Journal of Geophysical Research: Atmospheres

Simulation of entrainment and transport of dust particles within North America in April 2001 (“Red Dust Episode”)



[1] A size-resolved, multicomponent, regional-scale particulate-matter (PM) model named AURAMS (A Unified Regional Air-quality Modelling System) has been used to study the entrainment and transport of dust from the southwestern United States and northwestern Mexico to eastern North America during the so-called “Red Dust Episode” in April, 2001. Two different wind-blown-dust emission schemes, the Marticorena-Bergametti-Alfaro (MBA) scheme and the Shao scheme, were incorporated into AURAMS to simulate dust generation, and sensitivity analyses for various dust-emission-scheme parameters were performed. Comparison of the model results with satellite observations and surface measurements showed that the model simulation reasonably reproduced the temporal and spatial distribution of wind-blown dust particles during the episode period in the downwind area of Oklahoma but not in the source region. Both dust-emission schemes captured the main features of the dust transport. The dust-emission-scheme parameter most responsible for inaccurate prediction of wind-blown-dust emission in the source region in this study appeared to be soil moisture content. The soil grain size distribution and the soil plastic pressure were also shown to be important parameters that should be accurately estimated for better model performance. For further validation and reliable use of wind-blown-dust emission schemes, accurate field and remote sensing measurements of those parameters are imperative. The unusually fast transport of dust during the episode appeared to be due to vigorous vertical mixing and uplift of emitted dust. Appropriate parameterization of additional vertical mixing by sub-grid-scale convection may help to better predict the long-range transport of dust storms.

1. Introduction

[2] Soil dust produced by wind erosion in arid areas is one of the major components of natural atmospheric aerosols contributing to global aerosol mass load and aerosol optical thickness. Soil dust particles can impact climate either directly by altering the radiative balance in the atmosphere due to their high absorption coefficient [Haywood and Boucher, 2000; Sokolik and Toon, 1996; Tegen and Fung, 1994] or indirectly by modifying cloud nucleation and optical properties [Levin et al., 1996] and by catalyzing heterogeneous reactions [Dentener et al., 1996]. Dust storms in major desert areas can also significantly affect regional air quality. During such storms, fine dust particles can be lifted up to high altitudes due to high wind speed and intense turbulence and then transported far from source regions, even across oceans, affecting the air quality in regions far from the source area [Jaffe et al., 1999; Mahowald et al., 2002].

[3] Due to the climatic and environmental importance of soil dust, considerable efforts have been devoted to simulating the production and transport of soil dust aerosols at regional and global scales [Ginoux et al., 2001; Gong et al., 2003b; Liu et al., 2003; Zender et al., 2003]. A challenge in soil dust aerosol modeling is to parameterize accurately the emission rate of dust particles as a function of particle size from the aeolian processes involved in wind erosion.

[4] There are two major dust-emission schemes now available, i.e., the MBA scheme [Alfaro and Gomes, 2001; Marticorena and Bergametti, 1995] and the Shao scheme [Shao, 2001]. A detailed comparison of these two schemes was recently performed by incorporating them into a regional climate model and then simulating East Asian dust storms that occurred in March 2002 [Zhao et al., 2006]. Values of physical parameters needed by the two schemes were tuned using comparison with measurements. Although both schemes can predict dust emissions as a function of particle size, only total dust mass emission was calculated by the schemes and the size distribution was then adjusted to match the measured one. While both schemes captured the dust mobilization episodes after appropriate tuning for the physical parameters, they produced considerably different dust emissions and near-surface dust concentrations depending on soil moisture and vegetation coverage. With the limited conditions used in the study, it is hard to generalize the performance of the schemes. For a more thorough assessment, tests with different transport models at different geological locations are needed.

[5] In North America, the most important desert soil dust emission source region is the southern Great Plains area of northwest Texas and eastern New Mexico [Orgill and Sehmel, 1976; Pye, 1987] as well as the Chihuahuan Desert of northwestern Mexico. On 6 April 2001, intense dust storms developed in that area. The dust was then quickly transported over 2000 km northeastward to the northeastern United States and southeastern Canada, where it was collected the next day in a number of precipitation-chemistry samplers in Ontario, Quebec and Pennsylvania. Mineralogical and isotopic analysis of the dust, as well as ionic analysis of the precipitation samples, indicated that the dust was of similar mineralogical composition at most of the monitoring sites in Ontario [Vet et al., 2003]. Air mass trajectories, satellite images, and output from a three-dimensional transport-diffusion model that is used operationally for volcanic plume forecasting clearly pointed to the above mentioned source area as the source of the wet-deposited dust [D'Amours, 1998; Vet et al., 2003]. This particular dust-transport event was unusual both for the large amount of dust that was transported and for the great distance that it was transported, effectively across the North American continent [Doggett IV et al., 2002]. We have called this event the “Red Dust Episode” in recognition of the reddish to yellowish dust that was detected in the precipitation-chemistry samplers.

[6] As dust storms originating from local sources can cause significant air pollution in a region, numerical modeling is a good tool for predicting the development of dust storms and their impact on regional air quality. In this paper, a size-resolved, multicomponent, regional-scale particulate-matter (PM) model named AURAMS (A Unified Regional Air-quality Modelling System) is used to study the entrainment and transport of dust from the southwestern United States and the Chihuahuan Desert of northwestern Mexico to eastern North America during the Red Dust Episode period. The present study was performed to address the following questions:

[7] (1) Can a state-of-the-art regional air quality model accurately predict the development of dust storms and the subsequent long-range transport of dust if it is combined with available wind-blown-dust emission schemes?

[8] (2) Are the values of the physical parameters used in the dust-emission schemes universal? For example, are those parameters determined in our previous Chinese dust storm studies [Gong et al., 2003b; Zhao et al., 2006] also applicable to North American dust storms?

[9] (3) How do the MBA and Shao wind-blown-dust emission schemes compare in predicting dust storms from North American soils? Is one superior to the other?

[10] (4) What laboratory, satellite, and/or field measurements are needed to improve the model prediction of wind-blown soil dust emissions?

2. Model Descriptions

2.1. AURAMS Regional AQ Model

[11] AURAMS is a size- and composition-resolved PM modeling system developed by Environment Canada (EC). It consists of three primary components: an emissions processor; a meteorological driver model; and a chemical transport model (CTM). AURAMS was developed to provide a better understanding of PM and other regional pollutants in North America [Gong et al., 2006; McKeen et al., 2005].

[12] The AURAMS meteorological driver is the EC operational weather forecast model, GEM (Global Environmental Multiscale model) [Côté et al., 1998a; Côté et al., 1998b]. The regional configuration of GEM version 3.2.0 that was used for this study employs a horizontal grid size of 24 km on a rotated latitude-longitude grid. Meteorological fields required by the AURAMS chemical transport model are stored at every CTM advection time step, which is set at 900 s. The grid selected for this study for the AURAMS CTM was the 42-km horizontal grid on a secant polar stereographic projection true at 60°N that is shown in Figure 1. The 28 unevenly spaced vertical levels ranged from the surface to 25 km AGL, with 14 levels located in the first 2 km.

Figure 1.

The AURAMS CTM domain together with locations of the nine stations whose measurements of surface PM10 concentrations or aerosol optical depth have been compared with model results: (1) Albuquerque, NM (106.55°W, 35.11°N); (2) El Paso, TX (106.50°W, 31.77°N); (3) Ponca City, OK (97.07°W, 36.66°N); (4) Tahlequah, OK (94.99°W, 35.86°N); (5) Miami, OK (94.84°W, 36.92°N); (6) Blue Island, IL (87.68°W, 41.66°N); (7) Toronto, ON (79.40°W, 43.66°N); (8) Sevilleta, NM (106.89°W, 34.36°N); and (9) Konza, KS (96.61°W, 39.10°N).

[13] To track the time evolution of the size distribution and the chemical composition of atmospheric aerosol particles including dust aerosol, the Canadian Aerosol Module (CAM), a composition- and size-resolved aerosol module [Gong et al., 2003a], was incorporated into AURAMS. All major atmospheric aerosol processes, including generation, growth, coagulation, transport, and dry/wet removal, are included in CAM. A sectional method is employed in CAM to represent the aerosol size distribution. Twelve size sections with a constant section-spacing factor on a logarithmic scale between 0.01 and 40.96 μm particle diameter are used, and each section is assumed to be internally mixed. Nine aerosol chemical components are considered: SO4=, NO3, NH4+, elemental carbon, primary organic carbon, secondary organic carbon, crustal material, and particle-bound water.

[14] The simulations presented in this paper cover the 16-day period from 0600 GMT, 1 April to 0600 GMT, 17 April 2001. GEM was run in 30-h segments beginning at 0000 GMT on each day and starting from objectively analyzed fields produced by the Canadian Meteorological Centre of EC. The first six hours of each GEM simulation were treated as spin-up time and only fields for the remaining 24 h were supplied to the AURAMS CTM. The first two days of the AURAMS CTM run (i.e., 1 and 2 April) was used as a spin-up period to allow emissions and atmospheric concentrations to attain a quasi-equilibrium.

2.2. Wind-Blown-Dust Emission Schemes

2.2.1. MBA Scheme

[15] The MBA scheme is a size-segregated wind-blown-dust emission scheme based on a threshold friction velocity [Alfaro and Gomes, 2001] that was developed by combining several previous studies [Alfaro et al., 1997; Alfaro et al., 1998; Marticorena and Bergametti, 1995; Marticorena et al., 1997; White, 1979]. The effects of non-erodible elements [Marticorena and Bergametti, 1995] and soil moisture content [Fécan et al., 1999] are both incorporated. The most critical parameters for determining dust particle emission rates are friction velocity, soil-grain size distribution, surface roughness length, soil moisture content, and the emitted dust size distribution parameters. Of these parameters, the friction velocity and the soil moisture content are obtained from the meteorological driver (e.g., GEM) whereas the others are input parameters that have to be provided by other means. One can refer to Gong et al. [2003b] for the detailed formulation of this scheme.

[16] There are 12 soil-texture classes associated with the percentage ranges of three primary soil modes with the following diameter size ranges: (1) sand: 1–0.05 mm; (2) silt: 0.05–0.001 mm; and (3) clay: <0.001 mm [Cosby et al., 1984; Hseung, 1984]. A global data set of soil texture for these 12 classes at 1/30° latitude by 1/30° longitude resolution based on Zobler's assessment of FAO Soil Units [Zobler, 1986] was used to create soil-texture fields on the AURAMS CTM grid. The fraction of each soil-texture class was calculated for each AURAMS grid cell by aggregating from the original soil texture data set. For each soil mode of a soil-texture class, the parameters (geometric mean diameter, geometric standard deviation, and relative contribution) for a lognormal distribution of soil grain size in this study were then assumed from the previous modeling of Chinese dust storms [Gong et al., 2003b], with the expectation that these parameters would also be reasonably representative of soil types in North America.

[17] The 1 × 1 km2 USGS (US Geological Survey) Global Land Cover Characteristics Data has been adopted in CAM for dry deposition calculations with appropriate regrouping from the 232 detailed land-cover classes in the USGS data set to 15 broader land-cover classes [Zhang et al., 2001]. One can refer to Gong et al. [2003b] for detailed information on these 15 land-cover classes and the corresponding representative physical heights of roughness elements (h) used in CAM. The roughness length of non-erodible roughness elements, Z0, is linearly related to the physical height as follows [Marticorena, 2006, personal communication]:

equation image

where λ is the roughness density defined by Marticorena et al. [1997]. Since there are no data on roughness density available for North America, two possible approximations were tested: (1) λ always larger than 0.045; and (2) λ equal to the fractional coverage of the corresponding land-cover class in a grid cell. Comparison of the predicted dust emission using the above two different approximations showed that the difference caused by these different definitions for roughness density is negligible (<1%). This is because the main dust source regions in our wind-blown-dust emission scheme are desert areas, where the fractional coverage of the desert land-cover class is usually much larger than 0.045. In desert areas, the surface might be more or less homogeneous and the roughness density should be relatively high. Therefore in this study λ ≥ 0.045 is assumed throughout the spatial domain, which can be regarded as a lower limit for dust production.

[18] For desert regions, more detailed information on the roughness length is needed to calculate the rate of dust particle entrainment accurately. It is known that for smooth soils the soil roughness length can be approximated by the quantity Ds,max/30, where Ds,max is the geometric mean soil-grain diameter of the largest aggregate population [Greeley and Iversen, 1985]. This approximation was used throughout this study since no field measurements for soil roughness length were available.

2.2.2. Shao Scheme

[19] The Shao scheme adapted different parameterizations of the threshold friction velocity and the sandblasting mechanism while retaining the same saltation model. In contrast to the binding-energy-based parameterization of the MBA scheme, the Shao scheme estimates the dust emission rate from the volume of soil removed by saltating soil grains as they plough into the surface and disintegration of saltating soil grains themselves [Lu and Shao, 1999; Shao, 2001]. The vertical dust emission rate is assumed proportional to the horizontal saltation flux, with the proportionality depending on the soil plastic pressure, which represents the resistance of surface soil to sandblasting. Presenting a simplified dust-emission scheme, Shao [2004] suggested soil plastic pressure values of 1000 to 5000 Pa for sandy soils and 30,000 to 50,000 Pa for clay soils. By comparing the dust emission rate of the Shao scheme with that of the MBA scheme for dry soils (i.e., with no soil moisture), Zhao et al. [2006] showed that the Shao scheme yields comparable dust emission rates to the MBA scheme when the soil plastic pressure is set to 1000 Pa for sandy soils, 5000 Pa for silty and loamy soils, and 10,000 Pa for clay soils. Considering that these values are at or below the lower limits suggested by Shao [2004], and that the Shao scheme was shown to be more sensitive to soil moisture [Zhao et al., 2006], these lower values for soil plastic pressure were used initially in this study.

3. Preliminary Simulations and Sensitivity Analyses

[20] Preliminary AURAMS runs were made for North America. Figure 2 compares time series of measured surface PM10 concentrations at two different downwind locations in Oklahoma to those predicted by AURAMS using both the MBA and the Shao schemes. All of the measurement data for surface PM concentrations used in this paper were provided by the EC National Atmospheric Chemistry (NAtChem) Database ( It can be seen that for these preliminary AURAMS simulations, the run using the MBA scheme generally underestimated the surface particle concentrations whereas the run using the Shao scheme considerably overestimated them at times. In order to identify the impacts of dust-emission-scheme parameters on the regional model predictions and to improve the model performance, sensitivity analyses were conducted on several critical dust-emission-scheme parameters.

Figure 2.

Comparison of time series of surface PM10 concentrations predicted by AURAMS using both dust-emission schemes with measurements at two downwind sites in Oklahoma. The locations of the two measurement stations are shown in Figure 1. The legend “AIRS (BAM)” means that the observed data were reported to the Aerometric Information Retrieval Systems (AIRS) database of the U.S. Environmental Protection Agency using beta attenuation monitors (BAM).

3.1. Surface Wind Speed

[21] Wind-blown-dust emission takes place only when the surface wind speed exceeds a threshold value. Thus skill in estimating the surface wind speed field is critical for dust emission modeling. In this study, the wind speeds at 10 m above the surface predicted by GEM were used for wind-blown-dust emission calculation. In order to evaluate those 10-m wind speeds, we compared them with hourly field observations obtained at 82 stations throughout Texas for 24 h on 6 April 2001 in local time (Figure 3) by the Texas Commission on Environmental Quality ( The performance statistics of the modeled 10-m wind speed are presented in Table 1. It is shown in Figure 3 and Table 1 that the predicted 10-m wind speeds are in fairly good agreement with the observed wind speeds.

Figure 3.

Comparison of GEM-predicted 10-m wind speeds (m/s) at various locations in Texas on 6 April 2001 with the field measurements provided by the Texas Commission on Environmental Quality.

Table 1. Performance Statistics of GEM-Predicted 10-m Wind Speeda
  • a

    M and O represent modeled and observed values, respectively. Total number of data records was 652.

Mean fractional bias (= equation image)0.18
Mean fractional error (= equation image)0.28
Agreement index (= 1 − equation image)0.72

3.2. Soil-Grain Size Distribution

[22] Soil-grain size governs the emitted dust particle production by affecting the saltation flux and sandblasting efficiency [Alfaro et al., 2004]. Because fine soil grains are more easily entrained into saltation at low wind speeds and because their sandblasting efficiency is much higher than that of coarse grains, fine grains dominate emitted dust production.

[23] In the first AURAMS simulation using the MBA scheme (see Figure 2), it was provisionally assumed that the soil-grain size distribution could be determined by the characteristics of Chinese soils having the same soil textures. It is quite probable that an inappropriate assumption about soil-grain size distribution led to incorrect predictions by AURAMS using the MBA scheme. Indeed, it was reported that vertical dust mass flux measurements performed over a wide variety of sources in the U.S. southwest [Nickling and Gillies, 1989] agreed well with MBA scheme predictions obtained under the assumption that the geometric mean diameter (GMD) of soil grains is 210 μm and the geometric standard deviation is 1.8 [Alfaro et al., 2004].

[24] This unimodal assumption about the soil-grain size distribution was therefore applied to the MBA scheme as a sensitivity test and the results were compared with the first simulation in terms of the size distribution of emitted dust particles produced in the source region. For the Shao scheme, two different kinds of soil-grain size distributions are needed, i.e., the minimally and the fully dispersed size distributions, and observations of the latter are not available in the literature for the source region of this study. Therefore we retained the values for the soil-grain size distribution parameters suggested by Shao [2004] throughout this study. Thus only the results obtained using the MBA scheme are shown here. Figure 4a compares the two assumed normalized soil-grain size distributions. The Chinese soil shows a trimodal distribution with the coarse mode predominant whereas the dashed line represents a unimodal distribution with a GMD of 210 μm. In Figure 4b, emitted dust size distributions predicted with the two assumptions on soil-grain size are compared. For a fair comparison, the size distributions were divided by the same M*, which is the total emitted dust mass for the first case. It is shown in this figure that the new assumption about soil-grain-size distribution considerably enhances the emitted dust production.

Figure 4.

Influence of soil-grain size distribution on emitted dust production from the MBA scheme. (a) Two different assumptions on soil-grain size distribution: (1) soil-grain size is identical to that of Chinese soil with the same soil texture (solid line); (2) soil-grain size distribution for dust production estimation purposes can be regarded as a unimodal lognormal distribution with geometric mean diameter of 210 μm and geometric standard deviation of 1.8 (dashed line). (b) Emitted dust particle size distributions predicted by the MBA scheme with the above mentioned assumptions. M* represents the total mass of dust particles for case (1) and Dp represents emitted dust particle diameter.

3.3. Soil Moisture Content

[25] Another parameter that significantly affects the dust production rate is the soil moisture content (SMC). Water adhering to soil particles increases their mass and surface tension, thereby decreasing suspension and transport. SMC also enhances the strength of surface crusts and the stability of aggregates [Bradford and Grosman, 1982].

[26] SMC is determined in GEM from parameterizations of the land surface processes based on predicted surface heat and moisture budgets. The default land-surface-process scheme currently in GEM is an improved version of the Interactions Soil-Biosphere-Atmosphere (ISBA) scheme originally developed by Noilhan and Planton [1989]. However, the current GEM objective-analysis framework did not become operational until the fall of 2001, so that the analysis that was available to run GEM for April 2001 is not compatible with the ISBA option. Consequently, the older and simpler Force-Restore option was chosen for the GEM land-surface-process scheme. In this scheme SMC is not obtained by solving predictive equations but is instead kept constant during the integration for one day at its initial specified values based on statistical interpolation using air temperature and humidity.

[27] The GEM-estimated SMC values were compared with field measurement data from Texas, New Mexico, and Arizona that were provided by the Soil Climate Analysis Network (SCAN; The comparison results are shown in Figure 5a. It is clear from this figure that the GEM Force-Restore scheme tends to underestimate SMC significantly (by a factor of 2.345 on average) around the source area. To evaluate the effect of SMC on emitted dust particle production, simulations were carried out assuming a Chinese soil-grain size distribution for two different SMC values: (1) original GEM estimated values; and (2) values enhanced by a factor of 2.345 with both the MBA and Shao schemes. Results from the two schemes are compared in Figure 5b. The SMC effect is roughly a factor of 10 for both schemes and is larger for the Shao scheme as has been previously shown [Zhao et al., 2006].

Figure 5.

The effect of soil moisture fraction: (a) comparison of daily average SMC values estimated by GEM with measured values (5 cm depth) around the source area for the period 3–16 April 2001; (b) effect of SMC adjustment on size distribution of produced dust particles for both MBA and Shao schemes. M* represents the total mass of dust particles for the first case (GEM parameterization) and Dp represents emitted dust particle diameter.

3.4. Dust Emission Depth

[28] In this study, the wind-blown-dust emission module was incorporated separately into the CAM component of AURAMS that is processed after the turbulent vertical diffusion, whereas the emissions of dust (and other pollutants) from anthropogenic sources included in the U.S. and Canadian emission inventories are accounted for at the beginning of each model time step. Newly emitted dust particles are usually injected into the AURAMS CTM bottom layer and stay in that layer until they are transported upwards to an overlying layer or are removed by various mechanisms. One implication of this difference is that the wind-blown-dust emission is not immediately followed by vertical diffusion, which is usually vigorous during a dust emission event, and thus newly emitted dust particles are confined to the bottom layer when the aerosol dry deposition calculation is performed. This will likely cause some underestimation of the total dust loading transported to downwind regions.

[29] To investigate the impact of the placement of the vertical diffusion operator, the method used to inject the emitted dust particles was varied; that is, the emitted dust particles were injected (1) only into the bottom layer (1-layer injection), whose top is at 10 m, and (2) into the lowest three model layers (3-layer injection), whose top is at 90 m. The 3-layer injection was chosen because the typical height of the surface layer is between 50 and 300 m [Jacobson, 2005]. A comparison of AURAMS runs using the two injection methods showed that higher surface dust concentrations and vertical dust loadings (by up to a factor of 2) were predicted for 3-layer injection, except right at the source grid cells, where the predicted surface dust concentration was lower right after the dust emission events. This indicates that 1-layer injection can significantly underestimate the amount of dust transported from the source region due to overestimation of dry deposition.

4. Results and Discussion

[30] On the basis of the two preliminary AURAMS simulations and the additional five sensitivity-analysis simulations, two parameters in AURAMS were revised. To reduce the overestimation of the Shao scheme (see Figure 2), the soil plastic pressure values were adjusted back to the upper limit values suggested by Shao [2004], i.e., 5000 Pa for sandy soils, 20,000 Pa for silty and loamy soils, and 50,000 Pa for clay soils. The 3-layer injection method was adapted for further simulations to avoid the overestimation of dry deposition of wind-blown dust. Chinese soil-grain-size distribution (MBA scheme) and unmodified GEM SMC fields were retained at this stage. The effects of changes in these two parameters are discussed later.

[31] In this section, AURAMS model results based on revised dust-emission scheme parameters are presented and the production, transport and distribution of dust particles during the episode period are discussed. As the period modeled was not very long (14 days) and the spatial domain of interest was relatively small (regional scale), only hourly or daily measured data were used in this study for model evaluation.

4.1. Vertical Soil Dust Flux and Soil Moisture Content

[32] During the simulation period of this study, two distinct dust production periods were predicted by the model; 1500 GMT, 6 April–0300 GMT, 8 April (S1) and 1800 GMT, 10 April–0000 GMT, 12 April (S2). Except for these two periods, little dust production was predicted. Figure 6 shows the average vertical soil dust fluxes for these two dust-production periods as predicted by the MBA and Shao schemes. Because the two schemes use different parameterizations in the calculation of threshold friction velocity, vertical dust emission flux, and soil-related factors, they showed different soil-dust-flux spatial distributions. Nevertheless, it is clear that both schemes predict that dust particles are mainly produced in the southwestern U.S. region. This region, including Texas, New Mexico, Colorado, and Arizona, will be referred in this paper as the “source region” because most source areas are located in these four states and northern Mexico during the Red Dust Episode period. S1 dust emissions occurred near the border between the U.S. and Mexico as well as in New Mexico and Arizona while New Mexico and Colorado appeared the main source regions for S2. Figure 6 also shows the SMC for the same periods with a significant change in SMC between the S1 and S2 periods. High emitted dust flux is closely correlated to low SMC values. The wind speed did not show high correlation with dust flux (not shown), demonstrating that the dust storms in the Red Dust Episode period were SMC-limited rather than wind speed-limited. Recalling that SMC estimation was a critical weakness of the present study, the strong dependence of dust production on soil moisture could be the most important reason for model underperformance as will be discussed later.

Figure 6.

Spatial distributions of the average vertical soil dust flux (g/km2/s) predicted by (a) MBA and (b) Shao schemes compared with (c) SMC distribution, for the S1 (1500 GMT, 6 April–0300 GMT, 8 April) and S2 (1800 GMT, 10 April–0000 GMT, 12 April) periods.

4.2. Dust Column Loading

[33] Figure 7 compares the 3-h-average dust column loading (DCL) predicted by AURAMS with the TOMS aerosol index (AI) [Herman et al., 1997; Torres et al., 1998], where DCL is calculated over the full model vertical column. To a first approximation, the TOMS AI can be considered to be proportional to the integrated concentration of UV-absorbing aerosols such as dust [Hsu et al., 1996]. All of the TOMS AI data were measured at around 1900 GMT each day whereas the modeled DCLs are average values between 1800 GMT and 2100 GMT. A white background in the AI images means that there were no valid data on those spots. One can see in Figure 7 that both the TOMS AI and the AURAMS DCL fields show a thick aerosol layer in the source region on 7 April and 11 April during the S1 and S2 periods, respectively, although the location of the dust loading predicted by the model was not matched precisely by TOMS AI, which indicated that the most intense dust emission source was around the border of Texas, New Mexico, and Mexico. Both dust schemes predicted relatively weak dust production in that area (especially for the S2 period) and estimated significant dust emissions in eastern New Mexico and Colorado. It has been reported that the area around the border of Texas and Mexico was in the middle of a severe drought during the period (, and hence SMC values were very low. The GEM-estimated SMC, however, significantly increased in this area between the S1 and the S2 periods (see Figure 6), preventing the model from correctly predicting the dust emission during the S2 period. Therefore the lack of accurate information for SMC is likely a major reason for the mismatch, while uncertainty in soil-grain-size distribution may be another contributing factor.

Figure 7.

Comparison of model-predicted 3-h-average dust column loading (mg/m2) with the aerosol index measured by the TOMS satellite on 7 April and 11 April 2001.

[34] A comparison of model-predicted DCL with observations from another satellite (Geostationary Operational Environmental Satellite-10; GOES-10) was prepared. GOES-10 images showing the 11-μm channel (T4) and 12-μm channel (T5) brightness temperature difference (ΔT = T4T5) are available every three hours for the southwestern region of the U.S. for 6 April to 7 April, which includes the S1 period. One of these images is compared with model-predicted DCL fields in Figure 8. Both dust-emission schemes correctly predicted the dust production near the border of the U.S. and Mexico observed during the S1 period. It is also evident from this figure that the dust production in eastern Arizona and western New Mexico during S1 that was predicted by the MBA scheme was not observed by GOES-10. It did not support a better prediction by the Shao scheme over the MBA scheme, however, because a snapshot taken by another satellite, GOES-8, using a visible wavelength ( at that time showed scattered clouds in the above mentioned area (not shown). The presence of meteorological clouds (positive ΔT) prevents the detection of dust (negative ΔT). Because of this limitation, the T4T5 difference is likely to underestimate the actual area covered by airborne dust. In fact, the AI field shown in Figure 7 indicates that there was a significant amount of dust in this area. Therefore it is not clear which scheme performs better based only on the above comparison. This issue is visited again in the next subsection with regard to surface PM10 concentration.

Figure 8.

Comparison of model-predicted dust column loading (mg/m2) for both the MBA and Shao schemes with the (CH4–CH5) values measured by GOES-10 satellite during the S1 period.

[35] In Figure 9, time series of model-predicted DCL are compared with time series of aerosol optical depth (AOD) measurements at two AERONET ( network stations. It is noted that the agreement between model predictions and measurements was better at the downwind station (Konza, Kansas) than at the near-source station (Sevilleta, New Mexico). The large discrepancy between model predictions and measurements around the source region is discussed in the next section more in detail.

Figure 9.

Comparison of time series of model-predicted dust column loading (DCL) with AERONET aerosol optical depth (500 nm) measurements for the 1–16 April 2001 period at two stations: Sevilleta, NM (left) and Konza, KS (right). All measured data and model results are daily mean values averaged from 0000 GMT to 2400 GMT on each day.

4.3. Surface PM Concentrations

[36] Surface PM concentrations predicted by AURAMS were compared in the source region for two monitoring stations and at five downwind stations. These seven stations were selected because (a) they make continuous hourly or daily measurements and (b) they had complete records for the 1–16 April 2001 period.

[37] Figure 10 shows the comparison results in the source region. For Albuquerque, New Mexico the hourly measured data are compared with hourly averaged model outputs. For El Paso, Texas only 24-h measured data were available, so they are compared with daily averaged model results. In Albuquerque, which is located in central-western New Mexico (Figure 1), a concentration peak that was predicted by both dust-emission schemes during the S1 period was not supported by measurements. On the other hand, the PM10 concentration peak measured during the S2 period at Albuquerque was significantly underestimated by both schemes. In El Paso, which is located very close to the border of Texas, New Mexico, and Mexico (Figure 1), the dust production nearby during both the S1 and S2 periods was not predicted at all or was considerably underestimated by both schemes. This can be seen in Figure 7, where AURAMS underestimated the dust loading near the U.S.-Mexico border.

Figure 10.

Comparison of time series of model-predicted surface PM10 concentrations with measurements around the source area at two stations: Albuquerque, NM (left) and El Paso, TX (right). The label “Gravimetric” indicates that PM10 concentration was measured using a gravimetric method. The dates designated in the daily average plot represent 24 h from 0600 GMT of the previous day.

[38] The difficulty in precisely predicting the location and time of dust production events stems from the characteristics of the dust production process. Soil dust entrainment occurs only when the wind speed exceeds a threshold value, which is a complicated function of various parameters such as soil and meteorological conditions, and as such, it is intrinsically an intermittent process. For example, SMC continuously changes due to surface-atmosphere exchange as well as meteorological events like precipitation. As the prediction of precipitation, especially convective precipitation, is a weakness of numerical weather prediction models, predicted surface SMC fields are in turn subject to significant error. Therefore lack of reliable information on SMC can cause errors in the prediction of dust emission events as demonstrated in this study.

[39] Similar comparisons between model PM predictions and measurements at five downwind stations are shown in Figure 11. The impact of wind-blown dust particles can be discerned when the two model versions show a clear difference in their predictions. Compared to the source region, the agreement between measured and model-predicted concentrations is encouraging at downwind stations, especially in Oklahoma. It has been reported that the influence of different individual sources of dust, e.g., dry lake beds, diminishes during long-range transport due to merging of the individual dust plumes [Uno et al., 2006]. It is interesting to note that the dust particles produced during the S2 period produced two separate PM10 concentration peaks at downwind stations between April 11 and April 15. The models correctly reproduced this phenomenon, which is evidence that AURAMS can accurately predict the transport pattern of dust particles on a regional scale.

Figure 11.

Same as Figure 10 but for five downwind stations in Oklahoma, Illinois, and Ontario. The label “NAPS” means that it was measured by National Air Pollution Surveillance (NAPS) network of Environment Canada. The dates designated in the daily average plot represent 24 h from 0600 GMT of the previous day.

[40] The two dust-emission schemes predicted similar dust storm effects for S2, but produced quite a large difference for S1, with the MBA scheme agreeing better with the measurements. The difference between the predictions of the two schemes stems from the difference in the way that soil moisture is taken into account and the way that the vertical dust flux is calculated from the horizontal soil flux. It should be noted that the better agreement with observations of the MBA scheme in terms of the surface PM concentration at a limited number of stations cannot be regarded as a proof of better performance of the MBA scheme because there are still a number of remaining uncertainties in the model parameters.

[41] At the farthest downwind stations, especially in Toronto, Ontario, the influence of the dust storms was not apparent at surface either in measurements or in the model predictions. The two models showed noticeable differences on April 14, indicating the influence of dust storm emissions, but the PM concentration on that day was lower than on some other days. Moreover, the model results showed little difference from the “null model”, in which the wind-blown dust production was not taken into account (not shown). This needed more investigation because the DCL plots showed that dust plumes passed this site (not shown) and there are records of precipitation over Ontario and Quebec containing dust taken up in a large cyclone during the S1 period. The vertical distribution of total dust particles are plotted (Figure 12) at four stations to explain these seemingly contradictory results. It should be noted that “total dust” includes not only wind-blown dust produced from a desert area but also fugitive dust particles from other sources such as road dust and fly ash, which appear as diurnal fluctuations in this figure. It is clearly shown in Figure 12 that wind-blown dust particles had risen high enough in the troposphere before they passed the Toronto region that they did not affect surface dust concentration there (e.g., April 11). However, precipitation could transport dust from this upper level to the surface.

Figure 12.

Time evolution of vertical distribution of total dust particle concentrations (μg/m3) predicted by the MBA-scheme version of AURAMS at four locations for the 1–16 April 2001 period. Times are in GMT.

4.4. Effect of Model Parameters

[42] In Figure 13, surface PM10 concentrations predicted by AURAMS using the two dust-emission schemes and the set of parameter values summarized at the beginning of Section 4 are compared to results from two more AURAMS simulations for another set of combined dust-scheme parameters. For the MBA scheme, the finer, unimodal soil-grain-size distribution suggested by Alfaro et al. [2004], which enhances the dust emission, was combined with GEM SMC increased by a factor of 2.345, which suppresses dust emissions as compared to the default set (Chinese soil-grain-size distribution and unadjusted GEM SMC). It is shown in Figure 13 that these two model parameter changes yield very similar results because the two opposing effects effectively compensate for each other. This implies that there are many possible combinations of dust-scheme parameters that can provide simulation results comparable to measurements since the individual effect of some parameters can be quite large (as was shown in Section 3). The lack of reliable measurement data for these various dust-emission-scheme parameters prevents us from judging which set is best among the possible ones.

Figure 13.

Assessment of AURAMS predictions of PM10 time series for two sets of parameter combinations for both dust-emission schemes for the period 1–16 April 2001.

[43] The combination of the lower limit of the soil plastic pressure values and enhanced SMC was used as a modified parameter set for the Shao scheme. Figure 13 shows that the AURAMS run with this modified parameter set significantly underestimated the effect of the S1 storm, whereas its prediction of the S2 storm was comparable to the default set. This may testify to the superiority of the default set over the modified one, at least for the Shao scheme. However, this judgment, even if true, cannot be applied directly to the MBA scheme since the Shao scheme was poor in predicting the S1 storm compared to the MBA scheme. Also, the MBA and Shao schemes are almost totally independent in computing the dust emission rate.

[44] Sensitivity analyses results and the discussion presented in this subsection both call for the collection of accurate model parameters for better validation of the models and more reliable prediction of wind-blown soil dust production. The soil-grain-size distribution should be precisely measured in the source region. For accurate SMC data, remote sensing may have to be used. A more sophisticated modeling technique for solving the soil moisture equation (e.g., ISBA scheme employed in GEM beginning fall 2001) could be another way to reduce the model uncertainty. Appropriate soil plastic pressure values can be determined from comparison of model predictions with measurements once other parameters have been decided.

4.5. Wet Deposition in Northeastern North America

[45] This study was motivated by the unusually fast dust transport that occurred for the S1 period. Eight Canadian Air and Precipitation Monitoring Network (CAPMoN) samplers in Ontario, Quebec and Pennsylvania (an internetwork calibration site at State College) all collected unusual red dust in precipitation that occurred during the 24-h collection period of 1300 GMT, 7 April through 1300 GMT, 8 April. Analysis for soluble ions showed that the amount of calcium ion in these eight samples ranged from 0.4 to 6.2 mg/L, considerably higher than the annual average values of 0.24—0.92 mg/L for these sites and consistent with abnormally high soil aerosol. Total wet-deposited calcium ion ranged from 2.8 to 17.6 mg/m2. Thus it is of interest whether this fast transport was reproduced well by our simulation.

[46] Figure 14a shows the average DCL during 0300 GMT through 0600 GMT, 8 April and the accumulated wet deposition flux of soil dust during the 24-h CAPMoN sampling period predicted by AURAMS with MBA scheme. These figures clearly show that some wind-blown dust produced in the source area was predicted to be transported quickly to Ontario and Pennsylvania and wet-deposited there. The predicted amount of wet-deposited dust (5–20 mg/m2) is, however, similar to that of calcium ions dissolved in precipitation. The amount of total dust includes insoluble components as well and might be much larger than that of soluble calcium ion. Forward trajectory analysis showed that only dust particles lifted up as high as about 3 km in the source region could reach Ontario within 24 h (not shown). Similar trajectory analysis results had been reported in the literature [Vet et al., 2003]. Based on this trajectory analysis, as another sensitivity test, two more AURAMS simulations with the MBA scheme were performed in which the emitted wind-blown dust were injected into the 17 and 20 lowest model layers (from the surface to about 3 and 5 km AGL, respectively) instead of the 3 lowest layers, and the results for DCL and 24-h wet deposited dust were again plotted (Figures 14b and 14c). Compared to Figure 14a, significant increases in the amount of dust reaching Ontario are shown. The results obtained for 17- and 20-layer injection were also different from each other. Only the 20-layer version could explain the amount of dust wet-deposited in Ontario region.

Figure 14.

Average DCL (0300 GMT, 8 April–0600 GMT, 8 April) (left) and accumulated wet deposition flux of wind-blown dust (1300 GMT, 7 April–1300 GMT, 8 April) (right) predicted by AURAMS (both in mg/m2) with the dust emission depth of (a) 90 m, (b) about 3 km, and (c) about 5 km.

[47] This result may imply that the unusually fast transport observed for the Red Dust Episode have been caused by vigorous uplift of dust generated at the surface to mid-tropospheric altitudes in the source area. Meteorological analysis for the S1 period showed that strong near-surface winds associated with three major meteorological conditions, namely, a rapidly deepening surface cyclone in Colorado, a squall line that developed along a dryline across the Texas Panhandle, and a Pacific cold front entering from the west, caused widespread wind erosion in the source area during the Red Dust Episode [Doggett IV et al., 2002]. Surface and 500 hPa analysis indicated that the storm was vertically stacked from the surface to more than 5 km altitude [Decker and Martin, 2005]. This unique meteorological condition may have produced vigorous and deep sub-grid-scale convective uplifting, which is not fully taken into account by the spatial resolution of AURAMS used in this study. The role of convective transport in altering vertical dust distribution has recently been examined by Jung et al. [2005], who showed that the strong uplift due to sub-grid-scale convection could carry dust particles from the boundary layer to the middle and upper troposphere. Strong vertical updrafts associated with the squall line on the afternoon of 6 April would also have carried dust to mid-tropospheric levels.

[48] It should be noted that the dust emission depth is just one of many parameters, such as wind field, dust emission rate, turbulent mixing intensity, etc., that affect the amount of dust transported. Nevertheless, the sensitivity analysis performed in this subsection suggests that appropriate parameterization of additional vertical mixing by sub-grid-scale convection may help to better predict the long-range transport of dust particles.

5. Conclusions

[49] Two different wind-blown-dust emission schemes have been incorporated into a size- and composition-resolved regional PM model to simulate the Red Dust Episode. A number of model simulations and sensitivity analyses for various dust-emission-scheme parameters were then performed.

[50] The model simulation reasonably reproduced the temporal and spatial distribution of wind-blown dust particles during the episode period in the downwind area of Oklahoma but not in the source region itself. At stations farther downwind such as Toronto, Ontario, surface PM2.5 concentrations were not much affected by long-range transport of wind-blown dust whereas the model-predicted vertical dust distribution suggested that dust plumes passed this station at high altitude. Our regional air quality model combined with either of the two dust-emission schemes captured the main features of the transport of dust.

[51] Relatively large discrepancies between measurements and model predictions in the source region can be attributed to the intermittent characteristics of the dust entrainment mechanism and to inaccuracies in model parameters. The most critical model parameter responsible for inaccurate prediction of wind-blown-dust emission was shown to be the soil moisture content (for both schemes) followed by the soil-grain-size distribution (for the MBA scheme). Determination of soil plastic pressure (for the Shao scheme) can be made only after reliable SMC data are collected. Because of the complicated and often opposing effects of these parameters on dust emission rate, it is not easy to reach clear conclusions on the validity and universality of these parameters at this point. It is encouraging, however, that the model could reasonably simulate the dust storm episode using parameter values, which had been suggested, used, and/or fitted in previous studies at locations other than North America, without any major adjustments. For further validation and reliable use of dust emission models, accurate field and remote sensing measurements of those model parameters are imperative.

[52] The unusually fast transport of dust from the source area to northeastern North America during this episode appears to be due to vigorous vertical mixing and uplift of emitted dust in the source region to mid-tropospheric levels. Turbulent mixing associated with strong winds may partly account for this vertical mixing, but it is likely that additional vertical mixing caused by sub-grid-scale convection due to the severe meteorological conditions also has contributed. Appropriate parameterization of additional vertical mixing by sub-grid-scale convection may help to better predict the long-range transport of dust storms.


[53] The authors acknowledge the Environment Canada National Atmospheric Chemistry (NAtChem) database and its data-contributing agencies/organizations, including the Aerometric Information Retrieval Systems (AIRS) and National Air Pollution Surveillance (NAPS) networks for the provision of the measurement data for 2001 used in this paper. The Soil Climate Analysis Network (SCAN) is acknowledged for its provision of the soil moisture content data. The authors thank Doug Moore and David Meyer for their efforts in establishing and maintaining Sevilleta and KONZA_EDC sites, respectively, of the AErosol RObotic NETwork (AERONET). The authors are also grateful to René Servranckx of Environment Canada for the GOES satellite data processing as well as helpful discussions and comments on this study. The authors appreciate the comments provided by three reviewers that have resulted in an improved manuscript. Finally, S. H. Park was supported during this research by the Visiting Fellowship program of the Natural Sciences and Engineering Research Council of Canada.