The sampling site is located at the State Key Laboratory of Atmospheric Boundary Layer Physics and Atmospheric Chemistry (LAPC), Institute of Atmospheric Physics (IAP), Chinese Academy of Sciences (CAS) in urban Beijing (39∘58′ N, 116∘22′ E; Fig. 1). The site is approximately 0.8 km north of the Third Ring Road, 1.3 km south of the Fourth Ring Road and 0.2 km west of the Beijing–Tibet expressway, which are three transport arteries encircling Beijing, each with average traffic volumes of over 200 000 vehicles day per day in 2016 (Beijing Transport Institute, 2017); therefore, this site represents a typical urban site that is mainly surrounded by residential areas.

From 16 March 2016 to 16 March 2017, weekly atmospheric NH3 samples were collected at 16 heights on the 325 m meteorological tower using ALPHA passive samplers (adapted low-cost high absorption, Centre for Ecology and Hydrology, Edinburgh, UK) except for a few samples with slightly different durations due to tower maintenance schedules. The samplers operate on the principle of diffusion using an acid-coated filter to capture the NH3. A PTFE (Teflon) membrane is placed directly at the mouth of the sampler, forming a quiescent boundary layer in front of the sample membrane. A stable, turbulent-free diffusion path length is achieved behind the membrane, whilst allowing gaseous NH3 to diffuse through for capture and minimizing the sampling of NH4+ aerosol (Tang et al., 2014). NH3 was sampled at 2, 8, 15, 32, 47, 63, 80, 102, 120, 140, 160, 180, 200, 240, 280 and 320 m a.g.l. (above ground level). At each height, three ALPHA samplers were deployed under a PVC shelter to protect the samplers from rain and direct sunlight (shown in Fig. 1). NH3 samples were extracted with 10 mL high-purity water (18.2 MΩ-cm) and analyzed using a continuous-flow analyzer (Seal AA3, Germany). Three field (travel) blanks were prepared for each batch of samples, which were analyzed together with the abovementioned samples, and used to blank correct sample results and determine the method detection limit (MDL) values. MDL was calculated using the following equation: MDL ≥t×Sb×N1+N2N1×N2, where the t value is given at the 95 % confidence level for the appropriate of degrees of freedom, Sb is the blank standard deviation, N1 and N2 are the number of sample measurements (single measurement, N1=1) and the number of analyzed blanks, respectively. From the field blanks, the MDL was calculated to be 0.31 µg m-3 for a 1-week ALPHA passive NH3 sample. All lab measurements were conducted in the Key Laboratory of Plant-Soil Interactions, Chinese Ministry of Education, China Agricultural University. More details regarding the passive samplers and the related laboratory preparation and analysis can be found in Xu et al. (2015).

Meteorological parameters, including wind speed (WS), wind direction (WD), relative humidity (RH) and temperature (T), were obtained at all sampling heights except 2 m; the temperature was also not available at 8 m. WS and WD were measured using four-cup anemometers (model O1OC, Met One Instruments), and RH and T were measured using a T/RH sensor (model HC2-S3, ROTRONIC).

Repeated-measures analysis of variance (ANOVA) was used to test changes in the NH3 concentration along vertical profiles. When the ANOVA results were significant, the Tukey's honest significant difference (HSD) test was used to determine the significance of the difference between means with a significance level of P<0.05. The coefficient of determination was used to test the linear correlations with a significance level of P<0.05. All of the statistical analyses were conducted using SPSS version 23.0 (IBM Corp., Armonk, NY, USA).

Potential source contribution function analysis (PSCF) (Ashbaugh et al., 1985) of atmospheric NH3 was performed using MeteoInfo (TrajStat package) (Wang, 2014), where 72 h back trajectories arriving at the monitoring site (IAP tower) at each height were calculated every 3 h for the entire study period. The average NH3 concentration for each cluster was computed using the cluster statistics function. NH3 pathways could then be associated with the high concentration clusters. The number of trajectory segment endpoints falling in a grid cell (i, j) is nij. The number of trajectory endpoints associated with the data with NH3 concentrations higher than an arbitrarily set criterion for each height during the four seasons (75th percentile for NH3 was set here) is mij (Table S1 in the Supplement). The PSCF value for the ijth cell is then calculated as mij/nij. A weighting function Wij was applied to reduce the uncertainties of small values of nij (Polissar et al., 1999). Weighted PSCF values (WPSCF) were calculated by multiplying a particular Wij (≤1.00) if the total number of the endpoints for one grid cell was lower than 3 times the average of the endpoints per each cell. Higher WPSCF values indicate higher potential contributions of NH3 to the receptor site (IAP tower). Wij=1.0080<nij0.7020<nij≤800.4210<nij≤200.05nij≤10

The time series of weekly averages of NH3 concentrations from 16 March 2016 to 16 March 2017 are shown in Fig. 2. The weekly NH3 concentration across all heights averaged 13.3±4.8 µg m-3 during the year-long study period. Individual weekly concentrations ranged from 4.4 µg m-3 at 2 m to 25.3 µg m-3 at 32 m. Nearly all (99.6 %) of the weekly NH3 concentrations along the profile exceeded 5 µg m-3. Summer concentrations were generally the highest. Maximum NH3 concentrations mostly occurred between 32 and 63 m, decreasing both towards the surface and the top of the tower. Minimum concentrations mostly occurred at 2 and 320 m (Fig. S1 in the Supplement). Significant differences of annual average NH3 concentrations across the vertical profile were only found between the “maximum concentration” height and the top two heights, i.e., 280 and 320 m (Fig. 3i). Even at 320 m, the annual average NH3 concentration was still relatively high at 11.3 µg m-3 (Fig. 3i). During the whole observation period, the daily average boundary layer height was generally above 320 m, indicating that a good portion of the sampling occurred within a well-mixed boundary layer (Fig. S2).

Seasonal vertical concentration profiles exhibited fairly similar shapes to the annual average profile, although there were some important differences in absolute concentration values and the magnitude of vertical gradients within the profiles (Fig. 3). The average NH3 concentration across the profile from high to low was observed in summer (18.2 µg m-3), spring (13.4 µg m-3), autumn (12.1 µg m-3) and winter (8.3 µg m-3). Proportional declines of the NH3 concentration from the peak to higher and lower elevations differed between seasons: the greatest proportional decline was seen in autumn (28.1 % decrease from 63 to 320 m), followed by winter (23.8 %), summer (20.5 %) and spring (15.8 %) (Fig. S3).

Vertical NH3 concentration profiles varied substantially during the sampling period, along with vertical changes in meteorological parameters. Bivariate polar plots (Fig. 4) show that high NH3 concentrations below 47 m were mostly observed during periods with low wind speeds (<4 m s-1). As heights and associated wind speeds increased, the relationship between NH3 concentrations and wind speed weakened. For example, at 280 m, the highest concentration was observed when the wind speed was also high (up to an average of ∼15 m s-1).

Wind direction also plays an important role in air pollution transport. Transport from the northwest was typically associated with low NH3 concentrations at all heights, consistent with the absence of large emissions sources in the mountains northwest of Beijing. It is noteworthy that high NH3 concentrations at near-surface heights (8 and 15 m) always coincide with winds from the south, including the southeast and southwest directions. High NH3 concentrations appear to be associated with winds from the northeast from 32 m to 80 m. Above 80 m, winds from the south contribute more to high NH3 concentrations. Major regions of agricultural NH3 emissions are located south and east of Beijing.

To further investigate observed variability, we show the probability density function of NH3 concentrations in relation to the relative humidity (RH) and temperature (T) (Fig. 5). Clear positive relationships between T and NH3 concentrations were found at all heights from low RH to high RH. When T was low (T<12 ∘C), the NH3 concentration mostly fell below 10 µg m-3 under all RH conditions. The occurrence of high NH3 concentrations increased with T>12 ∘C, which is not surprising given that agricultural NH3 emissions increase with T; furthermore, higher T and lower RH also shift the equilibrium of the NH3(gas) + HNO3(gas) ↔ NH4NO3(particulate) system toward the gas phase. Statistically, a strong positive relationship was found between NH3 and T at all heights from the surface to the top of the tower (R2∼0.6; Fig. S4); both the slope and the correlation coefficients were similar across all heights. Although, a positive correlation between NH3 and RH and a negative correlation between NH3 and WS were found, the correlation coefficients were quite low.

Analysis of the relationship between local wind direction and NH3 concentrations does not fully clarify the potential source regions contributing to observed NH3 at the sampling site (Fig. S6). Some seasonal variations were observed, i.e., the frequency of high NH3 concentrations were greater under southerly winds than northwesterly winds in the spring, the increased frequency of high NH3 concentrations were associated with southerly and easterly winds in the summer and autumn, and NH3 concentrations still exceeded 5 µg m-3 during winter with relatively frequent winds from the northwest.

To examine the relationship between air transport and NH3 concentrations more rigorously, weighted PSCF (WPSCF) during the four seasons were calculated for several measurement heights (2, 63, 180 and 320 m) (Fig. 6). In summer, from the surface to the tower top, a strong influence from source areas to the south of Beijing was seen, coinciding with regions (e.g., Tianjin, Henan, Hebei and Shandong provinces) characterized by elevated anthropogenic emissions of NH3 (Fig. 1), largely from agricultural activities (Zhang et al., 2009; Gu et al., 2012). During summer, regions to the north and west of the monitoring site had low WPSCF values, whereas high WPSCF values to the south and southeast were common during spring. High WPSCF values were mainly located northwest and southeast of Beijing in autumn, while their WPSCF values were typically lower in winter than during other seasons.

It is important to remember that aerosol–gas partitioning can also strongly influence measured NH3 concentrations. To investigate seasonal phase changes between NH3 and NH4+, we define the NH3 gas fraction (FNH3 = the gaseous NH3 concentration divided by the sum of the gaseous NH3 and fine particulate NH4+ concentrations), where the concentrations are expressed in molar units. Monthly average partitioning for these reduced inorganic nitrogen forms from a nearby urban monitoring site, 10 km from the IAP tower, is plotted in Fig. S8. The NH3 gas fraction (FNH3) was found to be the highest in summer (0.83 in August) and the lowest in winter (0.36 in February). As expected, gas phase NH3 is favored in the warmer months, while particle phase NH4+ is favored in the cooler months, with a gradual transition. Weekly NH4+ concentrations at the tower were estimated using weekly NH3 concentrations divided by monthly FNH3, and WPSCF analysis of the sum of NH3 + NH4+ was then performed (see results in Fig. S9). Results of this total WPSCF (NH3 + NH4+) analysis yielded similar patterns to the NH3 WPSCF analysis for all heights and seasons, indicating the importance of the identified source regions for both the gaseous and particulate atmospheric forms of emitted NH3.

The North China Plain is a well-known “hotspot” for NH3 emissions due to the rapid development of industrialization, urbanization and intensive agriculture (Kang et al., 2016; Y. Zhang et al., 2010). In our study, high atmospheric NH3 concentrations (13.3±4.8 µg m-3) were found up to 320 m a.g.l. in urban Beijing (16 March 2016–16 March 2017), and were much higher than the average annual NH3 concentration (3.3±1.4 µg m-3) observed across a vertical profile at the 300 m rural BAO tower, USA (Li et al., 2017). Some studies of NH3 vertical distribution found that the NH3 concentration decreased significantly with height. For example, Tevlin et al. (2017) reported an overall increase in summertime NH3 mixing ratios toward the surface of 6.7 ppb or 5.1 µg m-3 (89 %) during the day and 3.9 ppb or 3.0 µg m-3 (141 %) at night. In the BAO tower study (Li et al., 2017), which also measured concentrations using passive (Radiello) samplers deployed for 1- to 2-week sample periods, the concentration profiles showed a similar overall vertical distribution: the minimum NH3 concentration was observed at the top of the tower, it slowly increased towards a peak concentration at ∼10 m, and a sharp reduction was then seen near the surface. By contrast, our results showed much smaller decreases in NH3 concentrations in the upper air in urban Beijing (Table 1), with only a 1.18 µg m-3 (9.5 %) average decrease from the surface to the top of the tower (Fig. 3i). The flatter shape of the Beijing vertical profile may reflect a combination of strong local (e.g., vehicle) and regional (e.g., industrial and agricultural emissions) sources (Figs. 2 and 6) in our study, the fact that deep mixing layers regularly enveloped the full height of the tower within the surface boundary layer so that all sources influencing the tower measurements were vertically well mixed (Fig. S2), and/or the averaging of more distinct profiles over the week-long sample periods. In contrast to the “rural” boundary layer above the fields surrounding the BAO tower, the mixing in the Beijing urban area could be greatly enhanced by larger surface roughness (e.g., the average urban building height is ∼50 m) and surface heating (Baklanov and Kuchin, 2004). Higher time resolution vertical profile measurements are needed in the future to untangle the influence of these potential factors.

Distinct seasonal variations in NH3 concentrations were found (Fig. 2), which were statistically most strongly associated with temperature rather than relative humidity or wind speed (Fig. S4). High temperatures enhance NH3 emissions from soil, applied fertilizers, animal waste, vertical mixing and increase volatilization of NH3 from NH4NO3 particulate matter (Bari et al., 2003; Ianniello et al., 2010; Li et al., 2014; Lin et al., 2006; Meng et al., 2011; Plessow et al., 2005; Walker et al., 2004; Zbieranowski and Aherne, 2012). While high (low) mixed-layer heights in spring and summer (autumn and winter) could dilute (concentrate) NH3 in the surface boundary layer (Fig. S3), average NH3 concentrations across the profile were actually high in summer/spring and low in winter/autumn, consistent with the strong temperature-driven seasonal variation of the NH3 concentration and the greater NH4NO3 particle formation during cold periods in autumn and winter. Conducting simultaneous measurements of fine particle composition at different heights in future studies would be valuable for more closely evaluating the influence of changes in phase-partitioning.

Li et al. (2017) found a vertical difference of approximately 75 % from the concentration peak near the surface to the top of the BAO tower in winter (Fig. 3j), and attributed this strong vertical gradient to the occurrence of low level temperature inversions which trapped emissions closer to the surface during this period. During our study in Beijing, the vertical gradient was only 28 % in winter (maximum concentration found at 32 m), consistent with a deeper average boundary layer. However, inversions did limit the vertical mixing of NH3 during some periods in Beijing. Examination of the thermal inversion layer probability at 06:00 and 15:00 LT (Fig. S7b and c) revealed that T inversions (0.22±0.26 ∘C) frequently occurred between 102 and 160 m. Consequently, persistent higher NH3 concentrations begin at a lower altitude (Fig. S7a) as also observed by Tevlin et al. (2017). Because the time resolution of our Beijing study was one sample per week, we could not catch the changes between the daytime and nighttime NH3 vertical mixing. Compared to NH3 monitoring in real time (Tevlin et al., 2017), weekly sampling smooths diurnal vertical distributions and makes it harder to identify the influence of local surface sources or sinks.

Surfaces can act either as sources or sinks of NH3, depending on the surface NH3 content, ambient NH3 concentrations, and local meteorology and surface type (Tevlin et al., 2017; L. Zhang et al., 2010 ). The maximum NH3 concentration occurrence at 2 m in Beijing and the concentration decrease with increased height may reflect an important surface source of NH3, although our limited time resolution makes such conclusions tentative. The influence of the evaporation of dew/precipitation may also be important. Some studies found that dew is both a significant nighttime reservoir/sink and strong morning source of NH3 (Wentworth et al., 2016; Teng et al., 2017).

Areas south of Beijing with high WPSCF values appear to be important NH3 source regions (Fig. 6), suggesting regional transport from high agricultural NH3 emission areas (e.g., Hebei, Henan, Shandong provinces) contributed significantly to atmospheric NH3 in the Beijing urban region. Consistently higher NH3 concentrations were observed during periods with winds from the southeast, south and southwest at all heights, especially in summer (Fig. S6). Although NH3 has a limited atmospheric lifetime with respect to dry deposition, concentrations in these agricultural NH3 source regions can be extremely high (Shen et al., 2011) while significant NH3 can be tied up in longer-lived ammonium nitrate particles that partially dissociate to release NH3 back to the gas phase in response to NH3 loss by dry deposition (Ianniello et al., 2011; Kang et al., 2016; Xu et al., 2017). The WPSCF (Fig. 6) and NH3 emissions distribution (Fig. 1a) both suggest the importance not only of regional transport from nearby areas, but also the potential for local emissions to play an important role in sustaining the high NH3 level in Beijing, e.g., vehicular traffic (Chang et al., 2016; Pan et al., 2018a). As discussed above, stagnant meteorological conditions with low WS and T inversions allow local emissions, such as those from urban traffic, to accumulate. Additionally, the topography of the mountains to the west and north of Beijing effectively traps polluted air over Beijing during southerly airflow, an effect reported in many Beijing particulate matter studies (Xia et al., 2016; Wu et al., 2009; Zhao et al., 2009).

Generally, NH3 source regions identified in the WPSCF analysis (Fig. 6) suggest that regional transport from the south exerts an important influence on Beijing NH3 concentrations throughout the year. The area south of Beijing (e.g., Hebei, Henan and Shandong provinces) is a hotspot of NH3 emission (Zhang et al., 2018), and half of NH3 emissions have been estimated to deposit as NH3 at urban sites in the North China Plain (Pan et al., 2018b). In addition, seasonal patterns of NH3 potential sources (Fig. 6) matched well with the seasonal surface NH3 concentrations in China (Zhang et al., 2018). In detail, NH3 concentrations were typically highest in summer, and south winds produced higher NH3 concentrations than other wind directions (Fig. S6). Spring and summer had a similar wind direction distribution (Fig. S6) and wind speeds (Fig. S5), but corresponding NH3 concentrations were lower in spring. This may reflect decreased emissions in regions to the south during cooler spring temperatures and the increased partitioning of NH3 into fine particles during this cooler season. As shown above aerosol–gas partitioning strongly influences NH3 concentrations; high FNH3 during warm periods, especially summer, favored greater NH3 gas concentrations due to the thermodynamic tendency for NH4NO3 to dissociate to NH3 and HNO3 at high temperatures. Although FNH3 was low in winter, indicating that NH4+ is the dominant NHx form in this cold season, winter NH3 concentrations across all heights still averaged 8.3±2.6 µg m-3, with a similar wind direction distribution as other seasons, except at high altitudes (i.e., 240 and 320 m; Fig. S6).