We have developed a Global Transport Model of Dust (GMOD) within a general circulation model, using comprehensive parameterizations of the emission and deposition processes from Wang et al. (2000). These parameterizations are modified to match the surface conditions and meteorological fields of the climate model. A 20-year simulation from the dust model predicts an average dust emission of 1935 ± 51 Tg a−1 and a global dust burden of 27.8 ± 0.8 Tg for particles whose radii are smaller than 10 μm. Comparisons with observations show that the GMOD reproduces reasonably well dust concentrations (mean bias MB of −0.67 μg m−3, normalized mean bias NMB of −8.0%, correlation coefficient of 0.96 at 18 sites), logarithmic total deposition (−0.62 g m−2 a−1, −36.0%, 0.84 at 251 sites), and aerosol optical thickness (−0.04, −26.7%, 0.80 at 16 sites). The simulated dust particle size distribution is consistent with observations; both have a volume median radius in the range 1.0–4.0 μm. We examine the temporal variation of dust transport on different timescales. The simulated interannual variability is small, but the seasonal variation is quite large in the Sahara Desert and central Asia. We pay special attention to the diurnal variation of dust; both observations and simulations show that dust mobilization is more active during the local daytime than nighttime. On a global and annual mean basis, the simulated ratio of the daytime maximum uplift to the nighttime minimum is 75. Both the dust burden and dry deposition of dust show a similar diurnal cycle peaking in the late afternoon.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The transport and climatic impact of mineral dust aerosol are major concerns in climate research today. Every year, about 200 to 5000 trillion grams (Tg) of dust mass are entrained into the air from arid and semiarid areas [Goudie, 1983]. The total mass of the suspended dust particles in the troposphere is about 20 Tg, which is nearly half of the total aerosol mass loading [Qian et al., 1999]. The haze and dust storms caused by these particles influence air quality and reduce atmospheric visibility. In addition to the environmental impacts, mineral dust aerosol has both direct and indirect effects on climate. Suspended dust particles can absorb and scatter shortwave and longwave radiation [Carlson and Benjamin, 1980; Miller and Tegen, 1998; Dufresne et al., 2002; Markowicz et al., 2003]. They can also influence cloud formation and modify the optical properties of clouds by acting as cloud condensation nuclei [Levin et al., 1996].
 The loading of dust aerosol is highly variable in space and time [Tegen and Fung, 1994], but global observational data are quite limited [Sokolik and Toon, 1996]. As a result, numerical simulations have become an effective way to study the climatic impact of dust aerosols [d'Almeida, 1986; Tegen and Fung, 1994; Ničković and Dobričić, 1996; Marticorena et al., 1997a; Wang et al., 2000; Ginoux et al., 2001; Woodward, 2001; Lunt and Valdes, 2002; Gong et al., 2003; Zender et al., 2003; Easter et al., 2004]. Tegen and Fung  simulated the global distribution of mineral dust with the three-dimensional tracer transport model of the Goddard Institute for Space Studies (GISS). The simulated seasonal variations of dust concentrations generally show reasonable agreement with the observed patterns. Wang et al.  developed a deflation module in the study of long-range transport of yellow sand over east Asia; the results of this study are in good agreement with observational records. Ginoux et al.  used assimilated data to drive the Georgia Tech/Goddard Global Ozone Chemistry Aerosol Radiation and Transport (GOCART) model and obtained a simulated dust climatological distribution, which includes dust concentration, deposition, optical depth, and size distribution. Zender et al.  designed a detailed Dust Entrainment and Deposition (DEAD) module to simulate the dust climatology in the 1990s. The module is embedded in a chemical transport model and driven by the National Center for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalyses [Kalnay et al., 1996]. These simulations mainly focused on annual and global mean emissions, burden, and seasonal variation of dust; few studies examined the diurnal variation of dust.
 In this study, a new dust model, designated the Global Transport Model of Dust (GMOD), is developed within a climate model. The model predictions are extensively compared with observations, including dust concentrations at low levels, deposition, dust optical thickness (DOT), and particle size distribution. We further investigate the temporal characteristics of dust transport on interannual, seasonal, and diurnal scales. In addition, this work is the first step of our studies on dust-climate interactions for paleoclimate and present-day climate, which will be reported subsequently.
 In the next section, a detailed description of the GMOD is presented. Section 3 presents the simulated dust climatology, including dust uplift, deposition, and optical thickness. In section 4, station-based observations are used for further evaluation. The observed data include low-level dust concentrations from the University of Miami Ocean Aerosol Network [Prospero, 1996a], dust deposition from the Dust Indicators and Records of Terrestrial and Marine Paleoenvironments (DIRTMAP) database [Mahowald et al., 1999; Kohfeld and Harrison, 2001], and aerosol optical depth and particle size distribution from the Aerosol Robotic Network (AERONET) [Holben et al., 1998; Dubovik et al., 2000]. Section 5 analyzes the temporal characteristics of simulated dust aerosols.
 The transport process of the GMOD is simulated with the tracer transport model in the IAP9L-AGCM. Different dust bins are transported independently in the model. The GCM is a C-grid point model with 4° × 5° horizontal resolution. It uses σ–p vertical coordinates (σ = ) to divide the atmosphere into nine levels, the top of which is set to 10 hPa. The dynamical framework for the model is a set of primitive equations discretized with the conservative format proposed by Zeng . Additionally, the model uses the Robert  filter to dampen the high-frequency noise caused by the leapfrog scheme used in the time integration [Zhang, 1990]. Using this method, the GCM is very stable in long-term simulations.
2.2. Dust Emission Scheme
 A large part of the uncertainties in dust simulation comes from the empirical parameterizations of the uplifting and removal processes. In this section, we will discuss the dust emission scheme used in the GMOD.
 Dust emission is associated with many environmental factors. Generally, dust storms are caused by strong, gusty winds associated with multiscale disturbances or convective activity. Dust mobilization is inhibited by surface-covering elements such as vegetation, snow cover, and rocks. It is also constrained by soil-binding conditions such as high soil moisture and high salinity. With these conditions, active dust-producing areas are confined to bare ground or sparsely vegetated ground in arid and semiarid places and to regions with strong winds (T. Tanaka, Global dust budget, in Encyclopedia of Earth, edited by H. Hanson, 2007).
 The parameterizations of dust uplift in models always take into consideration the above mentioned factors, though the combinations in formulation and weights of each parameter are diverse. Zender et al.  provided a review of the mobilization schemes that appear in contemporary dust models. There are generally two distinct classes. One group uses concise parameterizations that take into consideration the influence of surface wind velocity (and soil moisture content) with a macroscopical view, and the other group uses more complicated schemes to represent related microphysical processes. Although the latter group is more physically correct, the lack of a global distribution of the needed input factors restricts its accuracy.
 We use the empirical formulation proposed by Wang et al.  for the dust emission, which can be grouped in the first class as defined above,
where Qp is the uplift flux of particle size class p in units of kg m−2 s−1, C1 is the potential emission coefficient of the soil, C2 is an empirical constant set to 2.9 × 10−11 [Hu and Qu, 1997], u* is the surface wind friction speed calculated by the land-surface model in the GCM [Liang, 1996], RH is the relative humidity of the surface air as a fraction, which is closely related to soil wetness, and sp is the mass fraction of each size bin; this parameter will be defined later. Equation (1) indicates that there will be dust emission only when the surface wind friction speed exceeds the threshold u*t and the relative humidity of the surface air is below RHt. Faster winds and/or drier air induce more dust emission. Soils that are especially vulnerable to dust uplift or deflation could be incorporated through variations in C1, but the preferential soil region treatment of Ginoux et al.  has not been incorporated here.
 The threshold wind friction velocity u*t is an important factor for dust mobilization and is closely related to soil texture, surface atmospheric conditions, and the size of disturbed crusts [Marticorena et al., 1997b]. Some studies measure u*t directly in field experiments or wind tunnels [Musick and Gillette, 1990; Gillette et al., 1980; Batt and Peabody, 1999]. Based on these observations, several empirical parameterizations are proposed [Iversen and White, 1982; Marticorena and Bergametti, 1995; Shao and Lu, 2000]. However, most of these schemes are only appropriate for particles exceeding 50 μm in diameter because a lack of reliable experimental data restricts such validations for smaller particles [Shao and Lu, 2000]. Easter et al.  set the thresholds of surface wind friction speed to 25, 35, 45, and 75 cm s−1 for desert, arid/semiarid, scrubland/grassland, and other dust source regions. For simplicity, we employ a similar method used by Lunt and Valdes , by setting u*t to a uniform constant of 40 cm s−1 for all the model grids, although different soil textures and dust size distributions may correspond to distinct threshold values. The latter influences are reflected by the potential emission coefficient C1 and mass percentage sp in equation (1), respectively.
 The emission scheme in the GMOD is similar to that proposed by Gillette and Passi , which is also used in GOCART [Ginoux et al., 2001] except that, as noted above, we have not included their preferential source region treatment. Their formulation uses a wind speed of 10 m rather than the surface wind friction velocity. The threshold wind is determined by an empirical function, which includes the effect of surface wetness. The GMOD independently considers the soil moisture content by employing a linear relationship between dust emission and relative humidity of surface air, as shown in equation (1). The threshold RHt is set to 0.4, which is smaller than the value of 0.5 used by Ginoux et al. .
 The potential emission coefficient C1 represents the uplifting capability of the land surface. As is known, dust storms usually take place in arid and semiarid regions. However, the uplift dust mass is constrained by geographic conditions, such as vegetation, water, and snow cover [Lunt and Valdes, 2002; Zender et al., 2003]. In the GMOD, we consider such influences by relating C1 to the vegetation types of the GCM. Wang et al.  set this factor according to thirteen kinds of land conditions [see Wang et al., 2000, Table 1]. In the GMOD, we make some modifications by dividing C1 into two parts:
where α is a constant determined a posteriori. C′1 represents the emission potential in ratio to that of desert soils, which is listed in Table 1.
Table 1. Ratio Factor of Potential Emission Coefficient (C1′) in the GMODa
Index in AGCM
GMOD, Global Transport Model of Dust; AGCM, atmospheric general circulation model.
Mixed farming, tall grassland
Tall/medium grassland, evergreen scrubland
Short grassland, meadow and scrubland
Evergreen forest (conifer)
Mixed deciduous, evergreen forest
Tropical evergreen broadleaved forest
Medium/tall grassland, woodlands
Table 1 gives a description of the vegetation types in IAP9L-AGCM [Liang, 1996]. The global distribution of these plants is determined from a Koppen climate classification [Bi, 1993]. In the GMOD, there are only three kinds of soil that yield mineral dust. The emission capability is weighted as 0.1, 0.3, and 1.0 for tall grassland, meadow, and desert, respectively. A spatial nine-point smoothing is conducted to avoid the computational instability caused by abrupt changes in C′1. In this way, we obtained a global distribution of the potential emission coefficients, as shown in Figure 1.
Figure 1 reflects the location and intensity of dust sources in the GMOD. The largest dust sources are located in the Northern Hemisphere (NH), mainly in a broad “dust belt” that extends from North Africa, over the Middle East and central Asia, to China; this is consistent with observational data [Prospero et al., 2002]. The dust source in the Southern Hemisphere (SH) is located at approximately 30°S, with the largest center in the Australian Desert. There are remarkably few large-scale dust source regions outside of these regions. The source distribution shown in Figure 1 is in agreement with previous studies [Ginoux et al., 2001; Lunt and Valdes, 2002].
 The last factor in equation (1) that needs to be determined is sp, which is the mass percentage of each size class. This is potentially one of the most uncertain parameters in the GMOD and in other dust models. The cause of the uncertainty is not only due to lack of observations but also the fact that the dust particles change in size through agglomeration or interaction with other gases/aerosols during transport [Maring et al., 2003; Bauer and Koch, 2005; Zhang and Iwasaka, 2006]. At present, most models simplify the size distribution of mineral dust by dividing the total size spectra into several bins that do not vary. Similarly, the GMOD considers four bins of dust with idealized spherical shapes whose radii span from 0.1 to 10 μm. The mobilization of particles smaller than 0.1 μm by wind is limited because soil adhesion tends to form larger particles. On the other hand, those larger than 10 μm are always removed from the air in a few hours by their large gravitational settling velocity; this settling restricts their long-range transport and possible climate effects [Tegen and Fung, 1994].
 The dust spectra in the GMOD obey a power law distribution of n(r) ∝ r−3 as described by Mishchenko et al. , so that the mass content of each dust bin varies almost linearly with respect to the radius span. The four dust bins in the GMOD are 0.1–1.0 μm, 1.0–2.0 μm, 2.0–5.0 μm, and 5.0–10.0 μm, with corresponding effective radii of 0.39, 1.44, 3.27, and 7.21 μm. According to the definition proposed by the U.S. Department of Agriculture (USDA), the particles in the GMOD are clay (r ≤ 1.0 μm) and small silt (1.0 μm < r ≤ 10 μm), whose mass densities are 2.5 and 2.65 g cm−3, respectively [Hillel, 1982].
2.3. Dust Removal Processes
 Generally, there are two processes that remove dust from the atmosphere. One is dry deposition which is caused by gravity and turbulence, and the other is wet scavenging in and under clouds (i.e., wet deposition).
 Dry deposition is the most effective mechanism for removing large particles from the atmosphere [Tegen and Fung, 1994]. The efficiency of such sedimentation is measured by the deposition velocity, which is defined as follows [Wang et al., 2000]:
where Vg is the gravitational speed of the particle given by
where ρp and ρ are the densities of mineral dust and dry air, respectively. re is the effective radius of the particles, g is the gravitational acceleration, γ is the dynamical viscosity of air, and Cc is the Cunningham correction factor defined as [Seinfeld and Pandis, 1997]
where λ is the mean free path of air.
 Dust at all model levels is affected by gravity. For the particles at the lowest model level, however, turbulent mixing also has an effect in the deposition process. For equation (3), is the wind speed at the lowest level of the GCM, k is the von Karman constant (0.4), and Sc and St are the particle Schmidt number and Stokes number, respectively [Zhang et al., 2001].
 The formulation of the settling velocity in equation (4) indicates that Vg is quite sensitive to variation in the effective radius, which explains why larger particles are affected by a greater settling velocity. However, for the small particles (r < 1.0 μm), the Vd is so small that they remain suspended in the atmosphere for nearly 1 year if there is no other removal mechanism [Tegen and Fung, 1994]. In this case, wet deposition becomes an effective scavenging process. The wet deposition scheme in the GMOD can be described by the following equation:
where m is a scavenging coefficient, which is different among the four dust classes. They are empirically set to 1.0 × 10−4p, 1.2 × 10−4p, 1.2 × 10−4p0.83, and 1.5 × 10−4p0.83, respectively, where p is the total precipitation rate (mm h−1) including both convective and large-scale rainfall. The coefficients for wet deposition are comparable to observations [Volken and Schumann, 1993] and other simulation settings [Lunt and Valdes, 2002; Woodward, 2001].
 The GMOD also considers the possible influence of precipitation height. It is well known that most precipitation occurs below the tropopause, which means that there will be no wet removal when aerosols are suspended in the stratosphere. Tegen and Fung  give an empirical equation of H = 7 + 4 cos (2θ) (km) to define the vertical range for rain scavenging, where θ is the latitude. In their definition, wet scavenging takes place only below H. As a result, wet deposition could occur as high as 11 km for the equator region, where θ = 0. In contrast, the limit height is only 3 km in the polar regions. Similarly, wet deposition in the GMOD is restricted below the 4th model level which is about 300 mb, above which there are only high clouds that do not precipitate.
3. Model Climatology
 The performance of the GMOD is analyzed in this section. The model is driven by the meteorological fields from a GCM every 1 h. A sensitivity analysis shows that the model reaches equilibrium in several months (not shown). However, to obtain a more reliable result, we ran the model for 25 years, and the average of the last 20 years is considered to be the climatology. In reference to seasonality, four boreal seasons from spring to winter are denoted by MAM (March–May), JJA (June–August), SON (September–November), and DJF (December–February).
3.1. Global Dust Distribution and Budget
 The total emission of mineral dust is first estimated to determine the constant α in equation (2). Figure 2 shows a comparison of the climatological dust emission flux from different studies. Though different simulations generally have different size ranges of particles, the estimated dust uplift fluxes are within 1000 to 3000 Tg a−1. A similar error span in dust emission has been reported by the third assessment report of the Intergovernmental Panel on Climate Change (IPCC) [Penner et al., 2001]. In the GMOD, we set the a posteriori coefficient α to 10000 Kg s m−4. In this way, we obtain an average emission of 1935 Tg a−1 for the 20-year simulation.
 Given this definition of α, the spatial distribution of dust emission and deposition is studied and is shown in Figure 3. The uplift domain is generally consistent with the distribution of the ratio factor C′1, as shown in Figure 1. The source intensity is strengthened in the desert regions because the surface air over these regions is drier than the air elsewhere. About 94% of the total dust uplift takes place in the NH, most of which originates from the Sahara and the Taklimakan Desert. The largest source in the SH is the Australian Desert.
 The deposition domain is far larger than the uplift area. Since most of the dust sources are located in the NH, the sedimentation amount is correspondingly greater in this hemisphere. Smaller particles can be transported to high latitudes and contribute to the deposition over those regions. The total deposition is divided into the dry and wet contributions, as shown in Figures 3 (bottom). Most dry deposition occurs near the source regions because large particles easily settle, due to their large settling velocity. The dry deposition over the ocean regions is mainly composed of small aerosols. The westerlies near 30°N transport a great number of particles into the open sea, making a ring of deposition around the world (Figure 3, top right). The ring indicates that dust aerosols are transported around the globe, including across the Pacific [VanCuren and Cahill, 2002; Uematsu et al., 2003; Zhao et al., 2003] and the Atlantic [Lee, 1983; Prospero, 1996b] propagation.
 The predicted dust budget is listed in Table 2. It includes the total uplift, deposition, burden, and lifetime of dust. Table 2 shows that dry deposition accounts for a high percentage of the total deposition, especially for particles whose radii are larger than 5 μm. Wet deposition, which is the primary scavenging process for particles smaller than 2μm, comprises 33% of the total deposition. The percentage is smaller than the value of 41% given by Zender et al. , but larger than the value of 13% given by Ginoux et al. . The differences among the three studies are probably caused by the application of different scavenging coefficients [Zender et al., 2003]. In addition, the difference in particle size distribution may also contribute much to such inconsistency. For example, Table 2 shows that large particles (r > 5 μm) contribute 74% of the total dry deposition. However, Zender et al.  only considered particles smaller than 5 μm. These differences show the great uncertainty encountered in understanding the role of wet deposition in global dust transport. Observations of the ratio of the dry to wet deposition are quite limited on a global scale [Duce et al., 1991], which makes the validation of this aspect of the model difficult. As a result, more observations of dust dry and wet deposition are required to reduce model biases.
Table 2. Global Dust Budget
Effective R (μm)
Uplift (Tg a−1)
Dry (Tg a−1)
Wet (Tg a−1)
 The small mass of small particles keeps them suspended in the air for long periods of time. In contrast, large particles are removed from the atmosphere in about 1 day. This feature helps to increase the percentage of uplifted clay (r < 1.0 μm) partition in all. As a result, the particles in the four size bins make up 20%, 30%, 39%, and 11% of the total dust mass. The average lifetime of dust aerosols in the atmosphere is about 5.2 days, which is close to the result given by Luo et al.  and is consistent with other studies [Tegen and Lacis, 1996; Chin et al., 2002; Zender et al., 2003].
 A detailed investigation of the dust budget for different continents is shown in Table 3. Dust emissions in DEAD are as follows (units: Tg): Africa, 980; Asia, 415; Australia, 37; South America, 35; North America, 8 [Zender et al., 2003]. The GMOD has a stronger dust flux in source regions but a weaker intensity elsewhere, indicating differences in the dust source intensity in the two studies. The effects of dry and wet deposition are different for each continent. In Africa, Asia, and Australia, where there are large deserts, more large particles are available and, therefore, dry deposition comprises a high percentage of the total deposition. In Europe, where there is little dust emission, wet deposition is more than 5 times the dry deposition because most of the aerosols are small clay particles transported from Africa or Asia [d'Almeida, 1986; Prospero, 1996b]. In North and South America, dry and wet deposition are comparable.
 The dust budget in Table 3 reveals the main dust sources and continental sinks. The largest desert, the Sahara Desert, makes Africa the biggest source of dust in the world. Extensive arid and semiarid regions in Asia contribute large amounts of mineral dust aerosols to the atmosphere. In contrast, Europe is the primary continental dust sink, as many particles are brought to the ground by rain. A further comparison between the budget and deposition in Table 3 shows that large sinks always correspond to a high percentage of wet deposition. For example, the wet deposition accounts for 51%, 64%, and 86% of the total deposition in North America, South America, and Europe, respectively. This feature is a result of the combined effects of the geographic conditions and particle size distribution.
Table 4 shows the simulated dust deposition over the oceans. The main characteristic of oceanic deposition is that the wet deposition is greater than the dry deposition. The waters near the source regions, such as the North Pacific and the North Atlantic, receive more dust deposition than elsewhere. As a result, the deposition in the NH is generally larger than that in the SH. The seasonal variations of oceanic dust deposition are also shown in Table 4. The global oceanic deposition reaches a maximum in the boreal summer and a minimum in the boreal winter. The seasonal variation of deposition in the Atlantic is consistent with observations [Gao et al., 2001], both of which show peaks in JJA. The simulated dust deposition in the Pacific shows comparable intensities in MAM and JJA. These results differ from observations. For example, Holzer et al.  documented that the maximum transpacific transport occurs in the boreal spring. The underestimation of the dust deposition to the North Pacific in MAM indicates that the GMOD simulates weaker springtime dust sources in central Asia than observations. Such deviation has also been reported in other dust model evaluations [Ginoux et al., 2001; Woodward, 2001; Grini et al., 2005]. This underestimation may be induced by the underestimation of the anthropogenic sources in northern China and Mongolia. Tegen and Fung  considered the influence of the large domain of cultivated soil in the above regions and obtained a reasonable springtime aerosol optical thickness in east Asia.
Table 4. Oceanic Dust Deposition
Dry (Tg a−1)
Wet (Tg a−1)
 A comparison of the oceanic deposition with estimates from observations [Duce et al., 1991; Prospero, 1996a] and other models [Ginoux et al., 2001; Zender et al., 2003] is listed in Table 5. The largest oceanic deposition in the GMOD is in the North Pacific, which is consistent with Duce et al. , but different from the results of Prospero [1996a], Ginoux et al. , and Zender et al. , where the North Atlantic is the largest oceanic sink. The disagreement among models is probably caused by the differences in dust size distribution, simulated meteorological conditions, and the efficiency of the wet scavenging processes. However, the inconsistency between observations indicates that more in situ measurements are required to reduce the currently large biases in estimating dust oceanic deposition. Except for this disagreement, most of the results predicted by the GMOD are comparable to previous studies. For example, the simulated intensity of the dust deposition in the Atlantic is about 147 Tg a−1, close to the estimation of 140 Tg a−1 by Kaufman et al. . The deposition in the South Pacific, the South Atlantic, and the South Indian Ocean are close to that estimated by Duce et al. .
 The radiative forcing by mineral dust aerosols is sensitive to many factors, including the refractive index, height of the dust layer, dust particle size, and the dust optical depth [Liao and Seinfeld, 1998]. In this section, we will evaluate the predicted dust optical thickness (DOT) in the GMOD, which can be used as an indication of the predicted dust radiative effects.
 Generally, the DOT (denoted as τ in this paper) at a specific wavelength λ at one grid point can be calculated using the following equation [Tegen et al., 1997; Ginoux et al., 2001]:
where n denotes the sequence of the four dust bins in the GMOD, Mi, j, n is the dust burden of the nth bin at the (i, j) grid point, and Ke is the mass extinction coefficient, which is determined by Shi et al. :
In this equation, re and ρ are the effective radius and mass density of dust, respectively. Qe(λ) is the extinction coefficient of a spherical particle, which can be obtained from Mie theory [de Rooij and van der Stap, 1984; Mishchenko et al., 1999]. This depends on the dust size distribution and refractive index. The GMOD adopts a power law size spectrum [Mishchenko et al., 1999] for all dust classes (see section 2.2). For the refractive index, we set k = 1.53−0.0039i at 0.67 μm, which is interpolated from Woodward .
Figure 4 shows the DOT at 0.67 μm in different seasons. Clearly, the Northern Hemisphere has the largest optical depth in the boreal summer, with the maximum located in the Sahara Desert. The dust column density in central Asia also becomes the highest in this season, allowing the global mean of the DOT to reach as high as 0.049. However, as noted above, most observed dust events in Asia take place in the spring rather than in the summer. In the boreal winter, most intense sources in the NH become moderate and, as a result, the total global DOT drops to 0.014. The desert activity in Australia shows opposite seasonality, and, as a result, the DOT in this region is highest in DJF and lowest in JJA.
 The global averages of the DOT for the four different bins are 0.019, 0.007, 0.004, and 0.000. The largest percentage of the DOT comes from the smallest particles, though they comprise only 20% of the total dust mass. The second and the third dust bins show comparable optical depths. The DOT of the large particles is very small. The extinction coefficient Ke in equation (7), calculated by Mie theory, is inversely proportional to the particle size. This explains why small particles are responsible for a large DOT. For the same reason, the shortwave radiative effect of dust is mainly determined by small particles [Yoshioka et al., 2007].
 The annual global mean DOT at 0.67 μm is about 0.032 ± 0.001 in the GMOD. This value is consistent with the value of 0.030 ± 0.004 predicted by Zender et al.  and the value of 0.029 by Tegen et al.  at 0.63 μm. The optical thickness of mineral dust aerosols is comparable to the optical depth of anthropogenic sulfate, which is estimated in the fourth assessment report of the IPCC to be approximately 0.024 [Forster et al., 2007]. This result indicates that both anthropogenic and natural sources need to be considered when determining the total aerosol optical depth.
4. Station-Based Evaluation
 The simulated dust climatology is validated by observed data in this section. When discussing the model performance, we use the mean bias (MB), normalized mean bias (NMB), and correlation coefficient to quantify the results. MB and NMB are defined as follows [Liao et al., 2007]
where Pi and Oi are the modeled and the observed result at site i, respectively. n is the number of model-observed pairs for all qualified data.
 The comparisons between simulated and observed dust concentrations at low levels at all 18 sites are shown in Figure 5b. The MB of the simulation is −0.67 μg m−3, and the NMB is equal to −8%. The correlation between the modeled and observed results is 0.96. Generally, simulations in the North Atlantic and the Indian Ocean are in better agreement with observations than those at the other sites. This is probably due to the adoption of a reasonable source intensity in North Africa and the Arabian Peninsula in the model. The GMOD overestimates the dust density at sites 14–17 by a factor of 3–10. These stations are all located in the central Pacific. This overestimation is caused by an overestimation of the westerly wind and an underestimation of precipitation in this area as simulated by the GCM (not shown), both of which help to blow more dust over the Pacific. On the other hand, the modeled dust concentration at King George Island (site 10) is underestimated by a factor of 40. This site, located on the tail of the Antarctic Peninsula, is affected by dust particles that originate from South America [Li et al., 2008]. However, the GCM underestimates the source strength by setting the land as scrubland instead of desert.
 The monthly dust concentrations at each site are shown in Figure 6. Sites 1–7 are located in the North Atlantic and are generally influenced by the dust from the Sahara Desert. At Barbados (site 1), the observations indicate a maximum in May–July, but the GMOD predicts an active dust period from June to August. The simulated amplitude of the peak is about one half of that observed. As the nearest site to Barbados (site 1), Cayenne (site 2) has a similar dust amount but different seasonality, which peaks in the spring. The simulation at this site does not capture the springtime maximum but obtains a reasonable dust density in the other seasons. The simulated maximum dust concentration in Miami (site 3) appears in May, while the observed concentration peaks in July. The dust activity in Miami (site 3) is strongly affected by the dust from the Sahara Desert, which peaks in May in the GMOD. As a result, the simulated dust concentration shows a high value in May in Miami (site 3). In Bermuda (site 4), the model results and observations are in good agreement in terms of both seasonality and magnitude. Sites Izania (site 5) and Sal Island (site 6) are located along the west coast of North Africa. These sites contain the highest dust concentrations among the 18 sites. Izania (site 5) is located at an altitude of 2360 m above sea level, which is about 800 hPa in the free air. We use the mean value of the 7th and 8th model levels' dust density to denote the concentration at this site. The model results at both Izania (site 5) and Sal Island (site 6) do not capture the dust plume in early spring. Except for this disparity, the GMOD generally reproduces the seasonal variations and magnitudes at these sites. At Mace Head (site 7), which is located at a high latitude in the North Atlantic, the simulated seasonality is consistent with observations.
 Sites 8–10 are all in the southern oceans. Cape Point (site 8) is located at the southern tip of Africa. The observations at this site are incomplete. However, the range of results from the simulation is in reasonable agreement with the sparse record. Cape Grim (site 9) is close to Australia; the correlation coefficient between the observation and simulation at this site is about 0.85. This indicates that the GMOD reproduces both the magnitude and variability at this site well. The underestimation of the dust concentration at King George Island (site 10) was shown in Figure 5b and is caused by the application of an inconsistent surface type in South America.
 Kaashidhoo (site 11) is the only site located in the Indian Ocean. The GMOD reproduces a reasonable magnitude and seasonality of the dust concentration at this site, except for an overestimation of the springtime dust concentration. Okinawa (site 12) and Cheju (site 13) are in east Asia. The observed dust concentrations in these regions peak in the spring due to frequent dust storms [Zhao and Yu, 1990; Zhou et al., 2002]. Since the intensity and frequency of dust events vary from year to year, the observations in April at Okinawa (site 12) and Cheju (site 13) show large variability. The GMOD underestimates the dust maximum in the spring for both sites, which causes an underestimation of the springtime transpacific transport, as discussed in section 3.1.
 Sites 14–18 are in the central Pacific. As the results in Figure 5b show, the dust concentrations at these sites are overestimated. This is very similar to the result of Ginoux et al. . However, Lunt and Valdes  obtained a good agreement between model and observations in the Pacific but less agreement at the Atlantic sites. The difference between the results of Lunt and Valdes  and that of the GMOD is probably caused by the difference in the particle size span. As an ocean near dust sources, the Atlantic receives more large particles than the Pacific. However, the dust model discussed by Lunt and Valdes  only simulated submicron particles. This is reasonable for the dust simulation over the Pacific but causes large biases for the dust over the Atlantic.
 Generally speaking, the GMOD reproduces the dust concentrations at most sites to within a reasonable range. The major deficiency is that the simulated dust density over the central Pacific is higher than in observations, which is probably caused by the biases of the meteorological fields in the GCM.
 Dust deposition records from the DIRTMAP database are used for model validation in this section. This database is comprised of geologic dust records obtained from ice cores, marine sediments, and terrestrial (loess) deposits [Kohfeld and Harrison, 2001]. Figure 7 shows the distribution of the sampling sites in DIRTMAP. There are a total of 251 sites, most of which are located in the oceans between the middle latitudes of the two hemispheres. The largest land records are from the Chinese loess plateau, which has the highest deposition rate in the whole data set.
 The comparison of dust deposition between the DIRTMAP records and the simulation of the GMOD is shown in Figure 8. As it shows, the observed deposition varies greatly among different sites from about 0.001 to 1000 g m−2 a−1. For such a large data span, it is more reasonable to calculate the MB and NMB of logarithmic deposition values. In this way, the calculated MB is −0.62 g m−2 a−1 and the NMB is −36.0%. The log correlation coefficient is 0.84, higher than the result of 0.76 given by Mahowald et al.  and 0.82 given by Lunt and Valdes  for the same data.
 The simulated deposition in the Atlantic shows the best agreement, most of which is a factor of 0.5 to 2.0 times that of the observations. The simulation for sites in the Pacific and the Indian Ocean are also reasonable. The deposition near the polar region is small. The GMOD captures this feature and obtains reasonable deposition in both Greenland and Antarctica. The largest deviation between the model and observations appears in China, where the GMOD underestimates the deposition by a factor of 22. The depositions at these sites are mainly obtained from the loess records, which include many large particles. However, the GMOD only considers the transport of particles smaller than 10 μm. The difference in the dust size range may explain the great deviation between the model and observations over that region. This underestimation accounts for a large portion of the disparity between simulation and observations. After eliminating the deposition records in China, the MB and NMB reduce to −0.07 g m−2 a−1 and −8.0%, respectively.
4.3. Dust Optical Thickness
 In section 3.2, the simulated DOT is analyzed and compared with other model results. In this section, we evaluate the simulated DOT with station-based observations.
 The observed aerosol optical thickness (AOT) that we used for validation comes from AERONET [Holben et al., 1998; Dubovik et al., 2000], which is a federation of ground-based remote sensing aerosol networks. However, the observed AOT usually includes information about sulfate, black carbon, organic carbon, sea salt, and dust, and it is hard to distinguish the contribution of each component. To solve this problem, we set up a number of criteria in choosing observational stations. First, the sites are located on land and, therefore, the influence of sea salt is excluded. Second, the sites are located in uncultivated land, where there is little or no human activity such that the influence of anthropogenic aerosols such as sulfate and black carbon are excluded. Third, the sites are located in or around desert regions and, therefore, the signal of dust is dominant. With these criteria, we selected the 16 AERONET sites shown in Figure 9a.
 The chosen sites shown in Figure 9a are generally in arid or semiarid regions. The comparison between model and observations at these sites is shown in Figure 9b. Overall, most of the simulated results are reasonable; the MB, NMB, and R are −0.04, −26.7%, and 0.80, respectively. The largest deviation is that the model underestimates the DOT in sites 13–16 by over 50%; these sites are located in South Africa, North America, and South America, respectively. We checked the vegetation types in those regions and found that the GCM underestimates the source strength by setting the vegetation type as scrubland instead of desert. This deficiency also explains the discrepancy between the GMOD and observations at high southern latitudes that is shown in Figure 5.
 A detailed investigation of the validation at each site is shown in Figure 10. Sites 1–4 are located in or near the Sahara Desert. The GMOD reproduces the seasonality and magnitude of DOT at these sites well, where the average correlation coefficient is up to 0.86. The model underestimates the JJA peak in Dahkla (site 1) and Ras El Ain (site 4) and slightly overestimates the magnitude in Izana (site 3).
 Sites 5–7 are located in the Arabian Peninsula. The simulation in SEDE BOKER (site 5) agrees with observations, except that the model overpredicts the DOT in JJA by a factor of 2. The simulation in Solar Village (site 6) is in good agreement with station records including seasonality, magnitude, and variability. The modeled DOT at Hamim (site 7) is underestimated in JJA, which is opposite to the situation at SEDE BOKER (site 5).
 Sites 8–10 are located in central Asia. The GMOD shows a DOT peak during JAS (July–September) in these areas. This trend is consistent with observations in Irkutsk (site 9) as opposed to the case at Tomsk (site 8) and Dalanzadgad (site 10). The failure in differentiating the seasonality of the DOT at the above three sites is attributed to the low resolution of the GCM, which makes it difficult to discern the subgrid surface conditions and meteorological fields in central Asia.
 Lake Argyle (site 11) and Birdsville (site 12) are located in Australia. The DOT values at these sites show seasonality opposite to those of the sites in the NH, such as Izana (site 3) and Solar Village (site 6). This trend has been discussed in section 3.2. The simulated seasonality at sites 11–12 generally agrees with observations, especially at Birdsville (site 12), with a correlation coefficient of 0.8.
 The underestimation of the DOT at sites 13–16 was shown in Figure 9b. The probable cause is the inconsistent surface vegetation types used in the GCM.
4.4. Size Distribution
 The particle size distribution is very important for studying the radiative effect of dust aerosols. Small particles are more efficient in scattering solar radiation, but large particles are more absorbing of longwave radiation [Yoshioka et al., 2007]. The observed particle size distribution used in our validation is from AERONET [Dubovik et al., 2000]. Six stations, shown in Figure 9a, are chosen because relatively longer and more complete records are found at these sites. We compare the simulated seasonal dust particle size distribution with observations in Figure 11. The vertical coordinate of Figure 11 is the value of dV/dlnr, where V is the vertically integrated volume of the dust particles. Generally, the observed size distributions show two peaks. One is at a submicron radius (fine mode), and the other is at a supermicron radius (coarse mode). Since the GMOD only simulates particles whose effective radii are larger than 0.39 μm (Table 2), only the coarse mode is discussed in the following.
 Tamanrasset (site 2) is located in the central Sahara Desert. In the active dust seasons, the total dust amount is much higher than that in the inactive seasons. This explains why the maximum of dV/dlnr in JJA is about 10 times that in DJF. The simulation captures the peak value of the median size (1–4 μm) particles well, though the predicted volume is about 5 times larger than that of observations in DJF. A similar case appears in Izana (site 3), which is on the west edge of the Sahara Desert. This partly contributes to the overestimation of DOT in boreal winter at these two sites, as shown in Figure 10.
 Solar Village (site 6) is on the Arabian Peninsula. The simulated size distribution at this site agrees with the observations, both in shape and magnitude. This explains why there is a good agreement in DOT between the simulation and observations here.
 However, the simulated dust size distribution at Dalanzadgad (site 10) shows deviations from observations in both seasons. The simulated dust volume at this site is overestimated by a factor of 3 in JJA but underestimated by a factor of 1/2 in DJF for the median size particles. The GMOD also underestimates the value of dV/dlnr in boreal spring (not shown). This explains why the model does not show the correct seasonality of DOT at Dalanzadgad (site 10), as discussed in section 4.3. The volume median radius for the coarse mode at this site is larger than that in the Sahara Desert. This means that the particles retrieved in central Asia are larger than those in North Africa. This again demonstrates the cause for the underestimation of dust deposition in China, as discussed in section 4.2.
 As sites located in the SH, Lake Argyle (site 11) and Birdsville (site 12) show opposite seasonality in the volume size distribution as compared to stations in the NH. The total volume of dust in JJA is about 1/3 of that in DJF. However, such seasonality is mostly attributed to the larger particles because the size distribution of fine particles shows little differences between these two seasons. Simulated results show correct seasonality and spectral distribution at these two sites, though the simulated dust activity at Birdsville (site 12) is overestimated by a factor of 2–3 in DJF.
5. Temporal Variation of Dust Climatology
 Although there are deficiencies, the GMOD generally performs well at many locations. The model predicts reasonable dust climatology compared with observations, especially the seasonal variation of dust activity. Based on these results, the temporal characteristics of dust mobilization and transport are discussed in this section. Variabilities on different timescales, including interannual, seasonal, and diurnal scales, are combined to obtain a comprehensive understanding of dust climatology. Results from other models and observed data are used to verify our findings.
5.1. Interannual Variability
 For the 25-year simulation, we calculated the last 20-year modeled mean and standard deviation σ of each dust field to denote the dust climatology and interannual variability, respectively. The results, denoted as ± σ, are listed as follows: uplift, 1935 ± 51 Tg a−1; dry deposition, 1305 ± 35 Tg a−1; wet deposition, 629 ± 17 Tg a−1; burden, 27.8 ± 0.8 Tg. Tegen and Miller  used prescribed sea surface temperatures to drive a GCM for 15 years. They obtained a similar magnitude of 20 Tg a−1 in assessing the interannual variability of the dust emission.
 However, the interannual variability of both the GMOD and Tegen and Miller  is small compared with those models driven by the reanalyses. Zender et al.  estimated a dust emission of 1490 ± 160 Tg a−1 and a dust load of 17 ± 2 Tg from the DEAD model, which is driven by NCEP/NCAR reanalyses [Kalnay et al., 1996]. Ginoux et al.  used the assimilated meteorological data from the Goddard Earth Observing System Data Assimilation System (GEOS-DAS) [Schubert et al., 1993] to drive the GOCART model. They simulated the dust climatology in 1987–1990 and obtained a dust emission of 1815 ± 132 Tg a−1. Marticorena and Bergametti  simulated the interannual variability of the dust emissions of the Sahara Desert in 1991 and 1992 with the ECMWF and found a variation of about 80 Tg between the 2 years.
 The underestimation of the variability of the dust budget in the GMOD is caused by the small variation in the meteorological conditions simulated by the GCM. Both the GMOD and Tegen and Miller  used fixed climatological sea surface temperatures as boundary conditions, which reduces the interannual variability of the atmosphere. In addition, the prescribed seasonal SST neglects the El Niño-Southern Oscillation (ENSO) signals, which are very important for the interannual variation of dust activities [Gong et al., 2006; Hara et al., 2006]. As a result, the interannual variability of the mineral dust aerosols will be smaller than that from a model driven by the actual meteorological fields.
5.2. Seasonal Variability
 The interannual variability of the dust budget is not as large as its seasonal variation [Marticorena and Bergametti,1996; Mahowald et al., 2003]. With the shift of the solar incident point between the two hemispheres, the dust emission, deposition, and the burden of each continent vary from month to month. In this section, we will discuss the seasonal variation of dust activities in each continent.
 Before continuing our discussion, we must investigate the reasonableness of the simulated seasonal variation of dust activities in the GMOD. In section 4, we have utilized observed concentrations and AOT to validate the simulation outputs. The results show that the GMOD generally reproduces reasonable seasonality at most sites. To expand this validation from a regional scale to a global scale, we compare the simulated dust burden with the absorbing aerosol index (AAI) from the Total Ozone Mapping Spectrometer (TOMS) [Torres et al., 2002]. The AAI is a good index for differentiating absorbing aerosols (such as dust and smoke) and nonabsorbing aerosols (such as sea salt and sulfate) [Torres et al., 1998]. The ground-based comparison over the Atlantic shows that the TOMS AAI provides a remarkably accurate picture of mineral dust distributions in the atmosphere over both continental and oceanic regions [Chiapello et al., 1999]. Prospero et al.  used TOMS AAI as an indication of mineral dust aerosol to study the environmental features of global sources of atmospheric soil dust. Similarly, we make a correlation between the simulated dust burden and TOMS AAI climatology, as shown in Figure 12.
 The TOMS AAI climatology is derived from the 7-year average of each month from 1984 to 1990, when the index is most stable [Mahowald et al., 2003]. After excluding missing data, the climatological monthly AAI is correlated with the simulated monthly average dust burden at each spatial point to obtain a coefficient that denotes the level of similarity between the model and observations. The result in Figure 12 shows that the simulated dust burden is highly correlated with the observed AAI in most areas. High coefficients (>0.8) appear in the Sahara Desert, Arabian Peninsula, and the Atlantic, where a large dust concentration is observed and simulated. Although the GMOD overestimates the dust concentration over the central Pacific (Figure 5b), the simulated dust seasonality is reasonable. Low coefficients (<0.4) appear in east Asia, where the dust model fails to reproduce the springtime dust maximum. The largest deviations (or negative correlations) appear in the Southern Ocean, the Sahel region, the east Pacific, and the mid-Indian Ocean, where other absorbing aerosols, such as carbonaceous aerosol, contribute significantly to the total AAI. However, these regions comprise a small portion of the total observed domain. This indicates that the GMOD generally reproduces reasonable seasonality of dust burden.
 Based on this conclusion, we begin our discussion of the dust budget in each continent. It is commonly known that the dust column density is dependent on emission, dry deposition, and wet deposition. However, there are differences between these fields, and it is difficult to derive the characteristics of global dust uplift and deposition from aerosol products such as AOT or AAI [Mahowald et al., 2003]. Since the ground-based observations are limited by the spatial domain, the model result may be a suitable alternative.
Figure 13 illustrates the seasonal variation of the simulated dust budget in different continents. They are obtained by an area-weighted sum of dust emission, deposition, and the burden in each continent. The corresponding seasonal variability is shown in Table 6.
Table 6. Seasonal Variability of Continental Dust Budgeta
Up (Tg month−1)
Dry (Tg month−1)
Wet (Tg month−1)
The first value is the monthly mean, and the second value is the standard deviation in 12 months.
93 ± 40
52 ± 23
5 ± 3
10 ± 5
61 ± 45
37 ± 26
14 ± 8
7 ± 4
6 ± 4
3 ± 2
0.7 ± 0.2
0.6 ± 0.3
1 ± 1
1 ± 1
1 ± 0.5
0.4 ± 0.2
0.2 ± 0.1
0.4 ± 0.1
0.6 ± 0.3
0.3 ± 0.1
0.1 ± 0.1
0.6 ± 0.5
4 ± 3
0.5 ± 0.4
161 ± 80
109 ± 55
52 ± 25
28 ± 13
 As the driest continent in the world, Africa yields the most mineral dust aerosols every month. The simulated dust uplift is 93 ± 40 Tg month−1. Dust uplift begins in March and ceases in October, with a maximum in the period of April through July. This seasonal variation is well described by Prospero et al.  and is revealed by the ground-based and satellite observations in our previous analysis (see Figure 10 (sites 1–4)). Dry deposition is the primary settlement mechanism for dust particles, which has a seasonality similar to that of dust mobilization. Wet deposition is limited due to the scarcity of precipitation in the Sahara Desert. Though its magnitude is only about 10% of dry deposition (Table 6), it still peaks in the boreal summer, which is a synthesized effect of the dust and rainfall maxima occurring at the same time.
 Asia is also a large dust source in the NH. It includes large areas of deserts located in the Middle East, central and south Asia, the Taklimakan Desert, and Mongolia [Prospero et al., 2002]. The latter two regions are usually the origins of springtime dust storms in east Asia [Sun et al., 2001]. However, the simulated dust emission in Asia peaks during May–August, rather than in the boreal spring. The main reason for this is that the total dust uplift in Asia is mainly attributed to dust sources in the Middle East and central Asia, where the dust uplifts are most active in the boreal summer [Prospero et al., 2002].
 To illustrate the Asian dust budget more clearly, we divide Asia into three sections. Region A is west of 60°E, which includes the Arabian Peninsula and the Iranian plateau. Region B is within 60°E and 90°E, which includes arid/semiarid regions in central and south Asia. Region C is east of 90°E, which includes Mongolia and east Asia. Figure 14 shows the simulated dust budget in these three subsections. Generally, the dust uplift in the first two regions is much higher than that in Region C. Moreover, the dust emissions in Regions A and B both peak in JJA, which accounts for the dust maximum for all of Asia occurring in this season. The seasonality of dust activity in Regions A and B can be verified by Figure 12. However, the seasonal variation in Region C fails in depicting the springtime maximum. This explains the low correlation between the simulated dust burden and TOMS AAI in this region (Figure 12).
 The above analysis reveals that the GMOD underestimates the springtime dust activities in east Asia, and the primary reason may be the underestimation of the dust source intensity in Mongolia (Figure 1). In fact, all the dust storms in east Asia are associated with cold air outbreaks. Sun et al.  analyzed the routes of cold air outbreaks, which can be divided into three broad categories: north, northwest, and west. These three pathways account for 32%, 41%, and 27% of the cold air outbreaks, respectively. The first two routes pass over the Mongolian arid region and cause dust storms in the downstream regions [see Sun et al., 2001, Figure 3]. From this point of view, an accurate depiction of the Mongolian desert sources is key to the simulation of dust activities in east Asia. The underestimation of springtime dust activities in east Asia has also been reported in other models, such as in the works of Ginoux et al.  and Woodward . However, this problem is well solved by Tegen and Fung  by considering the influence from the large domain of cultivated soil in Mongolia.
 The depositions in regions A, B, and C have different characteristics. Regions A and B are dust source regions and, as a result, the dry deposition is much higher than the wet deposition. The opposite trend is observed in region C, which indicates that a large percentage of the dust that affects east Asia originates from outside sources rather than from local sources. The wet deposition in east Asia peaks during May–July, during the east Asian summer monsoon breakouts.
 An interesting phenomenon can be found when comparing the seasonal variabilities of Asia and Africa (Table 6). Though the latter yields more dust on average, its variability is smaller than the former. This means the dust activities in Asia show large seasonality, and they are not as sustained as those in Africa.
 The dust activities in Australia show seasonality opposite to that in the NH. Both uplift and dry deposition become active in October–February. Ekstrom et al.  studied the 40-year records from 95 stations in Australia and found that most of the dust storms occur in September–February. This seasonal variation is well reproduced by the GMOD. The simulated wet deposition is quite small in Australia and peaks in the austral spring, which is the wet season of Australia.
5.2.4. North America
 The simulated magnitudes of dust uplift and dry and wet deposition in North America are all about 1 Tg month−1, all of which peak during June–October. However, Prospero et al.  pointed out that the dust season in North America usually begins in April–May, reaches a maximum in June–July, and ends in August–September. The seasonality in the GMOD in this region seems to lag behind the observations by 2 months. This deviation may be caused by the simulation deficiency of the intra-annual variability in North America by the IAP9L-AGCM.
5.2.5. South America
 The simulated dust activities in South America are underestimated in the GMOD. This has been discussed in section 4.3. The simulated dust uplift is about 0.2 ± 0.1 Tg month−1, and it shows seasonality similar to that in Australia. The largest component of the dust budget in South America is wet deposition, which is the result of the trans-Atlantic transport of Saharan dust. Correspondingly, wet deposition peaks during June–September. The simulated dust budget pattern is different from the result given by Tanaka and Chiba , in which the seasonality of dust activity in South America is quite similar to that in Australia. This deviation may be caused by the differences in the source intensity.
 Dust activities are infrequent in Europe, the continent that has the least dust emission. However, particles originating from North Africa can transport and deposit there [d'Almeida, 1986; Prospero, 1996b; Ansmann et al., 2003]. Figure 13 shows that wet deposition is far larger than the uplift and dry deposition in Europe. The maximum dust burden in Europe occurs in JJA, when the Africa dust emission is the most active in the year. The dust burden over Europe is higher than that of North and South America because Europe is closer to the Sahara Desert, the biggest desert in the world.
5.3. Diurnal Variability
 The diurnal variation of dust activities is very important for the study of dust radiative forcing. Dust absorbs and scatters both shortwave and longwave radiation and emits longwave radiation. These processes reduce the downward shortwave radiation reaching the Earth's surface but enhance the downward longwave radiation. As a result, the total surface radiative forcing (solar plus infrared) is negative during the daytime, due to the loss of sunlight by absorption and backscattering, and positive during the nighttime, due to the continuous longwave heating compensation [Cautenet et al., 1992; Claquin et al., 1998]. Therefore, the study of dust diurnal variation helps to illustrate the radiative effect of dust aerosols more accurately.
 Observations and models both show that there are large variabilities in the dust daily cycle. For example, station records of dust storms in different regions show that dust events always take place in the daytime [Orlovsky et al., 2005; Wu and Ren, 2007], when the surface winds are stronger and the air is more unstable due to the development of turbulent vortices. Field experiments also reveal that the maximum DOT tends to occur in the late afternoon [Wang et al., 2004; Shen et al., 2007]. However, these observations are always limited by space and time. A better way to understand the diurnal variation of dust is by modeling. Though several numerical studies reveal the importance of dust diurnal variation [Luo et al., 2004; Miller et al., 2004], related work is still rare.
Figure 15 shows the long-term mean daily cycle of dust uplift in four model hours. Dust uplift becomes active during the local daytime. However, after sunset, most of the dust emissions gradually weaken and are totally constrained at night. A more quantitative estimation of the daily variation is conducted as follows. We first convert the model hour on each grid point to the local hour. Then, the variables at the same local hours are assembled to form a new global field, based on which the global mean of the dust budget and meteorological fields are calculated and shown in Figure 16. The daily fields shown in Figure 16 are all normalized to facilitate comparison among the fields. Figure 16a shows that the dust uplift arrives at high values during 0900–1800 local time, which is consistent with the result of Figure 15.
 The diurnal variation of dust uplift in the GMOD is determined by the variations of related meteorological fields. Equation (1) indicates that the uplift flux is dependent on the surface relative humidity (RH2) and friction velocity (UVR). These variables have a distinct daily cycle in the model, as shown in Figure 16b, which has been verified by 6-h daily NCEP/NCAR reanalyses [Kalnay et al., 1996]. As Figure 16b shows, the lowest RH2 appears in the local afternoon because the largest saturated air humidity caused by the surface warming is expected at this time. In contrast, the UVR becomes larger in the presence of the sunlight. This is because the surface warming during the daytime increases the airflow and turbulent momentum. The association of the lower RH2 and higher UVR leads to more dust emission in the daytime. Conversely, the surface cooling at night increases the RH2 but decreases the UVR, which limits the dust uplift to a very low level.
 The dry deposition flux in the GMOD is dependent on the dust density at the lowest model level. Since most of the dust is concentrated at low levels, the total dust column density is also highly dependent on the low-level dust amount. This explains why the diurnal variation of the dust dry deposition is similar to that of the dust burden (Figure 16a). Both curves vary with the change of uplift, with about a 1-h lag, which is equal to the time step in the dust simulation. The maximum dust burden appears in the late afternoon, which is a synthesis of the time lag and turbulent perturbation. Liu et al.  carried out a sensitivity study of turbulence in the planetary boundary layer (PBL). They found that the turbulence becomes very active in the local afternoon, especially between 1200 and 1800 LT.
 Wet deposition shows different temporal variation from uplift. It is greatest at local midnight (Figure 16a) because rainfall usually takes place in the evening (Figure 16b), when the temperature is lower and the humidity is higher.
 The magnitude of the diurnal variation is different among each dust budget (see Table 7 and 8). Since the related global observations are quite limited for validation, Table 7 has a more qualitative meaning. The largest variation can be found in the dust uplift. The maximum emission is about 75 times the minimum, and such variation is so significant that the diurnal variability is larger than the daily average dust uplift. Depositions show moderate diurnal changes. The largest deposition is about 1.5 times the smallest deposition. The dust burden shows the least variation, with a standard deviation of only 1% of its mean value. However, dust activities are always regional. This small change in the global domain may indicate large alterations in the local areas.
Table 7. Magnitude of Diurnal Variation for Each Dust Budget
0.33 μg m−2 s−1
0.004 μg m−2 s−1
0.120 ± 0.135 μg m−2 s−1
0.10 μg m−2 s−1
0.064 μg m−2 s−1
0.081 ± 0.011 μg m−2 s−1
0.046 μg m−2 s−1
0.034 μg m−2 s−1
0.039 ± 0.004 μg m−2 s−1
55.4 mg m−2
53.4 mg m−2
54.4 ± 0.7 mg m−2
Table 8. Illustration of the Characters in Figure 2
 The daily cycle of the dust budget shown in Figure 16 has been verified by the observations. In addition to the observations in the studies mentioned above, a more comprehensive data set is used here for further validation. The data used are the historical dust storm records of 753 stations in China from 1954 to 2007 [Zhou et al., 2002]. The temporal information, including the onset and cessation period, for dust storms at each site is collected. However, some sites have too few records, which will decrease the credibility of statistics; therefore, we apply a filter before analysis. Only 398 stations that have at least three dust storm records are chosen. The distribution of these qualified sites is shown in Figure 17a.
 Generally speaking, these chosen sites are distributed over North China, because most of the dust comes from Taklimakan and Mongolia. At each qualified station, we sort each dust event between 1954 and 2007 by the onset hour. These times are denoted by Beijing standard time, which is 8 h earlier than Greenwich Mean Time (GMT). We convert the times into local hours by considering the longitudinal difference between Beijing and each site. Finally, we sum the number of all the dust events in every local hour from each station. The result is shown in Figure 17b. This gives an indication of the most probable hour for the occurrence of a dust event.
Figure 17b reveals that most of the dust storms happen during the day, usually in the local afternoon. This feature is consistent with the simulation of the GMOD, as shown in Figure 16. In addition, both observations and simulations show double peaks in the daily cycle. One occurs at noon and the other is in the late afternoon. The first extreme may be caused by the evolution of meteorological fields that are helpful for dust mobilization, and the second extreme may be caused by the development of turbulent energy, as discussed before.
6. Summary and Discussion
 In this paper, a new global dust transport model, the GMOD, is developed and evaluated. The model uses emission and deposition parameterization schemes similar to Wang et al. . It is embedded in a general circulation model, the IAP9L-AGCM, and inherits the stable features of this climate model.
 The total uplift in a 20-year simulation is 1935 ± 51 Tg a−1, in which 1305 ± 35 Tg a−1 are dry deposited and 629 ± 17 Tg a−1 are wet deposited. The simulated dust burden is 27.8 Tg for particles smaller than 10 μm. The average lifetime for dust suspended in the atmosphere is about 5.2 days, which is similar to the results of Tegen and Fung  and Zender et al. , but smaller than the 7.1 days predicted by Ginoux et al. . The continental and oceanic dust budgets are also reasonable when compared with both observations and other models. The simulated annual global mean DOT at 0.67 μm is about 0.032, with the maximum occurring in the boreal summer and the minimum in the boreal winter.
 The simulated dust climatology is validated with both station-based and satellite observations. The GMOD reproduces observed dust concentrations (MB of −0.67 μg m−3, NMB of −8.0%, correlation coefficient of 0.96 at 18 sites), logarithmic total deposition (−0.62 g m−2 a−1, −36.0%, 0.84 at 251 sites), and aerosol optical thickness (−0.04, −26.7%, 0.80 at 16 sites) reasonably well. The simulated dust particle size distributions have a volume median radius in the range of 1.0–4.0 μm, similar to the observations. The simulated dust burden is also highly correlated with the TOMS AAI on the seasonal scale. The main deficiency of the model is an overestimation of dust concentration in the central Pacific and an underestimation of the dust activity in North and South America.
 Based on this validation, the temporal characteristics of dust transport are further explored. The interannual variability of the GMOD is smaller than that driven by reanalyses or assimilation data [Zender et al., 2003; Ginoux et al., 2001; Marticorena and Bergametti, 1996]. However, the seasonal variability is much larger. The maximum and minimum of dust concentration in the NH usually appear in the boreal summer and winter, respectively. However, the dust budget in the SH shows opposite seasonality, which peaks in the boreal winter. The seasonal variation of dust concentration in east Asia fails to capture the springtime maximum.
 The diurnal variation of dust mobilization is very important. Simulated dust uplift is more active in the local daytime due to the diurnal alteration of surface meteorological fields, such as relative humidity and wind friction velocity. The increase in turbulent energy in the afternoon helps increase the dust burden in the late afternoon. Wet deposition tends to be larger at midnight, when rainfall is more likely to occur. The magnitude of the daily cycle is different among each component of the dust budget. The emission shows the largest diurnal variability; the maximum global dust uplift is about 75 times the minimum global dust uplift. The dust burden shows the least variation. However, given the regional distribution of dust particles, the diurnal cycle of local dust loading is important.
 Though the GMOD provides a reasonable simulation of the dust transport around the world, several factors constrain its performance. As a model driven by a GCM, the GMOD is greatly influenced by the deviation of the meteorological fields from observations in the climate model. Luo et al.  performed a sensitivity study of the meteorological parameters on the dust simulation. They used both NCEP/NCAR reanalyses and National Aeronautics and Space Administration Data Assimilation Office (NASA DAO) reanalyses to drive the same dust model and obtained some important differences between the two results. From this point of view, the deviation in the meteorological fields may partly limit the effectiveness of the GMOD.
 Another possible influence originates from the uncertainties in the domain and the strength of dust sources, which is a common problem faced by other models [Kinne et al., 2003; Uno et al., 2006]. Dust sources in the GMOD are generally determined by the vegetation types of the GCM, which are not linearly related to the emission potential of soils. In addition, the GCM seems to omit the influence of agriculture on the alteration of plants and thus may underestimate dust sources in east Asia. From this point of view, a more reasonable and accurate distribution of the ratio factor of the potential emission coefficient C′1 may help enhance the model's performance.
 On the other hand, the GMOD has advantages for some fields. Sokolik et al.  concluded that the incorporation of regionally and temporally varying dust mineralogical composition into the GCM helps to decrease the large uncertainties in the assessments of radiative forcing by the natural and anthropogenic components of airborne mineral aerosols. In addition, dust can redistribute the energy in the atmosphere by reflecting, scattering, and absorbing radiation [Miller and Tegen, 1998]. These effects will, in turn, surely change the general circulation and affect dust emission and transport. Since the GMOD is embedded in the GCM, the climatic impact of mineral dust aerosols can be analyzed in a temporally varying and interactive way. These interactions will be the focus in part 2 of this paper (X. Yue et al., Simulation of dust aerosol radiative feedback using the GMOD: 2. Dust-climate interactions, submitted to Journal of Geophysical Research, 2009).
 We are grateful to Hong Liao for her helpful suggestions in debugging the model and writing the manuscript. June Yeung carefully proofread the paper and improved the language. Gan Luo kindly offered codes for programming. Joseph M. Prospero generously provided the observed dust concentration data from the University of Miami Ocean Aerosol Network. Zijiang Zhou kindly offered the historical dust storm records of 753 stations in China. Two anonymous reviewers offered helpful comments that greatly improved this article. This research was jointly supported by the National Key Program for Basic Research (“973” program) under grant 2006CB403705, by the Key Project of Chinese Academy of Sciences under grants KZCX2-YW-Q1-02 and KZCX2-YW-Q11-03, and by the National Science Foundation of China under grants 40631005 and 40620130113.