Measurements of total reactive nitrogen, Nr, were accomplished by catalytic conversion to NO and detection of the NO using a chemiluminescence instrument. This NO-O3 chemiluminescence instrument is a custom-built version of the common atmospheric monitoring instrument (Williams et al., 1998) and is calibrated directly with gas-phase standards of NO. All the Nr species were converted to NO or NO2 on a high-temperature catalyst, and the NO2 was subsequently converted to NO on a lower-temperature catalyst. The high-temperature catalyst system consisted of a quartz tube (13 mm OD × 11 mm ID × 35 cm L) packed with 36 platinum (Pt) screens (Shimadzu part no. 630-00105) run at high temperature (750 ∘C), shown in Fig. 1. The catalyst bed was confined to an 8 cm long section by dimples in the quartz tube, and that section was positioned so that the gas reaching it had been equilibrated to 750 ∘C, as confirmed by a thermocouple probe. The flow through the catalyst was set to 1 standard L min-1 via a downstream flow controller. The Pt surface area was 126 cm2 and the residence time was 0.1 s at 83.3 kPa and 750 ∘C. Platinum catalysts of this kind are also known to oxidize NO to NO2, which has been the source of problems with some previous systems that were designed to measure atmospheric ammonia (NH3) (Schwab et al., 2007). In our system, the Pt catalyst is followed by a molybdenum oxide (MoOx) catalyst consisting of a solid molybdenum tube (4.2 mm ID × 32 cm L) operated at 450 ∘C, to which an 8 standard cm3 min-1 flow of pure hydrogen was added to create a stable molybdenum oxide surface. Run in this manner, the MoOx surface did not require periodic treatment at higher temperatures under reducing conditions as described by Williams et al. (1998). The heated MoOx catalyst reduces the remaining NO2 to NO. The NO chemiluminescence detection scheme used for laboratory calibrations had a fundamental sensitivity between six and seven counts per part per trillion (pptv) and the detection limit determined by the background signal in zero air (ZA) was typically 15 pptv (4σ) for a 1 s measurement. The operation of this instrument often required considerable de-tuning to keep the instrument count rates below the rollover point of the photon-counting electronics (approximately 5 MHz) for the particle concentrations generated; thus the detection limit was closer to 0.1 ppbv (corresponding to 0.3 µg m-3 for aerosol nitrate).

Measurements of total carbon (Cy) were accomplished by catalytic conversion to carbon dioxide (CO2) and detection using a CO2 analyzer. The high-temperature (750 ∘C) platinum catalyst (Fig. 1) in the Nr system should quantitatively convert carbon-containing species to CO2 in the presence of air. Gas-phase carbon conversion across similar precious metals has been studied extensively (see for example the Pt catalyst used in Veres et al., 2010). The total flow through the Pt catalyst was set to ∼ 1.5 standard L min-1 and was then split before the MoOx catalyst. In our sampling scheme 0.5 standard L min-1 of flow was directed to a LI-COR 6251 (LI-6251; Lincoln, NE) CO2 analyzer, while the remaining flow, 1 standard L min-1, was directed through the MoOx catalyst and to the NO-O3 chemiluminescence detector as detailed in Sect. 2.1.1. Run in this manner, the conversion of compounds that contain both N and C atoms can then be measured simultaneously using the NO-O3 chemiluminescence detector and LI-6251 detector in parallel.

The LI-COR instrument was internally referenced to scrubbed ZA. At ambient CO2 levels, it is challenging to retrieve reliable measurements since the signal relative to the background abundance of CO2 is small. In order to evaluate organic carbon conversion efficiency, our approach relies on using ultrapure air for aerosol generation and carrier gas flow; therefore ambient CO and CO2 are eliminated. The LI-6251 was calibrated with sub-5 ppm CO2 standards (Scott-Marrin Inc., Riverside, CA) in ultrapure air. Due to the low signal levels and the uncertainty of the low-concentration CO2 standards, the overall uncertainty of the CO2 measurements below 1 ppmv presented in this work is ±10 % for 10 s averages.

A schematic of the PILS–ESI/MS is shown in Fig. 2. The PILS (Brechtel Manufacturing Inc., Hayward, CA) was developed by Weber et al. (2001) and collects water-soluble aerosol compounds by growing particles into liquid droplets in a supersaturated water environment and then collecting the droplets. A detailed description of the PILS can be found in Sorooshian et al. (2006). The PILS is an established water-soluble aerosol collection technique that has been coupled with various mass analysis methods and was used previously by other laboratories in instrument evaluation studies (e.g., Drewnick et al., 2003; Takegawa et al., 2005; Canagaratna et al., 2007).

The PILS liquid output flow was set to 100 µL min-1 and was continuously mixed with an acetonitrile flow (100 µL min-1). The 1 : 1 volume mixture of acetonitrile and water was directed toward the custom electrospray ionization source (at ∼ 10 µL min-1) of a commercial quadrupole mass spectrometer (Balzers Instruments, QMG 422) operated in negative-ion mode for online analysis of selected water-soluble organic and inorganic compounds. The electrospray interface involved sample injection at ambient pressure through a fused silica capillary tip (30 µM ID) with a 2.5 L min-1 N2 sheath flow at a spray voltage of -3.5 kV. The MS instrument was modified from the negative-ion proton-transfer chemical-ionization mass spectrometer (NI-PT-CIMS) described in Veres et al. (2008). The flow tube was replaced with a stainless steel capillary inlet connected to the front region (I; shown in Fig. 2) held at ∼ 300 Pa. Ions were focused across this region using a planar direct current (DC) ion carpet (Anthony et al., 2014) mounted in front of the orifice leading to the second region (II). The ions were then accelerated through the collisional dissociation chamber (CDC) and collimated in the octopole ion guide at a total pressure of ∼ 1 Pa (region II). The ions were transferred to the quadrupole mass spectrometer (region III). The electron multiplier detector was maintained at a pressure of less than 6.6 × 10-3 Pa.

The ESI/MS was calibrated using volumetrically and gravimetrically prepared liquid-phase standards of the anions associated with the target compounds (e.g., SO42-, NO3-, Cl-) (Sigma Aldrich, St. Louis, MO). Anion-specific calibration factors were calculated from linear least-square fits of multipoint calibration curves. The uncertainty in the slope resulted in a maximum uncertainty of ∼ 10 % for the compounds tested. The ESI flow rate, solvent composition, analyte chemical properties, and matrix effects potentially impact the ionization and transmission efficiencies of compounds (Kostiainen and Kauppila, 2009). For these reasons, experiments were performed under similar, or as close to identical, conditions as the calibrations for instrument evaluation. The limits of detection for the anions measured with the PILS–ESI/MS were below ∼ 0.1 µg m-3 for the current system and sampling conditions. Sorooshian et al. (2006) discuss volatility losses in the PILS for several inorganic species and reported negligible loss with a collection efficiency of ≥ 96 % for mass loadings of Cl-, SO42-, and NO3- ranging from 1 to 140 µg m-3. Additionally, Orsini et al. (2003) showed the collection efficiency of ≥ 95 % for particles as small as 30 nm diameter for a 15 L min-1 sample flow rate. Ammonium (NH4+) is the major ion susceptible to volatilization as shown in Ma (2004), who indicated an underestimation of ∼ 15 %. In this study, because we were operating in the negative-ion mode, we did not measure NH4+ directly.

The efficiency of conversion of several N-containing gases by the Nr catalyst was determined through addition of a number of representative compounds that were calibrated independently. The NO signal from the converted species was then compared to the signal from an NO in N2 standard (5.38 ppmv, Scott-Marrin Inc., Riverside, CA) that was used as the working standard for this project. Typical calibration levels were in the range of 50 to 100 ppbv as determined by the mass flow controllers used to mix the standard into the measurement stream. The standards used for each compound and their associated calibrations were as follows.

The NO2 standard stream was produced from the NO working standard through gas-phase titration with a small stream of O2 in which O3 had been produced by photolysis at 184.9 nm using a mercury discharge lamp. This technique is used routinely for NOx and NOy measurement systems (Williams et al., 1998) and allows straightforward determination of NO2 conversion provided care is taken not to over-titrate the NO stream to produce NO3 and therefore N2O5. The uncertainty in the NO2 conversion determination is simply the propagated errors in the subtraction of the signals before and after titration.

The ammonia (NH3) conversion was examined using two different NH3 sources, a gas mixture (3.1 ppmv in N2, Scott-Marrin), and a permeation device (Kin-Tek, La Marque, Texas). Care was taken with these standards to keep them under flow for periods of several days in order to insure any system surfaces were equilibrated. The calibration of these standards was accomplished with ultraviolet (UV) absorption spectroscopy at a 184.9 nm wavelength using an instrument described by Neuman et al. (2003), and based on absorption cross sections reported in the literature (Tannenbaum et al., 1953; Lovejoy, 1999; Froyd, 2002). The uncertainty of NH3 conversion was propagated based on the uncertainties in flow rate and UV absorption determinations.

The hydrogen cyanide (HCN) standard consisted of a commercial gravimetric mixture of HCN in N2 (10 ppmv, Gasco, Oldsmar, FL), which was mixed into the system using a mass flow controller. The specified uncertainty of this mixture was ±10 %, and the standard concentration was verified using long-path Fourier-transform infrared (FTIR) spectroscopy to within the stated uncertainty. The HCN standard was used to produce a gas-phase stream of cyanogen chlorine (ClCN) by reaction with chloramine-T, a nonvolatile chlorinating agent, which has been described previously (Valentour et al., 1974). To do this, a small stream (5 standard cm3 min-1) of the HCN standard was combined with humidified ZA (60 % RH, 30 standard cm3 min-1) over a bed packed with glass beads coated with a solution of chloramine-T. The glass beads were prepared by coating glass 3 mm outer diameter (OD) beads with a 2 g 100 mL-1 solution and packing ∼ 20 cm3 of them in a 12.7 mm OD PFA tube and flowing ZA over them until dry. The reaction was shown to be essentially 100 % (±10 %) using proton-transfer-reaction mass spectrometry (PTR-MS), when conducted in a humidified atmosphere (RH ≥ 60 %), and by FTIR analysis of the HCN and ClCN in the gas stream before and after chlorination.

The isocyanic acid (HNCO) standard was prepared according to the methods described by Roberts et al. (2010), in which the trimer, cyanuric acid, was thermally decomposed at 250 ∘C in a diffusion cell to produce a steady stream of HNCO, which was then calibrated using long-path FTIR spectroscopy. Initially, this source has the potential to produce NH3 as an impurity, most likely because of the presence of trace amounts of water. Keeping the source under flow and above 120 ∘C at all times when not in use was found to reduce the NH3 impurity to negligible levels (< 5 %), as measured using PTR-MS. The uncertainties in the HNCO standard were propagated from the uncertainties in the HNCO cross section (Northwest Infrared, PNNL), the NH3 subtraction, and flow rates. Standard streams of both nitrobenzene and trimethylamine were produced using gravimetrically prepared solutions and a commercial liquid calibration device (Ionicon, Innsbruck, Austria). The uncertainties in these liquid calibration standards were estimated from the propagated uncertainties in the solution concentrations and the liquid and gas flow rates.

The conversion of nitrous oxide (N2O) is a potential interference in the Nr method as N2O is not typically considered a reactive nitrogen compound in the troposphere. Several experiments were conducted to determine the extent of this potential interference using a 10.1 ppmv N2O standard. The resulting conversion efficiency ranged from 0.03 to 0.05 % in dry and humidified air, respectively. These can be considered upper limits for this interference as we cannot be completely sure that there were no Nr contaminants (e.g., NO2) in the N2O standard.

Several aerosols were generated including polystyrene latex spheres (PSLs; Nanosphere size standards, Thermo Fisher Scientific Inc., Waltham, MA), NH4NO3, (NH4)2SO4, (NH4)2C2O4 (Sigma Aldrich, St. Louis, MO), NH4Cl (J. T. Baker Chemical Co., Phillipsburg, NJ), and NaNO3 (Fisher Scientific, Hampton, NH). Aerosol particles were generated by atomizing aqueous solutions of pure compounds in distilled water (∼ 0.5–6 g L-1) using a custom-built Collison-type atomizer (similar to one reported by Liu and Lee, 1975) in a dry particle-free nitrogen or ZA flow. The output flow was dried using a silica gel diffusion dryer to a relative humidity (RH) less than 10 %. The dry polydisperse particles were then size-selected using a custom-built differential mobility analyzer (DMA; similar to one described by Knutson and Whitby, 1975). The DMA was operated at a sample flow of 0.3–0.5 volumetric L min-1 and a ratio of 10 : 1 between the sheath and sample flow. The particles were diluted with ultra-high-purity filtered ZA (range 1–10 L min-1) before entering a mixing vessel. In instances in which a mixing vessel was not available, a segment of smaller diameter tubing was added in-line to promote mixing prior to the flow being divided among the instruments. A condensation particle counter (CPC; 3022A, TSI Inc., Shoreview, MN) (Stolzenburg and McMurry, 1991) continuously measured the particle number concentration of the DMA output flow, following dilution. We measured the CPC flow rate pre- and post-sampling using a low-flow DryCal (Mesa Laboratories, Lakewood, CO) and estimate an uncertainty in the CPC flow rate calibration to be ±1 %. During several experiments, the aerosol flow was split and sampled by an ultra-high-sensitivity aerosol spectrometer (UHSAS; Droplet Measurement Technologies, Longmont, CO) to continuously measure the particle concentration and size distribution for particles with diameters between ∼ 63 and 1000 nm.

For the liquid concentrations, atomizer conditions, and DMA settings used here, the DMA output size distribution was a multi-peaked population consisting not only of singly charged particles but also particles with multiple (mostly two or three) charges that could contribute significantly to the overall particle mass. Hence the particle mass could not be calculated directly from the singly charged mobility diameter, particle density, and the CPC number concentrations. We generally used two methods to calculate the particle mass concentrations for these experiments. For the first method, the size distributions were measured using the scanning mobility particle sizer (SMPS; Wang and Flagan, 1990) function of the DMA (physical diameter, Dp = 1–1000 nm). We used the DMA transfer theory (Knutson and Whitby, 1975; Stolzenburg, 1988) with the steady-state charge distribution approximation of Wiedensohler (1988) to estimate the fraction of multiply charged particles contributing to the CPC number concentration for each diameter setting. There are a number of possible sources of uncertainty using these methods that may include particle losses, DMA transfer function uncertainty, counting uncertainty, and inversion errors. When comparing mass concentrations from the SMPS with those measured by the Nr system, issues with the SMPS-derived size distributions became apparent (discussed separately in Sect. 3.2.2). For the second method of calculating mass concentrations, we directly measured the diluted DMA output using the UHSAS. UHSAS particle sizing is a function of the amount of light scattered onto the photodetectors, which depends not only on the particle size but also on the composition-dependent particle refractive index (Bohren and Huffman, 1983; Liu and Daum, 2000; Hand and Kreidenweis, 2002; Rosenberg et al., 2012). The UHSAS manufacturer recommended calibration uses PSL microspheres, which are well characterized and have a known refractive index (n = 1.58) and shape. Because the UHSAS sizing is sensitive to particle refractive index, a new sizing calibration curve was produced for each studied particle type (i.e., refractive index) using a DMA to select particles for a range of known sizes (Kupc et al., 2018). These calibration curves were used to retrieve accurate particle size distributions that properly accounted for the multiply charged particles.

Differences in particle counting efficiency between the UHSAS and CPC are potentially important. Previous laboratory studies show UHSAS and CPC number concentration comparisons in excellent agreement (Cai et al., 2008; Kupc et al., 2018); however, occasionally only a ∼ 90 % counting efficiency for the UHSAS was observed when compared to the CPC. These differences are attributed to particle coincidence at high concentrations (> 1000 cm-3) and to inefficient particle mixing before reaching the instruments. Corrections for particle coincidence were applied (Kupc et al., 2018) though we expect differences due to particle mixing add an additional 10 % uncertainty to the measurements. The UHSAS- and CPC-measured particle number concentrations were generally within 10 % of each other; however, the CPC values did not require coincidence corrections and had a better signal-to-noise ratio. Due to problems with measuring SMPS size distributions and requiring coincidence corrections for the UHSAS number concentrations, we used the UHSAS size distributions with the total particle number taken from the CPC measurement to calculate particle mass from total volume and density.

The measurement of particle-phase Nr requires decomposition or volatilization of the solid material, followed by catalytic conversion to NO (or NO2). Broadly, there are three types of Nr-containing particles, with a range of thermal stabilities from volatile to refractory. First, there is considerable literature that indicates that small particles composed of two major semi-volatile species, ammonium nitrate (NH4NO3) and ammonium chloride (NH4Cl), will dissociate to constituents NH3 and HNO3 (and HCl) when modestly heated to temperatures < 100 ∘C (Huffman et al., 2009; Hu et al., 2011). These materials will be readily converted on high-temperature catalysts (e.g., platinum, Pt) as gas-phase NH3 and HNO3. The second type of Nr-containing particles include intermediate-stability compounds consisting mostly of nitro-organics (R-NO2), organic nitrates (RONO2), and amine and ammonium salts of acids. These compounds begin to decompose at relatively low temperatures. For example, thermal decomposition studies of bulk ammonium oxalate ((NH4)2C2O4) indicate that it begins to decompose at temperatures slightly above 200 ∘C (Usherenko et al., 1988). Similarly, bulk samples of ammonium sulfate ((NH4)2SO4) and ammonium bisulfate ((NH4)HSO4) decompose at approximately 150–250 ∘C depending on water content (Kiyoura and Urano, 1970). Given sufficient residence time, intermediate-volatility compounds will start to convert to gas-phase products in the hot inlet tubing and fully convert to NO (or NO2) on a hot Pt surface (750 ∘C). The third type of Nr-containing particles are composed of refractory salts such as sodium nitrate (NaNO3), which will be the most resistant to decomposition and require contact with high-temperature surfaces of the Pt catalyst. Studies of the thermal decomposition of NaNO3 on Pt surfaces indicate that NO is evolved starting at about 500 ∘C. In summary, the existing literature suggests that the thermal decomposition and conversion of Nr-containing particles to NO (NO2) is thermodynamically feasible provided there is sufficient residence time and surface area in the catalyst zone.

The atomizer output was diluted with particle-free nitrogen and ultrapure ZA; therefore, the Nr measurement should theoretically be attributed to particles only since no detectable gas-phase nitrogen is added to the sample stream. However, equilibration within the sample lines may result in outgassing and formation of gas-phase compounds affecting total Nr detection. Figure 3a shows the initial response of the Nr system in cleaned inlets for NaNO3. The Nr mass signal tracks the CPC-derived aerosol mass features closely as the aerosol source concentrations fluctuate. Additionally, as different particle sizes are selected by the DMA for (NH4)2SO4 (Fig. 3b), changes in the total Nr response are fast and precisely track the changes in the CPC signal. The potential gas-phase constituents equilibrating in the lines from aerosols in this study include HNO3, HCl, and NH3. If these compounds formed before reaching the Nr catalyst it is likely adsorption and desorption from inlets and tubing surfaces would occur (e.g., Neuman et al., 1999; Yokelson et al., 2003). As an example, the presence of NH3 in Fig. 3b (or HNO3 in nitrate-containing particles) would be indicated by a delayed and lengthened rise or fall in the Nr response with sudden changes to the input concentrations. However, the total Nr response precisely tracks the CPC signal on rapid timescales (a few seconds), suggesting that gas-phase NH3 was not present in significant quantities. In experiments at exceptionally high aerosol loadings of (NH4)2C2O4 (up to several parts per million volume of total Nr, i.e., several thousand micrograms per cubic meter) Nr signal “tailing” was observed, suggesting that NH3 was scavenging to the walls of the inlet before the heated quartz tubing.

Marx et al. (2012) reported calculated conversion efficiencies in air sampled from a small chamber for NaNO3, NH4NO3, and (NH4)2SO4 to be 78, 142, and 91 %, respectively. The authors suggested the overestimation of NH4NO3 was a result of its semi-volatile properties under ambient conditions that led to the formation of gaseous NH3 and HNO3 in the chamber. For these reasons, we limit the background artifacts and volatilization effects that may have occurred during chamber filling and sampling in Marx et al. (2012) by sampling immediately following solution atomization through conductive tubing at relatively high sample flow rates. Additionally, we use a DMA to size-select the atomized polydisperse aerosol to evaluate the particle conversion efficiency at several different diameters (100–600 nm in 50 nm increments) to investigate the volatilization effects and conversion efficiencies of smaller particles for the extended list of Nr-containing aerosols studied in our work.

The voltage scanning (SMPS) function of the DMA and number concentration measurements by the CPC are a conventional method to determine particle size distributions, and for calculating particle mass from total volume and density, assuming spherical particles. For the total nitrogen measurements, the total particle-bound Nr mixing ratios were retrieved and converted to mass concentrations for each corresponding salt. Figure 4a–d show the SMPS-calculated vs. Nr-measured mass concentrations (µg m-3) for particles of different composition and diameter. The plots show that a strong correlation (R2 > 0.98) and good agreement was obtained for smaller particles (50–200 nm) with slopes ranging from 0.86 to 0.97, while for larger particles (≥ 250 nm) the mass-calculated values from the SMPS-derived distributions were sometimes as much as > 50 % too high. The R2 for all particles including ≥ 250 nm ranged from 0.71 to 0.85 with slopes of 1.08–1.36.

For larger particles, we used a UHSAS to determine the size distribution of multiply charged species exiting the DMA. The SMPS inversion-derived size distributions were generally broader than the UHSAS size distributions, though agreement improved at increased scan times. Small differences in the size distribution recovered from the voltage scans at larger diameters (> 200 nm) affected the mass distribution considerably because particle mass scales with diameter cubed. A possible explanation is that we are not correctly accounting for the delay time from the DMA exit to the CPC; therefore the particle counts did not correspond to the correct size designated from voltage scanning and this likely skewed the size distribution relative to the true distribution (Collins et al., 2002). Methods for limiting these effects exist (Russell et al., 1995; Collins et al., 2002), including slower voltage scan rates. However, our results demonstrate the added challenges in particle mass determination using estimated size distributions from the SMPS method. For the remaining discussion, we measure the size distributions directly using the UHSAS with particle concentration measurements (by either the CPC or UHSAS) to evaluate the Nr particle conversion in the Nr system.

For the aerosol mass concentrations (µg m-3) calculated using UHSAS particle size distributions, we refer to these values as UHSAS-calculated mass. Comparisons of the mass directly measured as Nr versus UHSAS-calculated mass concentrations for atomized solutions of NaNO3, (NH4)2SO4, NH4Cl, and (NH4)2C2O4 are shown in Fig. 5 with orthogonal distance regression lines with slopes that range from 0.910 to 1.06 for concentrations from ∼ 0 to 70 µg m-3. The instruments are highly correlated (R2 = 0.90–0.99) and the fits indicate that for the salts tested there is quantitative conversion of particulate nitrogen, to within the combined uncertainties of the methods, independent of diameter (range: 100–600 nm). More detailed particle conversion efficiencies by size are shown in Table 2 for each aerosol tested. On average across all size ranges the results indicate 97 ± 7, 101 ± 5, 100 ± 10, and 93 ± 5 % particle conversion efficiencies for NaNO3, (NH4)2SO4, NH4Cl, and (NH4)2C2O4, respectively. The largest deviation from the one-to-one line occurred for (NH4)2C2O4, which may imply some ammonia loss, though the agreement is generally still within 10 % for most particle sizes.

For the case of NH4NO3, the UHSAS-measured size distribution peaked at significantly lower diameters than expected based on the DMA size selection. This difference has been reported previously (Cai et al., 2008; Womack et al., 2017), though to a lesser extent (∼ 8 %) than observed here (up to 30 %). Possible explanations for these differences could include vaporization and/or evaporation effects, residual water in the particles, surface effects, or differences in electrical mobility diameter and geometric diameter due to non-sphericity as discussed in DeCarlo et al. (2004). For these reasons, we made no attempt to characterize NH4NO3 behavior in either the DMA or UHSAS and refer to Sect. 4 for mass concentration comparisons of polydisperse aerosol measured using separate mass measurement techniques (both the Nr system and PILS–ESI/MS). It is worth noting that NH4NO3 is one of the more volatile compounds included in this study and it is reasonable to expect similar particle conversion efficiencies in the Nr system catalysts for NH4NO3 as the other species tested (Table 2).