Patterns of North African dust transport over the Atlantic: winter vs. summer, based on CALIPSO first year data

One of the most important factors that determine the transported dust effect on the atmosphere is its vertical distribution. In this study the vertical structure of North African dust and stratiform low clouds is analyzed over the Atlantic Ocean for the 2006–2007 boreal winter (December– February) and boreal summer of 2006 (June–August). By using the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) backscatter measurements over the dust routes, we describe the differences in dust transport between the seasons. We show a bi-modal distribution of the average dust plumes height in both seasons (it is less clear in the winter). The higher plume top height is 5.1 ±0.4 km, near the African coast line in the summer and 3.7 ±0.4 km in the winter. The lower plume merges with the marine boundary layer, in both seasons. Our study suggests that a significant part of the dust is transported near and within the marine boundary layer and interacts with low stratiform clouds.


Introduction
Mineral dust has a significant role in the biogeochemical cycles, climatic processes and human life.Dust particles interact directly with electromagnetic radiation and therefore alter the surface-atmospheric radiation budget (Ramanathan et al., 2001;Kaufman et al., 2002;Yu et al., 2006).By serving as clouds drops/ice crystals condensation/freezing nuclei dust may indirectly invigorate convective clouds height (Koren et al., 2005) or accelerate precipitations (Levin et al., 1996), and it can also change clouds lifetime, and albedo (Albrecht, 1989;Hansen et al., 1997;Twomey, 1974;Kaufman et al., 2005a).Solar energy absorbed by dust may change the atmospheric temperature profile and therefore change cloudi-Correspondence to: Y. Ben-Ami (yuval.ben-ami@weizmann.ac.il) ness (Hansen et al., 1997).After sinking or scavenging, dust can supply essential minerals to the biosphere such as in the Amazon rain forest (Swap et al., 1992;Reichholf, 1986;Koren et al., 2006) and phytoplankton in the oceans, hence it may enhance the reduction in atmospheric CO 2 (Falkowski et al., 1998) and it can modify the ocean albedo.Dust can also decrease snow and ice albedo and enhance their melting (Psenner, 1999).It may serve as a vehicle for cross oceans transfer of fungi or bacteria, and can be associated with human diseases or oceans catastrophes (Garrison et al., 2003).Nevertheless, our knowledge on basic questions concerning the dust loading (Cakmur et al., 2006) and sink (Tegen, 2003), as well as its chemical (IPCC, 2007;Engelstaedter et al., 2006) and radiative (Haywood and Boucher, 2000) properties, still lack with large uncertainties.
Previous remote sensing studies (e.g.Herman et al., 1997;Prospero et al., 2002;Israelevich et al., 2002;Middleton and Goudie, 2001;Kaufman et al., 2005b) and observations (e.g.Prospero and Lamb, 2003) identified that the North African dust activity is subjected to a strong seasonal cycle.As a consequence, the location of the dust sources, their activity pattern and the transport routes are affected.
The synoptic conditions over West of Africa and the mid Atlantic during the summer months were discussed in details in previous studies (Karyampudi and Carlson, 1988;Karyampudi et al., 1999;Schepanski et al., 2009).The dust outbreaks occur predominantly within the ridge region of passing easterly wave disturbances with a periodicity of 5-7 days (Prospero and Carlson, 1972).These dust outbreaks occur under the conditions of a thermal low that prevails over West Africa (centered around 17 • -23 • N) in this period (due to intense solar heating).The converging strong low-level flow around the low lifts the dust as a result of strong surface pressure gradient.Over this region there is a deep mixed layer that often extends up to 5-6 km in height.The airborne dust is carried westward in the Saharan Air Layer (SAL) between 1.5 to 5-7 km by the prevailing easterly flow in these latitudes (Prospero and Carlson, 1972;Karyampudi et al., 1999;Karyampudi and Carleson, 1988;Prospero and Nees, 1977).Over the ocean, at low level (beneath the SAL), there are relatively clean northeasterly trade winds in the marine mixed layer.The southern equatorial trough or Intertropical Convergence Zone (ITCZ) is located between 5 • -10 • N. In accordance with the meridional temperature gradient at mid levels heights, between the warm Saharan air and the cooler air in the equatorial zone, there is a strong vertical shear in the zonal wind between 850 and 650 mb, that manifests itself as a mid-level easterly jet (Karyampudi and Carlson, 1988).The 700 mb level winds represent this mid-level flow.Figure 1a presents the mean winds in the 700 mb level during the studied summer months.During these months the dust is mainly transported over the Atlantic Ocean towards the southern part of the United States, Caribbean and the north part of South-America.Often, it can be transported northerly by anticyclonic eddies over the Canaries or Azores Islands (Carlson and Prospero, 1972;Kaufman et al., 2005b).
The synoptic condition over the region during the winter months are less discussed in the literature.The synoptic systems move southward compared to the summer (Kalu, 1979;Schepanski et al., 2009) and they are less consistent.The thermal low over Africa has larger variance and is centered on latitudes 5 • -10 • N. The mid-level jet location is around 5 • S-7 • N. Figure 1b presents the mean wind direction in the 700 mb height during the studied winter months.The dust is transported south westward, toward the northern part of South America, while crossing the African coastline near the equator (Prospero et al., 2002).
During the winter months the West African Sahelian region is characterized by large areas of biomass burning fires mainly due to agricultural activities.In the same time low level easterly and north-easterly Harmattan winds transport the dust toward the biomass burning regions causing unavoidable external mixing.Ground based and aircraft measurements from the African Monsoon Multidimensional Analyses (AMMA, Formenti et al., 2008), the Dust Outflow Deposition to the Ocean (DODO, Formenti et al., 2008), the SAharan Mineral dUst ExperiMent (SAMUM, Ansmann et al., 2009), and the Dust And Biomass EXperiment (DABEX, Johnson et al., 2008a, b) showed that the winter atmospheric column may contain a multi layers structure of low level dust layer and elevated biomass burning layer that contains dust as well as external mixing of both types of aerosols.Moreover, they estimated the contribution of the dust mass, of elevated biomass burning aerosol layer, to be extremely high (72±16%, Formenti et al., 2008).Therefore, during the winter we describe the transport of plumes of dust-biomass burning aerosol mixture.
Until the launch of CALIPSO, the vertical distribution of dust over the Atlantic Ocean was studied mostly by short term in-situ (e.g.McConnell et al., 2008;Reid et al., 2003) and remote sensing (e.g.Karyampudi et al., 1999) measurements as well as models (e.g.Schepanski et al., 2009).The Cloud-Aerosol LIdar with Orthogonal Polarization (CALIOP) instrument on board the CALIPSO (as a part of the A-Train constellation) has opened an opportunity to study the transported dust vertical structure in a large number of events (sufficient statistics).A few recently published papers used the CALIPSO data to examine the vertical distribution of the dust over the Atlantic Ocean, but only for short term case studies (e.g.Liu et al., 2008a;Generoso et al., 2008).In this study we use the attenuation backscatter measurements, done by the CALIOP instrument on board the CALIPSO (Thomason et al., 2007) for a longer period of time.We characterize the average dust vertical distribution in the summer ( 2006) versus the dust and smoke in the winter (2006)(2007) along the main transport route in each season, where the impact of dust upon biogeochemical cycles, climatic processes and human life is the most significant.We also examine the heights of the dust and low stratiform clouds layers in order to understand what type of interaction should be expected there.

Methods
The measurements made by the CALIOP instrument are three calibrated lidar profiles: total attenuated backscatter at 532 nm and 1064 nm, and the perpendicular polarization component of the backscatter at 532 nm.The vertical resolution of the 532 nm and 1064 nm channels (in the lowest 8.2 km) are 30 and 60 m, respectively.Both channels have horizontal resolution of 33 km (Winker et al., 2007).
The polarization backscatter data attributed from CALIOP instrument can add significant information on aerosol shape and optical properties that can be used for aerosol classification.However, the noise level of most of the attenuated backscatter data was too high for (a) detecting the boundaries of the aerosol layer, and (b) to determining if we see smoke only, dust only or mixture.Classification process, based on this data, could not be robust.Moreover several recent papers suggested that over the ocean most of the aerosol packages are mixed.
For fast classification between aerosol and clouds and in order to mark the top and bottom of the dust layer, we hoped to use either the CALIPSO depolarization product or the Ver- tical Feature Mask (VFM, Vaughan et al., 2005).However we found out that for such detailed analysis the best results are obtained when the final aerosol layer is mask manually on each profile.Due to the high noise level of the attenuated backscatter data (especially backscatter data acquired during daytime), we used the depolarization product and the VFM as the sources for the initial classifications and then we manually determined the location of the aerosol layer.The data was not used, when the confidence level was low.The classification is based on the different backscatter patterns between clouds and aerosol.While aerosol plumes have relative weak but uniform signature, stratiform (low, marine stratocumulus or higher stratus) clouds have a much stronger and narrower backscatter signal and convective clouds has patchy backscatter pattern.The total attenuated backscatter at 532 nm data has shown the best contrast and served as the main reference data for classification.
The selected research area is between 40 • N to 20 • S, and 15 • E to 70 • W, during 2006 summer (June-August) and 2006-2007 winter (December-February).The selection of the study area was based on the dust transport region obtained by the MODIS total Aerosol Optical Depth (AOD) and aerosol fine fraction maps (Kaufman et al., 2005c) and in agreement with the dust distribution in previous studies (e.g.Prospero, 1999;Prospero et al., 2002;Herman et al., 1997;Kaufman et al., 2005b).This wide region covers the main dust transport routes for both seasons.MODIS aerosol fine mode fraction of optical depth is defined as the fraction of the total optical depth attributed to the fine mode (aerosol with diameter smaller than 1 µm, Kaufman et al., 2005b).The fine mode fraction parameter enables discrimination between dust or sea-salt and pollution or biomass burning aerosols.Dust aerosol is dominated by relatively large particles (diameter larger than 1 µm) and therefore will have a significant coarse mode fraction where the smoke from biomass burning and industrial aerosol are dominated by smaller particles (diameter smaller then 1 µm) and therefore are characterized by the fine mode fraction (Dubovik, et al., 2002).
The summer dust transport route is clearly shown in an area characterized by high AOD levels (contributed mainly from dust but also from marine aerosol) and low fine fraction values (Fig. 2, upper row).The winter dust transport (Fig. 2, lower row) is known to migrate southward to the border of the Sahara and the Sahel region (e.g.Kaufman et al., 2005b), where intensive biomass burning occurs.Therefore the coarse mode contribution of the dust is smoothed by the smoke, resulting in a blurrier coarse mode signature of the dust.Nevertheless, the high AOD (contributed mainly from dust and smoke, but also from marine aerosol) area of the winter flux is shown to have lower fine mode values surrounded by higher fine-fraction contributed by smoke from the southern parts of Africa and Brazil and pollution form North America.
Based on this MODIS data and taking into account the above field experiments results we decided not to distinguish between the dust and the smoke plumes during the winter.Therefore, for the winter analysis, we study the transport of a joint dust-smoke plumes.
For each backscatter vertical profile within the research area the top and the base of the aerosol plumes as well as the location of the low stratiform clouds were picked (the stratiform clouds are often too thin to distinguish between their bases and tops) manually.An example for the differences between the aerosol and cloud signature can be seen in Fig. 3.
When the location of the dust plume base was close to the top of the Marine Boundary Layer (MBL), we couldn't determine its exact location.Therefore, results of plume bases height and thickness (mainly of the lower plumes) may introduce error.Since the daytime measured data has higher level of noise due to the contribution of the solar radiation (Liu et al., 2008b) the aerosol signal in daytime profiles was enhanced by averaging the nearest 5 pixels (total of 1.65 km) along the horizontal dimension of the profile.The collected data of aerosol plumes and clouds heights, as well as the MBL depth, for both daytime and nighttime were sorted into 1 • ×1 • grid.The data was collected from 381 daytime and 499 nocturnal CALIPSO tracks.
To avoid statistical biases in areas with rare aerosol and clouds measurements, aerosol and cloud grid points with only one daily sample were removed from the dataset.Next, the average heights of the dust and dust-smoke plumes (and clouds) were defined statistically by analyzing the local top height distribution (see a detailed description in Sects.3.1 and 3.2).

Dust
Analyzing the aerosol plumes top height distribution reveals a bi-modal distribution in the summer.In the winter distribution there is a dominant dust-smoke high level plumes with a minor peak in lower altitudes (Fig. 4).The low altitude mode in both seasons contains marine aerosol beside the dust.Note that although the MBL depth (often the top of the layer is bounded by marine stratocumulus clouds (MSc), see Sect.3.2 and Fig. 3) appears in the histograms of the dust layers top height, it was excluded from the dataset in the next steps of the analysis.
In order to analyze the height distribution of the apparent bi modal distribution, we needed to separate the two modes.The best threshold for separation (after excluding the MBL samples) top heights was determined by looking for the threshold that will show the weakest sensitivity of the average top height to small threshold changes.Namely, we calculated the mean (low and high) dust top height (H l and H h ) for a range of thresholds (Tr) and the best threshold was determined by minimizing the top's height derivatives, T r = T r(min dH l,h dT r ). (1) From the above distribution analysis, the chosen thresholds for the aerosol average top heights are: 1) an upper aerosol plumes with top height above 2.7 and 1.6 km in the summer and winter, respectively; 2) a lower aerosol plumes with top height above the MBL (0.8 km, measured by the MSc height, see Sect.3.2) in both seasons and top height equal or below 2.7 and 1.6 km for the summer and winter, respectively.The analysis of the dust plumes over western parts of the Atlantic Ocean reveals that the separation between the two modes is less clear.Therefore, the Region Of Interest (ROI) was selected to be the first 50 • longitude from the African coastline over the Ocean (along the contours of Fig. 2).
Upper plumes: The dust top height distribution analysis along the longitudinal cross-section of the ROI, (Fig. 5a), shows that during the summer the average top height, when crossing the African coastline, is 5.1±0.4km.During the winter the upper dust-smoke plumes are lower and the mean height near the African coastline is 3.7±0.4km.As the upper dust plumes are transported westward they decrease in height on an average of 27 m per 1 • longitude during the summer and 23 m during the winter.Examination of the upper dust plumes average thickness (calculated as the difference between each plume top height and its corresponding base height along the ROI, Fig. 5a) showed that on average, the upper plumes are significantly thicker during the summer (3 to 3.6 km in the summer, versus 2 to 2.7 km in the winter).Lower plumes: Analysis of the lower dust plumes mean top height (along the ROI) shows an inverse trend (Fig. 5b), namely the dust crosses the African coastline near 1.5±0.25 and 1.1±0.35km in the summer and winter, respectively, and the average height increases (in an order of a few hundred meters) toward the western side of the Atlantic Ocean (in the winter the trend is less pronounced).
Comparing the seasonal spatial structure of the averaged dust plumes top height (Fig. 6a and b) to the structure of the MODIS AOD measurements over the Atlantic Ocean (Fig. 2, left panel) shows that during the summer, the maximal top height of the upper dust plumes is located north to the maximal AOD.Similar analysis of the winter season reveals that the location of the maximum top height of the dust-smoke plumes is located south to the maximal AOD location.A possible explanation to this difference is the African relief that may serve as an orographic barrier that lifts the transported dust: during the summer it may be the Atlas mountains, located in the north western part of Africa (Morocco, Algeria, and Tunisia), while in the winter the transported dust may be lifted by mount Oku (3000 m) and mount Cameroon (4070 m), both located in Cameroon.Note that the MODIS winter fine fraction distribution (Fig. 2, lower right side) shows higher fine fraction values in the southern region suggesting a larger contribution from smoke on the southern part of the average plume.

Low stratiform clouds
The examined clouds were low clouds (including Stratocumulus and Stratus clouds), with top heights lower than 2.5 km.These clouds height distributions along the ROI show a bi-modal distribution in the summer and one dominant mode in the winter season (Fig. 7).This can be interpreted as a separation between the MSc clouds that are bounded by the shallow MBL (as was observed during the sampling stage) and other stratiform clouds.The low MSc clouds height can be used as a proxy to the MBL top height.Like in the dust case, when analyzing the stratiform cloud height, the best threshold for separation between the clouds modes was determined by looking for the threshold that will show the weakest sensitivity of the average clouds mode height to small threshold changes.Based on this analysis the clouds height distribution was divided to two separate modes below and above 0.8 km.

Discussion and conclusions
In this study we used the CALIOP vertical backscattering profiles to examine the seasonal dust height distribution over the Atlantic Ocean during one boreal summer (2006) and one boreal winter (2006)(2007).
During the summer, North African dust transport over the Atlantic Ocean is characterized by two levels of plumes: (a) elevated plumes, mostly separated from the underneath MBL and undercut below by the trade winds, and (b) lower plumes, attached (from above) and most likely within the MBL.
Presumably, the elevated plumes are a result of intensive solar heating of the bright Saharan surface that encourage lifting of dust into the deep mixed boundary layer.As the plumes cross the coastline, they override the cooler and moister maritime air layer (Karyampudi et al., 1999).The lower dust layer may be a result of a combination of fresh emission from costal dust sources toward the ocean at low altitude and sedimentation of dust from upper plumes.
The average base and top heights of the upper dust plumes found in this study agree with results from previous studies (e.g.McConnell., 2008;Myhre et al., 2003) although some deviations exist.
Towards the western Atlantic Ocean, the apparent vertical structure is less clear, possibly due to descending and sedimentation of the upper layer and the rising of the lower layer (caused possibly by the east-west sea surface temperature gradient and its impact on the marine boundary layer depth).
In situ measurements conducted in the western Atlantic (on the edge of the ROI during the summer) (Reid et al., 2002(Reid et al., , 2003;;Maring, 2003) showed a chaotic vertical structure of transported dust.However, Reid et al. (2003) showed that over the western Atlantic, the chaotic vertical structure of the transported dust was favored with upper and lower (in the MBL) layers.These finding suggest that our difficulties to track the structure of the dust layers over the western Atlantic, was caused by the method of seasonal averaging because of the changing nature of these layers.
During the winter season, the bi-modal trend is less clear, the dust is advected off the coastline of Africa in lower altitudes compared to the summer.This may be a result of shallower boundary layer and weaker surface heat fluxes over the Sahel, where the dust is emitted towards the ocean during the winter (compared to the Sahara region where it is emitted during the summer).Comparison between the above cited AMMA results (that showed low and high concentration of dust in the lower and higher altitude, respectively) to our winter results, shows that our results may over estimate the height of the bulk part (in mass) of the transported dust.
In situ measurements that represent the winter months (Formenti et al., 2001) showed that the dust may reach South America either in a uniform vertical structure or as several plumes.
Westwards from the studied area (ROI, marked by contours in Fig. 2) we could not distinguish between the two dust layers.We assumed that the aerosols are transported in one dust layer that includes the marine boundary layer, in both seasons (in the winter it is a mixed dust-smoke plume), or in a chaotic vertical structure (not shown in this paper).
The average vertical depth of the summer upper dust plumes is 3 to 3.6 km with average top height of 5.1±0.4km near the African coast (15 • W).A decrease in height of 27 m per 1 • longitude yields dust top height of 3.3 km and a base touching the ocean around 82 • W (Caribbean Sea).
The winter upper dust and smoke plumes average width is 2 to 2.7 km, with average top height of 3.7±0.4km near the Gulf of Guinea (10 • E).A height slope of 23 m per 1 • longitude yielding dust-smoke plumes top height of 2.4 km and a base touching the ocean around 49 • W, near the Amazon.
The average summer lower dust plumes top height near the African coast is 1.5±0.25 km and it is 1.1±0.35km in the winter.In both seasons the lower dust plumes top heights slightly increase toward the western side of the Atlantic Ocean.
The stratiform low clouds height analysis showed a bimodal distribution as well.The MSc clouds (bounded by the MBL) contribute to the lower altitude clouds layer at average height of less than 0.8 km.The higher altitude (less common) startiform clouds, appear mostly in the summer.The presence of the lower dust plumes suggests a direct interaction between the clouds and the dust, as their co-existence was observed at the sampling stage.Moreover, on average, the base of the upper dust plumes will reach height of less than 1 km in longitude of 43 • W in the summer and longitude 7 • E in the winter, over the Atlantic Ocean where there are still significant regions covered by MSc cloud fields (Fig. 7).

Fig. 2 .
Fig. 2. Aerosol optical properties over the Atlantic Ocean, for the 2006 summer (upper row) and 2006-2007 winter (lower row), from the MODIS instrument onboared Terra.Left column: the aerosol optical depth (at 550 nm).Right column: the aerosol fine mode fraction.The contours mark the center of the seasonal plume.

Fig. 4 .
Fig. 4. Aerosol plumes top height distribution (for all CALIOP samples and before excluding the MBL tops) along the center of the plumes (see Fig. 2), for the summer (upper row) and winter (lower row).Each histogram covers an area of 10 • longitudes along the contour line (marked in Fig. 2), starting from the West North African coastline (right) toward the Americas (left).Samples for horizontal intervals of 330 m along the flight track.

Fig. 5 .
Fig. 5. Results of (a) the average dust height and thickness of the summer (red) and winter (blue) upper plumes, and (b) the average lower dust plumes top heights.Results of plume bases height and thickness (mainly of the lower plumes) may introduce error.Due to the expected error the bases of the lower dust plumes are not presented in Fig. 5b.All averages were calculated along the ROI, and after determination of the best thresholds.

Fig. 6 .
Fig. 6.Averaged seasonal spatial distribution of the upper dust plums top height for the summer (upper panel) and winter (lower panel) seasons; scale height in km.

Fig. 7 .
Fig. 7. Clouds top height distribution along the ROI, for the summer (upper panel) and winter (lower panel).Each histogram covers an area of 10 • longitudes along the contour line (marked in Fig. 2), starting from the West North African coastline (right) toward the Americas (left).CALIOP samples covered only low clouds (lower then 2.5 km).Samples for horizontal intervals of 330 m along the flight track.