Experiments were performed at the University of Helsinki in a 2 m3 atmospheric simulation Teflon (FEP) chamber. The COALA chamber (named after the project for which it was constructed: Comprehensive molecular characterization of secondary Organic AerosoL formation in the Atmosphere) was operated under steady-state conditions, meaning that a constant flow of reactants and oxidants were continuously added to the chamber, while chamber air was sampled by the instruments. Under the conditions used in this study, the average residence time in the chamber was ∼30 min, and the majority of conditions were kept constant for 6 to 12 h before changing to new conditions. These experiments focused on the characterization of the oxidation products arising from the α-Pinene (C10H16) ozonolysis. α-Pinene was used for the generation of oxidation products because it is the most abundant monoterpene emitted by the boreal forests and is one of the most important SOA precursors on a global scale (Jokinen et al., 2015; Kelly et al., 2018).

The experiments were conducted at room temperature (27±2 ∘C) and under dry conditions (RH < 1 %). An overview of the measurements, as well as the experimental conditions, are presented in Fig. 1. α-Pinene was introduced to the chamber from a gas cylinder, and steady-state concentrations of α-Pinene were varied from 20 to 100 ppb. As alkene ozonolysis yields OH radicals (Atkinson et al., 1997), in some experiments, ∼1500 ppm of carbon monoxide (CO) was injected to serve as the OH scavenger. Also, 10 to 50 ppb of O3 was generated by injecting purified air through an ozone generator (Dasibi 1008-PC) and monitored over the process of the campaign using a UV photometric analyzer (Model 49P, Thermo-Environmental). In the experiments performed in the presence of NOx, 400 nm LED lights were used to generate NO in the chamber from the photolysis of the injected NO2. The purified air ([O3/NOx] and [VOC] reduced to less than 1 ppb and 5 ppb, respectively), generated by an air purification system (AADCO, 737 Series, Ohio, USA) running on compressed air, was used as a bath gas. Temperature, relative humidity (RH) and pressure were monitored by a Vaisala Humidity and Temperature Probe (INTERCAP^® HMP60) and a differential pressure sensor (Sensirion SDP1000-L025).

We deployed five chemical ionization schemes to the COALA chamber in order to characterize the chemical composition of the gas-phase oxidation products formed from α-Pinene ozonolysis. In this section, we briefly present each instrument, summarized in Table 1. As each mass spectrometer has slightly different working principles, references to more detailed descriptions are provided. Specific benefits and limitations, which were not often discussed in earlier studies, are reviewed in Sect. 2.4. Each of the mass spectrometers were equipped with a mass analyzer manufactured by Tofwerk AG, either an HTOF (mass resolving power ∼5000) or long TOF (LTOF, mass resolving power ∼10000) version.

In the analysis, we focused primarily on the relative behavior of the ions measured by the different mass spectrometers. An absolute comparison was also performed, but this approach has a larger uncertainty, as the sensitivity towards every molecule is different in each of the mass spectrometers, depending on molecular size, functionality, proton affinity, polarizability, etc. We attempted a rough estimate of absolute concentrations for each instrument, despite the fact that, with around a thousand ions analyzed, it is evident that we make no claim for them to all be accurate. As will be shown, the concentrations of gas-phase VOCs and OVOCs vary up to 7 orders of magnitude, and therefore useful information can still be obtained even in cases where concentration estimates could be off by an order of magnitude. Details about instruments used in this study as well as calibrations and instrumental limitations are discussed in the following sections.

In order to estimate absolute concentrations of all detected molecules, each instrument's signals, using an averaging period of 15 min, were normalized to the reagent ion signals (to eliminate the influence of changes affecting all signals in the instruments, e.g., due to a degrading response from the detector), followed by multiplication with a scaling factor. The reagent ion quantity used for normalization is described below, separately for each instrument. Normalized ion count rates are reported as normalized cps and normalized counts per second (ncps). The scaling factors were derived differently for each instrument (details provided below). For iodide, nitrate, and amine, the same factor was used for all ions in the spectrum, while for the PTR instruments the factors were different depending on the type of molecule (e.g., VOC or OVOC). For the PTR instruments and the iodide, a duty cycle correction was applied to compensate for mass-dependent transmission due to the orthogonal extraction of the mass analyzers. The amine and nitrate were calibrated by scaling a wide range of mass to charge based on earlier studies, where duty cycle corrections had not been performed. Therefore, we did not apply such a correction for the atmospheric pressure ionization mass spectrometers. Finally, we emphasize that the scaling factors should not be compared between instruments as a measure of sensitivity, since multiple factors impact these values, including, for example, the specific normalization approach and the chosen extraction frequency of the mass analyzers.

The PTR-TOF was calibrated twice using a calibration unit consisting of a calibration gas mixture of 16 different VOCs (Apel-Riemer Environmental Inc., USA) that was diluted with clean air purified by a catalytic converter (1.2 L min-1 of zero air and 8 sccm of standard gas), producing VOC mixing ratios of around 7 ppb (parts per billion) (Schallhart et al., 2016). Sensitivities were calculated to be 12.31, 27.92, and 30.51 ncps ppb-1 based on the concentrations of monoterpenes, MVK (methyl vinyl ketone) and m-/o-xylenes. PTR-TOF signals were normalized using the sum of the first primary ion isotope at 21.0221 Th and the first water cluster isotope at 39.0327 Th (e.g., Schallhart et al., 2016). According to common practice, the sensitivities above were scaled to correspond to a situation where the total reagent ion signal equaled 106 cps.

The vocus was calibrated four times during the campaign using the same calibration gas mixture as used for the PTR-TOF. There was variability in the sensitivity during the campaign and therefore the uncertainty in the vocus results are slightly larger than normal. Sensitivities were highest for acetone, at maximum around 1800 and around 650 cps ppb-1 for monoterpenes. α-Pinene concentration was retrieved using the authentic standard, while the concentrations of the OVOC and C10H14H+ were estimated using the calibration factor of the MVK and sum of m-/o-xylenes, respectively. MVK and m-/o-xylene sensitivities was around 1700 and 700 cps ppb-1, respectively. Vocus signals were normalized using the primary ion signal at 19.0178 Th only, as the water clusters have a negligible effect on the ion chemistry inside the FIMR (Krechmer et al., 2018). Due to the high-pass filter that removes almost all the signal at 19.0178 Th, we do not report the normalized sensitivities (i.e., in ncps ppb-1) for the vocus in order to avoid direct comparisons with the PTR-TOF. Instead, the sensitivities above are given without normalization, although a normalization was used for the final data. For the uncertainty estimates, the same applies as listed above for the PTR-TOF.

The uncertainties for the compounds that were directly calibrated are estimated to be ±20 % for PTR-TOF and vocus. For other compounds, the uncertainties are much higher due to uncertain ionization efficiencies and potential fragmentation of the compounds with unknown structures. For example, we used sensitivity of MVK for all oxygenated monoterpenes, even though all those compounds may have very different fragmentation patterns, transmission rates and/or proton transfer reaction rates. Therefore, we refrain from quantitative estimates of the uncertainties for these species.

The iodide was calibrated twice during the campaign (15 and 23 December) by injecting known amounts of formic acid into the instrument. Due to unknown reasons, the response of the iodide decayed throughout the campaign, and therefore only data measured before 17 December, when a stronger drop occurred, were included for the direct comparison of the nonnitrate OVOCs. While normalization should compensate for this type of behavior, this particular instrument utilized a time-to-digital converter (TDC) acquisition card, which meant the primary ion peak was heavily saturated. Lacking any isotopic signatures for I-, we found that utilizing a region of the rising edge of the I- signal (126.5–126.65 Th) provided a reasonable correction to our data. The sensitivity without normalization was 1.0 cps ppt-1 for formic acid, and following the normalization, this sensitivity was applied for all ions throughout the period where iodide data were included in the analysis. We acknowledge that this brings with it a large uncertainty, as the iodide has sensitivities ranging over a few orders of magnitude depending on the specific molecule (Lee et al., 2014), and refrain from quantitative uncertainty estimates, as in the case for the PTR instruments above.

Standards for OVOC compounds measurable by the nitrate are still lacking, and this instrument was therefore not directly calibrated during the campaign. However, to be able to roughly estimate concentrations, a calibration was inferred by assuming that the molar yield of HOMs, i.e., molecules with six or more oxygen atoms, was 5 % during α-Pinene ozonolysis experiments. Different values have been reported for the HOM yield in this system, ranging from slightly above to slightly below 5 % (Ehn et al., 2014; Jokinen et al., 2014, 2015). Clearly such an approach yields large uncertainties, and we estimated it here to roughly ±70 %. Earlier work with more direct calibrations reported an uncertainty of ±50 % (Ehn et al., 2014) and the added 20 p.p. in this work reflects the increased uncertainty in scaling the sensitivity based on expected HOM yields. This method requires knowledge of the wall loss rate of HOMs in the COALA chamber, which was estimated to be 1/300 s-1 in our study. This estimate is based on a rough scaling to a slightly smaller chamber (1.5 m3) with active mixing by a fan, where the loss rate was measured to be 0.01 s-1 (Ehn et al., 2014). As our chamber is larger, and our mixing fan was only spinning at a moderate speed, we estimated the loss rates to roughly 3 times lower. The resulting calibration coefficient was 2×1010 molecules cm-3 ncps-1, which is similar to that in previous studies (Ehn et al., 2014; Jokinen et al., 2012). As for nitrate, the amine was also not calibrated directly, and in order to achieve an estimate of the concentrations measured by this instrument, we scaled the sensitivity of the amine to match that of the nitrate for specific HOM dimers (C19H28/30O12-17 and C20H30/32O12-17), which were found to correlate very well between the two instruments (as described in more detail in the Results section). This approach gave a calibration factor of 6×108 molecules cm-3 ncps-1, with similar uncertainty estimates to the nitrate. In the CI-APi-TOFs, the calibration factor is generally close to 1010 molecules cm-3 ncps-1, but as discussed later in Sect. 3.1, the amine reagent ion was considerably depleted during the experiments, which led to the relatively low calibration factor. As mentioned earlier, the scaling factors should not be compared directly between instruments. The lower value for the amine is a result of the normalization rather than an indication of higher sensitivity. This reagent ion depletion also means that the most abundant species were most likely no longer responding linearly to concentration changes, and therefore their concentrations can be off by an order of magnitude or more.

In this section we aim to highlight some of the limitations involved when characterizing and quantifying OVOCs measured by online mass spectrometers. The list below is not exhaustive but addresses several issues that are relevant for the interpretation of our results.

Concentrations of the identified compounds were estimated for all the different instruments, as described in Sect. 2.6. It should be noted that no separate inlet loss corrections were applied. The estimations for the results of PTR-TOF and the vocus are the most reliable as both instruments were calibrated using authentic standards with a proven method, while larger uncertainties in the total measured concentrations are expected for the iodide and the CI-APi-TOFs.

With around 1000 identified ions for each instrument, except for the PTR-TOF, we decided to focus our attention in this section on a few particular compound groups: the most abundant C10 monomers (i.e., C10H14/16On), C10 organonitrates (C10H15NOn) and dimers (C20H32On). For the nonnitrate compounds, the concentrations were measured during steady-state conditions on 9 December from 15:30 to 23:00 with [O3] =25 ppb and [α-Pinene] =100 ppb) during period I (Fig. 1 in blue). The organonitrate concentrations were compared using steady-state conditions from 20 December, from 02:45 to 07:45 with [O3] =35, [α-Pinene] =100 and NO =0.5 ppb, during period IV (Fig. 1 in purple). Figure 5a–d show the concentrations of the selected species as a function of oxygen number in the molecules. While we again emphasize that all the concentrations were only rough estimates, these plots painted a similar picture to the correlation analysis, as described in more detail in the next paragraphs.

Focusing first on the nonnitrate monomers (Fig. 5a–b), for compounds with zero or one oxygen atoms, the PTR-TOF agreed well with the concentration estimated by the vocus, while molecules with more than two oxygen atoms were already close to, or below, the noise level of the PTR-TOF. In contrast, as the number of oxygen atoms in the molecule reached two or more, the iodide signal increased and for most compounds showed concentrations similar to the vocus. These two instruments agreed on concentration estimates fairly well all the way up to an oxygen content of around nine oxygen atoms, where the measured signals were close to the instruments' noise levels. However, when comparing to the nitrate, which is assumed to have good sensitivity for HOMs with seven or more oxygen atoms, the concentrations suggested by the vocus and iodide for the O7 and O8 monomers were very high. We preliminarily attributed this to an overestimation of the concentrations of HOMs by these two instruments, possibly due to higher sensitivities towards these molecules compared to the compounds used for calibration (i.e., MVK). We also did not correct for potential backgrounds using the blanks for the iodide, although they were measured, since the variability in the blank concentrations (see also discussion in Sect. 2.4) was large enough to cause artificially high fluctuations in the final signals. Therefore, we opted to not include such a correction but also note that, even if half the signal at a given ion was attributable to background in the iodide, then it would only have a small impact on the logarithmic scales used in Fig. 5. Other possible reasons for this discrepancy was that the iodide and vocus were able to detect isomers that the nitrate was not, or that the nitrate sensitivity was underestimated. However, considering that the nitrate HOM signal was scaled to match a 5 % molar HOM yield, it was unlikely that the HOM concentrations can be considerably higher than this. Other estimated parameters involved in the formation and loss rates of HOMs also had uncertainties, but we did not expect any of them to be off by more than 50 %. This concentration discrepancy thus remained unresolved and will require more dedicated future studies.

Finally, the quantities estimated using the amine are significantly lower (1–2 orders of magnitude) for all monomers when compared to the other instruments. This was presumably related to the titration of the reagent ion, which meant that the majority of charged OVOCs will undergo multiple subsequent collisions with other OVOCs, potentially losing their charge in the process. The nitrate had, as expected, very low sensitivity towards less oxygenated compounds and its highest detection efficiency for HOMs (i.e., molecules with at least six oxygen atoms).

The organonitrate comparison in Fig. 5c suggested that both the vocus and the iodide were efficient at detecting these compounds, as both instruments agreed well (R2 > 0.7) for C10 organonitrates with 5 to 10 oxygen atoms. While organonitrates have been detected before using the iodide (Lee et al., 2016), this was the first observation in which the vocus also detected such compounds efficiently. However, we cannot exclude such compounds undergoing fragmentation within the drift tube as commonly observed in other PTR instruments (Yuan et al., 2017). For larger oxygen content, the nitrate again seemed to be most sensitive, showing clear signals above 10 oxygen atoms, where the previous instruments were already close to noise levels. The amine seemed worse at detecting organonitrates compared to nonnitrate monomers.

Neither of the PTR instruments were able to detect any dimers in this study within their measurement ranges (up to 320 Th for PTR-TOF). The amine and the nitrate were able to quantify the widest range of HOM dimers, while the iodide was able to detect less oxidized dimers (Fig. 5d). Based on the concentration estimates, the amine detection range also extended to less oxidized dimers than the nitrate, as has already been shown by Berndt et al. (2018). Dimers measured by the iodide were more abundant than the ones detected by the amine, but from the monomer comparisons we speculated that the amine might be underestimating concentrations, while the iodide might be overestimating them. With the data available to us, we can only speculate on the relative sensitivities of the instruments able to detect dimers, especially with the vocus providing no support to the comparison.

One aspect lending credibility to the amine dimer data, in addition to the good time series correlation with the nitrate, was the odd–even oxygen atom patterns visible both in the amine and nitrate data. Such a pattern is to be expected, since the 32 hydrogen atoms in the selected dimers indicate that they have been formed from RO2 radicals, one of which had 15 hydrogen atoms (which is what ozonolysis will yield, following OH loss) (Docherty et al., 2005; Lee et al., 2006; Ziemann and Atkinson, 2012), while the second RO2 had 17 hydrogen atoms (which is the number expected from OH oxidation of an alkene where OH adds to the double bond). The first RO2 from ozonolysis had four oxygen atoms, and further autoxidation will keep an even number of oxygen atoms, while the opposite was true for the OH-derived RO2, which started from three oxygen atoms. In other words, the major dimers from this pathway should contain an odd number of oxygen atoms after they are combined. In the case of C20H30On dimers, mainly formed from two ozonolysis RO2, the pattern was expected to show peaks at even numbers, which is also the case (not shown).

Odd–even patterns for the oxygen content were not visible in the iodide, but the reason remained unknown. It was possible that the dimers detected by the iodide might be formed via other pathways, where such a selectivity did not occur. This topic should be explored further in future studies, since dimers formed from the oxidation of biogenic compounds are important for new-particle formation, and it is therefore critical to accurately identify and quantify the formation and evolution of different types of dimers. To date, both dimers measured by iodide (Mohr et al., 2017) and nitrate (Tröstl et al., 2016) have been found to be important for particle formation from monoterpenes.

Figure 6 summarizes our results and depicts the performance of each mass spectrometer in detecting monomer and dimer monoterpene oxidation products. Molecules of C10H16On, C10H15NOn and C20H30On were provided as examples. We emphasized that the oxygen content alone was not the determining factor for whether a certain type of mass spectrometer will detect a compound, but we utilized this simplified representation in order to provide an overview of the performances of the different chemical ionization schemes. The results were primarily based on the correlation analysis from Sect. 3.1, and as apparent from the y axis, this comparison was only qualitative. However, our aim was to provide an easy-to-interpret starting point, especially for new CIMS users wanting to compare different available techniques.

For monomer compounds without N atoms, shown in Fig. 6a, the PTR-TOF was limited to the detection of VOCs, while the vocus was additionally able to measure a large range of OVOCs, up to at least five to six oxygen atoms. The iodide detected OVOCs with oxygen content starting from ∼3 atoms but did not seem to efficiently observe HOM monomers (i.e., C10HxO>7). While being a very promising instrument for a broad detection of OVOCs, the performance of the amine was limited in our study due to a significant drop in the reagent ion to ∼40 % of the total signal. Therefore, the amine was marked with a shaded region rather than a line, with the lower limit based roughly on its usefulness under the conditions we probed, while the upper limit was an estimate based on findings in a cleaner system with low loadings (Berndt et al., 2018). Finally, the nitrate was mainly selective towards HOMs. The detection and quantification of monomeric OVOCs containing five to eight oxygen atoms remained the most uncertain, since there were inconsistencies in both concentration and correlation between the nitrate, measuring the more oxygenated species, and the vocus and iodide, which detected the less oxidized compounds.

In Fig. 6b, the suitability for the different instruments was plotted for organonitrate monomers. The vocus efficiently detected the less oxidized organonitrates, while the iodide displayed good sensitivity for the same compounds, with the exception of the least oxygenated ones. For larger number of oxygens, the nitrate again seemed the most suitable method. For dimers (Fig. 6c), neither of the PTR techniques showed any ability to detect these compounds in our study. We did not extend the lines all the way down to n=0 for the compounds, as it was still possible that these methods can be able to detect the least oxidized and most volatile C20 compounds, which might not have been present during our experiments. The iodide showed some correlation with the nitrate but had good signals mainly in the range of dimers with four to eight oxygen atoms. The amine and nitrate correlated well for the most oxidized dimers, suggesting good suitability for dimer detection of HOM dimers. The amine concentrations stayed high, with the expected odd–even pattern in oxygen number, even at lower oxygen content than the nitrate, and therefore the suitability extended further towards lower O-atom contents. Again, the shaded area was based on a combination of our findings and those of Berndt et al. (2018).

The results in Fig. 6 are based on the α-Pinene ozonolysis system. While we will not speculate too much about the extent to which these findings can be extrapolated to other systems, certain features will remain similar for other atmospherically relevant reactions. For example, the most oxidized gaseous HOM species will likely have been formed through autoxidation processes, which means that they will contain hydroperoxide functionalities and could thus be detectable by the nitrate. Likewise, the HOMs, and in particular the dimers, will very likely have low volatilities, requiring high sample flows with minimal wall contact, as in the case of the Eisele-type CI inlets used in the nitrate and amine. Several other key features are also expected to be valid in different VOC–oxidant systems, and therefore we believe that our findings are also relevant for many other reaction partners.

As a final test for each instrument, we estimated how much of the reacted carbon (in ppbC) the different mass spectrometers can explain. As shown in Fig. 7, both the iodide and vocus seemed to capture most of the reacted carbon within uncertainties. The concentration determined using the vocus was overestimated, explaining more carbon than was reacted. Out of the largest contributors to the reacted carbon, pinonaldehyde (C10H16O2) was not efficiently detected by iodide, but otherwise most of the abundant molecules were quantified by both vocus and iodide. Any carbon lost by condensation to walls or particles would not have been quantifiable by any of the instruments in this study. While the nitrate was calibrated with an assumption that it can measure 5 % of the reacted α-Pinene, it only detected less than 0.1 of that amount. The reason was that the HOMs it can detect were quickly lost to walls (or particles), and thus the gas-phase concentration was not equivalent to the branching ratio of the VOC oxidation reaction. In fact, and as revealed by the slow changes in the times series in Fig. 7d, most of the carbon ultimately measured by the nitrate was semi-volatile, as such compounds accumulated and reached higher concentration in the chamber, unlike HOMs. Thus, while the nitrate was able to detect a critical group of OVOCs from an aerosol formation perspective, i.e., HOMs, for carbon closure studies (Isaacman-VanWertz et al., 2017, 2018), it will be of limited use. This again highlights the need to first determine the target of a study before deciding which CIMS technique is the most useful. For the closure comparison in our study, the overestimations emphasized the need to perform calibration with an extensive set of OVOCs, ideally with monoterpene-oxidation products, in order to better constrain the sensitivity of the products of interest. The study by Isaacman-VanWertz et al. (2018), as the only study to achieve full carbon closure during chamber oxidation of α-Pinene by OH, also successfully utilized voltage scanning to determine sensitivities of each compound.