea and mainly thin stratiform clouds are identified in the cloud cover. Similar results are also found when the AOD is substituted by the AI (not shown).
Spatial distributions of the difference of the cloud properties
CF
Applying Eq. (2) to the cloud parameters, the impact of low and high aerosol
loading (
Mean low-level liquid cloud properties plotted as a function of LTS
and AI
In an attempt to connect the link between aerosol and clouds with meteorology, we evaluated the variability of low-level liquid cloud properties as a function of aerosol conditions (AOD–AI) and lower troposphere stability (LTS). Figure 7 shows the cloud properties (LWP, CER, CF and COT) plotted as a function of the LTS and AI–AOD. While the CF shows a gradient for both directions of the LTS and the AI–AOD, the other cloud variables (LWP, CER, COT) are mainly affected by aerosols with little to no correlation with changes in the LTS. Higher aerosol values correspond to a smaller CER (Fig. 7b, f) and higher CF (Fig. 7c, g) and LWP (Fig. 7a), in agreement with the AIEs, except for the LWP (Fig. 7e), which decreases as a function of the AOD. The LWP (Fig. 7e) shows a non-monotonic response by increasing when the AOD ranges between 0.3 and 0.4 because at high aerosol concentrations the cloud droplets are smaller and less likely to precipitate, and furthermore the LWP slightly decreases. A possible explanation of a better correlation of the LWP with the AI than with AOD might be found by looking at the LWP vertical distributions in Fig. 4, which indicate a more distinctive separation of the LWP for the AI-based classes than for AOD.
Figure 8 illustrates the ACI estimate for the CER (Fig. 8a) and its
corresponding correlation coefficient
ACI estimates computed for the CER as a function of the LWP and by
applying both the AI and AOD as proxies for the CCN are shown
in
Applying the AOD as a proxy for the CCN, estimates of the ACI and
correlation coefficient for the CER and for the interval of the LWP between
20 and 60 g m
In this work we have studied the applicability of satellite-based information for quantifying the ACI over the Baltic Sea region. Distinct subregional differences were found in the estimates of the ACI related to the effective radius of cloud droplets. No clear ACI results were observed for the other cloud parameters, which suggests that these may be influenced by other factors, such as the local meteorological conditions. The meteorological conditions are represented here by the LTS, which was compared to the cloud parameters. The LTS is correlated with the CF, while no effect was observed upon the other cloud parameters. In particular, there is no clear evidence of the effect of LTS on the interaction between aerosols and cloud effective radius.
One of the key aspects of this study was to find out whether a rigorously filtered Level 3 MODIS data set can be applied for ACI studies at a regional level. As the northerly location of the region of interest here restrains the availability of the MODIS observations to the summer months (JJA), one of the challenges is the limited data coverage. Moreover, the selection of specific cloud regimes and the co-location of aerosol and cloud observations are additional essential key factors in building up a robust data set, which, however, further decreases the amount of data points available. As far as the authors know, no previous results on ACI from a satellite perspective are provided over this area.
This study shows that the different aerosol conditions characterizing the Baltic Sea countries have an impact on the ACI and this can be also observed on a regional scale. According to ACI theory, polluted atmospheric conditions are connected with clouds characterized by lower cloud-top pressure, larger coverage and optical thickness. However, the cloud effective radius strictly follows the AIE's theory only over Area 1 (the Baltic Sea), which also agrees with the results presented by Feingold (1997). As reported in this study, the CER retrieved in clean clouds is mainly affected by the LWP and aerosol presence, while when detected under polluted conditions, it additionally shows a high dependence on other factors.
The cleaner atmosphere characterizing Area 1 (the Baltic Sea) and Area 2 (Fennoscandia) reveals statistically significant and positive ACI estimates between the CER and AOD that are in agreement with the values obtained from ground-based measurements collected at the sites of Pallas and Hyytiälä in Finland and Vavihill in Sweden (Lihavainen et al., 2010; Sporre et al., 2014b), while over the more polluted Area 3 (central–eastern Europe), the sensitivity to locally determine the ACI is smaller. It can be assumed that more aerosols lead to a high concentration of the CCN and this lowers the average droplet radius, as can be seen in Fig. 3e when the radius is compared between areas located south (high aerosol load) and north (low aerosol load) of the Baltic Sea.
Our analysis of the ACI for the CER shown in Fig. 8 leads to the following
conclusions:
The lowest values of the ACI can be seen over Area 3. This is also the
subregion with the highest average AOD values, leading to the smallest cloud
droplet size. A further addition of aerosol particles and thus possibly also
CCN does not decrease the cloud droplet size any further. Most of the ACI
values are actually negative but very close to zero. The positive ACI values for Area 2 show that the addition of aerosols to a
relatively clean atmosphere does decrease the droplet size. The AI over the land areas in the study should be considered unrealistic
because the average inland AE can have values below 1. The average AE over Area 1 has values as high as 1.4 to 1.5. These values,
however, can be trusted and have been evaluated by Melin et al. (2013). The low CER over Area 1 requires further explanation. The most probable
cause for the low values, based on the MODIS cloud retrieval, is the
relatively low cloud-top height over the sea. As cloud droplets generally
grow in size from the cloud base towards the cloud top (McFiggans et al.,
2006), Fig. 4 confirms that the average CER increases with the decreasing
CTP. Furthermore, in Fig. 5 there is a distinctive lack of results for high
LWP values, indicating that there are fewer clouds at higher top heights.
These reasons altogether lead to low values of the CER over Area 1 since the
MODIS instrument retrieves the droplet radius at cloud top, and the top
height CER results are low when compared to the surrounding over-land
values. The ACI over Area 1 has considerably higher values than over the land
subregions, and there is a difference in the magnitude between the ACI
values determined using the AOD or AI. The clean maritime atmospheric
conditions lead to the high sensitivity of droplet size to changes in fine-particle concentrations. The AOD and AI difference in ACI, the latter being
the higher, indicates that the ACI is caused by fine particles as expected.
Another way to assess the aerosol-induced changes in cloud parameters would
be to analyse time series to find out whether dynamically decreasing or
increasing aerosol loading has an effect on clouds. This sort of approach
was not attempted in this work.
Another important result of this work is the comparison of the ACIs obtained using the AI and AOD, chosen as proxies for the CCN, in order to determine which option leads to more realistic results. Even though theoretically the AI would be a better parameter than AOD to indicate the presence of fine-mode aerosol particles, the impact of uncertainties of the derived AI might be substantial.
All data used in this study are publicly available. The satellite data from
the MODIS instrument used in this study were obtained from
The authors declare that they have no conflict of interest.
This research was funded by the Maj and Tor Nessling Foundation (grant no. 201600287). The authors also acknowledge the Academy of Finland Centre of Excellence (grant no. 272041). Edited by: F. Yu Reviewed by: two anonymous referees