Experiments were performed using chamber A of the two 200 m3 simulation chambers of the European PHOtoREactor (EUPHORE) facility in Valencia, Spain, to provide a common, homogeneous air volume for multiple NOx analysers to sample from. The EUPHORE chambers are formed from fluorine-ethene-propene (FEP) Teflon foil fitted with housings that exclude ambient light (Wiesen, 2001; Munoz et al., 2011). The chambers are fitted with large horizontal and vertical fans to ensure rapid mixing (timescale of 3 min). Instrumentation used comprises long path Fourier transform infrared spectroscopy (FTIR; for absolute and specific alkene/VOC measurements), monitors for temperature, pressure, humidity (dew point hygrometer), ozone (UV absorption) and CO (infrared absorption). NOx levels were measured using four independent chemiluminescent monitors plus (in the case of NO2) LP-DOAS absorption spectroscopy. All monitor sampling lines were of similar lengths and attached to one inlet sampling from the centre of the chamber.

Monitors 1 and 2 employed heated Mo catalysts, while 3 and 4 used photolytic NO2 converters (see Table 1). All NOx monitors were calibrated (in the range 0–100 ppb) at the start and end of the 2-week measurement period, using a multipoint calibration derived from a primary NO standard (BOC 5 ppm alpha standard; certified to the National Physics Laboratory (NPL) scale) in addition to single-point calibrations performed on a daily basis. NO2 calibration was achieved via gas-phase titration, using added ozone within the chamber. In some experiments, the calibrations and interference were confirmed with use of the EUPHORE LP-DOAS system to unequivocally identify and quantify NO2.

All experiments were performed with the chamber housing closed (i.e. dark conditions; j(NO2) < 2 ×10-6 s-1) at near-atmospheric pressure and ambient temperature. For most experiments, humidity was low (dew point ca. -45 ∘C). The experimental procedure, starting with a clean flushed chamber, was to add SF6 (as a dilution tracer) followed by successive aliquots of various alkenes and, in certain cases, additional species (H2O and CO), whilst recording the measured NO and NO2 levels over periods of 1–3 h. For some systems, ozone was added at the end of the experiment; under such dark, high-O3 conditions, we can be confident that negligible NO could actually be present in the chamber (e.g. from wall sources) and, hence, that any NO signal observed by the monitors was unequivocally an interference response (as any NO remaining would be rapidly consumed by reaction with O3). The potential interferant species investigated were cis-2-butene (C2B), trans-2-butene (T2B), tetra-methyl ethylene (2,3-dimethyl-butene or TME), α-terpinene, limonene, methyl chavicol (estragole) and terpinolene, with four to five additions of 20–50 ppb in each case, together with single- or dual-point interference measurements for ethene, propene, isobutene, isoprene, α-pinene, β-pinene and myrcene. Repeat experiments were performed for T2B, terpinolene and α-terpinene under conditions of increased humidity (up to ca. 30 % relative humidity – RH). Alkene mixing ratios introduced into the chamber are given in Table S1. Propene, C2B and T2B were supplied by Linde plc. (purity > 99 %), isobutene (purity > 99 %) and terpinolene (purity > 85 %) were from Fluka Analytical and TME (purity > 98 %), isoprene (purity > 99 %), limonene (purity > 97 %), α-pinene (purity > 97 %), β-pinene (purity > 97 %), α-terpinene (purity > 85 %), estragole (purity > 98 %) and myrcene (purity > 99 %) were from Sigma-Aldrich. All reagents were used as supplied.

The limit of detection (LOD) for each instrument was determined, under the actual experimental conditions, as three times the standard deviation of the NO and NO2 signal recorded each day from the empty chamber prior to the start of experiments (i.e. before addition of any reactants). The mean LODs determined for NO and NO2 are shown in Table 1. These LOD values are higher than those quoted by the manufacturers for monitors 1–4 (typically 2–100 ppt) but accurately reflect the actual performance of the instruments as used during these experiments. In the analysis which follows, in order to confirm that any change in the measured NO and NO2 mixing ratio for each alkene addition was not due to noise or drift and therefore came from signal, the readings were compared to the experimentally determined LOD for each instrument. Only in cases where the measured change was greater than the experimentally determined LOD were these readings used for determining an interference. The interference due to the VOC was determined by means of linear regression (least squares fit), with the slopes and their uncertainty and Pearson's correlation coefficients calculated in IGOR (see Tables 2 and 3).

Positive NO interferences were observed for those alkenes which reacted most rapidly with ozone and, hence, will be present within the monitor reaction chamber at different levels in the measurement and background modes. This interference is attributed to chemiluminescent emission following the alkene–ozone reaction and may be attributed to a combination of two factors, namely the formation of excited products in the alkene–ozone reaction, which emit chemiluminescence, coupled with the significant removal of some alkenes during the instrument background phase compared with the measurement phase through their reaction with (elevated levels of) ozone within the instrument, i.e. mechanism (4), outlined above.

Possible origins of this signal are the production of excited HCHO, vibrationally excited OH and electronically excited OH (e.g. Finlayson et al., 1974). While the long-pass filters used in chemiluminescence NOx monitors should preclude emissions from electronically excited species, vibrationally excited OH produced through the hydroperoxide mechanism is known to emit in the 700–1100 nm wavelength range (Finlayson et al., 1974; Schurath et al., 1976; Hansen et al., 1977; Toby, 1984) and would be detected as NO2. Specifications of the long-pass filters used in the chemiluminescence NOx monitors in this study are not reported in their respective user manuals but typically block light below ca. 600 nm, while typical PMT response characteristics are between 400 and 950 nm (Jernigan, 2001). Any chemiluminescence signal in the 600–950 nm wavelength range can therefore cause a potential interference.

The difference in the interference effect among monitors may then reflect differences in the conditions (e.g. ozone abundance, pressure and residence time) within the reaction cell and filter specifications. The relative magnitudes of the positive interference signals observed between the different monitors are consistent with this picture as the reaction chamber pressure is much lower for monitors 3 and 4 (ca. 1–10 Torr) compared with monitors 1 and 2 (ca. 300 Torr), leading to greater collisional quenching. Similarly, the addition of H2O, which would be expected to efficiently accept vibrational energy from OH radicals (Gerboles et al., 2003), was found to substantially reduce the apparent interference. In the experiments with higher humidity, a reduced interference (factor of ca. 2; see Table 2) was observed for all NO experiments for all instruments, except for TME for the photolytic converters for which an increase was observed. There is currently no recommended relative humidity under which calibrations should be performed for any of the instruments or within EU and EPA guidelines (AQEG, 2004; USEPA, 2002). However, the installation of permeation driers at the sample inlet should (in principle) reduce the impact of different H2O/relative humidity levels upon the quenching of NO2 or other species and are a common feature of most modern samplers (AQEG, 2004).

The most significant effects are the large positive NO interferences observed for the monoterpenes, namely α-terpinene and terpinolene, within monitors 1, 3 and 4. The criteria for an alkene to display such a positive interference (i.e. via mechanism 4) are that it reacts with ozone to produce excited products which exhibit a chemiluminescent signal at appropriate wavelengths. In addition, the alkene must have a sufficiently rapid reaction with ozone so that its mixing ratio is substantially reduced during the instrument background phase compared with the measurement phase, thus precluding the correct subtraction of the interference signal. The reaction rate constants for many alkenes with ozone are well known, allowing the calculation of a kinetic interference potential (KIP) ranking for this second factor, and are calculated by Eq. (1). 1KIP=100×1-exp⁡-k′t×k(Alkene+O3)k(NO+O3), where k=k[O3]t and [NO][NO]0=0.01 (i.e. 1 % of NO left after reaction with excess O3; see the Supplement for calculation details). The calculated KIP are shown in Table 4 as the percentage of a given alkene's potential chemiluminescent signal which would not be subtracted in the standard background cycle, under the assumption that the background cycle conditions (O3 mixing ratio and residence time) would be sufficient to remove 99 % of the NO present.

This ranking does not reflect the precise (relative) interference which is observed as it neglects structural features which will affect the yield (and state i.e. electronic or vibrationally excited) of the chemiluminescent products from the ozonolysis reaction, but it is consistent with the trend and relative magnitudes for the substantial positive interferences shown in Tables 2 and 3. For example, a lack of interference is observed for myrcene and limonene, both of which exhibit terminal C=C bonds (see Table 4) and, after reaction with ozone, lead to the production of the CH2OO Criegee intermediate (CI), which subsequently decomposes or undergoes rearrangement to form small yields of OH (Alam et al., 2011). The ozonolysis of internal alkenes, such as C2B and T2B, produce the CH3CHOO CI, which predominantly decomposes via the vinyl hydroperoxide mechanism, forming larger yields of OH (Johnson and Marston, 2008; Alam et al., 2013). Such chemically formed OH that produces a detectable signal may also be augmented by contributions from HO2 and RO2 converted into OH within the instrument by reaction with NO, especially in the NO2 channel of photolytic converter instruments.

The relationship between the KIP (Table 4) and measured NO interference (Tables 2 and 3) is illustrated in Fig. 5 and can be used for predicting the potential interference of a given alkene to the NO signal from a kinetic perspective. For example, α-humulene has a KIP of 94.54 %, which could give rise to a 1.7 %, 2.4 % or 10.2 % NO interference for monitors 1, 3 and 4, respectively. This estimate is, however, based on the rate constant of α-humulene alone and does not include any structural features such as the presence of terminal and non-terminal C=C bonds.

The above discussion considers only the interference effect arising from alkene chemiluminescent emission; further measurement impacts are also evident in the (negative) interferences apparent for other species/monitors in Tables 2 and 3. An inspection of Tables 2 and 3 shows smaller positive interferences, and some negative interferences, from alkenes in the NO2 measurements.

NO2 measurements using chemiluminescence approaches are usually obtained by measuring NOx (i.e. Σ(NO + NO2), after passing the sampled air through an NO2 converter) and subtracting the (independently determined) NO contribution. If the actual interference signal (additional chemiluminescence) during the NOx measurement mode arises solely from mechanism (4), ozonolysis chemiluminescence, then this would be expected to match that in the NO mode (subject to the alkene abundance not being altered in the NO2 conversion stage and the detection conditions for the NO and NOx phases being identical) and, consequently, would not affect the retrieved NO2 mixing ratio. Monitors 1, 2 and 3 used a single detection cell, alternating between NO and NO2 (NOx) modes, and measured the NO2∗ chemiluminescence signal under identical conditions (optical arrangement, filtering and pressure). The observed negative interference for NO2, therefore, may have arisen due to the removal of alkene by the Mo catalyst within the monitors.

For monitor 1 (TE42i-TL), the negative interference observed for NO2 was the same magnitude as that observed for the positive interference for NO, including the experiments with H2O and CO (see Fig. 6 and Tables 2–3). This response is thought to arise as a consequence of the calculation methodology, combined with removal of alkenes during the NO2 conversion by the Mo catalyst.

There are three modes of operation in monitor 1 (TE42i-TL), namely NO measurement, NO2 / NOx measurement and background (pre-reactor) measurement, given by Eqs. (2)–(4), respectively, as follows: 2sNO=sNOreal+Xi,3sNOx=sNOxreal+yXi,4sP=fXi, where sNO and sNOx are the NO and NOx signals produced by the chemiluminescence monitor, respectively, sNOreal and sNOxreal are the real NO and NOx signals, Xi denotes the interference alkene i, y is the fraction of the interferant (alkene) Xi remaining after the Mo convertor, sP denotes signal at the pre-reactor, and f is the fraction of Xi remaining after the pre-reactor. The mixing ratios of NO, NO2 and NOx are given by the following: 5[NO]=sNO-sPcNO,6[NO]=sNOreal+Xi-fXicNO,7[NO]=sNOreal+(1-f)XicNO,8[NOx]=sNOx-sPcNOx,9[NOx]=sNOxreal+yXi-fXicNOx,10[NOx]=sNOxreal+(y-f)XicNOx,11[NO2]=NOx-NOCE,12[NO2]=sNOxreal+(y-f)XicNOx×CE,-sNOreal+(1-f)XicNO×CE, where c is the span factor and CE represents the conversion efficiency. If we assume cNOx≈cNO≈c, then the following applies: 13[NO2]=sNOxreal+(y-f)Xi-sNOreal+(1-f)Xic×CE. From Eq. (13), it may be seen that if y=1 (i.e. if the interferant alkene abundance is not affected by passage through the Mo converter), then there would be no interference observed in the retrieved NO2, while if the interferant species is subject to removal during its passage through the converter, then y < 1 and a negative interference would be observed. Molybdenum oxide catalysts have been reported to efficiently isomerise alkenes at temperatures between 300 and 400 ∘C (Wehrer et al., 2003) and are also effective catalysts for the epoxidation of alkenes (Shen et al., 2019). The observed small negative interference effects (for monitors 1 and 2; the Mo converter units), in the absence of significant sampled NOx, may reflect the partial removal of the alkene in the converter.

The negative NO2 interference apparent for monitors 3 and 4 (photolytic converter instruments) is more difficult to rationalise (as no Mo catalyst is present). Under ambient conditions, where NOx is present, mechanism (3) may occur, as outlined below. In reality, the conversion efficiency for photolytic converters is substantially lower than 100 % (Reed et al., 2016) as a consequence of both the finite photolysis intensity achievable and the occurrence of the NO+O3 back reaction. If the instrument calibration factor for NOx is not equal to that for NO (see Eq. 12), or if alkene was removed in the convertor stage, then this will lead to different interferences for NO and NO2, as CE is also (significantly) less than 1. This trend is apparent in the values shown in Table 3, in particular for the instruments fitted with photolytic convertors. However, in the absence of sampled NOx, the observed less positive or even negative NO2 interference suggests that less alkene is present in the NOx mode. Direct photolysis of alkenes is unlikely to cause such a change, considering the photolytic converter wavelength envelope, but the photolytic production of HOx radicals (which then react with the alkene) may be responsible.

Monitor 4 (Air Quality Design, Inc. – AQD) used independent NO2∗ detection channels; tests were conducted using both channels for C2B and terpinolene systems and revealed significant differences between the two detectors (ca. 40 % lower interference response for NO in the NO2 detection channel). With two independent detection channels, NO2 may be determined from the NOx measurement by either subtracting the NO level obtained from the NO channel (method a) or via the difference in signal observed in the NO2 / NOx channel when turning the photolysis lamp on and off (method b). Under method (a), as employed for C2B and terpinolene, a lower positive interference from alkene chemiluminescence results as a consequence of the difference in the detection cell conditions (results marked ∗ in Table 3), while under method (b), as employed for the other alkenes studied here with the AQD system, the interference (from mechanism 4 alone) should cancel out (results marked with a # in Table 3).

The data presented in Figs. 2–4 and Tables 2 and 3 show both negative and positive interferences, while mechanism 4 alone would be expected to result in positive interference signals for NO for all alkenes. We therefore conclude that additional mechanisms are occurring. Under the conditions of these chamber experiments, the retrieval of additional NOy species can be precluded. The chamber wall source of HONO has been characterised and shown to produce ppt levels of HONO under the dark, dry conditions of these experiments (Zádor et al., 2005) and would be equally present for all experiments. We attribute the negative (or reduced positive) interference effects to a combination of mechanisms (1) and (3), where the quenching of excited OH (produced by alkene and ozone reaction) by alkenes (electron-rich alkenes have been shown to be effective quenchers; Gersdorf et al., 1987; Chang and Schuster, 1987) and the generation of HOx radicals within the instrument follow on from the ozonolysis reaction.

The alkene–ozone reactions are known to produce OH, HO2 and RO2 radicals both directly (e.g. Johnson and Marston, 2008), following the photolysis of other alkene–ozone reaction products (e.g. carbonyl compounds), and through OH–alkene reactions. Peroxy radicals promote the conversion of NO to NO2, altering the abundance of both species (the formation of NOx reservoirs, such as nitric acid and organic nitrates, will also occur but will be negligible on the timescale of the operation of most instruments).

The ozonolysis of the TME results in the production of OH with a close to unity yield (IUPAC, 2020), and if taking into account the above mechanism (4) only, might be expected to exhibit a large interference in NO mode. Table 2 shows no interference for monitors 1 and 2 (Mo convertor units) and negative and positive interferences for monitors 3 and 4 (photolytic convertor units), respectively, and so it is hard to rationalise (for NO mode). The addition of CO as a scavenger for OH led to an increase in the NO signal for all monitors. A possible origin of this signal is the chemiluminescence production of the excited intermediate HOCO (from the reaction of vibrationally excited OH, which is from the ozonolysis of TME, with CO), which has a temperature and pressure-dependent rate of reaction (Atkinson et al., 2006; Li and Francisco, 2000) and is consistent with the larger NO signal in the photolytic monitors (Table 2).