A height resolved global view of dust aerosols from the first year CALIPSO lidar measurements

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 113,, doi: /2007jd009776, 2008 A height resolved global view of dust aerosols from the first year CALIPSO lidar measurements Dong Liu, 1,2 Zhien Wang, 1 Zhaoyan Liu, 3 Dave Winker, 4 and Charles Trepte 4 Received 31 December 2007; revised 13 May 2008; accepted 21 May 2008; published 30 August [1] Based on the first year of CALIPSO lidar measurements under cloud-free conditions, a height-resolved global distribution of dust aerosols is presented for the first time. Results indicate that spring is the most active dust season, during which 20% and 12% of areas between 0 and 60 N are influenced by dust at least 10% and 50% of the time, respectively. In summer within 3 6 km, 8.3% of area between 0 and 60 N is impacted by dust at least 50% of the time. Strong seasonal cycles of dust layer vertical extent are observed in major source regions, which are similar to the seasonal variation of the thermally driven boundary layer depth. The arid and semiarid areas in North Africa and the Arabian Peninsula are the most persistent and prolific dust sources. African dust is transported across the Atlantic all yearlong with strong seasonal variation in the transport pathways mainly in the free troposphere in summer and at the low altitudes in winter. However, the trans-atlantic dust is transported at the low altitudes is important for all seasons, especially transported further cross the ocean. The crossing Atlantic dusty zones are shifted southward from summer to winter, which is accompanied by a similar southward shift of dust-generating areas over North Africa. The Taklimakan and Gobi deserts are two major dust sources in East Asia with long-range transport mainly occurring in spring. The large horizontal and vertical coverage of dust aerosols indicate their importance in the climate system through both direct and indirect aerosol effects. Citation: Liu, D., Z. Wang, Z. Liu, D. Winker, and C. Trepte (2008), A height resolved global view of dust aerosols from the first year CALIPSO lidar measurements, J. Geophys. Res., 113,, doi: /2007jd Introduction [2] Dust is one of main aerosol types in the atmosphere and plays an important role in modulating climate via a number of complex processes. Dust aerosols impact the radiation budget of the Earth atmosphere system both directly by scattering and absorbing solar and terrestrial thermal radiation, and also indirectly by modifying cloud optical properties and lifetimes. The shortwave radiative forcing of dust aerosols can be either positive or negative depending on the vertical distribution of dust, the underlying cloud cover, and the surface albedo. In contrast, the longwave radiative forcing is always positive [Tegen et al., 1996; Sokolik et al., 2001; Ramanathan et al., 2007]. Compared with anthropogenic aerosols, dust aerosols cover a larger area, have a longer life span, occur more frequently, and possess larger radiative influence [Husar et al., 1997]. During heavy dust events, the optical thickness of dust aerosols can reach 3.5 at a wavelength of 0.5 mm[pinker et al., 2001]. Moreover, dust aerosols can also 1 Department of Atmospheric Science, University of Wyoming, Laramie, Wyoming, USA. 2 Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei, Anhui, China. 3 National Institute of Aerospace, Hampton, Virginia, USA. 4 NASA Langley Research Center, Hampton, Virginia, USA. Copyright 2008 by the American Geophysical Union /08/2007JD interact with clouds and thereby influence the earth s climate indirectly. They may act as effective cloud condensation nuclei (CCN) and suppress precipitation [Rosenfeld et al., 2001]. They may also act as effective ice nuclei (IN) [Sassen, 2002; Zuberi et al., 2002], changing the number of clouds [Mahowald and Kiehl, 2003]. Depending on the temperature, dust aerosols can serve as CCN or IN, which may modify the global radiative forcing through changes in cloud properties [Mace et al., 2006]. The absorption of radiation in the ultraviolet (UV) region by dust aerosols may modulate the photochemical processes [Dickerson et al., 1997]. Since dust aerosols are a source of iron, and they may also modulate many marine biogeochemical processes [Fung et al., 2000]. [3] The radiative forcing due to dust aerosols has not been well quantified due to a poor understanding of their indirect effect, as well as limited observations on a global scale, especially on their vertical distribution. Current knowledge of global/hemispheric distribution of dust aerosols are mainly derived from satellite observations with passive remote sensors, e.g., the Advanced Very High Resolution Radiometer (AVHRR) [e.g., Carlson, 1979; Husar et al., 1997], Meteosat [e.g., Moulin et al., 1997], Total Ozone Mapping Spectrometer (TOMS) [e.g., Prospero et al., 2002] and Moderate Resolution Imaging Spectroradiometer (MODIS) [e.g., Kaufman et al., 2005]. However, these passive measurements have intrinsic limitations. For example, the AVHRR, Meteosat and MODIS measurements are mainly limited to oceans because the scattered radiance from most land surfaces is significantly 1of15

2 stronger than that from aerosols in the visible and near IR spectral regions. As a consequence, techniques based on passive measurements cannot provide reliable estimations of dust optical thickness over the source regions. TOMS can make aerosol content retrieval easier both over the land and ocean, taking the advantage of the UV surface reflectivity that is typically low and nearly constant over the continents as well as oceans [Herman et al., 1997; Torres et al., 1998]. However, TOMS cannot make an objective distinction between dust aerosols and biomass burning [Prospero et al., 2002] because both are absorbing aerosols in the UV spectrum. The TOMS retrieval of aerosol content is also strongly influenced by the vertical distribution of the absorbing aerosol layer [Torres et al., 1998; Chiapello et al., 1999]. Mostimportantly, these passive techniques cannot provide the vertical distribution of dust aerosols, which is important to evaluate the dust radiative forcing [Darmenov and Sokolik, 2005] and determine the indirect effect of dust through different cloud microphysics. Model intercomparisons have shown a large variation in transport characteristics and impacts partly due to the lack of vertical dust extent in the source areas [Sokolik et al., 2001]. Therefore for better understanding the generation and transport of dust, it is necessary to study the vertical and global distribution of dust aerosols. [4] The Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) satellite was launched successfully on April 28, 2006 [Winker et al., 2007]. The CALIPSO satellite carries a two-wavelength, polarization-sensitive lidar, named the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP). Information on the vertical distribution of dust aerosols provided by CALIOP will greatly help to refine their radiative studies on a global scale because of three main factors. First, by receiving the laser light backscattered from different layers of the atmosphere along the transmitted laser beam, CALIOP can accurately measure the global vertical distribution of aerosol and cloud [Winker et al., 2003]. Second, the capability of CALIOP to measure the depolarization ratio makes it possible to distinguish dust aerosols from other types of aerosol [Sassen, 2000]. Third, CALIOP can detect dust aerosols for any terrestrial surfaces during the day and night [Liu et al., 2006]. [5] In this paper, a global view of height-resolved dust aerosols distribution based on the first year of CALIOP measurements is provided, especially in the dust source regions. In section 2, the CALIPSO lidar level 1B and level 2 data products are introduced and the cloud screening and the identification methods of dust aerosols are described in detail. Section 3 presents new findings and discussions on the main dust source regions (i.e., North Africa, the Arabian Peninsula, the Indian Subcontinent and East Asia), which is then followed by a conclusion. 2. Data and Analysis 2.1. CALIPSO Lidar Data [6] CALIPSO flies as a part of a constellation of satellites called the A-Train [Stephens et al., 2002]. CALIOP is designed to acquire vertical profiles of two orthogonal polarization components of a backscattered signal at 532 nm and a total backscattered laser signal at 1064 nm from a near nadir-viewing geometry. Each CALIPSO lidar level 1B data product contains a daytime or nighttime portion data of an orbit. The primary products are three calibrated and geolocated lidar profiles: a 532 nm and 1064 nm total attenuated backscatter and a 532 nm perpendicular polarization component. These profile data are provided from 40 km to 2 km (relative to the mean sea level) for 532 nm and 30 km to 2 km for 1064 nm wavelength with variable vertical and horizontal resolutions within different altitude ranges. The downlinked highest sampling resolution is 30 m vertically and 1/3 km horizontally for the 532 nm data within the altitude range from 0.5 km to 8.2 km. The resolution for the 1064 nm data within this altitude range is 60 m vertically and 1/3 km horizontally. For other altitude ranges, the vertical and horizontal resolutions are 60 m and 1 km for km and 180 m and 5/3 km for km, respectively. The 532 nm data resolution is further reduced to 300 m vertically and 5 km horizontally within km. The finer resolutions are employed for the lower altitudes where aerosol and cloud generally have a larger spatial variability and stronger backscatter intensity. [7] The current CALIPSO lidar level 2 data products contain the cloud and aerosol layer reports along with column properties. Cloud layers found at 1/3 km, 1 km and 5 km resolutions are reported in the 1/3 km, 1 km and 5 km cloud layer products, respectively. Cloud layers found at 20 and 80 km spatial resolutions are also populated in the 5 km cloud layer product. Aerosol is currently only reported in the 5 km aerosol layer product, including layers found at 5, 20 and 80 km horizontal averaging. The layer properties include layer boundaries and optical characteristics. Temporal and geophysical location information is available for each vertical profile Data Analysis [8] The cloud screening and the identification of dust aerosols are two critical steps in the data analysis and are described in the following subsections Cloud Screening [9] Cloudy profiles in CALISPO lidar level 2 cloud layer products are identified based on the cloud-aerosol discrimination (CAD) method developed by Liu et al. [2004]. Color ratio, defined as the ratio of 1064 nm and 532 nm backscatters, is sensitive to the particle size of cloud and aerosol, and therefore is used to distinguish cloud from aerosol. In addition, the backscattered intensities from cloud are generally stronger than those from aerosol, which can also be used to separate cloud from aerosol. Cloudy profiles identified at the 1/3 km and 5 km horizontal resolutions are stored in the level 2 1/3 km and 5 km cloud layer products. First, the level 1B profiles are screened with the level 2 1/3 km cloud layer product. If one or more cloud layers are found in a 1/3 km profile, this profile is excluded from this aerosol study. Then, identified cloud-free level 1B profiles at 1/3 km resolution are averaged to 5 km horizontal resolution to improve signal-to-noise ratio (SNR). Cloudy conditions in these 5 km averaged profiles are further screened with the level 2 5-km cloud layer product. Only the remaining cloud-free profiles are searched for the presence of dust aerosols by checking 2of15

3 Figure 1. The probability density function (PDF) and cumulative PDF (right y-axis) of 1 2 km volume depolarization ratios for dust, biomass burning, and maritime aerosols observed during summer 2006 and for continental aerosols observed during winter The vertical line indicates the depolarization threshold of 0.06 selected to separate dust aerosols from other types. the layer averaged volume depolarization ratio (VDR) which is described in the next section. [10] We note that dense dust aerosol profiles over the source regions are often misclassified as cloud in the current data products because the dust optical properties (color ratio and backscatter intensity) are fairly close to the cloud during heavy dust loading episodes (optical depth >1 2). Hence a few heavy aerosol loading dust profiles are excluded in this analysis, resulting in a slight underestimation of dust aerosol occurrence over or near the source regions. Because heavy dust cases only count for a small fraction of the total dust events and high dust occurrences are limited over source regions, our results are not significantly affected by this misclassification Dust Aerosols Identification [11] Dust aerosols have a large linear depolarization ratio (the ratio of perpendicular to parallel polarization components measured by a lidar) due to the nonsphericity of dust particles which is different from other types of aerosol. Therefore the linear depolarization ratio is an effective parameter for the identification of dust aerosols [Murayama et al., 2001]. In this study, dust aerosols are first identified with a single threshold method based on the layer averaged VDR, which is the ratio of layer integrated perpendicular to parallel components at 532 nm based on the 5 km averaged level 1B data (the integrated perpendicular and parallel signals are calculated in the selected geometric height layers, i.e., km, km, km, km, km and km, above ground level, AGL). The VDR values reflect the contributions for molecular, dust and the other types of aerosol in the observed lidar backscattering signals. Although the VDR is not as good as the particle depolarization ratio (PDR) for dust identification, the VDR is more straightforward to calculate and can avoid potential errors in deriving the PDR due to the signal-to-noise ratio and attenuation correction. As discussed below, the VDR is sufficient enough to identify dust aerosol occurrence. The global variability of PDR is beyond the scope of this paper. [12] To select a proper threshold of VDR for the identification of dust aerosols, we derive the probability density functions (PDF) of VDR for four typical types of aerosol (i.e., desert dust, biomass burning, continental, and maritime). Seven regions are selected to represent these four aerosol types: desert dust aerosols (North Africa, 10 W 15 E and N; the Arabian Peninsula, E and N; Taklimakan desert, E and N), biomass burning aerosols (Central Africa, E and 15 S equator), continental aerosols (North America, W and N), and maritime aerosols (Pacific Ocean, W and 30 S 30 N; South Atlantic Ocean and South Indian Ocean, 45 W 120 E and S). Figure 1 shows the PDF of VDR for these four types of aerosol calculated from the 1 2 km measurements from June to August 2006 for desert dust, maritime and biomass burning aerosols and from December 2006 to February 2007 for continental aerosols. The reason for selecting the winter season to characterize continental aerosols is that the dust sources are less active in the northern United States. [13] Figure 1 clearly indicates that the desert dust aerosols are largely separated from the other three types of aerosol with larger VDR values due to irregular shapes of dust particles. The VDR distribution of the dust aerosols is centered at The maritime and winter continental aerosols have a VDR distribution peaking at zero because these types of aerosol are mostly composed of spherical particles. The biomass burning aerosols containing nonspherical but small black carbon particles have a nonzero peak at just Thus values of VDR significantly lager than zero indicate the presence of dust aerosols. Generally, the VDR is small for optically thin dust layers because in this case the molecular scattering (whose depolarization ratio is very small in the CALIOP case) dominates total parallel signals (molecular + particle scattering). Assuming a dust particle depolarization ratio of 0.35 and lidar ratio of 45 Sr, a VDR of 0.06 is equivalent to a particulate to molecular backscatter ratio of 0.26 (corresponding to an optical depth of for a 1 2 km aerosol layer and for a 3 6 km aerosol layer). Therefore a value of 0.06 is chosen as the threshold of VDR for identifying dust aerosols and is applied to the nighttime, cloud-free 5 km average data during the first year (June 2006 through May 2007) to derive the global dust distribution. Based on the cumulative PDF presented in Figure 1, only 1.4% of dust aerosols are missed and 2.3% of nondust aerosols are classified as dust aerosols with this threshold. All types of dust events, such as a dust storm (visibility < 1 km), blowing dust (1 10 km) and floating dust (<10 km), can be correctly detected by using the VDR threshold of However, when weak dust aerosols are mixed with heavy pollution the threshold may fail to detect them. Under this condition, the radiative effect of dust aerosols will be small compared to that of the pollution. [14] Although the VDR is not a perfect index for dust aerosol concentration since the depolarization ratio of dust aerosols is a function of particle shape and size, it does offer a relative measure of aerosol loading in terms of its contribution to total lidar backscattering signal related to molecular contribution. The mean dust PDR is 0.25, thus VDRs of and 0.2 suggest that dust contributions to the total lidar backscattering signals are 1.2 and 4.9 times that of the molecular contribution, respectively. The extinction coefficient of dust aerosols can be derived from 3of15

4 Figure 2. The seasonal maps of dust aerosols occurrence between 60 S and 60 N in four height layers derived from the first year CALIOP cloud-free nighttime measurements. CALIOP measurements to better measure their radiative impact, but it is also beyond the scope of this paper. vertical layers: km, 1 2 km, 2 3 km, and 3 6 km, AGL. The frequency of occurrence of dust aerosols (OCC) is defined as: 3. Results and Discussion OCCi ¼ Ni; dust =Ncf 3.1. The Dust Belt [15] The global distribution of dust aerosols in 1 1 grid boxes between 60 S and 60 N are calculated for four ð1þ where, Ni,dust and Ncf are the number of dusty profiles in the vertical layer i and the number of cloud-free profiles, 4 of 15

5 Figure 3. The seasonal maps of cloud-free profiles used in this analysis. The dust occurrence frequency over the region strongly affected by the SAA (white hole 45 0 S and W) is not available in this study. respectively, in a 1 1 grid box. For the 3 6 km layer calculation, dust occurrence in either the 3 4 km, 4 5 km or 5 6 km layer is counted as a dusty profile in 3 6 km layer. [16] The height-resolved, seasonal occurrence maps of dust aerosols are presented in Figure 2, which shows that dust covers a larger area in the northern hemisphere than in the southern hemisphere. The high-occurrence regions shown in Figure 2 indicate the major dust source regions which stretch from the western coast of North Africa to China, covering the Sahara and Sahel regions, the Arabian Peninsula, the northern India, the Tarim Basin and Gobi desert which concurring with those identified by TOMS [Herman et al., 1997; Prospero et al., 2002]. These major dust sources are located in the broad dust belt and are usually associated with topographical lows in the arid regions or on land adjacent to strong topographical highs or the intermountain basins as discussed in detail by Prospero et al. [2002]. In these source regions the annual rainfall is generally low, under mm. [17] Limiting dust identification to cloud-free conditions may underestimate dust occurrence over regions with frequent cloud cover. Seasonal maps of cloud-free profiles used in this study are presented in Figure 3, which clearly show a large regional contrast in cloud cover. A larger number of cloud-free profiles are available for all dust source regions, but cloud-free profiles are hard to find over the northern Pacific. Thus the impact of cloud on sampling is larger over transport regions than in the source regions. [18] Table 1 lists the percentage of area with OCC greater than 0.5, 0.3 and 0.1, respectively, from the equator to 60 N. The area impacted by dust is strongly dependent on the season. Spring is the most active dust season, with 20% of area between 0 and 60 N is influenced by dust at least 10% of the time. In contrast, dust aerosols only affect 12% of area in the winter. More importantly, 12% of area between 0 60 N is influenced by dust at least 50% of the time in the spring. Figure 2 and Table 1 indicate that significant dust aerosols are lifted above 2 km. 15.9% of area between 3 and 6 km layer is impacted by dust at least 10% of the time in spring and 8.3% of area between 3 and 6 km layer is impacted by dust at least 50% of the time during summer. Vertical extent of the dust is also highly dependent on the season. It peaks in 1 2 km in the fall, winter, and spring and 2 3 km in the summer. [19] Table 2 lists the mean values of dust OCC and VDR within the four height layers over six arid, semiarid, and dust deposit regions in the dust belt (North Africa, 10 W 30 E and 5 35 N; the Arabian Peninsula, 35 E 60 E and 15 N 30 N; the Indian Subcontinent, 65 E 90 E and N; Taklimakan desert, E and N; Gobi desert, E and N; Southeastern China, E and N). The boundaries of these regions are highlighted in Figure 4 over the mean surface altitude map. While the mean occurrence is a measure of how frequently dust aerosols happen in a given region and season, the mean VDR gives a sense of the intensity of dust aerosols. Table 2 and Figure 2 clearly indicate that North Africa and the Arabian Peninsula are the main dust sources in terms of dust occurrence and intensity. In general, the occurrence of dust aerosols decreases with height with a rate depending on season and region. Based on the VDR, their loading also decreases with height on average. [20] During the summer and spring, dust outbreaks are more active in this dust belt, and dust aerosols are often lifted up to 6 km AGL. In North Africa and the Arabian Peninsula, Table 1. The Percentage of Area With OCC Greater Than 0.5, 0.3, and 0.1 in the North Hemisphere Between 0 and 60 N OCC Height Summer Fall Winter Spring > km km km km > km km km km > km km km km of15

6 Table 2. Seasonal Means of OCC and VDR for Dust Aerosols in the Four Selected Height Ranges Over Major Dust Source or Deposit Regions Summer Fall Winter Spring Height OCC VDR OCC VDR OCC VDR OCC VDR North Africa 3 6 km km km km Arabian Peninsula 3 6 km km km km Indian Subcontinent 3 6 km km km km Taklimakan desert 3 6 km km km km Gobi desert 3 6 km km km km Southeastern China 3 6 km km km km summer is the most active season. Spring is the most active season in the Indian Subcontinent and Gobi desert region. In Taklimakan desert there are similar dust occurrences in the spring and summer, but the dust storms are stronger in terms of VDR in the spring than in the summer. The fine resolution vertical distribution of dust aerosols and their seasonal variations are presented the following sections for the major dust source regions. [21] These dust source regions are prone to the mobilization of dust aerosols. In Figure 2, the high dust occurrence regions over the ocean indicate the transport pathways from these source areas. The frequency of occurrence of dust aerosols normally decreases as they are transported away from the source regions due to deposition and dispersion. Figure 2 clearly shows two main dust transport pathways: the trans-atlantic transport of North African dust and the trans-pacific transport of Asian dust, which stretches the dust belt out toward the American continent. The North African dust is transported across the Atlantic all season long with strong seasonal dependencies of vertical and geographical distribution, while the trans-pacific transport of Asian dusts mainly occurs in the spring season [Husar et al., 2001]. The discontinuous pattern of the frequency of occurrence of dust aerosols over the northern Pacific may be caused by the limited cloud-free conditions available due to the high occurrence of cloud over this region. [22] In the Southern Hemisphere, CALIOP also detects some weak dust sources in South Africa and Australia, consistent with the TOMS measurements [Herman et al., 1997; Prospero et al., 2002]. We note that lidar signals over South America are strongly affected by the South Atlantic Anomaly (SAA). The lidar profiles affected by the SAA are successfully excluded by excluding profiles when the calibration constant uncertainty for 532 nm is larger than 2.5 Figure 4. The mean (2 ) topographic map based on surface altitudes under CALIPSO satellite tracks. The boxes indicate the boundaries of six selected dusty regions: (a) North Africa, (b) Arabian Peninsula, (c) Indian Subcontinent, (d) Taklimakan desert, (e) Gobi desert, and (f) Southeastern China. 6of15

7 Figure 5. The zonal (a) and meridional (b) mean vertical distributions of dust vertical occurrence, depolarization ratio, and 1064 nm attenuated backscattering coefficient over the North Africa region for four seasons, respectively. The mean dust layer top and surface altitudes are over plotted in the dust layer occurrence as solid lines with triangles and without triangles, respectively (provided by level 1B data). The region affected by the SAA is shown in Figure 3 as a large white hole over South America and nearby ocean. Therefore reliable dust occurrences over this region could not be derived with CALIOP data in this study North Africa [23] Arid regions in North Africa are one of the most prolific sources of atmospheric dust as indicated in Figure 2. Dust from these regions can be transported westward to America over the Atlantic Ocean, northward to Europe over the Mediterranean Sea, and eastward toward the Middle East over the Red Sea [Prospero et al., 1981; Moulin et al., 1997; Perry et al., 1997; Swap et al., 1992]. The emission and long-range transport of Sahara dust have been well documented but not their vertical distribution. With CALIOP data, we can explore the dust vertical distribution in terms of dust vertical occurrence, depolarization ratio, 1064 nm attenuated backscattering coefficient, and mean dust layer top height. The zonal and meridional means are calculated in 0.5 bins. [24] Dust layer base and top heights are determined based on depolarization data at 60 m vertical resolution. First, the height of maximum depolarization ratio in the highest dust layer (1 km resolution) is identified. Then the top of the dust layer is selected at a height where the depolarization ratio decreases below 0.06 searched from the maximum depolarization ratio height upward. Similarly, the dust layer base is identified by starting at the lowest dust layer and analyzing the depolarization ratio downward. Data points between the dust layer top and base with 1064 nm attenuated backscattering coefficient larger than km 1 sr 1 are used to calculate dust vertical occurrence (related to total dusty profile), mean VDR and mean attenuated backscattering coefficient. The results for the four seasons are shown in Figure 5. Because the molecular signal at 1064 nm is small, the magnitude of 1064 nm attenuated backscattering coefficient indicates aerosol loading in terms of laser backscattering intensity. However, it should be noted that attenuated backscattering coefficient at lower altitude is affected more strongly by attenuation than at a higher altitude. Thus weaker attenuated backscattering coefficient near the sur- 7of15

8 Figure 6. Same as Figure 5 except for the Arabian region. face under heavy dust condition does not indicate lower aerosol loading. [25] The distributions of dust vertical occurrence, which indicate the dust aerosol layer height statistically, show strong seasonal variations in their vertical extension over this region. During the summer and the spring seasons, mean dust layer tops are around 4 km with significant dust layers go as high as 6 km. A case study of an extensive dust event occurring in August 2006 over North Africa [Liu et al., 2008] has shown that the dust particles were lifted to 6.6 km over the source. Higher dust vertical occurrences in most profiles during the summer season indicate that dust aerosol layers are deeper in the summer than those in the other seasons. The winter season has the lowest mean layer top height at about 2 km, although a small portion of dust layers are lifted up to 4 km. The vertical aerosol loading based on attenuated backscattering signals at 1064 nm and depolarization ratio indicates that significant amount of dust aerosols are lifted into the free troposphere except in the winter season, when high aerosol loading is mainly trapped below 2 km and there is a noticeable decrease above 2 km. [26] The high occurrence and aerosol loading regions generally indicate the main source areas. Figures 2 and 5 shows a southward shift of dust-generating areas from N and 20 W 20 E in the summer to 0 15 N and 20 W 20 E in the winter. This shift seems consistent with the intertropical convergence zone (ITCZ) shifting among seasons, which is believed to be a main factor in controlling the southward shift of trans-atlantic dust transport [Prospero et al., 1981]. However, further study is needed to link the shifting of the dust-generating area with the atmospheric circulation in the ITCZ. The cell structure of the meridional means of dust vertical occurrence suggests that dust outbreaks are confined to some main localized regions. [27] Figure 2 clearly shows that the dust aerosols generated in North Africa are transported across the Atlantic all season long via the easterly waves in all four height layers that were analyzed. In summer, Sahara dust particles are transported furthest into North America. As documented in an earlier studies [Prospero and Carlson, 1972; Chiapello et al., 1995; Reid et al., 2002; Colarco et al., 2003], Figure 2 also shows that the significant dust aerosols are transported across the Atlantic above 2 km in summer. However, dust occurrences above 2 km decrease further away from the west coast of Africa while dust occurrences below 2 km increase or stay constant. This suggests that dust transport at the low altitudes 8of15

9 Figure 7. Same as Figure 5 except for the Indian region. is also significant in the summer season. During the winter season more dust aerosols are transported below 2 km. [28] Another seasonal difference is the range of dust aerosols carried across the Atlantic Ocean reaching North America or South America more in the summer and spring than in the winter. These seasonal differences in height layer and distance that dust aerosols are transported are consistent with the vertical structure of dust in the source region presented in Figure 5 (i.e., the higher the dust aerosols are lifted in the source region, the further and higher the dust aerosols are carried cross the Atlantic Ocean.) [29] Another noticeable feature of the trans-atlantic transport of dust aerosols is the southward shift of the dust band over the Ocean from the summer to the winter. In the summer, the dust band is mainly confined within 0 30 N while the dust band extends beyond 10 S in winter. This seasonal shift of the transported dust band is certainly related to the seasonal shift of dust outbreaking areas showed in Figures 2 and 5 as discussed previously. Prospero et al. [1981] attributed this seasonal change to the seasonal variation of the ITCZ The Arabian Peninsula [30] The zonal and meridional means vertical structures of dust aerosols over the Arabian Peninsula are presented in Figure 6. As revealed in Figures 2 and 6, dust outbreaks in this region are relatively less active in the winter than in the summer and spring. The dust occurrence maps during the winter (Figure 2) show that there are two active areas on the peninsula. One extends along the eastern side of the peninsula and the other is near the coast of Oman [Prospero et al., 2002]. Similar to North Africa, there is a strong seasonal variation of dust vertical transport. The mean dust layer top is close to 5 km during the summer and below 3 km during the winter. As showed in Figure 6, intense dust plumes can be lifted up to 6 km during the winter. As indicated by the depolarization and 1064 nm attenuated backscattering, dust layers are optically thicker in the spring than the other seasons. During the winter, there are few height bins with 1064 nm attenuated backscattering coefficient lager than km 1 sr 1, which are likely caused by cloud contamination. [31] Dust aerosols generated in this source region can also be transported to the Persian Gulf and the Arabian Sea [Ackerman and Cox, 1989; Husar et al., 1997; Herman et al., 1997; Torres et al., 1998]. Figure 2 shows a clear seasonal difference in the heights of dust aerosol transported to the Arabian Sea. During the summer and fall, dust aerosols are mainly transported above 1 km while significant dust aerosols are transported below 1 km during the spring which may due to the local monsoon circulation [Anderson and Prell, 9of15

10 Figure 8. Same as Figure 5 except for the Taklimakan region. 1992]. Figure 2 also shows that significant dust aerosols from the regions are also transported northward The Indian Subcontinent [32] Large areas of Pakistan and the northwestern India are quite arid, and extensive desert regions exist there, such as the Thar desert. The dust occurrence maps of the winter months show high occurrences over these source regions (Figure 2). Surface meteorological observations indicated that the maximum dust storm frequency is at Ganganagar (29.4 N, 71.7 E) [Middleton, 1986]. Identification of dust aerosols over this area (as well as many other Asian areas like southeastern China) from TOMS is challenging due to a high loading of pollutants [Herman et al., 1997; Prospero et al., 2002]. Dust activities in these main source areas are persistent all yearlong, which is clearly illustrated by the zonal and meridional means vertical structures of dust aerosols over Indian subcontinent (Figure 7). [33] During the summer, most of India is influenced by monsoons with plenty of precipitation and dominant southwesterly wind. The heavy dust aerosol region is largely confined to the northwest of 26 N. During the spring, dust aerosols are more intense and are transported to cover a large area. A similar seasonal variation of dust layer depth observed in Africa and Arabian regions is depicted in Figure 7. The zonal means of the dust layer top height also show a southward decrease in the fall, winter and spring, which reflects the southward dust transport and dynamical impact on dust transport in the region East Asia [34] Dust aerosols also originate in the deserts of Mongolia, northern China and Kazakhstan, where strong cyclone events and cold air outbreaks can kick up a large amount of fine, dry soil particles. The dust occurrence maps (Figure 2) clearly show two main source areas: the Tarim Basin area and the Mongolian Plateau [Xuan et al., 2004]. [35] The Tarim Basin consists mainly of the Taklimakan desert, one of the largest deserts in the world and one of the major dust sources [Merrill et al., 1989; Prospero et al., 2002; Zhang et al., 1997, 2003; Sun et al., 2000; Qian et al., 2002; Xuan et al., 2000]. The dust occurrence maps and Table 2 clearly show that dust activities over this area appear to be persistent almost all yearlong, reaching a maximum in the spring and a minimum in the winter. During the summer and fall, dust aerosols are mainly confined to the basin with occurrence frequency as high as 80% (Figure 2). This high occurrence concurs with other studies based on surface observations [Qian et al., 2002; Wang et al., 2006]. [36] The zonal and meridional means vertical structures of dust aerosols over the Tarim Basin are presented in 10 of 15

11 Figure 9. Same as Figure 5 except for the Gobi region. Figure 8. The zonal means indicate that intense dust layers are confined between 38 N and 42 N. The geographic setting of the basin surrounded by high mountains (the Tianshan Mts. in the north and the Kunlun Mts. in the south/ southwest) creates circulations in the basin which are favorable for dust to remain suspended for a long time in the air [Tsunematsu et al., 2005]. Although mean dust layer top height during the winter in this region is lowest when compared to other regions, the mean dust layer top is the highest in the spring, which is different from the summer observed in the African, Arabian, and Indian regions. Another noticeable difference between East Asia and the other regions is that dense dust aerosol layers can be found as high as 6 km in all seasons. The cell structure of the meridional means also suggest that there are several favorable locations for dust production, such as those around 81 E and between 84 E and 86 E. [37] The Gobi desert, located to the east of the Tarim Basin on the Mongolian Plateau, is another major dust source [Zhang et al., 1997, 2003; Sun et al., 2000]. Figure 2 shows lower dust occurrences over the Gobi desert, especially for the portion located in southern Mongolia, than those over the Tarim Basin area, except in the spring. Over the Gobi desert, a large portion of the fine particles from surface soils have been removed by Aeolian processes and primarily coarser particles remain and form a protective layer to prevent underlying fine material from wind erosion [Wang et al., 2008]. Therefore required wind speed threshold for dust emission is higher in the Gobi (7 m/s) than that in the Taklimakan (4 m/s). [38] The vertical structure of dust aerosols over the Gobi desert is presented in Figure 9. It shows a similar seasonal variation as observed in the Taklimakan desert with the highest dust layer mean top height (4 km) in the spring. In addition, dense dust aerosols can be lifted up to 6 km in all seasons, which is also similar to those observed in the Taklimakan desert. Studies show that the vertical extent of dust storms in these regions is strongly influenced by synoptic weather events [Tsunematsu et al., 2005]. [39] Dust aerosols generated in the Taklimakan and Gobi areas can be carried eastward by prevailing westerlies and can pass over China, North and South Korea, and Japan, as well as parts of the Russian Far East [Iwasaka et al., 1983; Uematsu et al., 1983; Zhang et al., 1997; Murayama et al., 2001; Uno et al., 2001; Chun et al., 2001; Natsagdorj et al., 2003]. This easterly dust transport is most active in the spring. As indicated in Figure 2, significant dust aerosols are carried further across the Pacific Ocean reaching North America [Duce et al., 1980; Shaw, 1980; Uno et al., 2001; Husar et al., 2001; Sassen, 2002]. This trans-pacific dust transport appears to occur mainly in the free troposphere, especially when they are transported further cross the 11 of 15

12 Figure 10. Simulated dust transport patterns by using the NOAA s HYSPLIT model with air parcels originated at 0.5 km over 38.5 N, 83.5 E in Taklimakan (a) and over 40.0 N, E in Gobi (b). 480-hour forward trajectories are calculated every 6-hour for the spring 2007 (March, April, and May). The simulated dust occurrences are calculated in 5 5 grids (see text for more information). Pacific Ocean. The trans-pacific dust transport also appears to happen in the winter. The lower dust occurrence transported across the northern Pacific may be caused by sampling bias due to high cloud occurrence, about also reflect a more complex transport pattern for dust generated in China as discussed below. [40] Figure 2 also shows that the eastward transport of dust aerosols in the spring is confined largely to land, forming a dust corridor starting from the northwest boundary to the southeast boundary of China, i.e., from the dust source areas across Hexi Corridor and the Loess Plateau to the southeastern China. In the winter, a high dust occurrence area is also observed over southeastern China, which is the historical dust depositional area [Zhang et al., 2003] or the floating dust occurrence area [Wang et al., 2005]. The suspended dust aerosols over this high occurrence area are optically thin as indicated by the mean VDR (Table 2), and appear to be transported mostly from the Tarim Basin source region. [41] To better understand the transport pattern of dust aerosols from the Taklimakan desert and Gobi desert regions, forward trajectory analyses are performed with the NOAA s HYSPLIT forward trajectory model [Draxler and Hess, 2004]. Initial air parcels used to start forward trajectory calculations are at 0.5 km AGL over 38.5 N and 83.5 E for the Taklimakan and over 40.0 N and Efor the Gobi, respectively. New air parcels are initialized every 6 hours from these locations during spring 2007 (March 1 May 30). For each parcel, a 480-hour forward trajectory is calculated. The simulated dust occurrences are calculated in 5x5 degree grid with all forward trajectories. If a trajectory passes a given grid box at a give height layer, dust aerosols are assumed to be observed. The dust occurrence frequency is then calculated as the ratio of total observed dust events in a given grid box at a given height layer to the total parcel number used for the forward calculations. The simulated air transport patterns for the two sources are presented in Figure 10, which agree with the observations quite well. Figure 10 shows that high dust occurrences observed over the southeastern China are from both sources and dusts from the Gobi desert can be transported further south than those from the Taklimakan desert. More importantly, dust particles that originated from these two important source regions can be carried by westerlies for a long distance. However, the forward trajectory analysis shows a relative low dust occurrence transport cross the northern Pacific, which agree with the observed magnitude. However, it also suggests that the observed trans-pacific transport pattern is affected by a limited cloud-free condition over this region. [42] Some dust events are observed over the Tibetan Plateau as showed in Figure 2. Tibet, often called roof of the world, is a high-altitude arid steppe interspersed with mountain ranges and large brackish lakes. The Tarim Basin is to the north of the plateau and the Gobi desert is to its northeast. The observed high occurrences in the spring are contributed by local sources and remote sources [Zhang et al., 2001]. With the CALIOP measurements in the summer 2006, Huang et al. [2007] showed evidence of dust transport to the plateau from the Taklimakan desert. Seasonlong forward trajectory analyses presented in Figure of 15

13 also suggest that significant amounts of dust from the Taklimakan desert are transported to the Tibet Plateau in the spring. The transported dust stacking up against the slopes of Tibetan Plateau can heat up the elevated surface air over the slopes by absorbing solar radiation. This elevated heat bump can further cause a large-scale circulation anomaly [Lau et al., 2006]. 4. Conclusion [43] Based on the first year of CALIOP measurements (June 2006 May 2007), a height-resolved global distribution of dust aerosols is presented for the first time. Based on the 1 km layer average depolarization ratio, dust aerosols, which include dust storms, blowing dusts, and floating dusts and even optically thin dust layers, can be effectively separated from other types of aerosol with a VDR threshold of Results indicate that spring is the most active dust season and 20% and 12% of areas between 0 and 60 N are influenced by dust at least 10% and 50% of the time, respectively. In contrast, dust aerosols in the winter are more confined to the source regions in the dust belt stretching from the western coast of North Africa to China. Dust aerosols are often lifted up to 6 km. Within 3 6 km height layer, 15.9% and 8.3% of areas between 0 and 60 N are impacted by dust aerosols at least 10% of the time in the spring and at least 50% of the time in the summer, respectively. The most influenced height layer is 1 2 km during the fall, winter, and spring seasons and 2 3 km during the summer. [44] The vertical distribution of dust aerosols amount and properties show a strong seasonal cycle in the major source regions. Over North Africa and the Arabian Peninsula, mean dust layer top is the highest in the summer and lowest in the winter. This seasonal change of dust layer top heights is similar to the seasonal variation of the thermally driven boundary layer depth. By contrast, mean dust layer top in the spring is the highest over the Taklimakan and Gobi deserts. This difference may reflect a different role of local meteorological conditions in controlling dust layer evolutions. The seasonal variation of dust layer thickness affects the dust transport distances from their sources. Dust mobilization is most active in the summer for the North Africa and the Arabian Peninsula and spring for the Taklimakan and Gobi deserts, coinciding with the deepest dust layer. [45] In the dust belt, the arid and semiarid areas of North Africa and the Arabian Peninsula appear to be the most persistent and prolific dust sources. African dust is transported across the Atlantic all yearlong with a strong seasonal variation in the transport pathways. CALIPSO lidar data confirms that dust aerosols are transported across the Atlantic mainly in the free troposphere in the summer and at low altitudes in the winter. However, results also show that trans-atlantic dust transport at low altitude is very important in all seasons, especially after crossing the midway of the Atlantic. The trans-atlantic dusty zones are shifted southward from summer to winter as noted earlier. CALIPSO lidar data reveals that there is a similar southward shift of dust-generating areas over North Africa, which may control the transport zone shift to some extent. Although dust storms over the Arabian Peninsula are active all season, dust aerosols from this region are mainly limited over land and nearby seas and oceans. Dust activities over the Indian Subcontinent show strong seasonal variations in dust production areas and the extent of southward dust transport due to the strong seasonal shifts of the wind pattern as well as the monsoons. [46] In East Asia, the Tarim Basin area (the Taklimakan desert) and the Gobi desert on the Mongolian Plateau are two major dust sources. Dust aerosols are produced more frequently in the Taklimakan desert than in the Gobi desert. During the summer and fall, dust aerosols from these two regions are confined to nearby the source region. During the spring and winter, significant dust aerosols from these two regions are transported over a long distance. Although their eastward dust transport is largely confined to the dust corridor starting from the northwest boundary to the southeast boundary of China, many strong dust storms can reach Japan and cross Pacific to North America in the spring. However, the trans-pacific dust transport occurs at lower frequency than the trans-atlantic dust transport. Season-long forward trajectory analysis shows a similar dust distribution pattern over East Asia and the Northern Pacific as observed and suggests that a large area is influenced by these two dust sources during the spring. [47] The observed large horizontal and vertical coverage of dust aerosols indicate the importance of them in the climate system through both the direct and indirect effects. The indirect effect of dust aerosols strongly depends on their vertical distribution. With a significant amount of dust lifted into the free troposphere, dust aerosols can significantly impact water and ice, as well as mixed-phase cloud properties in the dust source and transport regions. Therefore it is more complicated to evaluate the indirect effect of the dust aerosols here than those in the boundary layer. However, the combination of the dust observations from CALIPSO and the cloud observations from other A-train satellites, such as the vertical cloud profile from the Cloud- Sat satellite, offers the potential to study dust and cloud interactions globally. [48] Results presented here clearly illustrate the value of CALIPSO data in improving our understanding of dust production and transport and in better validating model simulations, which are important to evaluate dust impacts on the climate system. In this study, the occurrence and vertical distribution of dust aerosols is examined with nighttime data to take advantage of the better SNR of nighttime depolarization measurements. Therefore possible diurnal differences of dust aerosol vertical distribution are not reflected in the results, but could be studied with CALIPSO data in the future. Limiting this analysis to cloud-free conditions may underestimate the impact of dust in regions with high cloud occurrence. New approaches are being explored to reliably detect dust layers above cloudy layers and/or under optical thin clouds in the future. [49] Acknowledgments. This analysis is supported by the CloudSat project from NASA/JPL and CALIPSO project from NASA/LARC. The authors would like to acknowledge three anonymous reviewers for their constructive comments. References Ackerman, S. A., and S. K. Cox (1989), Surface weather observations of atmospheric dust over the southeast summer monsoon region, Meteorol. Atmos. Phys., 41, of 15