This study uses the two-way coupled WRF-CMAQ (WRF v3.8 and CMAQ v5.2) as the base platform to integrate the SOA-NH₃ uptake process to conduct regional meteorology and air quality simulations. WRF is the meteorology driver of the model, while CMAQ handles the air quality dynamics. A detailed description of the two-way coupled WRF-CMAQ model is presented in Wong et al. (2012). The publicly available version used in this study supports the Rapid and accurate Radiative Transfer Model for General Circulation Models (RRTMG) radiation scheme (Clough et al., 2005) for shortwave aerosol direct effects. It uses a core-shell model to perform the aerosol optics calculation. The aerosol indirect effects that result from interactions between aerosols and cloud formation are not considered in this version.

For WRF configuration, the Asymmetric Convective model version 2 (ACM2) (Pleim 2007) is used as the planetary boundary layer scheme with the Pleim-Xiu land surface model (Xiu and Pleim 2001). The Morrison double-moment scheme (Morrison, Thompson, and Tatarskii 2009) is used for the microphysics option of WRF, and version 2 of Kain-Fritsch convective parametrization (Kain 2004) is used for cumulus physics. Both longwave and shortwave radiations are solved by the RRTMG scheme (Iacono et al., 2008). For the current scenarios, WRF input is derived from the National Centers for Environmental Prediction (NCEP) North American Mesoscale Forecast System (NAM) 12 km analysis data (NCEP 2015). For the future (year 2050) climate scenarios, the RCP 8.5 National Center for Atmospheric Research (NCAR) Community Earth System Model (CESM) global bias-corrected CMIP5 dataset (Monaghan et al., 2014) is used to provide a midterm outlook and the worst scenario estimation, where CAM-Chem (Lamarque et al., 2012) is used for chemistry-climate modeling and MOZART as the chemical mechanism (Emmons et al., 2010).

The CMAQ model is configured using the Carbon Bond 2006 (CB06) chemical mechanism for gas-phase chemistry (Yarwood et al., 2010), including 127 species as detailed on the model’s website (Adams 2017). The aerosol dynamics are solved by the sixth-generation CMAQ aerosol module (AERO6) (Appel et al., 2013), which includes 21 inorganic species and 34 organic species (28 SOA and 6 primary organic species) as detailed on the CMASWIKI website (Pye 2016). The modeling domain is the same as Zhu et al. (2018), which covers the contiguous US (CONUS) with a 12 km × 12 km horizontal-grid resolution and a 29-layers logarithmic vertical structure. The initial and boundary conditions are downscaled from the Model for Ozone and Related Chemical Tracers (MOZART v4.0) (Emmons et al., 2010). Emissions are generated based on the 2014 National Emissions Inventory (NEI) version 2 (US EPA 2018) and spatially/temporally resolved using the Sparse Matrix Operator Kernel Emission (SMOKE, version 4.5) processor (US EPA 2017). Biogenic emissions are calculated using the Biogenic Emission Inventory System (BEIS) (Schwede et al., 2005). As a preliminary study and only accounting for the change in climate alone, the same emissions are used for both 2014 and 2050 cases.

The SOA-NH₃ uptake mechanism is incorporated into the AERO6 module using the same method as our previous study (Zhu et al., 2018). In AERO6, all particles are considered spherical and internally mixed, which means a homogeneous distribution of all chemical substances within the particle. However, as the uptake coefficient used in this study is measured from pure SOA particles, how the uptake coefficient could be changed with SOA mass ratios within the particle is unknown. In the absence of better information, we assume that the uptake coefficient is proportional to the SOA mass fraction within the particle. In general, the uptake of NH₃ by SOA is calculated based on the representative wet surface area concentration of SOA (S_SOA) and the reactive uptake coefficient γ . The calculation of S_SOA is based on the SOA mass ratios within the particle. Details of the calculation are presented in our previous work (Zhu et al., 2018). In general, the amount of NH₃ uptake by SOA is reduced from the bulk NH₃ concentration at each timestep before the all-other chemical mass-transfer calculation, and the mass of organic aerosol is kept constant during the process. The effective first-order rate constant for of the NH₃ uptake by SOA is calculated as follows: k = γ × v N H 3 × S S O A 4 where v N H 3 is the average speed of NH₃ molecules (609 ms⁻¹ at 298 K). This first-order rate constant is then multiplied by the gas-phase NH₃ concentration to determine the loss rate of NH₃ in each grid cell at each time step. As explained in Zhu et al. (2018) in this method, the direct change of SOA mass due to the NH₃ uptake is neglected to simplify the calculation. This assumption is supported by the experimental observation that SOA particles exposed to ammonia in a smog chamber did not change their size distribution but showed clear evidence of incorporation of organic nitrogen into the particles in online and offline mass spectra, as described in Horne et al. (2018). As the existing laboratory data are still insufficient to determine the exact uptake coefficient for individual SOA species, a range of uptake coefficient γ is selected for sensitivity studies in our previous work (Zhu et al., 2018) between γ = 10 − 5 to γ = 10 − 3 based on the values reported (Y. J. Li et al., 2015). To avoid redundant sensitivity analysis, we only select the highest uptake coefficient γ = 10 − 3 in this study to provide the largest possible impact estimation of the SOA-NH₃ uptake process.

In total, four simulations with two-way feedback are performed in this study, two for 2014 and two for 2050. For each year, there is one case considering the SOA-NH₃ uptake and one case without the uptake. Table 1 summarizes the naming and definition of these six simulation cases. July is selected as the simulation period, with the first week discarded as a spin-up. Our previous study (Zhu et al., 2018) showed that SOA-NH₃ uptake produced an additional impact on acid-catalyzed SOA formation during the summer (Pye et al., 2013), suggesting that July is a good choice for this simulation.

Model validation and performance analysis are quantified using the Atmospheric Model Evaluation Tool (AMET) (Appel et al., 2011) developed by the US EPA. AMET organizes, provides consistency, and speeds up the evaluation process for operational meteorological and air quality model simulations. It can match the model output for particular locations to the corresponding observed values from one or more monitor networks. For meteorological model evaluation, observations are obtained from NCEP Meteorological Assimilation Data Ingest System (MADIS) (NCEP 2019) and Baseline Surface Radiation Network (BSRN) (AWI 2019). For air quality model evaluation, measurement network data provided by AMET includes AErosol RObotic NETwork (AERONET) (NASA 2019), Ammonia Monitoring Network (AMoN) (NADP 2019), Air Quality System (AQS) (US EPA 2019a), Clean Air Status and Trends Network (CASTNET) (US EPA 2019c), Chemical Speciation Network (CSN) (US EPA 2019b), Interagency Monitoring of Protected Visual Environments (IMPROVE) (USGS 2019a), National Atmospheric Deposition Program (NADP), National Air Pollution Surveillance (NAPS) (Canada 2019), South-Eastern Aerosol Research and Characterization (SEARCH) (USGS 2019b), Tropospheric Ozone Assessment Report (TOAR) (Schultz et al., 2017). The model performance is evaluated and compared for the two 2014 cases in section 3.1.

The model performance of the 2014 simulations with and without ammonia uptake is evaluated for PM_2.5, NH₄ ⁺, and Organic Carbon (OC) against multiple observation networks detailed in section 2.2 (Table 2). The PM_2.5 model performance of both Base_14 and UpTk_14 satisfies the recommended performance criteria proposed by Emery et al. (2017), with normalized mean bias (NMB) < ± 30%, normalized mean error (NME) <40% and correlation >40%. The PM_2.5 is underestimated by 22.14% in the base case, while the NH₃-SOA uptake process reduced such underestimation bias to 17.08% across the U.S. The overall model error is also reduced, as most model improvements occur over the southeastern US, as presented in Figure 1A. The model performance for NH₄ ⁺ is also largely improved after including the NH₃-SOA uptake process, with model bias reduced from −37.01% to −27.39%, model error reduced from 44.43 to 36.86% and correlation increased from 47.74 to 56.91%. With the NH₃-SOA uptake process included in the model, the underestimation of OC is also significantly improved, and the model bias is reduced from −17.51% to −6.96%. With more than 60% of the observation sites show decreased model error, Figure 1B shows the OC prediction is largely improved over the Southern and Eastern US, especially for the northeastern metropolitan regions like New York and Washington D.C. As underestimating OC is a common problem for air quality models (J. Li et al., 2017) such an improvement on OC prediction is promising. It confirms the importance of including the NH₃-SOA chemistry in air quality models. However, the NH₃-SOA uptake process only has little to no impact on the meteorological side of model performance [Supplementary Figure S1 in the Supporting Information (SI)], such as the temperature and relative humidity (RH), only slight improvement is found for the stational pressure estimation. Although the overall averaged impact is minimum, impacts on some specific locations could still be visible and could influence air pollutant distribution through feedback. Those feedback impacts are driven by the changes in the short-wave radiation balance due to the changes in aerosol concentrations (Clough et al., 2005). In the following section, the impact on meteorological parameters will be more carefully examined.

By comparing simulations UpTk_14 and Base_14, more details can be revealed on the impact on NH₃ and PM_2.5. Figure 2A shows the time-averaged NH₃ concentration over the simulation domain for Base_14. Hot spots can be identified in regions with high agricultural NH₃ emissions (e.g., California, North Carolina, Iowa, and Idaho, Supplementary Figure S2 in the SI) or regions with intense wildfire activities (e.g., Washington, Oregon, and Texas, Supplementary Figure S3 in the SI). Figure 2B shows the NH₃ concentration difference between UpTk_14 and Base_14. The inclusion of the NH₃-SOA uptake process results in an overall decrease in NH₃ concentration throughout the domain, with the most significant decrease of almost 30% in California. The enhanced NH₃ reduction for California is most likely due to the unique spatial overlap between domains of high NH₃ and SOA concentrations (Supplementary Figure S4 in the SI). Among the states with top NH₃ emissions, Iowa is the only one increasing NH₃ concentration after including the NH₃-SOA uptake. A more detailed investigation shows this is most likely due to the meteorological feedback resulting from the changing air quality conditions. Figure 2D shows the changes of 10-m wind speed for Iowa after the addition of NH₃-SOA uptake. The overall reduction in wind speed indicates a loss in dispersion capability, especially over the northwest part of the state, where most of the NH₃ emissions are located (Figure 2C). As a result, more NH₃ is accumulated over those regions, resulting in the unusual increase of NH₃ concentrations (Figure 2E). This phenomenon also implies that the most effective meteorological factor that could impact the atmospheric chemical concentrations from the chemistry-atmosphere feedbacks would be the change in atmospheric dispersion capability.

The inclusion of NH₃-SOA uptake also resulted in significant changes on modeled PM_2.5. Figure 3A shows the changes of PM_2.5 and its components (e.g., NH₄ ⁺ and SOA) for the entire CONUS and four states with the highest NH₃ emissions. In general, PM_2.5 and SOA increased after the inclusion of NH₃-SOA uptake, while NH₄ ⁺ decreased. Among them, North Carolina shows the most significant decrease in NH₄ ⁺ concentration and the largest increase in SOA. However, western states like California and Idaho exhibit both NH₄ ⁺ and SOA decreases after implementing the NH₃-SOA uptake. Figure 4A shows the mean PM_2.5 concentration for Base_14. The high concentration along the southeast boundary of the domain is caused by foreign PM_2.5 that originated from the western Sahara Desert in Africa.

At the same time, some hot spots with high PM_2.5 concentrations are caused by wildfires, like those in Washington, Oregon, and Texas. The result of the PM_2.5 distribution shows a clear pattern with the majority of the pollution happening over the eastern part of the CONUS, except for California. Figure 4D presents the change in PM_2.5 concentration after introducing the NH₃-SOA uptake. The overall impact on PM_2.5 is smaller than the impact on NH₃. The 4% increase of domain-wide PM_2.5 concentrations is still significant, with peak changes as large as 160 μg/m³. Most of the increases occur over the eastern part of the CONUS, overlapped mainly with the region with high PM_2.5 concentration, except California. Such a distribution pattern is consistent with previous studies (Horne et al., 2018; Zhu et al., 2018). More detailed investigation shows that almost half of the PM_2.5 changes result from the increase of SOA concentrations (Figure 4E). The most impacted region states close to the lower Mississippi River (e.g., Louisiana, Alabama, Mississippi, and Georgia), which also have elevated SOA background concentrations (Figure 4B). The drive behind the increase of SOA concentration is a change in particle acidity as more NH₃ is consumed by the NH₃-SOA uptake process (Figure 4C). The acid-catalyzed SOA formation pathway of isoprene epoxydiols is enhanced by increasing particle acidity (Pye et al., 2013), which is also observed in our previous works (Zhu et al., 2018; Wu et al., 2021).

Interestingly, the situation is different in Florida, where a decrease in PM_2.5 and SOA concentration is predicted. Like the anomaly of increasing NH₃ in Iowa, this particular case also results from the meteorological feedback effects. Figure 4F shows the changes of 10-m wind speed after adding the NH₃-SOA uptake process, where a substantial increase in atmospheric dispersion is observed over Florida. Such an increase in wind speed clears out PM_2.5 and decreases its overall concentration. The exact magnitude correlations between the change in wind and air pollutant concentrations are likely to be affected by multiple factors (e.g., boundary lay high, topographic, etc.) and would be an interesting subject for future investigations.

Changes in background meteorological conditions under RCP 8.5 climate scenarios largely altered the atmospheric chemistry conditions in the model. From the SI, Supplementary Figure S5 shows the meteorological and air quality changes between Base_50 and Base_14. First, there is a general increase in the surface temperature and a decrease in 10-m wind speed over the continent. The particle acidity also experiences dramatic changes. Furthermore, there is some substantial increase in pH values over the southeastern of the US, largely overlapping with the high SOA region in Figure 4B. For air pollutants, the overall NH₃ concentration is decreased by 2.1%, with significant decreases in Washington, Oregon, and Texas. Decreases in those states have a direct connection with wildfire emissions. The plume rise or vertical distribution of wildfire emissions are calculated inline with the CMAQ model based on a Briggs plume rise formulation (Briggs 1982), which means the change in meteorological conditions (e.g., temperature, wind speed, and mixing height) could broadly impact the distribution of related emissions. In addition, the change in dispersion directions alters the distribution of NH₃ concentrations in California and North Carolina. Also, some significant increase in PM_2.5 concentration are observed in the southeast corner of the simulation domain, most likely due to the spread of the west Sahara original sandstorm from the boundary conditions under the new meteorology. Interestingly, SOA concentration decreases significantly over the southeastern CONUS, leading to a 15% decrease domain wide.

Figure 3B shows the impacts on air pollutants after including the NH₃-SOA uptake in the 2050 climate scenario. Compared to the impacts for the 2014 case (Figure 3A), the domain-wide NH₃ reduction is 30% smaller under the future climate scenario. One reason is that the lower background NH₃ concentration produced less NH₃ available for uptake. Another more important reason is the meteorological feedback due to the NH₃-SOA uptake changed the wildfire emission, and the gains in wildfire emissions counterbalanced the reduction caused by the NH₃-SOA uptake (Figure 5A). The significant increase in pollutants concentrations in Idaho is also caused by wildfire emissions (Figures 5A–D). For Iowa, the NH₃ anomaly is largely reduced as the overall wind speed increased by 1.7% (Figure 5F). Due to the decrease in background SOA concentration, the impact of SOA (Figure 5C) and particle acidity (Figure 5E) over the southeastern CONUS due to NH₃-SOA uptake is largely reduced compares to the 2014 case (Figure 4). Even with the additional gain from altered wildfire emissions, the increase of SOA is only about half of the 2014 case, and the decrease in pH is only 47%. Even for California, the impact caused by the NH₃-SOA uptake is also largely reduced. Only North Carolina shows similar impacts to the 2014 cases. Although, climate change could also impact the upstream emissions that have not been considered in this study. These preliminary results imply that under a warmer climate, the impact of the NH₃-SOA uptake might be reduced compared to current climate conditions. Also, the impact on smoke plume rise characters could be a significant factor during the meteorology and air quality two-way feedback, as observed in this study.