CALIPSO observations of transatlantic dust: vertical 1 stratification and effect of clouds 2

17 We use CALIOP nighttime measurements of lidar backscatter, color and depolarization ratios 18 during the summer of 2007 to study transatlantic dust properties downwind of Saharan 19 sources, and to examine the interaction of clouds and dust. Our analysis suggests that (1) 20 while lidar backscatter doesn’t change much with altitude in the Saharan Air Layer (SAL), 21 depolarization and color ratios both increase with altitude in the SAL; (2) lidar backscatter 22 and color ratio increase as dust is transported westward in the SAL; (3) the vertical lapse rate 23 of dust depolarization ratio, introduced here, increases within SAL as plumes move westward; 24 (4) nearby clouds barely affect the backscatter and color ratio of dust volumes within SAL but 25 not so below SAL. Moreover, the presence of nearby clouds tends to decrease the 26 depolarization of dust volumes within SAL. Finally, (5) the odds of CALIOP finding dust 27 below SAL next to clouds are about 2/3 of those far away from clouds. This feature, together 28 with an apparent increase in depolarization ratio near clouds, indicates that particles in some processes. The study examines four characteristics of dust layers: ( i ) the volume of air containing dust, ( ii ) lidar backscatter (related to optical thickness), ( iii ) color 30 ratio (roughly proportional to particle size) at least for spherical particles, and ( iv ) 31 depolarization ratio (characterizing particle shape, with larger values for irregular particles

). The answer to this question is likely not only to 2 improve our understanding of dust-cloud interactions but also yield better estimates of direct 3 radiative forcing. To that end, we present analysis of dust properties over the North Atlantic 4 Ocean-including near-cloud behavior-based on Cloud-Aerosol Lidar with Orthogonal 5 Polarization (CALIOP) data (Winker, et al. 2003). 6 CALIOP is a space based lidar system onboard the Cloud Aerosol Lidar and Infrared 7 Pathfinder Satellite Observations (CALIPSO) satellite launched in 2006. CALIOP data offers 8 many advantages for this study. First, since CALIOP uses a laser with a small footprint (~90 9 m in diameter on the ground), its aerosol data is not affected by the 3D radiative 10 enhancements of nearby clouds, which cause complications for instruments observing 11 reflected sunlight (Wen et al., 2007;Marshak et al., 2008;Várnai and Marshak, 2011). 12 Second, CALIOP provides backscatter depolarization information at 532 nm, which allows 13 one to distinguish (typically) non-spherical dust from (typically) spherical droplets (Sassen,   This paper characterizes dust particles using their attenuated backscatter coefficient (βʹ′), 29 attenuated color ratio (χʹ′), and depolarization ratio (δʹ′) values. Unless specified otherwise, βʹ′ 30 is the median of the vertically averaged 532 nm attenuated backscatter coefficients within 31 To discern changes in dust properties during transatlantic transport, we examine dust behavior 1 in the three regions shown in Figure 1: East (E) (0°-30° W), Middle (M) (30°-60° W) and 2 West (W) (60°-90° W). These three regions lie at different distances from the African dust 3 sources, and cover most of the dust paths from Africa to America during the summer of 2007. 4 5 3 Results and discussion 6 The spatial and optical characteristics of African dust vary during the transatlantic journey (e. 7 g., Liu et al., 2008;Huang et al., 2010). In this section we examine the variations in three 8 steps. First, the overall statistics of dust properties in the three geographic regions are 9 compared. This part focuses on the vertical distribution of dust samples in the CALIOP 5 km 10 resolution aerosol product, and on the vertical distributions of attenuated backscatter 11 coefficient, color ratio and depolarization ratio. We then analyze the relationships between 12 dust properties and cloud coverage in the three regions. Finally, we discuss the systematic 13 changes in dust properties that occur near clouds. 14 product, contain dust and have a CAD value between -70 and -100. In the Eastern (E) region, 20 more than 80% of dust is between 1.5 km and 5.5 km altitude, with the peak probability 21 around 3.5 km. This elevated dust distribution is a typical result of two confining inversions 22 below and above the SAL (Carlson and Prospero, 1972). The dust remains elevated in the 23 middle (M) region as well, although the mean elevation descends about 0.5 km. The similar 24 patterns in the E and M regions indicate that the driving forces keeping the dust at high 25 altitudes in region E persist in region M as well. If a 3-day average transport time from 26 Region E to M is assumed, the descending velocity from center of E to center of M is 27 estimated around 1.7 mm/s, which is consistent with the typical SAL average descending 28 velocity of 1-2 mm/s (Carlson and Prospero, 1972). Finally, in the West (W) region the 29 chances of finding dust decrease steadily with altitude, and dust is rarely found above 5 km. This section examines dust optical properties in the three regions. Since dust above 5 km is 7 rare, the analysis of dust optical properties will be limited to altitudes below 5 km. The 8 analysis uses the Level 2 CALIOP aerosol layer product, which provides averaged βʹ′, χʹ′, and 9

Dust properties in the three regions
δʹ′ values for each dust layer. These layer-average values are assigned to all altitude bins 10 within a dust layer when creating Figure 2 (and 4). 11 The results in Figure 2 show that dust properties vary with altitude differently within SAL 12 (from 1.5 km to 5 km in altitude of regions E and M), below the SAL (below 1.5 km in 13 regions E and M), and in region W. 14 In the SAL, median βʹ′ 532 values are nearly constant with altitude, but the medians of χʹ′ and δʹ′ 15 increase with altitude. Since the typically large and non-spherical dust particles imply large χʹ′ 16 and δʹ′ values, the observed dusty volume behavior suggests that the concentration of dust 17 increases with altitude and/or the concentration of non-dust marine aerosols in dusty volumes 18 decreases with altitude. 19 Below the SAL in the E and M regions, the median δʹ′ increases with altitude. This is the 20 result of dust mixing with non-dust marine aerosols in the moist air confined between the 21 marine surface and the inversion created by the dry and warm SAL aloft. Since the 22 concentration of wet marine aerosols is much higher below than inside the SAL, backscatter 23 from these aerosols contributes significantly to the lidar signals and reduce the depolarization 24 ratios of dusty volumes below the SAL. As discussed in the following paragraph, the mixing 25 of dust and marine aerosols in region W could be the major reason behind higher backscatter 26 and lower depolarization in region W than in E or M. 27 In region W, the median of βʹ′ 532 decreases with altitude and the medians of χʹ′ and δʹ′ increase 28 with altitude up to at least 4-5 km. This behavior arises from the dissipation of the SAL in the 29 region. As the SAL extends westward, its temperature drops and convection can bring in 30 7 moist air from below. Eventually the inversion confining the moist marine air at low altitudes 1 breaks up in region W, which allows the moist marine air to reach much higher altitudes. 2 Even so, the concentration of wetter and larger marine aerosols tends to decrease with 3 altitude, and so their contribution to the lidar backscatter of dusty volumes tends to be smaller 4 at higher altitudes. This results in higher median χʹ′ and δʹ′ values at higher altitudes even in 5 region W. 6 A comparison of dust properties in regions E, M, and W also reveals several features related 7 to SAL influences at different transport stages. For example, at most altitudes the medians of 8 βʹ′ 532 and χʹ′ increase westward, whereas the median of δʹ′ tends to decrease westward. These 9 trends arise from the fading of the SAL during the westward transport: As the air moves from 10 west Africa to the Caribbean, the SAL temperature keeps decreasing, which allows the 11 concentration of moist marine air mixed in from below to keep increasing. The presence of 12 more moist marine aerosols at west increases the median βʹ′ 532 and χʹ′ of dusty volumes and 13 reduces the median δʹ′ value. 14 However, some features in Figure 2 cannot be explained by contributions from non-dust 15 marine aerosols in dusty sample volumes, and are likely caused by changes in the properties 16 of dust particles instead. For example, Figure 2d shows that above 3.5 km, the median δʹ′ is 17 larger in region M than in region E. This cannot be explained by mixing from below, because 18 the larger contribution of mixed-in marine aerosols in region M would imply smaller (rather 19 than larger) δʹ′ values. Instead, the observed tendency is likely related to lower fall speed for 20 aspherical dust particles: as the more spherical particles fall faster, this leaves an increasingly 21 non-spherical dust population at high altitudes as the air moves to region M. The plausibility 22 of this scenario is also supported by simulations for highly irregular particles falling slower 23 than more spherical ones, because of greater air resistance (Ginoux, 2003). This issue will be 24 further discussed in Section 3.3.2. 25 We note that the slight increase of χʹ′ of dusty volumes from E to M (in Figure 2c This section examines the relationships between dust properties and cloudiness in the three 2 study regions. We characterize cloudiness through the cloud fraction (CF), defined for each 3 dust-containing 5 km-size column as the ratio of number of cloudy 0.333 km profiles to the 4 total number of 0.333 km profiles in the column. Simply put, if the number of cloudy 0.333 5 km profiles is m, the cloud fraction is m/15. We note that in addition to the cloud fraction 6 varying between 0 and 1, the relative location of dust and clouds within 5 km wide columns 7 can also vary ( Figure 3). We also note that unlike the conventional cloud fraction that is based 8 on 2-dimensional (2D) images, our definition here is based on 1-dimensional (1D) 9 measurements along the CALIPSO track. Although off-track clouds may influence dust 10 properties along the track even for CF 1D =0 (Várnai and Marshak, 2012), CF 1D is still a 11 generally useful indicator of cloud coverage. 12 The results in Figure 4 show that dust properties are closely related to CF in all three regions. 13 The main features of the relationship are as follows. 14 First, the top row of Figure 4 reveals that a smaller fraction of dust samples occurs under clear 15 skies in region M than in region E. This is because the SAL is warmer and drier in the East, 16 and so the conditions are less favorable for cloud formation in region E than region M. 17 Second, rows 2 and 3 in Figure 4 reveal that within each region, the median values of βʹ′ 532 18 and χʹ′ are larger for higher CFs. This feature is likely caused by aerosols getting hydrated and 19 swelling in humid regions containing clouds, although undetected cloud particles may also 20 contribute. The figure also shows that in regions E and M, the increase in backscatter and 21 color ratio is more pronounced below the SAL than inside it. The swelling is greater below 22 the SAL than inside it both because clouds and high humidity are more common below the 23 SAL, and because hygroscopic marine aerosols are fairly abundant at low altitudes even in 24 dust layers, whereas the SAL is dominated by less hygroscopic dust particles. 25 Third, within each region, two opposite trends of correlations appear between δʹ′ and CF: 26 Inside the SAL, the median δʹ′ of dust is always larger in clear sky than in cloudy skies; 27 whereas below the SAL, the median δʹ′ of dust is always smaller in clear sky than in cloudy 28 skies. The domains of these opposite behaviors can be separated in the fourth row of Figure 4  29 roughly at the crossing point of the red curve (CF=0) and the blue curve (0<CF<0.6). We 30 note that these crossing points are approximately at the altitude of the bottom of the SAL. The 31 opposite trends inside and below the SAL clearly indicate a different dust volume 1 depolarization response to increased humidity. The possible mechanisms affecting the 2 apparent depolarization ratio of dusty volumes below the SAL will be discussed in Section 3 3.4. 4

Features of dust volumes in the SAL under clear skies 5
Unlike the dusty volumes below the SAL, where the dust is mixed with humidified non-dust 6 aerosols, the SAL is dominated by dust particles. This subsection examines several features of 7 dust volumes inside the SAL in regions E and M. To reduce the effects of clouds, dust 8 volumes are limited to only those under clear skies. In addition, region W is excluded because 9 of its low number of dust samples inside the SAL (Figure 2a). The relationships in Figure 5 and the similarity of results from the two independent datasets 22 can be attributed to the steady altitude-dependence of backscatter, color ratio and 23 depolarization ratio inside the SAL. As discussed in Section 3.1.2, these altitude dependences 24 are likely caused by two mechanisms: (i) a decrease with altitude in the concentration of non-25 dust particles mixed in from below, and (ii) different fall speeds vertically separating the 26 relatively more spherical dust particles from the least spherical ones. This latter mechanism 27 dominates and is further explored in Section 3.3.2. 28 In addition, we note that the depolarization ratio is not only a function of the aspect ratio of 29 particles but is also related to their size (Mishchenko and Hovenier, 1995). Therefore the 30 relationship between δʹ′ and χʹ′ in Figures 5a and 5c may also indicate that backscatter from 1 larger dust particles can be more depolarized. 2

Relationship between the vertical increase in depolarization ratio
3 and longitude 4 As indicated in Figure 2, the depolarization ratio not only increases with altitude in the SAL, 5 but also has a larger increase rate in region M than E. As mentioned above, the increase may 6 come from more spherical and less spherical dust particles getting vertically separated 7 because of their different sedimentation speeds. The upward increase in δʹ′ could then be 8 stronger in region M simply because the sedimentation process has more time to work by the 9 time the dust reaches region M. 10 Numerous studies have demonstrated that the fall speed of atmospheric particles is related to 11 their shape (e.g., Ginoux, 2003). Since particles with irregular shapes have greater cross-12 sectional areas and drag-coefficients, they experience stronger drag force in the air-which 13 implies that more irregular particles fall slower than more spherical ones. Note that a sphere is 14 the most compact object (least surface area for a given volume) and it experiences least drag 15 for a given mass. (Here we assume that dust particle shape does not change systematically 16 with particle size, which also greatly impacts fall speed.) As a result, shape-induced vertical 17 separation will ensue as dust is advected westward, with irregular particles increasingly 18 predominant in the upper portions of SAL. At a constant altitude, this stratification is 19 expected to widen the dynamic range of depolarization ratios with downstream distance from 20 the dust source. This is indeed the case. 21 To that end, we divide regions E and M into sub-regions covering 10° wide longitude bands, 22 and examine the average difference between the depolarization ratios at 3 and 4 km altitudes 23 for each region. As shown in Figure 6, the average difference δʹ′ 4 km -δʹ′ 3 km keeps increasing 24 with the distance from the west coast of Africa. This result implies that the observed change 25 in volume depolarization ratio within SAL is most likely caused by the greater drag of 26 aspherical dust particles. 27

Dust volume properties near clouds 28
Relative humidity usually increases as clouds are approached and this causes nearby aerosols 29 to swell and get hydrated (acquire thin film of water) or even activated as haze (e.g., Twohy et 30 al., 2009b). Observing changes of dust characteristics near clouds can help improve our 31 understanding of the effect of high relative humidity and clouds on dust particles. Figure 4  1 has shown that the backscatter and color ratio of dust volumes increase with cloud fraction 2 both in and below the SAL, whereas the depolarization ratio changes with cloud fraction 3 differently in and below the SAL. This finding indicates that dust properties in and below the 4 SAL are different. This section further examines the near cloud behaviors of dust volumes in 5 and below the SAL. We note that although the base altitude of the SAL may vary during 6 westward transport, as shown in Figures 2 and 4, this analysis uses constant separation 7 altitude of 2 km for convenience. The analysis uses CALIOP Level 1 data to examine changes 8 in backscatter, color ratio, and depolarization as a function of distance to clouds at a 9 resolution of 0.333 km. In this analysis a 0.333 km resolution clear sky profile is considered a 10 dusty profile if it is included in one or multiple 5 km resolution dust layer(s). 11 Figure 7 illustrates the behavior of dust as a function of distance to clouds. The orange curve 12 corresponds to all aerosol samples while the black and green ones to high (in the SAL) and 13 low (below the SAL) dust, respectively. missing dust in about 1/3 of dust profiles that occur within 5 km from clouds. 28 Figures 7b-7d also show that backscatter, color ratio, and depolarization ratio all increase near 29 clouds for dust layers below 2 km, but they remain fairly constant for dust layers above 2 km. 30 The stable behavior in the SAL occurs because most clouds are below the SAL and have little 31 impact on dust in the SAL. In addition, the dust population in the SAL is dominated by 32 hydrophobic particles. For dust volumes below 2 km, the enhanced backscatter and color ratio 1 may come from the swelling of hygroscopic dust and non-dust particles in the humid air near 2 clouds, or even from cloud contamination. However, the depolarization ratio is expected to 3 decrease and not increase near clouds, as hydrated particles tend to be more spherical than dry 4 particles. Thus the apparent increase in depolarization ratio near clouds for dust volumes 5 below 2 km is somewhat counter-intuitive. A possible explanation is that the hydrated and 6 more spherical dust particles or those heavily mixed with marine aerosols are (mis)classified 7 as non-dust aerosols due to their reduced depolarization ratio; the remaining particle 8 populations will be dominated by hydrophobic dust particles that have irregular shapes and 9 hence higher depolarization ratios. 10 In principle, multiple scattering by undetected cloud fragments could also increase 11 depolarization, but this is likely insignificant, for two reasons: (i) the increase in backscatter is 12 too small to suggest strong multiple scattering near clouds, and (ii) the depolarization ratio of 13 all aerosols (orange curve in Figure 7d The study finds that the observed properties of dusty volumes are related not only to the 27 meteorological conditions in the three regions, but also to the speed and duration of dry and 28 wet sedimentation processes. The study examines four characteristics of dust layers: (i) the 29 volume of air containing dust, (ii) lidar backscatter (related to optical thickness), (iii) color 30 ratio (roughly proportional to particle size) at least for spherical particles, and (iv) 31 depolarization ratio (characterizing particle shape, with larger values for irregular particles 32 than for spherical ones). The results show that lidar backscatter and color ratio are smaller, 1 while the depolarization ratio is larger in the warmer and dryer East region. 2 The analysis reveals that the medians of depolarization ratio and color ratio generally increase 3 with altitude in the SAL. The rate of vertical increase in depolarization ratio is significantly 4 larger farther away from Africa's west coast. 5 We find the dusty volume optical properties related to cloud coverage, with backscatter and 6 color ratio increase with cloudiness of surrounding areas. The effects of cloudiness are most 7 prominent for dust below the SAL. The results highlight that sensitivity to cloudiness is very 8 different below and within SAL. 9 The results also reveal other differences between dust volume near-cloud behaviors inside and 10 below the SAL. In the SAL, the fraction of aerosol samples that contain dust doesn't depend 11 on the distance to clouds, neither the median lidar backscatter, color ratio, and depolarization 12 ratio. Below the SAL, the fraction of aerosol samples containing dust decreases near clouds,