The Adapted Low-Cost Passive High Absorption (ALPHA) samplers (Centre for Ecology and Hydrology, Edinburgh, UK), one of the most widely recognized passive sampling devices (PSDs) (Puchalski et al., 2011; Tang et al., 2001; Xu et al., 2015a), were used to collect ambient NH3 in this study. The ALPHA sampler is a circular polyethylene vial (26 mm height, 27 mm diameter) with one open end. The vial contains a position for a 25 mm phosphorous acid-impregnated filter and a PTFE (polytetrafluoroethylene) membrane for gaseous NH3 diffusion. In the current study, triplicate ALPHA filters were used to collect NH3 for IRMS at weekends (from Saturday to Monday) or weekdays (from Monday to Saturday) on the roof of a four-story building (12 m a.g.l.) on the campus of the China Agricultural University (CAU), Beijing (116.289∘ E, 40.032∘ N), before (18–20, 20–25, 25–27 October, 27 October–1 November, 1–3 November), during (3–8, 8–10, 10–15 November), and after (15–17, 17–22, 22–24, 24–29 November) the APEC summit (normally at 08:00 local time). The site of the CAU campus represents a general urban background with relatively low-rise buildings in the surrounding area, ca. 16.3 km northwest of Tiananmen Square (Fig. 1). In this study, the three filters of one sampling event were combined for a single analysis. Hourly mass concentrations of gases (CO, SO2, NOx, and O3) and PM2.5 in Beijing are averaged from the data obtained from the 14 state-controlled environmental monitoring stations across the city (http://datacenter.mep.gov.cn/).

The Ogawa passive sampling device (Ogawa & Co., FL, USA) is another popular PSD that has been successfully applied in the USA (Butler et al., 2015; Puchalski et al., 2011) and China (Chang et al., 2015; Meng et al., 2011) to determine the time-integrated NH3 concentrations. The Ogawa PSD is a double-sided passive diffusion sampler equipped with a diffusive end cap, followed by a stainless-steel screen, and a 14 mm quartz filter impregnated with phosphoric acid by the manufacturer. In this study, the Ogawa PSDs were used to collect NH3 emitted from sources for isotopic analysis.

Six NH3 emission sources were involved in this study, i.e., livestock (two pig sties), fertilizer volatilization (laboratory simulation of NH3 volatilization from urea-fertilized soil), human excreta (septic tanks of a residential building and a teaching building), waste water (sewage water treatment plant), solid waste (municipal waste transfer stations in a residential community and an educational area), and vehicles (a heavily used urban tunnel) (Tables 1 and S1 in the Supplement). NH3 emitted from fertilizer volatilization was collected in the laboratory of Fudan University. To minimize the mixture of ambient atmospheric NH3/NH4+ (see Figs. 2 and S1 for two examples) and to examine the potential influence of the macro-environment, non-laboratory samples were collected within a confined space in short periods (e.g., several hours) during both the warm (summer) and cold seasons (winter) between June 2014 and January 2015. Descriptions of the sampling processes are detailed in Table S1.

The ambient average NH3 concentration was less than 6 µg m-3 (Chang et al., 2016), which was between 1 and 3 orders of magnitude smaller than all emission sources we investigated (Table 1). Therefore, the interference of ambient air to all source samples is not a major concern in our study. HONO has been measured in significant concentrations in the UK and interfered with measurements of HNO3 by DELTA active samplers, causing us now to coat the instrument with NaCl to avoid this interference with HNO3 measurements. The Ogawa filters used for trapping NH3 were impregnated with citric acid. Using a discrete auto analyzer (Smartchem 200, AMS, Italy; the detection limit for NO2--N is 0.002 mg L-1), we recently analyzed nearly 100 passive filter samples collected from our passive ammonia network in Shanghai (Fig. 1). No detectable nitrite was found for almost all these samples (including samples collected from a busy road tunnel; Chang et al., 2016), which provides compelling evidence that passive collection of ammonia does not introduce interference of nitrite in this study.

In our pilot study, the sampling period of each source was optimized to avoid insufficiency of NH3 absorption, but more importantly, to minimize potential effects of N isotope fractionation. Taking the source of human excreta for example, the concentrations of NH3 in the exhausts of 15 ceiling ducts from collecting septic tanks in 13 buildings with 6 functions (2809-2661+5803 µg m-3) were higher than those in ambient air by 3 orders of magnitude (Chang et al., 2015). Therefore, the insufficiency of NH3 absorption is of no significant concern. However, the δ15N–NH3 values of daily samples varied widely (±10 ‰), suggesting that the isotope fractionation may occur during the process of sampling/storage. After many tests by trial and error, we found that a sampling period of 2 h could provide sufficient N–NH3 as well as avoid potential fractionation. Another example is the on-road traffic source. The δ15N–NH3 values of weekly (-11.9; -11.2 ‰), semi-monthly (-11.7; -12.5 ‰), and monthly samples (-12.0 ± 1.8 ‰; n=4) in the exit of Handan tunnel were almost identical, suggesting that passive NH3 collection is an effective approach for isotopic analysis procedures.

A newly developed chemical method for δ15N–NH4+ of low NH4+ samples was used in the current study. The detailed analytical procedures are given elsewhere (D. Liu et al., 2014). Briefly, this method is based on the isotopic analysis of N2O, which is much less abundant in the atmosphere than N2 and thus causes minimal atmospheric contamination. Filter samples were firstly soaked with ultra-pure water (18.2 MΩ cm). Concentrations of NH4+ were then analyzed using an ion chromatographic system (883 Basic IC plus, Metrohm Co., Switzerland) equipped with a Metrosept C4/4.0 cation column. The eluent was 1.0 mM HNO3+ 0.5 mM PDA. The detection limit for NH4+ was 0.0028 mg L-1. After the measurement of the NH4+ concentration, NH4+ in every sample was initially oxidized to NO2- by hypobromite (BrO-) in a vial. NO2- was then quantitatively converted into N2O by hydroxylamine (NH2OH) under strongly acid conditions. The produced N2O was analyzed by a purge and cryogenic trap system (Gilson GX-271, IsoPrime Ltd., Cheadle Hulme, UK) coupled to an IRMS (PT–IRMS) (IsoPrime 100, IsoPrime Ltd., Cheadle Hulme, UK) at the Stable Isotope Ecology Laboratory of Institute of Applied Ecology, Chinese Academy of Sciences.

Isotope ratio values are reported in parts per thousand relative to atmospheric N2 as follows: δ15N‰=15N/14Nsample-15N/14NN215N/14NN2×1000. Three international NH4+ standards (IAEA N1, USGS 25, and USGS26 with δ15N values of +0.4, -30.4, and +53.7 ‰, respectively) were used to correct for the reagent blank and drift during isotope analysis of the produced N2O. The standard deviation of δ15N measurements is less than 0.3 ‰. Moreover, to enhance our confidence in the results determined by the PT–IRMS method, the δ15N values of fertilizer-, vehicle-, human excreta-, and solid waste-derived NH3 were also examined by the EA-IRMS method at the Shanghai Institutes of Life Sciences, Chinese Academy of Sciences (Supplement Sect. S1).

Stable isotope mixing models offer a statistical framework to estimate the relative contributions of multiple sources to a mixture, such as food-web structure, plant water use, air pollution, and many other environments (Cole et al., 2011; Dai et al., 2015; Jautzy et al., 2015; Wang et al., 2016). A common problem, however, is having too many possible sources relative to isotopes to allow unique linear mixing solutions based on mass-balance equations. To this end, Phillips and Gregg (2003) developed the model IsoSource, which solves iteratively for feasible mixing solutions, and has been well tested in numerous studies (e.g., Cole et al., 2011; Dai et al., 2015; Jautzy et al., 2015; Wang et al., 2016). The model does not generate exact values for proportional contributions of each source, but instead provides a range of possible contributions or feasible solutions. The IsoSource addresses every possible combination of source proportions (summing to 100 %) incrementally (e.g., 1 %), then calculates the predicted isotope value for each combination using linear mass-balance equations. These predicted values are then examined to determine which ones fall within some tolerance range (typically 0.1 ‰) of the observed consumer isotope value, and all of these feasible solutions are recorded. The IsoSource model is available at https://www.epa.gov/eco-research/stable-isotope-mixing-models-estimating-source-proportions.

Considering the large number of possible sources for ambient NH3, we use multiple lines of evidence (prior information) to constrain the emission sources in the mixing model analysis: (1) there was no crop harvest activity in the North China Plain (NCP) during the APEC summit in Beijing. Besides, harvesting forests for fuelwood and timber has nearly disappeared in Beijing. Therefore, the contribution of biomass burning is considered minimal; (2) Beijing is 150 km inland from the Bohai Sea (the nearest sea) via Tianjin Municipality in the southeast. Therefore, the marine source in Beijing can be neglected; and (3) previous work indicated that miscellaneous NH3 sources like pets and household products are minor NH3 emissions in Beijing urban areas (Chang, 2014), which thus can be largely neglected.

In conclusion, ambient NH3 in Beijing during our study period has been shown to be due to four main sources: livestock production, N-fertilizer application, on-road traffic emissions, and waste-derived emissions. The δ15N average values for these four NH3 emission sources will be built and served as the baseline input to the IsoSource. For the traffic source, given that the relatively larger difference in terms of their δ15N values in different seasons (see Table 1), the wintertime average value of δ15N was used in this study because the APEC summit was held during winter. In brief, the N isotopic signatures for the sources of waste, livestock, traffic, and fertilizer are set as -37.8, -29.1, -16.5, and -50.0 ‰, respectively (Tables 1 and S1). The source increment and mass-balance tolerance parameter values of 1 % and 0.1 ‰, respectively, were applied. Model output files include all the feasible source combinations, with histograms and descriptive statistics on the distributions for each source. Results are expressed as box-and-whisker plots for these distributions.

Using N isotope as a tool to discriminate the contribution of various sources to ambient NH3 concentration requires (i) well-established N isotopic compositions of NH3 emission sources and (ii) well-constrained N isotope fractionation to allow separating different sources. In total of 44 NH3 source samples in this study, δ15N values and NH3 concentrations of these samples ranged from -52.0 to -9.6 ‰, and 33 to 6211 µg m-3, respectively (all data are presented in Table S1). These NH3 sources can be classified into four categories, i.e., fertilizer, livestock, traffic, and waste. For most sources, there was no significant difference in terms of their δ15N values in different seasons (Table 1), indicating the effectiveness of our sampling strategy. The work of Felix et al. (2013) addressed some of the same issues and we found and reported similar results for the range of δ15N–NH3 values (-56.1 to -2.2 ‰) from major NH3 emission sources (including livestock, marine, vehicle, and fertilizer sources) as those presented here. These two independent studies determining the δ15N values of major NH3 sources arrive at the same conclusion: NH3 emitted from volatilized sources has relatively low δ15N values, allowing them to be distinctly differentiated from NH3 emitted from traffic sources that are characterized by relatively high δ15N values.

Road tunnels are excellent locations to provide emissions from a large number of vehicles mixed with SCRs and TWCs under “real-world conditions” (T. Liu et al., 2014). Here we assume that the δ15N–NH3 values collected from a heavily used tunnel like the Handan tunnel (around 120 000 vehicles passing per day; an average of 65.9 µg m-3 NH3 or 12.4 times the levels of open environment (Chang et al., 2016) can be accepted as the isotopic signatures of vehicles in China. Nevertheless, we note the δ15N–NH3 values of vehicle exhausts that we collected from the Handan tunnel (-17.8 to -9.6 ‰; n=8) are lower than the δ15N–NH3 values determined by Felix et al. (2013) in the Squirrel Hill Tunnel, Pittsburgh (-4.6 and -2.2 ‰; n=2). This Sino–US difference may be partially attributed to a higher usage of TWCs in the traffic in the USA.

The δ15N values of NH3 from livestock (-31.7 to -27.1 ‰) and fertilizer (-52.0 to -47.6 ‰) that we measured are slightly lower than the range of δ15N–NH3 values collected monthly from two dairy barns (-28.5 to -22.8 ‰) and a cornfield treated with urea–ammonia–nitrate fertilizer (-48.0 to -36.3‰) by Felix et al. (2013). These ranges of δ15N values are a function of the initial δ15N values of animal waste and fertilizer and variations in the bacteria populations, as well as other factors (temperature, wind, pH, etc.) that influence kinetic fractionation rates associated with NH3 volatilization. Long-term (30 days) monitoring of δ15N and NH3 emissions of manure measured by Lee et al. (2011) indicated that the dynamics of N isotope fractionation may be complicating the usefulness of the isotope approach as a tool for estimating NH3 emissions in field conditions. In this sense, the shorter sampling period in our work should reflect the essence of δ15N values of livestock- and fertilizer-derived NH3.

As a normal metabolic process, the release of NH3 from human excreta has been well documented. However, most emission inventories involving human excreta have focused on pit latrines in rural areas of developing and middle income countries. In urban China, human excreta are typically stored in a three-grille septic tank under the building before disposal. After a series of anaerobic decomposition processes, a substantial amount of NH3 will be generated and emitted through a ceiling duct. In the present study, the concentrations of NH3 in the ceiling ducts (4509.3 ± 1248.0 µg m-3; n=8) outweigh those in the open air by 3 orders of magnitude, and the δ15N–NH3 values are seasonally consistent (-38.4 ± 0.9 ‰ in summer and -38.6 ± 1.0 ‰ in winter; Table 1), suggesting that human excreta may be an important and consistent source of NH3 in urban areas. These data suggest that emissions of NH3 from human excreta for an urban population of ∼ 21 million people in Shanghai contribute 1386 Mg NH3 annually to the atmosphere within the city, which corresponds to 11.4 % of the total NH3 emissions in the Shanghai urban areas (Chang et al., 2015). The δ15N values of wastewater-originated NH3 (-41.0 ± 0.9 ‰; n=8) are close to that of human excreta and also show no seasonal variation. Sampling in a stable and closed physical environment may be responsible for such a small range of isotopic variation. However, although also sampled in a closed environment, the δ15N values of municipal solid waste demonstrate a much greater variation (-37.6 to -29.9 ‰), which may be due to the variable composition of solid waste.

Hourly observations of major air pollutants (including PM2.5, NOx, CO, SO2, and O3) in Beijing are show in Fig. S2. The meteorological differences (e.g., temperature and wind speed) for the three periods pre-, during and post-APEC are not significant, suggesting that emission reduction strategies implemented during the APEC summit were successful (Fig. S2). It should be noted that several control measures, i.e., closing factories within 200 km of the city center and stopping the entrance of out-of-city vehicles, had been undertaken in Beijing and its neighboring regions before the summit. Therefore, the before–during comparison of some pollutants like SO2 are not in stark contrast in terms of their mass concentration (Fig. S2). The evolution of ambient NH3 mass concentrations measured at CAU shows a similar pattern with CO and NOx (Fig. S3). Before the opening of the summit (from 18 October to 3 November), NH3 concentrations averaged 9.9 µg m-3 and ranged from 6.9 to 11.0 µg m-3. During the summit session (from 3 to 15 November, the ending date of the summit is 13 November), this was reduced to 7.3 µg m-3 with a range of 5.8–8.6 µg m-3. After the APEC summit (from 15 to 29 November), the NH3 concentration levels rebounded to an average of 12.7 µg m-3 (ranging from 10.7 to 17.7 µg m-3). In other words, the NH3 concentrations were reduced by 20.0 % during the APEC summit compared with the period before it. Compared with the period after control, the concentrations were 74.5 % lower than that during the summit (Table 2).

On the basis of the δ15N values of NH3 emission sources and ambient δ15N–NH3 samples (Table 2), the ranges (within 5 and 95 percentiles) of relative contribution fractions of each NH3 source to the ambient atmosphere were modeled by the IsoSource and depicted in Fig. 3a–d. Of these, sources of traffic and fertilizer are better constrained than livestock and waste. This is because the δ15N–NH3 values of the ambient atmosphere are closer to the latter two. For example, after the APEC summit in our study period, the δ15N–NH3 values of the ambient atmosphere averaged -30.7 ‰, which is very close to the isotopic signature of livestock (-29.1 ‰), thus leading to weaker constraint with 5 and 95 percentiles ranges from near 0 to 0.7 (Fig. 3).

For each sample, information regarding the δ15N value, the NH3 concentration and its source contributions is listed in Table 2. Figure 4 illustrates the overall source contribution proportions before, during, and after the summit, in which traffic, waste, livestock, and fertilizer comprised 20.4, 25.9, 24.0, and 29.7 % of the whole period (Table 2), respectively. Specifically, the contribution of traffic initially decreased by 56.7 % and then doubled, and this represents the largest change among the four sources (Table 2). Considering the absolute contributions of different sources to ambient NH3, these results show that traffic is the most sensitive source to the emission control measures. This result was expected since over half the vehicles in Beijing were banned from entering the city during the APEC summit (note that before the summit, out-of-city vehicles had already been banned from entering Beijing). Based on the vehicle NH3 emission factor obtained from real-world tunnel tests, T. Liu et al. (2014) and Chang et al. (2016) recently reported that vehicles contribute 8.1 and 12.0 % to total NH3 emissions in the Pearl River Delta region and Shanghai urban areas, respectively. With a total of 5.5 million vehicles in 2014, the traffic fleet in Beijing was much more than other cities. Thus, vehicle sources might contribute more than 20 % of the total NH3 to the air within Beijing.

The waste-originated percentages in Figs. 3 and 4 remain fairly constant over the sampling period, appearing to be a stable and important NH3 contributor in Beijing. Compared with wastewater and solid water, NH3 emissions from human excreta through in situ sceptic tank system are far from quantified. Based on an extensive measurement campaign, we estimated that the population of ∼ 21 million people living in the urban areas of Shanghai annually emitted approximately 1386 Mg NH3, which corresponds to over 11.4 % of the total NH3 emissions in the urban areas (Chang et al., 2015). Non-agricultural sources – merged with waste and traffic NH3 emissions – collectively account for nearly 50 % of ambient NH3 before and after the APEC summit, which cannot be explained by previous work of emission inventories (e.g., Fu et al., 2013; Huang et al., 2011, 2012; Kang et al., 2016; M. Li et al., 2015; Zhang et al., 2009, 2010). Our results do not contradict a commonly held belief that agriculture is responsible for the vast majority (normally > 90 %) of total ammonia emissions at a regional scale. However, the results show that within urban areas, non-agricultural sources are very important. A consequence of the new results is that measures to improve air quality in urban areas of China need to include measures to reduce both agricultural and non-agricultural sources of NH3.

The NCP is one of the most intensive agricultural regions in China, enjoying a good reputation of “China's granary” (Ju et al., 2009). Situated on the northern edge of the NCP with mountains to the north and west, Beijing is a receptor of agricultural NH3 from rural areas. During our study period, crops in the NCP had been harvested and thus fertilizer application would have been very limited. However, our results show that fertilizer is the largest contributor, accounting for 29.7 % throughout the sampling periods (Table 2). One explanation might be the prevalence of intensive urban agricultural production with high nitrogen fertilizer input in the suburban areas. Beijing's increase in land area from 4822 km2 in 1956 to 16808 km2 in 1958 led to the increased adoption of peri-urban agriculture. Such ”suburban agriculture” contributed ∼ 70 % of non-staple food in Beijing, mainly consisting of vegetables and milk, to be produced by the city itself in the 1960s and 1970s (Cai, 2003). In the late 1990s, recognizing the importance of urban agriculture to sustainable urban development, Beijing's municipal government launched an official program encouraging multi-function urban agriculture in peri-urban areas by supporting the development of “agro-parks”, which not only produce food but also attract tourism and are used as educational tools (Cai, 2003). One of the recent experiments in urban agriculture is the Modern Agricultural Science Demonstration Park in Xiaotangshan Town, Changping District (Cai, 2003). Today, Beijing is leading the way in using smart-city technologies to make urban farming more sustainable. In addition to suburban agriculture, there are 17 golf courses with 2280 ha. greens in Beijing; some of them are located at the urban areas (Chang, 2014). The turf grass of golf course typically needs 200–400 kg N ha-1 yr-1 as N fertilizer to achieve high performance (Wong et al., 1998, 2002; Zhang, 2002), which should be considered as an overlooked NH3 contributor in Beijing.

The data set reported in this study represents, to the best of our knowledge, the first attempt to partition urban atmospheric NH3 sources. Considering the current nascent stage of partitioning NH3 sources using stable isotope approach, there are several unsolved problems that could potentially undermine the above-mentioned results. One of our fundamental assumptions in this study is that the measured NH3 was directly from NH3 emission sources. In other words, we treated the measured NH3 as the mixture of primary NH3 sources without “gas–aerosol conversion” fractionation. But in fact, NH3↔ NH4+ equilibrium will cause 14N to be preferentially associated with NH3 and 15N to be enriched in NH4+ of PM due to the stronger associative strength of 15N than 14N in NH4+ (Kawashima and Kurahashi, 2011; Yeatman et al., 2001). In Beijing, earlier studies confirmed that the atmosphere in Beijing was NH3-limited, suggesting that acidic gases could not be fully neutralized to form ammonium salts (Ianniello et al., 2010; Wang et al., 2016). Therefore, ammonium salts had much less opportunity to volatilize to NH3 to exert substantial isotopic effect through NH3↔ NH4+ equilibrium. Still, it is critical to develop a controlled laboratory system to fundamentally understand the characteristics and mechanisms of N isotope fractionations during the process of NH3 transformation (Li et al., 2012).

Several additional factors could introduce uncertainty in the solutions of isotope mixing model. Given the complexity of urban NH3 sources, no definitive solution exists in a linear mixing model with one isotope system tracer (δ15N) in the current study. Recommended future studies should include the combinations of different types of isotope ratio measurements and the adoption of more sophisticated Bayesian mixing models. The isotopic signature of sources like on-road traffic still remains uncertain. Some may argue that since NH3 (also NOx) is known to be a component of vehicle-emitted exhaust, why not collect vehicle-emitted NH3 directly from the tailpipes. To our knowledge the δ15N of vehicle-emitted NH3 has not been previously assessed. The tunnel test has a unique advantage in measuring the overall isotopic signatures of vehicle-emitted pollutants. We believe that the δ15N–NH3 values of the samples collected from Handan tunnel in this study are representative as the isotopic signatures of vehicles in China.

Despite the potential limitations in this study, given the importance of NH3 to PM2.5 formation, this work can be expected to enrich the discussion on the methodologies (including stable isotope analysis) in terms of identifying the largest NH3 sources in urban atmosphere where policy efforts relating to emissions abatement can be directed to deliver the largest impact.