For each sample, a part of the quartz-fibre filter was extracted three times with 5 mL of methanol (Optima^® grade, Fisher Scientific) under ultrasonic agitation in an ice-chilled bath for 30 min. The three extracts were combined, filtered through a Teflon^® filter (0.2 µm), and reduced by volume to approximately 200 µL under a gentle stream of nitrogen. Extracts of field blanks were prepared analogously.

Recent results show that the positive-mode ESI, which is less selective, can provide additional valuable information, especially regarding fresh emissions (Lin et al., 2018). Here, to compare with previous results, and, as the main aim is to characterize SOA, we have focused on the negative-mode, (-)ESI. The aerosol extracts were analysed using an ultra-high-resolution LTQ Orbitrap Velos mass spectrometer in the negative mode (Thermo Fisher, Bremen, Germany) equipped with a TriVersa Nanomate robotic nanoflow chip-based ESI source (Advion Biosciences, Ithaca, NY, USA). The analyses were conducted with an ionization voltage of 1.51 kV, a vaporizer temperature of 230 ∘C for the Zurich samples and 254 ∘C for the other samples, and a capillary temperature of 200 ∘C (with only a small impact on measurements; see Fig. S8 in the Supplement). The Orbitrap MS instrument was calibrated using an Ultramark 1621 solution (Sigma-Aldrich, UK). The accuracy of the m/z calibration was < 2 ppm. The mass resolution of the instrument was 100000 at m/z 400. Two or three replicate measurements were conducted for each extract (variability assessed in Fig. S9), and field blank extracts were analysed in the same way.

For each replicate measurement, approximately 38 mass spectral scans were averaged into one mass spectrum per measurement (representing 1 min measurement time). The gathered mass spectral information was processed with Xcalibur 2.1 (Thermo Scientific). In the first step, peaks exhibiting a signal below instrumental noise level were removed. We estimated the instrumental noise level as the 99.9th percentile of the signal recorded in regions of the mass spectrum where no signal is expected. In the second step, for all remaining peaks possible, molecular compositions were assigned to the signals using a tolerance level within ±5 ppm and the following constraints: 12C ≤100, 13C ≤1, 1H ≤200, 16O ≤50, 14N ≤5, 32S ≤2, and 34S ≤1. Among all the assigned molecular compositions, only molecular assignments in agreement with the following criteria described in detail in Kourtchev et al. (2013) were further considered.

Molecular assignments of a peak have to be within a defined mass difference from the measured mass (m/z). The tolerated difference between the molecular assignments and the measured mass of a peak was assessed based on the respective difference for nine known compounds for every analysed sample (on average 0.5 ppm).

Molecular assignments have to be in agreement with the following elemental ratios: (i) O:C ≤1.5, (ii) 0.3≤ H/C ≤2.5, (iii) 0≤ N/C ≤0.5, and (iv) 0≤ S/C ≤0.2.

Molecular assignments have to be consistent with a neutral formula with a positive-integer double-bond-equivalent (DBE, for any chemical formula CNCHNHONONNNSNS, where NC, NH, NO, NN, and NS represent the number of carbon, hydrogen, oxygen, nitrogen, and sulfur atoms): 1DBE=1-NH2+NN2+NC. Nitrogen-containing compounds have to be in agreement with the nitrogen rule.

Formulae including either or both 13C and 34S were only further considered in the presence of a counterpart including only 12C and 32S, respectively.

Peaks were only further considered if their intensity in the sample was at least 3 times larger than in the blank.

All properties, molar ratios, and chemical formulae presented in this paper refer to neutral molecules. Literature data were additionally also filtered with criterion (2) for comparability. Bulk elemental ratios (H/C, O:C, N/C, and S/C) and the number of carbons of the organic aerosol were computed as follows (Nizkorodov et al., 2011; Bateman et al., 2012): 2O:Cbulk=∑ixi×NO,i/∑ixi×NC,i,3H/Cbulk=∑ixi×NH,i/∑ixi×NC,i,4N/Cbulk=∑ixi×NN,i/∑ixi×NC,i,5S/Cbulk=∑ixi×NS,i/∑ixi×NC,i,6Cbulk=∑ixi×NC,i/∑ixi. O:C, H/C, N/C, S/C, and carbon number, as well as the probability distribution function (pdf) are weighted by the number of O, H, N, S, and C atoms in a compound i (NO,i, NH,i, NN,i, NS,i, NC,i) by the respective peak intensity xi such that the probability p of, for example, an H/C of h is defined as follows: 7pH/C=h=∑ixi|H/C=h/∑ixi. The compounds' aromaticity can be estimated by different parameterizations (e.g. aromaticity index, AI, and aromaticity equivalent, Xc). Xc (Yassine et al., 2014) has the advantage over AI (Koch and Dittmar, 2006) that (poly)aromatic compounds with significant alkylation are accurately classified. Therefore, we apply the widely used aromaticity equivalent (Xc) in this study, expressed as follows: 8Xc=3×DBE-m×NO+n×NS-2DBE-(m×NO+n×NS). Here DBE is the double bond equivalent (Eq. 1), NO and NS are the number of oxygen and sulfur atoms present in the molecule, respectively, and m and n are the number of oxygen and sulfur atoms involved in π bonds, respectively (both assumed to be 0.5, Yassine et al., 2014). For molecular formulae with an odd number of sulfur or oxygen, the sum (m×NO+n×NS) was rounded down to the closest integer and for compounds with DBE ≤m×O+n×SXc was set to 0 (Yassine et al., 2014). Compounds with Xc < 2.5 were considered non-aromatic, with Xc≥2.5 aromatic and with Xc≥2.7143 condensed aromatic (Yassine et al., 2014).

In order to understand the spectral profiles recorded in summertime Zurich, we start by describing similar spectra from Hyytiälä. The largest fraction of signal in Hyytiälä can be attributed to CHO compounds with H/C ∼1.5 and Cbulk 10.5 (H/C 1.48, Cbulk 9.7 for 2011 and H/C 1.50, Cbulk 11.3 for 2014; Figs. 4, 5, Table 2), characteristic of biogenic emissions. Compounds with 8 to 12 carbons (C8–C12), thought to arise from monoterpene oxidation, dominate the signals (Figs. 5, 6, 7). Additionally, CHO compounds with 13–16 carbons and 17–22 carbons significantly contribute to the signal. The C13–C16 compounds are thought to consist mainly of sesquiterpene oxidation products but may also be produced through reactions of monoterpene and isoprene RO2 radicals (Berndt et al., 2018), which is less probable in a boreal forest due to the low isoprene concentrations. Meanwhile, the latter class (C17–C22) of compounds is thought to consist mainly of dimeric oxidation products (e.g. Kristensen et al. 2016; Berndt et al., 2018; Frege et al., 2018). Larger oxidation products (C > 22) were detected but to a much lower extent.

The OA bulk composition in summertime Zurich was similar to that in Hyytiälä, especially in the C8–C12 range, and at both locations CHO compounds dominated the signal (Figs. 4, 5, 6). The (H/C)bulk was 1.49 for Zurich samples during summertime, which is similar to those for samples from Hyytiälä and to the oxidation products of BVOCs (e.g. α-pinene, pinic acid: C9H14O4, MBTCA, C8H12O6). However, compared to Hyytiälä, signals in Zurich were clearly shifted toward smaller molecules (Cbulk 7.6: Figs. 4, 6, 7, 8, Table 2). While a high fractional contribution to the signal of C13–C16 and C17–C22 compounds was observed in Hyytiälä 2014, these compounds contributed much less in Zurich (Figs. 5, 6, 7). Meanwhile, small molecules such as C4H6O5 (possibly related to malonic acid) and C5H8O5 (possibly related to hydroxyglutaric acid) exhibited a higher fractional contribution in Zurich during summer than in Hyytiälä 2014 (Figs. 4, 6, 7, 8). Some of these compounds were related to OH-radical-induced atmospheric aging of monoterpene SOA, especially at high NOx conditions, in ambient as well as in laboratory experiments (Zhang et al., 2018; Mutzel et al., 2015) but could also originate from other biogenic precursors such as isoprene (see Sect. 3.4.2). In the following, we will discuss the possible reasons for the differences.

The composition of BVOC emissions depends on various parameters, such as vegetation type and temperature. While many BVOCs lead to the formation of oxidation products characterized by H/C ∼1.5, Cbulk depends on the size of the carbon backbone of the initially emitted precursor and the degree of accretion. Thus, the composition of the BVOC emissions has an impact on Cbulk. Modelled biogenic emissions showed a higher isoprene (ISO, C5H8) to monoterpene (MT, C10H16) ratio in Switzerland than in Finland (Fig. S2, Jiang et al., 2019). The higher ISO/MT ratio in BVOC emissions in Zurich could contribute to the higher C3–C7 CHO compound contribution at this site (see above, Figs. 4, 5, 7, 8). Sesquiterpene/monoterpene emission ratios (SQT/MT) did not show a clear difference between Finland and Switzerland and is therefore not expected to be the reason for the observed enhanced abundance of C13–C17 compounds in Hyytiälä compared to summertime Zurich (see the NOx discussion in the next section). In addition, increased temperatures lead to higher BVOC emissions and may induce some effects on the ratio of ISO/MT (mainly driven by photosynthetic activity) and of SQT/MT (Zhao et al., 2017). Biotic stress acting on plants may influence the SQT/MT emission ratio as well. Zhao et al. (2017) reported ratios of 0.15 for unstressed and 3.5 for stressed plants at 22–25 ∘C. These various effects could contribute to the variability in the abundance of C13–C17 CHO compounds observed between Zurich (17 ∘C, average Tmax=21 ∘C) and Hyytiälä in 2011 (15 ∘C, average Tmax=18 ∘C) and 2014 (20 ∘C, average Tmax=24 ∘C).

In 2014, higher daily average temperatures (20 ∘C, average Tmax=24 ∘C) were recorded compared to 2011 (15 ∘C, average Tmax=18 ∘C). The mass spectral signature recorded in Hyytiälä in 2011 showed distinct differences with the observations made in 2014 at the same place and was also less similar to summertime Zurich (Fig. 7). In 2011 (lower T), C8–C12 compounds dominated the signal, while larger compounds (C13–C16 and C17–C22) contributed significantly to the signal in 2014 (Fig. 5). As highlighted by Kourtchev et al. (2016), the higher contributions of dimeric and trimeric BVOC oxidation products in 2014 could be related to higher precursor and SOA mass, which is in agreement with laboratory experiments presented in the same study. Temperature differences affect not only the emissions from the biosphere but also the ratio between particle- and gas-phase concentration of compounds as a function of their volatility. This could lead to an enhancement of less volatile dimeric compared to more volatile monomeric BVOC oxidation products at higher T (Fig. 8, Hyytiälä 2014 vs. 2011). Temperature also affects the particle-phase contribution of first- (more volatile, such as pinic acid) and second-generation (less volatile, such as MBTCA) gas-phase oxidation products (Zhang et al., 2010; Vogel et al., 2013; Müller et al., 2012; Donahue et al., 2012). This is consistent with the observed enhancement of MBTCA compared to pinic acid with rising temperatures (Fig. 8). This phenomenon partially explains the variability in the observed composition of monomeric BVOC oxidation products.

Kourtchev et al. (2016) observed an increasing fraction of smaller molecules (C3–C7) in the total observed signal from biogenic SOA at higher temperatures. The increase in the proportion of smaller compounds (C3–C7) occurs despite their increasingly high evaporation rates. This could be related to a higher fraction of first-generation products residing in the gas phase where they are prone to further oxidation, possibly also promoting fragmentation. Since the average temperature in Zurich during summer is 17 ∘C (average Tmax=21 ∘C) this would partially explain the enhancement of the fraction of lower-molecular-weight compounds (C3–C7) compared to Hyytiälä.

While laboratory monoterpene experiments show a large contribution of functionalized monomeric oxidation products to SOA, ambient measurements have revealed an enhancement of fragmentation over functionalized products with increasing NOx concentrations (Zhang et al., 2018). Fragmentation products of RO2+NO reactions and subsequent autoxidation could explain such observations. Since we observe a similar behaviour (Fig. 8) in this study, the higher (C3–C7) / (C8–C12) in summertime Zurich than in Hyytiälä can be related to enhanced NOx concentrations at urban sites (NOx summertime Zurich – 15 ppb; Hyytiälä – 0.5 ppb).

Dimeric monoterpene oxidation products (C17–C22) are mainly formed through RO2+RO2 reactions, while in the presence of NOx this reaction pathway is suppressed by radical termination reactions between RO2 and NO (Kristensen et al., 2016; Lehtipalo et al., 2018). This effect explains the considerable contribution of C17–C22 compounds in Hyytiälä, while they are largely absent in summertime Zurich (Figs. 5, 7, 8). Overall the enhanced NOx concentrations inhibit the formation of such dimeric C17–C22 compounds, leading to the smaller Cbulk in summertime Zurich than in Hyytiälä.

A fraction of the signal is related to compounds also consisting of nitrogen and/or sulfur (CHON, CHOS, and CHON) in summertime Zurich (32 %) and Hyytiälä (19 %, 2011 – 15 %; 2014 – 23 %). Both in Hyytiälä and summertime Zurich CHOS compounds contribute to the signal (Zurich – 14 %; Hyytiälä 2011 – 6 %; Hyytiälä 2014 – 22 %), while the summertime Zurich SO2 concentration (average 0.4 ppb) exceeds a typical Hyytiälä concentration (June–July–August Q75=0.15 ppb, not available during sampling periods). Such compounds were also detected in wintertime Zurich. This group of compounds exhibits a similar composition at both sites (H/C range of 1.3–1.9 and mostly C5–C12 compounds, though some larger compounds were also found in Hyytiälä 2014 with the 15 enhanced contributions of dimers, as expected). CHON compounds are enhanced in summertime Zurich (CHON: 13 %) compared to Hyytiälä (CHON – 4 %; 2011 – 7 %; 2014 – 1 %) but contribute clearly less than in wintertime Zurich. The CHON compounds have a similar H/C (0.8–0.9) at both sites but a higher Cbulk in Hyytiälä (C9–C12) than in summertime Zurich (C5–C8). CHONS compounds are observed in summertime Zurich (4 %) and cover an H/C range similar to CHOS compounds (1.4–1.9), but the signal can almost uniquely be explained by compounds with C9–C12 (most prominently C10H17O7NS). These compounds are largely absent in Hyytiälä (1 %), which might be explained by elevated NOx concentrations in Zurich.

Wood burning is an important wintertime source of OA in Central Europe (Herich et al., 2014; Lanz et al., 2010; Crippa et al., 2014; Zotter et al., 2014a; Daellenbach et al., 2017). In the following, filters from laboratory wood-burning experiments were used as a reference for understanding the influence of such emissions on wintertime pollution at different sites (Figs. 9, 10). During laboratory wood-burning experiments, aromatic CHON compounds with H/C between 0.8 and 1.0 and 5 to 8 carbon atoms (C5–C8) contribute a large fraction of the signal (43 %, Figs. 4, 5). Additionally, CHO compounds, not present in summer samples with C8–C12 and H/C 0.8–1.0, contribute significantly to fresh emissions (56 %). Since the contribution of CHON to biomass-burning aerosol increases at least initially during aging (65 %–78 % for aged and 43 % for fresh emissions), the relative contribution of CHO decreases (22 %–34 % for aged and 56 % for fresh emissions; Fig. 5). The composition during wood-burning episodes in Alpine valleys is similar to primary wood-burning emissions sampled in the laboratory (CHON – 47 %–58 %; CHO – 48 %–41 %; Fig. 5). In wintertime Zurich, the chemical composition is also characterized by a large contribution of CHON (43 %) and CHO (35 %) compounds to the signal but is smaller than at the Alpine valley sites (Fig. 5). However, additionally CHOS (12 %) and CHONS (10 %) compounds also contribute to the signal in wintertime Zurich, which is the case in neither the laboratory wood-burning experiments nor wood-burning episodes in the Alpine valleys (Figs. 5, 10). The composition of the laboratory wood-burning emissions, wood-burning episodes in Alpine valleys, and wintertime Zurich is clearly distinguishable from biogenic SOA by a higher contribution of CHON compounds (43 %–58 %) as well as a lower bulk H/C (0.93–1.24) and Cbulk (7.2–8.7) than for biogenic SOA (CHON – 1 %–4 %; H/C – 1.48–1.50; Cbulk – 9.7–11.3; Figs. 4, 5).

During the laboratory wood-burning experiments, compounds with H/C of 0.8–1.0 and 5–8 carbon atoms dominate the CHON family. In fact, only a few compounds, such as nitrocatechols and similar compounds (C6H5O4N, C7H7O4N, C8H9O4N), contribute to the signal. Compounds detected in a road traffic tunnel, which were related to vehicular emissions (C7H7O3N and C8H9O3N, Tong et al., 2016) were also detected in the primary wood-burning emissions but with much lower intensities than C7H7O4N and C8H9O4N. While SOA formation from vehicular emissions might proceed via the oxidation of aromatic species, the most important precursors in biomass smoke are oxygenated aromatics such as phenol, cresol, and catechol (Harrison et al., 2005; Platt et al., 2013; Bruns, et al., 2015, 2016, 2017; Schauer et al., 2001, 2002). While CHON compounds were already present in the fresh emissions and thus were directly emitted, the increasing (N/C)bulk indicates a strong additional secondary formation of such compounds (Figs. 4, 5, 9). In Magadino, San Vittore, and wintertime Zurich similar CHON compounds dominated the signal. Their composition suggests that the biomass-burning emissions observed in wintertime Zurich were further processed than in Magadino and San Vittore. During summer these compounds exhibited a much smaller contribution in Zurich (see more detail in Sect. 3.4 and Fig. 5).

The CHO compounds observed in the fresh wood-burning emissions during the laboratory experiments were characterized by a lower H/C (0.7 to 1.0) than biogenic SOA (1.5) (Figs. 5, 10). Anhydrous sugars with high H/C such as levoglucosan (C6H10O5, H/C 1.67, O:C 0.83), mannosan, and galactosan, directly emitted from cellulose pyrolysis during biomass burning were detected in the laboratory wood-burning emissions as well as in Magadino and San Vittore but only contributed little to the signal. A considerable amount of the CHO compounds in wood-burning emissions could be considered aromatic (Xc≥2.5) or even condensed aromatic (Xc≥2.7143, Fig. S4). These compounds are consistent with products from lignin pyrolysis (Fig. S4, Bertrand et al., 2017, 2018) and contributed significantly to the CHO signal recorded for primary wood-burning emissions but less for aged emissions (57 %, 49 %, 44 % of CHO aromatic; 5 %, 3 %, 3 % of condensed aromatic; and 38 %, 48 %, 53 % of non-aromatic for fresh, 10 h, and 30 h atmospherically aged emissions, respectively). The ambient wood-burning pollution showed a similar distribution of the CHO signal as aged laboratory wood-burning emissions (Magadino – 46 % aromatic, 6 % condensed aromatic, 48 % non-aromatic; San Vittore – 44 % aromatic, 4 % condensed aromatic, 52 % non-aromatic), and the detailed chemical composition was similar to the fresh laboratory wood-burning emissions (Fig. S4). With proceeding aging during the wood combustion experiments, more oxygenated compounds dominated the signal during the wood combustion experiments (centre at H/C of 1.0 and O:C of 0.5, Figs. 10, S4). These values are consistent with aqueous SOA from syringol–guaiacol–phenol formed through the reaction with hydroxyl radicals and excited states of organic compounds (Yu et al., 2014). Since wood burning is a known emitter of such compounds, it seems probable that the aged wood-burning emissions consisted of oxidation products of phenolic compounds (Schauer et al., 2001; Bruns et al., 2016, 2017). In wintertime Zurich, the CHO composition largely shows common features with aged laboratory wood-burning emissions (Figs. 10, S4). However, in wintertime Zurich the contribution of smaller CHO compounds (C3–C7) was higher (Figs. 5, 7, S3) and the contribution of aromatic compounds to the total CHO signal was clearly lower (22 % aromatic, 2 % condensed aromatic, 76 % non-aromatic) than for the laboratory wood-burning emissions (Figs. S3, S4, S5, S6). This indicates that additional processes leading to fragmentation play a role in the urban environment as already observed in summer. During summer the influence of aromatic compounds on the CHO signal was negligible (Zurich summer: condensed aromatic -0%, aromatic -3%, and non-aromatic -97%).

In wintertime Zurich, CHOS and CHONS compounds contributed significantly to the signal as opposed to the laboratory wood-burning experiments and Alpine valleys. These compounds had between 9 and 12 carbons and were characterized by an H/C between 1.5 and 2.0. During the laboratory wood-burning experiments (aging initialized with HONO) no SO2 was added during photochemical aging, which could explain the absence of these CHOS and CHONS compounds. On the other hand, a compound group with similar H/C and Cbulk was also found in summertime Zurich but not in biogenic SOA. Thus, they can be linked to neither biogenic SOA nor wood-burning emissions. The presence of this compound group indicates the importance of additional sources and/or processes in the formation of urban SOA.