Dust size distributions are characterized by the presence of a prominent coarse mode (particle size greater than 1 μm radius) in contrast to urban and biomass burning aerosols, which yield abundant fine-mode aerosols (particle size less than 1μm) [Dubovik et al., 2002]. Gravitational settling efficiently removes large particles, and consequently, the dust size distribution shifts to smaller radii with increasing transport time. Eck et al.  show that the dominance of one mode over the other can be measured with the Angstrom wavelength exponent α. The α values range from −0.5 to 0.5 in dusty environments; they are greater than 0.5 in polluted regions. Schepanski et al.  have imposed α< 0.6 to detect dust using the infrared channels of the Spinning Enhanced Visible and Infrared Imager instrument on the Meteosat Second Generation satellite. However, we would like to apply a more stringent criterion so as to screen out all scenes with any significant amount of fine-mode particles, either from other aerosol types or aged dust. Many measurements in dusty environments yield negativeα during dust outbreaks and near dust sources: from −0.2 to 0.04 (during three flights) in Niger [Osborne et al., 2008], −0.06 (three events) in Delhi [Singh et al., 2005], −0.5 (one event) in Spain [Cachorro et al., 2000], less than 0 at Birdsville (many events mostly in summer but also in fall) in the Lake Eyre basin [Radhi et al., 2010], and −0.24 (typical during dust storms) in Tengger Desert in northern China [Xin et al., 2005]. Cheng et al.  reported negative values 11.4% and 6.7% of the time over 5 years of measurements near Chinese dust sources at Dunhuang and Yulin, respectively.
 Although M-DB2 products are retrieved only over bright surfaces in the visible (to not be mistaken with brightness in the deep blue, which is always small), thereby excluding oceans, there may be scenes in coastal regions where sea salt concentrations are high. Because sea salt has a significant amount of coarse-mode particles, lowα values could result. To avoid this situation, we require that the single scattering albedo ω at 412 nm is less than 0.95. For scattering aerosols such as sea salt ω is near 1. This second criterion efficiently eliminates sea salt–dominated scenes. We should note that some dust sources contain a large percentage of salt, for example, the Aral Sea [Rudich et al., 2002]. Internal mixing of dust and salt will affect the optical characteristics of pure dust, but it is not clear how it could affect our results. The mixture will still absorb shortwave radiation, although more weakly. In the following sections, we will show that our scheme successfully detects dust near salty sources such as the Aral Sea, Owens Lake, and Great Salt Lake, among others.
4.1. Comparison With AERONET
 To evaluate our screening method using M-DB2 products, we apply it to the direct measurements of AOD made in the AERONET Sun photometer network. AERONET is a federated worldwide network of Sun photometers that are monitored and maintained at NASA Goddard Space Flight Center [Holben et al., 1998]. We use aerosol optical depth and the Angstrom exponent (440–670 nm) level 2 data, which are cloud screened and quality assured; these are available at http://aeronet.gsfc.nasa.gov. From all measurements collected between 2003 and 2009 and from all sites, we extract AERONET data between 12:00 PM and 3:00 PM local time. This provides a 3 h window centered at 1:30 PM, the local passing time of the MODIS instrument on Aqua. We only use M-DB2data within a 30 km window centered on the AERONET site. The spectral values of single scattering albedo are also retrieved by inversion of almucantar measurements [Dubovik and King, 2000]. The almucantar measurements are performed by keeping the same solar zenith angle while varying the azimuthal angle of the Sun photometer over 360°. However, the level 2 quality assured inversion products are computed only for AOD greater than 0.4. This condition would severely limit the number of collocated measurements. Therefore, the only dust criterion that we require of the AERONET data is that α < 0.
 Figure 2(top) shows the comparison between the spatially and temporally collocated mean AOD (550 nm) measured by AERONET and retrieved by M-DB2 algorithm, as well as, inFigure 2(bottom), the mean DOD (550 nm) extracted from AERONET and M-DB2 data. We found 195 and 13 AERONET sites with collocated measurements of AOD and DOD, respectively. There is a significant correlation between AERONET and M-DB2 for AOD as well as DOD. The root-mean-square differences are 0.11 and 0.26, while the mean absolute differences are 0.07 and 0.24 for AOD and DOD, respectively. Although AOD sites are widespread and include polluted regions with aerosol dominated by fine-mode particles, the screening method selects only sites known to lie nearby to dust sources. The largest biases in AOD and DOD are in California and Australia. In Africa, M-DB2 AOD is slightly overestimated while DOD is systematically underestimated by 25%–50%. The largest DOD value inFigure 2 corresponds to the Kanpur (India) site.
Figure 2. Comparison between AERONET and M-DB2 (top) aerosol optical depth and (bottom) dust optical depth at AERONET sites with collocated data between 2003 and 2009. (left) The standard deviation is added to the mean values. (right) The percent relative difference between M-DB2 and AERONET values, given using colored circles. The number of sites (n), correlation coefficient (r), root mean square difference (rmsd), and mean absolute difference (mean diff) are provided in the top left corner in Figure 2 (left).
Download figure to PowerPoint
 Table 1 gives the mean AOD and DOD values and the number of days with collocated measurements at the 13 AERONET sites. While there are hundreds of AOD measurements per site, the number of days with collocated measurements satisfying DOD conditions is dramatically reduced, ranging from 104 days in Agoufou (Mali) to a minimum of 8 days in Rogers Lake (California). Table 1shows that at all sites the mean DOD is significantly larger than the mean AOD, with the highest mean DOD (1.5 for M-DB2 and 1.4 for AERONET) at Kanpur (India). The lowest DOD values are observed at the Tinga Tingana and Birdsville sites, both of which are in the Lake Eyre Basin of Australia. The annual AERONET DOD is 0.16 for both sites; M-DB2 values are higher, 0.34 for Birdsville and 0.19 for Tinga Tingana. These data show, as expected, that the imposed conditionα< 0 effectively serves to identify dust events and to discriminate against air parcels containing fine-mode pollutants.
Table TABLE 1.. Collocated Mean Aerosol (AOD) and Dust (DOD) Optical Depth at 550 nm Measured by AERONET and Retrieved From M-DB2, at 13 AERONET Sitesa
|Cape Verde||Sal Island||16.73°N||22.93°W||314||0.43||0.36||10||0.99||0.69|
|Solar Village||Saudi Arabia||24.9°N||46.39°E||1287||0.34||0.31||55||0.88||0.54|
4.2. Seasonal Distribution
 Comparisons between monthly mean and standard deviation of AOD from AERONET and M-DB2 and DOD from M-DB2, calculated from measurements between 2003 and 2009, are shown inFigure 3. For this comparison, all M-DB2 data are selected within a 30 km window centered on the location of the AERONET sites but without restricting local passing time between 12:00 PM and 3:00 PM. The values inFigure 3are calculated by averaging all valid AERONET level 2 measurements and M-DB2 retrievals. M-DB2 monthly AODs are within the standard deviation of AERONET data in Africa, Arabian Peninsula, and India but largely overestimated in California, Australia, and Israel. These discrepancies were already apparent inFigure 2. Most sites in Africa are located within or approximate to the Sahel, one of the most active dust sources in North Africa (Dakar, Agoufou, Cinzana, Banizoumbou, and Soroa). The Cape Verde Islands site is located off the west coast of North Africa, under the path of much of the dust that emerges from North Africa. Tamanrasset is a mountain site (1377 m above sea level) located in the Sahara. At the Tamanrasset site, there is a distinct peak of M-DB2 DOD in June, in agreement with the measurements taken in 2006 byCuesta et al. at that location. The seasonal cycle at the Sahel sites differs from Tamanrasset in that dust is a significant contributor to AOD from January to July. The M-DB2 seasonality is supported by measurements made byRajot et al.  at Banizoumbou during the African Monsoon Multidisciplinary Analysis field campaign in 2006.
Figure 3. Monthly AOD from AERONET (mean, black dots; standard deviation, vertical line) and M-DB2 (mean, black bold line; standard deviation, grey shading) and DOD from M-DB2 (mean, brown line; standard deviation, brown shading) at 12 sites whose locations are given inTable 1.
Download figure to PowerPoint
 In Solar Village (Arabian Peninsula), there is a pronounced maximum of M-DB2 DOD in April-May; dust activity weakens rapidly in summer, reaching a minimum in winter, in accordance with observations ofSabbah and Hassan . At Kanpur (India), large amounts of dust are observed during the premonsoon season with M-DB2 DOD reaching 0.8 in May-June; in contrast, M-DB2 DOD shows no dust during the other seasons. The greatest discrepancies are seen at Sede Boker (Israel), Birdsville (Australia), and Rogers Lake (California), where M-DB2 AOD is largely overestimated.
 The 7 year mean seasonal variation of M-DB2 AOD and DOD at 550 nm is shown inFigure 4. DOD distribution is plotted for all values of DOD greater than 0.1; elsewhere, AOD is plotted so that the relative distribution is made visible. The Northern Hemisphere is clearly much more dusty than the Southern Hemisphere both in terms of the absolute values of DOD and the spatial coverage. The same is true for AOD. In both hemispheres, fall is the season with the lowest DOD values: September, October, and November (SON) in the Northern Hemisphere and March, April, and May (MAM) in the Southern Hemisphere.
Figure 4. Global distribution of M-DB2 seasonal mean aerosol optical depth (blue) overplotted by dust optical depth (red).
Download figure to PowerPoint
 There are many regions with DOD > 0.1 all year long. The most widespread dust activity is seen in North Africa, especially within the Sahel, a region that is broadly defined in terms of rainfall as the zone lying between the 100 and 500 mm isohyets [National Research Council, 1983], which in West Africa lies roughly between 14°N to 20°N. This region encompasses three phytogeographical divisions: the northerly Sahelo-Saharan zone (grass steppe), between the 100 and 200 mm isohyets; the Sahel proper (tree steppe), between the 200 and 400 mm isohyets; and the southerly Sudano-Sahelian borderlands (shrub savanna), extending to the 500 mm isohyet. The Sahel has been the focus of much interest because of the great increase in dust activity that occurred following the onset of prolonged drought in the early 1970s [Prospero and Lamb, 2003].
 In West Africa, DOD is consistently higher in the Sahel (including Senegal, Mauritania, Mali, Niger, and Chad) than in the Sahara, although most studies of dust activity have focused on major sources in the Sahara [Prospero et al., 2002; Schepanski et al., 2007]. However, Maurer et al.  pointed out that the Sahel region is one of the Earth's most wind erosion–prone zones because these soils, which largely overlie sand sheets, are intensively developed for agriculture and thus become vulnerable to wind erosion. The region is influenced by the dry Harmattan winds from the north and the monsoon flow from the Gulf of Guinea. These two flows converge at the surface along the Intertropical Discontinuity (ITD) and in the free troposphere along the ITCZ. There does not seem to be any seasonal variation of these hot spots in Figure 4 despite the fact that the ITD shifts from 5°N in winter to around 18°N in summer [Bou Karam et al., 2008]. This quasi-permanent maximum DOD in the Sahel may be a result of the combination of both emissions from local sources and transport from other upwind regions.Klose et al.  analyzed weather reports from 1983 to 2008 and found the existence of a zone of frequent dust events and high dust concentration in the Sahel. The dust events are reported as mostly dust in suspension, which suggests that transport from the Sahara to the Sahel is more important than local emissions. Nonetheless, weak dust sources in the Sahel may be significant as pointed out by the modeling study of Guelle et al. .
 It is interesting to note that most models do not produce a large amount of emissions from the Sahel. Among the 15 global dust models analyzed by Huneeus et al. , few reproduce the most southward displacement of the Saharan dust cloud in winter. This disparity may be related to the resolution of these models. Indeed, Bou Karam et al. observed haboob-type dust events during the passage of a density current that originated from a mesoscale convective system situated on the leading edge of the monsoon flow. Using a 20 km resolution model,Bou Karam et al.  suggested that emissions driven by strong surface winds associated with these density currents may contribute significantly to the total dust load over West and North Africa. However, using a model with similar resolution (25 km), Haustein et al. had difficulty simulating an observed dust storm in the Sahel associated with intense moist convection. Using a higher-resolution (7 km) regional model,Tegen et al.  could reproduce a heavy dust plume over the Bodélé depression, although the model underestimates wind speed over the region. Similar resolution will be necessary to confirm the importance of downdrafts from convective storms over the Sahel, as well as other monsoon regions.
 Some of the most intense hot spots are in the northern part of the Sahel. The best example is the much studied Bodélé depression (17°N, 18°E, 170 m) in Chad [Koren et al., 2006; Washington et al., 2006; Todd et al., 2007] which yields an annual mean DOD value greater than 0.75. This high value is in agreement with the mean AOD = 1.1 retrieved from Multiangle Imaging Spectroradiometer (MISR) and reported by Koren et al. . One of the few studies measuring DOD in this region [Osborne et al., 2008] reported values up to 0.8 during dust events over Niger.
 Over North Africa, there are regions where DOD < 0.1 and AOD > 0.25. They are mostly located in the Sahara where sulfate emitted by fossil fuel burning and transported from Europe [Lelieveld et al., 2002] may contribute significantly to AOD. Also along the southernmost areas of the Sahel, we expect carbonaceous aerosols from biomass burning to contribute significantly to AOD, especially in winter [Crutzen and Andreae, 1990].
 Over the Middle East, the regions with the highest and most widespread FoOs of DOD > 0.1 are in Mesopotamia and along the Persian Gulf in MAM. There are also a few local spots in the coastal regions of Yemen and Oman with mean DOD greater than 0.25, mostly in MAM and June, July, and August (JJA). In central Asia, DOD > 0.1 is found over the east Aral Sea, the southeast coastal region of the Caspian Sea, the eastern parts of Uzbekistan and Turkmenistan, and the southwest corner of Afghanistan, all of which are known sites of highly active dust sources.
 In Figure 4, India is characterized by a strong seasonal and latitudinal variation of DOD. The peak period for dust is March, April, and May (premonsoon), and the weakest period is in September, October, and November (postmonsoon). During the premonsoon period, DOD is >0.5 over the Indo-Gangetic basin. During the monsoon period (June, July, and August), the number of retrievals is too low (cf.Figure 1) to make any conclusion about DOD or AOD in Figure 4. But after the monsoon period and in winter, the region appears free of a significant amount of dust.
 Recently, Dey and Di Girolamo  derived a climatology of nonspherical aerosol optical depth over India using MISR data. For the most part, the seasonal variation and latitudinal gradients are similar, but the absolute values of DOD are more than a factor of 5 lower in their study. As shown above (Figure 3), we overestimate AOD by a factor of 2 in Kanpur in May and June, while in their study they underestimate AERONET AOD by a factor of 2 at Kanpur.
 Data are only consistently obtained in NW and NE China where the AOD and DOD distributions show a significant seasonal and spatial variation. DOD coverage and amplitude are at a maximum in spring, in agreement with previous studies [Sun et al., 2001; Wang et al., 2004]. DOD makes a significant contribution to total AOD only during this most active dust season and only in the NW region. It is notable that in NE China, AOD dominates DOD in spring, despite the fact that intense, large-scale dust events are common throughout the region. This is most likely related to the large contribution of fine-pollutant aerosol to optical depth and the low frequency of dust events, as discussed in the next section. The intense and widespread dust activity seen in NW China in MAM is mostly associated with basins that have been previously identified: Tarim, Qaidam, Junggar, and Turpan [Prospero et al., 2002]. Over these regions, the seasonal mean DOD varies between 0.1 and 0.5 in MAM but drops below 0.1 in most areas during the other seasons. In JJA, some areas of the Tarim and Qaidam basins are still dusty; in the Tarim, there are two strong sources (DOD maxima > 0.5), one in the NE and one in the SW of the basin. Ge et al. showed that M-DB2 retrievals agree relatively well with ground-based data during dust events in northwest China. They observed AOD varying from 0.07 to 2.5 during dust events, with M-DB2 performance improving with increasing AOD.Christopher and Wang  showed similar daily variations (from 0.2 to 1.5) during dust events over Dunhuang (40.1°N, 94.4°E), which is located near the Tarim Basin and Gobi dust sources.
 In North America, DOD > 0.1 are seen around Baja California and the southern high plains in Texas. Some hot spots with DOD > 0.25 are observed in MAM over the Salton Basin of southern California, the Gila Valley in southwest Arizona, along the Pecos River of southwest Texas, the Vizcano Desert of the central part of the Baja California, and the Playa de San Nicolas in the southern part of the Sonoran Desert in Mexico. In Europe, the only two regions with DOD > 0.1 are located in Spain's Meseta Central and Anatolia in Turkey, but only in JJA.
 In the Southern Hemisphere, Australia is the only continent that yields substantial areas with DOD > 0.1; dust activity is greatest in SON, Austral spring. Over some ephemeral lakes within the Lake Eyre Basin, DOD is higher than 0.25 from September to February. But, as seen in Figure 3, M-DB2 AOD and DOD are largely overestimated in Australia. The other regions in the Southern Hemisphere with DOD > 0.1 are mostly areas within deserts, e.g., the Namib (Namibia), Kalahari (Namibia), Atacama (Chile), and Sechura (Peru) Deserts.
4.3. Frequency Distribution
 In this section, we analyze the FoO of optical depth by region and season with the objective of developing a procedure to identify major dust storm days based on the relative frequency of magnitude of DOD and AOD. To this end we divided the continents into seven regions defined in Table 2. The number of samples per region is large and varies between 105 to 107 depending on the season. Table 3 provides the percent cumulative frequency for three values of optical depths (0.25, 0.5, and 1) and for each region and season.
Table TABLE 2.. Domain of the Continental Regions Considered in This Study
|Region||Longitude Range||Latitude Range|
Table TABLE 3.. Cumulative Frequency Distribution (Expressed in Percentage) of M-DB2 DOD and AOD (in Parentheses) ≤0.25, 0.5, and 1 Over Five Continental Regions for Each Seasona
|North America||21 (89)||78 (97)||98 (100)||9 (67)||53 (84)||90 (97)||11 (75)||55 (88)||90 (98)||23 (95)||78 (99)||98 (100)|
|South America||12 (69)||57 (85)||92 (98)||21 (92)||71 (97)||97 (99)||17 (86)||72 (94)||96 (99)||8 (60)||49 (79)||90 (96)|
|North Africa||13 (49)||51 (77)||91 (96)||3 (23)||25 (60)||84 (94)||4 (27)||27 (66)||89 (96)||23 (72)||71 (91)||97 (99)|
|South Africa||18 (72)||63 (88)||95 (99)||21 (90)||71 (97)||97 (100)||14 (77)||65 (88)||95 (98)||13 (65)||61 (83)||94 (97)|
|West Asia||12 (63)||52 (86)||90 (98)||4 (40)||31 (68)||82 (94)||12 (53)||46 (78)||88 (96)||18 (75)||69 (93)||97 (99)|
|Central Asia||9 (60)||58 (79)||90 (95)||4 (40)||32 (59)||76 (87)||8 (58)||52 (77)||87 (92)||14 (72)||71 (86)||95 (97)|
|East Asia||9 (60)||57 (75)||87 (91)||4 (52)||32 (66)||71 (85)||9 (69)||55 (83)||88 (94)||16 (76)||72 (87)||96 (96)|
|Australia||20 (90)||71 (96)||96 (99)||33 (99)||84 (100)||99 (100)||26 (96)||80 (99)||98 (100)||20 (79)||69 (92)||97 (99)|
 In all regions, for all seasons, and for all three optical depth values in Table 3, the cumulative frequencies of AOD are much greater than those of DOD. The frequency of AOD > 0.25 is generally lower than 50%, except during MAM in west and central Asia, while DOD is most frequently greater than 0.25 for all regions. Therefore, a DOD threshold DODthresh = 0.2 is selected to distinguish dust events from background aerosols.
 The global distribution of the number of days DOD > 0.2 for each season is shown in Figure 5. The global distribution of dust event days with DOD > 0.2 shown in Figure 5is broadly similar to that of the mean M-DB2 AOD and DOD shown inFigure 4. The most widespread occurrence and the highest frequencies are seen in North Africa. Within the Sahel, events with DOD > 0.2 occur at least seven times per season and more than 75% of the time in certain areas (e.g., Mauritania, Niger, and the Bodélé depression). Engelstaedter et al.  used visibility data to develop a global map of annual dust storm frequency which shows distributions in the Sahel that are remarkably similar to those in Figure 5, taking into consideration that our results are based on seasons. Other areas with frequent events are Mesopotamia in summer, the Iranian coastal region all seasons, eastern Uzbekistan and Turkmenistan in summer, and the Indo-Gangetic basin during premonsoon season. On the other hand, the number of dust events in Inner Mongolia and Mongolia appears to be low. Ground-based visibility data appear to support this low frequency of dust outbreaks in China. Using visibility data from 1988 to 2004 over the entire east Asian continent,Kurosaki and Mikami  showed that the frequency of dust outbreaks in China is greater than 4% (corresponding to 4 days per season in Figure 5) only in the Tarim Basin, the Gobi Desert, and the Loess Plateau. These are the regions in Figure 5where M-DB2 frequencies are greater than 7 days in MAM and JJA.
 In North America, the highest frequency of dust events is found in the southwestern U.S. and northern Mexico. Along the border between the U.S. and northern Mexico, events with DOD > 0.2 appear as frequently as 30% of the time in MAM. This is in agreement with the long-term record of visibility data at El Paso (Texas), where there is high frequency of blowing dust in spring [Novlan et al., 2007]. There is also considerable dust activity in the western great plains in MAM.