Fine Scale Modeling of Agricultural Air Quality over the Southeastern United States Using Two Air Quality Models . Part II . Sensitivity Studies and Policy Implications

Sensitivity simulations using CMAQ at various grid resolutions are evaluated. Compared with the simulations at 12and 4-km, the 1.33-km simulation shows large improvement in most meteorological predictions in July and some chemical predictions in January and July 2002. Limited improvements at 1.33-km and 4-km are attributed to current limitations in meteorological parameterizations and lack of accurate data for land use and emissions at a fine scale. NH3 plays an important role in PM2.5 formation, but the emission control strategies focus only on SO2 and NOx in the southeastern U.S. To understand the impact of NH3, NH3 to NH4 conversion and the chemical regimes of PM2.5 formation are examined. The conversion rates of NH3 to NH4 from CMAQ and CAMx simulations are 10–60% in January and 10–50% in July at and near major sources. The eastern North Carolina and northeastern Georgia are NH3-rich and the remaining areas are NH3-neutral in both months. To further assess the impact of NH3 emission reductions, the sensitivity of CMAQ to emission reductions is evaluated for four emission scenarios: reducing emissions of SO2, NOx, agricultural livestock-NH3 (AL-NH3) by 50%, respectively and collectively. The largest reductions of PM2.5 are by up to 19.2% in January and 18.3% in July when all these emissions are reduced by 50%. AL-NH3 reductions result in the largest decrease in January by up to 16%, dominated by a reduction in NH4NO3, while SO2 reductions result in the largest decrease in July (up to 11%) due to decreases in NH4 and SO4. This indicates that reducing AL-NH3 emissions together with SO2 and NOx emissions can reduce PM2.5 concentrations more than reducing emissions of SO2 and NOx alone, particularly in winter. Future emission control strategies for PM2.5 controlling should consider the reduction of NH3 emissions, in addition to the emissions of SO2 and NOx.


INTRODUCTION
As illustrated in Part I paper, air quality predictions show sensitivity to different air quality models due to their different model treatments.The model predictions are also sensitive to model configurations such as horizontal grid resolution and inputs such as emissions and boundary conditions.Mass et al. (2002) reported that the use of a finer grid resolution showed some improvement for some events (e.g., strong forced convection, diurnal circulations, and heavy precipitation).Queen and Zhang (2008) found that the simulation at a fine grid resolution of 4-km better captured the mesoscale convection thus predicted more accurate precipitation and wet deposition of species in summer than the simulations at 12-or 36-km grid resolutions.They also showed that the 12-km simulation performed the best for precipitation but the worst for wet deposition of PM 2.5 species in winter, implying that wet deposition is more sensitive to other factors (e.g., emissions) in winter than in summer.Wu et al. (2008a) reported that meteorological variables were less sensitive to the horizontal grid resolution than chemical variables, and that some species (i.e., ammonium (NH 4 + ) and nitrate (NO 3 -)) showed more sensitivity to grid resolution in winter than in summer.Some studies have shown that a coarser grid resolution provides similar or even better air quality predictions than a finer grid resolution (Mathur et al., 2005;Arunachalam et al., 2006;Cohan et al., 2006;Zhang et al., 2006;Queen and Zhang, 2008;Liu et al., 2010).The poorer model performance at a finer grid resolution can be attributed to inaccuracies or uncertainties in the required inputs (e.g., meteorology, emissions, land use) due to the limitation of current meteorological models in capturing fine-scale atmospheric processes and the lack of information (e.g., emissions and land use data) at a finer grid scale.The gridaveraging of emissions and land use data can influence model predictions.With a coarser grid resolution, the emissions are diffused into a larger grid cell instead of a smaller grid cell.For example, Mathur et al. (2005) reported a wider O 3 plume in central North Carolina (NC) using 36and 12-km horizontal grid resolutions than with a 4-km horizontal grid resolution, resulting in an overprediction of O 3 mixing ratios at most monitoring sites in the area.Cohan et al. (2006) also reported a premature diffusion of emissions in a coarser grid resolution and found that VOCsensitive regions showed more dependence on grid scale than NO x -sensitive regions.
The understanding of key air pollutant formation mechanisms and associated seasonal and regional variations is very important in developing effective region-specific emission control strategies.While O 3 and PM 2.5 share the common set of precursors, their formation mechanisms vary across the domain of interest because of different dependences on emission sources and meteorological conditions and other unique precursors (e.g., CO and CH 4 are precursors for O 3 only and NH 3 is a precursor for PM 2.5 only).Such a regionspecific variability should be included in the consideration of the emission control strategy design to prevent unwanted results.The formation of these pollutants is often non-linear, i.e., reducing the emissions of a precursor species may cause no change or even increase in the concentrations of secondary pollutants.For example, O 3 formation is NO xlimited in the southeastern U.S. because of the large biogenic volatile organic compounds (VOCs) emissions (Chameides and Cowling, 1995;Liao et al., 2007;Zhang et al., 2009a).As such, controlling NO x can reduce O 3 but controlling VOCs may have little impact on O 3 or even increase O 3 in this region.The formation of secondary PM 2.5 can be limited by one or more of its precursors including SO 2 , NO x , VOCs, and NH 3 (Ansari and Pandis, 1998;Tsimpidi et al., 2007;Pinder et al., 2007Pinder et al., , 2008;;Zhang et al., 2009b).In addition, regional transport of air pollutants plays an important role in local ambient air quality.Baker (2004) reported that decreases in NO x levels and NH 3 had more of a localized impact, whereas decreases in SO 2 levels affected PM 2.5 predictions over a larger region.
Current air pollution control strategies focus on the reduction of emissions of SO 2 , NO x , and VOCs, although NH 3 is also an important precursor of PM 2.5 .PM 2.5 and its secondary components such as sulfate (SO 4 2-), NO 3 -, NH 4 + , and secondary organic aerosol (SOA) respond differently to emission reductions, depending on chemical regimes (e.g., sulfate-rich (or NH 3 -poor), neutral (or NH 3 -neutral), or poor (or NH 3 -rich)) and atmospheric conditions (e.g., cold vs. warm, or dry vs. wet).Baker (2004) reported a decrease in NO 3 -when NO x and NH 3 were reduced by 30% and a decrease in SO 4 2-when SO 2 was decreased by 30%.Wu et al. (2008b) showed that removing NH 3 emissions from agricultural livestock (AL-NH 3 ) resulted in a significant reduction in NH 4 + and NO 3 -and had little impact on SO 4 2over NC.Reducing VOCs in the eastern U.S. can reduce organic matter (OM), but increase SO 4 2-and NO 3 - (Pun et al., 2008).Changes in PM 2.5 are more sensitive to NO x reductions in winter and SO 2 reductions in summer (Baker, 2004;Liao et al., 2007;Pinder et al., 2007;Tsimpidi et al., 2007).The gas ratio (GR) has been used to identify NH 3rich and NH 3 -poor regions (Ansari and Pandis, 1998): where TA, TS, and TN are the molar concentrations of total ammonia (NH 3 + NH 4 + ) (also called reduced nitrogen, NH x ), total sulfate, and total nitrate (NO 3 -+ HNO 3 ).GR > 1, 0-1, < 0 indicate NH 3 -rich, neutral, and poor regimes, respectively.The calculation of GR assumes that SO 4 2-is fully neutralized by NH 4 + to form (NH 4 ) 2 SO 4 which may not always be the case (i.e., in winter) (Pinder et al., 2008;Wu et al., 2008b).An adjusted GR (AdjGR) is defined by replacing the value of 2 in Eq. ( 1) with the degree of sulfate neutralization (DSN) to address this limitation (Pinder et al., 2008): where DSN is defined as When the amount of NH 4 + is sufficient to fully neutralize SO 4 2-, DSN = 2, Eq. ( 2) is thus the same as Eq. ( 1).The threshold values of AdjGR are the same as those of GR.In NH 3 -rich regions, where there is excess NH 3 after neutralizing SO 4 2-and NO 3 -(GR and AdjGR > 1), reducing NH 3 emissions may or may not result in change in PM 2.5 , depending on temperature, relative humidity, TN, as well as the percentage reduction in NH 3 emissions (Zhang et al., 2009b).Reducing SO 2 and/or NO x emissions in NH 3 -rich regions can lead to reduced PM 2.5 , but frees up more NH x , which can then be deposited to the surface, affecting vegetation, waterways, and soil nutrients.In NH 3 -poor or neutral regimes, reducing SO 2 will reduce SO 4 2-, but has the potential to increase NO 3 -by releasing NH 4 + to neutralize HNO 3 (Ansari and Pandis, 1998;Takahama et al., 2004).Similarly, reducing NO x will reduce NO 3 -, but has the potential to increase SO 4 2-.Reducing NH 3 can decrease NO 3 -and NH 4 + and the amount of NH x deposited to the surface.The complex interplay between secondary PM (i.e., NH 4 + , SO 4 2-, and NO 3 -) and their precursors (i.e., NH 3 , SO 2 , and NO x ) indicates the importance of the regional-specific emission control strategies for PM 2.5 .
This Part II paper examines the sensitivity of model predictions to horizontal grid resolutions and emissions of several PM precursors.The objectives are to assess model performance at different grid resolutions, identify potential areas of improvements, and to evaluate the relative effectiveness of reductions in emissions of PM precursors in reducing PM 2.5 and nitrogen deposition, in particular, whether reductions of AL-NH 3 emissions can lead to additional reduction in PM 2.5 concentrations in the southeastern U.S.

SENSITIVITY SIMULATION DESIGN
Two sets of sensitivity simulations are performed.In the first set, a MM5/CMAQ simulation is conducted at a 1.33km horizontal grid resolution over the central and eastern NC and compared with that at 4-km in Part I paper and a simulation at 12-km from the Visibility Improvement State and Tribal Association of the Southeast (VISTAS) (Morris et al., 2007) (see Fig. 1).All three simulations are evaluated against observations within the 1.33-km domain.Among all variables selected for statistical evaluation from the simulations at 4-km in Part I paper, CO, NO, NO 2 , the extinction coefficient (β ext ) and haziness index (HI) cannot be evaluated because no sites are located in the 1.33-km domain.Among the nine sites selected for temporal analysis for the simulations at 4-km in Part I paper, only five sites are within the 1.33-km simulation domain, including one urban site (Raleigh (RAL)), one rural site (Candor (CND), NC), and three sites in the eastern NC (Kinston-Lenoir (LCC), Clinton (CLT), and Jamesville (JMV)) with high NH 3 emissions.
In the second set, four MM5/CMAQ simulations are conducted at 4-km, with 50% of reduction in the emissions of an individual PM 2.5 precursor (i.e., AL-NH 3 , SO 2 , and NO x , one at a time) and combined (i.e., 50% of reduction in the emissions of all these species).The 4-km grid resolution is selected because it was suggested by the U.S. EPA for SIP modeling over areas with high gaseous precursor emissions and primary PM sources (U.S. EPA, 2007) and it is likely the grid resolution to be used for most states in near future as the computational resources at the state level increase.Tsimpidi et al. (2007) reported that lower NH 3 reductions (e.g., 25%) are less effective in reducing PM mass in regions of high NH 3 concentrations in July than higher reductions (e.g., 50%).A 50% reduction in AL-NH 3 is thus selected for the simulation over the southeastern U.S. where NH 3 concentrations are high.50% reduction in the emissions of SO 2 and NO x was derived based on differences between the VISTAS Phase II 2002 emissions and projected emissions for 2018.Table 1 shows the expected % change in the emissions of SO 2 and NO x in 2018 from six major source types including electric generating units (EGUs), non-EGU, mobile, non-road (NR) (e.g., construction equipment and farm machinery), area, and fire projected for seven VISTAS states within the 4-km simulation domain used by VISTAS (MACTEC, 2008).Among these source types, SO 2 is  expected to decrease by > 50% from EGUs, mobile, and NR and NO x is expected to decrease by > 50% from EGUs and mobile by 2018.The projected reductions in the emissions of SO 2 and NO x from non-EGUs and those of NO x from NR are much smaller.The emissions from area sources are projected to increase in all states and those from fires are projected to increase in all states except for Georgia (GA) and West Virginia (WV).50% reductions in SO 2 and NO x for all EGUs and mobile sources and 50% reductions in SO 2 for NR sources are therefore applied in the sensitivity simulations.An additional sensitivity simulation is conducted to combine 50% reductions in the emissions of SO 2 , NO x , and AL-NH 3 .

Meteorological Predictions
Table 2 summarizes the statistics for the meteorological variables over the central and eastern NC at all three grid resolutions.In January, compared with results from simulations at 12-and 4-km, the use of 1.33-km horizontal grid resolution does not improve the meteorological performance at the CASTNET or STN sites.The simulation at 12-km gives overall the best agreement with available observations.In July, the use of a 1.33-km grid resolution shows large improvement in temperature at 1.5-m (T1.5), wind speed at 10-m (WS10), and wind direction (WD10) at the CASTNET sites and precipitation (Precip) at the NADP sites, changing their NMBs from 4.1% to -0.4%, 23.5% to 16.2%, 17.8% to 10.9%, and 115.1% to 86.7%, respectively.All three simulations show overall similar temporal variations over most days for all meteorological variables in January and a better agreement with observations at 1.33-km in July (figures not shown).These biases would propagate into chemical predictions and also to some extent affect the results of the model sensitivity to emissions.

Chemical Predictions
Table 3 summarizes the statistics for chemical and visibility variables.In January, the maximum 1-hr and 8-hr O 3 mixing ratios are overpredicted by all three simulations with an overall best performance by the 4-km simulation, indicating that the initial and boundary conditions used for winter O 3 are appropriate.The higher emissions of NO x at 1.33-km coupled with less dispersion further increase the O 3 production and accumulation in comparison to the 4-km and 12-km simulations.The mixing ratios of HNO 3 and SO 2 , two important PM precursors, are overpredicted with slightly better performance for HNO 3 at 4-km and for SO 2 at 1.33-km than at 12-km, whereas another important PM precursor, NH 3 mixing ratios, are underpredicted with the best performance at 4-km.While the 12-km simulation gives slightly better agreement for PM 2.5 than those at finer grid resolutions, the performance for PM 2.5 components by the 1.33-km simulation is the best (except for total carbon (TC)).This is because more secondary PM mass is removed through wet deposition, resulting in less overprediction in NH 4 + and NO 3 -.In the meanwhile, higher emissions of SO 2 lead to higher concentrations of SO 4 2-, particularly at urban sites, reducing underpredictions.For some species (i.e., SO 4 2-, TC, PM 2.5 ), the differences in performance among the three simulations are relatively small, within 4%.Although MM5 does not show any improvement using the finer grid resolution in January, CMAQ does, indicating other factors such as the differences in emissions and wet deposition are more influential than meteorology on the sensitivity of CMAQ to horizontal grid resolution.In July, the 12-km simulation performs the best for all species except for NH 3 and TC, dry deposition fluxes of NO 3 -, and wet deposition fluxes of NH 4 + , SO 4 2- , and NO 3 -.Similar to January, more secondary PM mass is removed through wet deposition,  -that leads to insufficient ambient concentrations for dissolution and wet scavenging.The differences in performance statistics for most species among the simulations are much larger in July than in January, with a range of NMBs of more than 10%.In both January and July, the sensitivity of precipitation to horizontal grid resolutions affects the sensitivity of wet deposition, which in turn affects the ambient PM concentrations.For example, in January, the 12-km simulation performs the best in simulating precipitation and the wet deposition of NH 4 + and NO 3 -.Fig. 2 shows observed and simulated hourly O 3 mixing ratios at RAL, LCC, and CND at the three horizontal grid resolutions in July.The 12-km simulation gives higher maximum O 3 mixing ratios that are in generally better agreement with observations but also higher nighttime O 3 mixing ratios that are in worse agreement with observations at all sites.Fig. 3 shows observed and simulated PM 2.5 concentrations at RAL and LCC.In January at RAL, some improvements are found for a few days (e.g., January 5), but overall there is no much improvement in capturing the diurnal variations of hourly PM 2.5 using finer grid resolutions.
At LCC, however, much improvement is found in predicting the 24-hr average PM 2.5 concentrations at 1.33-km throughout the month.In July, all three simulations give similar predictions on most days at RAL and LCC.Fig. 4 shows observed and simulated NH 3 mixing ratios at CLT, LCC, and JMV.The 12-km simulation gives higher NH 3 mixing ratios than those at finer grid resolutions on most days at all sites, resulting in a worst overprediction at CLT in January but overall a better agreement at CLT in July and at the other two sites in both months.Fig. 5 shows hourly dry deposition fluxes of SO 2 , HNO 3 , NH 4 + , SO 4 2-, and NO 3 -in January and July at the site in CND by CMAQ at 12-, 4-, and 1.33-km grid resolutions and by the Multilayer Model (MLM) (Clarke et al., 1997).Noticeable differences are found for dry deposition fluxes of all species between the 12-km simulation and those at finer grid resolutions.For dry deposition fluxes of SO 2 and HNO 3 , all CMAQ simulations give higher values than those by MLM.The 1.33-km simulation indicates a significant improvement for HNO 3 in July, which gives the best agreement with MLM among all species.For those of NH 4 + , SO 4 2-, and NO 3 -, all these simulations fail to capture the maximum fluxes of all three species and underpredict the lowest fluxes during some periods in both months.domain.The faster conversion can be attributed to higher SO 4 2-concentrations in CAMx due to a weaker vertical mixing, a lower dry deposition flux, and different aerosol representation and microphysics treatments.The larger differences in July than in January between results from the two models can be attributed to different aqueous-phase concentrations of hydrogen peroxide (H 2 O 2 ) as a result of a weaker vertical mixing simulated by CAMx in July, which is one of the major species responsible for the conversion of SO 2 to SO 4 2-through aqueous-phase oxidation.NH 4 + is more likely to enter particles that contain SO 4 2-or NO 3 -to neutralize them.In January, both models simulate similar gas-phase concentrations of H 2 O 2 (and thus similar aqueousphase concentrations of H 2 O 2 ), resulting in similar concentrations of SO 4 2-and NH 4 + .In July, however, CAMx simulates up to 60% more H 2 O 2 in the gas-phase than CMAQ, which result in more aqueous-phase H 2 O 2 , and thus more SO 4 2-and NH 4 + through the aqueous-phase oxidation reactions.Wu et al. (2008a) reported conversion rates of 10-40% in August 2002 and 20-50% in December 2002 at or near the source using CMAQ version 4.4, which are consistent with the results in this work.Robarge et al. (2002) measured concentrations from October 1998 to September 1999 in Sampson County, NC and found that NH 4 + accounts for ~18% and ~27% of NH x in summer and winter, respectively, which are also comparable to the results in this work.
As shown in Fig. 7, there is a significant difference in AdjGR between the models in January.CMAQ gives values greater than 1 (NH 3 -rich) localized near the NH 3 sources, while CAMx indicates high AdjGR values across the majority of the domain (i.e., a larger area of NH 3 -rich than CMAQ).The higher AdjGR values are likely due to the much lower values of NO 3 -and HNO 3 in CAMx than in CMAQ, as indicated in Table 6(a) in Part I paper.The lower total nitrogen (TN) in the denominator of Eq. ( 2) results in an increase in AdjGR.In July, the spatial distributions of AdjGR by both models are similar, with a slightly larger NH 3 -rich region by CAMx than CMAQ, because of lower TN; however the differences between CMAQ and CAMx are much smaller in July than in January.The high NH 3 emissions in the eastern NC and northeastern GA result in large regions of NH 3 -rich conditions, while remaining areas are NH 3 -neutral in both January and July.This indicates that reducing NO x and SO 2 would act to reduce PM in those regions, but would also free up additional NH 3 that could be deposited closer to the sources.Thus, reducing NH 3 in addition to SO 2 and NO x could potentially provide additional environmental benefits besides reducing PM.

SENSITIVITY TO EMISSION REDUCTIONS USING CMAQ
The simulated values of AdjGR indicate that the eastern NC and northeastern GA are NH 3 -rich in both January and July.Figs. 8 and 9 show the percentage difference in O 3 and PM 2.5 , respectively, from the four sensitivity simulations in January and July.50% reduction in the emissions of SO 2 and AL-NH 3 has very little impacts on O 3 in both months.50% reduction in NO x emissions increases O 3 mixing ratios by 10-65.5% over urban areas due to a VOC-limited O 3 chemistry in January.It decreases O 3 mixing ratios by up to 6.7% over most areas except for a few urban centers such as Atlanta and Charlotte where O 3 mixing ratios increase by up to 16.7% in July, because O 3 chemistry is NO x -limited over most areas but VOC-limited over large urban centers (Zhang et al., 2009b).As expected, the simulation with combined emission reductions shows a similar impact to that with 50% reduction in NO x emissions.
The impact of these emission reductions on PM 2.5 is very different.50% reduction in the emissions of SO 2 decreases PM 2.5 concentrations over the entire domain by up to 3.9% in January and up to 11.7% in July through reducing SO 4 2concentrations.However, the released NH 4 + is consumed by NO 3 -.The increase in NO 3 -cancels out the reduced SO 4 2-, resulting in a small decrease in PM 2.5 in January.In July, there is a larger decrease in SO 4 2-but less NO 3 -available to consume NH 4 + .This results in some decreases in NH 4 + , particularly in regions of high NH 3 emissions, and a larger decrease in PM 2.5 than January throughout the domain.While SO 2 emission reductions have little impact on PM 2.5 in January, it is more effective in reducing PM 2.5 in July than the individual reductions of NO x or NH 3 , consistent with results of Tsimpidi et al. (2007) and Pinder et al. (2007).50% reduction in NO x emissions decreases PM 2.5 concentrations over most of domain by up to 5-6% in both months due to decreased NO 3 -and NH 4 + but slightly increases PM 2.5 concentrations by up to 0.9% over some areas in January due to a small increase in the concentrations of NH 4 + and SO 4 2-.These results suggest that reducing NO x by 50% alone in the southeastern U.S. does not result in a significant decrease in PM 2.5 in January or July and is not an effective control strategy.50% reduction in AL-NH 3 emissions decreases PM 2.5 concentrations over the entire domain by up to 16.2% in January and up to 7.4% in July, indicating a more important role of NH 3 in PM 2.5 formation under winter conditions.The decrease in PM 2.5 in both months is due to a decrease in both NH 4 + and NO 3 -in NH 3rich regions, consistent with results of Wu et al. (2008b).NO 3 -concentrations are lower in July than January, so the reduction of AL-NH 3 results in a smaller maximum reduction of PM 2.5 .The reduction of NH 3 and NH 4 + acts to decrease the pH of the aerosols, where the oxidation of SO 2 to SO 4 2is highly dominated by reaction with abundant H 2 O 2 (Seinfeld and Pandis, 2006) and results in an increase of SO 4 2-in some regions, which compensates the decrease in the concentrations of NH 3 and NH 4 + (thus the concentrations of PM 2.5 ).For combined emission reductions, the effects of 50% reduction in AL-NH 3 emissions dominate over those of the combined reductions in January and those of 50% reduction in SO 2 and AL-NH 3 emissions dominate in July, resulting in a decrease in PM 2.5 concentrations by up to 19.2% in January and by up to 18.3% in July.The largest reductions occur throughout the eastern NC and in the northeastern GA.The combined reductions result in a decrease of all three components (NH 4 + , SO 4 2-, and NO 3 -) throughout the domain, with larger reductions of NH 4 + and SO 4 2-in the regions of high NH 3 emissions.In January, the reduction of PM 2.5 in January is dominated by a reduction in NH 4 NO 3 resulting from the reduced NH 3 emissions.In July, the region of SO  of AL-NH 3 , NO x , and SO 2 emissions reduce PM 2.5 more than reducing any of them individually, with the reduction of NH 3 being the most effective control strategy in January and the reduction of NH 3 and SO 2 being the most effective control strategy in July.This indicates, in terms of air quality management policies in the southeastern U.S., reducing NH 3 emissions in July could be beneficial in reducing PM 2.5 in the region.In addition, Pinder et al. (2007) reported that reducing NH 3 may be more cost effective than implementing further reductions in SO 2 and NO x for regions that may require additional controls beyond initial SO 2 and NO x reductions to meet the national ambient air quality standards.SO 2 emissions are reduced; however the increase in NO 3 contribution to PM 2.5 concentrations is insignificant.The largest reductions in domain-wide average PM 2.5 in both months occur when the emissions of all three species are reduced by 50%.In January, the decrease is due to the decrease in AL-NH 3 emissions, decreasing NH 4 NO 3 is not much larger than that when AL-NH 3 emissions alone are decreased.In July, the decrease in PM 2.5 is larger than the decrease due to emission reductions in any individual species.The reduction of PM 2.5 is largely due to the reduced AL-NH 3 and SO 2 emissions.Although the percentage reduction of NO 3 -is large, the concentration of NO 3 -in summer is small and has little impact on the decrease of PM 2.5 concentrations.
Similar to the non-linear responses of secondary air pollutants to precursor emission reductions, the response of the dry and wet deposition fluxes to precursor emission reductions is non-linear, because they depend not only on the precursor concentrations, but also meteorological variables such as precipitation, wind speeds, temperature, and atmospheric stability.Figs. 10 and 11 show percentage differences in the dry, wet, and total deposition fluxes of total nitrogen (Tot-N) in January and July, respectively, when the emissions of SO 2 , NO x , and AL-NH 3 individually and collectively are reduced by 50%.In January, reducing SO 2 results in a slight increase (by up to 1.8%) in Tot-N dry deposition and a slight decrease (by up to 5.0%) in the Tot-N wet deposition over most areas, with the former impact dominating the impact on the total deposition of Tot-N.Reducing NO x emissions results in a moderate decrease (up to 17.2%) in the Tot-N dry deposition in most areas except for the source regions in NC and a decrease (up to 19.3%) in the Tot-N wet deposition in all areas, resulting in a reduction in the total deposition of Tot-N.Reducing AL-NH 3 emissions results in a large decrease (up to 53.3%) over source regions and a small increase (up to 6.3%) over the rest of areas in Tot-N dry deposition, a moderate decrease in Tot-N wet deposition in all areas, with the latter impact dominating the impact on the total deposition of Tot-N.Reducing emissions of all three species results in a moderate-to-large decrease in the dry (up to 51.7%), wet (up to 28.7%), and total (up to 42.5%) deposition fluxes of Tot-N throughout the domain.In July, reducing SO 2 results in a slight increase in dry deposition fluxes of Tot-N (by up to 6%) over the whole domain and in total deposition fluxes of Tot-N (by up to 3.6%) over most areas and a slight decrease (by up to 9.8%) in Tot-N wet deposition fluxes over most areas.Reducing NO x and AL-NH 3 emissions results in a moderate-to-large decrease in dry (up to 30.9% and 52.1%, respectively), wet (up to 32.7% and 32.3%), and total (up to 27.1% and 47.4%) deposition fluxes of Tot-N in the whole domain.Similar to the July results, reducing emissions of all three species results in a moderateto-large decrease in the dry (up to 51.8%), wet (up to 41.4%), and total (up to 48.5%) deposition fluxes of Tot-N throughout the domain.
The sensitivity of the Tot-N wet deposition fluxes to emission reductions may be somewhat affected by the large biases in precipitation and concentration predictions.For example, the significant overpredictions in Precip at 4km in July would make the Tot-N wet deposition fluxes more limited by its ambient concentrations, thus increasing the predicted sensitivity of the Tot-N wet deposition fluxes to emissions of NO x and NH 3 .On the other hand, the moderate-to-large underpredictions of NH 4 + and NO 3 concentrations tend to decrease the predicted sensitivity of the Tot-N wet deposition fluxes to the emissions of NO x and NH 3 .Therefore, the impacts of the model biases in Precip and concentrations of NH 4 + and NO 3 -on the simulated sensitivity may compensate to some extent, resulting in a smaller net impact than that of the individual bias.

CONCLUSIONS
The performance of MM5/CMAQ at horizontal grid resolutions of 12-, 4-, and 1.33-km is evaluated using available observations in the 1.33-km simulation domain.
Compared with simulation at 12-km, the use of a 1.33-km grid resolution in January generally degrades the performance of meteorological predictions but improves the performance of chemical predictions in terms of performance statistics of some species (e.g., SO 2 and secondary inorganic PM species) and temporal variation (e.g., O 3 at CND and PM 2.5 at LCC).It shows large improvement in July in most meteorological predictions (e.g., T1.5 at the CASTNET site, WS10, WD10, and Precip) and some chemical variables (e.g., dry deposition flux of HNO  + of 10-60% in January and 10-50% in July at and near the major sources, consistent with limited observations and other modeling studies in NC.CAMx gives a faster conversion rate in much larger areas in July due to higher SO 4 2-concentrations as a result of a weaker vertical mixing, a lower dry deposition flux, and different aerosol representation and microphysical treatments.CMAQ and CAMx give similar values of AdjGR in July but CAMx gives much larger NH 3 -rich regions and thus smaller NH 3neutral regions due to lower values of NO 3 -and HNO 3 in January.Both model calculations show that the eastern NC and northeastern GA are NH 3 -rich in both January and July. Sensitivity simulations are performed using MM5/CMAQ at a 4-km horizontal grid resolution with four emission reduction scenarios to determine most effective emission control strategies in the southeastern U.S. 50% SO 2 emission reduction from EGUs, mobile, and non-road sources has little impact on O 3 but a larger impact on PM 2.5 and nitrogen wet deposition in July than in January.The PM 2.5 reduction is dominated by SO 4 2-and NH 4 + reduction which is compensated to some extent by increased NO 3 -due to available NH 3 .Nitrogen dry deposition slightly increases, and nitrogen wet deposition slightly decrease.50% NO x emission reduction from EGUs and mobile sources results in an increase in O 3 due to a VOC-limited chemistry as reported in Zhang et al. (2009bZhang et al. ( , 2010) ) and an increase in PM 2.5 due to increased SO 4 2-(despite decreased NO 3 -and NH 4 + ) in January and a small decrease due to decreased NO 3 -, SO 4 2-, and NH 4 + in July.The decrease in NO 3 -also leads to moderate decreases in nitrogen dry and wet deposition fluxes in both months.50% AL-NH 3 emission reduction results in a larger decrease in PM 2.5 in January than July.The PM 2.5 reduction is dominated by NO 3 -and NH 4 + reduction in both months, but reduced NH 3 and NH 4 + in regions away from the AL-NH 3 sources result in an increase in pH which increases SO 4 2-in these regions in July.Nitrogen dry deposition decreases significantly at and near the source regions but slightly increases away from the sources.Nitrogen wet deposition decreases in both months, particularly at and near the sources.The combined reduction of emissions of SO 2 , NO x , and AL-NH 3 gives the largest decreases in PM 2.5 by up to 19.2% in January and by up to 18.3% in July due to decreased SO 4 2-, NO 3 -, and NH 4 + throughout the domain, despite an increase in O 3 due to a VOC-limited chemistry in January.These reductions also give the largest decrease in nitrogen dry and wet deposition throughout the domain.These results indicate that reducing AL-NH 3 emissions, in conjunction with already implemented SO 2 and NO x emission reductions, can further reduce PM 2.5 than reducing SO 2 and NO x emissions alone, while also preventing an increase in nitrogen deposition.Therefore, the reduction of NH 3 emissions should be incorporated in the integrated emission control strategies for future air quality attainment, particularly for regions with high emissions of NH 3.

Fig. 1 .
Fig. 1.The modeling domains at grid resolutions of 12-, 4-, and 1-km.The simulation at the 12-km grid resolution was performed by Morris et al. (2007) under the Visibility Improvement State and Tribal Association of the Southeast (VISTAS) program.The simulations at 4 and 1.33-km grid resolutions are performed in this work.

Fig. 8 .
Fig. 8.The percentage difference in the concentrations of O 3 simulated by CMAQ at a 4-km horizontal grid resolution in January (left) and July (right) when the emissions of SO 2 , NO x , and AL-NH 3 individually and their combined emissions are reduced by 50%.
terms of statistics but degradation in the performance of RH1.5 and other chemical predictions.The improvement at finer scales (1.33-km and

Table 1 .
Percentage changes in annual SO 2 and NO x emissions from each source type in each state within the 4-km domain in 2018 from the 2002 levels based on projected annual emissions reported by MACTEC (2008).

Table 2 .
Performance statistics of meteorological variables at 12-, 4-, and 1.33-km horizontal grid resolutions in January and July 2002.Bold numbers indicate the lowest NMBs for the corresponding variables.

Table 3 .
Statistics for gaseous and PM species at 12-, 4-, and 1.33-km horizontal grid resolutions in January and July 2002.Bold numbers indicate the lowest NMBs for the corresponding variables.

Table 4 .
Absolute (in μg/m 3 ) and percentage (%) changes in domain-wide average concentrations of PM 2.5 , NH 4 ) is not as much as expected, because of current limitations in some meteorological parameterizations (e.g., boundary layer schemes and land surface modules), and lack of accurate data for land use and emissions at a fine scale.CMAQ and CAMx give the conversion rates of NH 3 to NH 4