The transmission of NO2 through the hot glass inlets (with frit and packed with glass beads) of the ΣPNs and ΣANs channels was investigated by sampling NO2 in synthetic air (up to 10 ppbv) into the three 405 nm channels simultaneously. NO2 is depleted slightly in the hot inlets with a transmission of 98.5±2 % and 95.5±2 % found for the ΣPNs and ΣANs inlets, respectively (see Fig. 5). The NO2 transmission was measured before, during and after the field deployment of this instrument (corresponding to a period of several months), and was found to be constant and also independent of relative humidity.

The thermal decomposition of PNs leads to the formation of organic radicals and NO2. In an ideal situation, in which PNs decompose 100 % to NO2 and when NO2 does not undergo any further production or loss reaction, the total [NO2] measured in the ΣPNs channel is equal to the sum of ambient NO2 and PNs. Under the operating conditions of the CRDS instrument described here, the total [NO2] signal in the ΣPNs channel can, however, be biased by a number of reactions initiated by the organic radicals. The influence of these reactions has been described in the literature (Day et al., 2002; Paul and Osthoff, 2010; Thieser et al., 2016). Taking the example of peroxyacetyl nitrate (PAN), subsequent to its thermal decomposition (Reaction R1a), there are processes that lead both to the removal of NO2 (Reactions R1b, R5) and to its formation (Reactions R2, R3, R4). CH3C(O)O2NO2+M→CH3C(O)O2+NO2+MCH3C(O)O2+NO2+M→CH3C(O)O2NO2+M CH3C(O)O2+NO(+O2)→NO2+CH3O2+CO2CH3O2+NO(+O2)→HCHO+HO2+NO2HO2+NO→OH+NO2OH+NO2+M→HNO3+M Reactions (R2)–(R4) show that, in the most unfavourable scenario, if all organic radicals react with NO, each peroxyacetyl radical can lead to the formation of three NO2 molecules, biasing the result by the same factor. Likewise, the presence of very high NO2 could conceivably result in complete reformation of PAN (Reaction R1b). The size and sign of the bias thus depends on the relative concentrations of NO and NO2 and, most importantly, on the rate of loss of organic radicals to the reactor walls or via thermal decomposition. The bias resulting from these reactions can be reduced by minimising the residence time between thermal dissociation and detection and by making pressure-dependent recombination reactions inefficient by working at low pressures, e.g. by using LIF detection of NO2 (Day et al., 2002; Wooldridge et al., 2010). As CRDS instruments operate at higher pressures, to maintain sufficient sensitivity (typically from 0.5 to 1 bar), we have taken a different approach and optimised the surface losses of the organic radicals by modifying the surface-to-volume ratio in the heated inlets.

First experiments on PAN samples using glass wool as a radical scavenger resulted in the desired reduction in the rate of recombination of CH3C(O)O2 with NO2. Glass wool was, however, observed to greatly enhance (rather than reduce) the oxidation of NO to NO2. This is presumably the result of a surface-catalysed process as previously observed on powdered, aluminium silicate mineral dust samples (Hanisch and Crowley, 2003), and may be related to the formation of oxidised surface sites that can react with NO. This led us to test glass beads, which presumably have less reactive “defective” sites than glass wool but a sufficient surface area to remove a large fraction of the organic radicals (see below). We conducted a series of experiments with different mixtures of PAN and NO2 / NO (from 1 ppmv in synthetic air gas bottles) and analysis of the observations by numerical simulation, similar to that presented in Thieser et al. (2016).

The set of chemical reactions used in the numerical simulations is essentially the same as that described in Thieser et al. (2016). One exception is formation of NO2 via the reaction between O3 and NO, which Thieser et al. (2016) treated as occurring independently of other chemical processes. Here, this reaction is treated more rigorously by including it in the chemical simulation. The other difference is the use of two different values for a rate constant (ks) used to calculate the kinetic limitation on the radical uptake coefficient (γ, see below). In order to simulate all laboratory data sets, different values of ks (one for sections A and B and one for sections C and D; see Fig. S1 in Supplement) are required to account for the temperature gradient in the heated sections of the inlets. The heterogeneous loss of radicals to the reactor walls is based on a Langmuir–Hinshelwood mechanism, with the first-order rate constant (kw) given by Eq. (2). kw=γc‾A4 The uptake coefficient, γ, depends on both kinetic and diffusive limitations as described in detail by Thieser et al. (2016). Here we focus on the role of glass beads in enhancing the rate of uptake and reaction of organic radicals by increasing the surface area available for reaction (A). Without the glass beads, the value of A in the heated inlets is 3.5 cm2cm-3 (corresponding to a cylinder with internal diameter 1.2 cm). The presence of the glass beads, with an average diameter of 0.5 mm, results in a surface-to-volume ratio of ≈ 100 cm2cm-3. The exact determination of the temperature profile in the inlet is possible in the case of a classic cylindrical reactor, but is difficult to achieve with the presence of the glass beads and glass frit. An approximate profile in section D of the reactor (part of the cylindrical reactor between the fritted glass and the cavity inlet; see Fig. S1) was obtained by measuring the temperature between the fritted glass and the entrance of the cavity. It is assumed that the temperature in section B (containing the glass beads) increases from ambient to 448 K at the point where the gas enters section C (corresponding to the fritted glass section). This profile is also used to calculate the residence time in the ΣPNs channel of 1.3 s. To take into account the variations of temperature, pressure, specific surface area and diffusion radius along the glass reactor, a profile for each of these parameters is used as input in the numerical simulation.

Figure 6a shows results of a set of four experiments in which different amounts of PAN are mixed with different amounts of NO2 and monitored in the three channels. The y axis shows the total [NO2] measured in the ΣPNs channel ([NO2]448K) minus NO2 measured in NO2 channel ([NO2]Amb). The negative slopes of these data indicate that NO2 from PAN decomposition is lost at high mixing ratios of added NO2. A similar set of data, but with added NO rather than NO2, is displayed in Fig. 6b. In this case, the positive slopes indicate that NO2 is being formed from the reaction of organic radicals with NO. Note that the solid lines are the results of the numerical simulations (with identical mechanism) of NO2 formation and loss in both experiments. The mixing ratios of PAN listed are those required in the model to match the experimental data. In both data sets the slopes are weakest at low PAN concentrations, reflecting the fact that radical-NO or radical-NO2 reaction rates will be dependent on the radical concentrations. The mixing ratios of PAN cannot be derived by simple back extrapolation to zero NO2 or NO as this does not take into account the NO2 formed when PAN itself decomposes. At the highest mixing ratios of NO2, up to 25 % of the peroxyacetyl radical recombines with NO2 while the other 75 % is lost to the walls or did not react with NO2.

The analysis of the measurements shows that (without correction) NO2 would be overestimated by approximately 40 % for 3 ppb of PAN when sampling air containing 8 ppbv of NO. This is much lower that the ≈ 150 % reported by Thieser et al. (2016) for similar conditions in their ΣPNs channel, reflecting the increased loss of radicals on the glass surfaces.

In both series of experiments, the model reproduces the data for the whole range of [PAN], [NO2] and [NO] explored. A set of experiments was carried out to investigate whether the formation (or loss) of NO2 described above could be influenced by the presence of water vapour acting, for example, as a quencher of a surface reaction. A mixture of PAN plus NO2 was humidified to 60 % RH, with other operating conditions similar to the previous experiments. The numerical model reproduced the measurement data without the need to modify the wall loss rates or add any chemical processes involving H2O, suggesting that the gas and surface reactions taking place are not significantly influenced by adsorbed water at these temperatures. In order to correct field data for these biases, an iterative fitting procedure is used and described in detail in Sect. 4.2.

As first discussed by Thieser al. (2016), experimental and theoretical studies show that the higher temperature (648 K) of the ΣANs channel changes the chemical processes substantially compared to 448 K. ANs can now decompose to NO2 and an alkyl radical fragment, whereas the peroxyacetyl radical (from PAN decomposition) is thermally unstable. The NO2 generated in the ΣANs channel when adding NO2 / NO to PAN samples is displayed in Fig. 6c and d. The results show a greatly reduced effect of NO2 recombination or NO oxidation compared to the ΣPNs channel (Fig. 6a, b), largely resulting from the instability of the of CH3C(O)O2 radical. The model takes into account two main pathways for the fate of the peroxyacetyl radical (Reactions R6, R7) in which the acetyl radical (CH3CO) formed in Reaction (R6) can react further with O2 to reform a peroxyacetyl radical or an OH radical and an α-lactone (Reactions R8, R9) (Carr et al., 2011). CH2C(O)OOH can also decompose to OH and an α-lactone (Reaction R10) (Carr et al., 2007, 2011; Chen and Lee, 2010). If not lost to the walls, OH can react with NO2 to form HNO3 according to Reaction (R5). CH3C(O)O2+M→CH3CO+O2+MCH3C(O)O2→CH2C(O)OOHCH3CO+O2+M→CH3C(O)O2+MCH3CO+O2→OH+C2H2O2CH2C(O)OOH→OH+C2H2O2CH3CO+O2+M→CH3O2+CO+M The acetyl radical can also form a methylperoxy radical CH3O2 through the formation of a CH3 radical and CO. The net reaction is given by Reaction (R11). In the presence of NO, the peroxy radicals (CH3C(O)O2, CH3O2 and HO2), if not lost to the glass surface, can lead to the formation of NO2. To simulate the reactions in the ΣANs channel, the temperature profile was determined in the same way as the ΣPNs channel profile. Due to the higher gas flow velocity at the elevated temperatures, the pressure profile is slightly different than in the ΣPNs channel and gives a cavity pressure that is about 10 mbar lower. The results of the experiments involving addition of NO2 to PAN samples in the ANs channel are presented in Fig. 6c. For the range of PAN concentrations covered, the loss of NO2 by chemical recombination is about 5 % of the initial PAN. The initial PAN concentration used as input variable for the numerical simulations is, for all four experiments, about 3–5 % higher than that needed to fit the data obtained in the PNs channel (Fig. 6a), which is most probably related to model uncertainties. Since this 3–5 % difference is the same for all four experiments and is apparently not correlated with the amount of PAN, it is added to the calculation of total uncertainty for the PNs measurement. The results of the NO addition experiments are shown in Fig. 6d. The thermal decomposition of the peroxyacetyl radical reduces the positive bias due to NO oxidation so that a maximum factor of only 1.04 (compared to 1.40 in the ΣPNs channel) is obtained. The initial PAN concentration required to fit the NO addition data set at 648 K is 5–10 % higher than at 448 K, which is related to uncertainty in the [NO] mixing ratio used or a bias in the simulation.

In order to investigate the role of organic (alkyl) radicals generated from the thermal decomposition of ANs, a set of experiments was conducted in which a sample of i-propyl nitrate (C3H7ONO2) in synthetic air was mixed with different amounts of NO and NO2. Thermal decomposition of i-propyl nitrate (Reaction R12) is followed by Reactions (R13) and (R14) that generate HO2 and CH3O2, which can convert NO to NO2 and sequester NO2 as HO2NO2 and CH3O2NO2. C3H7ONO2→C3H7O+NO2C3H7O+O2→CH3C(O)CH3+HO2C3H7O+M→CH3+CH3CHO+M Figure 7a and b show that the addition of NO increases the amount of NO2 formed per AN, whereas the presence of NO2 results in a negative bias to the data. These effects are captured well for both data sets by the model simulations (blue lines).

At high temperatures, O3 decomposes according to Reaction (R15). Although most of the O atom produced reacts with O2 to reform ozone (Reaction R16), it can form NO+O2 in the presence of NO2 (Reaction R17). O3→O2+OO+O2+M→O3+MO+NO2→O2+NO The importance of this process is strongly dependent on the operating conditions of the instrument, recent studies showing that higher temperatures or pressures may result in a significant, negative bias from the ozone-pyrolysis-initiated reduction of NO2 (Lee et al., 2014; Thieser et al., 2016). Assuming that the steady-state concentration of the O atom in the heated inlets is dominated by Reactions (R15) and (R16) (k16[O2]≫k17[NO2]), the loss of NO2 can be approximated as follows: -dNO2=NO2O3k15k17k16O2t. The term in brackets is constant at constant temperature, pressure and flow rate through the heated inlet, and can be determined experimentally by sampling different mixtures of NO2 and O3 simultaneously through the ambient temperature channel and the two hot channels. Figure 8 shows the difference [NO2]Amb-[NO2]648 K (in ppbv) as a function of [NO2]×[O3] for three different levels of ozone (50, 115 and 185 ppbv). A linear fit through the data points yields a slope of 2.54±0.26×10-4 ppbv-1, which is < 50 % of the value found by Thieser et al. (2016). The difference can be attributed mainly to the use of a lower oven temperature (648 K in this work instead of 723 K). No reduction of NO2 is observed in the ΣPNs channel because thermal decomposition of O3 at 448 K is too slow. To illustrate the impact of the thermal dissociation of O3 in the ANs channel, we note that 5 ppbv of total NO2 in the presence of 50 ppbv of O3 results in the removal of 75 pptv NO2. Since the amount of O3 decomposed and NO produced in this process is negligible compared to ambient amounts, it does not affect the input conditions for the numerical simulations, and can therefore be treated separately using Eq. (3).

A positive bias to the measurement of NO2 may result from NO oxidation by O3 during the time it takes for the sampled air to flow through the inlets and cavities of the instrument (Reaction R18). NO+O3→NO2 The rate coefficient of this reaction is temperature-dependent, and the effect is larger at 448 and 648 K than at room temperature. To investigate this interference, different mixtures of O3 and NO were sampled simultaneously through the three channels. The results are shown in Fig. 9 where data from four experiments with different O3 mixing ratios (41, 62, 80 and 131 ppbv) are plotted. The solid lines are fits to the data, the slopes of which are second-order rate constants multiplied by a reaction time (cm3molecule-1). The rate constant for the gas-phase reaction between NO and O3 is listed as kO3+NO=9×10-19exp⁡(-850/T)×T2.25 cm3molecule-1s-1 (Atkinson et al., 2004). For the ΣPNs channel (448 K), using this expression results in an underestimation of the effective (measured) production of NO2 by a factor 1.06. For the ΣANs (648 K) channel the equivalent factor is 1.52. The most likely reason for this discrepancy is that some NO is oxidised in a surface catalysed process, presumably involving surface sites that are activated following interaction with O3. To account for this extra source of NO2, the O3 plus NO reaction is implemented in the chemical model (with modified rate expressions) for the 448 and 648 K data corrections (see Sect. 3.2.2 and 3.2.3), and corrected manually for the NO2 channel.

Atmospheric trace gases other than PNs and ANs can potentially be detected as NO2 in this instrument. As thermal decomposition of N2O5 to NO3 and NO2 has been shown to be 100 % efficient in our 383 K inlet in the 662 nm channels, we expect 100 % dissociation at the higher temperatures of the ΣPNs and ΣANs channels. This may represent a significant source of bias during the night when N2O5 mixing ratios can be large. However, as the five-channel instrument simultaneously monitors N2O5, this can be easily corrected. Here we consider possible HNO3 and ClNO2 interferences in the ΣANs channel.

HNO3 is a major reservoir of tropospheric NOx. Although HNO3 is thermally stable at temperatures below 700 K, Wild et al. (2014) report ≈ 95 % conversion of HNO3 to NO2 at 648 K, whereas Thieser et al. (2016) found about 10 % at 723 K. In order to test for unwanted detection of HNO3 when sampling from the hot inlets (with glass beads) in the five-channel instrument, we injected a sample of HNO3 in synthetic air directly prior to the flow division before entering the three 405 nm channels. In this way, possible losses of HNO3 in the inlet in front of the heated sections of the instrument were minimised. HNO3 was generated in a custom-built permeation source, which was calibrated by absorption spectroscopy at 184.85 nm, where HNO3 absorbs strongly. The permeation source generated several ppmv of HNO3 in a flow of 10 sccm, which, following dilution, provided a mixing ratio of ≈ 30 ppbv at the CRDS inlet. The permeation source also generated about 3 ppbv of NO2. Figure 4 (open circles) shows that NO2 arising from HNO3 decomposition could not be observed in either heated inlet, enabling us to set an upper limit to the decomposition efficiency of HNO3 to NO2 of < 0.5 %. This is in broad agreement with the previous measurement by Thieser et al. (2016), who saw no evidence for HNO3 decomposition below 650 K, and is in stark contrast to that reported by Wild et al. (2014). While we have no rigorous explanation of this, we note that the gas-phase thermal dissociation of HNO3 to NO2 at the temperature of these experiments is too slow to explain its formation, which suggests that it is probably surface-catalysed in the experiments of Wild et al. (2014), implying that different glass types or chemical history of the surface may influence the formation of NO2 significantly.

As reported previously, ClNO2, formed in the atmosphere in the reaction between N2O5 and chloride-containing particles, can be detected as NO2 in TD-CRDS instruments (Thaler et al., 2011; Wild et al., 2014; Thieser et al., 2016) The ClNO2 decomposition efficiency as a function of the oven temperature is plotted in Fig. 4. The ClNO2 sample was generated by flowing Cl2 in synthetic air over sodium nitrite crystals. In normal operating conditions (at 648 K), 90±3 % of the ClNO2 is decomposed to NO2. As ClNO2 mixing ratios are highly variable and can approach ppbv levels, ClNO2 may thus represent a serious limitation to ANs measurements, unless independent measurements are available to enable correction.

The detection limit for the 405 nm channels can be estimated in a similar manner to that described for the 662 nm channels. The 2σ standard deviation for consecutive zeros for the NO2, ΣPNs and ΣANs channels are respectively 59, 74 and 54 pptv. Using these values, we obtain detection limits for [NO2], [PNs] and [ANs] of 59, 94 and 80 pptv. The detection limits for [PNs] and [ANs] are obtained by error propagation on the NO2 and PNs channels and NO2 and ANs channel (see Sect. 4).

The total uncertainty associated with the [NO2] measurement is a combination of the uncertainties in the l-to-d ratio, humidity matching of the zero and ambient air and the correction for the NO+O3 reaction on the absorption cross section of NO2 and on N2O5 decomposition. The total uncertainty on the absorption cross section is estimated to 6 %, taking into account the error on the reference cross section (Voigt et al., 2002) and fluctuation in the laser emission spectrum. This value, when combined with the uncertainty associated with the l-to-d ratio, results in 6.5 % uncertainty. The uncertainty associated with the correction for NO2 formation in the NO+O3 reaction depends on ambient ozone and NO levels. Considering 10 % uncertainty for kO3+NO and 5 % uncertainty for [O3], [NO] and [NO2], [NO] and [O3] ambient levels of 1, 0.5 and 50 ppbv, we obtain an uncertainty for this correction of 0.1 % on the final [NO2] value. NO2 can be formed at room temperature by the slow thermal decomposition of N2O5. Taking a decomposition rate constant of 4.4×10-2 s-1 at 303 K for N2O5 (Atkinson et al., 2016) and a residence time in the NO2 channel of 1.7 s, we calculate that 0.08 of the N2O5 can decompose. The largest N2O5-to-NO ratio measured during NOTOMO was 0.17, which results in a maximum contribution of ≈ 1.5 % NO2 from N2O5 decomposition. As mentioned in Sect. 2.2, the upper limit for bias caused by humidity matching errors when zeroing is 10 pptv, which results in a final uncertainty for [NO2] of 8 % + 10 pptv.

The uncertainty of the total NO2 detected in the ΣPNs and ΣANs channels is a combination of the same uncertainties involved in the [NO2] measurement plus the uncertainty of the NO2 transmission through the heated glass beads. Taking the values listed in Sect. 3.2.1, we obtain an uncertainty for total NO2 detected of 7 %pptv for both ΣPNs and ΣANs channels. To correct for the radical plus NOx biases, the total NO2 signals measured in the two hot channels are used to constrain the chemical model described earlier. The uncertainty arising from these calculations depends on the total [NO2] measured in each channel, and is thus highly variable. Note that uncertainty arising from the humidity difference is cancelled out by the subtraction to obtain [PNs] and [ANs]. The uncertainty added by the calculation of the ambient [PNs] and [ANs] depends on the model uncertainties, as well as on the uncertainty associated with the trace gas concentrations involved in this calculation. The uncertainty associated with [ΣPNs] and [ΣANs] is related to ambient [NO2], [O3] and [NO]. The fact that the corrections rely on PAN and IPN chemistry led Thieser et al. (2016) to estimate the uncertainty associated with the model-derived corrections to be 30 % (max) of the overall correction factor. The final uncertainties on the [ΣPNs] and [ΣANs] value depend on the total NO2 signal in all three channels, and can vary significantly. Below, we discuss the correction factors required for the NO2, ΣPNs and ΣANs data sets obtained during the first field deployment of the instrument.