Interactive comment on “Sensitivity of nitrate aerosols to ammonia emissions and to nitrate chemistry: implications for present and future nitrate optical depth”

In this study, the authors describe an updated version of the nitrate aerosol scheme included in the GFDL atmospheric model, evaluate that scheme against observations, and analyse sensitivity simulations where emissions and chemistry options are varied. The paper is interesting and well-written. Figures illustrate the discussion well. The authors give an exhaustive view of the nitrate modelling capability of their model, and demonstrate deep understanding of the model and its sensitivity to changes. It is particularly useful to give budgets. I recommend publication after minor revisions are made to address the following comments.


Introduction
Ammonium nitrate (NH 4 NO 3 ) aerosols are produced by the reaction of nitric acid (HNO 3 ), a photochemical product of NO oxidation, and ammonia (NH 3 ).Emissions of NH 3 and NO are primarily from anthropogenic origin: fossil fuel combustion for NO and agriculture for NH 3 (i.e., Bouwman et al., 1997;Paulot et al., 2014).The formation of NH 4 NO 3 is favored by cold temperatures and high relative humidity (Stelson and Seinfeld, 1982).NH 4 NO 3 production competes with that of ammonium sulfate, which is generally more thermodynamically stable (Pinder et al., 2008), and that of coarsemode nitrate via heterogeneous uptake of HNO 3 on dust and sea salt (i.e., Zhuang et al., 1999;Jacobson, 1999;Jordan et al., 2003).
NH 4 NO 3 is an important component of surface particulate matter in the USA (i.e., Malm et al., 2004;Hand et al., 2012;Kim et al., 2014), Europe (i.e., Schaap et al., 2004), and Asia (i.e., Pathak et al., 2009;Ying et al., 2014), especially in winter.As NH 4 NO 3 rapidly volatilizes away from sources of NO and NH 3 and with warmer temperature, it is only predicted to make an important contribution to aerosol optical depth (AOD) over polluted regions (Park et al., 2014), with global annual estimates of nitrate optical depth ranging from 0.0023 to 0.025 (Bellouin et al., 2011;Shindell et al., 2013;Myhre et al., 2013;Hauglustaine et al., 2014).However, recent modeling studies have shown that NH 4 NO 3 may become the largest contributor to anthropogenic AOD by the end of the twenty-first century (Hauglustaine et al., 2014), following the projected increase of NH 3 emissions and decrease of SO 2 emissions.Such an increase of NH 4 NO 3 F. Paulot et al.: Sensitivity of nitrate optical depth to ammonia emissions and nitrate chemistry would offset some of the decline in anthropogenic aerosol radiative forcing over the twenty-first century (West et al., 1998;Adams et al., 2001;Liao et al., 2006;Bellouin et al., 2011;Henze et al., 2012;Shindell et al., 2013;Hauglustaine et al., 2014).
In this study, we aim to characterize the mechanisms controlling the response of NO − 3 optical depth to changes in anthropogenic emissions from 2010 to 2050.We focus in particular on how this response is modulated by the temporal and spatial variations in NH 3 emissions, the heterogeneous chemistry of HNO 3 , and the surface removal of nitrate aerosols.In Sect.2, we first describe a new configuration (AM3N) of the global chemistry-climate atmospheric model (AM3) from the Geophysical Fluid Dynamics Laboratory (GFDL), with revised treatments of sulfate and nitrate chemistry and aerosol deposition.We emphasize significant differences in the simulated budgets of SO 2− 4 , NO − 3 , and NH x ≡ NH 3 + NH + 4 between AM3N and the version of AM3 used for the Coupled Model Intercomparison Project (CMIP) 5.In Sect.3, we evaluate the simulated distribution of AOD, as well as SO 2− 4 , NO − 3 , and NH 3 concentrations at the surface and in precipitated water.In particular, we evaluate AM3 and AM3N against the extensive set of aerosol composition and optical properties routinely measured at Bondville (40.1 • N, 88.4 • W).In Sect.4, we examine the response of NO − 3 optical depth to projected changes in anthropogenic emissions in 2050 and its sensitivity to different treatments of removal and chemistry.

Model description
We use the GFDL-AM3 chemistry-climate model to simulate gas and aerosol chemistry.In its standard form, AM3 uses a finite volume dynamical core on a cubed sphere grid with 200 km (c48) horizontal resolution and 48 hybrid sigma pressure vertical layers (Donner et al., 2011).AM3 simulations were conducted for the Atmospheric Chemistry and Climate Model Intercomparison Project (ACCMIP) (Naik et al., 2013b) and as the atmospheric component of the GFDL coupled climate model CM3 for CMIP5 in support of the IPCC AR5.
The chemistry of AM3 has been described by Naik et al. (2013a) with updates to the gas-phase and heterogeneous chemistry (Mao et al., 2013b, c).Briefly, AM3 includes SO 2− 4 formation from gas-phase oxidation and the in-cloud reaction of SO 2 with O 3 and H 2 O 2 (Tie et al., 2005).Incloud production of SO 2− 4 is sensitive to cloud pH, which is calculated as a function of the concentration of SO 2− 4 (assumed to be entirely in-cloud water), NH 3 , SO 2 , HNO 3 , and CO 2 .NH 4 NO 3 formation is calculated following Stelson and Seinfeld (1982), but is assumed irreversible.Dry deposition and wet scavenging by large-scale and convective precipita-tion are described by Fang et al. (2011);Donner et al. (2011);Naik et al. (2013a).
Aerosol optical properties (i.e., extinction efficiency, single-scattering albedo, and asymmetry parameter) are described by Donner et al. (2011) and Strong et al. (2015).Sulfate is assumed to be fully neutralized by ammonium.Its size distribution is taken as log-normal following Haywood and Ramaswamy (1998) with hygroscopic growth based on pure ammonium sulfate (Tang and Munkelwitz, 1994) and capped at 95 % relative humidity.Aerosol activation into cloud droplets follows the parameterization of Ming et al. (2006).For radiative calculations, aerosols are assumed to be externally mixed except for sulfate and hydrophilic black carbon, which are assumed internally mixed (Donner et al., 2011).Nitrate is not considered for radiative calculations in AM3.
A new configuration of AM3 is introduced (referred to as AM3N hereafter) with the following changes aimed at improving the simulation of nitrate aerosols (see Sect. 3).
Aerosol chemistry -we use ISORROPIA to simulate the sulfate-nitrate-ammonia thermodynamic equilibrium (Fountoukis and Nenes, 2007).Equilibrium between gas and aerosol is assumed to be reached at each model time step (30 min), which is generally justified for PM 2.5 (Meng and Seinfeld, 1996).In-cloud oxidation of SO 2 is restricted to liquid clouds and we revise the calculation of cloud pH to account for the partitioning of HNO 3 /NO − 3 and NH 3 /NH + 4 between the gas phase and cloud water.
Heterogeneous chemistry -we include the heterogeneous uptake of HNO 3 , NO − 3 , N 2 O 5 , SO 2 , and H 2 SO 4 on dust particles (Table S1 in the Supplement).The uptake of HNO 3 , NO − 3 , and N 2 O 5 is assumed to be limited by alkalinity (Song and Carmichael, 2001).Following Fairlie et al. (2010), dust alkalinity is comprised of calcium and magnesium carbonates, with calcium and magnesium constituting 3 and 0.6 % (by mass) of coarse dust emissions (radius > 1 µm), respectively.Observations suggest alkalinity is primarily found in the coarse mode (Claquin et al., 1999); we assume that fine dust carries half as much alkalinity per kilogram as coarse dust.We also reduce the reaction probabilities (γ ) of N 2 O 5 , NO 2 , and NO 3 on aerosols relative to AM3 (Mao et al., 2013b) (see Table S1 in the Supplement and Sect.2.3.2).The implications of these changes for the budget of HNO 3 and aerosol NO − 3 are described in Sect.2.4.Nitrate optical depth -the optical properties and the mixing with black carbon of ammonium nitrate are assumed to be identical to those of ammonium sulfate.This approximation introduces an error in mass extinction at 550 nm of less than 20 % for relative humidity (RH) < 90 % and by less than 10 % between 90 and 95 % (Fig. S1 in the Supplement).The optical depth of NO − 3 associated with dust is expected to be small relative to fine-mode NO − 3 (e.g., Hauglustaine et al., 2014) and it is not considered here.
Dry deposition -similar to AM3, the dry deposition fluxes of gases and fine aerosols are calculated based on a monthly climatology of deposition velocities.We update this climatology to account for recent observations of rapid deposition of H 2 O 2 and some oxygenated volatile organic compounds, using the deposition velocities calculated in the GEOS-Chem chemical transport model as described by Nguyen et al. (2015).
Wet deposition -in AM3, aerosol removal by snow is treated like that by rain.In AM3N, water-soluble aerosols are not removed by snow, when the snow is formed via the Wegener-Bergeron-Findeisen mechanism (referred to as Bergeron mechanism hereafter), i.e., when water evaporates from liquid cloud droplets and condenses onto growing ice crystals.This treatment is consistent with observations (Henning et al., 2004) and similar to that used in other global models (Liu et al., 2011;Wang et al., 2011;Fan et al., 2012).Scavenging by snow formed via riming and homogeneous freezing is treated like that by rain.Gases are not scavenged by snow except HNO 3 (Neu and Prather, 2012).Convective plumes are discretized on a vertical grid that has finer vertical resolution than AM3 (Donner, 1993).The improved discretization of the convective plume has little impact on precipitation at the surface but increases the convective wet removal of tracers as we will show in Sect.3.

Emissions
We use anthropogenic emissions from the Hemispheric Transport of Air Pollution v2 (HTAP_v2) task force regridded to 0.5 • × 0.5 • for years 2008and 2010(Janssens-Maenhout et al., 2015).HTAP_v2 aircraft emissions are distributed vertically following Lamarque et al. (2010).Daily biomass burning emissions are based on the NCAR Fire INventory (FINNv1, Wiedinmyer et al., 2011) and emitted in the model surface layer.Average dust emissions are parameterized following Ginoux et al. (2001), as where C is a dimensional factor (µg s 2 m −5 ), S is the source function based on topography, u 10 m is the horizontal wind at 10 m (m s −1 ), u t is the threshold velocity (m s −1 ), and s p is the fraction of total dust emitted in the size class p as defined by Li et al. (2008).Over the 2008-2010 period, dust emission is 1640 Tg a −1 .This includes 1230 Tg a −1 from natural sources (S from Ginoux et al. (2001), C = 0.125 µg s 2 m −5 , u t = 1 m s −1 ), similar to the AEROCOM multi-model mean (Huneeus et al., 2011), and 410 Tg a −1 from anthropogenic sources (primarily over cropland and pasture from Ginoux et al. (2012b) with updated MODIS collection 6, C = 0.219 µg s 2 m −5 , u t = 3 m s −1 ).Isoprene emissions are calculated using the Model of Emissions of Gases and Aerosols from Nature (MEGAN, Guenther et al., 2006;Rasmussen et al., 2012).NO emissions from lightning are calculated as a function of subgrid convection (Horowitz et al., 2003).Differences in the treatment of convection in AM3N result in greater NO emissions from lightning in AM3N (5.6 Tg N a −1 ) compared to AM3 (5.2 Tg N a −1 ), with both estimates within the range of emissions inferred from observations (Martin et al., 2007;Murray et al., 2012).Other natural emissions, including soil NO x and soil and ocean NH 3 emissions, are described by Donner et al. (2011) and Naik et al. (2013a).Global total emissions of SO 2 , NH 3 , and NO x are listed in Table 1.

Sensitivity simulations
Considering the large uncertainty in the simulated nitrate optical depth and surface concentrations, we design a set of sensitivity simulations based on AM3N to characterize the sensitivity of nitrate and sulfate to key uncertainties in chemistry and in NH 3 emissions (Table 2).All simulations are run from 2007 to 2010, using 2007 to spin-up the model.To facilitate the comparison with observations and limit meteorological variability across model configurations, the model horizontal wind is relaxed to 6 hourly values from the National Centers for Environmental Predictions reanalysis (Kalnay et al., 1996) as described in Lin et al. (2012).

NH 3 emissions
Present-day -the largest source of NH 3 to the atmosphere is agriculture.Unlike anthropogenic emissions of other compounds, which are dominated by fossil fuel emissions, NH 3 emissions exhibit large seasonal variations, which reflect the seasonality of agricultural practices (e.g., fertilizer application) as well as the decrease of NH 3 solubility with temperature (Misselbrook et al., 2004;Pinder et al., 2006;Paulot et al., 2014).The HTAP_v2 inventory includes monthly variations in anthropogenic NH 3 emissions over North America, Europe, and parts of Asia, including Japan and China, but excluding India.Anthropogenic emissions of NH 3 previously used in AM3 simulations for ACCMIP and CMIP5 are constant throughout the year (Lamarque et al., 2010).To evaluate the impact of the seasonality of NH 3 emissions on NO − 3 , we remove all temporal variability in the anthropogenic emissions of NH 3 in simulation AM3N_ns.NH 3 emissions also exhibit diurnal variability (Pinder et al., 2006), which may affect the simulated concentrations of NH 3 and NH 4 NO 3 (Zhu et al., 2013;Van Damme et al., 2014b;Schiferl et al., 2014;Zhu et al., 2015).In AM3N_diu, we impose the NH 3 diurnal cycle of the regional LOTOS (Long Term Ozone Simulation) model globally (Schaap et al., 2004).The ratio between maximum emissions (13:00-14:00 local time) and minimum emissions (03:00-06:00) is 5.7.
2050 -anthropogenic NH 3 emissions for 2050 are estimated by scaling HTAP_v2 surface anthropogenic NH 3 emissions with national projections from the Representative Concentration Pathway 8.5 (RCP8.5)from 2010 to 2050 (Fig. 1), while keeping natural and biomass burning emissions at their present-day levels.We use the RCP8.5 scenario for 2050 (van Vuuren et al., 2011) as it most closely resem- a SO 2 emissions are 74.0Tg S a −1 including 16.0 Tg S a −1 from dimethyl sulfide (DMS) oxidation.b including 39.9 Tg N a −1 from anthropogenic sources, 3.9 Tg N a −1 from biomass burning, and 10.7 Tg N a −1 from natural sources (primarily from the ocean).
bles emissions from regional inventories over the 2000-2010 period (Granier et al., 2011).However, we do not use the RCP8.5 spatial distribution of NH 3 sources, as it differs notably from HTAP_v2 over many source regions such as India, the Nile delta, the Benelux, the California Central Val- ley, and the Saskatchewan (Fig. 1).These differences may reflect mapping errors for RCP8.5 NH 3 emissions from agriculture as noted by Lamarque et al. (2013).Our approach results in 18 % more anthropogenic emissions (60 Tg N a −1 ) than in RCP8.5 for 2050.

Heterogeneous chemistry
Wintertime production of HNO 3 in the northern midlatitudes' boundary layer is dominated by the uptake of N 2 O 5 on aerosols (e.g., Dentener and Crutzen, 1993;Tie et al., 2003;Lamsal et al., 2010).The probability for the heterogeneous conversion of N 2 O 5 to HNO 3 (γ ) remains uncertain (Chang et al., 2011), with field and laboratory observations showing that it is inhibited by aerosol nitrate and organics (Brown et al., 2009;Brown and Stutz, 2012;Wagner et al., 2013;Gaston et al., 2014), but enhanced by cold temperatures (Griffiths and Anthony Cox, 2009;Wagner et al., 2013).
To quantify the impact of the heterogeneous production of HNO 3 on aerosol NO − 3 , we neglect the heterogeneous production of HNO 3 via N 2 O 5 aerosol uptake in AM3N_nhet.We also neglect the productions of HNO 3 by NO 3 and NO 2 reactive uptake, as they may modulate the wintertime budget of NO y in polluted region (Paulot et al., 2013).Note that previous characterizations of NO − 3 optical depth also neglected the heterogeneous chemistry of oxidized nitrogen (e.g., Bellouin et al., 2011).
We also evaluate the impact of the heterogeneous chemistry on dust as it is not included in all models (e.g., Pye et al., 2009;Bellouin et al., 2011).In AM3N_ndust, we neglect the uptake of HNO 3 , N 2 O 5 , NO 3 , H 2 SO 4 , and SO 2 on dust.(Huebert et al., 1988;Wyers and Duyzer, 1997;Van Oss et al., 1998;Rattray and Sievering, 2001;Nemitz et al., 2004;Fowler et al., 2009;Wolff et al., 2010;Barbaro et al., 2015).This difference stems from gradients in temperature, RH, and HNO 3 within the boundary layer, which reduce the stability of NH 4 NO 3 near the surface.The volatilization of NH 4 NO 3 may result in an underestimate of the surface deposition of TNO 3 ≡ HNO 3 +NO − 3 , since v d (NH 4 NO 3 ) v d (HNO 3 ).As an upper bound, we assume that the surface removal of fine NO − 3 is limited by turbulent transport by setting v d (NO − 3 ) = v d (HNO 3 ) in AM3N_fdep.

Budget and global distribution
Table 1 shows the budgets of SO 2− 4 , NH x , and NO y in AM3 and AM3N for 2010.Here NO y is defined as the sum of all species that contained oxidized nitrogen.The budgets for all simulations are given in Table S2.
The lifetimes of SO 2− 4 , NH x , and NO y are significantly shorter in AM3N than in AM3.This decrease is driven in part by greater convective removal associated with changes in finer vertical discretization of convective plumes.For instance, the lifetime of SO 2− 4 with respect to convective removal decreases from 44 to 18 days.
For SO 2− 4 , the increased effectiveness of convective removal is partly offset by reduction in the removal by snow (Sect.2.2).The SO 2− 4 lifetime in both AM3 and AM3N falls within the range of AEROCOM models (3-5.2 days Schulz et al., 2006).Unlike AM3, AM3N includes ammonium in the calculation of cloud pH, which reduces the acidity of cloud droplets and favors the production of SO 2− 4 via in-cloud oxidation of SO 2 by O 3 .The production of SO 2− 4 via SO 2 + O 3 is 4.5 Tg S a −1 in AM3N, greater than the recent estimate of Sofen et al. (2011) (1.5 Tg S a −1 ).This discrepancy may reflect differences in cloud pH and lower H 2 O 2 concentrations in AM3N because of faster dry deposition for H 2 O 2 and effi-cient removal of HO 2 via aerosol uptake (Mao et al., 2013a).AM3N does not include production of SO 2− 4 via the aqueous reaction of SO 2 with O 2 catalyzed by iron and manganese or by the oxidation of SO 2 by stabilized Criegee intermediates (Mauldin III et al., 2012).The lifetime of SO 2 is 1.3 days in both AM3 and AM3N, similar to Sofen et al. (2011) and Lee et al. (2011).The overall conversion from SO 2 to SO 2− 4 (excluding SO 2− 4 on dust) is reduced compared to AM3 from 50 to 42 % and lower than the AEROCOM multi-model mean (62 %).
In AM3, NH 3 uptake by SO 2− 4 is solely controlled by kinetics without any thermodynamic limit, such that NH 3 burden is small (0.005 TgN) and NH 3 generally limits the formation of NH 4 NO 3 .In AM3N, the uptake of NH 3 by SO 2− 4 aerosols cannot exceed the thermodynamic limit calculated by ISORROPIA, which results in a greater NH 3 burden (0.11 TgN) and favors the production of NH 4 NO 3 .The shorter lifetime of NH x in AM3N than in AM3 reflects the change in the speciation of NH x and the faster dry deposition of NH 3 relative to NH + 4 .The lifetime of NH x in AM3N (2.5 days) is similar to that derived by Xu and Penner (2012) and Hauglustaine et al. (2014) (2.3 days).
AM3N and AM3 differ most strikingly in their simulations of NO y .The contribution of HNO 3 to the removal of NO y decreases from 81 % (AM3) to 56 % (AM3N).In contrast, the contribution of aerosols to NO y removal increases from 2 to 22 %.Recent studies (Hauglustaine et al., 2014;Xu and Penner, 2012) have found an even greater contribution of aerosols to the removal of NO y (> 30 %); this difference may reflect the lack of HNO 3 uptake by sea salt in AM3N.Organic nitrogen contributes 10 % of NO y removal in both AM3 and AM3N.The much lower fraction of NO y deposited as HNO 3 in AM3N relative to AM3 reflects both the increased production of NH 4 NO 3 and the uptake of HNO 3 on dust.The total heterogeneous production of HNO 3 by N 2 O 5 (9.7 Tg N a −1 ), NO 2 (0.6 Tg N a −1 ), and NO − 3 (0.4 Tg N a −1 ) uptake on fine aerosols is reduced by 50 % in AM3N relative to AM3.This decrease is primarily driven by reduced reaction probabilities for NO 2 and NO 3 uptake.In contrast, the change of γ N 2 O 5 from 0.1 (AM3) to 0.01 (AM3N) reduces the heterogeneous uptake of the large increase in the sulfate surface area in winter (see Sect. 3).The magnitude of the N 2 O 5 source of HNO 3 in AM3N is 3 times as large as reported by Hauglustaine et al. (2014).This may reflect greater reactive aerosol surface area in AM3N, as N 2 O 5 hydrolysis can take place on SO 2− 4 , BC, OC, and NO − 3 aerosols, while only SO 2− 4 is considered by Hauglustaine et al. (2014).Reduction in the simulated HNO 3 burden -driven by faster NO − 3 deposition (AM3N_fdep), heterogeneous uptake of HNO 3 on dust (AM3N_ndust), or reduced heterogeneous production of HNO 3 (AM3N_nhet) -increase cloud pH, which favors the oxidation of SO 2 by O 3 (Table S2).
Figure 2 shows the burden of fine NO − 3 , NO − 3 on dust, NH + 4 , and NH 3 in AM3N.The simulated global burdens fall within the range of previous estimates (Bauer et al., 2007;Feng and Penner, 2007;Pye et al., 2009;Pringle et al., 2010;Bellouin et al., 2011;Xu and Penner, 2012;Hauglustaine et al., 2014) for fine NO − 3 (0.04-0.11Tg N), NO − 3 on dust (0.07-0.41 Tg N), NH + 4 (0.21-0.27Tg N), and NH 3 (0.07-0.29 Tg N).The burden of fine NO − 3 peaks over China where it reaches over 5 mg N m −2 , with a secondary maximum over India.Fine NO − 3 burden is also elevated over northern Europe and the US Midwest, where agricultural activities are located close to large sources of oxidized nitrogen.Compared with the fine nitrate distribution from Hauglustaine et al. (2014) for 2000, AM3N simulates greater nitrate burden over Asia but lower burdens over Europe and the USA.These differences may reflect different spatial distributions of NH 3 emissions (Fig. 1).AM3N simulates large enhancements in NH 3 column over source regions such as India (where the burden reaches 12 mg m −2 ), northern China, the Netherlands, and the US Midwest, as supported by satellite observations (Van Damme et al., 2014a).This lends some support to the spatial allocation of anthropogenic NH 3 emissions in HTAP_v2 inventory, although observed enhance-  2. ments in NH 3 burden over the Po Valley and California are not captured by AM3N.

Bondville
We first evaluate the model against an extensive suite of observations collected at Bondville (40.1 • N, 88.4 • W; 213 m a.s.l.).Bondville is located in the vicinity of large sources of NH 3 and NO x , which result in elevated surface NO − 3 concentrations (Fig. 3) and make this site wellsuited to evaluate the representation of nitrate aerosols in AM3 and AM3N.Here we compare the model against observations of surface NO − 3 and SO Figure 3 shows the observed (black) and simulated monthly concentrations in surface air (left column) and in precipitated water (right column) for NO − 3 , SO 2− 4 , and NH 3 (NH + 4 for wet deposition) for AM3 and different AM3N configurations.Both NO − 3 and NH 3 concentrations are higher year round in AM3N than in AM3, as ISORROPIA enforces thermodynamic limitation on the uptake of NH 3 by SO 2− 4 .Observations show a spring peak in surface NH 3 concentrations, while both AM3 and AM3N simulate a summer peak.Bondville is surrounded by corn and soybean fields and NH 3 emissions associated with spring fertilizer application may be underestimated (Paulot et al., 2014).In summer, more efficient convective removal of SO 2− 4 in AM3N reduces the AM3 high bias for SO 2− 4 surface concentration and low bias for SO 2− 4 wet deposition.In winter, the low bias for surface SO 2− 4 concentration in AM3 is reduced as a result of less efficient removal by snow and increased in-cloud oxidation of SO 2 .AM3N_nhet and AM3N_fdep produce greater SO 2− 4 concentrations in winter than AM3N consistent with increased in-cloud oxidation of SO 2 by O 3 (Table S2).
NO − 3 shows a large positive bias in AM3N in winter (> 70 % in February).This bias can be reduced by either neglecting the heterogeneous production of HNO 3 via NO 2 , NO 3 , and N 2 O 5 (AM3N_nhet) or treating the deposition of fine NO − 3 like that of HNO 3 (AM3N_fdep).Conversely, neglecting the seasonality of NH 3 emissions (AM3N_ns), similar to simulations performed for ACCMIP and CMIP5, increases the bias for NO − 3 in winter.To analyze the factors controlling NH 4 NO 3 in the model, we calculate the gas ratio (GR) at each model time step.The GR was first proposed by Ansari and Pandis (1998) to diagnose the sensitivity of NH 4 NO 3 to its gas-phase precursors NH 3 and HNO 3 and is defined as . (2) GR defines three different regimes: (a) GR > 1, in which NH 4 NO 3 formation is limited by the availability of HNO 3 , (b) 0 < GR < 1, in which NH 4 NO 3 is limited by the availability of NH 3 , and (c) GR < 0, in which NH 4 NO 3 is inhibited by SO 2− 4 .We define the degree of limitation of NH 4 NO 3 by HNO 3 (L(HNO 3 )) as the fraction of the time when GR > 1.In winter, NH 4 NO 3 is most frequently limited by HNO 3 (L(HNO 3 )) = 78 % in AM3N).3 concentrations well but underestimates the frequency of high NO − 3 events, when NH 4 NO 3 exhibits significant sensitivity to NH 3 .Under these conditions, less volatilization of NH 4 NO 3 near the surface is expected as NH 3 is not depleted near the surface like HNO 3 .

AM3_nhet [NO −
3 ] is most consistent with observations at high [NO − 3 ], conditions under which N 2 O 5 heterogeneous uptake has been observed to be inhibited both in laboratory and field settings (Bertram and Thornton, 2009;Wagner et al., 2013).The ability of AM3N_fdep and AM3N_nhet to capture NO − 3 under different conditions emphasizes the need to represent the dynamic nature of γ (N 2 O 5 ) and TNO 3 surface removal.
Figure 5 shows the observed and simulated monthly AOD at Bondville.Observed AOD peaks in summer and reaches a minimum in winter.This seasonality is well captured by AM3 (top panel), while AOD in AM3N_fdep_diu (bottom panel) peaks in spring and is biased high in winter and fall.Biases in AOD may be caused by errors in aerosol abundance and speciation but also by errors in aerosol hygroscopic growth.Their relative contribution can be estimated by comparing observed and simulated aerosol extinction pro- files, under dry conditions (RH < 40 %) (Delene and Ogren, 2002;Sheridan et al., 2012;Esteve et al., 2012).Figure 6 shows that AM3N overestimates aerosol dry extinction in spring and fall, which suggests that the simulated aerosol abundance is overestimated.This bias may be caused by organic carbon or dust, which contribute over 30 % of the simulated aerosol dry extinction throughout the column in spring, summer, and fall (Fig. S2 in the Supplement).In winter and summer, AM3N is more consistent with the observed aerosol dry extinction profile than AM3.In particular, AM3 exhibits a low bias in winter and a high bias in summer, consistent with the biases for surface [SO 2− 4 ] and with the lack of extinction from NO − 3 , the largest contributor to AM3N dry aerosol extinction below 1000 m in winter (Fig. S2).The different biases of AM3 and AM3N against AOD and dry extinction in winter and summer suggest errors in the hygroscopic growth of aerosols.This is consistent with comparisons with twice daily soundings of temperature (Fig. S3) and relative humidity (Fig. S4) over Bondville, which show that AM3N is on average too humid in winter and spring and too dry in summer.In particular, AM3N overestimates the Dry extinctions are reported at standard temperature and pressure (273.15K, 1 atm).We multiply the modeled nitrate extinction by 0.8 to account for the evaporation of ammonium nitrate in the nephelometer (Bergin et al., 1997).The different model sensitivity experiments are described in Table 2. occurrence of high-humidity periods (RH > 90 %, Fig. S5), when aerosol hygroscopic growth is especially large.Modeled AOD would be especially sensitive to positive RH biases in winter since AOD winter is primarily controlled by SO 2− 4 and NO − 3 , which have stronger hygroscopic growth than organic carbon and dust.

Global evaluation
We broaden our evaluations of AM3 and AM3N using observations of surface  (2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014) to take advantage of the ongoing expansion of the network.We apply Grubbs' test (Grubbs, 1950) for each station to filter out possible outliers (95 % critical value).Table 3 shows the normalized mean bias (ratio of the mean difference between the model and observations to the mean observed value) and the correlation between the model and observations for each data set for AM3, AM3N.Evaluations of all AM3N configurations and seasonal com-  parisons (Table S3 and Figs.S6 to S18) are provided in the Supplement.
Table 3 shows that AM3 and AM3N exhibit similar normalized mean biases for SO 2− 4 surface concentrations and wet deposition in the USA and Europe.However, AM3N exhibits better correlation with observations, which reflects a large improvement in the simulated seasonality of surface SO 2− Surface [NO − 3 ] is generally overestimated in AM3N, especially over the USA (+100 %).Recent studies using a range of NH 3 emissions and different representations of aerosol thermodynamics and heterogeneous chemistry have also found large positive biases in simulated surface [NO − 3 ] ( Heald et al., 2012;Walker et al., 2012;Hauglustaine et al., 2014).Figure 7 shows the annual distribution of L(HNO 3 ) in AM3N.At the surface, NH 4 NO 3 formation is primarily limited by the availability of HNO 3 over continental regions, such as Europe, India, or northern China.Under HNO 3limited conditions, our analysis at Bondville suggests that increasing the deposition of TNO 3 (AM3N_fdep) can improve the simulation of surface [NO − 3 ].On a continental basis, we also find that AM3N_fdep_diu better captures surface [NO − 3 ] (+17 % bias in the USA) and we will focus on this configuration in the following.Note that the diurnal cycle of NH 3 emissions has a small impact on the simulated mean surface [NO − 3 ] concentration, but reduces surface [NH 3 ] and increases its export to the free troposphere.Figure S20 shows the observed and simulated diurnal cycle of [NO − 3 ] at the YRK site from the SouthEastern Aerosol Research and Characterization Network.NO − 3 exhibits a pronounced diurnal cycle with a maximum in the early morning and a minimum in the late afternoon (as a result of both thermodynamics and boundary layer height).AM3N and AM3N_diu capture the timing of the diurnal cycle well.As NH 3 emissions peak in the afternoon, the magnitude of the NH 4 NO 3 diurnal cycle in AM3N_diu is lower than in AM3N.Higher daytime concentrations of NH 4 NO 3 in AM3N_diu suggest that accounting for the diurnal cycle of NH 3 emissions may increase the magnitude of the radiative forcing associated with NH 4 NO 3 .
Figure 8 shows the average monthly variation of AOD from 2008 to 2010 over different regions as observed by MODIS (Remer et al., 2008) and MISR (Kahn et al., 2009) and simulated by AM3 and AM3N_fdep_diu.Although AM3 does not exhibit a large bias on a global scale (normalized mean biases lower than 10% for both MODIS and MISR), it fails to capture the seasonality of AOD over most continental regions.Over North America, AOD is biased low in winter and high in summer in AM3, consistent with the biases in surface [SO 2−  4 ].The spring bias may be exacerbated by insufficient transport of aerosols from Asia.AM3 is biased high over tropical land masses, consistent with insufficient convective removal of aerosols.AM3N_fdep_diu AOD shows improved correlations with observations over most continental regions (see also Fig. S19).The increased AOD in winter and spring can be partly attributed to nitrate optical depth, which accounts for over 30 % of AOD over North America.
Following Lee and Adams (2010) and Shindell et al. (2013), we further evaluate the performances of AM3 and AM3N in locations within the top decile of simulated NO − 3 and SO 2− 4 burden against observations from MODIS, MISR, and AERONET.AM3 AOD is biased high over high SO 2− 4 regions (+30 to 50 %) and low over high NO − 3 regions (−10 to −50 %) consistent with the analysis of Shindell et al. (2013).The bias over high SO 2− 4 regions is greatly reduced in AM3N (< 10 %), while the model exhibits a high bias against satellite AOD observations (10-20 %) but little bias against AERONET observations in high NO − 3 regions.More detailed comparisons with AERONET show that AM3N better captures AOD at high latitudes in spring (Fig. S19), which lends support to the changes made to the representation of incloud sulfate production and wet deposition.

Present-day emission
Figure 9 compares the contributions of SO 2− 4 , NO − 3 , OC, BC, dust, and sea salt to the global mean AOD in AM3 and AM3N_fdep_diu with previous estimates (Shindell et al., 2013;Hauglustaine et al., 2014).Present-day global mean AOD in AM3N_fdep_diu is 0.136, 16 % less than in AM3.All AOD components decrease as a result of more efficient convective removal, with the largest decrease for SO 2− 4 (−36 %).SO 2− 4 optical depth decreases most from AM3 to AM3N_fdep_diu over tropical regions, while it increases at high latitudes, consistent with changes in SO 2− 4 chemistry and removal.NO − 3 optical depth ranges from 0.0052 (AM3N_nhet) to 0.0078 (AM3N_ndust).Our best estimate is 0.0060 (AM3N_fdep_diu).The different treatment of reactive nitrogen results in similar changes in SO 2− 4 (0.002) and NO − 3 optical depth (0.003).The range of NO − 3 optical depths derived from AM3N (0.0052-0.0078) encompasses   recent estimates by Hauglustaine et al. (2014) and Bellouin et al. (2011), but differs significantly from the Goddard Institute for Space Studies (GISS) (0.023) and the Centre for International Climate and Environmental Research -Oslo (CI-CERO) (0.002) models.Shindell et al. (2013) reported that convective transport of NH 3 to the free troposphere, where NH 4 NO 3 is stable and sensitive to NH 3 (Fig. 7), is responsible for the elevated nitrate in the GISS model.Revisions of the treatment of NH 3 convective removal in GISS reduce the simulated present-day NO − 3 optical depth to 0.005 (S.Bauer, personal communication, 2015).Shindell et al. (2013) also showed that CICERO may overestimate SO 2− 4 optical depth, which would inhibit the production of NH 4 NO 3 by decreasing the amount of free ammonia ([NH x ] − 2[SO 2− 4 ]). Figure 10 shows the annual AM3N nitrate optical depth and its sensitivity to the treatment of NH 3 emissions and NO − 3 chemistry in AM3N.The sensitivity of NO − 3 optical depth to NH 3 seasonality is small and follows the patterns of NH 4 NO 3 limitations by NH 3 , with largest sensitivity over the eastern USA and in the outflow of continents The global sensitivity to NH 3 seasonality is a lower bound, since the sea-sonality of anthropogenic NH 3 emissions is not represented in important source regions (e.g., India, South America) in HTAPv2.We find greater sensitivity to the diurnal cycle of NH 3 emissions, which is attributed to increased transport of NH 3 into the free troposphere, where NH 4 NO 3 is more sensitive to NH 3 (Fig. 7) and more stable because of colder temperature.Decreasing HNO 3 production, either by neglecting its heterogeneous production (AM3N_nhet) or increasing the deposition of NO − 3 (AM3N_fdep), reduces the annual mean NO − 3 optical depth by 25 % globally.Regionally, NO − 3 in polluted regions is more sensitive to the heterogeneous production of HNO 3 because of the large aerosol surface area in these regions.Neglecting heterogeneous chemistry on dust results in a large relative increase of NO − 3 optical depth in dusty regions, but the increase of the global mean NO − 3 optical depth is small (13 %).This muted response is caused by low NH 3 sources near major natural dust sources.A notable exception is anthropogenic dust, whose sources are primarily associated with agriculture (Ginoux et al., 2012a).The proximity of NH 3 and anthropogenic dust sources results in 35 % greater sensitivity of NO − 3 optical depth to anthropogenic dust than to natural dust (per kilogram of dust).to changes in anthropogenic emissions from the present day to 2050.NH 4 NO 3 increases in the free troposphere but decreases near the surface, a vertical redistribution also noted by Hauglustaine et al. (2014).The decrease of surface NO − 3 in the midlatitudes is primarily driven by lower NO emission.Large differences in the seasonality, spatial distribution, and magnitude of anthropogenic NH 3 emissions in RCP8.5 (dotted line) and scaled HTAPv2 for 2050 have little impact on the simulated NO − 3 burden (< 10 %), which reflects the diminishing sensitivity of surface NH 4 NO 3 to NH 3 .However, NO − 3 remains sensitive to NH 3 in the free troposphere, where it can persist longer than in the boundary layer thanks to lower temperature.The solid line in Fig. 12 shows the impact of lower convective removal of NH 3 (achieved by neglecting the impact of pH on NH 3 solubility) on the NO − 3 burden.Over the 2008-2010 period, this results in a 40 % increase of the NO − 3 burden with a near quadrupling in the tropics, qualitatively matching the results of Hauglustaine et al. (2014) in this region.In 2050, the impact is much more pronounced and the simulated burden is more than twice as large as in 2010, a similar response to that found by Hauglustaine et al. (2014).Note that increasing NH 3 emissions from biomass burning and distributing these emissions vertically (Naik et al., 2013a) also increases tropical NO − 3 (not shown) but to a much lower degree (< 50 %).These results suggest that differences in the transport of NH 3 to the free tropo-sphere across models contribute to the variability in the projected NO − 3 burden and optical depth.Such differences may arise from differences in the parameterizations of convection (Folkins et al., 2006) as suggested by the much lower tropical NO − 3 burden in AM3N than in the LMDz-INCA model (Hauglustaine et al., 2014) but also from changes in the tropical circulation in response to climate change (e.g., Ma et al., 2012).

Conclusions
We have developed a new configuration of AM3 (AM3N) with revised treatment of nitrate and sulfate chemistry and deposition.We showed that AM3N better captures observed AOD than a configuration of AM3 similar to that used for ACCMIP and CMIP5.AM3N overestimates surface NO − 3 concentration especially in the USA.This bias may reflect neglect in AM3N of the dynamic nature of N 2 O 5 uptake and near-surface volatilization of NH 4 NO 3 .
We have evaluated the sensitivity of NO − 3 optical depth to poorly constrained aspects of NO − 3 chemistry (heterogeneous production of HNO 3 , uptake of HNO 3 by natural and anthropogenic dust, surface removal of NH 4 NO 3 ) and NH 3 emissions (diurnal cycle, seasonality).Globally, the formation of NH 4 NO 3 is more limited by HNO 3 than NH 3 , such that NO − 3 optical depth is more sensitive to the representation of the heterogeneous chemistry of HNO 3 than to uncertainties in NH 3 emissions.Simulated present-day NO − 3 optical depth ranges from 0.0054 to 0.0082, depending on the treatment of reactive nitrogen.Differences in the treatment of reactive nitrogen alone are unlikely to account for the large spread in estimates of present-day NO − 3 optical depth (0.0023-0.025).
We have examined the response of simulated NO − 3 optical depth to projected changes in anthropogenic emissions from 2010 to 2050 in RCP8.5.Depending on the configuration of AM3N (Table 2), NO − 3 optical depth varies from 0.0061 to 0.01 in 2050.The increase of NO − 3 (< 30 % relative to 2008-2010) is partly inhibited by greater limitation of NH 4 NO 3 production by HNO 3 at the surface due to lower NO emissions, more efficient removal of HNO 3 by dust, and a large decrease in the heterogeneous production of HNO 3 by N 2 O 5 (associated with lower aerosol surface area).In the Northern Hemisphere, the NO − 3 burden is projected to shift southward, following the increase of tropical NH 3 emissions and the decrease of NO emissions in the midlatitudes.This shift is associated with an increase of the NO − 3 burden in the free troposphere, where NH 4 NO 3 formation is limited by NH 3 .We suggest that the convective transport of NH 3 and its response to climate change (not considered here) play an important role in modulating the response of NO − 3 optical depth to changes in anthropogenic emissions.The complexity of the response of NO − 3 to changes in surface processes, chemistry, and convection indicates that the global trends of NH 3 ; see text).The blue, green, red, and cyan regions denote the NO − 3 burden located above 800 hPa, between 600 and 800 hPa, between 400 and 600 hPa, and below 400 hPa, with the partial burden in each pressure range indicated inset.The annual mean zonal burdens of NO − 3 simulated using AM3N_fdep_diu (dashed line), using AM3N with anthropogenic emissions from RCP8.5 for NH 3 (dotted line), using AM3N_ndust (dashed dotted line), and using AM3N with reduced convective removal of NH 3 (solid line) are also shown.The white circles in the top panel indicate the 2000 annual zonal mean NO − 3 burden simulated by Hauglustaine et al. (2014).emissions may not be a suitable proxy to estimate the future forcing from NO − 3 aerosols (Heald and Spracklen, 2015).We conclude that in addition to improvements to NH 3 emission inventories (e.g., bidirectional exchange of NH 3 , Zhu et al., 2015), observational constraints on the processes controlling the vertical redistribution of NH 3 and the response of NO − 3 to NH 3 in the free troposphere (e.g., magnitude of NH 3 emissions in the tropics (Aneja et al., 2012;Whitburn et al., 2015), biomass burning injection height (Val Martin et al., 2010), transport and removal of NH 3 in convective updrafts, heterogeneous chemistry on dust) and sensitivity studies to characterize their response to climate change are needed to improve estimates of present and future NO − 3 optical depth.
The Supplement related to this article is available online at doi:10.5194/acp-16-1459-2016-supplement.

Figure 1 .
Figure 1.Average annual emissions of NH 3 for 2010 (top row) and 2050 (bottom row) based on anthropogenic NH 3 emissions from HTAP_v2 (left column) and from RCP8.5 (right column).Non anthropogenic emissions (including biomass burning) are the same in all scenarios.Total annual emissions are indicated inset.

Figure 2 .
Figure 2. Annual mean burden of NO − 3 , NO − 3 on dust, NH + 4 , and NH 3 in mg N m −2 in AM3N from 2008 to 2010.Global burdens are indicated inset.The location of the Bondville site is indicated by a black cross in the upper left panel.

Figure 3 .
Figure 3. Observed (black) and simulated monthly concentrations of NO − 3 , SO 2− 4 , and NH 3 at Bondville (40.1 • N, 88.4 • W) in surface air (left panel) and precipitated water (right panel).Observations are averaged from 2006 to 2012, while model output is from 2008 to 2010.The vertical bars denote 1 standard deviation of the mean monthly observations.The different model sensitivity experiments are described in Table2.

Figure 4 .
Figure 4. Observed and simulated distribution of daily NO − 3 concentration at Bondville (40.1 • N, 88.4 • W) in winter (top panel) from 2006 to 2012 (observations) and 2008 to 2010 (model).The degree of HNO 3 limitation for NH 4 NO 3 formation (GR > 1) is shown in the bottom panel.The different model sensitivity experiments are described in Table2.

Figure 4
also shows that AM3N_nhet and AM3N_fdep produce different distributions of daily [NO − 3 ] although they have similar mean monthly [NO − 3 ] (top panel).AM3N_fdep reproduces observations at low NO −

Figure 6 .
Figure 6.Mean seasonal observed (black dots) and simulated surface and vertical profiles of aerosol dry extinction at Bondville (40.1 • N, 88.4 • W).The vertical profile show the average of all observations by the Airborne Aerosol Observatory from 2006 to 2009 collected during daytime (10:00-16:00 local time).Surface observations reflect the average of all daytime observations at the ESRL BND station from 2006 to 2012 with no local pollution.The model is averaged for daytime from 2008 to 2010.Horizontal lines show the 25th to 75th percentiles of observed dry aerosol extinctions.Dry extinctions are reported at standard temperature and pressure (273.15K, 1 atm).We multiply the modeled nitrate extinction by 0.8 to account for the evaporation of ammonium nitrate in the nephelometer(Bergin et al., 1997).The different model sensitivity experiments are described in Table2.
10 (0.71) * Model results are averaged from 2008 to 2010, while we use observations from 2006 to 2012, except for MODIS and MISR (2008-2010) and NH 3 observations in the USA (2007-2014).Detailed seasonal comparisons are presented in the Supplement.

Figure 7 .
Figure 7. Simulated degree of limitation of NH 4 NO 3 formation by HNO 3 (GR > 1) weighted by NH 4 NO 3 concentration at different pressure levels in AM3N for 2010.

Figure 8 .
Figure 8. Observed and simulated monthly AOD at 550 nm in different regions averaged over the 2008-2010 period.Circles show observations from MODIS (open circles) and MISR (filled circles).The solid and dashed black lines show the AOD simulated by AM3N_fdep_diu and AM3 respectively.We also show the simulated optical depths of sulfate (red), nitrate (cyan), dust (brown), organic carbon (green), black carbon (purple), and sea salt (blue) in AM3N_fdep_diu.The model is sampled to match the location and time of valid measurements by both MODIS and MISR in each region.Correlations between simulated and observed AOD are shown inset for AM3N_fdep_diu and AM3 (in parentheses).

Figure 11 .
Figure 11.Nitrate optical depth at 550 nm over the United States, Europe, China, and India for 2008-2010 (white bars) and 2050 (black bars) anthropogenic emissions for different configurations of AM3N.The thin red bar indicates the nitrate optical depth calculated using RCP8.5 2050 NH 3 emissions in AM3N.The relative changes between 2008-2010 and 2050 in NO − 3 optical depth and surface emissions of NH 3 , SO 2 , and NO are indicated for each region.

FFigure 12 .
Figure 12.Annual zonal mean distribution of NO −3 in AM3N with 2008-2010 anthropogenic emissions (top) and 2050 anthropogenic emissions (from RCP8.5 except for NH 3 ; see text).The blue, green, red, and cyan regions denote the NO − 3 burden located above 800 hPa, between 600 and 800 hPa, between 400 and 600 hPa, and below 400 hPa, with the partial burden in each pressure range indicated inset.The annual mean zonal burdens of NO − 3 simulated using AM3N_fdep_diu (dashed line), using AM3N with anthropogenic emissions from RCP8.5 for NH 3 (dotted line), using AM3N_ndust (dashed dotted line), and using AM3N with reduced convective removal of NH 3 (solid line) are also shown.The white circles in the top panel indicate the 2000 annual zonal mean NO −

Table 1 .
Simulated budget of SO 4 , NH x , and NO y in 2010.
x NH 3 emission (Tg N a −1 ) b

Table 2 .
Configurations of AM3N used in this study.
and [NH 3 ] in the USA (IMPROVE and AMoN) and Europe (European Monitoring and Evaluation Programme (EMEP)), [NH x ] and [HNO 3 ]

Table 3 .
Normalized mean bias and correlation coefficient (in parentheses) of monthly model results vs. measurements of surface concentrations of SO 2− 4 , NO − 3 and HNO 3 , NH 3 and NH x , concentrations of SO 2− 4 , NH + 4 , and NO − 3 in rain, and total aerosol optical depth at 550 nm from AERONET, MISR, and MODIS * .