The IAGOS NOx instrument – design, operation and first results from deployment aboard passenger aircraft

We describe the nitrogen oxide instrument designed for the autonomous operation on board passenger aircraft in the framework of the European Research Infrastructure IAGOS (In-service Aircraft for a Global Observing System). We demonstrate the performance of the instrument using data from two deployment periods aboard an A340-300 aircraft of Deutsche Lufthansa. The well-established chemiluminescence detection method is used to measure nitrogen monoxide (NO) and nitrogen oxides (NOx). NOx is measured using a photolytic converter, and nitrogen dioxide (NO2) is determined from the difference between NOx and NO. This technique allows measuring at high time resolution (4 s) and high precision in the low ppt range (NO: 2σ = 24 pptv; NOx : 2σ = 35 pptv) over different ambient temperature and ambient pressure altitude ranges (from surface pressure down to 190 hPa). The IAGOS NOx instrument is characterized for (1) calibration stability and total uncertainty, (2) humidity and chemical interferences (e.g., ozone; nitrous acid, HONO; peroxyacetyl nitrate, PAN) and (3) inter-instrumental precision. We demonstrate that the IAGOS NOx instrument is a robust, fully automated, and longterm stable instrument suitable for unattended operation on airborne platforms, which provides useful measurements for future air quality studies and emission estimates.


Introduction
Monitoring of NO x (= NO + NO 2 ) in the atmosphere is important for estimating the amount of natural and anthropogenic NO x emissions, for assessing air quality (e.g., for-mation of ozone and secondary aerosols) and concerning the climate impact of ozone. Ozone is a strong greenhouse gas and contributes to global radiative forcing (IPCC, 2007;Fahey and Lee, 2016) and to changes in global dynamics (Fueglistaler et al., 2014). Close to ground ozone has an impact on human health (Skalska et al., 2010) and causes ecosystem damage (Ainsworth et al., 2012); NO 2 by itself poses a public health risk as well. Therefore knowledge of the spatial distribution of NO x is important to identify the sources, sinks and its partitioning between NO and NO 2 in the atmosphere (Monks et al., 2009).
It is known that the global NO x budget contains contributions from natural sources of NO x -like lightning (LNO x ), biomass burning and soil emissions -as well as from anthropogenic sources, such as power generation, road transportation and aviation. Current knowledge of the global distribution of NO x and its emission estimates is based mostly on surface monitoring stations (Aerosols, Clouds and Trace gases Research Infrastructure, ACTRIS; https://www.actris. eu, last access: 29 November 2017), satellite measurements (Fishman et al., 2008;de Laat et al., 2014;Duncan et al., 2016) and model simulations (Ehhalt et al., 1992;Emmons et al., 1997).
The satellite retrievals provide tropospheric NO 2 columns, which are defined as the vertically integrated NO 2 number density between the surface and the tropopause. Satellite data users are provided with averaging kernels, which give the relationship between the true vertical profile and what is actually measured (Eskes and Boersma, 2003). The new experiment TROPOMI (TROPOspheric Monitoring Instrument) on Sentinel-5P provides global coverage with a spatial res-olution of 7 × 7 km 2 . The instrument covers spectral bands at different wavelengths, including bands in the ultraviolet (UV) spectrum up to the shortwave infrared (SWIR) spectrum. These bands are selected to measure the most relevant species in the troposphere and to improve cloud correction retrievals (Veefkind et al., 2012).
In the upper troposphere and lowermost stratosphere (UTLS), emissions from cruising passenger aircraft form another important source of NO x , with its source strength being determined from civil aviation traffic data and specific emission factors (Emmons et al., 1997;Rohrer et al., 1997;Schumann and Huntrieser, 2007;Ziereis et al., 2000;Gressent et al., 2016). Aircraft campaigns conducted in the past have made considerable progress in improving the estimate of the emissions of aviation (Schumann and Huntrieser, 2007;Lee et al., 2010;Wasiuk et al., 2016); in improving the estimate of LNO x emissions over different regions, summarized by Gressent et al. (2016); and in increasing knowledge of deep convectively lifted pollutants and their burden to ozone chemistry (Huntrieser et al., 2016). However, these and other research aircraft campaigns lack the statistical robustness of comprehensive seasonal and geographical coverage of the UTLS region.
Despite the progress made on modeling aviation's impacts on tropospheric chemistry, there remains a significant spread in model results (Lee et al., 2010). Parameterization of natural NO x emissions by lightning still has large uncertainty in global chemical transport models (e.g., Gressent et al., 2016). Brunner et al. (2005) and Prather et al. (2017) concluded that a better description of emissions, chemistry and sinks of NO x (and other key species) is needed to improve chemistry in the UTLS region in global chemistry models.
Using passenger aircraft equipped with instruments for measuring NO x as a measurement platform can help to link satellite and surface measurements, and to fill the UTLS gap where otherwise no regular in situ observations are possible. Global-scale NO x observations in the upper troposphere are particularly important regarding long-range transport of pollutants and its burden to regional air quality . Since 1994, the European Research Infrastructure IAGOS (In-service Aircraft for a Global Observing System, https://www.iagos.org/) has provided in situ observations of essential climate variables (temperature, water vapor, and ozone, and other species later on) on a global scale from the surface up to 13 km altitude . IAGOS builds on the former EU framework projects MOZAIC (Measurement of Ozone and Water Vapour by Airbus In-service Aircraft; Marenco et al., 1998) and CARIBIC (Civil Aircraft for the Regular Investigation of the atmosphere Based on an Instrument Container; Brenninkmeijer et al., 2007). Between 2001 and 2005, total odd nitrogen (NO y = NO and its atmospheric oxidation products, such as nitrogen dioxide, NO 2 ; nitric acid, HNO 3 ; and peroxyacetyl nitrate, PAN) was measured on MOZAIC (Volz- Thomas et al., 2005;Pätz et al., 2006), and since 2005 it has been measured on CARIBIC (Stratmann et al., 2016).
Based on the IAGOS data sets, Thomas et al. (2015) and Stratmann et al. (2016) presented the geographical distribution and seasonal variation of NO y at cruising altitude over the different periods, whereas Gressent et al. (2014) showed that the majority of large-scale plumes of NO y are related to long-range transport and only a minor fraction to LNO x in the UTLS over the North Atlantic region. On the other hand, Brunner et al. (2001) demonstrated from a 1-year climatology of NO x in the UTLS region, from the Swiss NOXAR (measurements of Nitrogen OXides and ozone along Air Routes) program the importance of and need for statistical robustness of comprehensive seasonal and geographical coverage of NO x measurements. However, NO 2 was mostly not trustable from these measurements (contamination, instrument failure) at that time, and therefore NO 2 is based on calculations of the photochemical state. This accounts also for the CARIBIC platform where NO 2 is only available from daytime calculation from the photochemical state (Stratmann et al., 2016).
Given its important role in atmospheric chemistry and the resulting needs for global-scale regular measurements, it was decided to develop a NO x -specific instrument for the operation in the framework of IAGOS, which we describe here. The most common measurement technologies for NO x are based on the chemiluminescence detection (CLD) for the indirect measurement of NO (Clough and Thrush, 1967;Ridley and Howlett, 1974;Drummond et al., 1985;Fahey et al., 1985). CLD instruments have often been coupled to a photolytic or catalytic converter to measure NO 2 and NO x by using a xenon lamp, blue-light converter or catalytic conversion of NO 2 into NO, prior to the CLD unit Ryerson et al., 2000;Nakamura et al., 2003;Pollack et al., 2010;Villena et al., 2012;Reed et al., 2016). NO 2 measurements at low-NO x conditions (below 0.1 ppbv) are close to the limit of detection (Yang et al., 2004), and depending on the installed converter each instrument might show interferences with other nitrogen-oxide-containing species (e.g., Reed et al., 2016).
To minimize these chemically driven interferences, recent instruments have been developed from optical techniques to measure NO 2 by light absorption with cavity ringdown spectroscopy (CRDS; Fuchs et al., 2010;Wagner et al., 2011), cavity-attenuated phase shift (CAPS; Kebabian et al., 2008), laser-induced fluorescence (LIF; Thornton et al., 2000) and differential optical absorption spectroscopy (DOAS; Platt and Stutz, 2008). However, most of these instruments have a detection limit above 0.1 ppbv, or the instrument size and weight is too large to be used for routine aircraft observations (Fuchs et al., 2010;Brent et al., 2015).
In the following, we present the technique, design, calibration and quality assurance (QA) of the IAGOS NO x instrument in Sect. 2, followed by details about the data processing (Sect. 3) and the instrument performance (Sect. 4). First ap-plications of the new instrument aboard an A340-300 aircraft of Deutsche Lufthansa are given in Sect. 5.

IAGOS NO x instrument Package 2b measurement system and calibration
The design of the IAGOS NO x instrument Package 2b (P2b) is based on the former MOZAIC NO y instrument described by Volz- Thomas et al. (2005) and Pätz et al. (2006), using the CLD method for NO with a photolytic converter to convert NO 2 into NO. When using passenger aircraft as platform, many conflicting needs have to be fulfilled: the instrument has to be fully automated, small and lightweight, with limited power consumption, and fulfill high safety standards (mechanical stability, electromagnetic interference and flammability specifications). Furthermore, easy access, simple installation and long deployment periods of up to 6 months have to be guaranteed, while it should measure at NO x mixing ratios as low as 0.1 ppbv and below with the highest possible temporal resolution, accuracy and reliability over the widely varying conditions of external temperature (−70 to +40 • C) and pressure (190 to 1000 hPa) in an unattended mode.
The IAGOS NO x instrument is installed on an IAGOS-CORE mounting rack, which is located in the avionics bay of an A340-300 aircraft (Fig. 1). The mounting rack provides all electrical, pneumatic and safety provisions required for operation. For data transfer the instrument is connected via Ethernet to IAGOS Package 1 (P1), which handles the data transfer for all IAGOS instruments on board . P1 is installed on every IAGOS-CORE aircraft and provides measurements of ozone, carbon monoxide, temperature, water vapor and a number of cloud particles (hydrometeors). It also records relevant parameters like position, static pressure, velocity, etc. from the avionics system of the aircraft . The uncertainty of ozone is given with 2 ppbv ± 2 %, and the uncertainty of water vapor is 5 % over liquid water Neis et al., 2015). Figure 2 shows the schematic flow and position of the major components of the IAGOS NO x instrument. The following sections present a detailed description of the detection method (Sect. 2.1.1); of the reaction cell and the photomultiplier (PMT) as the primary detector hosted in the NO detector (NOD) unit (Sect. 2.1.2); of the ozone generator (O3G); of the photolytic converter (Sect. 2.1.3); and of the inlet manifold (Sect. 2.1.5), residence time characterization (Sect. 2.1.6) and internal stability checks (Sect. 2.1.7) of the inlet, converter and calibration assembly (ICC). A description of the instrument operation is provided in Sect. 2.1.7. The NO detector sensitivity and the converter efficiency are determined in the laboratory (Sect. 2.2). Tables 1 and 2 pro- vide an overview of the instrument specification and the main instrument parameters.

The chemiluminescence detection method
The CLD method is a well-established technique to detect NO by reaction with excess ozone. NO x is measured by converting NO 2 into NO. This converted NO x is often called NO c at this stage.
In measuring mode (MM) the sample air is mixed with ozone in the reaction cell where NO is oxidized by Reactions (R1) or (R2). The photons released in Reaction (R3) are detected by a photomultiplier tube (Hamamatsu R2228P or Electron Tubes Enterprises 9828A, depending on the individual instrument) which is operated in photon-counting mode. In zero mode (ZM), ozone is mixed with the sample air before the pre-chamber (a 30 to 50 cm long 1/8 in. outer diameter stainless-steel tube) in order to oxidize most of the NO before it reaches the reaction cell. The volume and thus the sample residence time of the pre-chamber are adjusted such that 97 to 99 % of the NO is oxidized before the sample air reaches the reaction cell. The photon count rate in zero mode includes the background signal of the photomultiplier (caused by photons originating from the thermal radiation) and additional interferences from other chemical reactions (Drummond et al., 1985). The count rate is quite stable, ex-  Ca. 6 months Time resolution of photon count rate 10 Hz cept during takeoff, due to warming up (or cooling down) of different components in the instrument (e.g., ozone generator, PMT). The mixing ratio (X, X ∈ {NO, NO 2 }) is determined from the difference of the photon count rates measured in measuring mode and zero mode divided by the detector sensitivity (S NOD ) and the conversion efficiency (E PLC ) in the case of NO 2 : (1)

The detector and reaction cell
The chemiluminescence detector mounted in the NOD unit is similar to that described by Volz- Thomas et al. (2005). The PMT is cooled by four Peltier elements to temperatures below −10 • C at an instrument temperature (T Instrument ) of 20 • C. The reaction cell is separated from the PMT housing by a 1 mm thick window and a low-pass red-light filter. This setup provides thermal insulation and limits the light reaching the PMT to wavelengths below 600 nm. The space be-tween the cell window and the low-pass filter, as well as the PMT housing, is purged with a small flow of O 2 or synthetic air (0.2 mL min −1 ) to avoid condensation. The reaction cell is operated at a pressure of approximately 10 mbar. We learned from the MOZAIC NO y instrument that the cell does not require power-consuming temperature control because of the relatively stable temperature in the avionics compartment. The temperature is measured, however, in order to allow for potentially necessary corrections of the sensitivity.

O 3 generator
The ozone is generated in an oxygen flow (approximately 20 sccm) through a ceramic discharge tube with a coaxial inner stainless-steel electrode of 3mm diameter, which is connected to a HV transformer (18 kV, alternating current with a frequency of 250 Hz). The ceramic tubes are inserted in an aluminum housing which is connected to the ground. A silent discharge is generated in the oxygen flow, which produces 1.5 × 10 19 molecules min −1 of O 3 . The pressure in the discharge tube is kept constant between 1 and 1.2 bar and is monitored by a pressure transducer. More details are described by Volz- Thomas et al. (2005).

Photolytic converter
The photolytic converter (PLC) consists of a UV transparent borosilicate glass tube (25 mL), which is mounted behind the manifold. The tube is illuminated by four UV-light-emitting diodes (UV-LEDs, 395 ± 5 nm, 250 mA, 5 VA each, 20 VA total) to convert NO 2 in the sample air into NO by absorption of a UV photon. The UV-LEDs and the associated power transistors of the LED power supply are mounted on individual heat sinks, which are cooled by air entering through the bottom of the housing by means of an external fan. Laboratory tests showed that the air passing the PLC is heated by about 30 • C above the instrument temperature if the UV-LEDs are switched on (Fig. 3). The determination of the converter efficiency and the NO 2 photolysis frequency (J PLC ) of the UV-LEDs are shown in Sect. 2.2. Possible interferences are discussed in Sect. 4.

The inlet line, exhaust line and inlet manifold
The inlet line consists of a 90 cm long PFA tube with an outer diameter (OD) of 1/8 in. It starts in the Rosemount housing outside of the fuselage of the aircraft  and ends at the inlet manifold of the NO x instrument. The residence time within the inlet line is about 0.05 s; thus, losses due to the reaction of NO and O 3 to NO 2 are negligible. About 10 % (150 mL min −1 ) of the total inlet flow is sucked from the manifold into the analytic section of the instrument by means of two membrane pumps (Vacu- line, which starts at the end of the inlet manifold, provided with an exhaust manifold to gather all flows (e.g., internal calibrations) which have passed through the instrument. Outside the instrument the excess flow is guided through the exhaust line (PTFE tube of 60 cm length with 6 mm outer diameter) to the outlet port at the fuselage of the aircraft. The manifold is made of stainless steel and contains ports for pressure measurement and for the addition of zero air and calibration gas. The total residence time from the manifold to the NOD is between 2.5 s at cruising altitude and 12 s at sea level. Thus NO losses by Reaction (R1) with ozone in the ambient air need to be accounted for when the LEDs of the photolytic converter are switched off.

Instrument response characterizing
The response time of the instrument is important for the correction of NO titration by ambient O 3 during sampling and by fast changes of the ambient conditions (e.g., the aircraft crosses the tropopause). The response time of the instrument was characterized in the laboratory by repeating 10 injections of 2 s NO pulses of 7.1 ppbv into the inlet line at each full minute at 250 hPa inlet pressures (Fig. 4). The width (1/e) of the NO peak is 4 s, which represents a peak broadening of a factor of 2, and the delay is about 3 s at an inlet pressure of 250 hPa.

Internal stability checks
Inside the instrument, NO 2 is continuously produced from a permeation tube (PT, KIN-Tek, EL-SRT2-W-67.12-2002/U) placed inside a stainless-steel block, which is purged with a small flow (< 12 mL min −1 ) of oxygen (Revision 1) or synthetic air (Revision 2). The stainless-steel block is temperature-controlled at 40 ± 0.5 • C using a Pt100 sensor and PID controller. The NO 2 flow enters the inlet manifold and is only used for stability checks of the detector sensitivity. During flight, the calibration gas is normally pumped away through the exhaust and will not reach the sample flow. If this pump flow is disabled, the calibration gas will reach the analytic section for a stability check of about 5 min duration (Fig. 5).

Instrument operation
The IAGOS NO x instrument is designed for autonomous deployment over several months. It is synchronized during flight with the main package, P1. The time synchronization has been cross-checked using the ozone measurements from P1, which are also transferred every 4 s to the P2 instrument during operation mode. The software utilizes aircraft signals (currently weight on wheels) to switch between operation mode during flight and standby mode on the ground. The instrument operates in a strict cyclic way by switching the PLC on (NO c mode) or off (NO mode) and by flushing the air into the pre-chamber or directly into the reaction cell. During normal operation in flight the ambient air along the flight track is sampled. In addition to the PMT signal (recorded in 10 Hz), pressures, sample flow and temperatures at different positions are recorded as 1 min averages to monitor the state of the instrument. For in-flight system checks, the manifold is flushed in regular intervals with NO x -free gas or NO 2 calibration gas (approximately 10-15 ppbv, generated from a permeation tube). On the ground, the instrument is in standby mode and does not record data. The ozone generator (O3G) is switched off, and the valves to the pump and between manifold and exhaust are closed, which leads to a backward flow of synthetic air from the gas bottles through O3G, NOD and manifold to the inlet, in order to avoid contamination by polluted air at the airport. The different modes of the instrument are summarized in Table 3, and the cyclic measurements during flight are shown in Fig. 5.

Calibration
The detector sensitivity, the conversion efficiency and the photolysis rate coefficient are determined by external calibrations in the laboratory using procedures defined in the standard operating procedure (SOP) for P2b (see http:// www.iagos.org/iagos-core-instruments/package2b/, last access: 29 November 2017) and described in detail in the following subsections. In principle, the instrument is flushed with a known mixture of NO and synthetic air, and NO 2 produced by gas-phase titration (GPT). The mixing ratio is calculated from the measured flows of the NO calibration gas, oxygen and NO x -free zero air (see Sect. 2.2.3). The titration rate of the external GPT mixture is adjusted to 70-90 %. A simplified example of one calibration is shown in Fig. S2 in the Supplement. Note that the entire calibration procedure is performed at 250 hPa inlet pressure. Table 4 shows the uncertainties of laboratory calibrations for the deployment phases in 2015 and 2016.

NO detector sensitivity
The detector sensitivity (S NOD ) is determined from the photon count rates (CAL NO ) by flushing the instrument with a mixture of known NO mixing ratio (µNO) from the secondary standard (NO Standard ), synthetic air (SL) and oxygen (O 2 ): where Our NO working gas standard (10 ppmv NO mixed in N 2 , 5.0) is a secondary standard and is regularly referenced to the primary standard of the World Calibration Center for NO x at the Forschungszentrum Jülich. Up to now, deviations between both standards have been found to be smaller than 1 %. The uncertainty of the flow measurements is below 2 %. The uncertainty of the detector sensitivity (δS NOD ) from the calibrations is 2 to 3 %, accounting for the errors of the flow meters and the primary NO standard. As an example, for a detector sensitivity of 1000 cps ppt −1 the uncertainty is 30 cps ppt −1 .

NO 2 conversion efficiency and the NO 2 photolysis frequency
The conversion efficiency (E PLC ) is calculated from the measured NO and NO x signals during the calibration by external GPT (CAL GPT ) by switching the UV-LEDs in the PLC on and off (Table 3). Note that the instrument background using NO x -free gas and the signals from the pre-volume (zero mode) must be subtracted from all signals in measuring mode Typically, the conversion efficiency is between 75 and 85 %, depending on the ambient pressure. During a deployment period of 6 months the total uncertainty of the conversion efficiency is determined within 4 %. The photolysis frequency (j PLC ) of the UV-LEDs is calculated as follows: with τ being the residence time in the converter. The photolysis frequency of the UV-LEDs was stable at j PLC = 0.55 (±0.05) s −1 during the last eight pre-and post-calibrations at inlet pressure of 250 hPa. During flight, this value is used to calculate for each measured data point the conversion efficiency considering the residence time and the ambient pressure in the converter.

Zero air (NO x -free air)
In the laboratory zero air is generated using one of the following: a. dried and purified compressed air using a Parker Hannifin adsorption dryer (dewpoint temperature T d < −40 • C) and an additional active charcoal filter for removing NO x , ozone and volatile organic compounds (VOCs); b. pure O 2 (99.5 %) from gas bottles, which is also used for the ozone generator; c. synthetic air (Air Liquide).
The three zero-air types showed no differences in zero mode within measurement errors, which is in agreement with the finding of Volz- Thomas et al. (2005) for the MOZAIC NO y instrument. However, the difference between measuring mode and zero mode of instrument background signal is not equal to zero and has to be subtracted from the ambient measured signal (see Sect. 3).

Quality assurance
Within the IAGOS community it was agreed to flag data quality according to the criteria elaborated in the EU Seventh Framework Programme (FP7) project IGAS (IAGOS for the GMES Atmospheric Service; http://igas-project.org, last access: 29 November 2017; Gerbig et al., 2014). One major topic of this project was to develop QA and quality control (QC) rules, defined in SOPs in collaboration with the IAGOS user community. The flagging criteria are summarized in Table 5. Quality assurance is performed according to the SOP for P2b and is described briefly in the following. Shortly, before and after each deployment period, the entire instrument performance is checked, and necessary replacements or services of compounds are performed, based on the expected lifetime of parts or due to deteriorated performance. The calibration procedure includes determination of the detector sensitivity for NO and the conversion efficiency for NO 2 of the PLC using an external calibration setup with GPT; determination of the instrument background with internal zero air and external zero-air supply;  (NO,NO 2 or NO x -free) can be flushed into the inlet line before being sucked into the reaction chamber Cal_NOc ZM Same as above, flushing the air into the pre-volume  The following steps describe briefly how the mixing ratios of NO, NO 2 and NO x are calculated from the different instrument mode signals (PMT count rates) for each flight: 1. Interpolate a time series of the different zero mode signals (AA_NOc ZM or AA_NO ZM ) separately by using a running mean with a window size of 400 s. This time frame covers at least four NO c and NO mode cycles with the current setup and determines the baseline. The running mean was chosen because it performed best at the beginning and the end of the time series compared to other interpolation methods.
2. Subtract the interpolated zero mode signal from the measuring mode signals (ambient air, zero air etc).
3. Subtract the instrumental background signals (BG_NO MM and BG_NOc MM ) from the ambient measurement signals (AA_NO MM and AA_NOc MM ) to avoid artifact signals (Drummond et al., 1985).

Calculate ambient NO mixing ratio ([NO] AA ) by applying Eqs.
(1) and (2), where S NOD (t) is the timedependent detector sensitivity (determined in the laboratory before installation and after deinstallation). S NOD (t) slightly decreases with time (see Sect. 4).
6. Use nighttime NO measurements to correct possible offsets associated with the zero mode. Nighttime periods are identified using the actual position of the aircraft, time and altitude, by calculating the solar zenith angle. Angles larger 100 • are used to flag the data as nighttime. Daytime measurements are flagged using solar zenith angles < 80 • . In between, the measurements are within the twilight zone, where NO is not fully oxidized by ozone.
7. Flag each data point according to Table 5. 8. The data time resolution is provided at 4 s by calculating the median based on 10 Hz raw data for the individual four second periods to be consistent with the other measured compound time series within IAGOS. The time resolution corresponds therefore to a horizontal resolution of approximately 1 km at cruising altitude. We used the median of the corresponding time interval to avoid a statistical bias uncertainty (Yang et al., 2004).

Water vapor correction
The third-body quenching effect of water vapor molecules on the excited NO 2 molecules in the reaction chamber leads to a reduced signal depending on the amount of ambient water Ridley et al., 1992). The correction factor has to be applied using Eq. (9): with [H 2 O] being the water vapor mixing ratio in parts per thousand. In the laboratory we determined the humidity interference parameter of α = (2.8 ± 0.1) × 10 −3 , independent of whether the PLC was switched on or off, which is 35 % lower than the value of α = 4.3 × 10 −3 determined by Ridley et al. (1992) (Fig. S4). If a water vapor correction could not be applied (e.g., missing water vapor measurements), then the data within the PBL (lowest: 3 km above ground) are flagged as "limited" (Table 5).

Ozone correction
Within the sample line and the converter, Reaction (R1) is still active. Depending on the residence time the reaction will lead to an enhanced NO 2 / NO ratio. The residence time (τ ) in the inlet line is on the order of about 0.05 s, and corrections are negligible here. The residence time of the constant sample mass flow within the PLC is about τ = 2.5 to 12 s as a function of the ambient pressure. The ozone corrections are applied using the in situ ozone measurements from Package 1 and the photolysis frequency J PLC of the UV-LEDs (see Eqs. 5-7) as described in the SOP for NO x from AC-TRIS.
[  Atkinson et al., 2004) and the ozone concentration (ccm −3 ), which is calculated from the in situ measured ozone mixing ratio measured by the IAGOS P1 instrument and the ambient pressure. Figure 6 shows the correction factor for NO (NO corr = [NO] 0 / [NO] AA ) and for NO 2 (NO 2corr = [NO 2 ] 0 / [NO 2 ] AA ). NO increases by up to 25 %, and NO 2 varies in the range ±10 %, both depending on the ambient mixing ratio of ozone, temperature and pressure. Since the ozone correction is sensitive to the ozone mixing ratio, the residence time τ inside the PLC is determined for each instrument for the expected pressure range from 1000 to 180 hPa, which provides the correction function τ (p) to be used in Eqs. (10) and (11) (see Fig. S5). For the future generation of IAGOS NO x instruments, we plan to keep the residence time in the PLC at 3 s, independent from the inlet pressure, by using a critical nozzle.

Signal precision and limit of detection
The precision of the instrument is limited by the dark noise of the PMT caused by counting thermal radiation photons. The counting statistic is Poisson distributed. The background signal is subtracted from the ambient signal (see Sect. 3.1). Therefore, the limit of detection (LOD) is calculated from the 2σ statistical precision of the zero-air measurements in measuring mode (BG O 2 _NO MM ) and zero mode (BG O 2 _NO ZM ), which are integrated over 4 s (t = 4 s) following Feigl (1998): Here the different count rates of the photons are given in counts per seconds (s −1 ), and the unit of the instrument sensitivity is counts per second per pptv (cps pptv −1 ). We derive a detection limit of LOD NO = 24 pptv for NO and LOD NO 2 = 35 pptv for NO 2 for 4 s integration time for a sensitivity of 0.9 cps pptv −1 . By integrating the data over 1 min, the detection limits improve to LOD NO = 6 pptv and LOD NO 2 = 9 pptv.

Total uncertainty
The total uncertainty for each measurement point is calculated by error propagation following from Eq. (1): The uncertainty of the count rate in measuring mode (δMM), zero mode (δZM) and offset (δoffset) is determined from the baseline noise for NO and NO x measurements. Statistical precision (2σ ) of an individual 4 s data point is calculated by error propagation using Eqs. (4) and (5). The uncertainty of the detector sensitivity during calibration is 2 to 3 %, and the uncertainty of the converter efficiency is 4 to 5 %. Figure 7 shows the relative uncertainty (ratio of the total uncertainty to its measured value) as a function of NO and NO 2 in the range of observations during 2015. The relative uncertainty of an individual 4 s data point is dependent on the ambient mixing ratio and reaches NO values of 25 % at 0.2 ppbv and 8 % at 1 ppbv. For NO 2 the relative uncertainty is 50 and 18 %, respectively. Similar uncertainties were calculated for all observations in 2016. The total uncertainty in the low pptv range is mostly dominated by statistical precision of the signal detector.

Instrument performance
The quality of the IAGOS NO and NO 2 measurements depends on the knowledge of the detector sensitivity during the flight phase, the accuracy and precision of the instrument, and possible interferences. These issues are discussed in the following subsections.

Instrument performance drift during deployment
The IAGOS NO x instrument regularly showed a negative drift of the detector sensitivity during each deployment period of two counts per ppbv per day. This sensitivity drift was related to a slow degradation of the surface of the reaction cell during the deployment period. The sensitivity losses were corrected by applying a robust linear fit interpolation of the sensitivity between the pre-and post-deployment calibrations. The robust linear interpolation is confirmed by the internal stability checks of NO 2 during the deployment phase (Fig. 8) and well documented from the MOZAIC NO y measurements . The internal stability checks of NO 2 , however, are not used for determining the  Figure 7. Relative uncertainty of NO (black, day) and of NO 2 (blue, day; red, night) using all measurements (4 s) in 2015.
mixing ratios from the raw signal. It should be noted that final data (L2) are provided after the post-calibration. Therefore the instrument operation period is kept short to a maximum of 6 months.

Instrument intercomparison
The German Weather Service organized an intercomparison of instruments measuring NO / NO 2 / NO x mixing ratios within the framework of ACTRIS. Here 11 European laboratories participated with 17 different state-of-the-art NO, NO 2 and NO x instruments during a 2-week period in October 2016. Most of the time all instruments agreed well, and the results of this workshop will provide detailed crosssensitivities of each individual instrument compared to the reference CLD instrument of the World Calibration Center (WCC) NO x . The WCC NO x instrument (here after REF) was regularly calibrated during this campaign and is used as a reference. Figure 9 shows correlations of NO and NO 2 for the IA-GOS NO x and the REF instruments for ambient air measurements during 2 days of this campaign. The ambient air was distributed by a ring line of 20 m length, with residence times of approximately 5 to 6 s from the first to the last instrument and corrected for ambient ozone mixing ratio. Mixing ratios of NO were observed in the range of the detection limit and 6 ppbv. The correlation coefficient is higher than R 2 > 0.98 with a slope of 1.037 and an offset of −18 pptv. NO 2 was observed in the range of 0.5 to 10 ppbv with R 2 > 0.94 with a slope of 1.063 and an offset of −102 pptv. The NO 2 data are more scattered than NO data, which is related to the different cyclic measurements of NO and NO 2 by both instruments. Further results (e.g., chemical interferences) will be presented in a separate paper. This and future intercomparisons will assure the quality of the IAGOS NO x instrument.

Photolytic decomposition
It is known that photolytic decomposition of nitrous acid (HONO) can occur when using a photolytic converter for the detection of NO 2 with CLD instruments (e.g., . During the ACTRIS NO x side-by-side intercomparison the interference of HONO within the IAGOS NO x instrument was determined to be about 10 % of the NO measurements at 11 ppbv. In situ observations of HONO in the UTLS regions are very rare, and they report only a few ppt (Jurkat et al., 2011(Jurkat et al., , 2016. Thus, the interferences are mostly below the total uncertainties for NO and NO x . This is also the case for BrONO 2 and NO 3 . Both species can be decomposed within the photolytic converter. The concentrations of both species are too low (< 10 ppt) in the UTLS region; thus we expect no major impact on the NO 2 measurements (Avallone et al., 1995;Brown et al., 2007;Carslaw et al., 1997).

Thermal decomposition of NO 2 -containing species
The instrument temperature is measured and varies mostly between 15 and 22 • C during flight. With the aircraft being close to the ground, the instrument temperature can rise up to 30 • C in summer. However, the gas temperature inside the PLC increases when the LEDs are switched on. Laboratory measurements showed that the gas temperature in the converter is in the range of 40 to 70 • C at an instrument temperature of 30 to 35 • C (Fig. 3). From these experiments, we extrapolate a gas temperature inside the converter between 27 • C (300 K) and 47 • C (320 K) during flight. As a result, thermal decomposition of reservoir species containing NO 2 can lead to erroneously enhanced NO 2 measurements. Reed et al. (2016) showed that the PAN interference could be up to 8 and 25 % when using an actively cooled and a not actively cooled photolytic converter, respectively. In the laboratory, we found NO 2 enhancements of 30 % by mixing PAN with the sample flow (at 35 • C instrument temperature and pressure level of 250 hPa), which was quantitatively generated from a NO calibration gas by photolysis of acetone (100 ppbv) in a flow system (Pätz et al., 2002;Volz-Thomas et al., 2002). The result is in good agreement with theoretical calculations of the lifetime of PAN at the maximum expected temperature of 340 K (at 250 hPa) in the PLC, which predicts an interference of 27 % to NO 2 . However, temperatures in the PLC above 320 K are not expected during flight, because instrument and unit temperatures are much lower than in the laboratory, and thus PAN interferences should be less than 3 % for NO 2 . Table 6 provides an overview of possible interference to the NO 2 measurements over different temperature ranges of the typical reservoir species containing NO 2 (dinitrogen pentoxide (N 2 O 5 ), peroxynitric acid (HO 2 NO 2 , only during daytime), methyl peroxy nitrate (CH 3 O 2 NO 2 ), and peroxyacetyl nitrate (= PAN, CH 3 CO 3 NO 2 )) at cruising altitude (250 hPa).  Table 7. At this stage, parts of the IAGOS measurements are available only as L1 data (preliminary), which explains the large fraction of limited data in 2016. Progression of the data to L2 (final) is ongoing. Here, we show the first results as examples, to demonstrate the performance of the instrument. A detailed analysis will be presented in a separate paper once all data are finalized. Figure 11 shows the NO and NO 2 mixing ratio probability density functions during all nighttime flights at cruising altitude (p < 350 hPa). The NO mixing ratio is expected to be zero within the standard deviation (1σ ) of 25 pptv, which is equal to the statistical precision of the instrument at 4 s time   resolution. The quality of the IAGOS NO x measurement is determined not only by the instrument precision but also by the homogeneity and representative of the climatological data set. Therefore, the NO measurements at nighttime are used as an additional quality check during each flight. Sometimes, a small negative NO offset is found (NO < −10 pptv), which occurs due to subtraction of the zero-air signal from the net signal at very low mixing ratios of NO and NO x . However, the half width of the distribution is larger than the random noise of the detector, and therefore the NO mixing ratio offset value is assumed to be zero. The median mixing ratio of NO 2 is 138.6 pptv with a width range from 0 pptv to several hundred pptv. A comparable nighttime median NO 2 value of 141 pptv was observed for the 2016 deployment period in the UTLS region. During daytime, NO recovers by the photochemical balance with NO 2 , which leads to a median distribution for NO mixing ratios of 57 pptv (86 pptv in 2016) and for NO 2 mixing ratios of 78 pptv (47 pptv). The sum of daytime NO and NO 2 mixing ratios in 2015 is only 1 % smaller compared to the nighttime NO 2 median value, which is equivalent to NO x . Differences of daytime NO and NO 2 mixing ratios between 2015 and 2016 are related to different flight routes and flight levels (Fig. 10).

NO and NO 2 partitioning in the UTLS region
The NO x partitioning is now compared to previous observations obtained by NOXAR and by CARIBIC. Brunner et al. (2001) showed median NO x values of around 140 pptv (96 pptv) for summer (autumn) in the UTLS region over the North Atlantic in 1995 and 1996. The authors calculated NO x with the photochemical balance using only daytime observations of NO and ozone. These median NO x values can be confirmed by splitting the IAGOS measurements in 2015 into summer (165 pptv) and autumn (84 pptv), where the differences between the NO x median mixing ratios are less than  15 %. The NO x values from CARIBIC are also calculated with the photochemical balance method, but using only daytime observations of NO and ozone, and considering only tropospheric air (Stratmann et al., 2016). In summer the median NO x mixing ratio is close to 200 pptv, and in autumn 100 pptv, which is approximately 16 % larger than the values found in our IAGOS measurements.
The median of the IAGOS NO x mixing ratios agrees well with the calculated median mixing ratios of NO x from NOXAR and CARIBIC. However, previous studies identify an unexplained imbalance between the measured and calculated NO 2 in low-NO x regions, which was explained by interferences of NO 2 -containing species and the large uncer-tainty of the calculations (e.g., Crawford et al., 1996;Reed et al., 2016). Thus, the impact of interference from NO 2containing species on the IAGOS measurements requires further investigations, which will be performed once a larger data set is available.

Discussion of observed features in the UTLS
As a first showcase of what can be gained from the IAGOS NO x observations, Fig. 12 demonstrates a time series of all measured compounds for the flight from Düsseldorf to New York City on 23 August 2015. The measurements (CO, O 3 , NO, NO 2 etc.) are presented as 2 min median averages to reduce the noise, and the potential vorticity (PV) was calculated using ECMWF (European Centre for Medium-Range Forecasts) ERA-Interim (Dee et al., 2011) data interpolated along the flight track (Berkes et al., 2017).
We want to focus now on the first more pronounced peak of NO 2 starting at 23:00 UTC, where we suggest an intrusion of polluted air into the lowermost stratosphere. NO varies around 0 pptv during nighttime as expected, while a distinct strong peak of NO 2 is observed at 11.5 km altitude at 23:00 UTC which lasts for about an hour and is correlated with CO and relative humidity. The timely coincidence with high CO and H 2 O values indicates that this air mass is highly polluted compared to typical mixing ratios at this altitude. This large peak is observed above the local tropopause, which can be identified by the chemical and dynamical tropopause heights. The chemical tropopause is often reported at 120 ppbv of ozone, and within the NO 2 plume the ozone mixing ratio is mostly larger than 150 ppbv (Thouret et al., 2006;Sprung and Zahn, 2010). The location of the dynamical tropopause varies between 2.5 and 5 PVU within the NO 2 plume, which is above the commonly used 2 PVU defined location of the dynamical tropopause for the midlatitudes (Kunz et al., 2011).
The origin of this peak was identified using the Lagrangian transport model FLEXPART model. Here a rapid vertical transport from the surface by deep convection of a longrange-transported biomass burning plume could be identified. The FLEXPART model (version 9.02) was used to identify the region with the largest contribution from the surface using 5-day backward simulations from the particle dispersion (Stohl et al., 2005). FLEXPART results showed that a surface-based air mass was lifted from the northwestern US within the previous 4 days. Here near-surface emissions of NO and NO 2 from biomass burning could be identified using fire count maps from satellite images during that time (Fig. S3). These fire emissions contributed also largely to poor air quality in the mid-US at that time (Creamean et al., 2016;Lindaas et al., 2017). Further analyses are beyond the scope of this paper, but this showcase study already indicates the possibilities for air quality studies using the full amount of IAGOS observations.

Vertical profiles
Satellite column observations allow monitoring of NO 2 on a global scale, but the columns do not provide vertical resolution within the troposphere (although there have been recent cloud-slicing methods giving satellite NO 2 profiles on a cli-matological basis), and the satellite retrieval depends on assumptions on the vertical distribution of NO 2 (Bucsela et al., 2008;Boersma et al., 2011;Veefkind et al., 2012). Laughner et al. (2016) showed that the estimates of NO 2 at the surface can be largely uncertain in regards to the daily meteorology if the a priori profile for NO 2 is not well known. So far, only a few methods exist to provide in situ NO 2 profiles, however with some limitations (e.g., Piters et al., 2012). We believe that this assumption can be evaluated with in situ vertical profiles of NO 2 from IAGOS to improve the satellite retrievals, which has been successfully demonstrated for CO (de Laat et al., 2014) and ozone (Zbinden et al., 2013).
In total, more than 400 descent profiles of nitrogen oxides are currently available over several regions in 2015 and 2016 (Fig. 10). Figure 13 shows the statistical analysis of NO and NO 2 only at daytime over Düsseldorf Airport in summer (JJA) 2015. The vertical average was calculated in 50 hPa intervals from 200 to 1000 hPa. Median NO and NO 2 values reach up to 200 pptv in the UTLS region (9-12 km), which agrees well with the previous observations over the eastern North Atlantic shown by Ziereis et al. (1999Ziereis et al. ( , 2000. The median NO and NO 2 values in the mid-troposphere (5 to 9 km), where no major sources exist, vary between the detection limit and 100 pptv. The largest values of nitrogen oxides are measured near the surface, with values up to several ppbv. It should be noted that these values represent a highly polluted region with a huge amount of emissions from ground traffic, industry and aviation. In further studies, the unique  IAGOS NO 2 profiles will be used for a new satellite mission (TROPOMI, http://www.tropomi.eu, last access: 29 November 2017) and model evaluation (e.g., air quality).

Discussion and conclusion
The IAGOS NO x instrument (P2b) setup provides measurements of nitrogen oxide with good precision and accuracy, while its design and performance are highly constrained by aircraft safety considerations and the requirement for unattended deployment over several months. We presented the different components and the determination of the uncertainties. The relative uncertainty of an individual 4 s data point is dependent on the ambient mixing ratio; for NO it reaches 25 % at 0.2 ppbv and 8 % at 1 ppbv, and for NO 2 it reaches 50 and 18 %, respectively. So far only a few instruments are available which can be used for unattended aircraft observations over several months, because of the need of a high temporal resolution and a low detection limit and fulfillment of the safety requirements. The IAGOS NO x instrument has a shorter residence time (at cruising altitude) and much larger conversion efficiency of NO 2 to NO than instruments using xenon lamps in the 1990s, which dramatically improves the instrument accuracy (Ryerson et al., 2000). The detection limit of the IAGOS NO x instrument (LOD NO = 24 pptv and LOD NO 2 = 35 pptv at 4 s, 2σ and 0.9 cps pptv −1 detector sensitivity) is in the range of research-grade instruments used in research aircraft (e.g., CLD technique: LOD NO = 10 to 50 pptv and LOD NO x = 30 to 80 pptv at 1 s (Pollack et al., 2012); CRDS technique, 1 s, 2σ : LOD NO = 140 pptv and LOD NO 2 = 90 pptv (Wagner et al., 2011)).
A major advantage of the IAGOS NO x instrument is the provision of NO and NO 2 in situ measurements on a global scale with comprehensive seasonal and geographical coverage of the UTLS region, and the measurements of vertical profiles from cruising altitude down to the surface over different continents. The emerging data set permits statistically robust conclusions on the seasonal and geographical distribution of NO x . As a first example, the statistical analysis over the North Atlantic region shows lower median mixing ratios of NO and NO 2 in the UTLS compared to previous projects where NO 2 was determined with the photochemical balance, which is an indication that the possible interferences might be small if the amount of NO x has not changed over the recent years.
Possible interferences for NO from HONO could be estimated to the order of 10 %. The water vapor quenching effect on the NO signal was determined in the laboratory and is applied to the in situ measurements if water vapor measurements are available. Note that most of the time the aircraft samples in very dry air, where the correction is negligible. However, close to the surface the water vapor correction factor increases up to 10 % at 30 000 ppmv. We apply to the measurements pressure-and temperature-dependent ozone corrections, which have large effects on NO (up to 25 %). Thermal decomposition of NO 2 -containing species might be a major source of uncertainty to the observed NO 2 mixing ratios. This also includes the blue-light converter, where we aim to reduce the temperature dependency while it is switched on and off within the next instrument revision.
The global distribution of NO x in the UTLS region in combination with transport model calculation allows calculating impact ratios of anthropogenic compared to natural emissions and the concurrency of large-scale plumes. This will lead to a better understanding of the ozone chemistry in the highly climate-sensitive region of the UTLS. Vertical profiles of NO 2 show the expected C-shape profile, and the nearsurface data can be used to monitor air quality in the vicinity of airports. Further, the day-to-day variations can be provided to improve satellite a priori profiles in the future (TROPOMI, http://www.tropomi.eu/, last access: 29 November 2017).
The current setup of the IAGOS NO x instrument provided more than 800 h of observations and 400 profiles using only one passenger aircraft as platform within 2 years (each 6 months). In the near future the number of aircraft will increase, leading to a larger statistical robustness of comprehensive seasonal and geographical coverage of in situ NO and NO 2 measurements.
Data availability. The data used in this study will be available from the central IAGOS database on the IAGOS website (http://www. iagos.org, last access: 29 November 2017.
Competing interests. The authors declare that they have no conflict of interest.
Acknowledgements. The authors are grateful to Deutsche Lufthansa AG for providing special certification for the instrumentation on one A340-300, and in particular to Gerd Saueressig and to Markus Huf. We also thank enviscope GmbH, in particular Ralf Stosius and Gomolzig Flugzeug-und Maschinenbau for their excellent and continuous support within the IAGOS project. We gratefully acknowledge the continuous support by Andreas Volz-Thomas during the final development of the instrument and during the preparation of the manuscript. Without his fundamental work on the measurement of total odd nitrogen and nitrogen oxides in the MOZAIC and IAGOS programs, this instrument would not exist. Marlon Tappertzhofen, Torben Blomel, Marcel Berg, Benjamin Winter, Jennifer Gläser and Günther Rupsch are acknowledged for their enormous help in maintaining and calibrating the instrument. Dominik Brunner and Helmut Ziereis are acknowledged for fruitful discussions and as external reviewers for the SOP. Furthermore, we acknowledge ECMWF for providing meteorological analyses. Part of this project was funded by BMBF under IAGOS-D contract 01LK1301A. The IAGOS database is supported by AERIS (CNES and INSU-CNRS), where the IAGOS data are stored.
The article processing charges for this open-access publication were covered by a Research Centre of the Helmholtz Association.
Edited by: Folkert Boersma Reviewed by: two anonymous referees