Role of ammonia chemistry and coarse mode aerosols in global climatological inorganic aerosol distributions

We use an inorganic aerosol thermodynamic equilibrium model in a three-dimensional chemical transport model to understand the roles of ammonia chemistry and natural aerosols on the global distribution of aerosols. The thermodynamic equilibrium model partitions gas-phase precursors among modeled aerosol species self-consistently with ambient relative humidity and natural and anthropogenic aerosol emissions during the 1990s. Model simulations show that accounting for aerosol inorganic thermodynamic equilibrium, ammonia chemistry and dust and sea-salt aerosols improve agreement with observed SO 4 , NO 3 , and NH 4 aerosols especially at North American sites. This study shows that the presence of sea salt, dust aerosol and ammonia chemistry signiﬁcantly increases sulfate over polluted continental regions. In all regions and seasons, representation of ammonia chemistry is required to obtain reasonable agreement between modeled and observed sulfate and nitrate concentrations. Observed and modeled correlations of sulfate and nitrate with ammonium conﬁrm that the sulfate and nitrate are strongly coupled with ammonium. SO 4 concentrations over East China peak in winter, while North American SO 4 peaks in summer. Seasonal variations of NO 3 and SO 4 are the same in East China. In North America, the seasonal variation is much stronger for NO 3 than SO 4 and peaks in winter. 10–70% in polluted areas. Second, they increase modeled nitrate over oceans and reduce nitrate over Northern hemisphere continents. Third, they reduce ammonium formation over oceans and increase ammonium over Northern Hemisphere continents. Comparisons of SO 4 , NO 3 and NH 4 deposition between pre-industrial, present, and year 2100 scenarios show that the present NO 3 and NH 4 deposition are twice pre-industrial deposition and present SO 4 deposition is almost ﬁve times pre-industrial deposition. r 2006 Elsevier Ltd. All rights reserved.


Introduction
Tropospheric aerosols pose the largest uncertainties in estimates of climate forcing by anthropogenic changes to the atmosphere's composition (National Research Council (NRC), 1996). Atmospheric aerosols are usually mixtures of many components, partly composed of inorganic acid (e.g. H 2 SO 4 , HNO 3 ), their salts (e.g. (NH 4 ) 2 SO 4 Á NH 4 NO 3 ), and water (Charlson et al., 1978;Heintzenberg, 1989). A couple of years ago, multicomponent aerosol concentrations are not routinely calculated within global atmospheric chemistry or climate models yet. The reason is that simulations of these aerosol particles, especially those including semi-volatile components, require complex and computationally expensive thermodynamic calculations (Metzger, 2000;Metzger et al., 2002a). For instance, the aerosol-associated water depends on the composition of the particles, which is determined by gas/ liquid/solid partitioning, which is in turn strongly dependent on temperature and relative humidity (Metzger, 2000). This study focuses on the roles of ammonia chemistry in multicomponent aerosol formation and partitioning, and on the sensitivity of this partitioning to the presence of coarse mode natural aerosols, sea salt and mineral dust.
In the past two decades, much effort has been devoted to the development of methods for the calculation of aerosol properties that are difficult to measure. These properties include the aerosol phase composition (i.e., solid or liquid) and the aerosolassociated water mass. Most studies have focused on the dominant inorganic aerosol compounds such as sulfate, ammonium, nitrate, and aerosol water (Metzger et al., 2002b). These compounds partition between the liquid-solid aerosol phase and gasphase aerosol precursor gases such as HNO 3 , and NH 3 . Numerous inorganic thermodynamic models have been developed to represent these processes. Some of them used the box model to estimate gas/ aerosol partitioning (e.g., Wexler and Seinfeld, 1991;Kim et al., 1993a, b;Kim and Seinfeld, 1995;Meng and Seinfeld, 1996;Nenes et al., 1998;Clegg et al., 1998a, b;Jacobson et al., 1996;Jacobson, 1999;Meng et al., 1998;Sun andWexler, 1998, Pilinis et al., 2000;Trebs et al., 2005;Metzger et al., 2006), and some of them were implemented to simplify thermodynamic equilibrium model in global CTM model to simulate aerosol distributions (Metzger et al., 2002a, b;Rodriguez and Dabdub, 2004;EMEP, 2003;Lauer et al., 2005;Tsigaridis et al., 2006). The differences between our model and these models are the meteorological data, resolution, emission, transport, deposition, chemistry, etc. For example, Metzger et al. (2002b) used European Center for Medium-range Weather Forecasts (ECMWF), resolution is 2.5 Â 2.5; and Rodriguez and Dabdub (2004) used monthly mean meteorological data with resolution 5 Â 5. The numerical advection in our model is calculated by secondorder moments method (Prather, 1986). As we show below, these processes, such as meteorology, chemistry, etc., are important for self-consistent treatment of biogeochemical air-surface exchanges, e.g., N deposition. This paper is organized into four sections. Section 2 briefly describes the models used in this study. Section 3 compares our climatological predictions to observations and presents our sensitivity studies. Section 4 summarizes the study.

Global chemistry transport model
This study use the UC Irvine global chemistry transport model (UCICTM) (Prather et al., 1987;Jacob et al., 1997;Olsen et al., 2000;Bian et al., 2003;Bian and Zender, 2003) with an embedded aerosol equilibrium model (Metzger et al., 2002a). The UCICTM includes an O 3 -NO x -NMHC-SO 2 chemical scheme with 48 species, 95 chemical kinetic reactions, 22 photolytic reactions, and 9 aqueous reactions upgraded with ammonia chemistry, dust and sea-salt modules. Wild and Akimoto, 2001;Bian, 2001). Trace gas emissions are based on the AeroCom emissions Inventory Activity database (http://www.mnp.nl/edgar/ model/edgarv32/acidifying). A first-order rainout parameterization for soluble gases and particles is used for large-scale precipitation (Giorgi and Chameides, 1986). Scavenging of aerosol by convective precipitation is computed in the model as part of the convective mass transport operator (Bian, 2001); air pumped in wet convective updrafts loses a fraction of its aerosol to deposition before dispersing at the top of the updraft. We adopt here a 50% aerosol scavenging efficiency in shallow wet convection (extending up to $2600 m altitude) and a 100% scavenging efficiency in deep wet convection (Balkanski et al., 1993). Dry deposition of gases and aerosols is calculated with a resistance-in-series scheme (Wesely and Hicks, 1977), and gravity settling deposition for large dust and sea-salt particles. The numerical solution for advection and convection conserved the second-order moments of tracer distribution (i.e., quadratics plus cross terms). The meteorological fields used in this study are from the Goddard Institute for Space Study (GISS) general circulation model version II that is run with a resolution of 41 latitude by 51 longitude, and 9 vertical levels. These meteorological fields include 3-D (winds, temperature, water vapor, clouds, and convection) and 2-D (boundary layer properties) data at 3-h averages. The scope of this study is focused on the impacts of chemical transformation on gas-aerosol partitioning and distribution with a single climatological meteorological field. We did not explore the impacts of different meteorological fields and model resolution on simulations. The impacts of different meteorological fields and different resolutions were explored using same model by GISS GCM 9 levels (GISS9) and GISS GCM 23 levels (GISS23) and European Center for Medium-range Weather Forecasts (EC21) (Bian, 2001). In her experiments, the vertical resolution did impact Rn and Pb concentrations especially over land source regions. Overall using the meteorological fields with two different vertical resolutions gave similar conclusions for Rn's spatial and temporal distributions (Bian, 2001). The relative biases of simulated Rn column by GISS9 meteorological field are around 7-15% globally (Bian, 2001). Even Rn simulations by GISS9 meteorological field does not do well seasonal cycle over most land stations as simulated by GISS23 and EC21 meteorological fields, but the seasonal cycle simulated by GISS9 meteorological field are still reasonable at most sites (Bian, 2001, pp. 138, 160, 171-176).
Aerosols are included here in the calculation of photolysis rate using the multiple-scattering Fast-J scheme Bian and Prather, 2002), which explicitly accounts for aerosol and cloud optical properties. In each CTM layer the monthly mean aerosol extinction is combined with the 3-h cloud optical depths from the meteorological fields. Fast-J is computationally efficient, and the radiation field as a function of wavelength is calculated hourly throughout the entire column. Bian and Zender (2003) document the seasonal and regional roles of aerosol-influenced photolysis on important atmospheric oxidants. Instead of a prescribe tropopause (used to diagnose where tropospheric versus stratospheric chemistry was calculated) and an upper boundary flux O 3 , the CTM model dynamically diagnoses the tropopause by using an on-line, ozone-like tracer (Syn-O 3 ) with an effective source of 475 T g yr À1 in the highest level of the model and was removed at surface (McLinden et al., 2000;Hannegan et al., 1998). The model has been applied previously to simulations of both tropospheric and stratospheric chemistry and transport (Prather et al., 1987;Hall and Prather, 1993;Avallone and Prather, 1997;Jacob et al., 1997;Hannegan et al., 1998;Hsu et al., 2003;Olsen et al., 2000;McLinden et al., 2000McLinden et al., , 2003Bian, 2001;Bian et al., 2003;Wild et al., 2003Wild et al., , 2004Wild et al., , 2000. The tropospheric model has been evaluated in several publications: tropospheric O 3 and CO, NO x /NO y at Mauna Loa, and global peroxyacetylnitrate (PAN) profiles of Wild and Prather (2000), and Wild and Akimoto (2001), further O 3 and CO evaluations of IPCC 2001 (Prather and Ehhalt, 2001), and updated radon and lead simulations of Bian (2001).

Emissions
We map the AeroCom emission inventories (SO x , NO x ) to the model grid, preserving the second-order moments of the emissions. Additional emissions for CO and biomass burning sources are from Wang et al. (1998a). NO from lightning is based on the parameterization of Price and Rind (1992). A NO source from aircraft is also included (Baughcum et al., 1996). The ammonia cycle was calculated by adding gas-phase ammonia (NH 3 ) and aerosol ammonium (NH 4 ). Recent GEIA ammonia emissions inventory was used in the model (Bouwman et al., 1997). The total ammonia source was estimated to be 53.6 T g N yr À1 . And most of them are from domesticated animals (43.3%), and fertilizers (16.8%).

Mineral dust
The mineral aerosols sources are calculated using Dust Entrainment And Deposition . This mobilization scheme is based on the wind tunnel and in situ studies of Iversen and White (1982), Marticorena and Bergametti (1995), Gillette et al. (1998), andFecan et al. (1999). It is similar to those used in Tegen and Fung (1994), Mahowald et al. (1999), Ginoux et al. (2001), and Tegen et al. (2002) in that it is based on a wind threshold velocity and has a wind speed cubed relationship for Within each bin we assume log-normal distribution in aerosol sizes .

Sea salt
The dominant mechanism for sea-salt production over the open ocean is believed to be air bubbles bursting during whitecap formations (Blanchard and Woodcock, 1980). Sea spray is generated by the wind stress on the ocean surface. Air bubbles, which constitute the whitecaps resulting from breaking waves, burst at the water surface and produce small droplets by means of two mechanisms. Film drops are produced when the thin liquid film that separates the air within a bubble from the atmosphere ruptures. The remaining surface energy of the bubble, after bursting, results in a liquid jet that becomes unstable and breaks into a number of jet drops (Smith et al., 1993). The formation of film and jet drops is called the indirect mechanism. At wind speeds greater than 10-12 m s À1 , spume drops torn directly from the wave crests by the strong turbulence make an increasing contribution to the sea salt and dominate the concentration at larger particle sizes. The formation of spume drops is called the direct mechanism. We prescribe production of sea-salt particles based on the Monahan et al. (1986) and Smith et al. (1993) empirical parameterization of laboratory experiments for both mechanisms.

Heterogeneous reactions module
Observations continue to highlight the importance of heterogeneous reactions on aerosol surface. The reactions of N 2 O 5 and/or NO 3 on wet aerosol surfaces are likely to be responsible for the observed destruction reactions Based on the observations and models, we include SO 2 uptake on dust aerosol surface and form into sulfate, and NO 3 and N 2 O 5 uptake on dust surface and form into nitrate (Zhang and Carmichael, 1999), and N 2 O 5 uptake on aerosols (sulfate, nitrate, ammonium, and sea salt) form to HNO 3 (Dentener and Crutzen, 1993;Dentener et al., 1996;Bian and Zender, 2003). The heterogeneous reactions and uptake coefficients used in our model are same as Bian and Zender (2003). Uncertainties in uptake coefficients are large, up to three orders of magnitude for certain species (Michel et al., 2002); Underwood et al., 2001;Zhang and Carmichael, 1999;Bian and Zender, 2003). For example, recent studies report 2.0 Â 10 À5 ogo1.6 Â 10 À2 for HNO 3 (Goodman et al., 2000;Underwood et al., 2001). We apply the values in globally so that regional differences in uptake coefficients due to dust mineralogy and RH are neglected. Our results on heterogeneous uptake on aerosol for some species should be considered an upper bound since some of these species would be lost to heterogeneous reactions on other aerosol types.
Our model includes the optical effects of BC/OC species on photolysis but neglects heterogeneous chemistry on BC/OC. Heterogeneous chemistry of carbonaceous particles is complex (Seinfeld and Pandis, 1996, p. 708) and currently beyond the UCICTM capabilities. Since the uptake coefficients are highly uncertain, it is difficult to assess the impact of neglecting these reactions. But impacts of ignoring heterogeneous reactions on BC/OC could be important in high BC/OC emission areas, such as East Asia, South Africa and South America.

Aerosol equilibrium thermodynamics module
This study uses the Equilibrium Simplified Aerosol model (EQSAM) model (Metzger, 2000;Metzger et al., 2002aMetzger et al., , b, 2006. The module includes the ammonium-sulfate-nitrate-water and major crustal elements (Ca 2+ , Mg 2+ , K + ) and sea salt (Na + and Cl À ). EQSAM assumes that aerosols are internally mixed and obey thermodynamic gas/aerosol equilibrium. These assumptions are accurate under most atmospheric conditions considering the 1 h time steps used in UCICTM. The basic concept of EQSM is that the activities of atmospheric aerosols in equilibrium with the ambient air are governed by relative humidity (RH). Since the water activity is fixed by RH, the solute activity is, for a given aerosol composition, a function of RH; the molality depends on the water mass, which solely depends on RH. This is also approximately true for activity coefficients of salt solutes of binary and multicomponents solutions. Using the ''domain structure'' (Metzger et al., 2002a), and

A u t h o r ' s p e r s o n a l c o p y
taking into account that gas/aerosol equilibrium is only valid for certain domains where sulfate is completely neutralized, we can noniteratively calculate the aerosol composition, including aerosolassociated water. The equilibrium assumption further implies that the water activity (aw) of an aqueous aerosol particle is equal to the ambient relative humidity (RH), i.e., aw ¼ RH (Bassett and Seinfeld, 1983) which the UCICTM supplies. This is valid for atmospheric applications, since the ambient relative humidity is not influenced by the small water uptake of aerosol particles. Because sulfuric acid has a very low vapor pressure, it is assumed that it resides completely in the aerosol phase.
Certain salts, such as ammonium sulfate or ammonium nitrate, deliquesce if the relative humidity reaches a threshold value; below the value these salts may be crystalline. Deliquescence of various salt compounds is determined in EQSAM in the corresponding subdomains (Metzger, 2000;Metzger et al., 2002a). The deliquescence of salt aerosol depends on the ambient RH and temperature. For partitioning between the gas/liquid/solid aerosol phases, chemical equilibrium is determined by the temperature-dependent equilibrium constant.

Measurements
Comparisons of simulated total aerosol mass concentrations for sulfate, nitrate and ammonium with measurements from two different available databases are presented in this section. Aerosol observations made from 1987 to 1999 at more than 70 monitoring stations across North America are available from CASTNET website (http://www. epa.gov/castnet). In addition, a comparable dataset that corresponds to European measurements was obtained from EMEP (http://www.nilu.no/projects/ ccc/emepdata.html). EMEP reported measurements are obtained from annual averages that span from 1978 to 2000. The measurements at Bermuda, Oahu, Okinawa, and Cheju (University of Miami observation network) are used for the model evaluation. Model results represent atmospheric concentrations of species during a typical year of the 1990s. Therefore arithmetic means from CAST-NET and EMEP datasets and data from University of Miami observation network are used for comparison with model outputs instead of data from any specific year.

Global aerosol distribution
The following section analyzes the predictions of the UCICTM coupled with EQSAM. In addition, evaluation of the coupled model against available data at ground-based stations in North America and Europe is provided to assess model performance. Annually averaged concentrations at the surface level for major species simulated with the coupled UCICTM-EQSAM are shown in Fig. 1. The aerosol distributions are similar to the distribution of the aerosol precursors. High SO 2 and HNO 3 are over industrialized areas, such as East Asia, Europe, and North America. In general, the model reproduces well-known features of secondary aerosol distributions such as high sulfate, nitrate and ammonium over industrialized regions (Benkovitz et al., 1996;Chin et al., 1996;Rodriguez and Dabdub, 2004;Metzger et al., 2002b). For instance, annual average concentrations of sulfate, nitrate, and ammonium are high over industrialized regions, such as the east coast of China, central Asia, Europe and North America, consistent with high emissions in these regions. Their values could reach 6-15 mg m À3 for sulfate, 4-9 mg m À3 for nitrate, and 4-6 mg m À3 for ammonium. Dust concentrations are distributed over North Africa, Arabian peninsular, East Asia, and Australia, which is similar with the previously calculations Luo et al., 2003;Mahowald et al., 2003;Ginoux et al., 2001) (Fig. 1). Model underpredicts dust concentration over East Asia. Sea-salt concentrations are high over the middle latitude of Southern and Northern Oceans and Arctic regions (Fig. 1).
It is important to consider the fidelity of the model performance against direct station measurements of aerosol concentration. The root-mean square (RMS) absolute error RMS abs is computed as where x i are observed data and y i are modeled data. The RMS abs is a strict measure of absolute model bias against the observed aerosol concentration. The stations with greatest absolute concentrations dominate RMS abs when it is computed using untransformed (linear) data. We also computed RMS abs with logarithmically transformed data, i.e., (1), and we compared the two error estimates. The differences were small, so we only include RMS abs using untransformed (linear) data in Table 3. The final statistic we examined is the relative root-mean square bias, RMS rel . RMS rel is computed from the relative, rather than absolute, bias for each experiment The relative root-mean square biases are included in Table 3. The mean relative bias (MRB) is calculated and are included in Table 3 in order to show model simulation is in error for each experiment.
CASTNET and EMEP measurements of SO 2 , HNO 3 , and NH 3 (CASTNET does not measure NH 3 ) generally agree well with UCICTM predictions (Fig. 2). Linear correlation coefficients between observations and simulations of SO 2 and HNO 3 are over 0.69 and 0.67 in North America and 0.46, 0.54 in Europe, respectively. The correlation coefficients between observations and control simulations (CTL) are 0.81, 0.41, and 0.74 for sulfate, nitrate and ammonium, respectively in North America and 0.45, 0.67, and 0.46, respectively in Europe (Fig. 3, and Table 2). Both our UCICTM climatological results and previous study from a completely different CTM (IMAGE) and aerosol equilibrium model (SCAPE2) (Rodriguez and Donald, 2004), show that comparisons are better at CASTNET than at EMEP.
The discrepancy between the UCICTM and observations is larger over the European EMEP sites than the North American CASTNET sites (Fig. 3, also see Fig. 5a-5c and Tables 2 and 3), similar to the bias in Rodriguez and Dabdub (2004). Similar as our results, Rodriguez's comparisons show that data dispersion is larger over sits of EMEP (Europe) (Rodriguez and Dabdub, 2004). It could suggest that this discrepancy could not be caused by meteorological data, since we used different meteorological datasets. This could be due to emission biases or model deficiencies, e.g. due to missing aerosol compounds in EQSAM. The uncertainties of the combined sampling and chemi-cal analysis of EMEP measurements are range from 15% to 20% (http://www.nilu.no/projects/ ccc/ga/index.htm). The uncertainties of CASTNET measurements are around 5% for sulfate and 10-12% for nitrate and ammonium (http://cfpub. epa.gov/gdm/index.cfm). That could be one of reasons why model comparisons are better at CASTNET sites than at EMEP sites. To further evaluate model performance, we compare the correlations of species of sulfate, nitrate, and ammonia of observations and simulations to show the degree to which these chemical cycles are coupled ( Fig. 4 and Table 1).
Sulfate and ammonium are very tightly correlated in both observations ðr40:8Þ and in the UCICTM ðr40:9Þ. The observed correlation coefficient of SO 4 and NH 4 is over 0.9 at CASTNET (North American) sites, and falls to 0.81 including all CASTNET and EMEP sites. It suggests that variability of (NH 4 )SO 4 or/and NH 4 HSO 4 explain 80% of the variability of sulfate. There is high correlation between nitrate and ammonium (0.71 for observation and 0.57 for model simulation). This suggests roughly half of the variability of nitrate can be explained by NH 4 NO 3 variability. The small observed and modeled correlations between sulfate and nitrate (r ¼ 0:52 and 0.49, respectively) reflect the fact that sulfate and nitrate precursor sources are distinct in space and time. Although the correlations are high for observations and model simulations, but the slopes are different. One reason could be that all observation data were collected in continents, but model data include continent and ocean. The partitioning could be different between continent and ocean, since anion and cation abundance are quit different over continent and over ocean. In Fig. 4, we highlight corresponding grids where observations are available for model simulations.

Seasonal variation
Globally the largest sulfate concentration is persistently presented over East China year around. Observed SO 4 concentrations did not show stronger seasonal variation over European. Sulfate shows East China peak in winter (not shown here), while North American SO 4 peaks in summer (Fig. 5a). China uses more coal for heating in winter, which results in maximal winter SO 4 . We calculate continental mean seasonal cycle of sulfate that show the mean stations and model prediction for

A u t h o r ' s p e r s o n a l c o p y
Europe and North America (left and right top panels in Fig. 5a). Continental mean seasonal cycles also show that sulfate did not show strong seasonal variation in Europe and sulfate shows peak in summer in North America. Seasonal variations of NO 3 and SO 4 are the same in East China. In North America, the seasonal variation is much stronger for NO 3 than SO 4 and peaks in winter. This is due to the greater reliance in North America on gas and oil for winter heating.
The ammonium seasonal variation is similar to the sulfate seasonal variation, highest in winter

A u t h o r ' s p e r s o n a l c o p y
cycle at some sites, but not all sites. Comparison of Rn simulated with different meteorological fields (ECMWF, GISS21, and GISS9) shows that the GISS9 meteorology used here is most biased in the seasonal cycle over land stations (Bian, 2001, pp. 171-176). The lack of winter sulfate over North

A u t h o r ' s p e r s o n a l c o p y
America could be due to a large reduction of sulfuric acid production. In East Asia and especially China, sulfate may be higher in winter due to fossil fuel burning.
Regional variations in seasonal emission patterns have significant impacts on aerosol production and partitioning. Low nitrate in summer could be caused by nitric acid and nitrate are more effectively removed in summer than in winter. Ammonium concentrations are higher in winter at some sites and higher in summer at other sites, depending on NH 3 , HNO 3 , H 2 SO 4 and meteorological parameters. The model calculated higher sulfate and ammonium in January at EMEP sites ( Fig. 5a and c) could be caused by uncertainties in meteorological data, since dynamic fields tend to have similar impacts on different species (sulfate and ammonium). Again the model biases are smaller at CASTNET (North America) sites than at EMEP (Europe) sites (Fig. 5a-c). As discussed above, these discrepancies between sites in America and Europe may not be caused by meteorological data and transport. Biases can also be due to emissions, deposition, or chemical mechanisms.

Sensitivity analysis
We now show the results of three sensitivity experiments to assess important aerosol formation processes. The standard (CTL) model described above includes the thermodynamic aerosol equilibrium model and ammonia chemistry interacting with coarse model natural aerosols (dust and sea salt). The first sensitivity experiment (XPT1) turns off thermodynamic equilibrium model. The second sensitivity experiment (XPT2) turns off ammonia chemistry (XPT2), and the third sensitivity experiment (XPT3) omits dust and sea-salt aerosol.
Annual mean correlation coefficients of SO 4 , NO 3 , and NH 4 between observations and model simulations at EMEP and CASTNET sites worsen when inorganic aerosol thermodynamic partitioning is not applied especially for EMEP measurements (compare Figs. 3 and 6, Table 2). The EQSAM aerosol module improves correlations between SO 4 simulations and observations at both EMEP and CASTNET sites (Fig. 6, summarized in Table 2). EQSAM improves nitrate and ammonium simulations at CASTNET sites. However, accounting for inorganic aerosol partitioning produces little difference between CTL and XPT1 nitrate and ammonium simulations at EMEP sites (compare Fig. 6 to Fig. 3, summarized in Table 2) and it is unclear for the reason of this performance. The RMS and relative biases are strict measures of absolute model bias against the observed aerosol concentration. The regress parameters and model mean relative biases and RMS abs and RMS rel are summary in Tables 2 and 3. From Table 3, we can see that the root-mean square bias and mean relative bias of CTL are smaller than XPT1 experiment.
Experiments also show that H 2 SO 4 /H 2 O system dose not nucleate easily but NH 3 /H 2 SO 4 /H 2 O system does (Coffman and Hegg, 1995). Our results demonstrate that ammonia chemistry is clearly very important for aerosol formation and partitioning (compare Figs. 3 with 7, Table 2). This is consistent with Fig. 4, which shows correlations of sulfate and nitrate with ammonium range from 0.71 to 0.81 (observed) and 0.57 to 0.93 (simulated). Table 3 also shows bias of CTL is much smaller than bias of XPT2, which is the largest in CTL and all experiments runs. The positive mean relative biases for XPT2 in Table 3 reflect that the model calculated sulfate and nitrate are much smaller than observations, which is consistent with Fig. 7.
Coarse mode aerosols such as dust and sea salt provide ample surface area and liquid volume for significant heterogeneous chemistry (e.g., Dentener et al., 1996;Bian and Zender, 2003), also are the sources of crustal elements such as Ca 2+ , Mg 2+ , K + , Na + , and Cl À , which are important for aerosol partitioning (Kim et al., 1993a, b;Kim and Seinfeld, 1995;Meng and Seinfeld, 1996;Nenes et al., 1998).  Hence removing dust and sea salt (XPT3) test the effect of these coarse aerosols on sulfate, nitrate and ammonium aerosol formation. Fig. 8 shows that scatter plots of observation and XPT3 experiment. It can be seen that the correlations of XPT3 are almost similar to CTL simulation at EMEP sites, and worse than CTL simulation at CASTNET sites.

ARTICLE IN PRESS
Since sulfate proportional to dust and SO 2 is because SO 2 uptake on dust aerosol surface and form to sulfate by heterogeneous reactions Fig. 6. Sulfate, nitrate and ammonium comparisons with EMEP and CASTNET observations for experiment XPT1 (without thermodynamic equilibrium module). Compare to Fig. 3, statistics parameters are same as Fig. 2, but for XPT1.

A u t h o r ' s p e r s o n a l c o p y
included in our model. But other species such as NO 3 , and N 2 O 5 are not behavior same way as SO 2 in the heterogeneous reactions. For example, nitrate can be produced by NO 3 heterogeneous reactions on the aerosol surface, but NO 3 and N 2 O 5 are not necessarily removed as nitrate, but can also be lost from the aerosol as HNO 3 , and this decreases nitrate production. That may be the reason why the correlations of other species are same or better in XPT3 experiment. The presence of dust and sea salt (CTL) significantly increases sulfate production and concentration by 10-70% in polluted continental regions such as East Asia, Europe, and part of America (see panel 1 of Fig. 9). This is because SO 2 uptake on dust aerosol surface to form sulfate via heterogeneous reactions . The sulfate formation by heterogeneous reactions is proportional to dust and SO 2 concentrations and so sulfate concentration are higher in industrial areas, such as East Asia, Europe and North America. Sulfate increases less from 0 to 60 s than in the northern hemisphere, consistent with inter-hemispheric SO 2 gradient. From Fig. 5a, we can see that the sulfate decrease almost at every sites without dust and sea salt (XPT3). Nitrate formation decreases in polluted industrial regions, and increases over most ocean regions when dust and sea salt are present (Panel 2 of Fig. 9). More HNO 3 loss on the surface of dust and sea-salt aerosols in high pollution areas, and nitrate decreases in these areas could be due to less available HNO 3 to form nitrate, according to Eqs. (1) and (2). Over ocean regions, N 2 O 5 uptake may dominate nitrate formation.
Dust and sea salt increase the ammonium concentration in polluted regions and decreases ammonium over most ocean regions and in the southern hemisphere (Fig. 9, panel 3). This reduction over ocean is partly explained by the presence of cation (NaCl, MgCl 2 , CaCO 3 ) tied to sea salt and dust. The cation abundance makes it difficult for NH 3 to partition into the aerosol phase. In polluted regions, anion abundance (such as SO 4 , NO 3 ) allows NH 3 to easily partition into the aerosol phase.
Generally from previous validations, our simulations show model did the reasonable works, model captures seasonal cycles at some sites but not all sites, same as Rn simulations using same model and meteorological data (Bian, 2001). The biases calculated by model simulations and observation at EMEP and CASTNET are shown in Table 3. It can be seen that biases of sulfate and nitrate are the smallest for control run, and then XPT1, XPT3, and XPT2. The biases of ammonium are the smallest for XPT3 experiment. As mentioned earlier, our model includes optical effects of BC/OC species on photolysis and neglects BC/OC heterogeneous chemistry. We expect any biases due to neglecting BC/OC heterogeneous chemistry to be largest in and near BC/OC emissions areas.

. Deposition comparison
Deposition fluxes of SO 4 , NO 3 , and NH 4 aerosols are important for biogeochemistry cycles. Fig. 10 shows the distributions of total predicted sulfate, nitrate and ammonium deposition (dry plus wet). High deposition rates occur in and downwind of industrial areas such as East Asia, Europe and North America. We compare depositions of sulfate, nitrate and ammonium against CASTNET-inferred deposition, which CASTNET calculates using observed concentrations and deposition velocity (Fig. 11). It shows that model underpredicts depositions of sulfate, nitrate and ammonium, and are consistent of aerosol concentrations comparisons. Table 4 compares SO 4 , NO 3 , NH 4 , NH 3 and HNO 3 of our simulations with previous studies, and it shows us our results are comparable with previous works. Deposition of ammonia is comparable with ammonium deposition. Ammonia wet deposition is at same magnitude as ammonium, but ammonia dry deposition can be 3-4 times of ammonium deposition. That is why the ammonium deposition is about half of total ammonia emission. Additionally we compare SO 4 , NO 3 , NH 4 , NH 3 and HNO 3 deposition between present and pre-industrial, and year 2100 scenarios. For the pre-industrial simulation, we turn off all anthropogenic emissions. Year 2100 emissions follow the B1 and A1B Intergovernmental Panel on Climate Change (IPCC) scenarios (IPCC Special Report on Emissions Scenarios, http:// www.grida.n0/climate/ipcc/emission). The meteorological fields are present-day, i.e., the same as the control (CTL), in both pre-industrial and Year 2100 simulations. We find that the present deposition of NH 4 and NO 3 are twice pre-industrial deposition and present SO 4 deposition is almost five times preindustrial deposition.   averaged sulfate falls within a factor of two of the observations at about 95% of CASTNET sites. The discrepancy between model and observation is larger for European sites. This discrepancy may be independent of meteorological fields, since simulations by Rodriguez and Dabdub (2004)     Moreover, with ammonia chemistry, dust and sea salt significantly increases sulfate production and concentrations in polluted regions. In all regions and seasons, representation of ammonia chemistry is required to obtain reasonable agreement between modeled and observed sulfate and nitrate concentrations. Observed and modeled correlations of sulfate and nitrate with ammonium confirm that the sulfate and nitrate are strongly coupled with ammonium. Dust and sea salt increase nitrate concentrations over ocean regions where N 2 O 5 uptake leads to increased nitrate formation. Dust and sea salt reduce nitrate concentrations over industrial and very dusty regions (e.g., North Africa) since HNO 3 is lost on dust surface, and less HNO 3 is available to form nitrate in polluted regions. Dust and sea salt increase ammonium formation in industrial regions and in very dusty regions, and they reduce ammonium over remote oceans. This reduction over ocean is partly explained by the presence of cation (NaCl, MgCl 2 , CaCO 3 ) tied to sea salt and dust. The cation abundance makes it difficult for NH 3 to partition into the aerosol phase. In polluted regions, anion abundance (such as SO 4 , NO 3 ) allows NH 3 to easily partition into the aerosol phase. Our model neglects heterogeneous chemistry on BC/OC heterogeneous chemistry. We expect the largest biases due to neglecting heterogeneous reactions on BC/OC occur in high BC/OC emission areas, such as East Asia, South Africa, and South America.

ARTICLE IN PRESS
Consistent with previous studies, we find high deposition rates of sulfate, nitrate, and ammonium occur in and downwind of industrial areas such as East Asia, Europe and North America. Additionally, we find that present deposition of NH 4 and NO 3 are twice pre-industrial deposition, and that present SO 4 deposition is almost five times pre-industrial deposition. We are currently using these atmospheric forcings to quantify anthropogenic impacts on the oceanic nitrogen and sulfur cycles.