Open Access

Abstract. The Community Multiscale Air Quality (CMAQ) model is a state-of-the-science air quality model that simulates the emission, transformation, transport, and fate of the many different air pollutant species that comprise particulate matter (PM), including dust (or soil). The CMAQ model version 5.0 (CMAQv5.0) has several enhancements over the previous version of the model for estimating the emission and transport of dust, including the ability to track the specific elemental constituents of dust and have the model-derived concentrations of those elements participate in chemistry. The latest version of the model also includes a parameterization to estimate emissions of dust due to wind action. The CMAQv5.0 modeling system was used to simulate the entire year 2006 for the continental United States, and the model estimates were evaluated against daily surface-based measurements from several air quality networks. The CMAQ modeling system overall did well replicating the observed soil concentrations in the western United States (mean bias generally around ±0.5 μg m−3); however, the model consistently overestimated the observed soil concentrations in the eastern United States (mean bias generally between 0.5–1.5 μg m−3), regardless of season. The performance of the individual trace metals was highly dependent on the network, species, and season, with relatively small biases for Fe, Al, Si, and Ti throughout the year at the Interagency Monitoring of Protected Visual Environments (IMPROVE) sites, while Ca, K, and Mn were overestimated and Mg underestimated. For the urban Chemical Speciation Network (CSN) sites, Fe, Mg, and Mn, while overestimated, had comparatively better performance throughout the year than the other trace metals, which were consistently overestimated, including very large overestimations of Al (380%), Ti (370%) and Si (470%) in the fall. An underestimation of nighttime mixing in the urban areas appears to contribute to the overestimation of trace metals. Removing the anthropogenic fugitive dust (AFD) emissions and the effects of wind-blown dust (WBD) lowered the model soil concentrations. However, even with both AFD emissions and WBD effects removed, soil concentrations were still often overestimated, suggesting that there are other sources of errors in the modeling system that contribute to the overestimation of soil components. Efforts are underway to improve both the nighttime mixing in urban areas and the spatial and temporal distribution of dust-related emission sources in the emissions inventory.


Introduction
Mineral dust (also referred to as soil) can represent a significant portion of the measured particulate matter (PM), both fine PM with a diameter less than 2.5 µm (PM 2.5 ) 5 and coarse PM with a diameter less than 10 µm (PM 10 ). PM contributes to the deterioration of air quality and can lead to adverse health effects resulting in premature death (Dockery, 2009), degradation of pristine environments through reduced visibility (Malm et al., 1994) and radiation impacts by absorbing and/or reflecting solar radiation (Sokolik and Toon, 1996). As such, understanding the emission, transport and fate of dust in the environment is important for protecting human health and sensitive ecosystems, as well as assessing the impact of air quality on climate (e.g. surface temperature) due to radiative feedbacks from dust and PM.
The Community Multiscale Air Quality (CMAQ; Byun and Schere, 2006) model is a state-of-the-science air quality model capable of reproducing the emission, trans-Introduction such as dust from unpaved roads and agricultural tilling, and the chemical boundary conditions (BCs) now include WBD from long-range transport. In this study, the CMAQv5.0 model has been used to simulate the entire year 2006 for the continental United States (CONUS). The CMAQ model estimates of the trace elements comprising dust are evaluated against daily surface based measurements of 5 the same elements. In addition to the annual base simulation, several seasonal sensitivity simulations are performed in order to assess the impact that changes made to the emissions inventory, boundary conditions, and inclusion of the WBD mechanism have on the CMAQ model estimates of dust. Finally, several recommendations for further improving the CMAQ estimates of dust are discussed.

2 Model inputs and configuration
The CMAQ model requires inputs of gridded meteorological fields, emissions data and boundary conditions. For a regional or continental CMAQ model simulation, the meteorological fields are typically provided by a regional scale meteorological model, such as the Weather Research and Forecast (WRF; Skamarock et al., 2008) model. The in- 15 put emissions are typically derived from a standard emissions input database, such as the USEPA's National Emissions Inventory (NEI), for which base year inventories are available every three years. Finally, chemical boundary conditions are typically based off a larger, coarser CMAQ model simulation or from a hemispheric or global air quality simulation provided by a global chemistry model. The meteorological, emission and 20 boundary condition inputs used in the base CMAQ model simulation are described in this section.

Meteorological inputs
The meteorological inputs for the CMAQ model simulations were provided by a 2006 annual CONUS WRFv3.3 model simulation that utilized 12-km horizontal grid spacing Introduction and 35-vertical layers of variable thickness extending up to 50 hPa, with the top of the lowest model layer at approximately 20-m above ground level. Initial and boundary conditions for WRF were provided by the North American Model (NAM) available from the National Centers for Environmental Prediction. The WRF simulation utilized the Rapid Radiation Transfer Model Global (RRTMG) for long-and short-wave radiation (Iacono 5 et al., 2008), the Kain-Fritsch 2 cumulus parameterization (Kain, 2004), the Morrison microphysics scheme (Morrison et al., 2009), the Pleim-Xiu land-surface model (PX-LSM; Pleim, 2007 andXiu, 2003) and the Asymmetric Convective Model version 2 (ACM2; Pleim, 2007a, b) planetary boundary layer (PBL) scheme. Four dimensional data assimilation (FDDA) was used to constrain the model above 10 the PBL; however unlike previous WRF model simulations, no FDDA was used within the PBL, which results in an improved wind speed bias in the PBL as compared to WRF simulations which utilized FDDA throughout the troposphere (Gilliam et al., 2012). The raw WRF outputs were processed for the CMAQ model using version 4.0 of the Meteorology-Chemistry Interface Processor (MCIP; Otte and Pleim, 2009). A 10-day 15 spin-up period was utilized to eliminate the effects of the initial conditions.

Base emissions
The input emissions for the CMAQ model simulation are based on a 12-km national US domain with speciation for the Carbon-Bond 05 (CB05) chemical mechanism (Yarwood 20 et al., 2005 (Allen et al., 2012) were calculated using time dependent input meteorology and observations from the National Lightning Detection Network (NLDN). The raw emissions inputs were preprocessed for the CMAQ model using the Sparse Matrix Operator Kernel Emissions (SMOKE; Houyoux et al., 2000).

Anthropogenic fugitive dust emissions
Crustal elements such as Ca and Fe are present in anthropogenic and wind-blown fugitive dust, but may also be found in some fly ash and industrial process emissions (which are chemically similar to crustal emissions). The sources of AFD include unpaved road dust, paved road dust, commercial construction, residential construction, 15 road construction, agricultural tilling, livestock operations, and mining and quarrying. Unpaved road dust is the largest single emissions category within the non-point fugitive dust category, accounting for about one third of non-windblown fugitive dust emissions. This is followed in size by dust from tilling, quarrying and other earth-moving. Source apportionment studies have shown that AFD emissions contribute on the order 20 of 5-20 % of PM 2.5 and 40-60 % of PM 10 in urban areas that either have been or potentially may be unable to attain the National Ambient Air Quality Standards (NAAQS) for PM 2.5 and/or PM 10 (Watson and Chow, 2000). Conversely, air quality models suggest vastly higher contributions from current fugitive dust emission inventories, with contributions ranging from 50-80 % for PM 2.5 and 70-90 % for PM 2.4 and/or PM 10 (Watson 25 and Chow, 2000). Although dust makes up the majority of PM emissions, much of the emitted mass gets deposited on surfaces near the source at scales much smaller than the model grid cell resolution (Veranth et al., 2003;Etyemezian et al., 2004). This is not 1864 Introduction true of other sources which are either emitted at a higher elevation (e.g. power plant stacks) or are emitted in warm exhaust (e.g. from vehicles) which rises quickly and gets entrained into the air mass. To correct for the near-source removal of dust, emissions from these sources are typically multiplied by a transportable fraction as proposed by Pace (2005). This transportable fraction is applied on a per county basis to both PM 10 5 and PM 2.5 . PM 2.5 emissions in the NEI are reported as an annual total. In order for these emissions to be used in modeling applications, they need to be chemically speciated and allocated to finer temporal resolutions (e.g. each hour of the year). PM 2.5 emissions in the NEI are typically speciated into five chemical components (organic carbon (OC), elemental carbon (EC), sulfate, nitrate, and other). Recently, an improved speciation of the PM has been developed to include, in addition to the current PM species, a range of trace metals as well as separate non-carbon organic matter and metal-bound oxygen (Reff et al., 2009). The current temporal profile used by the EPA to allocate dust emissions to daily resolution assumes no monthly variability and no weekday/weekend 15 variation (http://www.epa.gov/ttn/chief/emch/index.html#2005). In essence, each day is represented identically throughout the year.
In this work, three changes were made to improve and diagnose the fugitive dust emission estimates used in chemical transport modeling. The first change involves improvements to the transportable fraction applied to the gridded emission inventory field. 20 Second, a new mapping of the temporal profiles is applied to fugitive dust emissions so that they vary by day of the year. Finally, the chemical speciation of PM 2.5 emissions is updated based on Reff et al. (2009). This allows for better source attribution of the measured trace metals.
In Pace (2005), the transportable fraction, (i.e. the amount that is not "captured" by 25 near-source removal), is calculated on a per county basis for 3 Regional Planning Organizations using the Biogenic Emission Inventory System (BEIS) version 2 county-level land use information (Byun and Ching, 1999). To improve the transportable fraction in CMAQ, it was recalculated at a 1-km resolution using the newer BELD3 database Introduction  , 2002) for all of the CONUS using five broad land use categories (e.g. forest, urban, sparsely wooded and grass, agricultural, and barren/water), generally resulting in an increase in the transportable fraction in the western United States and little change to the transportable fraction in the eastern United States (Pouliot et al., 2010). Table 1 shows the mapping of the BELD3 land use types to the five broad land 5 use categories and the associated capture fraction. A second improvement to the emissions was to modify the temporal activity factors used in the emissions processing. For each of the fugitive dust source categories, revisions were made to the monthly, weekly, and daily temporal profiles. The rationale for these temporal allocation changes is that the activity factors for associated sectors differ from the activity factors that have previously been assumed for the fugitive dust emissions. For example, a flat daily profile had previously been used for agricultural tiling. This has now been replaced by the same temporal profile used for the combustion emissions from agricultural equipment in the non-road mobile source sector, which is a more realistic representation of the daily activity pattern for agricultural 15 tilling. The temporal factors for each of the fugitive dust sectors have been harmonized with other components of the emission inventory and processing platform where appropriate (see http://www.epa.gov/ttn/chief/conference/ei19/session9/pouliot.pdf for additional details).
Finally, the speciation of PM 2.5 emissions from all sources, including the dust 20 sources, was updated. These updates to the speciation of PM 2.5 were based on the work of Reff et al. (2009), in which an inventory for trace metals from PM 2.5 was derived using EPA's SPECIATE database (EPA, 2006;Simon et al., 2010 dust, external combustion boilers (from electric generating units), paved road dust, construction, and mining and quarrying.

Chemical boundary conditions
The Chemical BCs for the CMAQ model simulation were provided by an annual 2006 GEOS-Chem (Bey et al., 2001) simulation. The GEOS-Chem simulation utilized the 5 pre-patch version 9-01-01 of the model with secondary organic aerosols enabled, and was run using 2.0 degree by 2.5 degree (latitude-longitude) horizontal grid spacing and 24-vertical layers. The simulation utilized GOES-5 meteorology and the default emissions based on the 2005 EPA NEI. Since GEOS-Chem and CMAQ use different names and definitions for a number of species, it is necessary to map the GEOS-Chem species to the CMAQ species. GEOS-Chem uses the Dust Entrainment and Deposition (DEAD) scheme with GO-CART source function (Zender et al., 2003;Ginoux et al., 2001;Fairlie et al., 2007) and transports WBD in four size bins with edges at 0.1, 1, 1.8, 3.0, and 6 µm radii. For use in BCs, the GEOS-Chem dust was speciated into trace metals as well as other 15 lumped species based on a composite of four desert soil profiles from SPECIATE. Dust in the smallest GEOS-Chem size-bin was matched to the CMAQ accumulation mode species (J mode) while the three larger GEOS-Chem size bins corresponded to CMAQ's coarse mode (K mode). to include the detailed speciation profiles described in Sect. 2.2.2, a representation of contributions from WBD, inclusion of NO emissions from lightning, an updated turbulent mixing scheme under stable conditions and an improved vertical advection scheme, as well as a number of additional updates to the model code structure. For additional details regarding the new features and enhancements in CMAQv5.0, please refer to the 5 release notes available for download along with the CMAQ model code.

CMAQ model configuration
The CMAQ model simulation covers the CONUS and parts of southern Canada and northern Mexico using 12-km horizontal grid spacing and 35-vertical layers matched to the WRF vertical layer structure. The CMAQ model simulations performed for this work utilize the CB05TUCL chemical mechanism, the ACM2 PBL scheme, the Euler Backward Iterative (EBI) solver, in-line plume rise for point sources, and employ the optional in-line photolysis calculation and NO emissions from lightning.
The two most important changes in the new version of the model that affect the estimates of dust are the updates to the aerosol chemistry and speciation, and the representation of the effects of WBD in the model. In addition, changes to turbulent 15 mixing and vertical advection also affect how dust is dispersed and transported in the model.
Enhancements to the aerosol module in CMAQv5.0 were directed both at improving the aerosol chemistry as well as speciation. Evaluation studies have revealed that the largest biases in CMAQ PM 2.5 results are driven by over predictions of the unspeciated 20 PM 2.5 , referred to hereafter as PM other (Appel et al., 2008); this component constitutes over half of the NEI for PM 2.5 using the old five-component chemical speciation scheme. Detailed speciation profiles derived from the work of Reff et al. (2009) were used to further subdivide emissions of PM other into primary ammonium (NH + 4 ), sodium (Na + ), chloride (Cl − ), selected trace metals (Mg, Al, Si, K, Ca, Ti, Mn, and Fe), and 25 non-carbon organic mass (NCOM).
The CMAQ transport and chemistry operators were further modified to explicitly represent these nine additional PM constituents. This additional speciation now allows for detailed characterization of the species, processes, and emission sector contributions Introduction to the model bias in primary and consequently total PM. The explicit treatment of Fe and Mn also allows for explicit representation of their catalysis effects on S(IV) to S(VI) conversion through aqueous chemistry, and consequently more consistent treatment of sulfate (SO 2− 4 ) production pathways in the model. The representation of gas/particle partitioning of chloride, ammonia and nitrate was 5 also improved through the incorporation of ISORROPIA version II (ISORROPIA II; Fountoukis and Nenes, 2007;Nenes et al., 1998Nenes et al., , 1999. In addition to more robust solutions compared to previous versions of ISORROPIA, ISORROPIA II includes calcium (Ca 2+ ), potassium (K + ), and magnesium (Mg 2+ ) ions, species abundant in seasalt and soil dust which can affect the partitioning of semivolatile inorganic species. The explicit representation of dust emission and PM composition simulated by CMAQv5.0 facilitates the expanded speciation and incorporation of ISORROPIA II.
In previous versions of CMAQ, contributions of natural WBD on airborne PM mass were not explicitly represented. CMAQv5.0 includes a module that dynamically estimates natural emissions of fine and coarse dust particles due to wind action over arid 15 and agricultural land.

Observation data
There are several sources of routine, ground based observations of PM that include observations of the speciated dust components. Both the Interagency Monitoring of PROtected Visual Environments (IMPROVE; http://vista.cira.colostate.edu/improve/) 20 and Chemical Speciation (CSN; http://www.epa.gov/ttnamti1/speciepg.html) networks provide surface measurements of total PM 2.5 and PM 10 , along with speciated PM 2.5 measurements of SO 2− 4 , NO − 3 , NH + 4 , Na + , Cl − , and the trace metals of Mg, Al, Si, K, Ca, Ti, Mn, and Fe. The IMPROVE network consisted of 161 sites in 2006, with the majority of the sites located in the western United States. The IMPROVE network sites 25 are typically located in rural areas, with a large number of the sites located in national parks, and as such the measurements tend to represent the background concentration Introduction with the CMAQ model estimates using the Atmospheric Model Evaluation Tool (AMET; Appel et al., 2011). The pairing is done without any interpolation of the model value to its location within the grid (a simple grid value to measurement value matching is used), and therefore the analysis presented is subject to the inherent incommensurability issues that arise when comparing model grid-cell averaged values to point mea-10 surements (Swall and Foley, 2009). Note that measurements that fall within the same grid cell are not averaged together, but instead paired individually to the same grid cell value.
Soil is not directly measured at the IMPROVE and CSN sites, but instead is derived from measurements of the various trace metals that are measured at each 15 site. The soil equation is useful as an aggregate measure of soil (as it could be tedious to examine each individual element separately) and accounts for the metalbound oxygen and K associated with the included elements. The equation for computing soil is shown in Eq. (1) and is based on the soil calculation equation used by the IMPROVE network (http://vista.cira.colostate.edu/improve/publications/graylit/ 20 023 SoilEquation/Soil Eq Evaluation.pdf). The same equation is used for the IM-PROVE and CSN networks, as well as in the CMAQ post-processing to define soil.
A recent study by Indresand and Dillner (2012) showed that Si and Al measurements from the IMPROVE network are misreported when the sulfur (S) to Fe (S/Fe) ratio is 25 large. This is due to low-energy spectral interference by S in the x-ray fluorescence spectrometry (XRF) instrument used at the IMPROVE sites. They examined IMPROVE data from 2008 and found that when the observed S/Fe ratio was less than 8, which 1870 Introduction constituted 49 % of the data, the reported Si and Al value were not affected by the S interference. For S/Fe ratios greater than 8 but less than 70 (47 % of the data), the Si value was over-reported by up to 100 % and the Al value was either over-reported by 50 % or incorrectly reported as below detection limit. For S/Fe ratios greater than 70 (4 % of the data), the Si value was over-reported by a factor of 2 or more, while the Al 5 value was misreported by ±50 % or more. They advise using the IMPROVE Si and Al data cautiously when the S/Fe ratio is large (while those data are included in the analysis here, no strong conclusions are made based on those particular data). The CSN measurements do not suffer from the same issue as the IMPROVE measurements due to lower measured S concentrations (due to a lower flow rate) and better peak baseline to the winter, with soil concentrations overestimated in the eastern United States and unbiased to slightly underestimated in the western United States. Similar spatial trends for the summer and winter were reported by Tong et al. (2012). Overall, soil is consistently overestimated in the eastern United States throughout the year, while in the western United States soil estimates tend to fluctuate between a slight underestimation to slight overestimation. Air-born soil in the eastern United States is primarily the result of anthropogenic sources, with a smaller contribution from natural WBD, whereas the western United States has a greater contribution to soil from WBD and long-range transport. Several possible reasons for the overestimation of dust in the eastern United States include AFD emissions that are too high in the model, an urban transportable fraction of dust that may be too large or too small, a contribution to soil from WBD may be overestimated (should be small for eastern United States), and the modeled PBL height in urban areas may be too low due to insufficient heat retention in urban areas (i.e. urban heat island effect). Several of these issues will be 5 discussed further Sect. 5.  Fig. 4 is the large overestimation of Mn and Ca concentrations during the nighttime hours (the same trend is seen for the other trace metals). Similar overestimations have been observed in other primary emitted species (e.g. NO and CO) in urban areas. This is due to the tendency of the WRF model to underestimate the overnight mixing in urban areas, possibly due to PBL heights that are too low or a min-20 imum eddy diffusivity (Kz min ) that is too small, which results in an over-concentration of pollutants near the surface, ultimately leading to high model biases. A number of the trace metals are emitted in urban areas from industrial operations that are continuous throughout the day and night. As such, those elements would tend to be overestimated during the night due to the insufficient mixing in the model. Work is currently underway 25 to improve the nighttime mixing in urban areas by adding impervious surface information (e.g. pavement) into the WRF model in order to capture the heat retention in cities which ultimately would improve the representation of mixing during stable conditions in these urban environments.

Effect on sulfate chemistry
In previous versions of the CMAQ model, aqueous phase SO 2− 4 production via the metal catalysis oxidation pathway was calculated using prescribed background concentrations of Fe(III) and Mn(II). As CMAQv5.0 contains predicted values of Fe and Mn, these tracked concentrations are now used to estimate the Fe(III) and Mn(II) val-5 ues for the metal catalyzed oxidation pathway. In addition to using model estimated values of Fe and Mn, the rate constant for in-cloud SO 2 oxidation via metal catalysis was also updated in CMAQv5.0 following Martin and Good (1991). Additional details regarding the implementation of the new treatment of crustal elements in the sulfate chemistry in CMAQ can be found in Sarwar et al. (2013). 4 concentrations is small in the summer, as expected, since SO 2− 4 production is predominantly due to oxidation by H 2 O 2 and OH. In winter, when the levels of these oxidants are lower, the contribution of the aqueous Fe/Mn catalysis reaction pathway becomes important. As a result, the change from the old crustal treatment in 5 CMAQv4.7 to the new treatment in CMAQv5.0 has the greatest impact on SO 2− 4 concentrations in the winter. It should also be noted that the relatively good agreement between CMAQ and observed concentrations (Fig. 3) builds confidence in the ability to integrate these mode estimated concentrations into the CMAQ chemistry.

10
The time series in Fig. 1 shows several episodes in the summer where the model grossly underestimates the observed soil concentrations during high observed concentrations of soil at both the IMPROVE and CSN sites. Three distinct episodes of the high observed soil concentrations were identified to have occurred on 13 July, 28 July and 3 August 2006. Figure 6 presents spatial plots of observed and model simulated 15 average soil concentrations from the three days identified above. The observed soil concentrations are highest in the southeastern United States, with observed concentrations greater than 4 µg m −3 (a number of sites having mean concentrations greater than 10 µg m −3 ) extending from Florida to central Texas, north into the Great Lakes region and into the Northeast. While the CMAQ model estimates the highest soil con-20 centrations in the same regions, most of the sites have mean concentrations less than 4 µg m −3 , with only a few of the sites having mean concentrations above 4 µg m −3 . The observed mean concentrations in the southeast United States for the three days are in strong contrast to the average mean observed soil concentration for July and August (not shown), which is typically less than 4 µg m −3 for the same region.

25
One possible cause of the high observed soil concentrations in July and August could be transported dust from the African continent, particularly from the Sahara desert region (Perry et al., 1997;Bates et al., 2008) Figure 7 shows the average surface-level soil concentration from 25 July through 5 August 2006, which covers two of the three high soil concentration episodes in the eastern United States identified from Fig. 1. For the 12-day period, the GEOS-Chem derived BCs capture high concentrations of soil (up to 20 µg m −3 ) along the southeast-10 ern boundary of the CMAQ model domain, which spread westward toward the CONUS. However, while the BC inputs include elevated soil concentrations, the high concentrations of soil do not progress far enough west to reach the eastern United States, with most of the soil being removed before it makes it to the coastal and inland areas (although some relatively high soil concentrations are observed in Florida). 15 One possible reason the high soil concentrations are not estimated correctly for the interior eastern United States may be due to an overestimation of convection and precipitation by the WRF model off the southeast coast of the United States and along the Gulf Stream, which results in excessive removal of dust by wet deposition. Figure 7 presents the WRF accumulated precipitation for the period of 25 July through 5 August 20 2006. The WRF model estimates for precipitation are large off the east coast of the United States, as well as in parts of the Caribbean, over Florida and in the Gulf of Mexico. Unfortunately, little data exist with which to verify the precipitation estimates over the water. However, Park et al. (2006) andFairlie et al. (2006) note a similar underestimation of dust concentrations in Florida by the GEOS-Chem model due to excessive 25 wet deposition (also the result of an overestimation of convective precipitation by the meteorological model).
Another reason for the lack of westward transport of the dust from the boundary may be due to incorrect wind flow or wind flow that is too weak off the southeast coast and in the Gulf of Mexico, which would result in the dust being advected in the wrong direction or settling out into the ocean before it reaches land. Additional sensitivity analyses with the WRF simulation are needed to confirm this as a possible cause of the transported dust issue within the CMAQ model simulation.

5
Several CMAQ model sensitivities were performed to further assess possible reasons for the noted discrepancies between modeled and observed trace element concentrations. Specifically, three separate model simulations were performed for the March through May time period, when observed soil concentrations are the highest, one with the emissions of anthropogenic fugitive dust removed, one with the WBD feature turned 10 off, and one with both the AFD and WBD removed. These sensitivity simulations are compared to the base model simulation for the same time period, and the impact of each change on the model estimates is assessed.

Effect of AFD emissions updates
Figures 8 and 9 present time series of observed soil concentrations from the IMPROVE 15 and CSN networks, respectively, and the corresponding model simulated soil concentrations for several CMAQ model sensitivities for the spring period. As noted previously in Fig. 1, the base CMAQ model simulation soil concentrations are slightly overestimated for most of the period compared to the observed soil concentrations from the IMPROVE network, with the largest overestimations from late-April to late-May. How-20 ever, for the CSN, soil concentrations are typically overestimated by 1 µg m −3 or more for the entire period in the base simulation. The largest source of soil (trace metals) emissions in the emissions inventory (based on total soil emissions) is AFD (75 %), followed by electric generation units (EGUs; 13 %), industrial mineral processes (2 %), industrial fuel consumption (2 %) and industrial metal production (2 %). is an overestimation of the AFD emissions as well as likely issues with their diurnal temporal allocation.

Effect of WBD mechanism
The CMAQ model estimates for the simulation in which the effects of WBD were removed (NoWBD) are lower than the base model simulated soil concentrations, but not 15 as low as the simulation where AFD emissions were removed. The largest decrease in soil concentrations in the NoWBD simulation compared to the base simulation occur in the late March through mid-April period (Figs. 8 and 9), indicating the effects of WBD in the model are significant during that period. Conversely, there is very little difference between the base simulation and the NoWBD simulation from mid-April through the 20 end of May, suggesting the WBD effects are small. Figure 10 presents a spatial plot of the absolute change in bias between the base CMAQ simulation and the NoWBD simulation for the IMPROVE and CSN sites. Expectedly, the largest impact on the model simulated soil concentrations occurs in the arid/semi-arid regions of the southwestern United States, with the bias slightly higher 25 in the NoWBD simulation versus the base simulation. There is little to no impact to the bias in the eastern United States, where the bias in soil was highest in the base simulation (Fig. 2). These results suggest that an overestimation of WBD is not responsible 1879 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | for the high model estimated soil concentrations in the eastern United States in the spring, where the effects of WBD should be small anyway.

Effect of both AFD and WBD
The final sensitivity performed removes the effects of both WBD and AFD emissions (referred to simply as NoDust) in order to assess the cumulative impact that those 5 two updates to modeling system had on the model estimates of soil. The NoDust model simulation always has lower soil concentrations than the base simulation and the NoWBD simulation (Figs. 8 and 9). There are several days during which the effects of either the AFD emissions (e.g. 24 March) or WBD (e.g. 2 April) dominate the change in soil concentrations compared to the base simulation. In addition, soil concentrations 10 are underestimated during the periods from 3 March to 12 March and 14 April to 23 April in the NoDust simulation, which are the only periods when soil is underestimated at the IMPROVE network sites (Fig. 8).
It is clear from the plots in Fig. 10 that the change in bias in the NoDust simulation is dominated by the removal of the AFD emissions, with the effect of removing 15 WBD limited in time and space. Since AFD emissions and WBD should be nonzero, the result of improved model performance when those emissions are removed entirely suggests that there are other errors in the modeling system (e.g. emission inputs) that contribute to an overestimation of soil. Wind-blown dust generally constitutes a small, temporally and spatially localized contribution to the soil concentrations in the model, transporting the high soil concentrations from the boundary into the interior United States, which may be due to an overestimation of convective activity (e.g. precipitation) in the WRF model simulation which results in too much deposition. However, more analysis is needed to determine the exact cause of the underestimation of soil in the CMAQ model during these dust events.

5
In addition to the base model simulation, several model sensitivity simulations were also performed for the spring period to assess the impact of uncertainties in AFD emissions and natural WBD dust emission estimates on the model estimates of soil. As expected, removing the AFD emissions resulted in substantially lower model soil concentrations. Similarly, removing the effects of WBD emissions also lowered the model  Etyemezian, V., Ahonen, S., Nikolic, D., Gillies, J., Kuhns, H., Gillette, D., and Veranth, J.: Deposition   Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Malm, W. C., Sisler, J. F., Huffman, D., Eldred, R. A., and Cahill, T. A.: Spatial and seasonal trends in particle concentration and optical extinction in the United States, J. Geophys. Res., 99, 1347Res., 99, -1370Res., 99, , 1994. Martin, R. L. and Good, T. W.: Catalyzed oxidation of sulfur dioxide in solution: the ironmanganese synergism, Atmos. Environ., 25, 2395-2399, 1991 Morrison, H., Thompson, G., and Tatarskii