A modified version of the WRF-Chem model developed by Li et al. (2010, 2011a, b) is used in this study. WRF-Chem provides the online-coupled meteorology, transport, gas-phase chemistry, photolysis, deposition, and aerosol–radiation framework, whereas aerosol processes are represented by the CMAQ/Models-3 AERO5 modal aerosol module incorporated within WRF-Chem. Gas-phase chemistry is represented by the SAPRC-99 mechanism. The WRF-Chem simulations are conducted for November 2017 over the NCP. The base simulation (BASE) does not include the additional NOCs-forming NH3–SOA pathway. Three sensitivity simulations use the γ values of 10-3, 10-4, and 10-5 (S(-3), S(-4), and S(-5), respectively). Further details of the model configuration and simulation domain are provided in Sect. S1, Table S1 in the Supplement, and Fig. 1.

Primary NOC emissions are not explicitly resolved in the anthropogenic emission inventory, and the model does not include a separate primary-NOCs tracer. Some nitrogen-containing organic material may be implicitly included within bulk primary organic aerosol (POA) or lumped organic-gas emissions, but it cannot be separately speciated or diagnosed as NOCs. The modeled NOCs concentration therefore represents only the product of the implemented NH3–SOA reactive-uptake pathway and should not be interpreted as total ambient NOCs. Because the measured NOCs burden can include primary emissions and products from other secondary pathways, the model–measurement comparison provides a conservative upper limit for the contribution of this pathway.

The BASE simulation is evaluated using a comprehensive suite of observations, including hourly criteria pollutants (PM2.5, O3, NO2, and SO2) from national networks and daily filter measurements of aerosol chemical components in the NCP. In situ measurements of NH3 and NOCs and Infrared Atmospheric Sounding Interferometer (IASI) satellite products are also used to evaluate model performance. Measurement details are provided in Sect. S2. Model performance is quantified using mean bias (MB), root mean square error (RMSE), and the index of agreement (IOA), as described in Sect. S3.

Total heterogeneous uptake of a gaseous species by an aerosol particle may involve gas-phase diffusion, surface accommodation, solvation, condensed-phase diffusion, and chemical reaction (Crowley et al., 2010). The present parameterization does not resolve these steps separately or represent the complete particulate uptake and fate of NH3. For consistency with previous laboratory and modeling studies, we retain the notation γ but interpret it here as an effective coefficient for the implemented NOCs-forming pathway. A literature-based range of γ=10-5-10-3 is selected based on the laboratory results of Liu et al. (2015) and previous modeling studies (Zhu et al., 2018; Wu et al., 2021). The corresponding first-order loss rate is incorporated into the model for the sensitivity simulations and is calculated as 1κ=γ×υNH3×SSOA4 Here, γ is the net probability that an NH3 collision with an SOA-containing particle ultimately contributes to a particulate organic-N product through the implemented pathway. υNH3 is the mean molecular speed of NH3, and SSOA is the wet surface-area concentration of SOA.

In this study, modeled NOCs are a lumped representation of reduced-nitrogen organic products that may form when NH3 reacts with reactive organic components of SOA, including carbonyl- or dicarbonyl-containing compounds. After formation, the modeled NOCs undergo dry and wet deposition.

In the real atmosphere, NH3 can react with various SOA species to form NOCs. To simplify the simulation while retaining key chemical characteristics, SOA species are divided into three main categories based on their precursor origins: anthropogenic volatile organic compounds (VOCs), biogenic VOCs, and primary organic components. Each category is assigned a representative molecular weight, which is used in calculating NOCs concentrations. The molecular weights for these SOA categories are adopted from a previously established model framework (Li et al., 2011b).

Figure S1 in the Supplement presents the diurnal profiles of observed and simulated near-surface PM2.5, O3, NO2, and SO2 concentrations averaged across monitoring sites in the NCP during the simulation episode. The model well captures the temporal variations of these pollutants, with IOA values of 0.93, 0.84, 0.83, and 0.67, respectively. Figure S2 compares the simulated and observed spatial distributions of their mean concentrations. The simulated spatial patterns are generally consistent with observations from ambient monitoring stations. Detailed validation results are provided in Sect. S4.

The simulated mass concentrations of key aerosol components are also evaluated against observations, including sulfate, nitrate, ammonium, elemental carbon (EC), primary organic aerosols (POA), and SOA (Fig. 2). The model generally performs well in simulating the sulfate, nitrate and ammonium concentrations, with IOAs of 0.74, 0.87, and 0.85, respectively, but underestimates the sulfate concentration appreciably, with a mean bias (MB) of -2.7µgm-3. The model also reasonably reproduces POA, EC, and SOA concentrations, with the IOA of 0.59, 0.62, and 0.53, respectively. It is worth noting that the model tends to underestimate the carbonaceous aerosols, particularly with regard to POA with a MB of -2.8µgm-3.

The simulated NH3 vertical column densities (VCDs) show a generally good consistency with the IASI retrievals during the simulation period, with an IOA of 0.66 and MB of 0.26×1016moleccm-2, indicating that the model generally reproduces the observed NH3 VCDs with minimal systematic bias (Fig. 3a). Figure 3b displays the comparison of simulated and observed NH3 concentrations at Xianghe station. The model demonstrates a good performance in simulating near-surface NH3 concentrations, with the IOA of 0.66 and the MB of -0.01ppb. The monthly variation in NH3 emissions is strongly influenced by agricultural practices and environmental conditions such as temperature and wind speed. During the simulation period, the observed NH3 concentrations have exhibited several pronounced peak events in mid-November that the model fails to reproduce. These anomalies are likely associated with localized biomass burning activities (primarily crop and residential wood combustion), which are prevalent in the rural-urban fringe of the NCP during the transition to the heating season (Pan et al., 2018). These episodic emissions explain the discrepancy between simulations and observations.

Overall, the comprehensive validation against multi-source datasets indicates that the model exhibits a good performance in capturing the spatiotemporal variations of both criteria pollutants and NH3, providing a reliable foundation for the subsequent sensitivity analysis.

Figure 4a shows the simulated distribution of average near-surface NH3 concentrations during the simulation period. Quantitatively, the modeled NH3 concentrations in the NCP range from 0.24–50.9 µgm-3, with an average of 15.6 µgm-3. These simulated NH3 concentrations are within the range of observations reported in recent studies (Ianniello et al., 2010; Pan et al., 2018; Wang et al., 2022a; Zhang et al., 2021). For example, Ianniello et al. (2010) have reported that the observed daily average NH3 concentrations in Beijing range from 0.20–44.38 µgm-3, and Pan et al. (2018) have reported a regional annual average of 13.4 µgm-3 over the NCP in 2015. Notably, the simulated NH3 peak concentrations are concentrated in southern Hebei and eastern Henan provinces, which has been further verified by recent emission constraints, showing sustained high nitrogen loading in these specific agricultural corridors (Liu et al., 2022b; Yang et al., 2023).

Figure 4b–d shows the net changes in NH3 concentrations in S(-3), S(-4), and S(-5) relative to BASE. In S(-3) and S(-4) (Fig. 4b and c), the pathway decreases NH3 throughout the NCP. The domain-average decrease is approximately 31.4 % (4.9 µgm-3) in S(-3), comparable to the sensitivity reported by Wu et al. (2021), and 5.5 % (0.86 µgm-3) in S(-4), with a maximum local reduction of 26.7 %. In S(-5) (Fig. 4d), the decrease is not spatially uniform, and small positive changes occur in southeastern parts of the domain. Nevertheless, the domain-average NH3 concentration decreases by 0.62 % (0.096 µgm-3), with a maximum local reduction of 5.2 %.

Considering that NH3 serves as the sole precursor of ammonium aerosols, the perturbation of NH3 concentrations due to its uptake on SOA directly influences the formation of ammonium aerosols in the atmosphere. Figure 5a shows the pattern comparison of average simulated and observed concentrations of ammonium aerosols during the simulated episode. Both the observation and simulation show high ammonium concentrations in the south of the NCP, generally exceeding 8 µgm-3. In the S(-3) (Fig. 5b), although the NH3 uptake reaction decreases NH3 concentrations by over 30 % in the NCP, it only reduces the ammonium concentration by 5.5 % or 0.86 µgm-3. The ammonium aerosol is formed by neutralizing sulfuric and nitric acids in the atmosphere. The observed and simulated high NH3 level in the NCP indicates that NH3 is excessive after neutralizing inorganic acids to form ammonium salt. Hence, an over 30 % decrease in NH3 concentrations only causes a 5.5 % reduction in ammonium aerosols in the NCP, an NH3-rich region. However, in the NH3-poor region, i.e., in Shanxi province and over oceans, ammonium concentrations decrease considerably.

In S(-4) and S(-5) (Fig. 5c and d), ammonium responses are spatially heterogeneous, with a slight increase north of Beijing. On average, ammonium concentrations decrease by 0.48 % (3.1×10-2µgm-3) and 0.048 % (3.1×10-3µgm-3), respectively.

Although reactive uptake itself is strictly a sink for gas-phase NH3, the difference between the sensitivity and BASE simulations represents the net response of all coupled chemical, thermodynamic, and dynamical processes rather than the uptake tendency alone. A two-way coupled WRF–CMAQ study of the same NH3–SOA uptake process have similarly predicted a local increase in NH3 over Iowa, which is attributed to uptake-induced meteorological feedback that reduces the 10 m wind speed and atmospheric dispersion, thereby promoting NH3 accumulation in a high-emission region (Zhu et al., 2022). WRF-Chem studies over the NCP have also shown that aerosol–radiation feedbacks can perturb surface temperature, planetary boundary layer development, and wind fields, leading to spatial redistribution of near-surface pollutants (Bei et al., 2017; Wu et al., 2019). In addition, the ISORROPIA thermodynamic module recalculates the equilibrium partitioning between gaseous NH3 and particulate NH4+. Changes in temperature, relative humidity, aerosol liquid water, particle acidity, and nitrate availability may promote NH4NO3 dissociation and partially replenish gas-phase NH3 (Nenes et al., 1998; Pye et al., 2020; Nenes et al., 2020). Therefore, the small positive NH3 values in a limited number of grid cells do not indicate direct NH3 production by the uptake reaction, but likely reflect secondary redistribution, aerosol–radiation interactions and gas–particle repartitioning within the coupled modeling system.

The decrease in ammonium shifts the thermodynamic equilibrium of NH4NO3 toward the gas phase and can reduce particulate nitrate. The model generally reproduces the nitrate distribution (Fig. 6a), with concentrations exceeding 10 µgm-3 and accounting for more than 20 % of PM2.5 in parts of the NCP. In S(-3), nitrate decreases by 4.6 % (0.70 µgm-3) across most of the domain. The responses in S(-4) and S(-5) are smaller and spatially non-uniform because of coupled aerosol–radiation and thermodynamic feedbacks. The domain-average nitrate changes are -0.56% (-8.5×10-2µgm-3) and -0.037% (-5.6×10-3µgm-3), respectively.

The reduction in available NH3 can also lower aerosol pH, indirectly suppress aqueous-phase SO2 oxidation, and potentially decrease sulfate formation (Zhu et al., 2018; Zhang et al., 2020; Zhang et al., 2026). Because NH3 is abundant in the NCP, this pH perturbation is expected to be small. Decreases in nitrate and ammonium may also reduce aerosol liquid water (ALW), thereby hindering sulfate formation. The model generally underestimates observed sulfate (Fig. 7a), but the simulated sulfate changes induced by this pathway remain small: the domain-average decreases are 1.1 %, 0.16 %, and 0.059 % in S(-3), S(-4), and S(-5), respectively. At smaller γ values, the sulfate response becomes increasingly spatially heterogeneous (Fig. 7b–d).

In all sensitivity simulations, the NH3 loss calculated specifically for the implemented reactive-uptake pathway is converted into a particulate organic nitrogen product while SOA mass is not depleted. The simulated SOA concentration averages 6.8 µgm-3 in the NCP, although urban SOA is generally underestimated (Fig. 8a). S(-3) produces an unrealistically high mean NOCs concentration of 65.5 µgm-3, much higher than the simulated SOA concentration (Fig. 8b). The mean NOCs concentration in S(-4) remains high at 11.3 µgm-3 (Fig. 8c). By contrast, S(-5) produces 1.2 µgm-3, within the magnitude of the measured NOCs burden (Fig. 8d). To place this value in context, Wang et al. (2022b) reported an average NOCs concentration of 1.27 µgm-3 at a regional site in Hebei, and Wang et al. (2024) reported daily concentrations of 0.71–3.89 µgm-3 over the Tibetan region.

Figure 9 compares measured and simulated daily NOCs concentrations at Xianghe in November 2017. The observed monthly average NOCs concentration is 1.52 µgm-3, ranging from 0.65–2.24 µgm-3. In the S(-3), the model simulates unrealistically high monthly average NOCs concentrations of 85.5 µgm-3 at the station. In the S(-4), the simulated concentration is 14.7 µgm-3, which remains well above the observation. Only in the S(-5) with the γ of 10-5, the simulated concentration is 1.6 µgm-3, which is within the observed range. Therefore, although large γ values are expected to enhance NH3 uptake and accelerate NOCs formation, the γ of 10-5 might be more reasonable to simulate the NH3 uptake on SOA when the simulation is constrained by the observations of NH3 and NOCs as well as other air species.

Laboratory studies indicate that NH3 uptake by SOA is finite, with about 10 % or less of the SOA material reacting with NH3 to form reduced-nitrogen products. Liu et al. (2015) further showed that γ decreased from initial values of approximately 10-2-10-3 toward 10-5 during continued NH3 exposure. This decay was attributed to reduced accessibility of reactive sites and condensed-phase diffusion limitations as the particles aged. Accordingly, the very high NOCs concentrations in S(-3) and S(-4) should not be interpreted as sustainable atmospheric production. Rather, these cases illustrate the overestimation that can arise when large initial coefficients are applied continuously without explicitly resolving progressive depletion or shielding of reactive organic sites.

In the atmosphere, the measured NOCs burden is also influenced by processes other than the pathway examined here. Secondary pathways include reactions of ammonia or ammonium with particulate organic acids (Laskin et al., 2009; Wang et al., 2017; Zhang et al., 2025), while primary sources include biomass burning, coal combustion, vehicle emissions, biogenic emissions, and soil dust (Cheng et al., 2006; Harrison et al., 2005; Lu et al., 2019; Wang et al., 2022b). These additional sources further limit the fraction of the measured NOCs burden that can be assigned to the implemented pathway.

Considering both the laboratory evidence for finite and declining SOA reactivity and the ceiling provided by the measured NOCs burden, we interpret γ=10-5 as a conservative upper limit for sustained operation of the NOCs-forming pathway under the present conditions.

In S(-5), the pathway has an almost negligible effect on inorganic aerosol over the NCP. Under NH3-rich conditions, only a small fraction of NH3 is consumed by this pathway, leaving most NH3 available for neutralization of inorganic acids. Consequently, the domain-average PM2.5 concentration decreases by only 0.018 % (8.8×10-3µgm-3), and the spatial response is heterogeneous because of coupled aerosol–radiation and thermodynamic feedbacks (Fig. 10).

Discrepancies from previous studies (Zhu et al., 2018; Wu et al., 2021) may reflect differences in chemical mechanisms, aerosol treatments, simulation periods, seasons, and baseline chemical environments. All sensitivity cases in the present study use the same gas-phase mechanism, aerosol module, emissions, and meteorological configuration, so their differences isolate the perturbation introduced by the added pathway within this framework. The γ=10-5 case is used to characterize the air-quality response of an observationally conservative upper limit representation, not to derive a universal physical coefficient. Under the NH3-rich conditions examined here, this pathway has little influence on NH3, inorganic aerosol, or PM2.5. Even in S(-3) and S(-4), the domain-average PM2.5 decreases are only 0.99 % (2.0 µgm-3) and 0.26 % (0.12 µgm-3), respectively.