Source appointment of nitrogen in PM2.5 based on bulk δ15N signatures and a Bayesian isotope mixing model

Abstract Nitrogen isotope (δ15N) has been employed to differentiate major sources of atmospheric N. However, it remains a challenge to quantify contributions of multiple sources based on δ15N values of the N mixture in atmospheric samples. This study measured δ15N of bulk N in PM2.5 at an urban site of Beijing during a severe haze episode of 22–30 January 2013 and a background site of Qinghai, north-western China from 6 September to 15 October 2013, then applied a Bayesian isotope mixing model (SIAR, Stable Isotope Analysis in R) to analyse the N sources. At Beijing site, δ15N values of PM2.5 (−4.1‰ to +13.5‰, +2.8 ± 6.4‰) were distributed within the range of major anthropogenic sources (including NH3 and NO2 from coal combustion, vehicle exhausts and domestic wastes/sewage). At Menyuan site, δ15N values of PM2.5 (+8.0‰ to +27.9‰, +18.5 ± 5.8‰) were significantly higher than that of potential sources (including NH3 and NO2 from biomass burning, animal wastes, soil N cycle, fertilizer application and dust N). High molar ratios of to and/or in PM2.5 at the background site suggested that the equilibrium of NH3 ↔ caused apparent 15N enrichments in ammonium. Results of the SIAR model showed that 39 and 32% of bulk N in PM2.5 of Beijing site were contributed from N emissions of coal combustion and vehicle exhausts, respectively, whereas N in PM2.5 at Menyuan site was derived mainly from N emissions of biomass burning (46%) and NH3 volatilization (34%). These results revealed that the stoichiometry between NH3 and acidic gases plays an important role in controlling the bulk δ15N signatures of PM2.5 and emissions of reactive N from coal combustion and urban transportation should be strictly controlled to advert the risk of haze episodes in Beijing.


Introduction
Urban air pollution is a globally challenging issue. Nitrogen (N) emissions play a key role in the formation of atmospheric particulates, especially secondary N-containing aerosols (Huang et al., 2014). Ammonia is the precursor of ammonium (NH + 4 ) and readily reacts with available acids formed by SO 2 and NO 2 , and also it can be transformed to organic N or amines (Ge et al., 2011). Nitrogen oxides are major precursors of both inorganic (as nitrate ions (NO − 3 )) and organic (as organic NO − 3 ) N aerosols (Berkemeier et al., 2016). Therefore, the source apportionment of N in PM 2.5 is always of significance for better understanding origins of particulates and haze pollution (Guo et al., 2014). Stable isotopes of N (i.e. δ 15 N values) have been used to trace major sources and processes of atmospheric N (Heaton, 1986;Michalski et al., 2004;Kendall et al., 2007;Pavuluri et al., 2010;Savarino et al., 2013). The analysis of bulk δ 15 N in PM 2.5 is a quick method compared to δ 15 N measurements of inorganic and organic N components (Widory, 2007;Hegde et al., 2015;Bikkina et al., 2016), and it does also provide valuable information on δ 15 N of dry N deposition (Yeatman et al., 2001a; 1 *Corresponding authors. e-mail: liuxueyan@tju.edu.cn (X.-Y. Liu); baizp@craes.org.cn (Z.-P. Bai) 2 Y.-L. WANG ET AL. than that in rain NH + 4 and gaseous NH 3 (Yeatman et al., 2001a(Yeatman et al., , 2001bJia and Chen 2010;Felix et al., 2013). In a hypothetical model by Heaton et al. (1997), the δ 15 N of particulate NH + 4 stabilized at values of 33‰ (an enrichment coefficient) higher than that of NH 3 when NH 3 ↔ NH + 4 equilibrium was achieved at 25 °C. However, mechanisms for atmospheric NH 3 ↔ NH + 4 equilibrium in the field circumstances are poorly understood, which is particularly important for interpreting the δ 15 N variations of PM 2.5 at locations dominated by NH + 4 -N. This study measured bulk δ 15 N of PM 2.5 at an urban site (Chinese Research Academy of Environmental Sciences (CRAES), Beijing, northern China) and a national atmospheric background monitoring station (Menyuan, Qinghai province, northwestern China). Based on bulk δ 15 N of PM 2.5 and major N sources, a Bayesian isotope mixing model (SIAR, Stable Isotope Analysis in R)  was used to estimate the proportions of different source contributions to N in PM 2.5 and to evaluate anthropogenic N emissions during the haze events in Beijing. As inorganic N in the atmosphere of both sites was dominated by NH 4 -N, we hypothesized that 15 N enrichments in PM 2.5 relative to dominant sources were mainly derived from the NH 3 ↔ NH + 4 equilibrium (assumed as 33‰) (Heaton et al., 1997;Li et al., 2012).

Study sites
The sampling site in Beijing is located in the courtyard of CRAES (40°04′ N, 116°42′ E), at Lishuiqiao South of Beiyuan Road. Due to rapid urbanization and economic development, the vehicle exhausts and energy consumption are large in Beijing, resulting in deterioration of air quality. Atmospheric PM 2.5 in Beijing was characterized by high contributions of secondary components from anthropogenic origins (Sun et al., 2006). Secondary inorganic ions (such as SO 2− 4 , NH + 4 and NO − 3 ) were the dominant contributors in PM 2.5 of Beijing (Zhang et al., 2013). During the sampling period (January 2013), Beijing suffered from the worst PM 2.5 pollutions Heaton et al., 2004;Elliott et al., 2007Elliott et al., , 2009. At a background site, the δ 15 N of PM 2.5 allows us to examine the impacts of emissions from non-point sources' agricultural N emissions on the N chemistry of regional atmosphere, whereas at an urban site, it can imprint the anthropogenic N emissions.
The bulk δ 15 N in atmospheric particulates is mainly determined by the δ 15 N of N precursors (Aggarwal et al., 2013;Hegde et al., 2015). Often, reported δ 15 N values of typical inorganic N sources (Table 1) can be used in studies on sources and fates of atmospheric N (Elliott et al., , 2009Kendall et al., 2007;Kawashima and Kurahashi, 2011;Michalski et al., 2014). For PM 2.5 , dust is a primary N source (Zhang, 2010;Huang et al., 2014). At background site, NO 2 and/or NH 3 from microbial N cycle, fertilization application and animal wastes are strongly 15 N-depleted Li and Wang, 2008;, while N emissions from biomass burning (Kawashima and Kurahashi, 2011;Divers et al., 2014) are typically 15 N-enriched. At urban site, most N sources of PM 2.5 are anthropogenic. The NH 3 from animal wastes (including sewages; Heaton, 1986), coal combustion and vehicle exhausts (Felix et al., 2013), as well as NO 2 from vehicle exhausts  showed negative δ 15 N values, but NO 2 from coal combustion had exclusively positive δ 15 N values (Felix et al., 2012).
Besides, the bulk δ 15 N in atmospheric particulates is influenced by the isotopic fractionations during gas (g) ↔ particle (p) exchange processes. However, isotope effects between N precursors and the aerosol N remain unclear, especially in field conditions. Isotope effects are assumed to be more significant between NH 3 and NH + 4 , and much smaller in the case of NO 2 and aerosol N (Yeatman et al., 2001a;Kawashima and Kurahashi, 2011). This assumption was supported by small differences in mean δ 15 N values between roadside NO 2 (5.7‰) and local aerosol N (6.8‰) (Ammann et al., 1999;Pearson et al., 2000). Although the kinetic isotope effect of NH 3 → NH + 4 reaction is small at the beginning, it becomes significant when NH 3 ↔ NH + 4 equilibrium attains and causes a preferential enrichment of 14 N in NH 3 and 15 N in NH + 4 of aerosols (Heaton et al., 1997;Fukuzaki and Hayasaka, 2009;Li et al., 2012). This explained generally higher δ 15 N values of NH + 4 in aerosols  Felix et al. (2013 in history (http://cleanairinitiative.org/portal/node/11599), registering the highest PM 2.5 hourly concentration of 886 μg/ m 3 (http://www.nasa.gov/multimedia/imagegallery/image-feature2425.html). The background site is located on the Daban Mountain (37°36′ N, 101°15′ E) in Menyuan county, north-eastern of Qinghai province, which is one of the 14 National Background Stations established by the Chinese Ministry of Environmental Protection in 2012. It has a typical Plateau continental climate, with an altitude of 3295 m above sea level, lower than the average of the Tibetan Plateau (about 4000 m). The mean annual temperature and precipitation amount are 0.8 °C and 520 mm, respectively. Agricultural activity is not intensive locally, except in low-altitude areas far away from the Daban Mountain in Menyuan. The sampling period (6 September-15 October 2013) is within the harvesting period after an intensive fertilization and pronounced biomass burning. The mean hourly temperature was 6.5 °C (3-11 °C) during the study period. There is no fossil fuel emission locally, with limited road traffic on the national highway of G227.

Sample collection and chemical analyses
PM 2.5 was collected using a pre-baked quartz filter (47 mm in diameter) and aerosol sampler (Leckel, MVS6, Germany) equipped with a size-segregating impactor. The operating air flow rate was 38.3 L/min. To collect sufficient PM 2.5 sample for bulk δ 15 N analyses, sampling was conducted for every 47-71 h at Menyuan (n = 14) and for 23 h at Beijing (n = 14). Filter blanks were also collected following the same procedure. The PM 2.5 mass on each filter was gravimetrically measured using microbalance (AWS-1, COMDE DERENDA, Germany, approved by European Standard) after being desiccated for at least 24 h under controlled temperature (20 ± 1 °C) and humidity (50 ± 5%). All filter samples were immediately stored at −20 °C prior to chemical analyses.
Concentrations of bulk N in PM 2.5 (mainly including NH + 4 , NO − 3 and organic N) were measured using three punches (ca. 0.53 cm 2 for each) of the filter in a vario MACRO cube (Elementar Analysensysteme GmbH, Germany) with an analytical precision of 0.02%. Based on N contents, bulk δ 15 N values of about 50 μg N in each PM 2.5 sample were determined by a Thermo MAT 253 isotope ratio mass spectrometer (Thermo Scientific, Bremen, Germany) coupled with an elemental analyzer (Flash EA 2000). IAEA-N-1 (Ammonium sulphate; δ 15 N = 0.4‰), USGS25 (Ammonium sulphate, δ 15 N = −30.4‰) and IAEA-NO-3 (Potassium nitrate; δ 15 N = +4.7‰) were used as standards for the calibration of δ 15 N values. The average standard deviation for replicate analyses of an individual sample was ±0.1‰. The δ 15 N in PM 2.5 was expressed in parts per thousand (per mille) by multiplying them by 1000: where R = 15 N/ 14 N for samples and standard (atmospheric N 2 ).
The concentrations of NO − 3 , NH + 4 and SO 2− 4 in PM 2.5 were measured during the sampling period at both sites by an ambient ion monitor (AIM-IC system: Model URG 9000B, URG Corporation, USA). It draws air in through a PM 2.5 sharp-cut cyclone at a volumetric flow controlled rate of 3 L/min to remove the larger particles from the air stream. The real-time instruments installed at both the stations have a detection limit of 0.05 μg/m 3 . Gases such as SO 2 , NH 3 and HNO 3 are stripped from the air stream by passing through a liquid parallel plate denuder with continuously replenished solvent flowing across the surface. Then, the PM 2.5 air stream is constrained into a supersaturated steam condensation coil and cyclone assembly and grown hygroscopically for collection. Enlarged particles are dissolved in water solutions for anion chromatographic analysis every hour following 60 min of ambient sampling. Concentrations of NO 2 were measured using a NO-NO 2 -NO x chemiluminescence analyzer (Model 42i, Thermo-Fisher Scientific). The instruments were operated and maintained properly to ensure data integrity. Scheduled quality control procedures included daily zero and span checks, weekly precision checks and data validations.
In this study, agricultural and biogenic N emissions were not considered as the major sources of bulk N in PM 2.5 of Beijing for two main reasons. First, the urban site is located in the centre of Beijing city cluster in CRAES. During the severe haze events occurring in Beijing, several studies have shown that aerosols have been mainly influenced by anthropogenic sources. Second, as the sampling of Beijing PM 2.5 was conducted in the winter time, contributions of NO 2 from microbial N cycle, NH 3 emission from seawater (δ 15 N = −8‰ to −5‰ in Jickells et al., 2003) and lightening NO x (δ 15 N = −0.5‰ to +1.4‰; Hoering, 1957) were quite small, with relatively lower contribution than anthropogenic N sources to the formation of near-surface PM 2.5 , especially in urban circumstances.
To date, δ 15 N values of various NO 2 and NH 3 emissions are unavailable in China. However, according to source δ 15 N data compiled from previous studies (Table 1, Fig. 1), δ 15 N values were distinctive between most typical sources, which have been broadly used in isotopic tracing or partitioning of atmospheric N deposition (e.g. Elliott et al., 2007Elliott et al., , 2009; Kawashima and Kurahashi, 2011). In this study, we did not use δ 15 N data of trations of NO 2 averaged 4.3 ± 1.3 μg/m 3 at the background site (Table 2). Ambient NH 3 and SO 2 concentrations were not available at the Menyuan site (37°36′ N, 101°15′ E; 3295 m); however, these concentrations were reported as 4.8 μg/m 3 and 0.31 μg/m 3 , respectively, at Waliguan (a global baseline station, 36°30′ N, 100°10′E, 3816 m), another background site in Qinghai (Carmichael et al., 2003). The estimated molar ratio of ambient NH 3 to (NO 2 + 2 * SO 2 ) averaged 2.7 at the background site (Table 2).

Major sources of N in PM 2.5 of Beijing
According to the source appointment of PM 2.5 at Beijing during the severe haze episode of January 2013 (Huang et al., 2014;Zhang et al., 2015), the following six dominant sources can be assigned for bulk N of PM 2.5 . S1: N from dust, S2: NO 2 from coal combustion, S3: NH 3 from coal combustion, S4: NO 2 from vehicle exhausts, S5: NH 3 from vehicle exhausts, S6: NH 3 from animal wastes (mainly domestic wastes and sewages). The box encompasses the 25th-75th percentiles, whiskers are SD values. The line and square in each box mark the median and arithmetic mean values, respectively. The number of jittered replicate δ 15 N data (dots around the boxes) is 1-34. Mean and SD values of source δ 15 N data were used in the SIAR model. δ 15 N values of N from dust were assumed as those of surface soils  according to the air mass backward trajectories (Fig. 2). SOURCE APPOINTMENT OF NITROGEN IN PM 2.5 BASED ON BULK δ 15 N SIGNATURES quite low in the PM 2.5 of Beijing. First, the low molar ratios of ambient NH 3 to (NO 2 + 2 * SO 2 ) as 0.3 (Table 2) reflected a relatively thorough neutralization of NH 3 by acidic gases, producing relatively more stable ammonium salts of NH 4 NO 3 , NH 4 HSO 4 and (NH 4 ) 2 SO 4 . Second, the molar ratios of NH + 4 to (NO − 3 + 2 * SO 2− 4 ) were calculated as 0.8 (Table 2), indicating a full fixation of NH + 4 by existing NO − 3 and SO 2− 4 for PM 2.5 of Beijing. In the calculation, NH + 4 is the actual molar concentrations of NH + 4 in PM 2.5 while the (NO − 3 + 2 * SO 2− 4 ) in PM 2.5 represents the concentrations of NH + 4 that can be fixed by NO − 3 and SO 2− 4 . More often, due to the high emissions of anthropogenic SO 2 and NO 2 in urban environments, NH 3 , after converting to NH + 4 , reacts mainly with acids formed by SO 2 and NO 2 , with little opportunity of NH 3 losses from PM 2.5; thus, no substantial 15 N enrichment in NH + 4 of PM 2.5 (Yeatman et al., 2001a;Pavuluri et al., 2010;Kawashima and Kurahashi, 2011) is observed. Consequently, bulk δ 15 N values of PM 2.5 at Beijing were mainly controlled by the mixing of N sources with inappreciable isotopic effects.

Major sources of N in PM 2.5 of Menyuan
According to the molar ratios of ambient NH 3 to NO 2 (ca. 3.0) or NH + 4 to NO − 3 (ca. 3.3) in PM 2.5 at Menyuan (Table 2), inorganic N in both ambient atmosphere and PM 2.5 were dominated by NH 3 and NH + 4 , respectively. Moreover, δ 15 N values of PM 2.5 did not assemble those of dust N and/or natural N (mainly NO 2 from N cycle) emissions; instead, they were much higher than those of potential sources (Table 1, Fig. 1). More likely, agricultural and biogenic NH 3 sources should be important to bulk N of the background PM 2.5 . Hence, we assigned major N sources of PM 2.5 at the background site as follows: S7: N from dust, S8: NO 2 from biomass burning, S9: NH 3 from biomass burning, S10: NH 3 from animal wastes, S11: NH 3 from fertilizer application, S12: NO 2 from microbial N cycle.
As bulk δ 15 N values of PM 2.5 at Beijing were distributed within those of major sources (Fig. 1), no substantial isotopic effect between N sources and bulk N of PM 2.5 at Beijing was assumed. In particular, as inorganic N of PM 2.5 was dominated by NH + 4 (with a mean molar ratio of NH + 4 to NO − 3 of 2.5; Table  2), the isotope effect of NH 3 ↔ NH + 4 equilibrium is considered  Carmichael et al. (2003), He et al. (2014), Wei et al. (2015). Data of NH 3 and SO 2 were cited from the background site of Waliguan in Qinghai Province (Carmichael et al., 2003).

Beijing (CRAES site)
Menyuan, Qinghai n-NH 3 /(n-NO 2 + 2*n-SO 2 ) 0.3 2.7 ( Fig. 1). The regulation of acidic gases-to-NH 3 stoichiometry on the reaction and isotopic effect between NH 3 and NH + 4 was supported by a positive correlation between δ 15 N values and NH + 4 ∕(NO − 3 + 2 * SO 2− 4 ) ratios in PM 2.5 (Fig. 3). Accordingly, a net isotopic effect of NH 3 (g) ↔ NH + 4 (p) at equilibrium (ε eq ) (33‰; Heaton et al., 1997) was considered in the SIAR model for the background PM 2.5 (details down in Section 4.2). However, it should be noted that isotope effects for the atmospheric NH 3 (g) ↔ NH + 4 (p) equilibrium in the field circumstances remain unclear. The value of 33‰ is the only empirical one for 15 N enrichment in particulate NH + 4 (Heaton et al., 1997). Experimental studies have been conducted on the isotope fractionations of NH 3 volatilization (e.g. Li et al., 2012), but it is uncertain what factors can be used to make corrections of the isotope effects for background PM 2.5 . Further studies are strongly needed to verify the relationships between the isotope effects and the ratio of NH 3 (g) to NH + 4 (p), which may be a feasible factor to make a correction of the isotope effects.  each source to the mixture . It can substantially incorporate the uncertainties associated with multiple sources, fractionations and isotope signatures (Moore and Semmens, 2008;Davis et al., 2015). In our estimations, uncertainties should be evaluated for the δ 15 N variabilities of bulk N in PM 2.5 and N sources, isotopic effect of the NH 3 (g) ↔ NH + 4 (p) equilibrium.
By defining a set of N mixture measurements on J isotopes with K source contributors, the mixing model can be expressed as follows (Parnell et al., 2010): where all F values sum to 1 (unity), X ij is the isotope value j of the mixture i, in which i = 1, 2, 3, …, N and j = 1, 2, 3, …, J; S jk is the source value k on isotope j (k = 1, 2, 3, …, K) and is normally distributed with mean μ jk and standard deviation ω jk ; F k is the proportion of source k estimated by the SIAR model; c jk is the fractionation factor for isotope j on source k and is normally distributed with mean λ jk and standard deviation τ jk ; and ε ij is the residual error representing the additional unquantified variation between individual mixtures and is normally distributed with mean 0 and standard deviation σ j . A detailed description of this model can be found in Moore and Semmens (2008), Jackson et al. (2009) andParnell et al. (2010). To estimate the contributions of N sources in the PM 2.5 samples at two study sites (n = 14 for each), one isotope (j = 1) (δ 15 N of bulk N) and six potential N sources (as discussed in Sections 4.1 and 4.2: S1-S6 for Beijing and S7-S12 for Menyuan) (Fig. 1) are utilized. δ 15 N values of replicate PM 2.5 samples at each study site were analysed in the SIAR model as one group.
Our estimation showed that the contribution of NO 2 (F NO 2 ) reached 41 ± 11% in bulk N of PM 2.5 in Beijing, which was much higher than F NO 2 at the background site (22 ± 10%) ( Table  3). The mean ratios of F NH3 to F NO 2 were about 1.6 and 4.4 for PM 2.5 at Beijing and at the background site, respectively (Table  3), which generally followed the molar ratios of NH + 4 to NO − 3 in PM 2.5 (Table 2). Aqueous phase reaction experiments have shown that atmospheric NO 2 and NH 3 potentially react with organic compounds to form organic N (Ge et al., 2011;Pavuluri et al., 2015), which might contribute to the high secondary organic aerosols during the study haze event in Beijing (Huang et al., 2014). In Beijing, anthropogenic N in PM 2.5 averaged 81% of its bulk N and was mainly derived from N emissions of fossil fuel combustions, with the highest contribution (ca. 25%) from NO 2 of coal combustion (Table 3; Fig. 4). The N emissions from coal combustion showed higher contributions (ca. 39%) than traffic emissions (ca. 32%), fossil-derived NO 2 contributed

Using the SIAR model to partition bulk N in PM 2.5
The proportional contributions (F, %) of major sources to N in PM 2.5 are estimated using the SIAR model. This model uses a Bayesian framework to establish a logical prior distribution based on Dirichlet distribution (Evans et al., 2000), and then to determine the probability distribution for the contribution of  behind the kinetic and equilibrium isotope effects should be explored and properly considered in future studies. more N (ca. 39%) than fossil-derived NH 3 (ca. 30%) (Table 3; Fig. 4). Comparable contributions (ca. 14-16%) were observed between NH 3 from coal combustion and NH 3 from traffic emissions, between NH 3 and NO 2 from vehicle exhausts (Fig. 4). Accordingly, fossil-derived NH 3 emissions substantially contributed to urban PM 2.5 pollution; regulatory controls of N emissions from coal combustion and urban transportation are important to advert the risk of severe haze episodes in Beijing. The N in PM 2.5 at the background site was mainly contributed by N emissions from biomass burning (46 ± 10%) and NH 3 volatilization (34 ± 12%) ( Table 3). The contribution of NH 3 from biomass burning (29 ± 6%) was comparable with the total contributions of NH 3 from animal wastes and fertilizer application (ca. 35%) ( Table 3). Biomass burning contributed less N as NO 2 (17 ± 10%) than as NH 3 (29 ± 6%) to N of PM 2.5 at the background site (Table 3). Higher production of NH 3 than NO 2 from biomass burning has been documented previously (Crutzen and Andreae, 1990). A burning experiment by Lobert et al. (1990) showed higher emission ratios of NH 3 (ca. 3.8%) than that of SO 2 (ca. 0.3%) during biomass burning. The emission factors of NH 3 were ca. 2-5 times higher than that of SO 2 from various types of biomass burning (Andreae and Merlet, 2001).

Remarks
This study attempted to quantify major sources of N in PM 2.5 based on bulk δ 15 N analysis using a Bayesian isotope mixing model. The isotopic effect of NH 3 ↔ NH + 4 equilibrium was recognized under the condition of lower acid gases relative to ambient NH 3 , which was a main reason for higher bulk δ 15 N of PM 2.5 than potential sources at the background site. Based on the estimations of SIAR model, PM 2.5 of Beijing derived N mainly from coal combustion and vehicle exhausts, while background PM 2.5 derived N mainly from biomass burning and NH 3 volatilization. Regulatory controls of N emissions from coal burning and urban transportation are important and effective steps to reduce the risk of severe haze episodes in Beijing. However, emissions of N from non-fossil emissions (particularly biomass burning) in broad rural areas should be stressed to meet a rigorous reduction of reactive N emissions in China.
Although δ 15 N interpretation using the SIAR model provided proportional contributions of major sources to bulk N in PM 2.5 , further investigations are needed to validate the assumptions and boundary conditions in this work. Particularly, δ 15 N analyses of gaseous N emissions should be implemented for reducing the uncertainties of source δ 15 N values. So far, isotopic studies on gaseous N emissions from typical anthropogenic and natural emissions are still sparse globally, especially in China. Isotope effects revealed in conversions between NO and NO 2 , NO x and NO − 3 (Monse et al., 1969;, NH 3 and NH + 4 (Heaton et al., 1997) and the regulatory mechanisms