Active speciated NHx collection was conducted using a glass honeycomb denuder-filter pack sampling system (ChemComb speciation cartridge) during all campaigns. This collection system has been extensively described for its ability to speciate between reactive inorganic gases and particulate matter for offline concentration determination (Koutrakis et al., 1988, 1993). Briefly, ambient air is drawn into the sampler, and reactive gases are removed under laminar flow conditions such that radial mixing can only be achieved via diffusion-based processes. Particulates, with their much lower diffusion velocity compared to gases, cannot migrate to the walls of the denuder during the residence time within the unit and are collected on a downstream filter pack. The samplers are also held vertically to limit the potential for gravitational settling of particles onto the denuder surfaces, such that particulates do not contribute to the denuder extract (Ali et al., 1989). The sampler consisted of a PTFE-coated inlet to minimize reactive gas loss, a PM2.5 impactor plate, a base-coated honeycomb denuder (2 % carbonate (w/v) +1 % glycerol (w/v) in 80:20 water–methanol (v/v) solution), acid-coated honeycomb denuder (2 % citric acid (w/v) +1 % glycerol (w/v) in 20:80 water–methanol (v/v) solution) to collect NH3, and a filter pack to collect pNH4+. The base-coated denuder was used to remove atmospheric acids (e.g., HNO3, SO2, and hydrochloric acid (HCl)) as a precaution to reduce collection-related gas–particle interactions. Recently, this sampling system has been shown to quantify δ15N(NH3) with a precision (±1σ) of ±0.8 ‰ from laboratory experiments and field testing (Walters and Hastings, 2018). The PTFE-coated air inlet (∼4 cm) has been shown to lead to a negligible loss of NH3 and induce insignificant δ15N(NH3) fractionation (Koutrakis et al., 1993; Walters and Hastings, 2018). The samplers were directly exposed to ambient air without the use of an additional inlet tubing to prevent the loss of NH3.

In the first measurement campaign (near-highway monitoring in summer of 2017), pNH4+ was collected using a single Fluoropore PTFE membrane filter (Millipore, 1.0 µm pore, 47 mm diameter). However, due to potential loss of semivolatile NH4NO3, all subsequent campaigns utilized a nylon filter (Cole-Parmer, 0.8 µm pore, 47 mm diameter) that has been shown to collect and retain pNO3- quantitatively (Yu et al., 2005). A significant fraction of pNH4+ collected on denuded nylon filters may volatilize (Yu et al., 2006), such that a backup acid-coated (5 % citric acid (w/v) in water) cellulose filter (Whatman, 8 µm pore, 47 mm diameter) is used to capture any volatilized NH3 from the collected particles and/or NH3 breakthrough during conditions of denuder saturation (Walters et al., 2019). All collections were conducted at a flow rate of 10 Lmin-1 using a mass-flow controller (Dakota mass-flow controller 6AGC1AL55-10AB2; precision ±1 %) attached to an oilless vacuum pump (Welch 2546B-01). All denuder and filter preparation, handling, and extraction techniques have been previously described (Walters et al., 2019; Walters and Hastings, 2018) and are summarized in the Supplement (Sect. S1).

The concentrations of the denuder and filter extraction solutions were analyzed using a combination of standardized colorimetry and/or ion chromatography analytical techniques. Colorimetric analysis included measurement of [NH4+] based on the indophenol blue method with absorbance detection at 625 nm (e.g., US EPA Method 350.1), as well as [NO2-] via diazotization with sulfanilamide dihydrochloride followed by detection of absorbance at 520 nm (e.g., US EPA Method 353.2) that was automated using a discrete UV–visible spectrophotometer (Westco SmartChem discrete analyzer) at Brown University. These analyses were conducted for all samples collected in the US (i.e., near-highway and mobile measurements). Pooled standard deviations (SDs) (±1σ) of replicate measurements of quality control standards were ±0.35 µmolL-1 (n=48) and ±0.23 µmolL-1 (n=60), and the average relative SDs (RSDs) were 1.3 % and 0.81 % for [NH4+] and [NO2-], respectively. All samples collected in the Shenyang tunnel were analyzed for [NH4+], [NO2-], [NO3-], and [SO42-] using ion chromatography (Dionex™ ICS-600) at the Institute of Applied Ecology, Chinese Academy of Sciences. Cations were quantified using a Dionex™ CS12A column and CQ12A guard column with 10 mmolL-1 methanesulfonic acid as the eluent. Anions were quantified using a Dionex™ AS22 column and AQ22 guard column with 4.5 mmolL-1 sodium carbonate and 1.4 mmolL-1 sodium bicarbonate as the eluent. For all quantified ions, the RSD was less than 1.5 %. The limit of detection (LOD) values of the quantified ions were no higher than 0.5, 0.2, 2.0, and 1.5 µmolL-1 for [NH4+], [NO2-], [NO3-], and [SO42-], respectively. The measured [NH4+] was used to calculate the concentrations of NH3 and pNH4+ in the traffic plumes, while [NO2-] was quantified because it will interfere with nitrogen isotopic analysis of NH4+ (Zhang et al., 2007), but [NO2-] was never measured above the LOD. The gases collected on the base-coated denuder were generally below detection limits and were not reported in this work.

The quantification of δ15N(NH4+) was performed separately for the acid-coated honeycomb denuder, the particulate filter, and the acid-coated cellulose filter extraction solutions, corresponding to NH3, pNH4+, and volatilized pNH4+ (and/or NH3 breakthrough during denuder saturation conditions), respectively. Briefly, δ15N(NH4+) was measured based on an established offline wet-chemistry technique involving hypobromite (BrO-) oxidation and acetic acid/sodium azide reduction (Zhang et al., 2007), which was conducted for samples with [NH4+] > 2 µmolL-1. Samples were diluted to at least 10 µmolL-1 of NH4+ and then oxidized to NO2- using BrO- in an alkaline solution as previously described (Zhang et al., 2007). After a reaction time of at least 30 min, the reaction was stopped by 0.4 mL addition of 0.4 molL-1 sodium arsenite to remove excess BrO-. The concentration of the product (NO2-) was then measured to confirm the quantitative conversion of NH4+ to NO2-. The product (NO2-) was reduced to nitrous oxide (N2O) using sodium azide buffered in an acetic acid solution based on previously described chemical protocols (McIlvin and Altabet, 2005).

Samples were then analyzed for their δ15N(N2O) composition using an automated N2O extraction system coupled to a continuous-flow isotope ratio mass spectrometer for m/z 44, 45, and 46 measurements. These measurements were conducted at Brown University for samples collected within the USA and at the Institute of Applied Ecology, Chinese Academy of Sciences, for samples collected within the Shenyang tunnel. In each batch analysis, samples were calibrated relative to internationally recognized N isotopic NH4+ reference materials. These reference materials underwent the same chemical processing as the samples and were used to correct for isotopic fractionation and blank effects resulting from the chemical conversion of NH4+ to N2O. At Brown University, two international reference materials were used that included IAEA-N2 and USGS25 with δ15N(NH4+) values of 20.3 ‰ and -30.3 ‰, respectively (Böhlke et al., 1993; Gonfiantini, 1984). Repeated measurements of these reference materials yielded SDs (±1σ) of ±0.65 ‰ (IAEA-N2; n=25) and ±0.73 ‰ (USGS25; n=25) and an overall pooled SD of ±0.69 ‰ (n=50). At the Institute of Applied Ecology, Chinese Academy of Sciences, three reference materials were used that included IAEA-N1, USGS25, and USGS26 with δ15N(NH4+) values of 0.4 ‰, -30.3 ‰, and 53.7 ‰ (Böhlke et al., 1993; Gonfiantini, 1984), respectively. These materials had measured SDs of ±0.53 ‰ (IAEA-N1; n=8), ±0.24 ‰ (USGS25; n=8), and ±0.45 ‰ (USGS26; n=8) and an overall pooled SD of ±0.42 ‰ (n=24). All N isotopic compositions are reported relative to reference standards using delta (δ) notation in units of per mill (‰). 1δ(‰)=1000RsampleRref-1, where R is the ratio of the heavy to light isotope (i.e., 15N/14N) for the sample and reference, respectively. Atmospheric nitrogen (N2) is the established international delta-scale reference for N isotopic composition.

The targeted analytes were corrected for field blanks, and ambient air concentrations were then calculated based on the volume of sampled air and reported in units of parts per billion by volume (ppbv) and microgram per cubic meter (µgm-3) for NH3 and pNH4+, respectively. The effective volume of air sampled by the ALPHA passive sampler was calculated as the following: 2V=DAt/L, where V is the volume of sampled air (m3), D is the NH3 diffusion constant (= 2.09 × 10-5 m2s-1), L is the stationary air layer within the sampler (= 0.006 m), t is the time of exposure (h), and A is the cross sectional area (= 3.4636 × 10-4 m2) (from ALPHA Sampler User Instructions).

The method detection limit (MDL) for [NH3] and [pNH4+] determination for the active sampling technique (i.e., denuder-filter pack) was calculated as 3 times the SD of the field blanks. The MDL was reported based on the typical collection times and reported separately for each sampling environment (Table 1). The reported [NH3] and [pNH4+] precisions using the denuder-filter pack sampling device were based on five separate replicate sample collections conducted at the near-highway stationary site and expressed as the relative SD (RSD %) (Table 1). The error bars of [NH3] and [pNH4+] quantified using the denuder-filter pack in subsequent figures represent the ± RSD % when above the MDL. Some collections had [pNH4+] below the MDL, and these samples were reported as 0.5±0.5 MDL. Multiple passive samplers (i.e., ALPHA) were always simultaneously collected, such that RSD % was not explicitly determined, and results were reported as x‾±1σ of the multiple collections.

Significant NH4+ field blanks were found on the acid-coated honeycomb denuder and the acid-coated cellulose filter. A subset of these blanks was analyzed for δ15N(NH4+) and found to have relatively consistent values of -9.6±1.3 ‰ (n=3) and -10.9±1.4 ‰ (n=3) for the acid-coated honeycomb denuder and acid-coated cellulose filter, respectively. Corrections for δ15N were made based on mass balance to account for the observed blanks as previously described (Walters et al., 2019): 3δ15N(NH4+)sample=δ15N(NH4+)total[NH4+]total-δ15N(NH4+)blank[NH4+]blank[NH4+]total-[NH4+]blank, Blank δ15N(NH4+) corrections were made for all samples when the fraction of the field blank (fBlank=[NH4+]blank/[NH4+]total) was less than 30 % of the total collected NH4+, as the propagated δ15N uncertainty generally did not exceed ±2.5 ‰ for this fBlank value. Samples with an fBlank that exceeded 30 % were not reported for δ15N. This requirement as well as the azide method detection limit (i.e., 2 µmolL-1) limited our ability to quantify δ15N(pNH4+) for samples collected at the near-highway monitoring site and mobile on-road measurements, such that only δ15N(NH3) was reported for the collections conducted in the USA. The collection media blank also impacted the mobile δ15N(NH3) measurements, as 6 out of 20 samples had a fBlank > 30 %. Error bars reported for subsequent δ15N values represent the propagated uncertainty that includes the collection uncertainty and the blank contribution. Replicate collected samples at the near-highway site indicated that δ15N(NH3) from NH3 collected using an acid-coated denuder had an average reproducibility within 0.8 ‰ (n=5) (Table 1), consistent with previous field measurements (Walters and Hastings, 2018).

Seasonal NHx collections at the near-highway monitoring site were performed under a variety of environmental conditions (Table 2). Overall, the near-highway [NH3] ranged between 5.8 and 20.2 ppbv during summer and 2.4 and 20.9 ppbv during winter at the near-highway monitoring location (Fig. 1a). The average [NH3] (x‾±1σ) was 14.0±4.0 ppbv (n=32) and 12.0±4.8 ppbv (n=22) for summer and winter, respectively (Table 2), which was not found to be significantly different (p>0.05). Diel [NH3] patterns were observed during both summer and winter, with significantly lower values occurring during the night and/or early morning collection period (Table 2). The dependence of [NH3] on the vector-averaged wind direction is shown in Fig. 2. Overall, the near-highway monitoring site was downwind of I-95 for 51 out of 54 NHx collection periods (Fig. 2). The [NH3] was significantly lower when the wind direction indicated the monitoring site was upwind of I-95 compared to when downwind of I-95, with averages of 5.8±2.7 ppbv (n=3) and 13.6±4.2 ppbv, respectively (p<0.01). Strong positive linear correlations were found between [NH3] and the mean [CO] during each collection period during summer (r=0.736, p<0.01) and winter (r=0.821, p<0.01), with slopes (mol mol-1) of 0.025±0.005 and 0.027±0.005, respectively (Fig. 3). These observed relations between [NH3] and [CO] were similar to previously reported values of 0.031±0.005 from on-road measurements in New Jersey and California in the United States using high-resolution open-pathway sensors (Sun et al., 2014, 2017), as well as 0.031 to 0.038 based on fitted NHx and CO slopes from aircraft measurements in the Californian South Coast Air Basin (Nowak et al., 2012). The similarity of these measurements indicated that the traffic plumes measured in this study were representative of previous literature reports in the USA, and the active collection of NH3 using a denuder-filter pack sampling technique was suitable for reproducing accurate [NH3] under traffic plume environmental conditions.

Roadside [pNH4+] ranged from 0.045 to 0.938 µgm-3 and from 0.117 to 2.327 µgm-3 during summer and winter, respectively (Fig. 1b). The average [pNH4+] was 0.302±0.208 µgm-3 (n=32) and 0.530±0.468 µgm-3 (n=22) during summer and winter, respectively (Table 2), which was significantly different (p<0.05). During winter, a [pNH4+] outlier of 2.327 µgm-3 was identified based on a Grub's t test (p<0.05). However, even with the removal of this outlier, the seasonal [pNH4+] average was found to be significantly different (p<0.05). NHx speciation was quantified as f(NH3): 4fNH3=[NH3](mol)NH3+pNH4+(mol). Overall, f(NH3) ranged from 0.889 to 0.996 during summer and from 0.878 to 0.986 during winter (Fig. 1c), indicating that NH3 was the dominant NHx species during both summer and winter. The average f(NH3) was 0.972±0.022 (n=32) and 0.944±0.029 (n=22) during summer and winter, respectively (Table 2). The average seasonal f(NH3) was found to be statistically different (p<0.05), indicating a greater extent of NH3 partitioning to pNH4+ during winter. Significant correlations were observed between f(NH3) and relative humidity for both summer (r=-0.533, p<0.01) (Fig. S4) and winter (r=-0.613, p<0.01) (Fig. S5).

The measured δ15N(NH3) ranged from 2.6 ‰ to 9.3 ‰ and from 4.9 ‰ to 10.1 ‰ during the summer and winter, respectively (Fig. 1d). The δ15N(NH3) averages were 6.4±1.7 ‰ (n=32) and 8.1±1.4 ‰ (n=22) during summer and winter, respectively (Table 2), which were significantly different (p<0.05). The dependence of δ15N(NH3) on the vector-averaged wind direction is shown in Fig. 2. Overall, the δ15N(NH3) values were not found to be significantly different when the monitoring site was upwind or downwind of I-95, with averages of 7.6±1.4 ‰ (n=3) and 7.1±1.8 ‰ (n=51), respectively (p>0.05), which is likely due to the proximity of the sampling location to air masses significantly influenced by vehicle emissions. No statistical difference was found between the collection period and δ15N(NH3) during the winter (p>0.05), but significantly lower δ15N(NH3) values were observed during the summer for the night and/or early morning sample (00:30 to 06:30) (p<0.05) (Table 2). Significant correlations between δ15N(NH3) and f(NH3) were observed for both summer (r=0.349, p<0.05) (Fig. S4) and winter (r=0.535, p<0.05) (Fig. S5). However, these correlations were found to be impacted by influential f(NH3) values during the summer and winter of 0.889 and 0.878, respectively (Figs. S4 and S5). Removing these influential f(NH3) observations resulted in an insignificant correlation between δ15N(NH3) and f(NH3) for both summer (r=0.300, p>0.05) (Fig. S4) and winter (r=0.378, p>0.05) (Fig. S5).

Tunnel temperature and relative humidity conditions remained relatively consistent throughout our sampling campaign and averaged 19.3±1.6 ∘C and 35.4±6.7 %, respectively (Table 3). Due to the elevated concentrations in the tunnel, the amount of collected NH3 on the acid-coated honeycomb denuders averaged 406±125 µg, indicating the laboratory-determined operative capacity of ∼400 µg was often exceeded (Walters and Hastings, 2018). The citric acid coated filter collected no more than 275 µg of NH3, which was within the laboratory-determined operative capacity of at least 350 µg (Walters et al., 2019). Thus, our NHx measurements are expected to be accurate, but there could be uncertainty in the NHx speciation; this is because the NH4+ extracted from the acid-coated denuder and nylon filter will have a low bias due to denuder saturation and pNH4+ volatilization, respectively, and NH4+ extracted from the acid-coated filter will derive from both NH3 breakthrough and NH3 volatilized from the nylon filter. Therefore, our concentration results and analysis of samples collected in the Shenyang tunnel will focus on [NHx]. Overall, [NHx] ranged from 64.4 to 210.6 ppbv and averaged 132.5±45.8 (n=21) (Fig. 4a; Table 3). An obvious [NHx] diel cycle was observed in which higher concentrations occurred during periods when the tunnel was open compared to sampling periods in which the tunnel was closed to vehicle passage, with averages of 136.8±18.8 ppbv (n=7), 181.2±23.0 ppbv (n=7), and 79.4±14.4 ppbv (n=7) for the 06:00 to 14:00, 14:00 to 22:00, and 22:00 to 06:00 LT collection periods, respectively (Table 3).

We have estimated f(NH3), assuming that the pNH4+ in PM2.5 was linked to the SO42--NO3--NH4+ thermodynamic system and that the influence of other ions (e.g., Na+, Ca2+, or Cl-) had a negligible impact on the chemistry of this system (Shah et al., 2018). Ion-mass balance was utilized to calculate the expected [pNH4+] for each collection period based on the measured [pNO3-] and [pSO42-] (Fig. 4b) from the aqueous filter extracts: 5[pNH4+](mol)=(2[pSO42-]+[pNO3-])(mol). Utilizing the ion-mass balance approach, f(NH3) was estimated to range between 0.856 and 0.997 and averaged 0.956±0.038 (Fig. 4c; Table 3). NHx speciation was also estimated using ISORROPIA, which is a gas–aerosol equilibrium partitioning model (Fountoukis and Nenes, 2007; Nenes et al., 1998). Model inputs included the measured [NHx], [pNO3-], and [pSO42-] and average relative humidity and temperature for each collection period, and the model was run in the forward direction in the metastable state. The f(NH3) was then calculated based on the model output of [NH3] and [pNH4+] (Table S1 in the Supplement). Overall, there was a near-exact agreement in f(NH3) between the ion-mass balance and the ISORROPIA approaches, noting that ISORROPIA was not used for the first five collection periods due to the absence of relative humidity and temperature data (Fig. 4c). Overall, this analysis indicated that NHx in the tunnel was primarily in the form of NH3, consistent with the near-highway stationary observations.

The measured δ15N from NH4+ extracted from the acid-coated denuders, nylon filters, and acid-coated filters averaged 6.0±5.6 ‰ (n=21), 1.0±10.7 ‰ (n=21), and -20.0±10.1 ‰ (n=21) (Fig. S6), respectively. These δ15N differences, to some degree, reflect differences in the δ15N of ambient NH3 and NH4+ but are difficult to interpret due to the ambiguity in NHx speciation. Since NHx speciation was not achieved in the tunnel collections due to denuder saturation, our reported isotopic results and analysis will focus on δ15N(NHx), with the expectation that it primarily represents NH3. The δ15N(NHx) was calculated for each sampling period using mass balance: 6δ15N(NHx)=fNH4+-denuder,δ15N(NH4+)denuder,+fNH4+-nylonδ15N(NH4+)nylon,+fNH4+-acidfilterδ15N(NH4+)acidfilter, where fNH4+-denuder, fNH4+-nylon, and fNH4+-acidfilter represent the fractions of NH4+ extracted from the denuder, nylon filter, and acid-coated filter, respectively, for each sampling event. Overall, the δ15N(NHx) ranged from -1.6 ‰ to 9.2 ‰ (Fig. 4d) and had a numerical average of 2.9±2.5 ‰ (n=21) (Table 3). There was a strong diel cycle in δ15N(NHx) in which the 22:00 to 06:00 LT collection period that included the period the tunnel was closed to vehicle passage (i.e., 23:00 to 05:00 LT) resulted in a statistically lower δ15N(NHx) of 0.1±1.3 ‰ (n=7), relative to the 06:00 to 14:00 and 14:00 to 22:00 LT collection periods that averaged 3.6±1.0 ‰ (n=7) and 4.8±2.0 ‰ (n=7), respectively (p<0.05) (Table 3).

Overall, the on-road [NH3] ranged from 2.3 to 23.2 ppbv and averaged 7.3±4.7 ppbv (n=20) (Fig. 5b). The highest [NH3] were found to occur during collection periods near urban cores that included Boston, MA, Providence, RI, New York City, NY, and Washington, DC (Fig. 6a). The on-road [NH3] was significantly correlated with [CO] (r=0.821, p<0.01), and the linear relationship between NH3 and CO had a slope (NH3(mol):CO(mol)) of 0.026±0.005 (Fig. 3), which was similar to the near-highway relation of 0.025±0.005 and 0.0270±0.005, observed during summer and winter, respectively. On-road [NH3] was found to be significantly correlated with vehicle speed (r= -0.673, p<0.01) (Fig. S7). On-road [pNH4+] ranged from 0.047 to 0.710 µgm-3 (Fig. 5c) and averaged 0.204±0.176 µgm-3 (n=20). NHx speciation indicated that NH3 was the dominant species, which is consistent with our stationary observations, as f(NH3) ranged from 0.800 to 0.987 (Fig. 5d) and averaged 0.934±0.050 (n=20).

On-road δ15N(NH3) ranged from -3.0 ‰ to 10.1 ‰ (Fig. 5e) and averaged 5.7±3.5 ‰ (n=14). On-road δ15N(NH3) was not found to be significantly correlated with f(NH3) (r=0.249, p>0.05) or average vehicle speed (r= -0.179, p>0.05) (Fig. S7). Spatial mapping of δ15N(NH3) indicated the highest values near urban cores (Fig. 6b). Each collection period was categorized as either a trucking or highway route using the percentage of annual average daily truck traffic contributions to annual average daily traffic (U.S. Dept of Transportation, 2013) similar to that previously described (Miller et al., 2017). Routes on our on-road measurements with diesel trucks that comprised at least 25 % of the annual average daily traffic and at least a yearly average of 8500 diesel trucks per day were identified (U.S. Dept of Transportation, 2013), which were located on rural highways typically outside of urban areas. This categorization technique was used to qualitatively identify differences in vehicle fleet compositions during our measurements since real-time vehicle count data were not collected. Two sampling collection periods were identified as a trucking route, including (1) from outside Harrisburg, PA, to New Smithville, PA, along I-81 and I-78 and (2) from Kirkwood, PA, to Colliersville, NY, along I-81 and I-88. Though the number of measurements conducted on trucking routes was limited in this case study, the average on-road δ15N(NH3) values on highway and trucking routes were 6.9±1.9 ‰ (n=12) and -1.5±1.6 ‰ (n=2), respectively, which were found to be significantly different (p<0.01).

A comparison between the active and passive collection of NH3 for concentration and δ15N(NH3) characterization is summarized in Table 4. The active collection sampling technique resulted in [NH3] of 12.0±1.2 ppbv and 127.1±12.5 ppbv over the entire winter near-highway and Shenyang tunnel sampling campaigns, respectively. These concentrations were calculated from the total collected NH4+ over the sampling campaign divided by the total volume of collected air for each respective campaign, and the reported uncertainty represents the RSD of the active collection technique of 9.8 %. We note that [NH3] in the Shenyang tunnel determined using the denuder-filter pack was not measured directly but was calculated from the measured [NHx] and estimated f(NH3). The passive collection resulted in [NH3] values of 11.6 ± 1.4 ppbv (n=4) and 124±3.6 ppbv (n=3) during winter at the near-highway monitoring location and in the Shenyang tunnel, respectively, which was in close agreement with the active collection technique. The mass-weighted δ15N(NHx) values using the active collection technique were 8.0±1.1 ‰ and 3.5±0.8 ‰ during winter at the near-highway monitoring location and in the Shenyang tunnel, respectively, where the uncertainty represents the propagated error (Table 4). We note that the tunnel δ15N(NH3) technically represents δ15N(NHx); however, due to the elevated estimated f(NH3), δ15N(NHx) ≈ δ15N(NH3). The passive NH3 collection technique resulted in an average δ15N(NH3) of -7.7±0.1 ‰ (n=4) and -11.7±0.3 ‰ (n=3) during winter at the near-highway monitoring location and in the Shenyang tunnel, respectively, which was found to be significantly different from the δ15N(NH3) measured using the active collection for each sampling campaign (p<0.01). The δ15N(NH3) difference values between passive and active collection were calculated to be -15.7±1.1 ‰ and -15.2±0.9 ‰ during winter at the near-highway monitoring location and in the tunnel in Shenyang China, respectively (Table 4), indicating a consistent δ15N(NH3) offset between the active and passive sampling collection techniques.

Here we assess the drivers behind the δ15N(NH3) variabilities measured within each sampling campaign, including the seasonal difference measured at the near-highway monitoring site, the temporal variation observed during summer at the near-highway site and the Shenyang tunnel, and the spatial patterns observed from the on-road measurements. We hypothesize that the observed δ15N(NH3) variabilities could be related to (1) f(NH3) partitioning, (2) NH3 dry deposition, (3) background NH3 contributions, and/or (4) vehicle fleet composition differences.

Previously, it has been theoretically estimated and shown from field observations and laboratory studies that isotopic N equilibrium and reactions between NH3 and NH4+ can scramble the 14N and 15N distributions between these molecules, leading to the preferential partitioning of 15N into NH4+ (Kawashima and Ono, 2019; Savard et al., 2017; Urey, 1947; Walters et al., 2018). A significant positive correlation between f(NH3) and δ15N(NH3) during both summer and winter was observed at the near-highway monitoring location, which is consistent with influences from N isotopic equilibrium reactions. However, the δ15N(NH3) and f(NH3) relations were affected by a single influential f(NH3) value during both summer and winter, and removal of these points resulted in an insignificant relation between δ15N(NH3) and f(NH3) (Figs. S4 and S5). The temporal tunnel variability is not likely to be driven by f(NH3) partitioning influences as the estimated f(NH3) was not found to be significantly different between periods when the tunnel was open or closed (p>0.05), indicating a significant change in NH3 and pNH4+ partitioning did not occur during these periods. Thus, we do not expect f(NH3) partitioning and NH3 reactive sink to have played a significant role in the δ15N(NH3) variability observed at the various sampling sites. We note that the influence of N isotopic exchange reactions on δ15N(NH3) depends on the degree of NH3 and pNH4+ partitioning. Typically, f(NH3) was observed to be > 0.934, which would limit the influence of equilibrium exchange reactions to alter the measured δ15N(NH3) values. We also note that there is an equilibration time needed before N isotopic equilibrium between NH3 and pNH4+ is achieved, but this rate is currently unknown. Thermodynamic gas–fine-aerosol equilibrium has been calculated to have an equilibration time on the order of tens of minutes to several hours, depending on ambient conditions and particle characteristics (Meng and Seinfeld, 1996). Assuming a similar equilibration rate for N isotopic exchange between NH3 and pNH4+ would indicate that complete N isotopic equilibrium would likely not be achieved near NH3 emission sources, which is consistent with our observations.

NH3 dry deposition was not expected to contribute to the observed variability in the well-ventilated sampling conditions at the near-highway monitoring location and the on-road measurements. These measurements were conducted close to the emitted NH3 (e.g., typically within 5 m at the near-highway monitoring site), which should have minimized NH3 loss via dry deposition (Asman et al., 1998). However, NH3 dry deposition may have played an important role under the closed sampling environment of the tunnel and may explain the observed δ15N(NHx) temporal variability with higher values observed when the tunnel was open (4.2±1.7 ‰, n=14) compared to samples collected during periods when the tunnel was closed to traffic (0.1±1.3 ‰, n=7) (Table 3). Previously, lower NH3 emission ratios were reported from traffic plumes in tunnels relative to on-road highway measurements, which was concluded to result from contributions of NH3 dry deposition on the tunnel surfaces (Sun et al., 2017). If NH3 dry deposition is influenced by N isotopic equilibrium reactions between NH3 and the surface deposited NH4+, this would have resulted in NH3 depleted in 15N as it is removed from the atmosphere, resulting in lower δ15N(NH3) values (Walters et al., 2018). Indeed, a previous NH3 absorption–desorption study on minerals has shown the preferential removal of 15NH3 from the gaseous phase, with the degree of 15N depletion of the gaseous NH3 dependent upon the adsorbed NH3 amount (Sugahara et al., 2017). Thus, as the traffic plume ages in the absence of fresh emissions, we would expect NH3 dry deposition influences and the potential for N isotopic exchange reactions between the air and tunnel surface to be most significant, which might explain the lower δ15N(NHx) values observed during periods when the tunnel was closed. Dry deposition of NH3 during the day in the tunnel could have also impacted the measured δ15N(NH3) values, but the constant emission of NH3 likely resulted in nonequilibrium conditions, such that N isotopic equilibrium between the ambient NH3 and surface deposited NH4+ would not have been fully achieved.

Background NH3 contributions are important to identify as a possible driver of δ15N(NH3) variability. At the near-highway monitoring site, wind sector analysis found no statistical difference in δ15N(NH3) when sorted by wind direction for either summer or winter (Fig. 2). This indicates that transport from local NH3 point sources other than vehicle emissions played a minor role in the seasonal δ15N(NH3) difference. Additionally, the similar seasonal relations between [NH3] and [CO] at the near-highway monitoring site (Fig. 3) indicate that seasonal variations in background NH3 influences at the near-highway monitoring site were minor. While dilution by background air into the Shenyang tunnel during the periods when the tunnel was closed to traffic should be considered a driver of the temporal δ15N(NHx) variability, the average [NHx] and f(NH3) values were not consistent with significant mixing in of background air. When the tunnel was closed, [NHx] averaged 79.4±14.4 ppbv (Table 3), which is elevated compared to urban background [NH3] measurements previously reported from a megacity in China (Beijing) during winter of 5.22±3.75 µgm-3 (or 6.9±4.9 ppbv) (Ianniello et al., 2010). Additionally, f(NH3) was elevated during the collection period that the tunnel was closed, averaging 0.937 ± 0.045 (Table 3), consistent with local emissions rather than contributions from background air that tends to have a lower f(NH3) value such as reported to be typically below 0.6 during November based on data collected from Beijing, China (Zhang et al., 2018). Thus, we do not expect the background NH3 contribution to have played a significant role in the tunnel temporal δ15N(NHx) variability. Furthermore, we do not expect background NH3 contributions to have played a significant role in the spatial δ15N(NH3) variability observed from the on-road measurements in the northeastern USA. While lower δ15N(NH3) values in non-urban regions might be consistent with an increased contribution from background agricultural emissions, which tend to have a low δ15N(NH3) signature (e.g., -31 ‰ to -14 ‰; Hristov et al., 2011), we expect these temperature-dependent emissions to be minimal during the winter when the on-road measurements were conducted.

Vehicle fleet compositions could have a strong influence on the measured δ15N(NH3) variabilities if gasoline- and diesel-powered engines, which utilize different types of NOx reduction technologies that lead to NH3 emissions (Suarez-Bertoa and Astorga, 2018), have different δ15N(NH3) emission signatures. Categorization of our on-road collection routes into highway routes and trucking routes resulted in statistically significantly different δ15N(NH3) values of 6.9±1.9 ‰ (n=12) and -1.5±1.6 ‰ (n=2), respectively, supporting the idea that the δ15N(NH3) spatial variation was influenced by fleet composition. This would also be consistent with previous findings that vehicle fleet composition was the main driver of spatial on-road variability observed for δ15N(NOx) (Miller et al., 2017). Vehicle fleet NH3 emissions driven by reduction technologies may have also influenced the seasonal δ15N(NH3) difference observed at the near-highway monitoring location. Under cold ambient conditions of -7 ∘C, diesel-powered vehicles equipped with the selective catalytic reduction (SCR) technology were reported to have minimal emission of NH3. In comparison, gasoline-powered vehicles equipped with a three-way catalytic converter (TWCC) were reported to have increased NH3 emissions relative to warmer conditions at 23 ∘C (Suarez-Bertoa and Astorga, 2018). Vehicle fleet composition may also explain the significantly lower δ15N(NH3) values during the summer night and/or early morning collection period at the near-highway monitoring site (Table 2). Vehicle fleet composition was not monitored in this study, but a previous study has reported relatively higher truck traffic compared to gasoline vehicles from near-highway measurements during the night and/or early morning before morning rush hour (Wang et al., 2018). A lower δ15N(NH3) signature from diesel emissions compared to gasoline, as supported by our on-road measurements, would explain both the seasonal differences in δ15N(NH3) and the temporal δ15N(NH3) variability observed primarily during summer. To date, there are neither direct tailpipe measurements of δ15N(NH3) from gasoline- and diesel-powered vehicle nor an explanation for the expected δ15N(NH3) signatures of vehicle-derived emissions. Future work is needed to evaluate direct tailpipe δ15N(NH3) signatures from gasoline- and diesel-powered vehicles to test our hypothesis. We note that while there was a statistically significant seasonal difference in the measured δ15N(NH3) at the near-highway monitoring site, the absolute difference of ∼ 1.7 ‰ was small.

A comparison between active and passive sampling was conducted to evaluate the performance of the varying NH3 collection techniques. Overall, remarkably similar [NH3] values were determined using the active (i.e., denuder-filter pack) and passive (i.e., ALPHA) sampling techniques (Table 4). The finding of similar [NH3] values between passive (ALPHA) and active sampling of NH3 is consistent with previous comparisons (Day et al., 2012; Pan et al., 2020) and provides support that passive collection of NH3 may be a convenient approach for spatial documentation of near-surface [NH3].

While the two sampling techniques produced consistent [NH3], significant differences in δ15N(NH3) were observed. The mass-weighted δ15N(NH3) values using the active sampling technique were 8.0±1.1 ‰ and 3.5±0.8 ‰, while the values using the passive sampling technique were -7.7±0.1 ‰ (n=4) and -11.7±0.3 ‰ (n=3) at the near-highway site (winter) and in the Shenyang tunnel, respectively (Table 4). The measured traffic-derived δ15N(NH3) values via the passive sampler were similar to previous measurements utilizing a similar sampling approach that included measurements in a tunnel in the USA and a tunnel in China with reported values of -3.4±1.2 ‰ (n=2) (Felix et al., 2013) and -14.2±2.6 ‰ (n=8) (Chang et al., 2016), respectively. While our passive δ15N(NH3) values were generally consistent with previous reports, there are large offsets between the passive and active sampling techniques that were calculated to be -15.7±1.1 ‰ and -15.2±0.9 ‰ at the near-highway site and in the Shenyang tunnel, respectively (Table 4). These δ15N(NH3) offsets between passive and active NH3 collection techniques are in agreement with a value of -15.4 ‰ observed from urban background measurements conducted in Beijing, which has been concluded to be due to a diffusive isotope fractionation (Pan et al., 2020). Overall, the large δ15N(NH3) offsets observed between passive and active NH3 collection and the potential δ15N(NH3) bias in the passive collection of NH3 have several important implications. The majority of reported δ15N(NH3) source signatures have been characterized using passive sampling techniques and might be biased by approximately -15.5 ‰ under the environmental conditions during our sampling periods. These previous measurements could potentially be corrected, but further characterization of the passive sampler δ15N(NH3) offset is needed.

The measured δ15N(NH3) traffic plume signatures utilizing the active sampling technique demonstrate an overall range from -3.0 ‰ to 10.1 ‰ (Fig. 7). Our analysis indicated that δ15N(NH3) variability was influenced by fleet composition and NH3 dry deposition in aged vehicle plumes measured in a tunnel. Thus, for deriving an urban traffic plume δ15N(NH3) signature, we have considered measurements conducted under fresh plume conditions and on or near highway measurements as representative of urban vehicle NH3 emissions. These observations included the near-highway measurements conducted during both summer and winter, the mobile on-road measurements conducted on highways, and the Shenyang tunnel during operation. While there are δ15N(NH3) differences between sampling environments for this subset of observations (Fig. 7), the absolute difference in the mean δ15N(NH3) was quite small (generally within ∼ 3 ‰) and may reflect actual differences in urban vehicle fleet compositions. Overall, the constrained observations assumed to be representative of urban vehicle emissions reduces the δ15N(NH3) variability with a range of 2.1 ‰ to 10.1 ‰ (Fig. 7). The constrained δ15N(NH3) has a combined numerical average of 6.6±2.1 ‰ (n=80) (Fig. 7), which was found to not significantly differ from a normal distribution (Kolmogorov–Smirnov test of normality, p=0.528), and it is suggested to be the urban vehicle-derived traffic plume δ15N(NH3) source signature.

The recommended δ15N(NH3) vehicle-derived traffic signature of 6.6±2.1 ‰ (n=80) has a narrower range and higher value than previously reported vehicle signatures of -17.8 ‰ to 0.4 ‰ (Chang et al., 2016; Felix et al., 2013; Smirnoff et al., 2012). The difference between the recommended δ15N(NH3) vehicle-derived source signature and previous reports by Chang et al. (2016) and Felix et al. (2013) was found to be caused by a δ15N(NH3) bias from passive NH3 collection that was suggested to be driven by a diffusion isotope effect. The recommended δ15N(NH3) vehicle-derived source signature was also found to be statistically different from a previous report that actively sampled NH3 using a filter pack collection system, which reported an average δ15N(NH3) of -2.1±1.9 ‰ (Smirnoff et al., 2012). Differences between our recommended δ15N(NH3) value and previous reports by Smirnoff et al. (2012) are difficult to identify and may be related to differences in vehicle fleet compositions. Additionally, we note that this difference may be related to the potential for a positive sampling artifact associated with filter pack collection using a particulate filter and subsequent acid-coated filter for separate pNH4+ and NH3 collection, respectively, as volatilization of the collected pNH4+ could have resulted in a NH3 collection bias (Yu et al., 2006). Indeed, previous laboratory experiments have shown that NH3 volatilized from NH4 and NO3 particles collected from filters have a δ15N(NH3) value lower than the δ15N (pNH4+) by 28.6±2.7 ‰ (Walters et al., 2019). Thus, pNH4+ volatilization could have artificially lowered the reported δ15N(NH3) value and may explain the lower δ15N(NH3) values reported in Smirnoff et al. (2012) compared to our results.