Aerosol vertical distribution in dust outflow over the Atlantic: Comparisons between GEOS-Chem and Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO)



[1] Vertically resolved attenuated backscatter from the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) mission and aerosol optical thickness (AOT) from the Moderate-resolution Imaging Spectroradiometer (MODIS) are used to characterize the Saharan dust outflow over the Atlantic and to evaluate results from a global chemical and transport model (GEOS-Chem). We first document two events of dust plume transport from the Sahara to the American coast representative of summer and winter conditions. Observed and model-derived attenuated backscatter profiles agree qualitatively well, suggesting that the altitudes of the aerosol layers over the Atlantic are well reproduced by the model. In both the model and the observations, dust plumes extend up to 8 km in summer and up to 4 km in winter over the Atlantic close to the source regions and gradually descend throughout their travel over the Atlantic. We find however that, in summertime, observed AOT are overestimated by the model close to the source regions and underestimated in the remote regions, suggesting a too weak transport toward the western Atlantic and/or a too strong deposition over the Caribbean Sea. We then use our simulation to characterize the deposition flux of dust in this region. Half of the dust deposited on the surface of the open ocean in 2006 in this area occurs during summer, 20% during each of winter and spring, and 10% in the fall. During a 1-week dust episode in July 2006 that we investigated, 5 Tg of dust were deposited (4% of the annual total in the region).

1. Introduction

[2] Atmospheric mineral particles (dust) are abundant in the atmosphere and have multiple and complex interactions with climate [e.g., Harrison et al., 2001]. Dust affects the Earth's radiative budget through reflection and absorption of both the incoming solar radiation and the outgoing infrared radiation. The sign of the solar and infrared forcing may be different, which further increases the level of uncertainty related to the dust net (solar + infrared) radiative perturbation [e.g., Tegen and Lacis, 1996; Sokolik and Toon, 1998; Boucher and Haywood, 2001; Forster et al., 2007]. For example, Lau and Kim [2007] recently suggested that the high atmospheric dust loading over the Atlantic in 2006 could be responsible for 30–40% of the negative anomaly in the underlying sea surface temperature owing to the solar attenuation by dust. As other aerosol particles, dust affects atmospheric chemistry in several ways [Jacob, 2000; Arimoto, 2001]. For instance, dust in the atmosphere can impact the concentrations of trace gases such as ozone [Bonasoni et al., 2004]. Aerosols act as cloud condensation nuclei, and therefore have strong implications for cloud microphysics [Rosenfeld, 2006]. The specific role of dust in altering precipitation has been discussed by Rosenfeld et al. [2001] who suggested a possible desertification feedback loop between the lift of dust in the atmosphere and a decrease in precipitation. Dust is also believed to influence the formation and intensity of Atlantic tropical cyclones [Dunion and Velden, 2004; Wu, 2007]. Finally, dust is a source of nutrients for marine and terrestrial ecosystems as it contains iron, which is a limiting factor of the phytoplankton growth [Jickells et al., 2005; Mahowald et al., 2005].

[3] Dust export and deposition are in turn strongly controlled by large-scale meteorology and climate variability [e.g., Prospero and Nees, 1986; Moulin et al., 1997]. All year, African desert dust is transported westward to the North Atlantic [e.g., Prospero and Carlson, 1980; Prospero et al., 1981; Swap et al., 1996; Chiapello and Moulin, 2002; Colarco et al., 2003b; Zhu et al., 2007] mostly under the influence of trade winds. In late winter and spring, dust plumes extend to the northeastern coasts of South America remaining south of 5°N, while they are shifted north to the Caribbean Sea and can extend up to about 20°N during summer [Moulin et al., 1997]. On the basis of meteorological analysis in the Capo Verde islands, Chiapello et al. [1995] showed the existence of low-altitude transport during winter in the northeastern Atlantic due to the restriction of easterly winds below 3 km. During summer, meteorological conditions allow particles transported over the Atlantic to reach altitudes above 5–7 km [Carlson and Prospero, 1972; Prospero and Nees, 1977].

[4] The altitude at which the aerosols are lifted and further transported directly influences their lifetime and also, to some extent, their interaction with clouds and radiation. The transport patterns of dust layers (e.g., altitude) over the oceanic regions could also have implications for the amount of material that can be deposited on sea surface and therefore for the oceanic biological productivity. The vertical structure of Saharan dust plumes exported to the Mediterranean basin has been investigated using ground-based lidar by Hamonou et al. [1999] who found layers at different altitudes between 1.5 and 5 km. Karyampudi et al. [1999] used the Lidar In-Space Technology Experiment (LITE) to investigate the Saharan dust transport over the eastern Atlantic. However, the LITE measurements were only acquired for a few days in September 1994. In contrast, the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) instrument on board of the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) mission [Winker et al., 2003] (launched in spring 2006) allows the continuous monitoring of vertically resolved attenuated backscatter on a global scale, providing unique information about the vertical distributions of aerosols. In this paper, we use the CALIOP observations (as well as those from other satellites) to (1) evaluate the representation of aerosol vertical profiles in a global chemical and transport model (CTM) in Saharan dust outflow over the Atlantic ocean and (2) characterize the seasonal variations in transport pattern of dust (with respect to vertical and geographical distributions) using both observations and model outputs.

[5] The manuscript is organized as follows. Section 2 describes the CTM and the different set of observations used in this study. Section 3 compares the observed and simulated transport patterns of dust for two specific episodes (July 2006 and February 2007) and on a seasonal basis. Section 4 presents the resulting dust deposition during that period and compares our result with previous estimates. Finally, section 5 presents a summary and conclusions of our work.

2. Data and Method

2.1. GEOS-Chem Model

[6] We used the GEOS-Chem model (version v7-03-06, to conduct a global three-dimensional simulation of coupled oxidant-aerosol chemistry for 2006 with a 2° × 2.5° horizontal resolution and 30 vertical levels. The model is driven by assimilated meteorology from the Goddard Earth Observing System (GEOS-4) of the NASA Global Modeling and Assimilation Office (GMAO), which includes winds, temperature, surface pressure, water content, clouds, precipitation, convective mass fluxes, mixed layer depth and surface properties with a 6-h temporal resolution (3-h for surface variables and mixing depths). The model includes a coupled simulation of tropospheric ozone-nitrogen oxides-hydrocarbon chemistry [Bey et al., 2001] and of tropospheric aerosol types including sea salts [Alexander et al., 2005], mineral dust [Fairlie et al., 2007], sulfate-nitrate-ammonium aerosols [Park et al., 2004], carbonaceous aerosols [Park et al., 2003], and Secondary Organic Aerosol (SOA) following Chung and Seinfeld [2002]. Aerosol optical thickness (AOT) at 550 and 865 nm is calculated online assuming lognormal size distributions of externally mixed aerosols and as a function of the local relative humidity to account for hygroscopic growth of all hydrophilic species as described by Martin et al. [2003]. Note that in the rest of the paper, we refer to GEOS-Chem (or simulated) AOT as the sum of optical thickness due to all aerosol components (included in the model) unless specified otherwise. We refer to Generoso et al. [2007] for a detailed description of aerosol sources other than dust sources.

[7] The standard version of GEOS-Chem includes the mineral dust entrainment and deposition (DEAD) scheme of Zender et al. [2003] as described by Fairlie et al. [2007]. The dust is distributed in four size classes (with radius bins of 0.1–1.0, 1.0–1.8, 1.8–3.0, 3.0–6.0 μm). Dry deposition accounts for gravitational settling [Seinfeld and Pandis, 1998] and turbulent dry transfer of particles to the surface [Zhang et al., 2001]. Dust is in addition subject to wet deposition, which includes both scavenging in convective updrafts and rainout/washout from large-scale precipitation [Liu et al., 2001]. We conducted a simulation using this setup starting in July 2005 and analyzed the outputs from January 2006 to February 2007 as the GEOS-4 meteorological fields are only available up to that date. We compared the model outputs to available ground-based measurements from the Aerosol Robotic Network (AERONET, available at and we found that the AOT were systematically overestimated over northern Africa and southern Europe in 2006 (see Figure 1). These results are consistent with those found by Auvray [2006] over Africa and southern Europe and by Fairlie et al. [2007] over the United States. Following the work and recommendations of Auvray [2006], we calculated the dust emissions with the DEAD scheme (using the dust classes mentioned before) but the total emissions were scaled on a monthly basis to those provided by Laurent [2005] for the oriental and occidental Sahara (east and west of 13°E, respectively). As described by Laurent et al. [2005], the dust emission scheme of Laurent [2005] has been improved upon that of Marticorena and Bergametti [1995, 1996] by using space-borne observations to estimate roughness lengths, improved soil size distributions and textures for Saharan soils, and ECMWF 40 year Re-Analysis (ERA-40) wind fields at a high resolution of 1° × 1°. We refer to Laurent [2005], Laurent et al. [2005], and Marticorena et al. [2004] for further details. The resulting dust emissions in northern Africa (Figure 2) amounts to 676.7 Tg (676.7 × 1012 g) in 2006 (while the standard version resulted in 2121.7 Tg). Even though the emissions reported by Laurent [2005] correspond to a climatology for the period 1996–2001, scaling our dust emissions to those values significantly improved the agreement between simulated and measured AOT over northern Africa and southern Europe both in terms of seasonal variations and amplitudes (Figure 1). Some discrepancies remain despite this fairly good agreement. For example, at Izana the model tends to overestimate the AOT especially in summer which may be related to too strong dust sources. We acknowledge the fact that our resulting dust emission scheme is highly parameterized. However, as several studies have suggested that dust distribution downwind of source regions is largely influenced by the dynamical transport rather than the magnitude of dust emissions [Tegen and Miller, 1998; Mahowald et al., 2003; Colarco et al., 2003a], we argue that our model (that uses assimilated meteorology for 2006) is a suitable tool to examine dust transport events over the Atlantic.

Figure 1.

Monthly mean observed AOT at (top) 865 nm and (bottom) 550 nm from AERONET network (blue circles) and simulated by the GEOS-Chem model using the DEAD scheme scaled to data of Laurent [2005] (black triangles) and of the standard version (dotted lines) from January 2006 to February 2007 (see text section 2.1 for further details). Error bars show the standard deviation of observed AOT. The locations of these stations are shown on the map in Figure 3 by up- and down-pointing triangles that correspond to measurements performed at 865 and 550 nm, respectively.

Figure 2.

Dust emissions as computed in GEOS-Chem using the DEAD scheme [Zender et al., 2003] with total amount scaled on a monthly basis to the emissions provided by Laurent [2005] for (top left) January, (top right) April, (bottom left) July, and (bottom right) October 2006. The monthly emissions from the region in the black box (10°N–35°N; 16°W−33°E) are indicated in Tg (1012 g) at the bottom left corner of each panel.

2.2. MODIS

[8] The Moderate-resolution Imaging Spectroradiometer (MODIS) developed by the National Aeronautics Space Administration (NASA) provides multispectral measurements that are used to retrieve aerosol properties over land and ocean using separate algorithms. AOT (τ) retrievals over ocean [Tanré et al., 1997; Remer et al., 2002] are estimated to be accurate to within 0.03 ± 0.05τ [Remer et al., 2005]. Over land, AOT retrievals are possible over dark (vegetated) surfaces but not available over bright land (i.e., desert or snow/ice-covered surfaces) [Kaufman et al., 1997; Chu et al., 2002]. Over dark land surfaces, the uncertainty is on the order of 0.05 ± 0.15τ [Remer et al., 2005]. The data used in this study come from Terra Collection 005 and consist of daily total AOT at 550 nm globally gridded at 1° × 1° horizontal resolution, and regridded here at the 2° × 2.5° model resolution.


[9] CALIPSO is a joint program between NASA and Centre National d'Etudes Spatiales (CNES) [Winker et al., 2003]. CALIPSO was launched in April 2006 and is part of the Afternoon or “A-Train” satellite constellation that presently consists of several satellites flying in formation around the globe. CALIPSO carries three instruments which measure aerosol and cloud properties: the Imaging Infrared Radiometer (IIR), the Wide Field Camera (WFC), and the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP). CALIOP is a two-wavelength lidar that provides high-resolution attenuated backscatter vertical profiles at 532 and 1064 nm [Winker et al., 2004, 2007]. The instrument acquired data with a horizontal resolution of 333 m and a vertical resolution of 30–60 m.

[10] In order to investigate the aerosol vertical distribution over the Atlantic, we used the CALIOP attenuated backscatter data at 532 nm. We used the level-1 product that includes the scattering by atmospheric molecules, clouds, and aerosols (at the time of this study, the level-2 product that distinguishes between aerosol and cloud layers still suffered from significant shortcomings [Labonne et al., 2007]). To reduce noise in the raw observations of attenuated backscatter, we first dereased the data resolution in the horizontal direction by averaging nine measurement points (initially acquired at the original resolution of 333 m) which results in data with an ∼3-km resolution. We further decreased the horizontal resolution by sampling 800 measurement points over the portion of the track of interest (i.e., in the present study, 0°–35°N over the northern Atlantic) so that the final resolution is of approximately 5 km. Little information is lost in this spatial smoothing as aerosol layers in our region of interest present a rather large spatial extension compared to the horizontal resolution of the instrument. We then also decreased the resolution in the vertical dimension by averaging the data over 120 altitude bins resulting in a vertical resolution of approximately 90 m. We removed the molecular scattering using a theoretical estimate which depends on the altitude but is almost independent of the spatial location and time, in particular over the tropics. Those various steps result in a smoothed CALIOP attenuated backscatter product that accounts for both the effects of clouds and aerosols.

[11] To compare the model results to the observed attenuated backscatter that accounts for the effects of clouds and aerosols, we expressed the latter as follows [see, e.g., Wang et al., 2007]:

equation image

where LR is the Lidar Ratio (here defined as the extinction-to-backscatter ratio) expressed in sr, σ the extinction at altitude z in m−1 and exp(−2equation imageσ(z)dz) is the transmission (the two-way extinction of the atmospheric column). We calculated both extinction and transmission from simulated daily mean optical thickness. To distinguish between the role of aerosols and clouds in the model, attenuated backscatter was calculated separately on the basis of aerosol and cloud optical thicknesses. (Note however that the resulting attenuated backscatter quantities cannot be summed for a direct comparison to observed attenuated backscatter.) The uncertainties associated to LR for aerosols in general and for dust in particular are quite large. For example, Mona et al. [2006] found a mean value of 22.6 ± 0.6 sr for LR at 532 nm in the case of Saharan dust layer. Karyampudi et al. [1999] reported LR values in the range 25 to 35 sr for Saharan dust while indicating that dust transported over the eastern Atlantic may have values on the high side of this range. In the present study, we used an average LR of 40 sr following recommendations from Cattrall et al. [2005] and in agreement with estimate provided by Liu et al. [2008] who also examined a dust event over the Atlantic during the summer 2006. Using lower values of LR for aerosols (e.g., 30 sr) would not affect our results in terms of dust plume altitudes in the Saharan outflow. For clouds, Chen et al. [2002] reported values of LR of 29 ± 12 sr for cirrus clouds. Using the CALIPSO data, Hu [2007] found LR at 532 nm generally less than 11 sr for water clouds and greater than 14 sr for cirrus clouds. We chose a LR of 15 sr to convert the GEOS-Chem cloud optical thickness into attenuated backscatter. For consistency with the aerosol case we refer to GEOS-Chem cloud optical thickness rather than GEOS-4, although the cloud properties come from GEOS-4 meteorological fields. Note that the cloud optical thickness from GEOS-Chem is given at 1000 nm. Because of the different assumptions and uncertainties mentioned above, the attenuated backscatter derived from the simulated optical thickness is used for qualitative (rather than quantitative) evaluation of the vertical profiles of clouds and aerosols in the present study.

3. Transport Patterns of Dust Layers in the African Outflow

3.1. Daily Variations

[12] The MODIS aerosol products show frequent dust export events over the Atlantic during summer 2006. Lau and Kim [2007] have recently reported that the dust loading in 2006 was significantly higher than in 2005 over the Atlantic and presented large fluctuations from June through September as a result of frequent dust outbreaks. Despite the apparent high frequency of dust outbreaks, it is not obvious to find export events that can be followed from the source region to the Caribbean conjointly in all the observation data sets because of the presence of clouds and the sampling frequency of the various satellites. Here, we present one episode of dust transport that was monitored by both the MODIS and CALIOP sensors over the Atlantic during seven days (30 June to 6 July 2006). During this episode the scenes were cloud-free most of the time, which is of particular importance when interpreting the CALIOP data as they include both cloud and aerosol signals. As mentioned in section 2.1, our simulation captures well the observed AOT in July 2006 over most of the AERONET sites located close to the source regions (Figure 1). Figure 3 presents a comparison between a 7-day (30 June to 6 July 2006) average of MODIS and GEOS-Chem AOT over the Atlantic Ocean, and indicates that the general patterns are relatively well captured by the model (correlation of r = 0.8 between MODIS and GEOS-Chem within the region highlighted in Figure 3 (black box). The model however overestimates AOT over the eastern tropical Atlantic (east of 30°W) by 30% but underestimates AOT in the remote regions (west of 30°W) by 20% during this period. The maximum AOT simulated over the Caribbean appears to be slightly shifted to the south in comparison to observed AOT, a problem also noticed in previous models [e.g., Colarco et al., 2003b]. In addition, aerosols are found in the MODIS observations west of 60°W in the latitudinal range of 10 to 20°N while the model does not simulate aerosols west of 60°W. We suggest that this could be due to (1) too weak transport of dust from the source regions to the Caribbean (which would result in an accumulation of dust over the eastern tropical Atlantic) and/or (2) too strong deposition over the Caribbean region.

Figure 3.

Seven-day average of (top) MODIS and (bottom) GEOS-Chem AOT (30 June to 6 July 2006). The model results and MODIS were sampled and averaged in a similar manner. The white areas correspond to missing data. The black (5°N–27°N; 70°W–17°W) and white (0°–30°N; 17°W–35°W) boxes delimit two regions which are used later in this study. The location of the AERONET stations used in this study is indicated by triangles and labeled with the three first letters of their name (see Figure 1). Up- and down-pointing triangles correspond to stations where measurements are performed at 865 and 550 nm, respectively.

[13] We further examine the 7-day episode using the CALIOP products. The coverage of CALIPSO is limited to individual tracks separated by approximately 25° in the longitudinal dimension and the track position varies from one day to another. We selected one CALIPSO track per day (from 30 June to 6 July 2006) that samples the dust plume as it travels from the source region to the eastern Atlantic. Figure 4 shows the CALIOP and the model-derived attenuated backscatter profiles sampled along those selected satellite tracks. In some cases, the CALIOP observations are affected by the presence of clouds. As clouds located at high altitudes generate a strong backscatter (often larger than 0.01 km−1 sr−1), they can totally attenuate the signal from layers underneath. For instance, on 30 June, clouds located above 10 km (cirrus) attenuate the lidar signal and prevent the observation of the aerosol layer between 6°N and 17°N of latitude. The model representation of cloud patterns is highly complex and still subject to large uncertainties. We note however that the attenuated backscatter derived from clouds in the model presents similarity with the signal that can be attributed to clouds in the CALIOP products (recall that the CALIOP data we are using in the present work include signals from both aerosols and clouds). For example, the low clouds (1–2 km) that are simulated over most of the oceanic regions can be seen in the CALIOP profiles even though their distribution is much more scattered than those from the model. At higher elevations (between 4 and 8 km), the model shows cloud patterns on the southern part of the study area, in agreement with the lidar observations. Finally, some simulated high clouds are also consistent with the observations (e.g., on 30 June around 10°N above 10 km, on 5 July).

Figure 4.

Dust transport over the Atlantic from (top to bottom) 30 June to 6 July as seen by CALIOP and GEOS-Chem. (first column) GEOS-Chem AOT at 550 nm. The green lines show the CALIOP track; in the remaining columns, attenuated backscatter (a.b.) is only shown along the red portion of the satellite track; (second column) CALIOP attenuated backscatter profiles at 532 nm; (third column) attenuated backscatter profiles derived from GEOS-Chem and solely due to aerosols; (fourth column) attenuated backscatter profiles derived from GEOS-Chem and solely due to clouds. The units of the attenuated backscatter profiles are km−1 sr−1.

[14] Despite the presence of those clouds, the transport of the aerosol layer is clearly seen in the CALIOP observations. On 30 June, the satellite sampled the atmosphere close to the dust source region. An aerosol layer that extends from the surface to approximately 8 km of height is seen both in the observations and in the model. As it travels to the Caribbean, the dust plume descents following a typical behavior previously reported by, for example, Colarco et al. [2003b], Liu et al. [2008], and references therein. The plume extends approximately up to 6 km on 1 July, 5 km on 5 July and below 4 km on 6 July. Colarco et al. [2003b] reported that the descent of the plume can be attributed to both particle sedimentation and the general descent of air between Africa and the Caribbean. The maximum of the model-derived backscatter also decreases in amplitude, suggesting that deposition of particles occurs over the ocean. Note however that on 5 and 6 July, missing data in the CALIOP products do not allow a direct comparison with the maximum values of the model-derived backscatter (south of 17.5°N). For regions that are not affected by clouds, there is remarkably good match between the main features in the simulated and observed aerosol layers, suggesting that the transport of dust occurs at a reasonable altitude.

[15] We also analyzed a dust transport event during winter conditions using the same data. We selected a dust export from the African to the South American coast from 13 to 17 February 2007. The episode is seen by MODIS sensor and by the GEOS-Chem model, although the simulated AOT are much larger than the satellite data (by a factor of 3) (not shown), which may be due to too large sources during the winter season as already discussed in the following section. We show the comparisons between the CALIOP and the model-derived attenuated backscatter in Figure 5. The vertical extension of the aerosol plume clearly differs from that seen during the summer conditions. Over the Atlantic, the plume remains below 4 km during this episode. Note that the presence of clouds in the model coincide with regions where the lidar signal is highly saturated, indicating again a reasonable representation of clouds in the model.

Figure 5.

Dust transport over the Atlantic from (top to bottom) 13 to 17 February 2007 as seen by CALIOP and GEOS-Chem. (first column) GEOS-Chem AOT at 550 nm. The green lines show the CALIOP track; in the remaining columns, attenuated backscatter (a.b.) is only shown along the red portion of the satellite track; (second column) CALIOP attenuated backscatter profiles at 532 nm; (third column) attenuated backscatter profiles derived from GEOS-Chem and solely due to aerosols; (fourth column) attenuated backscatter profiles derived from GEOS-Chem and solely due to clouds. The units of the attenuated backscatter profiles are km−1 sr−1.

3.2. Seasonal Variations

[16] Figure 6 presents the seasonal variation in AOT over Africa and the Atlantic Ocean as observed with MODIS and simulated with GEOS-Chem. In January, the comparison between simulated and observed AOT suggests a possible overestimate of dust sources. The aerosol outflow is mainly seen between 0° and 20°N, and includes a contribution from biomass burning occurring in the southwest of Africa (see auxiliary material Figure S1 for the contribution of nondust aerosol). In spring and summer, dust aerosols contribute most to the outflow. A shift in latitudes is observed in the continental outflow in comparison with the winter season, as already described for instance by Moulin et al. [1997]. As mentioned in section 3.1, in summer, the model tends to overestimate the AOT over the ocean nearby the African coasts and underestimate them in the Caribbean.

Figure 6.

Monthly average of (top) MODIS AOT and (bottom) GEOS-Chem AOT for January, April, July, and October 2006. The model was sampled and averaged according to the time and location of MODIS observations. The white areas correspond to missing data.

[17] In the following, we use the 7-month data set from CALIOP to analyze the seasonal variations in vertical extent of dust outflow over the Atlantic at two specific longitudes, i.e., close to the dust source regions and further away over the Atlantic. At 35°W longitude, the dust transport in the model is limited to latitudes south of 30°N (see auxiliary material Figure S2). We therefore limit the area to 0°–30°N for our analysis. We first select the eastern part of the Atlantic closer to the source regions (see white box in Figure 3). Time series of attenuated backscatter profiles as derived from CALIOP and from GEOS-Chem model outputs are presented in Figure 7 from March 2006 to February 2007 along with the attenuated backscatter derived solely from the simulated dust aerosols. The model clearly shows a difference in altitude of the aerosol layers between the summer−early fall (from June to September) and the winter−early spring (from January to April) in this area, which confirms our analysis of individual events. Dust contributes largely to those variations. During the summer season, the model indicates the presence of aerosol from about 6 km down to the surface throughout the season, with a maximum load at around 4 km. Similar patterns are seen in the CALIOP time series, although the lidar observations also show a strong backscatter at around 0.5–1 km (which may correspond to the top of the boundary layer) and may be due to the presence of marine aerosols and/or low-level clouds. Recall that we use one single value of LR for the transformation of simulated AOT into model-derived attenuated backscatter, which may result in errors if different types of aerosols (e.g., sea salt, carbonaceous aerosols from biomass burning) are present. During the winter season, the model indicates that dust is more largely found below 3 km, which is also consistent with the CALIOP measurements. From January to March 2007, the model presents high values of attenuated backscatter at some places that are not consistent with the CALIOP observations and likely reflect the overestimate of dust load in the model as discussed in section 3.1 in the case of the winter episode. The vertical structures are consistent with other published results [e.g., Zhu et al., 2007; Chiapello et al., 1995].

Figure 7.

Time series of attenuated backscatter profiles (km−1 sr−1) as derived from the CALIOP products and from the GEOS-Chem model outputs (AER, all aerosols included; DUST, dust aerosol only). The GEOS-Chem data are averaged every day within 0°N–30°N and 17°W–35°W (white box in Figure 3); the CALIOP data are averaged along all the satellite tracks that fall into those limits. The period covers 12 months of model data (from March 2006 to February 2007), while observations are only available for 7 months.

[18] Farther away from the source regions (Figure 8), over the Caribbean, CALIOP observations show a strong backscatter at around 0.5–1 km throughout the 7-month period, similarly to what is seen close to the dust source region (Figure 7). The summertime aerosol layer located at around 4 km close to the dust source region is not apparent anymore over the Caribbean although a few episodes with enhanced attenuated backscatter can be seen in July and August. The model-derived attenuated backscatter profiles are too weak compared to CALIOP in particular during the summertime although enhanced signals in the observations do coincide with the arrival of dust outbreaks in the model. This suggests that, as mention in section 3.1, deposition during the plume transport may be too strong in the model. Errors induced by using constant LR values throughout the region and period examined here could contribute to the disagreement between model results and observations but only to a limited extent since only small changes in LR were reported during the transport over the Atlantic (from 41 ± 3 sr at 532 nm at locations near the sources to 45.8 ± 0.8 sr over the Gulf of Mexico) [Liu et al., 2008].

Figure 8.

Time series of attenuated backscatter profiles (km−1 sr−1) as derived from the CALIOP products and from the GEOS-Chem model outputs (AER, all aerosols included; DUST, aerosol only). The GEOS-Chem data are avaraged every day within 0°N–30°N and 60°W–80°W; the CALIOP data are averaged along all the satellite tracks that fall into those limits. The period covers 12 months of model data (from March 2006 to February 2007), while observations are only available for 7 months.

4. Implications: Deposition of Dust

[19] The atmospheric input of desert dust to the oceans is a dominant source of iron to the surface of the open ocean and contributes to maintain the oceanic primary production and CO2 uptake as iron is an essential nutrient for all organisms. As aerosol distributions are strongly inhomogeneous, deposition pattern also varies significantly with space and time. Various approaches (e.g., measurements, modeling) have been used to estimate aerosol deposition but the uncertainties are still likely to be on the order of a factor 10 [Mahowald et al., 2005; Jickells et al., 2005]. Although passive satellite observations provide aerosol horizontal and temporal distributions from which deposition can be approximately derived, they do not provide direct observations of aerosol surface fluxes. Direct flux measurements using sediment traps are sparse in time and space and can lead to large errors because of the high spatial variability of aerosol deposition. Models can be used to estimate the deposition flux but this flux is clearly sensitive to the accuracy of the horizontal and vertical distribution of dust in the model. The continuous and global monitoring of aerosol profiles supplied by CALIPSO provides an additional constrain to indirectly validate models in terms of aerosol deposition flux. We argue that the comparison of our model results to passive and active spaceborne observations has demonstrated a reasonably good agreement with the limitation that too strong deposition may occur during the transport over the Atlantic (see section 3.1).

[20] Global estimates of dust deposition range from 1000 to 2000 Tg per year [Jickells et al., 2005]. Our model yields a global estimate (land and ocean) of 1120 Tg for 2006 with more than half of that amount being deposited in the tropical Atlantic, northern Africa and the Mediterranean basin. The amount of dust deposited in the tropical Atlantic in GEOS-Chem is consistent with previous estimates based on results from different modeling studies [see Jickells et al., 2005, and references therein]. Our model yields 151 Tg in 2006 in this area (5°N–27°N and 100°W–17°W). The simulated wet deposition is relatively strong in the Caribbean and off the South American coast (auxiliary material Figure S3), i.e., over the region where we suggest that deposition may be overestimated. Figure 9 presents the spatial distribution of the simulated deposition flux representative for each season. The values obtained in the Atlantic are consistent with Jickells et al. [2005]. The results of our simulation indicate that half of dust (46%) is deposited during summer, 20% in winter, and spring and less than 10% in autumn. According to our simulation, 5 Tg of dust were deposited in the tropical Atlantic (5°N–27°N and 100°W–17°W) during the episode of Saharan dust export from 30 June to 6 July 2006 (see section 3.1); most of these (80%) being deposited off the coast of South America and in the Caribbean Sea (west of 30°W) (see auxiliary material Figure S4).

Figure 9.

Accumulated GEOS-Chem dust deposition (g m−2) over January, April, July, and October 2006.

5. Summary and Conclusions

[21] This study is a first attempt to compare profiles of attenuated backscatter from CALIOP to results from a global chemical and transport model. We focus on African dust export from January 2006 to February 2007 and we investigate the seasonal variations of dust transport over the Atlantic with a focus on the vertical extension of the plumes. Several episodes of dust outbreaks that occurred during this period could be followed from regions nearby the Saharan sources up to the Caribbean conjointly in products provided by MODIS and CALIOP. We discussed in particular a 7-day episode in July 2006 and a 5-day episode in February 2007 to contrast the summertime versus wintertime vertical extension of the plumes. We find that the model-derived vertical profiles of attenuated backscatter match reasonably well the CALIOP products as the plumes are traveling over the Atlantic Ocean. This indicates that the altitudes at which the plumes travel are fairly well represented in the model. The observation and simulation results support that dust plumes extend up to 8 km in summer over the Atlantic, while the maximum altitude reaches approximately 4 km in winter. We also find evidence that the plumes gradually descend throughout their travel over the Atlantic Ocean. Comparisons between model results and observations from MODIS and CALIOP suggest however that too little dust reaches the Caribbean region in the model, especially in the summer, which may be due to (1) too weak transport from the source region to the Caribbean and/or (2) too strong (wet) deposition over the Caribbean regions. Finally, although evaluating the representation of clouds in the model was not foreseen as a main objective of this study, we find the simulated cloud patterns to be temporally and spatially consistent with the observations. This is an encouraging preliminary result as a quantitative assessment of the cloud-aerosol interactions and of their impact on climate requires a good representation of both aerosol and cloud vertical distributions in models.

[22] We then apply our simulation to quantify dust deposition over the northern Atlantic Ocean as dust is an important nutrient for marine and terrestrial ecosystems. We estimated that 1120 Tg of dust were deposited globally in 2006 (over land and ocean). According to our simulation, more than half of the global amount was deposited in the Atlantic, northern Africa and southern Europe with 13% (151 Tg) being deposited over the tropical Atlantic (5°N–27°N and 100°W–17°W) in 2006. There were strong seasonal variations, almost half of the deposition of dust in the tropical Atlantic (46%) occurred in summer, 10% in the fall, 20% in winter and spring. The 7-day dust episode that we investigated in this study yields 5 Tg of dust deposited in the Atlantic (4% of the annual amount for that region). Note that these values are likely to be an upper range estimate as dust deposition may be overestimated in our simulation in that specific region of the world.

[23] Although this study describes a preliminary work in terms of the analysis of aerosol vertical profiles over an extended region and time period, it demonstrates that the CALIOP products (used in conjunction with the MODIS products for example) can provide an additional set of information for the evaluation of chemical and transport models. Additional work is in progress in order to perform more quantitative comparison between model results and the CALIOP products.


[24] The CALIPSO data were obtained from the NASA Langley Research Center Atmospheric Science Data Center via the ICARE network ( The authors gratefully acknowledge the ICARE team for providing and maintaining the computational facilities to store and treat the CALIPSO data. The GEOS-Chem model is managed by the Atmospheric Chemistry Modeling group at Harvard University with support from the NASA Atmospheric Chemistry Modeling Analysis Program. The MODIS data used in this study were acquired as part of the NASA's Earth Science Enterprise. The algorithms were developed by the MODIS Science Teams. The data were processed by the MODIS Adaptive Processing System (MODAPS) and are archived and distributed by the Goddard DAAC. The authors gratefully acknowledge the AERONET principal investigators and their staff for establishing and maintaining the eight sites used in this investigation. The authors are also very grateful to Laurent Benoit for providing the dust sources from his Ph.D. work.