An estimate of aerosol indirect effect from satellite measurements with concurrent meteorological analysis

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 115,, doi: /2010jd013948, 2010 An estimate of aerosol indirect effect from satellite measurements with concurrent meteorological analysis Wenying Su, 1 Norman G. Loeb, 2 Kuan Man Xu, 2 Gregory L. Schuster, 2 and Zachary A. Eitzen 1 Received 26 January 2010; revised 9 June 2010; accepted 17 June 2010; published 28 September [1] Many studies have used satellite retrievals to investigate the effect of aerosols on cloud properties, but these retrievals are subject to artifacts that can confound interpretation. Additionally, large scale meteorological differences over a study region dominate cloud dynamics and must be accounted for when studying aerosol and cloud interactions. We have developed an analysis method which minimizes the effect of retrieval artifacts and large scale meteorology on the assessment of the aerosol indirect effect. The method divides an oceanic study region into 1 1 grid boxes and separates the grid boxes into two populations according to back trajectory analysis: one population contains aerosols of oceanic origin, and the other population contains aerosols of continental origin. We account for variability in the large scale dynamical and thermodynamical conditions by stratifying these two populations according to vertical velocity (at 700 hpa) and estimated inversion strength and analyze differences in the aerosol optical depths, cloud properties, and top of atmosphere (TOA) albedos. We also stratify the differences by cloud liquid water path (LWP) in order to quantify the first aerosol indirect effect. We apply our method to a study region off the west coast of Africa and only consider single layer low level clouds. We find that grid boxes associated with aerosols of continental origin have higher cloud fraction than those associated with oceanic origin. Additionally, we limit our analysis to those grid boxes with cloud fractions larger than 80% to ensure that the two populations have similar retrieval biases. This is important for eliminating the retrieval biases in our difference analysis. We find a significant reduction in cloud droplet effective radius associated with continental aerosols relative to that associated with oceanic aerosols under all LWP ranges; the overall reduction is about 1.0 mm, when cloud fraction is not constrained, and is about 0.5 mm, when cloud fraction is constrained to be larger than 80%. We also find significant increases in cloud optical depth and TOA albedo associated with continental aerosols relative to those associated with oceanic aerosols under all LWP ranges. The overall increase in cloud optical depth is about 0.6, and the overall increase in TOA albedo is about 0.021, when we do not constrained cloud fraction. The overall increases in cloud optical depth and TOA albedo are 0.4 and 0.008, when we only use grid boxes with cloud fraction larger than 80%. Citation: Su, W., N. G. Loeb, K. M. Xu, G. L. Schuster, and Z. A. Eitzen (2010), An estimate of aerosol indirect effect from satellite measurements with concurrent meteorological analysis, J. Geophys. Res., 115,, doi: /2010jd Introduction [2] Anthropogenic aerosols affect the global energy budget by scattering and absorbing sunlight (i.e., the direct effect), and by increasing the cloud droplet number concentration (i.e., the indirect effect). Twomey [1977] states that the presence of anthropogenic aerosols increases the cloud droplet number concentration and thus the optical depth of 1 Science Systems and Applications Inc., Hampton, Virginia, USA. 2 NASA Langley Research Center, Hampton, Virginia, USA. Copyright 2010 by the American Geophysical Union /10/2010JD clouds (under constant liquid water path assumption), which leads to an increase of cloud albedo; the presence of anthropogenic aerosols also increases the absorption coefficient, leading to a decrease in cloud albedo. His calculations suggest that the brightening effect dominates for thin to moderately thick clouds, whereas the darkening effect dominants for sufficiently thick clouds. This is known as the first aerosol indirect effect (AIE), or Twomey effect. Furthermore, smaller cloud droplet effective radius (R e ) results in a reduced precipitation formation rate and potentially an enhanced cloud lifetime and cloud fraction. This is referred to as the second aerosol indirect effect or cloud lifetime effect [Albrecht, 1989]. 1of14

2 [3] The aerosol indirect effects are difficult to observe from space because clouds are controlled by dynamical and thermodynamical processes (to first order). We stress that the Twomey effect is conceptualized under constant liquid water path (LWP). When this constraint is not applied, cloud albedo tends to increase with decreasing droplet size for optically thick clouds (optical depth >15), but decreases with decreasing droplet size for optically thin clouds [Han et al., 1998]. However, Quaas et al. [2004] found a very similar relationship between cloud top radius and aerosol index (product of AOD and Angstrom exponent) with and without constant LWP constraint. They did not discuss the effect of constraining LWP on cloud albedo. Jones et al. [2009] indicated that cloud albedo can either increase or decrease for thin clouds. The lack of consensus on this topic demonstrates the need for further investigation. [4] Han et al. [1994] conducted a global survey of cloud particle size and found that R e in continental water clouds are about 2 3 mm smaller than R e in marine clouds, and R e are about 1 mm smaller in marine clouds in the Northern Hemisphere than in the Southern Hemisphere. However, no hemispheric differences are observed in cloud fraction based upon multiyear MODIS measurements [Kishcha et al., 2009]. Recent studies have concentrated on the relationships between aerosol and cloud properties. Column aerosol number concentrations derived from four months of AVHRR aerosol optical depths (AOD) and Angstrom exponents over the global oceans have been correlated with cloud properties to identify the first AIE [Nakajima et al., 2001; Sekiguchi et al., 2003]; these studies found that R e is negatively correlated with column aerosol number concentration and cloud optical depth (COD) is positively correlated with column aerosol number concentration, but LWP is almost independent of aerosol number concentration. Note the aerosol and cloud retrievals used in these studies are based upon resolution AVHRR data. [5] Kaufman et al. [2005] pointed out that the coarse spatial resolution of AVHRR retrievals ( ) cannot resolve smaller clouds which are more susceptible to aerosol effect. They therefore analyzed 1 km resolution data from the MODerate resolution Imaging Spectroradiometer (MODIS) in four regions of the Atlantic ocean to study the aerosol effect on shallow water clouds. They found that cloud droplet size decreased while cloud fraction (F c ) and AOD increased, and concluded that most of these changes were associated with aerosol changes rather than meteorological conditions. Model simulations indicate that aerosolinduced changes are solely responsible for decreases in droplet size with increasing AOD, while increases in F c with increasing AOD are dominated by changes in the dynamical regime, not by aerosol indirect effects [Lohmann et al., 2006]. [6] To minimize the influence of large scale meteorology, Loeb and Schuster [2008] use a sampling strategy which sorts the aerosol and cloud properties within a 5 5 region into two populations according to whether AOD retrievals in 1 1 subregions are less than or greater than the 5 5 regional mean AOD. They find that the population with larger AOD have systematically higher cloud cover. They also find that the population with larger AOD have smaller cloud droplet effective radius only when the AOD differences between the two populations are larger than 0.08, whereas there are no significant differences in several meteorological parameters between the two populations. [7] The negative correlation between satellite retrieved aerosol properties and cloud droplet effective radius has been attributed to the first AIE, although this correlation was not derived under constant LWP [Nakajima et al., 2001; Breon et al., 2002; Sekiguchi et al., 2003]. However, cloud droplet size change should not be interpreted as observational evidence of the first AIE, because a reduction in droplet size by itself will cause a decrease in scattering cross section and cloud albedo [Han et al., 1998]. Cloud albedo will increase as cloud optical depth increase through the decrease of R e under constant LWP according to the following relationship [Stephens, 1978]: COD ¼ 3LWP 2 l R e ; where r l is the density of water. Note this equation is valid for clouds have a vertically uniform droplet size distribution. [8] Moreover, satellite retrieved aerosol and cloud properties have biases. Cloud free pixels used for aerosol retrievals are brightened (or shadowed) by reflected light from surrounding clouds [Cahalan et al., 2001; Wen et al., 2006, 2007; Varnai and Marshak, 2009]; this is the so called cloud adjacency (or 3 D) effect. Wen et al. [2006, 2007] used Monte Carlo simulations of TOA reflectance to show that the overall cloud 3 D effect is to enhance the reflectance in clear regions around broken clouds, which leads to systematically higher AOD estimates for pixels closer to clouds. Varnai and Marshak [2009] analyzed a large MODIS data set, and confirmed that cloud 3 D effect is responsible for a large portion of the enhanced clear sky reflectance near clouds. In addition, studies using airborne data seem to indicate that AOD enhancement in the vicinity of clouds from aerosol swelling and in cloud processing are much smaller than that from cloud 3 D effect [Su et al., 2008; Redemann et al., 2009]. The 3 D cloud induced enhancement of reflectance increases as surrounding COD and F c increase, and the enhancement is more pronounced at shorter wavelengths; consequently, this results in an enhancement of the Angstrom exponent. [9] Cloud retrievals are not immune from retrieval biases, either. Threshold retrievals of cloud properties assume that all cloud containing pixels are overcast, and tend to underestimate COD and overestimate R e for partially cloudy pixels [Han et al.,1994;coakley et al.,2005;matheson et al.,2006]. This overcast assumption results in biases of cloud properties that decrease as cloud cover increases. [10] The biases in aerosol and cloud retrievals discussed above, if not taken into account, can mistakenly be interpreted as aerosol indirect effect, or can cause aerosol indirect effect to be overestimated. For a region with a few clouds of a certain optical depth (and therefore dominated by partially cloudy pixels), the reflectance enhancement associated with the 3 D effect in the cloud free column is small. Hence, the AOD and Angstrom exponent retrievals are likely close to their true values (if there are no other sources of biases). Applying threshold cloud retrieval to partially cloudy pixels results in overestimated R e and underestimated COD. Adding clouds of the same optical depth and effective radius to this region (by increasing cloud fraction), the reflectance ð1þ 2of14

3 enhancement associated with the 3 D effect in the cloudfree column increases as well. This will cause the AOD and Angstrom exponent retrievals to be higher than their true values. However, increasing the cloud fraction also decreases the biases in cloud properties (as pixels become truly overcast and dominate the scene). Therefore as cloud fraction increases, retrieval biases will indicate larger apparent AOD, larger apparent Angstrom exponent, larger COD, but smaller R e, even if the actual aerosol and cloud properties (COD and R e ) remain constant. [11] This study aims to minimize the effects of satellite retrieval biases on aerosol induced changes in cloud properties and TOA albedo (a). We use back trajectory analysis to classify aerosols in broken low level cloud fields as oceanic or continental origin. Aerosol and cloud properties, and TOA albedo associated with oceanic and continental origin are stratified by vertical velocity at 700 hpa and estimated inversion strength to constrain the dynamic and thermodynamic conditions. These properties are further constrained by F c and LWP, which ensures the biases in these retrieved properties are nearly the same between the two origin populations, thus minimize the retrieval biases in the property differences between the two populations. It is expected that AODs, CODs and a associated with aerosols of continental origin are higher than their counterparts associated with aerosols of oceanic aerosols, whereas the reverse is true for R e. Estimating the differences between the two populations is the goal of this study. 2. Data and Method [12] We investigate the effects of aerosols on cloud properties and TOA albedo by combining radiative fluxes from the Clouds and the Earth s Radiant Energy System (CERES) instrument [Wielicki et al., 1996] with the cloud and aerosol properties retrieved from the MODIS measurements. These data sets are obtained from the Terra Rev1 Single Scanner Footprint TOA/Surface Fluxes and Clouds (SSF) product [Loeb et al., 2005], which includes cloud property retrievals from Minnis et al. [1998] and aerosol property retrievals from the MOD04 product [Remer et al., 2005]. We use daily 1 1 gridded data off the west coast of Africa (0 30 S and 15 W 10 E) for the years 2001 through The small study domain ensures minimum geographical impact on clouds, since cloud properties near the coast could differ from those in the open ocean (see more discussion of Figure 3). Aerosol retrievals within each 1 1 grid box are weighted by clear sky fraction and cloud retrievals within each 1 1 grid box are weighted by cloud fraction to provide the respective grid box mean values. We consider only grid boxes that have single layer low level clouds with both aerosol and cloud retrievals. Clouds with effective pressures larger than 680 hpa are defined as low level clouds. Cloud property retrievals for LWP 20 g m 2 are highly uncertain; therefore they are not included in this study. The upper limit of the LWP used in this study is 200 g m 2, though less than 1% of the data are associated with LWP > 120 g m 2. ECMWF ERA Interim reanalysis [Uppala et al., 2008], which is available at 6 h interval with grid size of , are used to monitor the dynamic and thermodynamic conditions. We limit our investigation to the months of April, May, and June (before the start of the biomass burning season) in order to minimize the effect of overlying absorbing aerosols on the cloud R e retrievals [Haywood et al., 2004]. [13] A back trajectory provides information about the movement of an air parcel through time before it reaches the grid box where the trajectory is initialized. We use trajectory analysis to classify aerosols as either oceanic or continental. We run parcel back trajectory using HYbrid Single Particle Lagrangian Integrated Trajectory Model (HYSPLIT) [Draxler and Rolph, 2003] for each 1 1 grid box in our study region. The trajectory calculations are started at the altitude of 1500 m and extend 72 h prior to the time of observations. We choose 72 h for the duration of back trajectory calculations based upon the residence time of tropospheric dust and sulfate aerosols provided by global models [Textor et al., 2006]. We classify the aerosols in a given grid box as oceanic if the air mass in the grid box was never over a continent in the previous 72 h; otherwise as continental. Hence, the back trajectory analysis separates the grid boxes into two populations of oceanic and continental origin. This approach is very different from previous studies which sort the cloud properties by aerosol without identifying the origin of the air masses [e.g., Nakajima et al., 2001; Sekiguchi et al., 2003; Lebsock et al., 2008]. [14] The utility of using back trajectories to separate oceanic aerosols from continental aerosols is demonstrated in Figure 1, which shows histograms of AOD (at 550 nm) and Angstrom exponent (calculated using AOD at 550 and 870 nm) associated with grid boxes of oceanic and continental origins over our study region. The total number of grid boxes classified as oceanic origin is 97,062, and the total number of grid boxes classified as continental origin is 13,396. The variability in oceanic AOD seen in Figure 1a is mainly caused by the fluctuating wind speed at the ocean surface, since sea salt aerosols are generated by sea spray at the ocean surface. Note that the grid boxes identified as having a continental origin tend to have higher AODs (mean = 0.26) than grid boxes with oceanic origin (mean = 0.11), which is consistent with the strong anthropogenic sources over the continents. The prominent anthropogenic aerosol type over the study period for this region is sulfate aerosols, since we chose the months before the start of the biomass burning season. Additionally, dust aerosols are located to the north of our study region and contribute very little to the continental aerosols. The mean AOD (0.11) of those grid boxes associated with oceanic origin agree well with the multiyear climatology from MODIS over tropical south Atlantic [Remer et al., 2008]. Angstrom exponent is often used as a qualitative indicator of aerosol particle size, with small values indicating large particles [Schuster et al., 2006]. The Angstrom exponent tends to be larger for grid boxes with continental influence (mean = 0.84) than those identified as oceanic origin (mean = 0.50); this indicates that the continental aerosols are smaller than the oceanic aerosols, as expected. [15] Cloud properties are affected by the dynamic and thermodynamic conditions. The dominate dynamical and thermodynamical parameters associated with cloud properties are vertical velocity and inversion strength [Bony et al., 2004; Bony and Dufresne, 2005; Klein and Hartmann, 1993; Wood and Bretherton, 2006; Medeiros and Stevens, 3of14

4 Figure 1. Probability distribution functions of aerosol optical depth and Angstrom exponent associated with oceanic origin (solid line with circles) and continental origin (dashed line with triangles). 2010]. The occurrence frequencies of different cloud regimes depend strongly on the large scale atmospheric circulation. We use the vertical velocity at 700 hpa (w 700 ) as a proxy for large scale ascending and descending motions to constrain the dynamic regimes. The validity of using w 700 to constrain dynamic regimes has been demonstrated extensively [e.g., Bony et al., 2004; Bony and Dufresne, 2005; Norris and Iacobellis, 2005; Medeiros and Stevens, 2010]. The amount of low level clouds is closely related to the inversion strength [Klein and Hartmann, 1993; Wood and Bretherton, 2006; Medeiros and Stevens, 2010], because inversion at the top of the boundary layer is the favorable condition for cloud formation and maintenance of existing clouds. We use the estimated inversion strength (EIS) to constrain the thermodynamic conditions [Wood and Bretherton, 2006]. Doing so, we assume that these two parameters can potentially capture the large scale meteorology. Indeed, Medeiros and Stevens [2010] demonstrated that inversion strength and vertical velocity can be used to categorize cloud and precipitation regimes. They found that the amount of low level clouds over the tropical oceans increases as LTS but is nearly independent of vertical velocity, and most of the precipitation is associated with the strongest upward motion but is almost independent of LTS. ERA Interim reanalysis data are used to obtain w 700 and to calculate EIS, and these two parameters are matched to the Terra observational time to within 3 h. Lower tropospheric stability (LTS) can also be used to constrain thermodynamic state [Klein and Hartmann, 1993; Matsui et al., 2004; Lebsock et al., 2008], but Wood and Bretherton [2006] demonstrated that EIS is a better predictor of low level cloud amount than LTS under a wide range of climatological conditions. [16] Ultimately, we wish to separate the meteorological effects from the aerosol effects on cloud properties and TOA albedo. We do this by using w 700 and EIS to stratify the cloud properties and TOA albedo for our two classifications of oceanic and continental aerosols. The data are stratified by three ranges of w 700 (hpa d 1 ): 60 to 0 for convective regime, 0 to 60 for subsidence regime, and 60 to 120 for strong subsidence regime; and by three ranges of EIS (K): 0 to 2, 2 to 4, and 4 to 6, with larger EIS corresponding to more stable conditions. There are about 20% of the data fall in the convective regime when using 6 hourly reanalysis for the study region, though when the reanalysis data are averaged to monthly time scales, the convective regime and strong subsidence regime do not exist. Means and standard errors are calculated for each EIS w 700 bin for AOD, cloud properties, and TOA albedo associated with continental and oceanic origin. Standard error is calculated as the standard deviation divided by the square root of the observational days of a given bin. 3. Meteorological Effects on Cloud Properties [17] To illustrate the meteorological effects on cloud properties, we use the grid boxes in the study domain that are associated with aerosols of oceanic origin. First we only use grid boxes with AODs less than 0.2, which accounts for 87% of the total oceanic samples. The corresponding mean AOD, R e, F c, LWP, a, and occurrence frequency of the nine EIS w 700 bins are shown in Figure 2. There are very small AOD differences across the EIS w 700 domain (Figure 2a). For a given w 700, R e is smaller in more stable conditions (larger EIS). This demonstrates that R e is sensitive to thermodynamic conditions, since the AODs are nearly identical across the EIS domain. Matsui et al. [2004] and Lebsock et al. [2008] also observe that R e decreases as LTS increases under constant aerosol indices, because a strong inversion corresponds to weak updraft within the cloud layer, which inhibits cloud droplet growth. The sensitivity of R e to w 700 was not included in previous studies. Figure 2b shows that R e tends to be larger in ascending air. This is because clouds in a more convective regime (w 700 < 0) tend to be higher than clouds in a subsidence regime (w 700 > 0), and the cloud droplet spectrum shifts toward larger sizes at higher altitude [Rogers and Yau, 1989]. Additionally, the notable LWP gradient across w 700 domain could also contribute to the R e gradient. [18] One could also argue that clouds closer to the coast have smaller R e than clouds further out to the sea, therefore the longitude distribution of each EIS w 700 bin could contribute to the R e gradient. Figure 3 shows the longitude distributions of three w 700 bins for EIS ranges from 2 to 4 K and the longitude distributions of three EIS bins for w 700 ranges from 0 to 60 hpa d 1. The longitude distributions of the three w 700 bins and the three EIS bins are quite similar in general, though the smallest w 700 bin has a slightly higher 4of14

5 Figure 2. Grid boxes of oceanic origin with aerosol optical depths less than 0.2 are used to calculate the means of (a) aerosol optical depth, (b) cloud droplet effective radius (mm), (c) cloud fraction (%), (d) LWP (g m 2), (e) TOA albedo, stratified by estimated inversion strength (EIS) and vertical velocity at 700 hpa (w700), and (f) the occurrence frequency (%) of each bin. frequency while the smallest EIS bin has a slightly lower frequency near the coast. However, we do not expect these small longitude distribution differences affect the Re distribution across the EIS w700 domain. [19] Figure 2c shows that Fc tends to be larger in stable conditions. This is because moisture evaporated from the sea surface will accumulate and gradually reach saturation when there is an inversion at the top of the boundary layer; and once single layer low level clouds have been formed, convection is easily maintained due primarily to the strong radiative cooling at the cloud top [Klein and Hartmann, 1993]. However Fc is not very sensitive to w700. LWP increase as w700 decrease, because convective clouds are generally thicker, and LWP is larger under more stable Figure 3. Grid boxes of oceanic origin with aerosol optical depths less than 0.2 are used to calculate the longitude distributions of vertical velocity at 700 hpa (w700) and estimated inversion strength (EIS): (a) three w700 bins for EIS ranges from 2 to 4 K and (b) three EIS bins for w700 ranges from 0 to 60 hpa d 1. 5 of 14

6 Figure 4. Differences of (a) aerosol optical depth (AOD), (b) cloud droplet effective radius (mm), (c) cloud fraction (%), (d) LWP (g m 2 ), and (e) TOA albedo ( 10 3 ), stratified by estimated inversion strength (EIS) and vertical velocity at 700 hpa (w 700 ). The difference is defined as property associated with all AOD minus property associated with AOD < 0.2. (f) The occurrence frequency (%) associated with AOD >0.2 of each bin. Bins shown in white indicate that the differences are not significant. conditions (larger EIS) than under less stable conditions. TOA albedo increase as EIS increase, largely caused by the relationship between F c and EIS. In addition, albedo also increase slightly as w 700 decrease. [20] We also calculate the means of the relevant variables without constraining the AODs, and compare these with results in Figure 2. Differences for AOD, R e, F c, LWP, and albedo are shown in Figure 4 (along with the occurrence frequency of the bins with AOD > 0.2). Bins that are shown in white indicate differences that are not significant (differences are smaller than the standard errors of the differences). By definition, unconstrained AODs are larger than the constrained AODs across all EIS w 700 bins. The AOD differences are larger for less stable conditions than for more stable conditions (larger EIS). We also observe significant reductions in R e when we do not constrain AODs. The reduction is greater for less stable conditions than for more stable conditions, which coincides with the large increase of AOD and reduction in LWP under less stable conditions. Cloud fraction increases after we remove the constraint on AOD (except for one bin), and the increment is larger for less stable conditions than for more stable conditions. Note the F c are very similar with and without the constraint on AOD (differences are less than 4%, Figure 4c). It is likely that the retrieval biases in R e for these two cases are similar, because the magnitude of the threshold retrieval biases depends on F c. Hence, the R e differences are unlikely affected by the retrieval biases. We observe a slight LWP reduction associated with increased AODs for less stable conditions, and no significant LWP changes for more stable conditions. TOA albedo also increases as AOD increases except for one bin, with larger albedo enhancement for less stable conditions than for more stable conditions. [21] These differences could be explained by aerosol indirect effects. More aerosols lead to more cloud condensation nuclei, therefore smaller R e under constant LWP conditions (first aerosol indirect effect). Furthermore, smaller R e results in a reduced precipitation formation rate and potentially an enhanced cloud lifetime and F c (second aerosol indirect effect). However, meteorological effects could also play a role. Most of the AODs that are larger than 0.2 are associated with the less stable conditions, which tend to be associated with greater air sea exchange and vertical transport. Strong vertical mixing could mean more entrainment of ambient dry air, and therefore more evaporation which would lead to smaller R e and LWP. More evaporation would also indicate reduced F c, but this is not the case (Figure 4c). [22] We calculate the slopes of the linear fit between logarithm of these variables and the logarithm of AOD for each EIS w 700 bin. The slopes show large sensitivity to EIS but only vary slightly among w 700 regimes for a given EIS range. Hence, we provide the slopes and their 95% confidence intervals for the three EIS ranges in Table 1. The 6of14

7 Table 1. Slopes and Their 95% Confidence Intervals of Linear Fit Between Logarithm of Relevant Variables and Logarithm of AOD for Three Estimated Inversion Strengths a EIS, K R e F c LWP Albedo [0,2] 0.26 ± ± ± ± 0.02 [2,4] 0.21 ± ± ± ± 0.03 [4,6] 0.14 ± ± ± ± 0.03 a EIS, estimated inversion strength; AOD, aerosol optical depth; R e, cloud droplet effective radius (mm); F c, cloud fraction (%); LWP, liquid water path (g m 2 ). slopes between ln(r e ) and ln(aod) are all negative and the magnitude of the slope is larger under less stable conditions. The slopes between ln(f c ) and ln(aod) are all positive and the magnitude of the slope is greater under more stable conditions. Interpretation of this strongly positive relationship between F c and AOD remains controversial. Four potential reasons (second AIE, covariance of large scale meteorological dynamics, aerosol swelling in high relative humidity environment, and cloud 3 D effect) responsible for this relationship are discussed by Quaas et al. [2009]. The slopes of ln(a) and ln(aod) are all positive and also increase as EIS increases, since a is very sensitive to F c. The slopes between ln(lwp) and ln(aod) are either negative or not significant. [23] Under less stable conditions, there is more entrainment of dry air above the boundary layer. Increased aerosols do get mixed throughout the boundary layer because of strong vertical motion, therefore R e is more sensitive to aerosol changes than under more stable conditions. Additionally, reductions of LWP as aerosol loading increases can also suppress cloud droplet growth resulting in smaller R e. Cloud fraction is to first order controlled by mixing of dry air from above, so it is less sensitive to aerosols under these conditions. Under more stable conditions, only a small fraction of the aerosols reaches the cloud top because of the strong inversion, therefore satellite retrieved R e is less affected. However, the cloud droplet size distribution in the lower part of the cloud layer does interact with the aerosol, resulting in less drizzle. Hence, a relatively small increase in LWP (which could also contribute to the smaller sensitivity of R e to AOD under stable conditions), and a relatively large increase in cloud fraction. 4. Mean Aerosol and Cloud Properties Associated With Oceanic and Continental Origin [24] We have shown the effects of increased oceanic aerosol loading on cloud properties and TOA albedo. We also demonstrated that R e is sensitive to both EIS and w 700, and F c is sensitive to EIS. Hence, it is important to stratify the aerosol/cloud properties and TOA albedo by EIS and w 700 to constrain the dynamic and thermodynamic conditions. [25] We now consider cloud properties and TOA albedo associated with aerosols of continental origin and compare these with those that are associated with aerosols of oceanic origin. We restrict our analysis to a small domain (0 25 S and 0 10 E) along the coast, where most of the continental aerosols are identified. For this small domain, the total number of grid boxes classified as oceanic origin is 22,785, and the total number of grid boxes classified as continental origin is 10,839. Figure 5 shows the mean AODs associated with oceanic and continental aerosols stratified by EIS and w 700 (Figure 5, top). Note that the continental AODs (Figure 5, top right) are larger than the oceanic AODs (Figure 5, top left) across all EIS w 700 bins by about 50%. AOD is also sensitive to EIS, as both oceanic and continental aerosols indicate larger AODs for less stable conditions. This could be caused by the minimized air sea exchange and vertical turbulence transport under stable conditions. Finally, we do not observe any consistent patterns across the w 700 regime in either frame. [26] Swelling of hygroscopic aerosols under high relative humidity (RH) conditions tends to increase AOD. Can we attribute the differences between oceanic and continental AOD and the distinct gradients of oceanic and continental AOD across the EIS regime to the differences in RH? Mean RH at 700 hpa from ERA Interim reanalysis associated with oceanic and continental grid boxes stratified by EIS and w 700 are shown in Figure 5 (bottom). Note we only consider grid boxes with single layer low level clouds, therefore the RH at 700 hpa represents the clear sky humidity above the cloud top. The RH associated with aerosols of oceanic origin ranges from 32% (under less stable condition) to 12% (under more stable condition), and the RH associated with aerosols of continental origin ranges from 43% to 17%. Aerosol swelling under the low humidity environment (i.e., from 15% to 45%) results in a rather small increase ( 5%) in extinction according to our calculation. This small increase in extinction due to RH change is not sufficient to explain neither the AOD difference between continental and oceanic origin nor the AOD gradients across the EIS regimes for both oceanic and continental origin. Furthermore, we note that the RH associated with gird boxes of continental aerosols is higher than that associated with grid boxes of oceanic aerosols. Large eddy simulations indicate that the edges of polluted clouds are more humid than those of clean clouds because of higher evaporation rate associated with polluted cloud droplets [Jiang et al., 2009], which lend credence to our results. [27] Figure 6 shows the mean R e and LWP associated with aerosols of oceanic and continental origin (their aerosol optical depths are shown in Figure 5), stratified by EIS and w 700. The mean R e associated with oceanic aerosols are larger than the mean R e associated with continental aerosols for all the EIS w 700 bins, consistent with expectation. Likewise, the mean LWP associated with oceanic aerosols are generally larger than the mean LWP associated with continental aerosols, except for one EIS w 700 bin. The overlying air is very dry for both oceanic and continental grid boxes (Figure 5). Ackerman et al. [2004] pointed out that in this case LWP is reduced as cloud droplet concentrations increase because the cloud top entrainment dominates. We observe that R e increase with increasing stability (i.e., as EIS increases) for both continental and oceanic aerosols, and the LWP increase with increasing stability for two out of the three w 700 ranges for both continental and oceanic aerosols. The increases in R e as EIS increase are likely caused by the decreases in AOD as EIS increase (Figure 5, valid for all of the w 700 ranges) and the increases in LWP as EIS increase (valid for some of the w 700 ranges). Decreases in AOD result in less cloud 7of14

8 Figure 5. (top) Aerosol optical depth and (bottom) relative humidity (%) associated with aerosols of (left) oceanic origin and (right) continental origin, stratified by estimated inversion strength (EIS) and vertical velocity at 700 hpa (w 700 ). condensation nuclei and increases in LWP supply the water needed for cloud droplet growth. Hence, larger droplet size under more stable conditions. In addition, entrainment of ambient dry air caused by strong vertical mixing under less stable conditions could also contribute to the smaller R e under less stable conditions than under more stable conditions. 5. Differences Between Continental and Oceanic Origins [28] The mean AOD and R e associated with aerosols of continental origin are different from their counterparts associated with aerosols of oceanic origin. If we assume that the continental population is dominated by anthropogenic aerosols, we can assess the anthropogenic aerosol effect on cloud properties and TOA albedo. Hence, we seek to identify measurable differences between the populations of oceanic and continental aerosols. We define the difference of a given property as: DX i ¼ X c i X o i ; where X can represent any measurable (AOD, R e, etc.), the subscript i denotes each EIS w 700 bin, and the superscripts ð2þ c and o denote grid boxes of continental or oceanic origin. The standard error of the difference is defined as: qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi i ¼ ð c i Þ2 þð o i Þ2 ; ð3þ where s is the standard error for the given population. If a given property exhibits DX i > s i, we regard the difference as statistically significant. One advantage of using this approach is that retrieval biases are generally similar for the two populations, so the difference is essentially free from retrieval biases Differences Derived Without Constraining the Cloud Fraction [29] First we calculate the differences between the two populations without restricting the cloud fractions in either of them. As the first AIE is established under constant LWP [Twomey, 1977], we stratify the differences by LWP to test the first AIE. Four LWP ranges are used: 20 40, 40 60, 60 80, and >80 g m 2. Figure 7 shows D R e corresponding to DAOD for three LWP ranges; note the significant reduction in R e associated with aerosols of continental origin for all EIS w 700 bins and all LWP ranges. The reduction is 1 2 mm for LWPs between 20 and 40 g m 2, and increases up to 8of14

9 Figure 6. (top) Cloud droplet effective radius (mm) and (bottom) liquid water path (g m 2 ) associated with aerosols of (left) oceanic origin and (right) continental origin, stratified by estimated inversion strength (EIS) and vertical velocity at 700 hpa (w 700 ); corresponding aerosol optical depths are shown in Figure 5. 4 mm for LWPs between 60 and 80 g m 2. The reduction in R e is almost independent of the increment in AOD for the smallest LWP range considered here. One possible explanation is the simultaneous increase of CCN and relative dispersion () as DAOD increases [Liu and Daum, 2002]. Increase in " tends to increase R e while increase in CCN tends to decrease R e. Under small LWP, these two competing effects compensate each other and therefore result in a near constant R e reduction. There is no clear dependence of R e reductiononeisandw 700. Lohmann et al. [2006] simulated the changes in R e from the smallest to highest AOD using a global model with and without the aerosol Figure 7. Cloud droplet effective radius (R e ) differences versus aerosol optical depth (AOD) differences for three estimated inversion strength (EIS) bins and three 700 hpa vertical velocity (w 700 ) bins. Difference is defined as continental origin minus oceanic origin. Difference is calculated by constraining the liquid water path (LWP, in unit of g m 2 ) but not the cloud fraction (F c ). Symbols are the mean differences, horizontal bars are the standard errors of AOD difference, and the vertical bars are the standard errors of R e difference. 9of14

10 Table 2. Averaged Differences Calculated for Four Liquid Water Path Ranges Without Restricting the Cloud Fractions a Variables 20 < LWP < LWP < LWP 80 LWP > 80 DAOD 0.13 ± ± ± ± 0.03 DR e (mm) 1.6 ± ± ± ± 0.9 DCOD 0.78 ± ± ± ± 0.91 DF c (%) 15.0 ± ± ± ± 3.9 DLWP (g m 2 ) 0.5 ± ± ± ± 7.1 Da ± ± ± ± n l a Averaged differences (as defined in equation (4)) between variables associated with continental origin and associated with oceanic origin and the standard errors for these differences. LWP, liquid water path (g m 2 ); AOD, aerosol optical depth; R e, cloud droplet effective radius; COD, cloud optical depth; F c, cloud fraction; a, TOA albedo; n l, occurrence frequency of each LWP range. indirect effect, in an effort to disentangle the microphysical from the dynamical effect. They found that aerosol induced changes are solely responsible for the decrease in R e with increasing AOD. This finding could explain the lack of sensitivity of R e reduction to EIS and w 700 shown in Figure 7. [30] Based upon the difference for every EIS w 700 bin under each LWP range, we calculate the averaged difference under each LWP range from: DX j ¼ð X9 i¼1 m i DX i Þ j ; where DX i is the difference for every EIS w 700 bin (denoted by the subscript i) under a given LWP range (denoted by subscript j), and n i m is the occurrence frequency of every EIS w 700 bin under the given LWP range. Since the focus here is to evaluate the effects of polluted continental aerosols on cloud properties and TOA albedo in our study region, we calculate n i m from EIS w 700 bins associated with aerosols of continental origin. [31] Averaged differences of AOD, R e, COD, F c, LWP, and a are listed in Table 2. Note that the averaged R e reduction associated with continental aerosols is 1.6 mm for the g m 2 LWP range, and increases to 3.3 mm for the g m 2 LWP range. According to equation (1), reduction in R e will cause an increase in COD associated ð4þ with continental aerosols under constant LWP (which is indeed the case). Table 2 also indicates that grid boxes associated with aerosols of continental origin have significantly higher cloud fractions than grid boxes associated with aerosols of oceanic origin under all LWP ranges, which is consistent with the finding of Mauger and Norris [2007]. Recall that the retrieval biases in AOD, COD and R e are sensitive to cloud fraction in the grid box, so higher cloud fraction in one population than the other could skew the differences (since the biases in one population are larger or smaller than the biases in the other, and differencing the two will not cancel out the biases). Furthermore, higher cloud fraction in continental population also contributes to the albedo difference, which makes it difficult to attribute the albedo enhancement to the first AIE. Hence, we further restrict the analysis to those grid boxes that have cloud fractions larger than 80%, which reduces the sample numbers to 7392 and 3958 for the oceanic and continental origin Differences Derived Under Constrained Cloud Fraction [32] Figure 8 shows DR e corresponding to DAOD using data from grid boxes with F c > 80%. Note only those EIS w 700 bins that consist of at least 10 days of observations are included in the difference figures. Compared to Figure 7, the DR e shown in Figure 8 are reduced by 1 mm for all Figure 8. Cloud droplet effective radius (R e ) differences versus aerosol optical depth (AOD) differences for three estimated inversion strength (EIS) bins and three 700 hpa vertical velocity (w 700 ) bins. Difference is defined as continental origin minus oceanic origin. Difference is calculated by constraining the liquid water path (LWP, in unit of g m 2 ) and the cloud fraction (F c > 80%). Symbols are the mean differences, horizontal bars are the standard errors of AOD difference, and the vertical bars are the standard errors of R e difference. 10 of 14

11 Figure 9. Cloud optical depth (COD) differences versus aerosol optical depth (AOD) differences for three estimated inversion strength (EIS) bins and three 700 hpa vertical velocity (w 700 ) bins. Difference is defined as continental origin minus oceanic origin. Difference is calculated by constraining the liquid water path (LWP, in unit of g m 2 ) and the cloud fraction (F c > 80%). Symbols are the mean differences, horizontal bars are the standard errors of AOD difference, and the vertical bars are the standard errors of COD difference. LWP ranges. This accounts for about 50% decrease in D R e for LWPs less than 60 g m 2, and about 30% decrease in DR e for LWPs of g m 2. This is because a large percentage of data are from grid boxes with F c < 80% for small LWP, which is not the case for large LWP. [33] Figure 8 confirms that the addition of polluted continental aerosols over ocean leads to reduced R e. Although the reduction in R e demonstrates that aerosols significantly affect the microphysical properties of clouds, the Twomey effect of enhanced cloud albedo can only be realized by increased cloud optical depth. Hence, Figure 9 shows the DCOD corresponding to DAOD for three LWP ranges derived using grid boxes with F c > 80%. Grid boxes with aerosols of continental origin have significantly higher COD than those with aerosols of oceanic origin for all LWP ranges (except for two cases), and DCOD increases as LWP increases. Large DCOD are generally associated with small EIS bins (less stable condition), and these differences are about 0.4 to 1.0 smaller than those obtained without the cloud fraction constraint (not shown). [34] These changes in COD produce significant changes in the TOA albedo calculated from the CERES measurements. Figure 10 shows the TOA albedo differences (Da) corresponding to AOD differences for three LWP ranges using grid boxes with F c > 80%. The albedo differences mirror the COD differences remarkably well. This lends credence to the COD differences that we derive, given that the albedo is calculated from CERES TOA upwelling fluxes and COD is retrieved from MODIS reflectances. TOA albedos of grid boxes containing aerosols of continental origin are significantly higher than those of grid boxes containing only aerosols of oceanic origin, except for one case in each LWP range. [35] We also calculate the averaged differences of the relevant variables for each LWP range with the F c > 80% constraint (see equation (4)); the results are listed in Table 3. The averaged reductions in R e and enhancements in COD are much smaller (by 20% 100% comparing to Table 2) after we constrain the analysis to grid boxes with F c > 80%. Averaged differences of TOA albedo (a) listed in Tables 2 and 3 differ by a factor of 2 to 4. The TOA albedo numbers are derived from CERES measurements, and are not subject to biases that affect the R e and COD retrievals. The overestimation of the a increment when we do not constrain F c (Table 2) is mainly due to the significantly higher cloud fractions of grid boxes associated with aerosols of conti- Figure 10. TOA albedo differences versus aerosol optical depth (AOD) differences for three estimated inversion strength (EIS) bins and three 700 hpa vertical velocity (w 700 ) bins. Difference is defined as continental origin minus oceanic origin. Difference is calculated by constraining the liquid water path (LWP, in unit of g m 2 ) and the cloud fraction (F c > 80%). Symbols are the mean differences, horizontal bars are the standard errors of AOD difference, and the vertical bars are the standard errors of albedo difference. 11 of 14

12 Table 3. Averaged Differences Calculated for Four Liquid Water Path Ranges Using Grid Boxes With Cloud Fractions Larger Than 80% a Variables 20 < LWP < LWP < LWP 80 LWP > 80 DAOD 0.09 ± ± ± ± 0.03 DR e (mm) 0.7 ± ± ± ± 1.0 DCOD 0.41 ± ± ± ± 0.96 DF c (%) 1.6 ± ± ± ± 1.4 DLWP (g m 2 ) 0.7 ± ± ± ± 8.0 Da ± ± ± ± n l a Averaged differences (as defined in equation (4)) between variables associated with continental origin and associated with oceanic origin and the standard errors for these differences. LWP, liquid water path (g m 2 ); AOD, aerosol optical depth; R e, cloud droplet effective radius; COD, cloud optical depth; F c, cloud fraction; a, TOA albedo; n l, occurrence frequency of each LWP range. nental origin than of grid boxes associated with aerosols of oceanic origin. After we constrain the analysis to grid boxes with F c > 80%, the cloud fractions differ by % under each LWP range. Our radiative transfer calculations indicate that a 2% cloud fraction difference will cause a difference in TOA albedo, which is within the standard error under each LWP range. Differences of R e and COD listed in Table 2 are larger than those in Table 3. This could be caused by larger retrieval biases in the oceanic population (lower F c associated with aerosols of oceanic origin than with continental origin when we do not constrain F c ). It also could be resulted from larger susceptibility of shallow cumulus than of stratocumulus clouds [Oreopoulos and Platnick, 2008]. [36] Tables 2 and 3 show averaged differences weighted by EIS w 700 occurrence frequencies for several LWP ranges; this demonstrates the effect of continental aerosols on maritime clouds for clouds with similar meteorological conditions. Next, we determine the overall effect of continental aerosols on clouds in the region by weighting the averaged differences with the LWP occurrence frequency: DX ¼ X4 j¼1 l j DX j; l where DX is the overall difference, and n j is the LWP occurrence frequency. In this case, we derive the LWP occurrence frequencies using all low level clouds associated with continental aerosols over the tropical south Atlantic ocean without requiring concurrent aerosol retrievals, and therefore include overcast cases as well. The occurrence frequencies for the four LWP bins are listed in Tables 2 and 3. These four LWP bins account for 45% of the population, and the rest (55%) of the population is associated with LWP less than 20 g m 2. The smallest LWP bin (<20 g m 2 ) is not included in the overall differences because of the large cloud retrieval uncertainties associated with low LWP. However, the differences are very small for this LWP bin when we constrain cloud fraction, so exclusion of these LWP has very small effect on the overall differences. [37] The overall differences of R e, COD, F c, LWP, and a derived with and without constraining the cloud fractions are listed in Table 4. Recall negative (positive) differences indicate that property associated with aerosols of continental origin is smaller (larger) than that associated with aerosols of oceanic origin. Since all differences are derived by constraining the LWP, it is expected that the LWP of the two ð5þ populations exhibit no difference (which is indeed the case). We observe significant reduction in R e, and significant increase in COD and a. The overall differences in R e, COD, and a are larger when we do not constrain cloud fraction than when we do constrain cloud fraction (>80%). That is, the overall difference in R e is reduced from 1.0 to 0.5 mm after we constrain the cloud fraction, and the overall difference in COD is reduced from 0.6 to 0.4 and the overall difference in a is reduced from to [38] We note that grid boxes associated with aerosols of continental origin have higher cloud fractions than grid boxes associated with aerosols of oceanic origin by about 10% (Table 2), which contributes to the large differences in R e, COD, and a in Table 4 when we do not constrain cloud fraction. However, the two populations have very similar cloud fractions after we limit our analysis to grid boxes with F c > 80% (Table 3). Hence, the differences of R e, COD, and a that we derived with constrained cloud fraction are very unlikely biased by retrieval artifacts, and most likely represent the effect of anthropogenic aerosols on cloud properties and TOA albedo under high cloud cover conditions. Note we only address the effect of anthropogenic aerosols on R e and COD under constrained LWP, the possible aerosol effects on LWP, F c, and cloud height are not included. 6. Conclusions [39] Satellite retrieved AOD, COD, and R e are frequently used to study the possible effect of increased aerosol burden on cloud properties. However, retrievals of AOD, COD, and R e are subject to biases depending on the cloud fraction Table 4. Overall Differences Calculated Without Restricting F c in Grid Boxes and With Restricting F c in Grid Boxes to Be Larger Than 80% a Variables All F c F c > 80% DR e (mm) 1.0 ± ± 0.2 DCOD 0.62 ± ± 0.15 DF c (%) 5.9 ± ± 0.5 DLWP (g m 2 ) 0.0 ± ± 0.9 Da ± ± a Overall differences (as defined in equation (5)) between variables associated with continental origin and associated with oceanic origin and the standard errors for these differences. R e, cloud droplet effective radius; COD, cloud optical depth; F c, cloud fraction; LWP, liquid water path; a, TOA albedo. 12 of 14