While the use of a catalytic converter or a dynamic humidification of zero air helped to reduce differences in RH between C2 and C3, small differences were still observed. Since the concentration of OH inside the reactor is driven by water photolysis, a small difference in RH can lead to significantly different OH levels between C2 and C3, and as a consequence to an artifact in the C2 measurement (Sinha et al., 2008). Therefore, a correction is directly applied to the pyrrole concentrations measured during C2 as proposed in Dolgorouky et al. (2012). To assess this correction, experimental determinations of the C2 sensitivity to humidity were performed measuring C2 at various RH before, during, and after field campaigns. These tests were made by introducing various flow rates of dry zero air (from 50 to 300 sccm) inside the sampling line. The dilution of humid ambient air with dry zero air allowed RH to be altered over a large range (typically 20–60 %, Fig. 2).

To track relative humidity during these experiments and during ambient measurements of OH reactivity, we use the ratio between m/z 37 (cluster ion H3O+⚫H2O) and m/z 19 (H3O+) monitored by PTR-ToFMS. Indeed, H3O+ ions can cluster in the drift tube of PTR-MS instruments (de Gouw and Warneke, 2007) to form water clusters (H3O+(H2O)n) whose levels depend on relative humidity inside the PTR-MS reactor. A linear relationship was found between RH and the m/z 37-to-m/z 19 ratio (referred to as m37/m19 ratio in the following) during laboratory tests (not shown).

As mentioned above, OH radicals are artificially generated in the sampling reactor from water photolysis using a mercury lamp. A drawback of this method is the formation of a similar amount of HO2 radicals, since hydrogen atoms are formed in the water photolysis process, which then quickly react with oxygen to form HO2. When ambient NO is sampled inside the reactor, these HO2 radicals can be rapidly converted into OH radicals. This secondary formation of OH leads to differences in OH levels between C2 and C3, and therefore to an artifact in the C3 measurement. To assess the required correction for C3 values, different amounts of NO (from 6 to 120 ppb) were introduced inside the reactor while sampling humid zero air. Similar experiments were conducted by adding ethane and isoprene together with NO in the sampling reactor, leading to VOC-induced reactivity values of 22.2 and 36.6 s-1, respectively. These experiments have been conducted at different apparent pyrrole-to-OH ratios ranging from 1.6 to 3.9 and determined by Eq. (2). This ratio is used to gauge the kinetic regime in the CRM reactor (Sinha et al., 2008). For simplicity, this apparent pyrrole-to-OH ratio (Supplement S3) is referred to as the pyrrole-to-OH ratio in the following. In practice, this ratio is adjusted by changing RH in the reactor, which in turn leads to a change in OH levels. pyrroleOH=C1C1-C2 While NO2 is not expected to lead to the formation of secondary OH inside the reactor, its conversion into NO through photolysis or other chemical processes can cause an artifact. To test the effect of NO2 on C3 measurements, we followed the same procedure as described above for NO. Different amounts of NO2 (from 60 to 410 ppb) were introduced inside the CRM reactor when sampling humid zero air at pyrrole-to-OH ratios ranging from 1.6 to 3.2.

Measured OH reactivity values are calculated using Eq. (1). In this equation, pseudo-first-order conditions are assumed for pyrrole; i.e., pyrrole concentrations are at least several times higher than OH concentrations. However, operating conditions used for CRM instruments do not comply with this assumption, and the calculated values (Eq. 1) need to be corrected. To assess this correction, several gas standards (ethane, ethene, propane, propene, and isoprene) were introduced inside the CRM reactor at different concentrations. This allows comparing the calculated OH reactivity generated by the standards (reactivity ranging from 6.5 to 65 s-1) to the OH reactivity measured using Eq. (1). These experiments have been conducted at pyrrole-to-OH ratios ranging from 1.4 to 2.6.

Figure 3 shows experiments conducted to quantify the C3 dependence on NO due to the spurious formation of OH from HO2+ NO. These experiments were made at different pyrrole-to-OH ratios by sampling humid zero air. A total of four laboratory experiments (three shown in Fig. 3) were conducted at pyrrole-to-OH ratios ranging from 1.6 to 3.9, covering the typical range of ratios observed during ambient measurements (generally from 1.6 to 2.2).

The variation of C3 (ΔC3) is computed as the difference between an expected C3 and the measured C3, since a decrease in the pyrrole mixing ratio is observed when NO increases. The expected C3 is calculated using measured levels of C1 and C2, and an expected OH reactivity due to NO (from 1.1 to 30.9 s-1). Knowing these three terms, one can calculate the expected C3 from Eq. (1). To indicate the level of correction brought to the OH reactivity measurements, the absolute change in total OH reactivity for the experiment conducted at a pyrrole-to-OH ratio of 2.2 is also given in Fig. 3 (right axis).

As expected, a decrease of the pyrrole mixing ratio is observed when NO is introduced inside the reactor at a constant pyrrole-to-OH ratio. However, the variation of C3 with NO is not linear and varies with the pyrrole-to-OH ratio. Indeed, the difference between the expected and measured C3 indicates a plateau at high NO mixing ratios. In addition, the amplitude of this NO artifact increases with the pyrrole-to-OH ratio, i.e., with decreasing OH concentrations in the reactor, C1 being kept constant for all experiments.

A quadratic regression forced through the origin was applied to fit the observations (solid lines in Fig. 3). It is interesting to note that the parameters for the quadratic regression vary linearly with the pyrrole-to-OH ratios (see middle and bottom panels in Fig. 3). It is thus possible to interpolate the parameters from the quadratic regression to the pyrrole-to-OH ratios observed during field measurements to calculate the correction to apply to C3 (Eqs. 4–7): C3corrected=C3measured+ΔC3,withΔC3=aNO2+bNO,witha=a1pyrroleOH+a2andb=b1pyrroleOH+b2,soΔC3=a1pyrroleOH+a2NO2+b1pyrroleOH+b2NO. The experiment performed using dry zero air (pyrrole-to-OH ratio of 3.9) is not taken into account in the linear regressions displayed in the bottom panels of Fig. 3 because a deviation from the linearity is observed. Not considering this point is acceptable since completely dry conditions are never observed during ambient measurements, for which pyrrole-to-OH ratios are always lower than 2.6. These experiments were also carried out by adding different gas standards (ethane and isoprene) inside the reactor at the same time as NO. The gas standard additions were adjusted to get OH reactivity values from VOCs ranging from 22.2 to 36.6 s-1. These experiments and the discussion on the effect of adding a VOC are given in the Supplement (Fig. S2 in the Supplement). Briefly, no clear impact was found on the NO artifact, suggesting that the correction characterized above is suitable for ambient measurements. This correction can be applied during field measurements using pyrrole-to-OH ratios that are continuously monitored by the CRM instrument and measured ambient mixing ratios of NO.

Figure 4 (top panel) displays the changes in C3 with NO2 mixing ratios inside the reactor for three different pyrrole-to-OH ratios. As for NO, the introduction of NO2 in the reactor leads to a decrease of the pyrrole mixing ratio. ΔC3 also appears to be nonlinear, with NO2, exhibiting a plateau at mixing ratios higher than approximately 150 ppb. However, no clear difference is observed for experiments conducted at various pyrrole-to-OH ratios.

The artifact caused by NO2 may be due to its conversion into NO before or within the reactor. To determine the fraction of NO2 converted into NO, we calculated the amount of NO needed (based on the experiments presented in Sect. 5.2.1) to explain the changes observed in C3 when NO2 was introduced inside the reactor. Based on the entire set of NO2 experiments, these calculations led to a conversion ranging from 16 to 37 %, with an average value of 24 % (±9 %, 1σ) (Fig. 4 bottom panel).

Further work was performed to study the conversion of NO2 inside the MD-CRM instrument. A NOx analyzer (Thermo Environmental Instruments, model 42C) was connected to the reactor exhaust instead of the PTR-MS, while NO2 was introduced at the reactor inlet. Since the sampling flow rate of the NOx analyzer was 600 sccm, 460 sccm of zero air was added in the sampling line of the NOx analyzer to only sample 140 sccm from the reactor, similar to the sampling flow rate from the PTR-MS instrument. Large mixing ratios of NO were observed at the exit of the reactor (between 25 and 30 % of total NOx) when the mercury lamp was off, while low NO mixing ratios (∼ 8.7 % of total NOx) were observed at the reactor inlet. This result indicates that NO2 is not converted into NO by photolysis but rather by heterogeneous chemical processes, probably on stainless-steel pieces upstream and downstream of the glass reactor. The replacement of all the stainless-steel pieces in the setup is planned in the future to avoid, or at least to reduce, this NO2 conversion.

Using a similar approach to that for NO, the correction to be applied in C3 for NO2 can be calculated using a quadratic regression, shown in Fig. 4, independently of the pyrrole-to-OH ratio and using the measurements of ambient NO2. It is worth noting that the amplitude of the correction is significantly lower for NO2 compared to NO.

Box model simulations were compared to experimental observations discussed above. It is worth noting that pyrrole-to-OH ratios reported for the simulations were calculated using the same approach as during laboratory experiments, i.e., using Eq. (2). As already mentioned, the calculations do not lead to the real pyrrole-to-OH ratios (calculated from concentrations of pyrrole and OH used to initialize the model) but to apparent ratios since C1–C2 is not the total OH mixing ratio inside the reactor but the amount of OH reacting with pyrrole. Since the mechanisms include self- and cross-reactions of radicals, all the OH introduced in the model does not react with pyrrole, and true pyrrole-to-OH ratios are lower than the measured apparent ratios. A comparison of real and apparent ratios is given in the Supplement (Fig. S3). For instance, an apparent ratio of 2 corresponds to a real ratio of approximately 1 for simulations conducted under dry conditions with the simple mechanism and initial mixing ratios of OH and HO2 set at the same value in the model.

Initial OH mixing ratios were set in the simulations to reproduce apparent pyrrole-to-OH ratios observed during laboratory experiments. Real mixing ratios of OH inside the reactor were also determined experimentally by introducing a large amount of isoprene (3 ppmv) in the presence or absence of OH. OH mixing ratios were calculated from the consumption of isoprene and compared to levels set in the model to reproduce laboratory observations (see Supplement, Fig. S4). It was found that OH mixing ratios set in the model agree within uncertainties with experimental determinations, indicating that initial conditions used in the model are representative of the real OH levels inside the reactor.

Experimental results related to the NO artifact, as well as simulations performed using the two mechanisms described in Sect. 3 (simple mechanism and MCM), are displayed in Fig. 5.

Both models predict a change in C3 that is similar to laboratory observations when NO increases inside the reactor. Indeed, the models predict that ΔC3 first increases with NO and levels off after the addition of a certain amount (> 90 ppbv). This behavior indicates that all HO2 radicals are titrated when a threshold of NO is reached and that further addition of NO does not cause any additional formation of OH through the NO + HO2 reaction. Furthermore, both models also predict that ΔC3 becomes larger at higher pyrrole-to-OH ratios, similar to experimental observations. A potential reason for this behavior is that the concentrations of both OH and HO2 are lower at higher pyrrole-to-OH ratios, since the pyrrole concentration is held constant in all experiments and simulations (C1 = 55 ppb). Lower concentrations of radicals lead to a slower reaction rate between OH and HO2. As a consequence, OH radicals formed from NO + HO2, even in smaller quantities, will preferentially react with pyrrole rather than HO2, leading to a larger change in ΔC3.

Significant differences are found between the simulations conducted using the two mechanisms. Indeed, simulations performed using the simple mechanism lead to an overestimation of the NO interference by up to 27 %, while simulations performed using MCM lead to an underestimation of 10 % at most. These differences lie in the way the chemistry of organic peroxy radicals is treated. In the simple mechanism, each reaction of OH with an organic compound gives the same RO2 radical, which propagates to HO2 after reaction with NO and O2 without any other byproduct. In MCM, a more complex chemistry is included since specific peroxy radicals are formed for each reacting organic molecule and since closed-shell byproducts are generated from the peroxy radical reactions, which can further react with OH. Since, the simulations from the two mechanisms encompass the experimental results, one can conclude that the lack of speciation for RO2 radicals and of secondary chemistry in the simple mechanism does not allow for correct reproduction of the laboratory observations, while the secondary chemistry included in the MCM, in particular the proxy used to account for the pyrrole chemistry (BZFUONE), is not fully representative of the chemistry occurring inside the CRM reactor.

It is interesting to note that both mechanisms lead to coefficients (a and b) of the quadratic regressions (ΔC3 vs. pyrrole-to-OH) similar to those observed for the laboratory experiments (see bottom panels of Fig. 5).

Since all the simulations described above were conducted under dry conditions, the influence of humidity was tested by repeating the same simulations at a relative humidity of 100 % (see Fig. S5). A decrease of ΔC3 of less than 10 % was observed at all pyrrole-to-OH ratios. This trend can be due to a water enhancement of the HO2 self-reaction rate, reducing the secondary formation of OH, and hence the NO artifact. In practice, the pyrrole-to-OH ratios are directly linked to relative humidity inside the reactor since OH levels depend on the amount of water available for photolysis. Simulation results that have to be compared to experimental results are between these two extreme cases (dry and RH saturated), being closer to dry-condition results at higher pyrrole-to-OH ratios and vice versa. However, only small differences are observed between dry and RH-saturated conditions, and the simulations made under dry conditions are suitable for this comparison.

The effect of adding gas standards (isoprene and ethane) in the simulations has also been investigated and is displayed in Fig. S6. Simulations made using MCM suggest a small dependence of ΔC3 (∼ 6.1 ± 1.1 %) on OH reactivity for values in the range 20–40 s-1, especially at high NO concentrations (NO > 100 ppb). This difference is small and is within measurement uncertainties.

Simulations presented above were performed assuming no O3 in the reactor. However, photolysis of O2 may occur inside the reactor due to the UV lamp and may lead to a significant O3 concentration. The influence of O3 on simulated NO artifacts was tested using an initial O3 mixing ratio of 200 ppb with the MCM (see Fig. S12.2). The 200 ppb of ozone corresponds to the value measured at the exhaust of the reactor under dry conditions, using an ozone analyzer (Environment S.A., model O3-42M). Adding ozone in the simulations only leads to a small decrease of approximately 3 % for the NO artifact, independent of the pyrrole-to-OH ratio. It is interesting to note that the presence of hundreds of parts per billion (ppb) of ozone in the reactor might also lead to additional production of OH through ozone photolysis, producing O(1D), which then quickly reacts with H2O to form two OH radicals. Therefore, OH radicals present in the reactor may come not only from H2O photolysis but also from O3 photolysis.

Figure 6 displays experimental observations of the measurement bias caused by not operating the instrument under pseudo-first-order conditions. This figure compares OH reactivity values calculated from the addition of a gas standard to values measured by the MD-CRM instrument. The measurements were derived using Eq. ( 1) and corrected for changes in humidity between C2 and C3 (see Sect. 5.1). NOx species were not added in these experiments.

The top panel of Fig. 6 shows results from the addition of three different gas standards (isoprene, ethane, and propene), characterized by OH rate constants spanning almost three orders of magnitude (1.0 × 10-10, 2.4 × 10-13, and 2.9 × 10-11 cm3 molecules-1 s-1, respectively), at a pyrrole-to-OH ratio of 1.4. This figure indicates a linear relationship between the measured and the calculated OH reactivity values. The slope F of a linear regression represents the correction factor that has to be applied to the measured OH reactivity values (see Eq. 8): ROHtrue=F∗ROHmeasured, where ROHtrue is the calculated total OH reactivity, based on the concentrations and OH reaction rate constants of the gas standards, and ROHmeasured is the OH reactivity measured by the MD-CRM instrument, corrected for changes in relative humidity between C2 and C3. For ambient measurements, ROHtrue will be the measured OH reactivity corrected for not operating the instrument under pseudo-first-order conditions.

The correction factors determined at a pyrrole-to-OH ratio of 1.4 indicate that the reactivity of the gas standard plays a role in the observed bias. Higher correction factors are found for more reactive compounds. Indeed, at a pyrrole-to-OH ratio of 1.4, the correction factor determined for isoprene is 7.6 % higher than for propene and is 50.6 % higher than for ethane. However, the middle panel of Fig. 6 indicates that the differences observed between the different gas standards are lower at a pyrrole-to-OH ratio of 2.3, with the correction factor derived for isoprene being only 1 % different than for ethane.

Experiments performed at various pyrrole-to-OH ratios (four ratios: 1.4, 1.6, 1.8, 2.3; not shown) indicate that the relative difference between the correction factors determined using ethane and isoprene range from 1 to 58 % and confirm the strong dependence on the pyrrole-to-OH ratio, with negligible differences for ratios higher than 2.3. These two gas standards represent extreme cases since ethane is one of the least reactive VOCs in the atmosphere (reaction rate constant with OH: 2.4 × 10-13 cm3 molecules-1 s-1), while isoprene is one of the most reactive (reaction rate constant with OH: 1.0 × 10-10 cm3 molecules-1 s-1). It is interesting to note that ambient air is a mixture of a large number of compounds with reaction rate constants ranging between those of ethane and isoprene. The correction factor should therefore be calculated as an averaged correction factor determined using a VOC exhibiting a slow rate constant with OH such as ethane and a VOC exhibiting a fast rate constant such as isoprene. The uncertainty in the averaged correction factor can then be estimated from the difference observed in F for these two VOCs.

Results from the addition of the same gas standard (isoprene) at three different pyrrole-to-OH ratios (1.4, 1.8, and 2.3) are shown in the bottom panel of Fig. 6. Correction factors derived from these experiments increase when the pyrrole-to-OH ratio decreases. Indeed, the highest correction factor (1.59) is found for the lowest pyrrole-to-OH ratio (i.e., 1.4), and the lowest correction factor (0.84) is found for the highest pyrrole-to-OH ratio (i.e., 2.3). A decrease of the correction factor with the pyrrole-to-OH ratio is consistent with a kinetic regime getting closer to pseudo-first-order conditions (OH≪ pyrrole) and therefore to correction factors closer to 1.

Results from the addition of other gas standards (ethene, k=8.5×10-12 cm3 molecules-1 s-1, and propane, k=1.09×10-12 cm3 molecules-1 s-1) are consistent with the results shown in Fig. 6 and confirmed the trends discussed above (Fig. S7).

From these experiments, it appears essential to take into account the pyrrole-to-OH ratio in the correction to be applied to the measured OH reactivity values. Figure 7 shows how the correction factor varies with the pyrrole-to-OH ratio. Correction factors shown in this figure were derived from single experiments using different standards (ethane, ethene, propane, propene, and isoprene) as shown in Fig. 6. This figure gathers laboratory and field experiments performed over 7 months. The gas standards were chosen to cover a large range of reaction rate constants with OH (from 2.4 × 10-13 cm3 molecules-1 s-1 for ethane to 1.0 × 10-10 cm3 molecules-1 s-1 for isoprene) to take into account the impact of the gas reactivity on the correction factor.

From the linear relationship observed in Fig. 7, the correction factor to be applied to the measurements can be calculated using Eq. (9) and the pyrrole-to-OH ratio that is monitored during field or laboratory measurements. F=-0.16Pyr/OH+1.27 Within the small range of pyrrole-to-OH ratios generally encountered during field campaigns, typically from 1.6 to 2.2, the correction factors determined using Eq. (9) range between 0.9 and 1.0. Therefore, an averaged value of the correction factors can be considered instead of using a pyrrole-to-OH dependent correction (Fig. 7). It is interesting to note that this correction takes into account the artifact caused by not operating the instrument under pseudo-first-order conditions as well as any unknown artifact that could impact the OH reactivity measurements, such as radical segregation and wall losses of radicals inside the reactor.

Simulations performed using both mechanisms described in Sect. 3 (the simple mechanism and MCM), as well as a mechanism similar to the one described by Sinha et al. (2008), which does not account for radical–radical reactions, are displayed in Fig. 8. The latter is referred to as the two-reaction mechanism in the following.

The top panel of Fig. 8 displays the results from the addition of three different gas standards (isoprene, ethane, and propene) using the MCM mechanism and a surrogate standard for the simple mechanism, characterized by different reaction rate constants with OH (1.0 × 10-10, 2.4 × 10-13, 2.9 × 10-11, and 5.0 × 10-12 cm3 molecules-1 s-1, respectively). These simulations were performed at a pyrrole-to-OH ratio of 1.4. It is interesting to note that simulations performed using MCM for a gas standard exhibiting an OH rate constant of 5.0 × 10-12 cm3 molecules-1 s-1 are similar to that observed with the simple mechanism, suggesting that the detailed secondary chemistry included in MCM is not important in the modeling of this correction. These results show that the correction factor should increase with the OH rate constant, independently of the mechanism used (the MCM or the simple mechanism). This trend is consistent with laboratory observations discussed above.

Simulations conducted with the simple mechanism and the two-reaction mechanism are displayed in the bottom panel of Fig. 8 for a surrogate gas standard (k=5.0 × 10-12 cm3 molecules-1 s-1) at three different pyrrole-to-OH ratios (1.4, 1.9, and 2.9). The correction factor derived with the simple mechanism decreases with increasing pyrrole-to-OH ratios as observed during the laboratory experiments. As mentioned previously, these observations are consistent with a chemical system getting closer to pseudo-first-order conditions (OH ≪ pyrrole) when the pyrrole-to-OH ratio increases. In contrast, an opposite trend is observed when the two-reaction mechanism is used. This result highlights the importance of accounting for radical–radical reactions in the mechanism to describe the complex chemistry occurring in the reactor. Both mechanisms, the simple mechanism and MCM, lead to correction factors that converge to unity when the pyrrole-to-OH ratio increases (i.e., OH decreases).

Figure 9 (top panel) shows how the correction factor changes with the pyrrole-to-OH ratio. Simulated values stem from simulations conducted using the box model including the MCM mechanism and constrained with ethane or isoprene under dry conditions, as well as under wet conditions (RH = 100 %) for ethane. The simulated correction factors for ethane under dry conditions are higher than the measurements by 72 and 54 % at pyrrole-to-OH ratios of 1.4 and 1.9, respectively. These differences are even larger for isoprene (143 and 80 % at pyrrole-to-OH ratios of 1.4 and 1.9, respectively). However, a similar trend is observed: a decrease of the correction factors with increasing pyrrole-to-OH ratios. Performing simulations at 100 % of relative humidity improves the agreement but still fails short of reconciling simulations and measurements. Furthermore, a RH of 100 % is not likely inside the reactor, and the real conditions are between these two extreme cases.

Several hypotheses (segregation between the reactants, RO2+ OH reactions (Fittschen et al., 2014), uncertainties in reaction rate constants of radical–radical reactions, higher or lower proportions of HO2 compared to OH, formation of O3 inside the reactor) were tested in the simulations to try to reconcile simulated results and laboratory observations (see Supplement S8 to S13). Unfortunately, none of these hypotheses seems to fully explain the disagreement, even if accounting for (i) uncertainties in reaction rate constants of radical–radical reactions and (ii) a potential lower proportion of HO2 compared to OH allow improving the agreement. The combination of these two hypotheses (lower proportion of HO2 by 25 % and reaction rate constants of OH + HO2 reduced by 20 %) leads to an agreement within 15 % (not shown).

The model inability to reproduce laboratory observations may be due to (i) the approach used to perform the simulations (see section 3), (ii) a misrepresentation of the secondary chemistry for pyrrole, and (iii) an impact of the flow dynamic inside the reactor. Improving model simulations would require investigating the pyrrole chemistry, coupling the chemical mechanisms to a CFD (computational fluid dynamics) model, and considering that a constant production of OH from the injector would lead to the reaction of OH with byproducts and peroxy radicals previously formed inside the reactor.

Nevertheless, similar behaviors observed between simulations and experiments give us confidence in experimentally derived correction factors. It is worth noting that the dependence of F on the reactivity of added gas standards is reduced at higher pyrrole-to-OH ratios (see Fig. 9, bottom panel). A similar trend is observed for the water dependence. It would therefore be beneficial to run the CRM instruments at high pyrrole-to-OH ratios to reduce the uncertainty introduced by the correction applied to account for not operating the instrument under pseudo-first-order conditions, keeping in mind that a higher pyrrole-to-OH ratio leads to a lower OH mixing ratio in the reactor, which in turn worsens the detection limit. Working at pyrrole-to-OH ratios ranging from 1.7 to 2.0 seems to be optimal for the MD-CRM instrument.