**1. Introduction**

Nitrate aerosol is a fine particulate matter (PM2.5) component produced from the reaction of gas-phase nitrate (nitric acid; HNO3) and ammonia (NH3). During a haze event, nitrate aerosol often contributes to the total observed particle mass by as much or more than the organic fraction across East Asia [1–4]. As severe haze events have increased across East Asia [5], nitrate contributions to PM2.5 mass have also increased in polluted urban areas [1,6].

The production of nitrate aerosol in urban areas is a ffected by vehicular emissions such as nitrogen oxides (NOx = NO + NO2) and NH3 [7–9]. These vehicular emissions are highly concentrated and are transported by turbulence from the complex geometry of buildings, causing steep gradients of pollutant concentrations [10–13]. Nitrate aerosol chemistry is highly nonlinear and follows the concentrations of precursor gases and humidity, so the formation and distribution of nitrate aerosol are not spatially uniform [14,15]. Most modeling studies on nitrate aerosols are based on regional or global air-quality models that have clear limitations due to their coarse resolution [16–18]. Therefore, to accurately investigate nitrate formation in urban areas, fine-scale simulations that can conserve highly concentrated emission plumes and turbulence are necessary.

Meanwhile, policies regarding vehicular nitrate control focus on reduction of NOx (e.g., through banning diesel vehicles) [19,20], based on studies of regional or global air-quality modeling. However, studies from observation campaigns have often reported that other vehicle emissions are much more important than NOx emissions for nitrate production in urban areas [21–23]. Link et al. investigated the secondary formation of ammonium nitrate from vehicle exhaust using sampling and laboratory experiments in the Seoul Metropolitan Region (SMR) [21]. They found that the secondary production of ammonium nitrate from diesel exhausts is much lower than that from gasoline. They concluded that the NH3 source from gasoline vehicles could be more important than NOx emissions, indicating that SMR is an NH3-limited environment. Wen et al. studied nitrate formation during a severe PM2.5 pollution period and reported that high NH3 concentrations in the early mornings significantly accelerated the formation of fine particulate nitrate [23]. In addition, they found that the increased rate of nitrate aerosol had a strong positive correlation with ozone (O3) concentrations at night, indicating the essential role of oxidants in nitrate formation. Studies also reported that volatile organic compound (VOC) concentrations control nitrate formation by a ffecting the O3 levels [22,24]. These studies indicate that nitrate aerosols are a ffected by a complex chemical condition that involves NOx, NH3, and oxidants, whereas policy tends to focus exclusively on NOx control. Thus, understanding the favorable conditions for nitrate formation in urban areas is crucial for the design of air-quality policies.

Thus, this study investigates the distribution of nitrate aerosols and the favorable conditions for nitration formation in urban streets using a microscale coupled computational fluid dynamics (CFD)–chemistry model that can reproduce the turbulence from complex building geometries. Sensitivity simulations were conducted to examine the sensitivity of emissions to nitrate production by changing the emissions of precursor gases for nitrate aerosols and oxidants. The sensitivity simulation results reveal what significant factors lead to nitrate aerosol problems in urban streets.

#### **2. Model Description and Simulation Set-Up**

## *2.1. Model Description*

A coupled CFD–chemistry model was used based on that proposed by Kim et al. [25]. The CFD model is based on the Reynolds-averaged Navier–Stokes equation (RANS) model and assumes a three-dimensional (3-D), nonrotating, nonhydrostatic, and incompressible airflow system [26]. This model was previously used to examine the flow and dispersion of both reactive gas pollutants [25,27] and reactive aerosol pollutants [25,28].

The model's chemical mechanism includes a full tropospheric NOx–Ox–VOC chemistry scheme from a global 3-D chemical transport model (GEOS-Chem V11-1) [29]. GEOS-Chem was initially developed to solve global air chemistry issues; however, application of the GEOS-Chem model has now been extended to the regional scale. The GEOS-Chem model can successfully explain urban air quality, including cases of severe haze over East Asia [1,30,31]. The chemical scheme includes 140 species and 393 reactions, among which 61 reactions are photochemical. Among the 140 species simulated in the chemistry module, the CFD model transports 65 chemical tracers. Radical species with very short chemical lifetimes are not transported. Photolysis rate coe fficients are calculated using the Fast-JX radiative transfer model [32,33].

The model also calculates aerosols that include sulfate, nitrate, ammonium, black carbon, and organic carbon [34,35]. Sulfate formation generally occurs via two pathways: the gas-phase oxidation of SO2 by OH and the aqueous-phase oxidation of SO2 by ozone and hydrogen peroxide. The CFD model only accounts for the gas-phase oxidation of SO2 by OH because it lacks an atmospheric physics module that simulates hydrometeors such as clouds and rain.

Nitrate and ammonium aerosol were calculated by partitioning the total NH3 and HNO3 between the gas and aerosol phases. We used the ISORROPIA-II model as a thermodynamic equilibrium model for aerosol partitioning [36,37] and employed it to calculate the thermodynamic equilibrium of a K<sup>+</sup>–Ca2<sup>+</sup>–Mg<sup>2</sup><sup>+</sup>–NH4 <sup>+</sup>–Na<sup>+</sup>–SO4 <sup>2</sup>−–NO3 −–Cl−–H2O aerosol system based on the NH3, HNO3, and SO4 2− concentrations.

The model also includes the production of HNO3 via heterogeneous chemistry between aerosols and gases following Jacob (2000) [38]. The reactions that contribute to HNO3 production by heterogeneous chemistry can be written as:

$$2\text{ NO}\_2 \rightarrow \text{HNO}\_3 + \text{HONO} \tag{\text{R1}}$$

$$\text{NO}\_3 \rightarrow \text{HNO}\_3 \tag{\text{R2}}$$

$$\text{N}\_2\text{O}\_5 \to 2 \text{ HNO}\_3.\tag{\text{R3}}$$

The uptake coefficients, γ, of R2 and R3 were 10−<sup>4</sup> and 0.1, respectively, following Jacob (2000) [38]. For R4, γ was set to 0.01 as suggested by Zhang et al. (2012) and Walker et al. (2012) [39,40]. N2O5 is essential for night-time nitrate aerosol chemistry; the production and loss reactions of N2O5 can be written as:

$$\rm NO\_2 + NO\_3 + M \rightarrow N\_2O\_5 + M \tag{R4}$$

$$\rm N\_2O\_5 + M \to NO\_2 + NO\_3 + M \tag{R5}$$

$$\rm N\_2O\_5 + hv \to NO\_3 + NO\_2. \tag{R6}$$

The simulation of carbonaceous aerosols follows the GEOS-Chem model [35]. The primary carbonaceous aerosol follows the passive tracer without any chemical reactions. However, the model resolves primary Black Carbon (BC) and Organic carbon (OC) with a hydrophobic and a hydrophilic fraction for each (i.e., making four aerosol types) for deposition processes. All sources emit hydrophobic aerosols that then become hydrophilic with an e-folding time of 1.2 days, as per Cooke et al. (1999) [41]. Although secondary organic aerosol (SOA) chemistry is not considered in the model, we treat boundary inflow and the transport of SOAs. However, the model does not account for either dust or sea-salt aerosol.

The dry deposition of gases and aerosols was simulated with a standard big-leaf resistance-in-series model [42]. The model accounts for the dry deposition of 46 species, including aerosols. The CFD model has no atmospheric physics module that simulates hydrometeors (e.g., clouds and rain); thus, wet deposition was not calculated, following Kim et al. (2012) [25].

## *2.2. Simulation Set-Up*

We assumed a street canyon selected for a simulation located in SMR, one of the most polluted cities, to investigate nitrate formation in urban streets. The domain size was 120 × 80 × 100 m in the x, y, and z directions, respectively. The grid intervals in all directions were 2 m, and the building height was 20 m with unified aspect ratios for all street canyons. Figure 1 shows the detailed structure of the simulation domain.

First, we set a control run (CNTL hereafter) with standard emissions. We estimated the pollutant emissions from average traffic volume in 2017 obtained from the Traffic Monitoring System (http: //www.road.re.kr), which provides traffic statistics in SMR. On that basis, the daily traffic volume for each street was assumed to be 15,130 vehicles day−1. The monthly averaged diurnal variation in traffic volume was also obtained from the Traffic Monitoring System. Vehicular emissions were computed using the emission factor from the Clean Air Policy Support System (CAPPS) emission inventory [43] and calculated by multiplying the mean ratios of vehicle sizes in Korea. The emission factors following vehicle size and the ratios of vehicle size are summarized in Tables 1 and 2, respectively. The calculated averaged emissions per vehicle were 0.10 g km−1, 0.15 g km−1, 0.015 g km−1, and 0.011 g km−<sup>1</sup> for NOx, CO, VOC, and NH3, respectively. NOx emissions were separated into NO and NO2 emissions at a 10:1 ratio by volume [44]. Total VOC emissions were further speciated using the method developed by EMEP/EEA (2016) [45]. Table 3 lists the emission rate and ratios of speciated VOC in the CNTL simulation. All vehicular emissions were emitted at z = 1 m from 4 m wide area sources located at the center of the streets. Other emissions in the model domain were not considered

in the simulation. Previous studies have reported that the largest proportion of emissions in SMR is vehicular emissions [43,46,47].

**Figure 1.** Schematic diagram of the coupled computational fluid dynamics (CFD)–chemistry simulation domain for the control run (CNTL) simulation.

**Table 1.** Emission estimates from vehicles used in the coupled CFD–chemistry model simulations for the CNTL simulation. The text discusses details about the species emission estimates.


**Table 2.** Ratios of vehicles following size and fuel type used in the CNTL simulation obtained from the Traffic Monitoring System.


For meteorological conditions, we used values for the SMR in winter that provided favorable conditions for nitrate formation [48]. To simulate diurnal changes in buoyancy and their effects on transport and chemical reaction rates in the model, we used the hourly temperature and relative humidity obtained from the Seoul station of the Korea Meteorological Administration (http://web. kma.go.kr/eng/). Table 4 shows the hourly temperature and relative humidity used in this study. The wind speed on the rooftop was assumed to be the observed seasonal mean value for SMR, which is 2.4 m s<sup>−</sup>1. The wind direction is westerly (most frequently observed in winter) and is perpendicular to

the street canyon [49]. The ambient wind speed and direction were fixed during a one-day simulation. The following vertical profiles of the wind, turbulent kinetic energy (TKE), and TKE dissipation rate were imposed:

$$\mathcal{U}\mathcal{U}(z) = \begin{array}{c} \mathfrak{u}\star\cos\theta\\ \kappa \end{array} \ln\left(\frac{z}{z\_0}\right) \tag{1}$$

$$V(z) = \frac{u\_\* \sin \theta}{\kappa} \ln \left(\frac{z}{z\_0}\right) \tag{2}$$

$$W(z) = \begin{cases} 0 & \text{if } z = 0\\ \end{cases} \tag{3}$$

$$k(z) = \frac{u\_\*^2}{C\_\mu^{1/2}} \left(1 - \frac{z}{\delta}\right)^2\tag{4}$$

$$\kappa(z) = \frac{\mathbb{C}\_{\mu}^{3/4} k^{3/2}}{\kappa z} \tag{5}$$

Here, *u*<sup>∗</sup>, *z*0, and κ represent the friction velocity, roughness length (=0.05 m), and von Karman constant (=0.4), respectively; *C*μ is an empirical constant (=0.0845), and θ is the wind direction. The surface and top boundary pressures in the model were assumed to be 1013.15 hPa and 993.72 hPa, respectively.

**Table 3.** Emission rates per vehicle and ratios of speciated volatile organic compound (VOC) following the method developed by EMEP/EEA (2016).


**Table 4.** Diurnal variations of the observed hourly surface temperature and relative humidity used in this model.


For the rooftop boundary conditions of the species, we used a reanalysis dataset from the Monitoring Atmospheric Composition and Climate (MACC) project in SMR with 6 h diurnal variations [50]. In addition, we used the boundary conditions of the species from the GEOS-Chem simulation with 6 h diurnal variations at SMR in winter for the species that were not provided in the MACC reanalysis dataset [51]. The initial conditions of the species were assumed as the concentrations of the boundary conditions in the first step. We conducted 48 h model simulations for each case: the first 24 h for the model spin-up, and the results from the last 24 h were used. The chemical and dynamical time steps were 1 min and 1 s, respectively.

Sensitivity simulations were conducted to examine the effect of emissions on nitrate aerosols in urban streets. Twelve sensitivity simulations were set by changing the vehicular emissions of NOx, VOC, and NH3. Each simulation was conducted with different emissions by multiplying the original emission by 0.25, 0.5, 2.0, and 4.0 for each species. We named the sensitivity simulations "species name" × "multiplying factor." For example, simulations named NOx × 0.25, NOx × 0.5, NOx × 2, and NOx × 4 indicate multiplying the vehicular NOx by 0.25, 0.5, 2.0, and 4.0, respectively, while other emissions were fixed.

#### *2.3. Model Validation*

The coupled CFD–chemistry model in this study was thoroughly validated for the flow and dispersion of passive tracers in street canyons by comparing the results from this model with those from a wind tunnel, an idealized numerical study, and fluid experiments [25,27,52]. Park et al. (2015) found good agreements when comparing the model with wind tunnel data and experimental data by implementing improved wall functions for the momentum and thermodynamic energy equations in the CFD model to more accurately represent the effects of solid–wall boundaries [27].

The coupled CFD model also evaluated the dispersion of reactive pollutants compared with idealized simulations and field campaigns [25]. Kim et al. (2012) applied the coupled CFD–chemistry model to simulations using the same building configuration as in Baker et al. (2004) [25,53]. Their results showed that the concentrations of NOx and O3 have a pattern and magnitude consistent with the simulated concentrations by Baker et al. (2004) under steady-state O3–NO–NO2 photochemistry. Kim et al. (2012) reproduced reactive pollutants on Dongfeng Middle Street, Guangzhou, China, using a full tropospheric NOx–Ox–VOC chemistry scheme and compared the results to a field campaign by Xie et al. (2003) [13]. The coupled model, with the full photochemical mechanism, also successfully captured the time variation in the observed CO concentrations for both upwind and downwind sites in the Dongfeng Street canyon. However, the coupled model overestimated NOx concentrations compared to observations by Xie et al. (2003) due to estimating excessive NO emissions from traffic volume, implying the necessity of utilizing an accurate emissions inventory [13].

The coupled CFD model has also been used to evaluate the dispersion of reactive aerosol in street canyons [28]. Kim et al. (2019) evaluated the composition of PM1 in summer and winter in a street canyon by comparison with the field campaign in Elche, Spain, by Yubero et al. (2015) [28,54]. The model generally captured seasonal variations of PM1 in the street canyon. We evaluated the seasonal variation in nitrate concentration by comparing our model results to those of Yubero et al. (2015) [54]. Four simulations were conducted to represent the four seasons (spring, summer, autumn, and winter) in Elche, Spain. The street was approximately 7 m wide and surrounded by buildings that were approximately 25 m in height. The domain size was 20 m × 40 m × 50 m, and the number of grid points was 42 × 82 × 52. The meteorological conditions used were the observed seasonal mean values during the campaign periods. Pollutant emissions were estimated from traffic volume obtained from the Elche Traffic Office [54]. Vehicular emissions were computed using Spain's emission rates in EMEP/EEA (2016). The detailed model configuration generally followed that of Kim et al. (2019) [28].

Figure 2 shows the observed and simulated nitrate concentrations by season. The observed nitrate concentrations were highest in winter and lowest in summer due to the thermal evaporation of nitrate aerosols, showing 1.3 and 0.3 μg m<sup>−</sup>3, respectively. The observed nitrate concentrations in spring and autumn were 0.4 and 0.5 μg m<sup>−</sup>3, respectively, falling between those in winter and summer and indicating the dominant effect of temperature on nitrate aerosol. The simulated nitrate concentrations were 1.0, 0.8, 1.0, and 1.4 μg m<sup>−</sup><sup>3</sup> in spring, summer, autumn, and winter, respectively, showing clear seasonal variation. The simulated concentrations reproduce the observed seasonal variation in nitrate concentration. However, the simulated nitrate concentration in summer was twice that of the observed magnitude, indicating the weaker thermal evaporation of ammonium nitrate predicted by the model. The model captured the magnitude of the observed nitrate in winter located within the standard deviation of the observed nitrate concentration. These results indicate that the model successfully calculated the nitrate aerosol in cold environments.

**Figure 2.** Comparisons of nitrate concentrations in the spring, summer, autumn, and winter cases between this work and the previous results of Yubero et al. (2015) (see Figure 1 in Yubero et al.). The values are averaged for the analysis period at the sample station. The black solid line is the observed concentrations and the red dashed line is the simulated concentrations. The error bars indicate the standard deviations of the observed nitrate aerosol in each season.
