Nitrogen oxides, including NO x (NO x ≡ NO + NO 2 ) and higher oxides (RO 2 NO 2 , RONO 2 , HNO 3 , N 2 O 5 , ClNO 2 , and other molecules) both influence and are affected by atmospheric oxidation rates. NO is directly emitted by anthropogenic and natural sources and is generally quickly converted to NO 2 . The concentrations of NO and NO 2 have direct effects on the production of O 3 , a pollutant in the troposphere, and on ambient concentrations of OH and peroxy radicals (e.g., RO 2 ), molecules which in turn affect the lifetime of NO x (e.g., Kenagy et al., 2018;Romer et al., 2016). The higher oxides of nitrogen can contribute to aerosol through the formation of condensable organics, inorganic nitrate (e.g., Rollins et al., 2013), and inorganic nitrite in some cases (Guo et al., 2016); serve as terminal sinks or reservoirs of NO x that are transported downwind of initial emission sources; and are diagnostics of our understanding of emissions and chemistry.
As a consequence of their central role in atmospheric chemistry, many independent approaches to observing NO x , the higher oxides (collectively known as NO z ), total reactive nitrogen (NO y = NO x + NO z ), and the aerosol-phase oxidized nitrogen have been developed. Examples include chemiluminescent (CL) detection of NO (Ridley & Grahek, 1990); laser-induced fluorescence (LIF) detection of NO 2 (George & O'Brien, 1991;Thornton et al., 2000); cavity ring-down spectroscopy (CRDS) applied to the direct detection of NO 2 and NO 3 (Brown, 2003;Wagner et al., 2011); luminol detection of NO 2 (Drummond et al., 1991;Wendel et al., 1983); and chemical ionization mass spectrometry (CIMS) applied to detection of HNO 3 (Neuman et al., 2002), N 2 O 5 , and ClNO 2 (e.g., Kercher et al., 2009), as well as organic nitrates and organic peroxy nitrates (Beaver et al., 2012;Lee et al., 2014), ion chromatography for detection of nitrate and nitrite (Dibb et al., 1996;Orsini et al., 2003), particulate organic nitrate by thermal dissociation-laser-induced-fluorescence (TD-LIF; Rollins et al., 2010) and aerosol mass spectrometry (AMS; e.g., Fry et al., 2013;Kiendler-Scharr et al., 2016), and inorganic nitrate by AMS (e.g., Canagaratna et al., 2007).
In addition to the primary detection of species indicated above, many of these techniques have been coupled to strategies for converting all or selected fractions of NO y to lower oxides (e.g., NO 3 , NO 2 , or NO) that can be readily measured. Examples include photolytic conversion of NO 2 to NO (Walega et al., 1991), catalytic conversion of NO y to NO (Walega et al., 1991), and thermal conversion of classes of NO z to NO 3 , NO 2 , or NO (Day et al., 2002;Paul et al., 2009;Thieser et al., 2016;Wagner et al., 2011;Wild et al., 2014). Some reactive nitrogen species present measurement challenges as the molecules of interest can react on, adsorb, or desorb from the walls of sampling lines or partially thermally dissociate as air is sampled and analyzed. Aerosol sampling efficiency through inlets is difficult to quantify and may be quite different for each inlet/species pair. In particular, aerosol transmission is lower for Teflon inlets designed for gas-phase sampling, due to the loss of charged particles (Deming et al., 2019). Thus, sampling of aerosol NO y species can affect interpretation of observations and comparison of different instruments. The detected aerosol fraction depends on the ambient size distribution and humidity, as well as the specific aircraft and inlet location (even for the same instrument and physical inlet).
A number of previous papers have compared methods for detection of nitrogen oxides. While the studies listed below are not a comprehensive review of reactive nitrogen intercomparisons, they do represent those that have focused on similar sets of instruments (CIMS, LIF, CL, and CRDS) or those that have taken place in the winter season. For NO 2 , comparisons relevant to this work include Thornton et al. (2003) in which LIF and CL detection of NO 2 were found to agree within 5% during the ground-based 1999 Southern Oxidant Study, where NO 2 ranged from 0.75 to 60 ppbv, and Fuchs et al. (2010) who found agreement for NO 2 (with NO 2 up to 75 ppbv) within 3% during the SAPHIR NO3COMP chamber comparison, which included CL, LIF, and three cavity-enhanced absorption techniques. A CRDS comparison with CL sampling ambient air (having NO 2 up to 60 ppbv) for 6 days in 2009 found NO 2 and NO agreed within 1% and 3%, respectively (Fuchs et al., 2009). In a recent paper, Javed et al. (2019) compared LIF, CL, CRDS, LP-DOAS (long-path differential optical absorption spectroscopy), and CE (cavityenhanced)-DOAS NO 2 instruments at a forested site in Germany. They reported agreement within the experimental limitations and instrumental uncertainties over the ambient concentration range of 0.13 to 22 ppbv.
For higher oxides, comparisons could be made between individual species (e.g., PAN or HNO 3 ), or grouped species. An example is the laboratory comparison of n-propyl nitrate (0 to 20 ppbv) measurements by LIF with thermal dissociation (TD-LIF) and CL using a molybdenum oxide catalytic converter, which found agreement within 1% (Day et al., 2002). During a field study at Blodgett forest (where NO y ranged from 0.5 to 5 ppbv), the same instruments agreed within 7% for NO y measurements (Day et al., 2002). Also at Blodgett forest, during the BEARPEX (Biosphere Effects on Aerosol and Photochemistry Experiment) 2009 study, two thirds of the TD-LIF summed organic nitrate measurement (0.05 to 1 ppbv) were accounted for by the CF 3 O -CIMS measurements of individually identified biogenic-derived nitrates, with the remainder one third not identified as individual species but consistent by their mass to contain nitrogen (Beaver et al., 2012). Wooldridge et al. (2010) found TD-LIF sum of PANs and summed measurements of individual PAN species agreed to within 10% for a number of ground-based and airborne campaigns where PAN compounds were generally 0.2 to 2 ppbv, with extreme values up to 8 ppbv. CRDS with thermal dissociation (TD) agreed within 1% with a CL instrument for NO y (up to 6 ppbv) during the 2013 Southern Oxidant and Aerosol Study and within 14% of a sum of NO y components at the 2013 Uintah Basin Winter Ozone Study (Wild et al., 2014) where NO y was generally less than 30 ppbv but did reach 60 ppbv extremes. Some of the latter difference is likely due to the lack of organic nitrate measurements to include in the Uintah Basin Winter Ozone Study sum (Lee et al., 2014).
Although these comparisons show that many different approaches to nitrogen oxide detection can deliver measurements that are accurate and cross-calibrated to within a few percent, there remain concerns that different approaches may not be fully consistent with each other and that newer approaches (e.g., I -CIMS and CRDS) have not been extensively compared to better established methods, particularly during airborne campaigns. The Wintertime INvestigation of Transport, Emissions, and Reactivity (WINTER) field experiment over the northeast United States during February and March 2015 provided an opportunity to compare seven different instruments, which measured one or more constituents of NO y from aboard the National Center for Atmospheric Research (NCAR)-National Science Foundation (NSF) C-130 aircraft. The lower temperatures of the winter season and the extensive nighttime flying provided a wider range of conditions than prior experiments, particularly those leading to the formation N 2 O 5 and ClNO 2 . In this paper we present instrument intercomparisons of NO, NO 2 , NO y , NO z , N 2 O 5 , ClNO 2 , and HNO 3 . Two instruments included O 3 measurements; however, the focus here is confined to nitrogen oxide species. In most cases the agreement is better than the stated uncertainties, an indication that the uncertainties are estimated conservatively.
The WINTER campaign (www.eol.ucar.edu/field_projects/winter) took place in February and March 2015. Based out of the NASA Langley Research Center in Virginia, 13 research flights with NCAR-NSF C-130 aircraft were conducted over the eastern United States with roughly equal attention given to daytime and nighttime. The instruments used in this study are each described briefly below and summarized in Table 1. After the WINTER campaign, a laboratory experiment in July 2015 was conducted to explore the TD-LIF and I -CIMS measurements of ClNO 2 . In these experiments, ClNO 2 was produced by reaction of gas-phase N 2 O 5 over a NaCl salt bed (Kercher et al., 2009).
NO 2 and other oxidized nitrogen classes were measured by a LIF instrument. (The LIF instrument is often referred to as TD-LIF whether or not the TD step described below is involved.) For WINTER, ambient air was sampled at 4.1 L/min through a 28-cm-long, 0.4-cm i.d., PFA Teflon tube warmed to 45 °C to maximize transmission of HNO 3 and RONO 2 . The tip was 24 cm from the aircraft outer surface at 90°to the flow. A 532-nm Nd:YAG laser with 20-ns pulses at 15 kHz excites NO 2 , and fluorescence >700 nm is imaged onto photon counting modules. Time gating is used to discriminate against laser light scattered from inside the cell, as well as from air molecules and aerosol particles (George & O'Brien, 1991;Thornton et al., 2000). An NO 2 standard (4.39 ppmv, 5% uncertainty, Praxair Certified Standard grade) diluted in zero air to deliver mixing ratios within 0-20 ppbv was used to calibrate the instrument by overflowing the inlet at selected times during the flights. A number of NO 2 standard cylinders are kept in the laboratory and intercompared regularly. Each calibration begins with quickly flowing (and diverting to an exhaust line) enough gas through the regulator to purge it of gas that has been in contact with the regulator internal surfaces for more than a few minutes. In addition, calibrations were done frequently enough to keep the mass flow controller and tubing passivated at the cylinder concentration. A correction for fluorescence quenching by water (which quenches five times faster than nitrogen or oxygen) was applied to the data to account for sampling air with ambient humidity but calibrating with dry mixtures. In addition, an empirical correction for nonlinearity arising from multiple photons arriving in the same time gate being counted as one was fitted to laboratory-generated NO 2 concentrations up to 500 ppbv.
The compound classes ∑PNs (total peroxynitrates, RO 2 NO 2 , which are typically dominated by peroxy-acyl nitrates, RC(O)O 2 NO 2 ), ∑ANs (total alkyl and multifunctional nitrates, RONO 2 ), and ∑HNO 3 , were measured by TD and LIF detection of the NO 2 product (Day et al., 2002). These utilize separate fused quartz
Table 1 Instrument Details Instrument Species Uncertainty Detection limit TD-LIF a NO 2 5% 20 pptv RO 2 NO 2 + N 2 O 5 10% 30 pptv RONO 2 + ClNO 2 25% 49 pptv HNO 3 + NO 3 25% 65 pptv NO y minus NO 10% 48 pptv CL b NO 30 pptv ± 10% NO y 100 pptv ± 50% CRDS c NO 4% 35-140 pptv NO 2 3% 25-114 pptv NO y 12% 50-380 pptv N 2 O 5 12% 1.3-4.4 pptv I -CIMS d N 2 O 5 30% 0.4 pptv ClNO 2 30% 0.6 pptv HNO 3 30% 7.0 pptv AMS e Particle NO 3 ¯35% 11 pptv Note. Values taken from version R3 1-s merged data files reported for Wintertime INvestigation of Transport, Emissions, and Reactivity (available at data.eol.ucar.edu), except for the more recently updated aerosol mass spectrometry version R2 data. Detection limits are for 1σ, 1 s; except for cavity ring-down spectroscopy, which is 2σ, 1 s; and aerosol mass spectrometry, which is 1σ, 1 min.
tubes heated to T PNs = 190 °C, T ANs = 350 °C, and T HNO3 = 540 °C. These temperature set points were determined from previous studies on the range of dissociation temperatures in combination with laboratory studies using the specific setup for WINTER (Wooldridge et al., 2010). At T PNs , all peroxy nitrates will be dissociated (Wooldridge et al., 2010). Also at T PNs , N 2 O 5 has been shown to be dissociated to NO 2 and NO 3 , resulting in one molecule of NO 2 detected per molecule of N 2 O 5 (Fuchs et al., 2012;Wood et al., 2003). At T ANs , in addition to ΣPNs, alkyl and multifunctional nitrates including particle-phase organic nitrate, as well as ClNO 2 , also dissociate (Wooldridge et al., 2010). At T HNO3 , all nitrogen oxides, including HNO 3 , particle-phase volatile inorganic nitrate (Day et al., 2002;Wooldridge et al., 2010), and NO 3 (Womack et al., 2017), have dissociated. The largest source of NO 3 is N 2 O 5 dissociation. Mixing ratios for each class of compounds are determined by subtracting the mixing ratio of the next lowest temperature channel. That is, NO 2 = NO 2 (T ambient ), ∑PNs = NO 2 (T PNs )-NO 2 (T ambient ), ∑ANs = NO 2 (T ANs )-NO 2 (T PNs ), ∑HNO 3 = NO 2 (T HNO3 )-NO 2 (T ANs ). The NO 2 (T HNO3 ) individual channel is a measurement of NO y , excluding NO and HONO (as HONO dissociates into NO, which is not detected here), and is labeled as "NO y minus NO" in the data set. All heated channels also measure nonrefractory particle-phase nitrogen oxides that are transmitted by the inlet, with particle-phase organic nitrate observed in the ∑ANs channel and inorganic nitrate in the ∑HNO 3 channel. Particle-bound peroxynitrates would appear with ∑PNs, but observations have shown no evidence for their presence.
The uncertainty for NO 2 is ~5%, which includes both the uncertainty in the NO 2 standard and uncertainty in the dilution and delivery. In addition, as detailed in Day et al. (2002), uncertainties in the higher oxides include the uncertainty of subtraction of the signal from lower temperature channels, as well as any sampling, heated channel chemical interferences, and inlet transmission losses. The potential humidity dependence of the inlet losses for ∑ANs and ∑HNO 3 are not separately quantified but included in the stated uncertainties. We estimate these are 10%, 25%, and 25% for ∑PNs, ∑ANs, and ∑HNO 3 , respectively, and 10% for the NO y minus NO measurement. More precise uncertainty estimates take into account the concentrations of the related species (Day et al., 2002). For example, 1 ppbv of HNO 3 is more uncertain if it is observed along with 50 ppbv NO 2 than with of 0.5 ppbv NO 2 . An additional uncertainty term arises if the dissociation of a species group does not occur completely within its temperature bounds for a particular heated tube. For example, it has been reported that a small fraction of ∑HNO 3 may dissociate below T ANs and thus appear as ∑ANs instead of ∑HNO 3 (Pusede et al., 2016 reported 2.5%;Womack et al., 2017, reported 2% to 6%;Thieser et al., 2016, reported 10%;Sobanski et al., 2016, reported <0.5%). Data are available for all WINTER research flights for NO 2 and ∑PNs, and all except research flights (RFs) 5-6 for ∑ANs, and RFs 2-3, 5-7 for ∑HNO 3 , and RF03 for NO y minus NO.
NO and NO y were measured using a CL instrument (Ridley & Grahek, 1990). The CL reaction of NO with excess O 3 in a reaction vessel designed to operate at 8-10 Torr with 1,000 sccm flow produces photons that are counted using a dry-ice cooled photomultiplier tube. An NO standard (5% uncertainty) diluted in zero air (Walega et al., 1991) is used to calibrate the instrument in flight. NO y is detected in a separate channel by catalytic conversion to NO. The gold catalyst is operated at 300 °C, and CO is added to the flow. The catalyst was cleansed preflight by heating the gold surface to 500-550 °C for 30-50 min (Bollinger et al., 1983;Fahey et al., 1986;Walega et al., 1991). A warmed (30 °C) Teflon inlet with a flow rate of 1 slpm is rear-facing to reduce particle intake. Uncertainties for NO and NO y are 30 pptv ± 10% and 100 pptv ± 50%, respectively. NO and NO y data are available for all flights except RFs 7, 10, and 11.
NO, NO 2 , N 2 O 5 , and NO y were measured using a cavity ring-down spectrometer (CRDS). CRD spectroscopy exploits the long path within an optical cavity to achieve high sensitivity in direct absorption. Light that coupled to an optical cavity formed from two high reflectivity mirrors (e.g., R > 99.99%) is rapidly extinguished, leading to a single exponential decay of light intensity exiting the cavity (Brown, 2003). Measurement of the characteristic time constant with and without an absorbing trace gas provides a quantitative measure of the absorber concentration. This instrument is referred to here as CRDS, rather than TD-CRDS, unless the TD component is being emphasized. The CRDS deployed during WINTER measured NO, NO 2 , and NO y in four separate channels by direct absorption of NO 2 at 405 nm (Wild et al., 2014). In the 10.1029/2019JD030700
Journal of Geophysical Research: Atmospheres first channel, NO is quantitatively converted to NO 2 by excess O 3 prior to detection (Fuchs et al., 2009). Ambient NO 2 concentrations measured on the second channel are subtracted to derive ambient concentrations of NO. During WINTER, NO and NO 2 were sampled through a 0.198-cm i.d. Teflon line, held at a constant volumetric flow rate of 2.7 lpm. NO y is measured as NO 2 on a separate channel after TD in a nichrome-wrapped, 650 °C heated quartz tube inlet, described by Wild et al. (2014). After thermal reduction, excess O 3 is added to quantitatively convert NO to NO 2 . Nonrefractory particles that enter the heated inlet are vaporized and any oxidized nitrogen molecules present are converted to NO 2 (Wild et al., 2014;Womack et al., 2017). All measurement channels were zeroed every 3-5 min during WINTER by the overflow addition of zero air at the front of each inlet. Both inlets were downward facing at 90°to the airflow. Calibrations of each channel were conducted before and after each research flight and on nonflight days using standard additions of NO 2 generated by conversion of known quantities of O 3 to NO 2 in excess NO, as described by Washenfelder et al. (2011). These calibrations varied by less than 2% over the course of the campaign.
Instrument uncertainty, error, and lower detection limits were compound dependent. CRDS NO and NO 2 data from RFs 1-4 were scaled to the ratio of CRDS/CL NO y measurements due to an inlet and zeroing error that affected those flights. Uncertainties on these flights were 9% and 8% for NO and NO 2 , respectively. For all remaining flights, the uncertainty for NO and NO 2 were 4% and 3%. Limits of detection during WINTER varied by flight and were between 1 Hz) for NO 2 , NO, and NO y , respectively. These limits include the linear sum of the measurement precision and the variation between zeros. These limits are larger than those quoted in Wild et al. (2014) from ground-based data. CRDS measurement uncertainties of these species were determined from the repeatability in the NO 2 calibration procedure (~3%). An additional 1% uncertainty in the NO measurement is associated with uncertainty in sample dilution from the O 3 addition. The NO y uncertainty of 12% has been previously determined by Wild et al. (2014) based on in-field comparisons to other NO y instruments and is consistent with estimated uncertainties in the thermal conversion efficiencies of common NO y components.
Ambient N 2 O 5 concentrations were measured on an additional channel of the CRDS instrument. In the inlet, prior to entering the measurement channel, N 2 O 5 is thermally dissociated at 130 °C and then measured as NO 3 via direct absorption at 662 nm (Wagner et al., 2011). Measurement zeros were performed every ~3-7 min during WINTER by the addition of excess NO to chemically convert NO 3 to NO 2 . The 0.16-cm i.d. Teflon inlet for N 2 O 5 was also downward facing at 90°to the airflow. The CRDS measurement of N 2 O 5 was calibrated off-line as described in Fuchs et al. (2008) and Wagner et al. (2011) by the addition of constant N 2 O 5 , in order to determine the inlet transmission efficiency and thermal conversion efficiency of N 2 O 5 in the measurement channel. During WINTER, the uncertainty was 12% with a flight-dependent limit of detection of between 1.3 and 4.4 pptv (2σ, 1 Hz). The total uncertainty of N 2 O 5 is subject to uncertainties in the temperature dependence of the NO 3 absorption cross section (σ NO3 , used to calculate ambient concentrations from measured ring-down times; ~5%) and variation between individual calibrations (~6.5%).
CRDS NO, NO 2 , NO y , and N 2 O 5 data are available for all WINTER flights except for N 2 O 5 on RF11.
A high-resolution time-of-flight chemical ionization mass spectrometer with iodide-adduct ionization (HRToF-CIMS, referred to as I -CIMS here) measured N 2 O 5 , ClNO 2 , and HNO 3 (Lee et al., 2018;Lee, Lopez-Hilfiker, et al., 2014). Intake is at 22 slpm through a 1.6 cm i.d., 40-cm-long polytetrafluoroethylene Teflon inlet after which compounds are ionized with iodide and then detected according to their mass to charge ratios. The inlet tip had a rear facing 45°cut and was unheated to avoid disturbing the gas-particle equilibrium. Water vapor was added to the ion-molecule reaction region to minimize the effect of ambient humidity variations on sensitivity. A correction factor was applied to account for the small residual water dependence (B. H. Lee et al., 2018). The instrument background was determined by periodic displacement of ambient air with ultrahigh purity nitrogen. An in-flight calibration system produced 15 N 2 O 5 (or Cl 15 NO 2 with wetted NaCl bed conversion) from O 3 and a gas cylinder with 15 NO 2 . The output was sampled off-line by the independently calibrated CRDS and TD-LIF instruments and found to agree to within 1.3%. For further details see Lee et al. (2018). During WINTER, the 1-s 1-σ limits of detection for N 2 O 5 , ClNO 2 , 10.1029/2019JD030700
Journal of Geophysical Research: Atmospheres and HNO 3 were 0.4, 0.6, and 7.0 pptv, respectively, with a calibration uncertainty of 30% for each (Lee et al., 2018).
Nitrate from submicron particles (PM 1 ) was measured by a highly customized Aerodyne high-resolution time-of-flight aerosol mass spectrometer (referred to here as AMS; Canagaratna et al., 2007;DeCarlo et al., 2006;Hu et al., 2017;Knote et al., 2011). Details on the AMS configuration in WINTER can be found in Schroder et al. (2018). Briefly, particles were sampled through an NCAR HIMIL inlet (Stith et al., 2009) at 10 slpm into a pressure-controlled region and then focused with an aerodynamic lens. Nonrefractory particle species are vaporized upon impact with a 600 °C porous tungsten vaporizer under high vacuum, and vapors are ionized by electron ionization. Ions (for particulate nitrate: NO + , NO 2 + , and HNO 3 + ) are then detected by time-of-flight mass spectrometry (Δm/m~2,500 at m/z 46). Instrument response was calibrated after nearly every flight with monodisperse ammonium nitrate particles, with ambient collection efficiency estimated according to Middlebrook et al. (2012). In mixing ratio units, the average uncertainty and detection limit for 1-min data were 35% and 11 pptv, respectively (Schroder et al., 2018).
Additional particulate nitrate measurements were made by a Particle-Into-Liquid Sampler with ion chromatography (PILS-IC) that measures submicron inorganic nitrate (Orsini et al., 2003) and a particle filter with postflight ion chromatography detection that measures inorganic nitrate in aerosol up to 4 μm in diameter (Dibb et al., 1996(Dibb et al., , 2000)). Guo et al. (2016) andSchroder et al. (2018) show agreement within 43% for these three techniques for particulate nitrate, with disagreements largely attributable to differences in size ranges and type (inorganic vs. organic and refractory vs. volatilizable) detected by each instrument. In addition, the PILS reported that some particulate nitrite was present during WINTER, which could also be part of the difference as the AMS does not distinguish between nitrite and nitrate.
Only AMS particle-phase nitrate (data set variable "Nitrate_Inorg") is shown in comparisons below, as comparisons were similar for all three of the nitrate measurements and our emphases here is on the gas phase species.
The time scales for 1-s data were aligned prior to comparison by synchronizing features in the time series. These shifts were typically less than 2 s. Analysis was done with the 1-s data set for NO 2 , NO, and NO y comparisons and a 1-min average for comparisons involving particulate nitrate. Data are from the merged data sets (Revision 3, available at data.eol.ucar.edu) except for the AMS 1-min time scale data set that was updated after assembly of the merge (also available at data.eol.ucar.edu). All best fit lines are bivariate fits assuming unit weights for each variable.
The comparison between CRDS and LIF measurements of NO 2 is shown in Figure 1. RFs 1-4 are plotted with a separate color (red) as the CRDS data for these flights were scaled to the CRDS/CL NO y ratio, as discussed earlier. RF13 is also removed due to a period of anomalous NO x data from one of the instruments (with NO<0 and NO 2 >NO y ). At the highest concentrations (>30 ppbv), there is a slight deviation from linearity with the CRDS slightly higher than the LIF. The degree of nonlinearity varied flight by flight and is more pronounced for flights affected by the CRDS zeroing issue (i.e., RFs 1-4).
Correlations of individual flights are generally better than for the overall comparison. The comparison for RF05 is shown in Figure 2 as an example for a flight with overall excellent agreement (i.e., slope = 1.00).
From a time series of RF05 in Panels c and d, however, it can be seen that some, but not all, altitude changes can momentarily perturb the agreement. For each of RFs 5-12, the slopes ranged from 0.97 to 1.04. This confirms that flight-to-flight variations in the correlation slope are larger than variations within each flight. On RFs 1-4 the CRDS inlet issue caused higher uncertainty in that measurement and contributed to differences within a flight of as much as 9% between the two measurements. Panels (e) and (f) in Figure 2 show an expanded view of the two instrument's response to a narrow plume. There is no clear difference in time response for the 1-Hz data reported here. Absolute differences may be attributable to the physical separation of the two inlets (LIF sampling forward and from the right side and CRDS sampling from further aft and the 10.1029/2019JD030700
Journal of Geophysical Research: Atmospheres bottom of the aircraft). Overall, the comparison between these two NO 2 instruments suggests that both perform within stated uncertainties but that the LIF NO 2 had better precision during WINTER. The LIF was also calibrated in flight, whereas the CRDS was calibrated between flights.
The comparison of CRDS and CL for NO, excluding RFs 1-4 and RF13, has slope of 1.16, as shown in Figure 3. Excluding points less than the stated detection limits would yield a slope of 1.15. The fitted intercept was near zero, though the presence of a small offset is apparent in the log-log plot. The slope difference is Table 1). 10.1029/2019JD030700 Journal of Geophysical Research: Atmospheres larger than the combined instrument uncertainties of 11%. The same two instruments agree better for their measurements of NO y (discussed below). The NO comparison is linear at the highest concentrations observed. Figure 4 shows a comparison of the measured NO 2 /NO ratio from both the CL and CRDS NO instruments to that predicted from a photo-stationary state calculation for a daytime flight (Research Flight 2 over the Ohio River Valley). The measured NO 2 /NO ratio uses CRDS NO 2 , although the use of LIF NO 2 gives the same result. Calculated NO 2 /NO = k[O 3 ]/j NO2 , where k is the temperature dependent rate constant for the reaction of NO with O 3 (Sander et al., 2011) and j NO2 is the measured photolysis rate constant for NO 2 from a spectroradiometer. Inclusion of the GEOS-Chem model HO 2 from the data merge and assuming [RO 2 ] = [HO 2 ] increases the slopes in Figure 4 by at most 1% each. The figure indicates that the CL NO is more consistent with the predicted partitioning of NO x . Results from other daytime flights are similar. The CL NO measurement, which is not based on a difference, has better precision, and agrees with a photostationary state calculation, is the preferred NO measurement for WINTER.
The NO comparison for RF12 shown in Figure 5 as an example flight that has less scatter than the overall comparison. The slopes for each individual flight excluding RFs 1-4 and RF13 were 1.14 to 1.16 with R 2 values >0.95 with the exception of RF05 (a nighttime-only flight with a single, sharp, NO plume), which had a slope of 1.00 but an R 2 of only 0.16.
There were three NO y measurements aboard the aircraft during the WINTER campaign: the CRDS NO y , the CL NO y measurement, and the TD-LIF measurement of NO y minus NO. The CRDS versus CL observations for all flight data has a slope of 1.01 ± 0.00 (Figure 6).
The example NO y comparison from RF06 shown in Figure 7 demonstrates a correlation with little scatter and a slope close to the overall 1:1 correlation. The CL measurement is approximately 2% higher than the CRDS for much of this flight, but this difference is not consistent from flight to flight. The population of points at NO y near 0.2 ppbv in the CRDS but spanning 0.01-0.1 ppbv in the CL instrument is a period early in the flight (approximately 22:00 to 22:30) when the two instruments show an offset of approximately 0.2 ppbv but are otherwise correlated.
The slopes on each flight vary from 0.93-1.12, all with correlation coefficients of at least 0.98 except for 0.90 for RF13. While the measurements agree within their combined uncertainties for all flights, the interflight variability is larger than for the CRDS versus LIF NO 2 comparison and smaller than the flight-to-flight variance in the CRDS versus CL NO comparison. It is not clear on the basis of this comparison which instrument exhibits greater accuracy, nor the reason for the flight to flight variability. However, the precision on the CL NO y was generally superior to that of the CRDS (see Table 1).
Since the TD-LIF does not measure NO, the signal from its highest temperature channel, [NO 2 (T HNO3 )], corresponds to NO y without NO. The TD-LIF "NO y minus NO" is compared to CRDS and CL measurements in Figure 8. For CL and CRDS, each respective instrument's NO measurement is subtracted from its respective NO y . The CRDS versus TD-LIF comparison has a slope of 0.94 for the full range of data and 0.91 using just data below 30 ppbv (Figure 8a). Note that restricting the fit range to less than 30 ppbv to avoid the effect of nonlinearity in NO at high concentrations only omits 0.2% of the points. Figure 9 shows one of the periods when the highest NO y values were sampled (a close transect of a power plant plume) and illustrates the disagreement also appears in NO z (using CL NO z = CL NO y -CL NO -LIF NO 2 ; using NO 2 from CRDS gives essentially identical results). The lower panel of Figure 9 illustrates that O x (=O 3 + NO 2 ) is well conserved in the fresh plume as the O 3 lost to the reaction with NO is regained as NO 2 . Section 3.5 below provides a more explicit comparison of the total and speciated NO z measurements derived from multiple instruments.
Figure 10 provides a summary of the comparisons between the TD-LIF, CRDS, and CL measurements of NO 2 , NO, and NO y on a flight-by-flight basis. Mean comparisons of NO 2 between the LIF and CRDS instruments show agreement within stated uncertainties, both for the average of all comparisons and for the comparisons between individual flights. Both of these instruments appear to adequately represent NO 2 on the scale over which it was observed during WINTER. By contrast, mean comparisons between NO from the CRDS and CL instruments lie outside of the range of combined uncertainties, and this disagreement is
10.1029/2019JD030700
NO y can be dominated by NO 2 and NO, such that an NO y comparison is largely a comparison of those species. This can make it difficult to evaluate measurements of more highly oxidized nitrogen compounds. Here, we compare NO z = NO y -NO 2 -NO to determine the extent of nitrogen budget closure for higher nitrogen oxides during the WINTER campaign. Nitrogen budgets for two example flights are shown in Figures 11 and 12, which illustrate time series of the principal individually measured components of NO z during WINTER: An example of a flight with NO z dominated (70±30%) by HNO 3 is RF06, which took place on February 22-23 off the eastern seaboard throughout the night. Overall, the sum of components tracks well with NO z (Figure 11); however, the TD-LIF NO z is noticeably higher than the CRDS and CL NO z measurements. On average, the sum of NO z components comprises 105% of CRDS NO z , 89% of CL NO z , and 77% of TD-LIF NO z . At times, N 2 O 5 is a substantial portion (averaged 4%, but ranged up to 25%), sometimes but not always coinciding with a rise in ClNO 2 . 10.1029/2019JD030700 An example flight where the major fractions of NO z were N 2 O 5 and ClNO 2 is shown in Figure 12. This was RF08 and took place on 1 March off the eastern seaboard straddling sunrise by a few hours on either side. Agreement between the sum of components with NO z is slightly better than for RF06, with the sum comprising 98%, 102%, and 81% of NO z for the CRDS, CL, and TD-LIF instruments, respectively. TD-LIF NO z is still higher than CRDS and CL, which track more closely to each other than in the prior example. In the early 10.1029/2019JD030700 10.1029/2019JD030700
Journal of Geophysical Research: Atmospheres
On the basis of these two comparisons, there is no clear recommendation for NO z closure. The CRDS and CL total NO y instruments, with NO x subtracted, agree with the summed NO z to within the stated uncertainty of this difference, irrespective of the uncertainty in the sum of speciated components (described in more detail below). The TD-LIF NO y compared to summed NO z is outside of the stated uncertainty in Table 1 for the subtraction between the HNO 3 and NO 2 channels. However, the sum is likely to be an underestimate of the total NO z . Therefore, this comparison shows only that the agreement between total and speciated NO z is likely within stated uncertainties for all instruments. 10.1029/2019JD030700 10.1029/2019JD030700 .05 ppbv) were distributed between approximately 0 and 1 ppbv, with a median of 0.22 ppbv, while measurements of ∑PNs when N 2 O 5 was not present had a very similar distribution, also with a median of 0.22 ppbv. While the I -CIMS reports some higher values than TD-LIF measurements (Figure 13c), scaling the I -CIMS measurements by 0.89 to agree with CRDS N 2 O 5 would bring the data into agreement with the assumption provided above that ∑PNs will be greater than or equal to N 2 O 5 .
Although the CRDS and I -CIMS instruments agreed generally within stated uncertainties, the disagreement and flight-to-flight variability between the CRDS and I -CIMS instruments is worse than in previous, published intercomparisons. The reason for the disagreement is not clear from these data but warrants future investigation and intercomparison activities. The TD-LIF data are more consistent with the CRDS than CIMS, in general, but are themselves subject to potential interference from other species, rendering any conclusion from this comparison exercise uncertain. Use of either the CRDS or CIMS N 2 O 5 data from WINTER is recommended, but using the disagreement as a measure of uncertainty, as has been described in the analysis of McDuffie, Fibiger, Dubé, Lopez Hilfiker, et al. (2018).
ClNO 2 is expected to be detected by the TD-LIF in the ∑ANs difference along with RONO 2 . Using the rate recommended by Baulch et al. (1981), Wooldridge et al. (2010) calculated that ClNO 2 would dissociate over a range of 300-425 °C in the TD-LIF ovens, which most closely overlaps with the ∑ANs temperature range of 300-400 °C. Since the dissociation range for HNO 3 was 475-625 °C, with no overlap with the range for ClNO 2 or RONO 2 , there should be a temperature set point at which all ClNO 2 will have dissociated while all HNO 3 remains intact. This would completely separate the ClNO 2 and HNO 3 signals with ClNO 2 detected in the same channel as RONO 2 . The actual temperature ranges for dissociation depend on the particular oven setup in each experiment, including especially the residence time in the heated section, so experiments are done for each instrument configuration to determine temperature set points in between ranges of dissociation. Thaler et al. (2011) used 450 °C for their TD-cavity ring down ClNO 2 instrument, which was the same as the temperature needed for CH 3 ONO 2 . The shape of the dissociation curve did not match that calculated using the dissociation rate, and the authors speculated that reactions on the heated fused silica surface may have an impact. Thieser et al. (2016), who also used a higher gas flow rate than that used by TD-LIF, observed ClNO 2 and RONO 2 s were fully dissociated by 450 °C. The TD-LIF temperature set Journal of Geophysical Research: Atmospheres points for WINTER were chosen based on experiments with n-propyl nitrate and HNO 3 directly prior to the campaign to determine the best temperature set point to avoid species overlap. Following the campaign, these were verified to still be appropriate.
After the WINTER campaign, additional experiments were conducted in July 2015 with the TD-LIF and I - CIMS instruments sampling the ClNO 2 calibration source (Kercher et al., 2009) in the laboratory. Both instruments sampled air coming from the same oven as the temperature was gradually raised from room temperature to 600 °C. The instrument responses to the heater temperature are opposite, since I -CIMS measures ClNO 2 directly and the TD-LIF measures the NO 2 TD product. Figure 14 For the TD-LIF, ClNO 2 appears to dissociate fully by 350 °C, suggesting that all ClNO 2 would be observed in ∑ANs. However, the temperature for full dissociation observed by the I -CIMS is 400 °C. The difference should not be due to gas-phase recombination of NO 2 and Cl because Cl atoms react very rapidly even with background levels of methane, etc. For example, the calculated Cl atom lifetime in the presence of 1.9 ppmv CH4 is <25 ms at 350 °C, which is short compared to the 50-ms residence time in the heated section. An alternate explanation for the difference could be surface reactions in the tubing, as the tubing from the oven to each instrument was of different lengths. This suggests that all the ClNO 2 would be observed in ∑ANs, but the temperature set point is marginally too low and some portion of ClNO 2 may be observed as ∑HNO 3 .
Observations during WINTER, however, show periods of time where the I -CIMS ClNO 2 measurement is higher than ∑ANs, which should include total RONO 2 + ClNO 2 . Figure 15 shows a time series where this is the case from RF08, during which some of the highest ClNO 2 concentrations were observed.
Given that the temperature set point was on the edge of ClNO 2 dissociation, a possible explanation is that some ClNO 2 was not dissociated at that temperature and was detected instead in the hotter ∑HNO 3 channel.
To explore this possibility, in Figure 15 we compare the sum of ∑ANs and ∑HNO 3 with the sum of their major components, ClNO 2 , HNO 3 , and N 2 O 5 , to determine if the gap persists or is closed. However, the ∑ANs -ClNO 2 difference (Panel b) largely persists with the (∑ANs+∑HNO 3 ) -(ClNO 2 +HNO 3 +N 2 O 5 ) difference (Panel d), suggesting the temperature set point was not the principal reason for the disagreement on this day.
The proportion of ClNO 2 signal that is not seen in the ∑ANs measurement was variable during WINTER.
For measurements where ClNO 2 is at least 100 pptv, the median ∑ANs signal as a percentage of total ClNO 2 is 67% but varied widely with a standard deviation of 55%. Due to this ClNO 2 measurement discrepancy, data were not reported for ∑ANs or ∑HNO 3 during WINTER when ClNO 2 was present above 0.05 ppbv. We further note that the ∑ANs may be subject to an interference from the fraction of HNO 3 that dissociates at the set temperature of the ∑ANs channel oven (Womack et al., 2017). As there were no other ∑ANs measurement against which to compare, this paper does not attempt to assess the accuracy relative to other instruments of the ∑ANs measurement from WINTER.
The TD-LIF ∑HNO 3 measurement includes gas-phase HNO 3 , any nonrefractory inorganic particulatephase nitrate that is sampled through the inlet tubing, and NO 3 dissociation (where the NO 3 is primarily from N 2 O 5 →NO 2 +NO 3 ; Womack et al., 2017). To compare ∑HNO 3 to I -CIMS HNO 3 , we select data when N 2 O 5 is less than 50 pptv. Figure 16a shows the comparison of TD-LIF ∑HNO 3 with I -CIMS HNO 3 with an overall slope of 0.76 where the TD-LIF is higher than CIMS. The same comparison but to I -CIMS HNO 3 + pNO 3 -shown in Figure 16b has a slope of 1.02 and a higher R 2 . The improvement in correlation when including the particulate nitrate suggests that the TD-LIF observes most of the submicron nitrate aerosol. The efficiency of sampling of nitrate aerosol by NO y inlets is an important analytical issue for these instruments. We expect a high sampling efficiency for submicron aerosol for the TD-LIF inlet described in section 2.1 but that has not been verified experimentally. For the wintertime DISCOVER-AQ study in the using CRDS N 2 O 5 (black) and I -CIMS N 2 O 5 (red), and (d) the (∑ANs+∑HNO 3 ) -(ClNO 2 +HNO 3 +N 2 O 5 +pNO 3 -) difference with red and black as in (c).
10.1029/2019JD030700
Journal of Geophysical Research: Atmospheres
The comparisons here indicate that instruments used during WINTER for measurements of nitrogen oxides are consistent with each other to within 20% or better. Often, correlations agree better than the combined instrument uncertainty, indicating that the reported uncertainties of individual instruments are conservative estimates. For most species, agreement within an individual flight was better than for the comparison of all campaign data. Measurements at mixing ratios that are close to instrument detection limits were not a focus of this campaign. Based on the analysis above, we make the following summary statements/recommendations. These recommendations are specific to the WINTER campaign and are not intended as general recommendations for other intercomparisons.
1. Comparison of CRDS and LIF NO 2 showed agreement to within stated uncertainties both in the average and the flight to flight comparisons. Both measurements are considered reliable for WINTER. 2. Comparison of CRDS and CL NO was outside of stated uncertainties. A cross check against a photostationary state calculation suggests the error is in the CRDS NO measurement, although we are unable to identify the cause of such an error. Thus, the CL NO measurement, where it is available, is recommended for WINTER. 3. Comparison of CRDS and CL NO y measurements agree within stated uncertainties, such that either measurement may be considered reliable for WINTER. The precision of the CL NO y measurement was in general superior. The TD-LIF instrument measured the sum of all NO y components except for NO.
Comparison to the CRDS NO y -NO and CL NO y -NO showed agreement to well within stated uncertainties below 30 ppbv, but a nonlinear relationship above this level that may be related to the uncertainty in the CRDS NO noted above. 4. Closure of the NO z (NO y -NO x ) budget between the total NO y instruments and the sum of the major speciated components agreed to within 20%. There is no recommendation for a preferred/more accurate instrument. 5. Comparison of CRDS and I -CIMS N 2 O 5 showed agreement to within stated uncertainties but considerable variability from flight to flight that is not explained here. The comparison was not as good as previous, published intercomparisons. While the TD-CRDS has an inherent, stable instrument response in that it is a direct absorption measurement, the I -CIMS was routinely calibrated and did not show variation that could explain the discrepancy. Therefore, we recommend further field or lab inter-comparison of these instruments. 6. In prior TD-LIF deployments, conditions favoring the formation of N 2 O 5 and ClNO 2 were very rare, but such conditions were specifically targeted during WINTER. We found evidence that the ClNO 2 signal was not entirely observed in the expected channel on the TD-LIF instrument. Further characterization of the TD behavior of ClNO 2 is needed. 10.1029/2019JD030700 Journal of Geophysical Research: Atmospheres SPARKS ET AL. 10,499 7. While the efficiency of in-flight particulate nitrate sampling by the TD-LIF was not quantified, including particulate nitrate brings the overall TD-LIF ∑HNO 3 versus I -CIMS HNO 3 comparison into agreement to within 2%.