3.2. Dust Emission in General Circulation Models
 The dust emission scheme for a global model requires (1) a dust source function to represent the location and relative erodibility of dust sources and (2) a parameterization of the mobilization process. Satellite observations are useful for detecting dust source regions. The dust source function used in this study is from Ginoux et al.  and Prospero et al. . Global bare soil regions are identified as potential dust sources from a 1° × 1° vegetation data set constructed from the advanced very high resolution radiometer (AVHRR) [DeFries and Townshend, 1994]. For each grid box, the efficiency for emitting dust is parameterized in terms of its local topography relative to surrounding grid boxes. That is, grid boxes that are in relative topographic depressions are assumed to have preferentially collected erodible sediments, and so are stronger dust sources than topographically elevated grid boxes. This approach has shown good consistency between the resulting global dust source function map (Figure 1) and dust aerosol locations observed with the Total Ozone Mapping Spectrometer (TOMS) aerosol index product [Ginoux et al., 2001].
 For this study, we consider two different in situ parameterizations of the dust mobilization process applied to our model grid. The first is based on the current GOCART scheme, from Ginoux et al.  (hereinafter referred to as the GOCART scheme). As an alternative, we have implemented a version of the DEAD scheme from Zender et al. . Both schemes parameterize dust emission in terms of the surface wind speed and distribute the emitted aerosol over a size distribution discretized by several size bins. Wind tunnel experiments have found that the horizontal flux of saltating soil particles is proportional to a power of the surface friction speed [Ivesen and White, 1982]. Marticorena and Bergametti  developed a semiempirical parameterization for this relationship, accounting for the confounding effects of soil moisture and vegetative cover. Both the GOCART and DEAD schemes use this parameterization of the dry threshold wind speed, but they diverge at this point.
 In the GOCART scheme, the dry threshold wind speed is computed as a function of aerosol particle size according to the size bins chosen. This threshold is then modified to account for soil moisture, following Ginoux et al. . The emissions are then computed in terms of the 10 m wind speed so that emission occurs for each size bin only when the 10 m wind speed exceeds the threshold wind speed as determined in Marticorena and Bergametti . The equation for emissions is thus
where F(r) is the mass flux of aerosol emitted into a size bin of radius r, C is a tuning constant in units of kg s2 m−5 used to set global dust emissions to a desired value, S is the spatially dust source function shown in Figure 1, s(r) represents the efficiency of the soil at emitting particles of size r, U10 m is the 10 m wind speed, and Ut is the size-dependent threshold wind speed from Marticorena and Bergametti  that has been modified for the presence of soil moisture w [after Ginoux et al., 2001].
 By contrast, the DEAD scheme connects the threshold wind speed to the initiation of saltation rather than direct aerosol injection. Sandblasting caused by saltation is the main dust entrainment mechanism for sustained emission [Shao and Raupach, 1993] and makes the emission physics of the DEAD scheme more satisfying. Unfortunately, determining soil grain saltation requires knowledge of the particle size distribution of the parent soil bed, which is not well known at global scales. For our implementation of DEAD, we follow Zender et al.  by assuming that the parent soil contains a fixed monomodal soil particle size distribution of optimally sized particles (D = 100 μm, ut* = 0.209 m s−1) and compute the horizontal saltation flux of those. Therefore, the threshold formulation from Marticorena and Bergametti  is used to determine the initiation of soil particle saltation as a function of surface properties and wind friction speed. The threshold is increased for soil moisture following Fécan et al. , as well as to account for the loss of atmospheric momentum to nonerodible objects within the soil (e.g., rocks, vegetation), where we assume a fixed drag efficiency for all model grid cells, following Marticorena and Bergametti . The surface friction speed u* is increased to account for the transfer of momentum to the surface from saltating particles, known as the Owen effect [Gillette et al., 1998]. The aerosol mass injected is proportional to the horizontal saltation flux, which is computed in terms of the threshold wind speed and the wind friction speed (not the 10 m wind speed, as in the Ginoux scheme):
where again F(r)′ is the mass flux of aerosol into a size bin of radius r, C′ is a global tuning constant that also incorporates the efficiency with which the horizontal saltation flux translates to vertical aerosol mass flux, S is the same dust source function used in equation (1), s(r)′ is the aerosol particle size distribution, u* is the surface friction speed from the land model that has been modified for soil moisture w.
 Comparing equations (1) and (2), we see that the dust emission flux in both schemes is approximately proportional to the third power of the wind speed for wind speeds exceeding some threshold. In the GOCART scheme, the relevant wind speed is the 10 m wind speed U10 m, while in the DEAD scheme the relevant wind speed is the surface friction speed u*. We note that both schemes use the same threshold speed parameterization of Marticorena and Bergametti , but apply it differently. The formulation in the GOCART scheme implies that Ut = ut*, with modifications for soil moisture content. Although this is not strictly correct, it captures the observed behavior that higher surface wind speeds are needed to mobilize smaller aerosol particles. The parameterization in DEAD is more physically satisfying in that it explicitly accounts for saltation and sandblasting, but it is itself a simplification in that it neglects variability in soil particle size distributions, distributions of erodible surfaces within grid cells, and differences in the efficiency of horizontal-to-vertical mass flux transfer that depend on soil type. Grini and Zender  modified DEAD to evaluate the effects of sub-grid-scale winds and different soil bed particle-size distributions, showing that these modifications affect simulated dust mass concentrations, optical depths, and the fraction of coarse particles, but not the timing of dust events. Unfortunately, the global variability of these properties is poorly known, particularly over the Saharan source region.
 For both emission schemes, we distribute the emitted aerosol mass across five transported size bins. On the basis of Tegen and Lacis  and Ginoux et al. , we choose the following radius bins: 0.1–1, 1–1.8, 1.8–3, 3–6, and 6–10 μm. The sub-bin particle size distribution of each bin follows from Tegen and Lacis  in that we assume dMass/(dln r) = constant. This determines an effective radius for each bin which is used in our emission and settling calculations: 0.73, 1.4, 2.4, 4.5, and 8 μm. Additionally, the first (smallest) size bin is further divided into four sub-bins for the purposes of optics, following Tegen and Lacis . For the GOCART scheme, the mass emitted to each bin is computed independently, based on how the wind speed exceeds the threshold for that bin. The soil particle size distribution enters as s(r) as in equation (1) (following Tegen and Fung ), where the mass of emitted clay particles (0.1 < r < 1 μm) is assumed to be one tenth of the total mass of emitted silt (particles of radius > 1 μm), that is s = 0.1 for the smallest bin. The four silt bins (1–1.8, 1.8–3, 3–6, and 6–10 μm) are each assigned a mass fraction of s = 0.25. We note that the emitted particle size distribution is dynamically determined in the GOCART scheme in that the threshold is computed for each size bin independently. In contrast, the DEAD scheme imposes a fixed trimodal lognormal distribution on the emitted aerosol that is based on the observed background dust particle size distributions near Saharan dust sources [D'Almeida, 1987].
 For both emission schemes, dust loss processes are parameterized in GEOS-4 following Ginoux et al.  and Chin et al. . Dry deposition of dust particles by gravitational settling and turbulent mix-out is the dominant loss process, while wet deposition and convective scavenging become more significant after larger particles have fallen out during transport. Dust optical properties used in GOCART follow from the Global Aerosol Data Set [Köpke et al., 1997].
 Table 1 summarizes the GOCART and DEAD emission schemes. Aside from the difference in winds used to parameterize the emission process, the major difference in the two schemes is that the emitted particle size distribution is fixed in the DEAD scheme, whereas it is dynamically generated in the GOCART scheme depending on the difference between the surface wind speed and the size-dependent threshold wind speed. However, because of how the threshold wind speed is applied in the GOCART scheme, the threshold speeds are generally much smaller than the 10 m wind speed (Ut ≪ U10 m), so that in practice there is little dynamical variation in the emitted size distribution. Both schemes have a drawback in that they are both in situ parameterizations that have been applied to a model grid. We use box-averaged parameters (i.e., wind speed and soil moisture) to represent the microscale processes that modulate dust emissions and cannot account for subgrid variability. Additionally, because both emission schemes have been applied to our model grid, global tuning constants are necessary to set the total global emissions. For these simulations, GOCART emissions have been tuned to match the mass budget of emissions from Ginoux et al. , which were shown to produce reasonable aerosol optical thickness (AOT) values as described by P. Colarco et al. (submitted manuscript, 2009). DEAD emissions were scaled so that the resultant regionally averaged AOT over North Africa was the same as the regionally averaged GOCART AOT over North Africa during the NAMMA period (August–September 2006).
Table 1. Emission Scheme Comparison
|Source function||bare topographical depressions [Ginoux et al., 2001]||bare topographical depressions [Ginoux et al., 2001]|
|Dry emission threshold speed||wind tunnel experiments [Marticorena and Bergametti, 1995]||wind tunnel experiments [Marticorena and Bergametti, 1995]|
|Threshold speed modifications||soil moisture content [Ginoux et al., 2001]||soil moisture content [Fécan et al., 1999] and nonerodible objects [Marticorena and Bergametti, 1995]|
|Wind parameter used to determine emitted mass flux||10 m wind speed||surface friction speed|
|Flux equation [F(r), F(r)′]||F(r) = C * s(r) * * (u10 m − ut (r, w))||F(r)′ = C′ * S * s(r)′ * u*3 * |
|Bin dependent mass fractions||s(r) [Tegen and Fung, 1994]||s(r)′ [D'Almeida, 1987]|
| reff = 0.73 μm (0.1–1 μm)||0.1||0.112|
| reff = 1.4 μm (1–1.8 μm)||0.25||0.232|
| reff = 2.4 μm (1.8–3 μm)||0.25||0.296|
| reff = 4.5 μm (3–6 μm)||0.25||0.277|
| reff = 8 μm (6–10 μm)||0.25||0.064|
|Constants [C, C′]||C = 0.375 × 10−9 kg s2 m−5||C′ = C″ * α = 1.780 × 10−5 kg s2 m−5, where C″ = 3.716 × 4 kg s2 m−5 and α = 0.0479 is the sandblasting mass efficiency after assuming a globally uniform mass fraction of clay particles of 0.2|