Modiﬁcation of a conventional photolytic converter for improving aircraft measurements of NO 2 via chemiluminescence.

. Nitrogen oxides (NO x ≡ NO + NO 2 ) are centrally involved in the photochemical processes taking place in the earth’s atmosphere. Measurements of NO 2 , particularly in remote areas where concentrations are of the order of pptv (parts per trillion by volume), are still a challenge and subject to extensive research. In this study, we present NO 2 measurements via photolysis-chemiluminescence during the research aircraft campaign CAFE Africa (Chemistry of the Atmosphere - Field Experiment in Africa) 2018 around Cabo Verde as well as the results of laboratory experiments to characterize the photolytic converter 5 used. We ﬁnd the NO 2 reservoir species MPN (methyl peroxy nitrate) to produce the only relevant thermal interference in the converter under the operating conditions during CAFE Africa. We identify a memory effect within the conventional photolytic converter (type 1) associated with high NO concentrations and rapidly changing water vapor concentrations, accompanying changes in altitude during aircraft measurements, which is due to the porous structure of the converter material. As a result, NO 2 artifacts, which are ampliﬁed by low conversion efﬁciencies, and a varying instrumental background adversely affect 10 the NO 2 measurements. We test and characterize an alternative photolytic converter (type 2) made from quartz glass which improves the reliability of NO 2 measurements in laboratory and ﬁeld studies.


Introduction
NO x (nitrogen oxides) represent the sum of NO (nitric oxide) and NO 2 (nitrogen dioxide) which can rapidly interconvert in the atmosphere in the presence of sunlight and O 3 (ozone) as shown in Reactions (R1) and (R2) (Jacob, 1999). R2) j N O2 is the photolysis frequency for NO 2 in Reaction (R2).
Several studies have shown that the Leighton ratio as presented in Equation (1) is only valid for highly polluted environments whereas in other regions, besides O 3 , oxidized halogen species and peroxy radicals (HO 2 and RO 2 ) significantly contribute to the oxidation of NO to NO 2 and require an extension of the Leighton expression for a reliable calculation of PSS NO 2 25 concentrations as presented in Sect. 2.4. (Mannschreck et al., 2004;Griffin et al., 2007;Ma et al., 2017;Reed et al., 2016). NO x can be emitted from either natural and anthropogenic sources, with the latter dominating globally. Natural emissions include for example biogenic soil emissions, biomass burning and lightning. Anthropogenic emissions are mainly from combustion processes in vehicles or from power and industrial plants which contribute almost two thirds to the global burden Ganzeveld et al., 2002;Logan, 1983). Nitrogen oxides are, together with volatile organic compounds, 30 important precursors to tropospheric ozone which can be a hazard to plant, animal and human health, causing multiple diseases regarding the cardiovascular and respiratory system (Nussbaumer and Cohen, 2020;Nuvolone et al., 2018;Lippmann, 1989). NO x additionally promote the formation of acid rain (through conversion to HNO 3 ) -hazardous to many ecosystems -and are a threat to human health themselves (Boningari and Smirniotis, 2016;Greaver et al., 2012). Beyond that, NO x control the abundance of OH radicals which regulate the oxidizing capacity of the atmosphere (Levy, 1971;Lelieveld and Dentener, 35 2000). Due to the health implications and the impact on atmospheric photochemical processes, it is highly relevant to measure and monitor ambient NO x concentrations with sophisticated instruments which provide reliable concentration measurements, especially also in remote areas where NO and NO 2 are low. More specifically, this requires a low instrumental background which -particularly for NO 2 -is often impacted by unwanted chemical processes which can lead to artifact signals (Reed et al., 2016;Andersen et al., 2020;Jordan et al., 2020). 40 Many different measurement techniques have been deployed to measure nitrogen oxides such as cavity enhanced absorption spectroscopy (and variants, e.g. cavity attenuated phase shift spectroscopy (Ge et al., 2013;Kebabian et al., 2005), cavity ring down spectroscopy (O'Keefe and Deacon, 1988) and others (Zheng et al., 2018)), differential optical absorption spectroscopy (Hüneke et al., 2017;Winer and Biermann, 1994) and laser induced fluorescence (Thornton et al., 2000;Javed et al., 2019) for NO 2 or absorption spectroscopy for NO (Ventrillard et al., 2017). However, detection of NO and NO 2 via chemiluminescence 45 (CLD) is likely the most common technique for the measurement of nitrogen oxides in the atmosphere and is distinguished by the simultaneous in-situ measurement of both, NO and NO 2 , low detection limits and the deployability in research aircrafts at high altitudes for measurements in the upper troposphere (Pollack et al., 2010;Reed et al., 2016;Tadic et al., 2020). The measurement principle is based on the reaction of nitric oxide and ozone which yields electronically excited NO 2 (NO 2 *) which (along with physical quenching) returns to the electronic ground state by fluorescence whereby a photon of a wavelength 50 > 600 nm is emitted, which can be detected by a photomultiplier tube. The resulting signal is proportional to the initial NO concentration (Clough and Thrush, 1967). For nitrogen dioxide detection, NO 2 is first converted to NO. The standard method for this conversion is the use of a catalytic converter, in which NO 2 passes through a heated molybdenum converter where it is reduced by Mo to NO (Mo + 3 NO 2 → MoO 3 + 3 NO). However, high temperatures (300 -350 • C) in the converter along with catalytic surface effects lead to interferences with other atmospheric compounds that can be converted to NO 2 such as HONO 55 (nitrous acid), HNO 3 (nitric acid) or PAN (peroxyacyl nitrate) and bias the measurement (Demerjian, 2000;Villena et al., 2012;Jung et al., 2017). An alternative and widespread method is the use of a photolytic converter (photolysis-chemiluminescence: P-CL), also referred to as blue light converter, which utilizes LEDs emitting at a wavelength of around 395 nm to dissociate NO 2 to NO (Pollack et al., 2010;Reed et al., 2016;Tadic et al., 2020;Ryerson et al., 2000). Interferences (as described above) in the blue light converter are still possible, but to a significantly lesser extent. Reed et al. (2016) investigated potential 60 interferences in a photolytic converter which are related to the presence of PAN, methyl peroxy nitrate (MPN, CH 3 O 2 NO 2 ) or pernitric acid (PNA, HO 2 NO 2 ). These compounds are NO 2 reservoir species and their decomposition (to NO 2 ) is dependent on the temperature, the pressure and the residence time in the blue light converter (Nault et al., 2015;Fischer et al., 2014). Please note that none of these compounds are photolyzed in the blue light converter and only subject to thermal decomposition (Reed et al., 2016;Tadic et al., 2020). Generally, increasing temperature and residence time promote the decay of thermally unstable 65 trace gases and the release of NO 2 which is further described in Section 2.5 (Reed et al., 2016). With increasing residence time in the converter and high atmospheric HONO/NO 2 ratios photolysis of HONO could become relevant as recently shown by Gingerysty et al. (2021).
The CLD detects a signal (which we call NO c signal) which is composed of the ambient NO concentration and the ambient NO 2 concentration multiplied by the conversion efficiency C e according to Equation (2).
The conversion efficiency describes the fraction of NO 2 that is converted to NO in the converter and can be thought of as the NO yield from NO 2 . Its value is dependent on the optical output of the LEDs as well as the NO 2 residence time and the pressure in the converter. C e is therefore in competition with unwanted formation of NO 2 from NO 2 reservoir species. For example, a longer residence time increases the conversion efficiency, but could potentially increase the amount of NO 2 reservoir species 75 that decay in the converter, which takes place according to first order kinetics which is described in more detail in Section 2.5. The NO 2 concentration is calculated from the difference in the signal with and without use of the photolytic converter: Sadanaga et al., 2010;Tadic et al., 2020;Ryerson et al., 2000).
While NO measurements are generally reliable and well-understood, NO 2 measurement techniques utilizing the conversion of NO 2 to NO are subject to extensive research. Hosaynali Beygi et al. (2011) found a strong deviation from the Leighton 80 ratio at low NO x concentrations between 5 and 25 pptv despite the inclusion of HO 2 , RO 2 and halogen oxides suggesting the occurrence of a so far unknown atmospheric oxidant. Frey et al. (2015) also reported higher measured NO 2 /NO ratios than expected from PSS based on measurements in Antarctica and hypothesized the presence of an additional oxidant or a measurement bias. This is in line with findings and suggestions by Silvern et al. (2018) based on observations during the aircraft campaign SEAC 4 RS over the United States of America. Reed et al. (2016) examined the described deviation through the lab-85 oratory investigation of potential NO 2 interferences of thermally unstable trace gases such as peroxyacyl nitrate (PAN) within the photolytic converter in comparison to laser-induced fluorescent NO 2 measurement and found that this could contribute to the higher than expected NO 2 concentrations measured by P-CL instruments. Jordan et al. (2020) investigated interferences in a photolytic converter made from quartz glass and showed how the converter conditions affect the conversion efficiency and the artifact signal (caused by NO 2 reservoir species). The correct adjustment of the conditions, preferably including low pressure, 90 high flow rates and small temperature variations, can minimize interferences which was also concluded by Reed et al. (2016). Andersen et al. (2020) reported the measurement of a significant NO 2 measurement bias during ground-based observations in the remote marine tropical troposphere with a conventional blue light converter which was related to its porous walls. They were able to eliminate this effect by implementation of a photolytic converter made from quartz glass which reduced the overall measurement uncertainty by around 50 %. The use of quartz glass in a blue light converter was also reported by Pollack et al. 95 (2010) who compared the commercially available converter BLC-A manufactured by Droplet Measurement Technologies to other photolytic converters.
An additional challenge is the significant decrease in the NO 2 /NO ratio with altitude. At the surface at daytime, NO 2 concentrations are approximately two to four times higher than NO concentrations. The NO 2 /NO ratio decreases by around one order of magnitude when going from the lower to the upper troposphere which increases the uncertainty when deriving NO 2 mixing 100 ratios using Equation (2) (Travis et al., 2016;Silvern et al., 2018;Logan et al., 1981). At the same time, the concentration of NO 2 reservoir species such as PNA or MPN is significantly higher in the upper troposphere compared to that at the surface and consequently interferences are more likely to occur at high altitudes (Nault et al., 2015;Kim et al., 2007). These aspects result in particularly strict requirements regarding airborne NO 2 measurements.
In this study, we describe a modified blue light converter (BLC) (type 1) originally purchased from Droplet Measurement

105
Technologies, which we have deployed in NO 2 measurements via photolysis-chemiluminescence during the research aircraft campaign CAFE Africa (Chemistry of the Atmosphere: Field Experiment in Africa) and also in laboratory investigations. We show how high NO concentrations and rapidly changing water vapor concentrations affect the instrumental background and induce a memory effect which cannot be corrected retrospectively. This is particularly relevant to aircraft measurements where water vapor concentrations are subject to rapid changes due to variations in flight altitude, but also to all other application areas.

110
The photolytic converter and similar designs are widely used for field measurements of NO 2 all across the world (e.g. Andersen photolytic converter entirely made from quartz glass (type 2). Highly reflective properties are achieved by an outer mantel made from optical PTFE (polytetrafluoroethylene, also known as teflon). The type 2 quartz converter shows promising results in the laboratory regarding its application in field studies for more reliable NO 2 measurements. We do not claim to be the first to present an alternative quartz glass converter for P-CL measurement of NO 2 . However, we are first to point out the technical difficulties in the application of conventional NO 2 converters in airborne studies and believe the presented results to 120 be a guidepost for future NO 2 aircraft measurements via photolysis-chemiluminescence.
2 Observations and methods

Instrument
All NO x measurements were performed using a modified two-channel chemiluminescence instrument originally purchased from ECO Physics, Dürnten, Switzerland (CLD 790 SR) as described by Tadic et al. (2020) (Figure 2 presents the instrument 125 schematic) operated at a total gas flow of 3 SLM, equally divided into the two channels. NO concentrations are measured in the first channel, also referred to as the NO channel, through formation of NO 2 * via reaction with O 3 . The resulting excited NO 2 * emits a photon (> 600 nm) detected by a photomultiplier tube, preamplifier set up and recorded as counts per second.
The second channel, also referred to as NO c channel, is structurally identical except for the implementation of a photolytic converter which converts a known fraction of NO 2 to NO prior to the reaction with O 3 and is operated at a constant pressure  Figure   1a. The converter was designed for airborne applications. Originally, the inner material is made of porous, optically active 135 PTFE (polytetrafluoroethylene) for providing highly reflective properties. To reduce surface effects the converter was equipped with a quartz cylinder covering approximately half of the PTFE surface (the gas still gets in touch with the PTFE surface in the ring channel and through the head piece). Please note that this modification was made prior to the CAFE Africa research campaign within a limited time frame and did not have the desired outcome. The sample gas enters the converter sideways into the ring channel and reaches the inner tube via the PTFE head piece which has four circular recesses, one for each UV LED.

140
The sample gas outlet proceeds analogously. The inner volume of the converter is V = 78 cm 3 which gives a residence time of t = V × 60 s min −1 F × p p standard = 78 cm 3 × 60 s min −1 1500 cm 3 min −1 × 110 hPa 1013 hPa = 0.34 s. The conversion efficiency for this type 1 photolytic converter operated under the conditions described above is approximately 20 % (j = 0.66 s −1 ) which was determined via gas phase titration (GPT) of NO with ozone. The results obtained with the described type 1 converter were compared to an alternative photolytic converter completely made from quartz glass (type 2) which is shown in Figure 1b reservoir species increases with increasing pressure in the converter which can be seen in Figure S1a of the Supplement. On the other hand, a higher conversion efficiency would be desirable for improved accuracy of the measurement. The main difference between the two converters is that the sample gas flow does not have direct contact with the porous surface of the material for the type 2 quartz converter. Additionally, the sample gas flow in the type 2 quartz converter does not have contact with the LEDs which likely minimizes the sample gas heating and consequently the thermal interferences when passing through 160 the converter. The reaction chambers (where the reaction of NO and O 3 takes places) are operated at a constant temperature of 25 • C and a pressure of 9 -10 mbar in order to minimize quenching of NO 2 * by other molecules. The dry ozone flow is humidified with water vapor for maintaining a constant humidity level at all times.
Besides the photons emitted from relaxation of NO 2 *, the PMT signal also includes detected photons from interference reactions, for example the reaction of O 3 with alkenes (Alam et al., 2020), as well as a dark current signal. Therefore, a pre-165 chamber measurement is operated for 20 seconds every 5 minutes where ozone is added to the sample gas flow. The residence time in the pre-chamber allows for the reaction of O 3 and NO and the relaxation of NO 2 * before entering the main reaction chamber (pre-chamber efficiency > 96 % for the NO channel and ∼ 100 % for the NO c channel). It is not long enough to convert interfering compounds which then occurs in the following main chamber. Consequently during pre-chamber measurements, the PMT signal only includes the interfering signal and the dark current signal (Ridley and Howlett, 1974; ECO PHYSICS AG, . We subtracted the interpolated signal obtained during pre-chamber measurements from the signal detected during mainchamber measurements in order to obtain the signal generated from NO. The material of both the pre-and the main-chambers is gold-plated stainless steel. The instrumental background of each channel is determined via zero (synthetic) air measurements from a gas cylinder and can be converted to mixing ratios using calibration measurements with a known NO concentration which defines the sensi-175 tivity (counts s −1 per ppbv (parts per billion by volume) of each channel towards NO as shown in Eq.
(3) (after pre-chamber corrections). The signal detected from zero air measurement (counts(zero air)) is subtracted from the signal detected from NO calibration (counts(NO calibration)) and divided by the absolute concentration of the NO calibration (c(NO calibration)) to calculate the sensitivity. Dividing the signal detected from zero air measurements by this value gives the instrumental background concentration in mixing ratios, e.g. ppbv. The precision is determined from the reproducibility of the NO calibrations 180 and is 3 % (1σ). The NO concentration is 4.96 ± 0.21 ppmv which gives a 4 % uncertainty on the used secondary standard.
The resulting NO calibration mixing ratio is 15.8 ± 0.7 ppbv. The detection limit is given by the reproducibility of the zero air measurements which is around 5 pptv for the NO channel and the NO c channel using the type 2 quartz converter. The detection limit is higher when using the type 1 converter, but difficult to determine due to the observed memory effects and estimated at > 10 pptv.
Please note that the utilized zero air can include a trace concentration of NO x . The manufacturer specifies the maximum concentration of NO x to be 0.1 ppmv (parts per million by volume) (Westfalen Gas Schweiz GmbH).

CAFE Africa field experiment
The

195
The sample line temperature was approximately 25 • C. More details on the campaign can be found in Tadic et al. (2021).
NO and NO 2 were measured via photolysis-chemiluminescence with the instrument described in Sect. 2.1 using the type 1 conventional blue light converter equipped with the quartz glass cylinder, operated at a temperature of 313 K and a pressure of 105 hPa (0.32 s residence time). Please note that it was not possible to measure the temperature inside the converter. Instead, the temperature of the gas outflow from the converter in the ring channel was measured which we equate to the inner temperature.

200
Zero air measurements and NO calibrations using a secondary NO standard (cylinder concentration of 1.187 ± 0.036 ppmv and calibration mixing ratio of 2.97 ± 0.09 ppbv) were performed regularly to determine the variability in the instrumental background and the sensitivity of the channels. The ambient measurement was interrupted every 1 -2 h by one minute zero air measurement, followed by one minute NO calibration and another one minute zero air measurement. These calibrationbackground-cycles (CB-cycles) were performed 4 -6 times during each measurement flight. We linearly interpolated these

Further measurements
Additional measurements of atmospheric trace gases during CAFE Africa including O 3 , CO, CH 4 , HO 2 , OH, NO 2 and water vapor as well as the photolysis frequencies j N O2 and j P N A were used in this study. O 3 was measured via UV absorption and chemiluminescence with the FAIRO (Fast AIRborne Ozone) instrument (total measurement uncertainty of 2.5 %, Zahn et al.

220
(2012). CO and CH 4 were measured via quantum cascade laser absorption spectroscopy (total measurement uncertainty of 4.3 % and 0.3 %, respectively, Schiller et al. (2008)). HO 2 and OH were measured with the custom-built HORUS (HydrOxyl Radical measurement Unit based on fluorescence Spectroscopy) instrument via fluorescence spectroscopy (Novelli et al., 2014;Marno et al., 2020). Please note that these data are still preliminary and the measurement uncertainty is estimated at 50 %.
Additional NO 2 concentrations for comparison were measured via differential optical absorption spectroscopy (miniDOAS) 225 with a detection limit of about 5 pptv and an uncertainty depending on the altitude and cloud cover of typically 40 pptv (Hüneke et al., 2017;Kluge et al., 2020). Water vapor was measured via direct absorption by the tunable diode laser system SHARC (Sophisticated Hygrometer for Atmospheric ResearCh) (accuracy of 5 %, detection limit typically in the range of 2 -3 ppmv) (Kaufmann et al., 2018). The photolysis frequencies were calculated from actinic flux densities measured with a spectral radiometer (Meteorologie Consult GmbH, Metcon, Koenigstein, Germany) (uncertainty < 15 %) (Bohn and Lohse, 2017).

230
Please note that all measurement data were converted to a uniform timescale with a 1 min time resolution as a basis for this analysis.

NO 2 calculations
For calculating the photostationary state NO 2 concentrations during CAFE Africa, we assume that NO 2 production occurs through reaction of NO with O 3 (Reaction (R1)), HO 2 (Reaction (R3)) and RO 2 (Reaction (R4)). Tadic et al. (2021) showed 235 that RO 2 is well represented by CH 3 O 2 during CAFE Africa via model simulations (80 % at 200 hPa altitude and up to 90 % below) which we therefore use as surrogate for describing all organic peroxy radicals. In analogy to Leighton (1961), we describe NO 2 loss by photo dissociation as shown in Reaction (R2). Other loss pathways for NO 2 for example via OH can be neglected (< 1 %) (Bozem et al., 2017).
NO 2 concentration in photostationary state can therefore be obtained via Equation (5) whereas the concentration of CH 3 O 2 is calculated by help of Equation (6) which was derived by Bozem et al. (2017). For the calculation via Equation (6) we assume that CH 3 O 2 and HO 2 formation occur through CH 4 and CO oxidation, respectively. We estimate an uncertainty of around 20 % 245 resulting from these assumptions. Propagating the measurement uncertainties of HO 2 , CH 4 and CO suggests a 50 % uncertainty in the calculated CH 3 O 2 data. The NO 2 PSS data have an uncertainty of 22 % regarding the trace gas measurements according to Gaussian error propagation (uncertainty of rate coefficients is considered negligible).
The temperature dependent rate coefficients were obtained from the data sheets of the IUPAC Task Group on Atmospheric Chemical Kinetic Data Evaluation (2021) (Atkinson et al., 2004(Atkinson et al., , 2006.

Calculation of NO 2 reservoir species
We consider the NO 2 reservoir species PAN (peroxyacetyl nitrate), MPN (methyl peroxy nitrate) and PNA (pernitric acid). PAN 255 was measured during CAFE Africa via chemical ionization mass spectrometry (CIMS) (Phillips et al., 2013). MPN and PNA were not measured and instead estimated via photostationary state calculations as suggested by Murphy et al. (2004). PNA (HO 2 NO 2 ) production occurs through reaction of HO 2 and NO 2 (R5) while PNA loss is described by Reactions (R6) Equation (7). k is the rate coefficient for each reaction and j P N A is the photolysis frequency for Reaction (R7).
270 MPN production and loss terms are in analogy to PNA as shown in Reactions (R9)-(R11) except for the reaction with OH which is negligible (Nault et al., 2015;Browne et al., 2011;Murphy et al., 2004;Bahta et al., 1982). The calculation in PSS is performed via Equation (8). k represents the rate coefficients and j M P N is the photolysis frequency for Reaction (R11). During CAFE Africa, only the photolysis frequency j P N A was evaluated because reliable molecular data for MPN were missing. As suggested by Murphy et al. (2004) we assume identical UV cross sections of MPN and PNA and therefore j M P N to be identical 275 with j P N A .
In the photolytic converter PNA, MPN and PAN can decompose to NO 2 depending on the temperature, the pressure and the residence time t according to first order kinetics. The resulting NO 2 artifact is determined via Equation (9).
Gaussian error propagation gives an uncertainty of 55 % for the calculated PNA and MPN data. We use the residence time according to the volume and the flow rate in the photolytic converter as described earlier. The actual value could deviate from the calculated one due to unknown flow-dynamics and temperature gradients. Assuming 30 % uncertainty in the residence time gives an overall uncertainty of around 60 % in the NO 2 formed from PNA and MPN in the photolytic converter.  (8) and (7)) as well as PAN measurements during CAFE Africa. MPN concentrations were 295 close to zero at low altitudes up to 10 km and increased above, reaching 57 ± 40 pptv between 13 and 14 km altitude. The concentration increased further aloft but had a large variability. PNA mixing ratios were low below 8 and above 12 km altitude and showed peak concentrations of 54 ± 21 pptv between 9 and 10 km. PAN increased from ground level to mid-range altitudes with a maximum of 383 ± 283 pptv at 4 -5 km. Concentrations subsequently decreased with altitude, reaching 92 ± 44 pptv at 14 -15 km. Figure 3b shows the NO 2 artifact concentrations resulting from thermal decomposition of the reservoir species in 300 the type 1 blue light converter according to first order decay. It can be seen that only relevant artifact signals originated from MPN of which more than 50 % decomposed to NO 2 at the conditions present in the converter. 3 % of the ambient PNA was  converted to NO 2 . Even though atmospheric PAN concentrations, particularly at mid-range altitudes, were high, temperature, pressure and residence time in the blue light converter were too low for PAN to decay to NO 2 . As an overview, Figure 3c shows the temperature-dependent decay (1 -c/c 0 ) of the discussed NO 2 reservoir species in the converter. The calculation is 305 based on constant pressure (105 hPa) and residence time (0.32 s). The temperature in the converter is shown with the black dashed line. Increasing temperature increases the decomposition share. It can be seen that, for PAN and PNA, the converter temperature would need to be significantly higher to observe a relevant decay (10 % decay of PNA at > 50 • C and of PAN at ∼ 80 • C). In contrast for MPN, small changes in the temperature have a strong effect on the decomposing share (4 % per • C at the steepest point). We show the time-and pressure-dependent decay of PAN, PNA and MPN in Figure S1 of the 310 Supplement. PAN and PNA decay only slightly depends on pressure at the given temperature and residence time. Please note that the residence time and the pressure are correlated, which we have neglected in this calculation. Based on these results, we recommend the implementation of a monitoring system for both temperature and pressure within the photolytic converter which is difficult to implement in the type 1 blue light converter, but allows for a more accurate calculation of the decomposing share and consequently a reliable correction of the NO 2 signal. 315 We have subtracted the NO 2 artifact signal arising from the decomposition of MPN and PNA from the CLD NO 2 concentrations. Please note that the data coverage for the NO 2 artifact from MPN is 55 % and from PNA is 48 % (difference due to OH data coverage). We have interpolated the data used in the following sections to reach full coverage of the CLD NO 2 concentrations. Sometimes the data were incomplete at the start or the end of a measurement flight in which case we considered the averaged NO 2 artifact signal according to the vertical profile shown in Figure 3b as a function of the altitude.  and decreased with altitude up to 2 km. They were constant with 15 ± 16 pptv between 2 and 10 km altitude and agreed to within ∼ 85 % to the calculated values. Concentrations increased again above reaching 54 ± 31 pptv between 11 and 12 km and decreased aloft with values similar to PSS NO 2 between 14 and 15 km. Average NO 2 concentrations measured by the CLD were 49 ± 76 pptv below 10 km altitude where decomposition of reservoir species did not play a role and decreased with altitude above. Figure 4b shows the calculated difference in NO 2 concentrations between PSS calculations and miniDOAS 330 measurements in black, and between PSS calculations and CLD measurements in gray. It is notable that NO 2 concentrations from PSS and miniDOAS measurements were nearly identical apart for a difference with a maximum value of 48 ± 4 pptv between 10 and 13 km altitude. In contrast, CLD NO 2 concentrations were higher by 45 ± 62 pptv compared to the calculation up to 10 km altitude and lower at higher altitudes with a maximum deviation of more than 100 pptv between 14 and 15 km. Figure 4a shows that the NO 2 CLD mixing ratios are negative at high altitudes. This is an indicator of a wrongly measured 335 instrumental background signal in the second channel. If the determined instrumental background was too high, Eq. (4) could return underestimated or even negative NO 2 concentrations. However, the CLD NO 2 data were not generally too small, but even enhanced at lower altitudes compared to PSS and miniDOAS data which may indicate the contribution of additional factors which we investigate in the following by the help of NO, H 2 O and NO 2 concentrations in the course of selected measurement the data are already negative before the subtraction of decomposing NO 2 reservoir species.

Influence of atmospheric water vapor
Atmospheric water vapor concentrations are highest at ground-level and decrease with increasing altitude. As an example, the vertical concentration profile of atmospheric water vapor during CAFE Africa is shown in Figure S2 of the Supplement.
Accordingly, altitude changes during aircraft measurements introduce rapid changes in relative humidity to the instruments 345 on-board. altitude, rising to 9.5 ± 0.7 × 10 3 ppmv on average after reaching 3.9 km (+ 15 minutes). Similar observations were made for MF12 at 18:30 UTC and for MF10 at around 13:30 and 18:00 UTC, in each case accompanied by a decrease in altitude and 375 an increase in water vapor concentrations. The observed NO 2 peaks appeared only for the CLD measurement, and not for the values from PSS calculation or miniDOAS measurement which underlines the instrumental cause. The time series for the measurement flights MF13, MF14 and MF15 shows similar results and can be found in Figure S4 of the Supplement.
We hypothesize that these observations were influenced by a surface effect in the type 1 blue light converter which has a highly porous inner surface as described earlier. This material can adsorb atmospheric compounds, such as NO, and desorb 380 them at a later stage (for example supported by an increase in humidity), which we will refer to as memory effect in the following. In a series of laboratory studies, we have investigated the impact of NO concentrations and humidity on the effects described above and particularly in regard to the instrumental background.

Laboratory experiments and implications for CAFE Africa
We propose that the memory effect described in the previous section is strongly affected by NO molecules and is dependent on 385 changes in the introduced relative humidity. In order to show this, we have conducted different experiments in the laboratory to investigate the instrumental background produced by the photolytic converter in the NO c channel. Beside NO and H 2 O, we suggest that one or more additional factors affect the observed instrumental background signal which are connected to the light of the LEDs and which we discuss at the end of this section.
For the first set of experiments, we exposed the converter (LEDs switched-on) to 16 ppbv of NO for 2 h followed by 4 h 390 zero air measurements. The first experiment was carried out under dry conditions, sampling NO and zero air directly from the gas cylinder. For the second experiment, we introduced water vapor by passing zero air through a washing bottle with deionized water before entering the instrument. The thus obtained relative humidity was ∼ 95 % at ambient temperature and decreased over time with decreasing water temperature (through evaporation). For the third and the fourth experiment, zero air was humidified only for the zero air measurement and the NO measurement, respectively. We repeated the latter introducing a 395 lower relative humidity of ∼ 35 % and obtained the same result. in the NO c channel which were performed during CAFE Africa and used for the data processing according to Equation (4) were consistently too high as they were only run for two times one minute (per CB-cycle) and therefore did not represent the actual instrumental background but an artifact signal. In the laboratory, we observed the strongest effect for a dry instrumental background measurement following a humid NO measurement which was a likely scenario for the ambient monitoring during 415 CAFE Africa as zero air for an instrumental background measurement was sampled from a gas cylinder (dry conditions) and the measured ambient concentration was subject to ambient meteorological conditions. This would explain the occurrence of negative NO 2 concentrations obtained from CLD measurements as mentioned earlier. These experiments also suggest that water molecules might not just promote adsorptive or desorptive processes, but participate in the surface allocation themselves and compete with NO. Following this hypothesis, the surface spaces would fill with NO during an NO measurement under dry Because of the conversion efficiency of 24.4 % of the type 1 blue light converter, the signal difference from the NO and the 430 NO c channel was multiplied by a factor of around four (compare Eq. (4)). Therefore, the resulting NO 2 signal was distorted by four times of the actual desorbed NO explaining the large peaks accompanying altitude descents. As we do not observe these effects in the NO channel (and the two channels are structurally identical), we can exclude that any of the humidity effects are caused by components in the instrument other than the photolytic converter.
We repeated the same laboratory experiments using the type 2 photolytic converter. The resulting temporal development of the 435 instrumental background is presented in Figure 6b which shows significant improvements compared to the type 1 converter.
The instrumental background in the NO c channel was many times smaller for the type 2 quartz converter with mixing ratios of around 10 to 15 pptv and was, most importantly, constant over time. For stationary long-term experiments it could be possible to measure the instrumental background on the scale of hours. However for aircraft measurements and the accompanying rapid air mass changes due to the high aircraft velocity, it is vital to obtain a reliable instrumental background measurement within 440 a short time interval, which would be possible with the type 2 converter. Furthermore, changes in humidity did not seem to impact the measurement as all four experiments show the same result. This too, suggests the suitability of the type 2 converter in aircraft measurements or generally field studies which are impacted by high and changing humidities. Performing zero air measurements after NO 2 measurements had a similar outcome for each of the applied converters. The instrumental background measurement in the type 1 blue light converter showed a decreasing trend over time while it was constant and significantly 445 smaller in the type 2 quartz converter.
Our assumption that the observed effect is -at least partly -associated with NO molecules is supported by an experiment where we heated the type 1 blue light converter with switched-off LEDs with a heat gun and observed a sharp increase in the NO c channel during zero air measurement (following NO calibration measurement). The increase in temperature promoted the desorptive process and had to include NO molecules. Otherwise, the CLD would not have detected any increase in the signal 450 as the converters' LEDs were switched-off and NO 2 could not form NO via the photolytic reaction. We show the result of the heating experiment in Figure 7a. The converter surface was heated for two minutes (under constant movement of the heat gun) at a distance of around 10 cm during zero air measurement. We estimate the surface temperature to not have exceeded 200 • C. We observed a peak NO concentration of 2 ppbv (NO c instrumental background was 0 ppbv). In comparison, Figure   7b shows the experiment repeated with the type 2 quartz converter which showed a small increase in the NO c signal, too, but 455 approximately one magnitude smaller compared to the type 1 blue light converter. The qualitative outcome of this experiment was the same with the LEDs switched-on as well as with preceding NO 2 (instead of NO) measurement. Please note that a direct comparison of experiments regarding adsorptive and desorptive processes with switched-on and -off LEDs is difficult because the operation of the LEDs increases the temperature within the converter which -as shown above -strongly impacts the surface allocation.

460
Beyond that, we performed an experiment to investigate how NO calibration measurements affect subsequent zero air measurements in response to different NO concentrations. Figure 8 shows the influence of the preceding NO concentration level on the following first 5 minutes of zero air measurement. We have performed 30 minute NO calibrations with NO concentrations between 0.25 ppbv and 10 ppbv. Red data points represent the instrumental background of the NO channel and green data points show the instrumental background of the NO c channel. Instrumental background concentrations in the NO channel were 465 independent of preceding NO concentrations. In contrast for the type 1 blue light converter, instrumental background concentrations in the NO c channel increased with increasing NO concentrations and leveled off for high values as shown in Figure   8a. Measured NO concentrations during CAFE Africa were between 0 and 1 ppbv and were therefore situated in the rising part of the curve. That shows that instrumental background measurements during CAFE Africa were not only too high, but also depended on the preceding NO concentration. We tried to retrospectively correct the NO 2 data with a lower instrumental 470 background as obtained from laboratory investigations after several hours of zero air sampling. However, it was not possible to quantify the effect of varying, preceding NO levels. Additionally, the impact of humidity had the exact opposite effect on the NO 2 measurements. While the higher than actual instrumental background led to lower than actual NO 2 concentrations, increases in humidity triggered higher than actual NO 2 concentrations. Figure 8b shows that the development of the instrumental background in the NO c channel did not depend on the preceding NO concentrations for the type 2 quartz converter.

475
The instrumental background was 12 ± 1 pptv and constant over the whole experiment. The detected instrumental background While we have shown above that NO and humidity strongly affect the instrumental background measurements in the type 1 blue light converter, it is likely that there are other factors contributing to the observed effects, too. When switching off 480 the LEDs in the type 1 converter, the observed instrumental background decreased rapidly (too rapidly for a sole temperature effect) which we present in Figure S6a of the Supplement. This suggests that the light of the LEDs impacts the instrumental background in the NO c channel. Many other compounds can be photolyzed to form NO, such as PAN, ClNO 2 or BrONO 2 .
However, their absorption cross sections suggest no interference at 397 nm, the spectral output maximum of the LEDs (Reed et al., 2016;Pollack et al., 2010). Only small interference could occur with HONO and NO 3 at the edge of the spectral output 485 and this would require the presence of these compounds in the converter which should not be the case for the described laboratory investigations, but is conceivable given the memory effect observations. For comparison, Figure S6b shows that the effect of switching the LEDs on and off during zero air measurement is marginal when using the type 2 quartz converter. We have performed an uptake experiment for HNO 3 (nitric acid) to investigate the adsorptive capacity of the converters. HNO 3 in zero air (2500 sccm) was first routed through a bypass and after reaching a constant signal, the gas flow was changed to 490 include the converter. The HNO 3 concentration behind the converter was monitored via chemical ionization mass spectrometry (CIMS). Figure S7a and S7b of the Supplement show the resulting adsorption behavior for the type 1 and the type 2 converter, respectively. When directing the gas flow through the type 1 converter, the detected HNO 3 flux decreased rapidly by around a factor of four and we did not observe the signal return to its initial value within 40 minutes (which is when we terminated the experiment). This indicates a high absorptive capacity or a decay of HNO 3 in the converter (or both). Integration of the HNO 3 495 flux shows that the converter took up approximately 1.7×10 16 HNO 3 molecules in the considered time frame. In contrast for the type 2 converter, the HNO 3 flux decreased, too, but returned to its initial value within 10 minutes while it adsorbed only ∼1.7×10 15 HNO 3 molecules. The observed adsorption capacity can be minimized by coating the quartz surface with FEP (fluorinated ethylene propylene) which provides a highly hydrophobic surface (Neuman et al., 1999;Liebmann et al., 2017).
The number of adsorbed HNO 3 molecules was by a factor of 4 -5 smaller compared to the non-coated quartz converter ( Figure   500 S7c). We did not observe any differences between the coated and the non-coated quartz converter regarding the experiments investigating the role of NO and humidity presented above. This uptake experiment shows the high adsorptive capacity of the type 1 blue light converter in comparison to the type 2 quartz converter. In the case of HNO 3 , which could have been adsorbed by the type 1 converter e.g. during stratospheric measurements, we hypothesize a potential source of NO or NO 2 through for example surface-catalyzed chemistry, possibly involving the light of the LEDs or elevated temperature. Again, this experiment 505 underlines the superiority of the type 2 quartz converter over the type 1 blue light converter and suggests the applicability for ambient and airborne measurements.

Conclusions
In this study, we have investigated a modified conventional blue light converter (type 1) with a highly reflective and porous inner surface made from optical PTFE regarding its application during the research aircraft campaign CAFE Africa which took 510 place in August and September 2018 around Cabo Verde. We have identified a memory effect in the blue light converter which is affected by humidity, especially by rapid changes in water vapor concentrations, as well as preceding NO levels, which is particularly relevant for the low NO 2 /NO ratio in the upper troposphere. More specifically, this includes the subtraction of a fluctuating, higher than actual instrumental background in the NO c channel yielding negative NO 2 values as well as humidity triggered spontaneous desorption of stored molecules appearing as large NO 2 peaks, both of which effects are amplified by 515 the low conversion efficiency. The high adsorptive capacity regarding other atmospheric trace gases such as HNO 3 and the light of the LEDs could additionally play a role in the observed effects. Because of the complex correlations between these parameters it is not possible to retrospectively correct the NO 2 signal measured during CAFE Africa to receive reliable data. If a conventional blue light converter is still in use, we would suggest to avoid constant altitude changes in aircraft applications.
Instead, we highly recommend the application of an alternative photolytic converter made from quartz glass (type 2) in order 520 to prevent the gas flow from contact with the porous PTFE surface which can additionally be coated with FEP to obtain highly hydrophobic properties. Laboratory results indicate a high suitability of the alternative converter in aircraft measurements which -looking into the future -should be investigated in detail in order to improve in-field NO 2 measurements. With an improved instrumental background, other important questions of current atmospheric NO 2 research such as deviations from photostationary state NO 2 in remote locations or interferences from NO y species could be addressed and investigated more 525 easily.
Africa. Photolysis frequencies were received from BB. DM, MM, RR and HH measured and provided the OH and HO2 data. KP ad FK mesaured and provided the NO2 miniDOAS data. FO measured and provided the O3 data. H2O data were received from MZ. RD and JNC measured and provided the PAN data. HF, JL and HH had a large contribution in the operation and planning of the aircraft campaign.
Competing interests. Hartwig Harder is a member of the editorial board of the journal.