All measurements were conducted in July 2014 (15–27) during the F-BEACh 2014 (Fichtelgebirge – Biogenic Emissions and Aerosol Chemistry) field campaign. The measurement site was located in a rural area at an altitude of 766 m a.s.l. in the Fichtelgebirge mountain range in southeastern Germany (BayCEER Waldstein-Pflanzgarten, 50∘08′35′′ N, 11∘51′49′′ E; operated by the University of Bayreuth). The site is surrounded by a mostly coniferous forest which is dominated by Norway spruce (∼ 90 %). The canopy height and displacement height are ∼ 23 and ∼ 15 m respectively. A mixture of larch, beech, maple, and pine accounts for the rest of the tree population (Staudt and Foken, 2007). All instruments were arranged closely with inlet heights of 4–6 m above ground and at a distance of less than 10 m. Solely, VOC cartridges were sampled in and above canopy level at a distance of ∼ 200 m from the other instruments.

The AeroFAPA ion source was used in combination with an ion trap mass spectrometer (LCQ Deca XP Plus, Thermo, San José, CA, USA) for real-time analysis of ambient organic aerosol particles. Since a detailed description of the technique can be found elsewhere (Brüggemann et al., 2015), only a brief description will be given here. In general, AeroFAPA-MS is a soft-ionization technique which allows the online detection of organic compounds in aerosol particles. The negative mode, which was applied throughout the field study, is selective towards acidic compounds, such as carboxylic acids and (nitrooxy) organosulfates. For the analysis, aerosol particles were drawn from a height of ∼ 4 m above ground into the manifold of the AeroFAPA at a flow rate of 0.9 L min-1. Before reaching the AeroFAPA-MS, the aerosol stream passed an activated charcoal denuder in order to remove gaseous species from the aerosol sample. Evaporation of organic aerosol components prior to ionization was supported by heating the inlet to 200 ∘C. Although heating is a common approach for aerosol evaporation and analysis, it should be noted here that recently Lopez-Hilfiker et al. (2016) reported on the decomposition of accretion products upon heating. Therefore, contributions of such decomposition products cannot be completely ruled out here. A helium glow discharge plasma was used to generate excited helium atoms and primary reagent ions which ionized the compounds of interest in the so-called afterglow region. During the campaign, a current of 55 mA was used, resulting in a discharge voltage of ∼ 400 V. In addition, a potential of -15 V was applied to the exit capillary of the AeroFAPA to enhance ion transmission. The resulting analyte ions, typically [M-H]-, were then sampled and detected by the mass spectrometer. Fragmentation and adduct formation was only observed to a minor extent under laboratory conditions (< 5 % of signal for [M-H]-), however, such processes might have influenced the observed signals to a greater extent during the field campaign due to ambient conditions. A voltage of -15 V was applied to the mass spectrometer inlet capillary, equaling the potential of the AeroFAPA's exit capillary. The tube lens was held at 0 V. All mass spectra were recorded in automatic gain-control mode with 300 microscans spectrum-1, giving roughly one full scan mass spectrum (m/z 130–500) per minute. The maximum ion trap injection time was set to 200 ms. MSn experiments were performed to elucidate the structure of the detected compounds. Data were recorded using XCalibur 2.0.7. Background subtraction of the acquired mass spectra was conducted by measuring a blank sample for half an hour every day. For the subsequent data analysis all files were converted to text files and analyzed using Matlab (R2014b, Mathworks Inc., USA). In order to compare and correlate data from different instruments a unified time vector was created with time intervals of 10 min. Thus, all signals, except the filter sample data, are average values for 10 min.

A high-resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS, Aerodyne, USA; Canagaratna et al., 2007), was used to measure the submicron mass concentrations and size distributions of nonrefractory particulate organic matter, sulfate, nitrate, ammonium, and chloride. The AMS was located in an adjacent laboratory container and connected to a sampling line with a PM10 inlet located at ∼ 6 m a.g.l. (above ground level). Relative humidity (RH) on the sampling line was maintained below 35 % using a Nafion^® dryer. A chemical-dependent collection efficiency (CDCE) was applied on the AMS data according to Middlebrook et al. (2012). The quality control of the data acquired by the AMS was made according to Poulain et al. (2014).

Gas-phase concentrations of HOMs and sulfuric acid were measured ∼ 4 m above ground using a CI-APi-TOFMS (chemical ionization atmospheric-pressure interface time-of-flight mass spectrometer). A detailed description of the instrument can be found elsewhere (Jokinen et al., 2012; Mutzel et al., 2015). Briefly, an 241Am source was used to produce nitrate ions which were electrostatically guided into the sample flow of the inlet (length of 28 cm, inner diameter of 1.6 cm) to give nitrate clusters with gas-phase compounds present in the sampled air (sample flow of ∼ 10 L m-3). Then, the resulting clusters were transferred into the high vacuum region and detected by TOFMS. Calibration of the instrument was performed using sulfuric acid detection via H2SO4 + (HNO3)nNO3- (n = 0, 1, 2, 3) (Eisele and Tanner, 1993; Mauldin et al., 1998) with a calibration factor of 1.85 × 109 molecules cm-3 (Berndt et al., 2014). Diffusion-controlled wall losses in the sampling tube of 12 % were taken into account, using a diffusion coefficient of 0.08 cm2 s-1. The detection limit was about 104 molecules cm-3. Due to differences in ion transmission for sulfuric acid and HOMs, an uncertainty factor of 2 is estimated for the given HOM concentrations.

Particle number size distributions were measured ∼ 6 m above ground with a twin scanning mobility particle sizer (T-SMPS) custom built by TROPOS (Leipzig, Germany) according to the design recommended by Wiedensohler et al. (2012). The instrument includes membrane dryers to keep the RH below 40 % both in the sample and the sheath flow. The aerosol sample is brought to bipolar charge equilibrium using a commercial 85Kr neutralizer and sent to a Hauke-type differential mobility analyzer (DMA). The mobility diameter range from 10 to 710 nm was scanned in 71 size bins with a time resolution of 5 min. The closed-loop sheath flow rate was set to 5 L min-1 while the sample flow was directed to a model 3772 condensation particle counter (TSI Inc., Shoreview, Minnesota, USA) for particle detection with a flow rate of 1 L min-1.

VOCs were actively sampled on commercial two-stage cartridges filled with Tenax TA/Carbograph 5TD (Markes International, Cincinnati, Ohio, USA) for 30 min with a flow rate of 0.1 L min-1 in and above the spruce canopy at 12 and 31 m a.g.l. for subsequent offline gas chromatographic analysis. Samples were taken during daytime from 09:00 to 20:00 CET on 4 selected days during F-BEACh 2014. Ozone scrubbers coated with potassium iodide were used to minimize oxidation of collected compounds. After sampling, the cartridges were sealed immediately with metal caps, placed in a screw-cap PTFE container, and kept refrigerated until analysis. In the laboratory, VOCs were analyzed using standard thermal desorption gas chromatography with flame ionization detection (TD-GC-FID). The sample cartridges were thermally desorbed (200 ∘C), pre-focused on a Peltier-cooled trap (-15 ∘C), and injected onto an Rxi-5ms column (30 m, 0.32 mm, 1.00 µm, Restek, Bad Homburg, Germany) in a Sichromat 1 (Siemens AG, Germany) gas chromatograph. Monoterpenes (i.e., α-/β-pinene, d-limonene, Δ-3-carene, camphene) were quantified using authentic standards.

Only a brief description of the preparation and analysis of filter sample extracts will be given here. For more details, the reader is referred to the Supplement.

Filter samples were taken twice a day on tetrachloroethylene-coated borosilicate filters. The sampling time was ∼ 8 h for daytime filters (09:00–17:00 CET) and ∼ 16 h for nighttime filters (17:00–09:00 CET). After sampling, the filters were stored at < -18 ∘C until analysis. For the extraction procedure, a filter sample was cut into pieces and extracted using a methanol/water solution. After sonication and evaporation to dryness, the residue was dissolved in a solution of acetonitrile/water (1 : 4). To compensate for losses during the processing, an average recovery rate was determined for pinic acid, which served as a surrogate for the quantification of other monoterpene oxidation products. Here, an average recovery rate of 85 % was found and applied to the detected organic compounds. The liquid chromatography (LC) separation was conducted on a C18 column which was coupled to a high-resolution mass spectrometer (Q-Exactive, Thermo Scientific, Germany; resolving power of R = 7 × 104 at m/z 200). Ionization was carried out using electrospray ionization (ESI) in the negative mode. Each sample was measured in triplicate.

The obtained LC-MS data were analyzed by a commercial non-target screening software (Sieve 2.2, Thermo Scientific, USA). For the elemental formula assignments, the following isotopes and conditions were used: 12C (0–50), 1H (0–100), 16O (0–40), 14N (0–4), and 32S (0–4). The mass tolerance was set to ±5 ppm. Afterwards, the obtained compound list was checked for chemically unreasonable formula assignments, such as the absence of hydrogen in carbon-containing compounds or impossible O / C ratios (O / C < 3; 0.1 < H / C < 6).

In total, the automated non-target analysis of the filter samples by LC-MS resulted in 695 compounds, which showed significant signal intensities after background subtraction. In order to identify characteristic compounds for the organic aerosol fraction from this relatively large number only those signals were selected for the subsequent analysis which showed an integrated peak area of > 107 a.u. for at least two separate filter samples, resembling a signal to noise ratio of ≥ 12. As an additional criterion, the formula assignment for these signals had to show a mass accuracy in the range of ±2 ppm. Eventually, these thresholds led to 93 compounds in the nominal mass range of m/z 133–387, which were identified from the data analysis (Fig. 1 and Supplement). In general, the entire group of CHO compounds can be assigned to the class of organic acids since all measurements were carried out in the negative ion mode which is selective towards acidic compounds. Among the identified organic acids several common biogenic SOA marker compounds were detected, such as pinic acid (m/z 185.0819, [M-H]-) (Yasmeen et al., 2011) and terpenylic acid (m/z 171.0663, [M-H]-) (Claeys et al., 2009). The concentrations of all SOA marker compounds in PM2.5 were estimated using pinic acid as calibration standard. Despite similar chemical structures, the ionization efficiencies might, however, differ to a certain extent among these marker compounds. For example, in a post-calibration experiment the response of the MS for MBTCA was found to be ∼ 80 % of the one for pinic acid. Furthermore, the composition of the LC eluent can have additional effects on the actual ionization efficiencies. Therefore, the given values should rather be taken as semi-quantitative and in the case of MBTCA considered as a lower limit. Table 1 gives an overview of the identified marker compounds and their average concentrations during the campaign period. A comprehensive list of all detected compounds is given in the Supplement.

The most dominant marker compounds during the campaign period were MBTCA and 3-carboxyheptanedioic acid with estimated concentrations of 13.8 (±9.0) and 10.2 (±6.6) ng m-3 respectively. While MBTCA depicts a major oxidation product of α-/β-pinene (Szmigielski et al., 2007; Müller et al., 2012), 3-carboxyheptanedioic acid is a major oxidation product of d-limonene (Jaoui et al., 2006). These findings suggest that the site was strongly influenced by biogenic emissions consisting mainly of α-/β-pinene and d-limonene and their corresponding oxidation products. This hypothesis is further supported by monoterpene measurements which showed relative mixing ratios of 38 % α-pinene, 23 % β-pinene, 19 % d-limonene, 12 % Δ-3-carene, and 8 % camphene. The median mixing ratios of the sum of these five monoterpenes were 0.8 ppbV above the canopy and 1.6 ppbV within the canopy. In addition, from a comparison with the MEGAN emission model (Guenther et al., 2012), the five monoterpenes α-/β-pinene, d-limonene, Δ-3-carene, and camphene are estimated to contribute about 80 % to the total monoterpene emissions at the F-BEACh site. Besides monoterpene emissions, previous studies at the site have shown that average mixing ratios for isoprene are typically in the range of 0.27–0.50 ppbV. An overview on typically VOC mixing ratios at the site can be found in Klemm et al. (2006) and others (Grabmer et al., 2006; Graus et al., 2006).

Despite these relatively high mixing ratios for monoterpenes and isoprene, Plewka et al. (2006) already observed that early-generation oxidation products of isoprene and terpenes account only for a small part of the total organic carbon content of the particles at the site. In agreement to this work, similar concentrations in the lower ng m-3 range were found during the F-BEACh study for early-generation monoterpene oxidation products, e.g., pinonic acid (c = 2.9 (±2.8) ng m-3) and pinic acid (c = 4.7 (±2.5) ng m-3). However, since MBTCA and 3-carboxyheptanedioic acid are known to be formed via photooxidation of their monoterpene precursors (Jaoui et al., 2006; Szmigielski et al., 2007; Müller et al., 2012), this observation might indicate the occurrence of fast photochemical aging processes, eventually resulting in high abundances for these compounds during the campaign period. This hypothesis is also in agreement with high solar radiation values observed at the site, typically showing a maximum around midday at an average of 393 (±15) W m-2. In addition, real-time measurements of the organic aerosol fraction and trajectory calculations also suggest a photochemical source for these compounds, as will be discussed later on in the text. Besides several monoterpene oxidation products, a marker compound for sesquiterpene oxidation, i.e., β-nocaryophyllinic acid, could also be identified (van Eijck et al., 2013). However, the contribution of sesquiterpene oxidation products on particle composition cannot be estimated here because the observed concentrations were typically below the quantification limits.

As can be seen from Fig. 1, several sulfur- and nitrogen-containing compounds were found on the filter samples, i.e., CHOS, CHON, and CHONS. Similar to the CHO group, all these compounds have to exhibit a certain acidity, which allows the detection as [M-H]- ions in the negative ion mode. Therefore, the CHOS and CHONS compounds were assigned to organosulfates and nitrooxy organosulfates, respectively, which contain an acidic organic sulfate (R-OSO3H) functionality. It should, however, be noted that hydroxysulfonates are isobaric with organosulfates, and thus might contribute to a certain extent to this class. The CHON group might possibly comprise acidic organonitrates, although no further evidence can be given here. While only 4 % of the number of compounds were classified as CHON compounds about 47 % of the compounds belong to either the CHOS or the CHONS group. This large number of organosulfates and nitrooxy organosulfates is, however, not surprising since these compound classes are ubiquitously found in organic aerosol particles and readily accessible for deprotonation via electrospray ionization (Iinuma et al., 2007; Surratt et al., 2007, 2008; Hallquist et al., 2009; Altieri et al., 2009; Schmitt-Kopplin et al., 2010; Gómez-González et al., 2012; Lin et al., 2012; Shalamzari et al., 2013; O'Brien et al., 2014; Staudt et al., 2014; Nozière et al., 2015; Riva et al., 2015, 2016a, b). A comprehensive list of all sulfur- and nitrogen-containing compounds is given in the Supplement.

Several of the identified sulfur-containing compounds were already studied in the past and found in field and laboratory studies (Liggio and Li, 2006; Surratt et al., 2007, 2008; Altieri et al., 2009; Kristensen et al., 2011, 2016; Nguyen et al., 2012; Lin et al., 2012). In general, it is assumed that organosulfates and nitrooxy organosulfates have a mixed biogenic/anthropogenic origin, possibly involving particulate sulfuric acid, SO2, NOx, and radical-initiated chemistry (Surratt et al., 2008; Zhang et al., 2009; Nozière et al., 2015). As can be seen from Table 2, among the CHOS compounds several HOOS were found on the filter samples. This recently described compound class exhibits O / C ratios greater than 1.0 and is possibly connected to the presence of gas-phase HOMs and, thus, might have implications for new particle formation processes (Ehn et al., 2014; Mutzel et al., 2015).

Real-time analysis of aerosol particles reaching the site was carried out using a HR-ToF-AMS and the recently described AeroFAPA-MS (Brüggemann et al., 2015). While the AMS was used for a general classification of the aerosol particles' components in ammonium, sulfate, nitrate, chloride, and organics, the AeroFAPA-MS was resolving the organic fraction on a molecular level. In summary, the majority of the aerosol particle mass was classified as organic compounds (63.4 %), followed by sulfate (21.1 %), ammonium (8.7 %), and nitrate (6.7 %).

Figures 2 and 3 show the concentrations of organics, measured by the AMS, in comparison to the total ion current (TIC) measured by the AeroFAPA-MS. It is assumed that the TIC, i.e., the sum of all detected ions, only shows signals for organic compounds since inorganic species are typically not volatilized and ionized by the AeroFAPA ion source. Additionally, the particle number size distributions and the main directions of 96 h backward trajectories, calculated by HYSPLIT (HYbrid Single-Particle Lagrangian Integrated Trajectory; Draxler and Rolph, 2013), are given for the campaign period. As can be seen from Fig. 2, the signals for the 93 identified compounds by LC-HRMS and the signals of AeroFAPA-MS generally follow the same trends and are in agreement with the trend for the total organic aerosol mass, measured by the AMS. All three instruments show a maximum of signal intensities during the night of 21 July, which can be explained by particles with relatively large diameters (median diameter ∼ 150 nm) from regional sources reaching the site. This observation is further supported by HYSPLIT backward trajectories, exhibiting rather low altitudes and trajectory lengths (Supplement), as well as a strong increase in sulfate during the night (see also Fig. 5b). For the organic aerosol fraction a maximum concentration of 16.9 µg m-3 was determined by the AMS for this period. The days before 21 July are mainly characterized by trajectories coming from western Europe and northern Germany while afterwards the trajectories are arriving almost exclusively from eastern and northeastern Europe, i.e., Estonia and Russia.

Deviations between the signals for organics of the AMS and the total signals of the AeroFAPA-MS are mostly observed during nighttime, which is possibly due to the formation of non-acidic compounds, such as alcohols, aldehydes, or ketones, possibly formed by nighttime nitrate radical chemistry, eluding detection by AeroFAPA-MS. In contrast, compounds containing organic bonded sulfate, such as organosulfates or nitrooxy organosulfates, are readily measured by the AeroFAPA-MS and LC-HRMS, whereas the AMS cannot differentiate between inorganic sulfate and organic bonded sulfate. The same bias is typically observed for measurements of inorganic nitrate and organonitrates. Thus, all sulfate and nitrate signals of the AMS were assigned to the inorganic fraction, possibly leading to an underestimation of the organic aerosol mass (Liggio and Li, 2006; Farmer et al., 2010; Vogel et al., 2016).

In order to estimate the portion of organic compounds in aerosol particles that was measurable by AeroFAPA-MS, the signals of AMS organics and the TIC of the AeroFAPA-MS were plotted against each other. As depicted in Fig. 3a, the data of the two instruments exhibit a linear correlation for the entire campaign period. By calculation of a linear regression fit a correlation coefficient of R2 = 0.83 was determined, indicating that about 83 % of the variability of the organic aerosol mass can be explained by the AeroFAPA-MS signals. Furthermore, the AeroFAPA-MS signals ([M-H]-) of the 93 compounds, which were previously identified from LC-MS data as characteristic contributors to the organic aerosol fraction, were plotted as a function of the organic aerosol mass, determined by the AMS (Fig. 3b). Similar to the correlation of the TIC of the AeroFAPA-MS to the organic aerosol mass, a linear correlation was found for the 93 signals (R2 = 0.80). Taking the well-established AMS as a reference and neglecting measurement uncertainties, this correlation indicates that about 80 % of the organic aerosol's variability can be explained by these 93 signals, supporting the hypothesis that these compounds reflect the general behavior of the organic aerosol fraction at the site. Nonetheless, it should be noted that from the AeroFAPA-MS data an unambiguous formula assignment is not possible due to the unit mass resolution.

Figure 4a shows the summed AeroFAPA-MS mass spectra over the entire measurement period. As can be seen from this figure, the AeroFAPA-MS spectra support the aforementioned hypothesis that the composition of the particle phase reaching the site were influenced by BSOA marker compounds such as 2-hydroxyterpenylic acid (m/z 187, [M-H]-), MBTCA and 3-carboxyheptanedioic acid (both m/z 203, [M-H]-). Moreover, signals for pinonic acid (m/z 183, [M-H]-) and pinic acid (m/z 185, [M-H]-) remain quite low over the entire campaign period, as it was already observed for the filter samples, confirming the already mentioned low concentration of primary oxidation products at the sampling site. The ratio of signals for MBTCA and pinic acid, which can be used as aging proxy for organic aerosols (Vogel et al., 2016), shows an average value of 5.76, but even ratios of > 32 were observed for single days (Fig. S8 in the Supplement). Although this ratio is very specific for the instrumental setup of the AeroFAPA-MS and its ionization mechanisms, these extremely high values suggest that mainly air masses with aged aerosol reached the site. In agreement, HYSPLIT trajectory calculations reveal that arriving air masses typically traveled several days over land with distances of > 1500 km at low altitudes, often within the boundary layer (Figs. S1 to S6). Moreover, the majority of the trajectories are accompanied by high solar radiation without any precipitation along their way, leading to a high degree of solar radiation and, therefore, photochemical processing of the transported air masses (see Supplement). Since MBTCA exhibits a rather long atmospheric lifetime of ∼ 10 days (Nozière et al., 2015), the observed high abundance of MBTCA might, therefore, not be solely the result of rapid photochemically driven oxidation near the sampling site but could be influenced to a certain extent by long-range transport of organic aerosols. It should also be noted that there are two trajectories (20 and 21 July, Fig. S1) traveling along the Czech–German border mostly over coniferous forest for at least 24 h before arriving at the Waldstein site. Previously, these southeasterly wind directions have been related to regional new particle formation events observed at the site (Held et al., 2004).

In addition to the detection of lower molecular weight BSOA marker compounds, the averaged mass spectrum exhibits several signals in the higher m/z range at significant abundances. These signals might correlate to larger and more oxygenated compounds, as also observed from the LC-HRMS data. For example, AeroFAPA-MS signals at m/z 357 are in agreement with the LC-MS signals for a compound at m/z 357.1559 ([M-H]-) with the molecular formula C17H26O8. This compound was previously identified as a dimeric oxidation product of α-pinene (Yasmeen et al., 2010; Beck and Hoffmann, 2016). Additional AeroFAPA-MS signals in this mass range may also correspond to the formation of sesquiterpene oxidation products as described recently by others (van Eijck et al., 2013; Chan et al., 2011; Zhao et al., 2016). As an example, signals at m/z 255 might indicate the presence of β-nocaryophyllinic acid (C13H20O5), which is supported by LC-MS signals for this compound at m/z 255.1238 ([M-H]-).

Further agreement between LC-MS and AeroFAPA-MS in the higher m/z-range is also observed for nitrogen-containing compounds. In general, most acidic monoterpene oxidation products show odd m/z ratios in the mass spectra of the AeroFAPA-MS, since they only contain carbon, hydrogen, and oxygen atoms and are detected as [M-H]- ions. However, the sum of AeroFAPA-MS spectra also exhibits elevated signals at even m/z ratios, such as m/z 308, showing a high linear correlation (R2 = 0.76) to the organic aerosol mass measured by the AMS (Fig. 4b). According to the nitrogen rule, these signals correspond to nitrogen-containing compounds with an odd number of nitrogen atoms. To identify this compound, the LC-MS data were checked for signals at the nominal m/z ratio 308. In fact, a nitrogen-containing compound with the chemical formula C11H18O9N (m/z 308.0987, [M-H]-) was detected from the filter analysis at this nominal m/z ratio. Since the AeroFAPA-MS as well as the LC-ESI-MS are selective towards acidic compounds, this signal possibly indicates the presence of a highly oxidized nitrogen-containing carboxylic acid, such as a nitrooxy carboxylic acid. Similarly, a signal at m/z 250 is found in the AeroFAPA-MS spectra, showing a linear correlation to the AMS data. In this case, however, the signals of the LC-MS analysis exhibited quite low abundances. Nonetheless, one significant signal was identified at m/z 250.0208, representing C7H8O9N ([M-H]-). Due to the high oxygen content of these two compounds and their low corresponding signals from the LC-MS analysis, it is assumed that they possibly decompose during sampling, transport, storage, or processing of the filter samples. Moreover, nitrooxy compounds are also known to be prone to nucleophilic substitution by SO42-, forming the more stable organosulfate derivatives (Darer et al., 2011). Thus, online detection methods such as AeroFAPA-MS might allow a more reliable detection of such highly oxidized nitrooxy carboxylic acids in organic aerosols. Nonetheless, different ionization efficiencies and in-source formations might also have a significant effect on the detection of such compounds and should be investigated further in the future.

In order to investigate the presence of monoterpene-derived HOOS using real-time data of the AeroFAPA-MS, several representative compounds were chosen which were previously identified from the filter analysis (Table 2). The selection procedure for these representative compounds was based on certain criteria. Firstly, the HOOS were grouped according to their number of carbon atoms per molecule into C6 to C10 compounds. HOOS exhibiting carbon numbers < 6 were discarded since they are likely to be decomposition products of larger HOOS or isoprene-derived compounds. Secondly, since the AeroFAPA-MS exhibits only unit mass resolution the signals of the representative HOOS had to show a higher intensity on their nominal m/z ratio than any other signal for the LC-MS data. Moreover, special care was taken that no HOOS with identical nominal m/z ratios but different carbon chain lengths were chosen as representative, as is the case, for example, for C9H13O9S and C10H17O8S (both at nominal m/z 297). Eventually, the high mass resolution data of the LC-MS analysis were checked for HOOS that meet these criteria.

As can be seen from Table 2 and Fig. S7, for each of the C7 to C10 HOOS classes one appropriate compound was found. Nonetheless, no signal of the C6 HOOS matched the selection criteria for the AeroFAPA-MS signals. Therefore, signals for HOOS containing six carbon atoms will not be discussed here in order to avoid wrong assignments. For the C7 HOOS the signal at m/z 239.0231 was chosen, representing C7H11O7S ([M-H]-). This signal shows the highest abundance of all HOOS compounds from the filter measurements, as it was already observed by Mutzel et al. (2015), and almost no other signal was detected in significant abundances at this nominal m/z ratio. For the C9 HOOS an intense signal at m/z 267.0543 (C9H15O7S, [M-H]-) met the criteria and was selected. This finding is also in agreement with previous studies in which the C7 and C9 HOOS have been identified in laboratory and field measurements by Surratt et al. (2008). The m/z ratio at 285.0284, embodying C8H13O9S ([M-H]-), was chosen as representative for the C8 HOOS. Here, it should be noted that this signal was observed in lower abundances and some additional, but less distinct, signals were found at this nominal m/z ratio. A similar case was observed for the C10 HOOS for which the signal at m/z 327.0390 (C10H15O10S, [M-H]-) was chosen as representative. In general, for all of the selected compounds it was assumed that additional minor signals on the same nominal m/z ratio have negligible effects on the overall signal intensity for HOOS over the entire measurement period. Furthermore, it was assumed that AeroFAPA-MS detects the selected HOOS as [M-H]- ions and thus at the same nominal m/z ratio as for the LC-MS data. A comparison between the time traces of HOOS detected by LC-MS and AeroFAPA-MS is given in the Supplement (Fig. S10), showing similar trends for the two techniques and supporting the suitability of the selected criteria. In the following only the signals of these four representative HOOS are discussed.

Figure 5a depicts the signals for HOOS of the AeroFAPA-MS, which were plotted as a function of each other and checked for linear correlations among them. In addition, the particulate sulfate concentrations and relative RH are given by the marker size and the color code respectively. In general, all four HOOS classes show a linear correlation to each other, suggesting similar sources for these compounds. However, the group of C7 HOOS exhibits significant lower correlation coefficients of 0.51, 0.55, and 0.52 to the C8, C9, and C10 HOOS respectively. This decreased correlation might indicate that the source for the C7 class is somewhat different to the larger HOOS. In fact, while the signal for C7H11O7S- (m/z 239.0231, [M-H]-) shows the highest abundances of all HOOS from the filter sample analysis, as it was also reported by Mutzel et al. (2015), the AeroFAPA-MS measurements exhibit only low signals for this compound, further suggesting a different source than for the other HOOS. Possibly, this compound is a decomposition product of the larger HOOS compounds and is formed over time on the filter surface during sampling, storage, and/or processing of the sample. This hypothesis is further supported by comparing time traces for HOOS on single days where the signals for the larger HOOS classes differ clearly from the signals for the C7 HOOS (Fig. S9).

As can be seen from Fig. 5, all HOOS classes yield the strongest signals for high particulate sulfate concentrations, but no linear correlation could be observed over the entire campaign period except for single days. Moreover, the most intense signals for HOOS were observed during high RH periods, which coincided with a strong increase of the particulate sulfate concentrations, following the same trend as for the HOOS signals (Fig. 5b). This finding might suggest that aqueous-phase chemistry plays a major role for HOOS production, as it is known for other OS compounds (Herrmann et al., 2015). Furthermore, it should be noted that for the entire campaign period the particle acidity was very low and rather stable (average of HAer+ = 7.4 nmol m-3), indicating the presence of partially or even fully neutralized particles (Fig. S11). In contrast to previous studies, which suggest aerosol acidity to be one of the main factors driving organosulfate formation (Surrat et al., 2007, 2008; Iinuma et al., 2009; Gaston et al., 2014), no such effect was observed here. This result is, however, not contradicting previous findings but rather indicating that even at low particle acidities HOOS formation can be observed, as will be discussed in the following.

During the high RH period from 21 to 23 July, the sulfate concentrations and signals for HOOS show a linear correlation (R2 = 0.70). Figure 5b depicts the time trace of the signals for the HOOS and the particulate sulfate concentrations. Additionally, the RH is given by the color code. Within the first hours of 21 July the signals for HOOS and the sulfate concentrations still show some minor deviation; however, starting roughly from 09:00 CET both time traces follow almost exactly the same trend for the rest of this period, for which RH values mostly exceed 80 %. During such high RH periods dissolved HSO4- might react with HOMs after phase transition via a nucleophilic attack to give HOOS, as was already proposed previously (Mutzel et al., 2015). An additional or alternative pathway for the formation of HOOS might be the hydrolysis of hydroperoxide-containing HOMs as recently reported for methylglyoxal-, isoprene-, and alkane-derived hydroperoxides (Lim and Turpin, 2015; Riva et al., 2016a, b). Assuming that RH is a one of the main factors driving the uptake of gas-phase species into the particle phase (Shiraiwa et al., 2011, 2013), the required rapid phase transition of gas-phase HOMs is further supported by the observed trend for the sum of HOMs, measured by the CI-APi-TOFMS. In contrast to the diurnal behavior observed for other days of the campaign, the HOM concentrations show very low concentrations during this high humidity period (Fig. S12). As depicted in the figure, until ∼ 11:00 CET of 22 July the RH values are rather high (RH > 60 %). During this period, i.e., ∼ 18:00 CET of 21 to ∼ 11:00 CET of 22 July, the signal for gas-phase HOMs is hardly correlated to either the HOOS signal or the sulfate concentration and shows rather low abundances. However, as soon as the RH decreases to values below 60 %, the gas-phase concentration of HOMs exhibits an immediate and strong increase, roughly tripling the sum of HOMs within ∼ 1.5 h. It should be noted that some strong precipitation around midnight of 21 July led to the observed strong decrease in particulate sulfate as well as HOOS concentrations.

In order to further investigate the formation of HOOS and the role of possible precursor HOMs, single m/z ratios of the CI-APi-TOFMS were analyzed in more detail for 17 and 24 July. Since during the high RH periods the gas-phase concentrations of HOMs were extremely low due to a rapid phase transition (Fig. 6b), only these 2 days with low RH values were chosen for a further data analysis and discussion. In general, during dryer periods, signals with an odd m/z ratio in the mass spectra of the CI-APi-TOFMS dominate the sum of gas-phase HOMs (Fig. S12). As it was reported previously (Ehn et al., 2014; Jokinen et al., 2014), several of these compounds represent peroxyradicals (RO2⚫) which may act as precursors for closed-shell HOMs. Remarkably, the signals for three of the four identified RO2⚫ (i.e., C10H15O6⚫, C10H17O7⚫, C10H15O8⚫) and HOOS follow the same trends during the dry periods, possibly revealing a connection between these species. Solely the signals for C10H15O10⚫ (i.e., m/z 357, [M + NO3]-) exhibit a different behavior (Fig. S13). As an example, Fig. 6a depicts the time traces of a C10 HOOS (m/z 327, C10H15O10S, [M-H]-) and the most abundant RO2⚫, i.e., C10H15O8⚫ (m/z 325, [M + NO3]-), which might represent a possible precursor species for this HOOS compound. In addition, the gas-phase concentration of H2SO4 (divided by 4) and RH are given. As can be seen for 17 and 24 July, the signal for RO2⚫ is increasing with time, showing its maximum at ∼ 4 × 106 molecules cm-3 for both days around 11:00 CET, and is afterwards slowly decreasing again. In each case, the signal for the C10 HOOS follows the concentration of the RO2⚫. While the observed coinciding concentration profiles are not unambiguous for the limited available dataset, there might be a certain connection between RO2⚫ and the observed HOOS. In contrast, time series for a possible closed-shell HOM precursors, such as C10H16O10 (m/z 358, [M + NO3]-), show only a weak agreement with signals for HOOS (Fig. S14). As previously suggested, for example, by Kurtén et al. (2015), RO2⚫ contain acyl peroxy functionalities which might possibly undergo a nucleophilic attack by HSO4-, forming the corresponding HOOS, which has been discussed for closed-shell HOMs earlier by Mutzel et al. (2015). Such a mechanism would explain HOOS formation coupling to RO2⚫ in the particle-phase and/or at the interface. However, knowledge on the existence of such formation pathways still needs to be much better explored.

For both days a significant amount of gas-phase H2SO4 was also present, showing maximum values of ∼ 1.6 × 107 and ∼ 1.3 × 107 molecules cm-3 for 17 and 24 July respectively. It should be noted that in contrast the particulate sulfate concentration was rather high during 17 July (maximum at ∼ 5.7 µg m-3) but quite low for 24 July (maximum at ∼ 2.8 µg m-3). In principle, gas-phase H2SO4, which gets rapidly dissolved in the aqueous phase during high RH periods, might present an additional sulfur source for HOOS generation during such dryer periods. However, correlations during the campaign period were only observed for single days when high HOOS signals coincided with elevated gas-phase H2SO4 concentrations (Fig. 6b). Thus, due to the limited dataset and ambient conditions the exact reaction mechanisms for HOMs and HOOS cannot be discriminated here. In general, it can also be expected that the presence of precursor HOMs is the rate-limiting step in HOOS production, since concentrations for particulate sulfate as well as gas-phase H2SO4 are typically significantly higher. The aforementioned reaction mechanisms for the high RH periods, i.e., nucleophilic attack by HSO4-, might therefore still play an important role for the dryer periods.