The concentrations of VOCs were measured with an in situ thermal desorption–gas chromatograph–mass spectrometer (TD–GC–MS; Perkin Elmer Inc., Waltham, US). Studied compounds were hydrocarbons with 5 to 15 carbon atoms, which are known to be important SOA precursors. Based on their most probable origin, compounds were classified as aVOCs and bVOCs even though some bVOCs are also known to have anthropogenic sources. The studied aVOCs consisted of aromatic hydrocarbons (benzene, toluene, ethylbenzene, p/m-xylene, styrene, o-xylene, 3-ethyltoluene, 4-ethyltoluene, 1,3,5-trimethylbenzene, 2-ethyltoluene, 1,2,4-trimethylbenzene, and 1,2,3-trimethylbenzene). The analyzed bVOCs were isoprene, monoterpenoids (α-pinene, camphene, β-pinene, Δ3-carene, p-cymene, 1,8-cineol, and limonene), sesquiterpenes (longicyclene, iso-longifolene, β-caryophyllene, and α-humulene), and an oxidation product of β-pinene (nopinone).

VOCs were collected into the cold trap (Tenax TA 60–80/Carbopack B 60–80) of the thermal desorption unit (TurboMatrix 350, Perkin-Elmer Inc., Waltham, US) connected to a gas chromatograph (Clarus 680, Perkin-Elmer Inc., Waltham, US) coupled to a mass spectrometer (Clarus SQ 8 T, Perkin-Elmer Inc., Waltham, US). The hydrophobic cold trap was kept at 25 ∘C for the removal of humidity; 30 min samples were taken with a 1 h time resolution and a flow rate of 40 mL min-1. The main flow going to the instruments through fluorinated ethylene propylene (FEP) tubing (ca. 5 m length, i.d. 1/8′′) was approximately 0.8 L min-1. To also enable the measurements of highly ozone-reactive terpenes, a heated stainless-steel tube was connected to the main flow path to remove ozone before sampling (see Hellén et al., 2012a). For calibration, standards were injected as methanol solutions into sorbent tubes (Tenax TA 60–80/Carbopack B 60–80), methanol was flushed away in nitrogen (6.0) flow, and the tubes were thermally desorbed and analyzed as samples. Five-point calibration curves were used. For isoprene calibration, a gas standard from National Physical Laboratories (UK) was used. The method has been described in detail by Helin et al. (2020).

Size-resolved chemical composition of submicron particles, i.e., organics, sulfate, nitrate, ammonium, chloride, and refractory black carbon (rBC), was determined with a soot particle aerosol mass spectrometer (SP-AMS; Aerodyne Research Inc., Billerica, US; Onasch et al., 2012). The SP-AMS collected data with 2 min time resolution, of which half of the time the instrument operated in a mass spectra mode (mass concentrations) and half of the time in a particle time-of-flight (PToF) mode (mass size distributions). The measured particle size range of the SP-AMS is roughly from 40 nm to 1 µm. A collection efficiency (CE) of 1 was applied to the data as with this value total PM1 from the aethalometer (equivalent black carbon, eBC) and the SP-AMS (excluding rBC) was comparable with that from the differential mobility particle sizer (DMPS) operating at the site. The CE of 1 was larger than that usually calculated for the AMS (Middlebrook et al., 2012) or SP-AMS (Onasch et al., 2012), which could be due to the inaccuracy in the ammonium nitrate calibration. A relative ionization efficiency (RIE) of 0.1 was used for rBC based on the calibration with REGAL black (REGAL 400R black pigment, Cabot Corp.). However, due to the considerable uncertainties related to the quantification of rBC with the SP-AMS (e.g., imperfect laser beam alignment), black carbon (BC) concentrations presented in this paper are taken from the aethalometer. The SP-AMS data were analyzed with IGOR 6.37, SQRL 1.62A, and PIKA 1.22A software.

Equivalent black carbon measurements were conducted using a dual-spot aethalometer (AE33, Aerosol d.o.o., Ljubljana, Slovenia), which allows real-time measurement of aerosol light absorption at seven wavelengths (370–950 nm; Drinovec et al., 2015). The sampling flow rate was set to 5 L min-1, and the inlet cut-off size was 1 µm (sharp-cut cyclone, BGI model SCC1.197). The time resolution was 1 min. The filter tape was a M8060 and consisted of tetrafluoroethylene (TFE)-coated glass fiber filters.

The sources of eBC can be examined by analyzing the absorption spectra of light-absorbing material in particles as particles from fossil fuel and biomass combustion are characterized by different spectral dependencies. The source apportionment method based on the light absorption at different wavelengths is usually called an aethalometer model (Sandradewi et al., 2008) based on the multi-wavelength optical instrument typically used in the measurements. In the aethalometer model, BC concentrations from wood burning (BCwb) and fossil fuel combustion (BCff) are estimated by the following equations: 1BCwb=babs(470nm)-babs(950nm)⋅470950-αff470950-αwb-470950-αffbabs(950nm)⋅eBC and 2BCff=eBC-BCwb, where babs is an aerosol light absorption coefficient given by the AE33 at the wavelengths of 470 and 950 nm. Absorption Ångström exponents (α) of 1.1 and 1.6 were applied to fossil fuel (αff) and wood burning (αwb), respectively, as those values have been previously optimized for the measurement site (Helin et al., 2018).

Submicron particle number size distributions were measured using a DMPS (Knutson and Whitby, 1975). The DMPS includes a differential mobility analyzer (DMA; Vienna type), used for particle sizing, and a condensation particle counter (CPC; A20 Airmodus, Helsinki, Finland) for obtaining particle number concentrations for each size bin. The time resolution of the DMPS was 9 min, and the scanned particle size range was 6 to 1000 nm (mobility diameter, Dm), but due to a power source issue, the three smallest stages of the DMPS were excluded from data, and the size distribution was calculated only for the size range of 10–1000 nm. The DMPS size distribution was compared to that of the electrical low-pressure impactor (ELPI+, Dekati Ltd., Tampere, Finland; Järvinen et al., 2014), which operated at the site during the last week of the measurement campaign (5–12 September 2019). The number size distributions from the DMPS and ELPI+ were similar, indicating that the DMPS data were reliable after excluding the smallest stages of the DMPS. The DMPS number size distribution was converted to the mass size distribution by assuming spherical particles and a particle density of 1.42 g cm-3, which has been shown to be the average density of submicron particles at the site (Barreira et al., 2021).

Basic air quality parameters were also measured at the site. The concentration of NOx was measured by an APNA-370 analyzer (Horiba, Kyoto, Japan), O3 by using an ambient O3 monitor (APOA-370, Horiba, Kyoto, Japan), CO by APMA-360 (Horiba, Kyoto, Japan), and PM2.5 and PM10 concentrations by a tapered element oscillating microbalance (1405 TEOMTM, Thermo Fischer Scientific, Waltham, US) with a time resolution of 1 min. The mass concentration of coarse particles (PM2.5–10) was calculated by subtracting PM2.5 from PM10. Of meteorological parameters, temperature (T), relative humidity (RH), and precipitation were measured at the Helsinki supersite, while wind speed and wind direction were measured at a meteorological station above the roof level (53 m above the land surface) located approximately 900 m northwest of the measurement site. The mixing height was calculated using the model (MPP-FMI) presented by Karppinen et al. (2000). In a previous study of Järvi et al. (2023), they have found that the concentration levels at the street canyon are more affected by traffic rates, whereas in surrounding areas meteorological conditions dominate pollutant levels.

In order to investigate a long-range transport (LRT) episode detected at the site on 9–11 September 2019, the origins of the air masses were calculated using the NOAA HYSPLIT model (Stein et al., 2015; Rolph et al., 2017); 96 h back trajectories were calculated for every 6 h at a height of 100 m a.s.l. (above sea level).

Production rates of oxidized compounds (OxPRs) from VOCsi were calculated from their concentration, the concentration of the oxidant, and their respective reaction rate: 3OxPR=d[products]dt=∑VOCikOH+VOCi[OH]+kO3+VOCiO3+kNO3+VOCiNO3, where ki is the reaction rate coefficient of a VOC with an oxidant (OH, O3, or NO3), and [VOCi] is the concentration of the corresponding VOC or oxidant. Details of the reaction rate coefficients used in this study can be found in Table S1 in the Supplement. Concentrations of O3 were from the local measurements, while OH and NO3 radical concentrations were modeled using the ARCA box model as described in Sect. 2.4.

The SP-AMS data set was analyzed for the sources and types of OA with a positive matrix factorization algorithm (CU AMS PMF tool v2.08D; Paatero and Tapper, 1994; Ulbrich et al., 2009). The number of factors was varied from two to eight (Fig. S2), and the solution obtained with six factors provided the most reasonable results. The factors were identified as two hydrocarbon-like OA (HOA) factors referred to as HOA-1 and HOA-2, one semi-volatile oxygenated OA factor (SV-OOA), one low-volatility oxygenated OA factor (LV-OOA), one LV-OOA factor from long-range transport (LV-OOA-LRT), and a coffee roastery OA factor (CoOA). HOA-1 and HOA-2 correlated only moderately in terms of time series (Pearson R=0.42) and mass spectra (R=0.69), and therefore, they were supposed to represent different types of OA and were not combined. The results for seven and eight factors did not provide any additional information; in the seven-factor solution, HOA-2 was split further into two factors, whereas in the eight-factor solution LV-OOA-LRT was also divided into two identical factors. Two periods of high OA concentrations were excluded from the PMF data matrix. Those periods were (1) from 02:00 to 04:15 LT on 31 August 2019 (Saturday) and (2) from 23:00 LT on 31 August 2019 to 01:20 LT on 1 September 2019 (Saturday–Sunday night). The average OA and PM1 concentrations during period (1) were 65.3 and 67.1 µg m-3 and during period (2) 47.5 and 52.6 µg m-3, respectively. During those periods OA consisted purely of hydrocarbon fragments (similar to the HOA-1 factor), but when those cases were included in the data set, they distorted the calculation of the campaign and diurnal averages as well as the PMF analysis. The source for the high HOA-1 concentrations was not found, but since CO, CO2, and NOx concentrations did not increase during those periods, the source was not likely to be any typical combustion process.

A PMF solution with six factors was investigated for the rotational freedom by varying fpeak, a tool that allows a single one-dimensional transect through the multidimensional solution space to be explored (Ulbrich et al., 2009), and for the accuracy with bootstrapping and multiple seeds (Figs. S3 and S4). These validation tests showed that the six-factor solution was very stable. Also detailed figures on the residuals for the six-factor solution are given in Fig. S5. Residuals show that there was a small amount of unexplained mass during early morning, afternoon, and evening, and in terms of the mass spectra, the largest relative residuals were noticed for larger m/z's that had the smallest absolute signal values.

Besides OA, PMF was also applied to the mass spectra of organics accompanied by NO+ and NO2+ ions to explore the presence of organonitrates in the mass spectra of the PMF factors. As in the OA PMF analysis, PMF was run with up to eight factors with NO+ and NO2+ ions (hereafter called the OA+NO+/NO2+ solution). The seven-factor OA+NO+/NO2+ solution corresponded closely to the six-factor solution with OA since the seventh factor in the OA+NO+/NO2+ solution represented inorganic ammonium nitrate, consisting mostly of NO+ and NO2+ ions and contributing only 1 % to the total OA signal. The comparison of the PMF solutions for OA (six-factor solution) and OA+NO+/NO2+ (seven-factor solution) in terms of high-resolution mass spectra and mass concentrations is presented in Figs. S6 and S7. For the POA factors (HOA-1, HOA-2, CoOA), the correlation was good for both mass spectra and mass concentrations (R=0.994–1.00), while for the oxygenated OA factors (especially for LV-OOA and LV-OOA-LRT) there were small differences between the solutions. The OA+NO+/NO2+ solution was utilized only to assess the contribution of organonitrates to the PMF factors (Sect. 3.5.2), and all the other data shown in this paper were obtained from the PMF solution for OA (OA solution with six factors).

The sources of OA were investigated by the PMF analysis. PMF extracted six different types of OA at the traffic environment (Fig. 6), of which three can be considered to be POA (HOA-1, HOA-2, and CoOA). HOA-1 had a contribution of 14 % to OA. The mass spectrum of HOA-1 was similar to that from engine emissions, which have the largest signal for the hydrocarbon ions C4H9+, C3H7+, C4H7+, C3H5+, C5H9+, and C5H11+ at mass-to-charge ratios (m/z's) of 57, 43, 55, 41, 69, and 71, respectively (Canagaratna et al., 2004). HOA-1 also displayed a similar diurnal trend with the traffic-related components BCff and nitrogen oxides (Fig. 7a) having the maximum in the morning; however, BCff increased more sharply in the morning, whereas HOA peaked later at ∼ 09:00 LT. Also, the evening rush hour peak detected for BCff and NOx (15:00–17:00 LT), was not apparent for HOA-1. Overall, HOA-1 correlated only moderately with BCff (R=0.52) and NOx (R=0.53). Also the total concentration of aVOCs correlated with HOA-1 (R=0.67). Both species peaked during morning rush hour and in the evening, but relative to the morning peak, aVOCs had larger concentrations in the evening than HOA-1 (Fig. 7b). In terms of mass size distributions, HOA-1 was likely to be found in a relatively small particle size. Unit-mass-resolution m/z of 57, used as a surrogate for HOA-1, peaked at ∼100–150 nm during the period of high traffic emissions, but there was also a second mode at larger particle size (250–450 nm; Fig. S13).

HOA-2 had a smaller contribution to OA than HOA-1 (10 %), and it was more oxygenated than HOA-1. Also HOA-2 had a clear pattern for the hydrocarbon ions, but it was different from HOA-1. HOA-2 had more signal at lower m/z's, with the largest signal being for C3H5+, C3H3+, C4H7+, and C3H7+ at m/z's of 41, 39, 55, and 43, respectively, indicating that HOA-2 had more double bonds in the hydrocarbon ions, and therefore they were less saturated than in HOA-1. The mass spectrum of HOA-2 also had a distinct signal for the oxygenated ions C2H4O2+ and C3H5O2+ at m/z's of 60 and 73, respectively, that are usually considered to be typical ions for biomass burning OA (BBOA; Alfarra et al., 2007). In this study, these fragments are unlikely to be related to biomass combustion, as similar to HOA-1, HOA-2 peaked in the morning between 08:00 and 09:00 LT; however, the morning peak was smaller for HOA-2 than for HOA-1. At nighttime, the concentrations of HOA-1 and HOA-2 were almost similar. HOA-2 correlated moderately with BCwb (R=0.41), but the correlation was stronger for HOA-1 and BCwb (R=0.59) since BCwb also had a clear morning peak. This is possibly explained by the fact that the aethalometer model cannot resolve BCwb and BCff completely. Furthermore, traffic also emits carbon, which absorbs at near-ultraviolet and lower-visible wavelengths (brown carbon), which can be attributed to BCwb in the aethalometer model, regardless of its original source. The contribution of biomass burning to OA and eBC was likely to be small since the measurements were carried out in late summer, when ambient temperature was still quite high, and the site was located in the area with apartment buildings with no wood stoves or wood-heated saunas.

It can be speculated that HOA-2 and HOA-1 were related to the emissions from different types of vehicles. In the previous studies, it has been shown that, for example, the exhaust emission of diesel–electric hybrid and ethanol buses equipped with exhaust aftertreatment systems can contain C2H4O2+ (m/z of 60) and C3H5O2+ (m/z of 73) ions in their mass spectra (Saarikoski et al., 2017). It should be noted that SOA produced from the vehicle emissions is also likely to include these oxygenated ions (Timonen et al., 2017); however, it has been shown that modern exhaust aftertreatment systems reduce SOA emissions in general (Karjalainen et al., 2019).

Another source for POA at the traffic site was the local coffee roastery. The mass spectrum of CoOA had pronounced peaks at m/z's of 55, 67, 82, and 109, corresponding to the ions of C3H5N+, C3H3N2+, C4H6N2+, and C5H7N3+, respectively, those ions being characteristic for caffeine in the AMS mass spectra (Timonen et al., 2013). As can be seen from the time series, CoOA was detected sporadically, while most of the time its concentration was near zero. On average, CoOA comprised 7 % of total OA, but during its maximum concentrations, its contribution to OA was as large as 80 %. Regarding diurnal trends, OA from the coffee roastery was detected mostly between 07:00 and 14:00 LT, which agreed with the operation hours of the roastery. CoOA did not correlate with any of the inorganic SP-AMS species. Based on the wind direction data, CoOA was clearly associated with the south sector from the measurement site (Fig. S14), which is the direction of the coffee roastery (Fig. S1). CoOA has been observed earlier in Helsinki at the SMEAR III station (1 % of OA; Timonen et al., 2013), but compared to SMEAR III, CoOA concentration and contribution was much larger at the Helsinki supersite because it was much closer to the coffee roastery (the Helsinki supersite is ∼600 m north of the roastery, while SMEAR III is ∼1.5 km northeast of the roastery). Regarding the mass size distributions, m/z of 109, a characteristic unit-mass-resolution m/z for CoOA, peaked at ∼300 nm during an intense coffee roastery emission event (Fig. S13).

The concentrations of CoOA might have been overestimated in this study. Compared to the PM1 mass from the DMPS, the sum of eBC from the aethalometer and the SP-AMS species (excluding rBC) was clearly larger when the coffee roastery emissions dominated OA (Fig. S15). This is likely due to the larger relative ionization efficiency for organics in the coffee roastery emissions since a constant RIE value (default 1.4) was used for organics regardless of the composition. For LV-OOA, the impact of RIE was opposite to CoOA as PM1 from the SP-AMS and aethalometer was smaller than that from the DMPS when the LV-OOA fraction had the largest values. For the other PMF factors, the impact of RIE was less clear. Another reason for higher PM1 from the SP-AMS and aethalometer could be the enhanced collection efficiency in the SP-AMS. A constant CE of 1 was used in this study, but the collection efficiency calculated by Middlebrook et al. (2012) resulted in a CE varying in the range of 0.45–0.65. Even so, the CE did not seem to explain the difference in PM1 between the DMPS and the sum of the SP-AMS and aethalometer (Fig. S16).

Although there were several restaurants near the measurement site (Fig. S1), cooking-related OA was not found at the site. That can be explained by the fact that there were no street kitchens or outdoor dining places near the site. However, based on the method presented in Mohr et al. (2012) to estimate cooking OA in the ambient data set, the ratios of f55 to f57 and fC4H7+ to fC4H9+ for CoOA were close to those of cooking OA (Fig. S17). Therefore, it is possible to explain CoOA as cooking OA if meteorological data, prior knowledge of the local sources, and reference mass spectra are not available.

The largest fraction of OA (78 %) consisted of three types of oxygenated OAs that were likely to be related to SOA. Two of the OOA factors had similar mass spectra, with the largest signal for the CO2-related ions CO2+ and CO+ at m/z's of 44 and 28, respectively, and the ion C2H3O+ at a m/z of 43. Based on their mass spectra, these OOAs were classified as LV-OOAs. However, the time series of two LV-OOAs differed clearly as one of the factors had more stable concentrations throughout the measurement period, whereas the other one increased clearly at the end of the measurement campaign, when the air masses came from eastern Europe and Russia (9–11 September 2019; Fig. S9). Therefore, this LV-OOA is called LV-OOA-LRT. The LRT episode was defined by large concentrations of inorganic species, namely sulfate, nitrate, and ammonium, and LV-OOA-LRT had a strong correlation with sulfate (R=0.87) and ammonium (R=0.81), while the corresponding correlations with LV-OOA were less significant (R=0.35 and R=0.53, respectively.) LV-OOA had a rather flat diurnal trend, while LV-OOA-LRT had smaller concentrations from 02:00 to 10:00 LT than at the other times of the day. The contributions of LV-OOA and LV-OOA-LRT to OA were 10 % and 20 %, respectively. According to wind direction and speed data, LV-OOA-LRT was mostly related to western winds, whereas LV-OOA was associated with the south and southwest direction (Fig. S14).

During the LRT episode, the mass size distributions of unit-mass-resolution m/z of 44, representative of LV-OOA, peaked at ∼450 nm (Fig. S13). The large size was probably due to the condensation of gaseous species on particles during transport and aging, increasing their size. Previous studies have also shown that the largest average particle sizes are observed for atmospherically processed particles that have grown, for instance, during the long-range transport of the air mass (Niemi et al., 2005; Timonen et al., 2008).

The third OOA factor was classified as semi-volatile OOA as it had the largest signal for the ion C2H3O+ at a m/z of 43 followed by the CO2-related ions. Of all six PMF factors, SV-OOA had the largest campaign-average contribution to OA with a fraction of 40 %. The SV-OOA concentration was smaller from 11:00 LT to midnight than at the other times of the day, similar to nitrate, suggesting its semi-volatile character. Additionally, SV-OOA had a small increase around 14:00 LT in the afternoon that could be due to the SOA formation in the afternoon. SV-OOA was at least partly related to biogenic SOA (discussed later in detail). During large biogenic emissions (see explanation in Fig. S13), unit-mass-resolution m/z of 43 peaked at ∼350 nm. Compared to the other sources, the size for a m/z of 43 was larger than the m/z of 57 for traffic emissions and the m/z of 109 for coffee roastery emissions but smaller than the m/z of 44 for the LRT episode.

Of the six PMF factors, LV-OOA-LRT was clearly the most oxygenated factor and had the largest oxidation state (Table 1). That was expected as long-range-transported OA had already spent several days in the atmosphere and was exposed to the oxidants before arriving in Helsinki. Also LV-OOA was highly oxygenated, whereas SV-OOA was much less oxygenated. As anticipated, primary OA sources were the least oxygenated factors. The same pattern was observed when the PMF factors were placed in the O:C–H:C space (Van Krevelen diagram in Fig. 8); POA factors were located at smaller O:C and larger H:C values than the OOA factors.

Oxidation of VOCs under various environmental conditions produces a variety of gas- and particle-phase products that are relevant for atmospheric chemistry and SOA production. To describe this, the production rates of oxidized compounds were calculated for studied VOCs from the VOC reactions as described in Sect. 2.3. The main local sources of oxidation products were O3 oxidation (OxPRO3) of bVOCs (66 % of total OxPR) and OH radical oxidation (OxPROH) of aVOCs and bVOCs (25 % of total OxPR; Fig. 9). OxPRO3 stayed relatively constant over the day, while OxPROH was significant only during daytime. This is expected since OH radicals are produced in photochemical reactions only during light hours. NO3 radical oxidation had only 8 % contribution to total OxPR. In an earlier study in a forest environment in Finland, O3 oxidation was found to be a major oxidation pathway (Hellén et al., 2018), and even with much higher mixing ratios of aVOCs, it was also the case here. However, at least part of the bVOCs detected here are emitted from anthropogenic sources, such as personal care and cleaning products. Nonetheless, this describes only the local situation in the traffic environment, and with a bit more regional perspective, the situation may change.

On average, aVOCs produced 18 % of the oxidation products at the site and had a major contribution to the OH radical oxidation (72 % of OxPROH). Oxidation of aVOCs with NO3 radicals had a low contribution (0.3 %) and no contribution to O3 oxidation. The major aVOC for OxPR was p/m-xylene (36 % of aVOC OxPROH), followed by toluene (10 % of the aVOC OxPROH). Even though the contribution of trimethylbenzenes was only 13 % of the aVOC OxPROH, their impact on SOA formation may still be significant due to their higher SOA formation potentials.

The contribution of bVOCs to total OxPR was 82 %. O3 oxidation was the main oxidation pathway for bVOCs, with 82 % contribution. Contributions of OH and NO3 radicals were 9 % and 10 %, respectively. OxPRO3 was driven by three monoterpenes, (α-pinene, limonene, and terpinolene) and a sesquiterpene (β-caryophyllene), with 20 %, 13 %, 33 %, and 26 % contributions to OxPRO3, respectively. For OxPROH, isoprene and α-pinene were the most significant bVOCs. NO3 radical oxidation was driven by four monoterpenes (α-pinene, 3Δ-carene, limonene, and terpinolene), which had a 92 % contribution to total OxPRNO3. It is clear that bVOCs with lower concentrations and sesquiterpenes with very low concentrations (∼0.004 µg m-3) also have a significant effect on local chemistry and, due to their high SOA yields (e.g., Lee et al., 2006), possibly also on the SOA formation. During summertime, when bVOC emissions from vegetation as well as their ambient air concentrations are higher (Hellén et al., 2012b), their contribution is expected to be even more significant.

There are also other VOCs, e.g., non-methane hydrocarbons and oxygenated volatile organic compounds (OVOCs) including ethanol, acetone, formaldehyde, and acetaldehyde, in urban air with possibly even higher concentrations (e.g., Hellén et al., 2006) than VOCs measured in this study. Most of them are more volatile and less reactive, and even with high concentrations, their oxidation products are not expected to have significant impacts on local SOA formation. As recent studies on volatile chemical products (McDonald et al., 2018; Coggon et al., 2021; Pennington et al., 2021) show, it is highly probable that there are also other VOCs (e.g., siloxanes and intermediate volatile organic compounds, IVOCs) which could contribute to SOA production. They estimate that volatile chemical product (VCP) emissions, which are not traditionally considered to be a significant VOC source, may be as high as traffic emissions. However, the total OH reactivity measurements in urban ambient air indicate that missing OH reactivity in urban areas has not been this high (see the review by Yang et al., 2016). VCP emissions include lots of different compounds, but part of the VCP emissions are aromatics and terpenes, which were also measured in this study and are measured in most total OH reactivity studies. This could also explain why actual missing reactivity in urban air has not been as high as missing VCP emissions.

SOA formation was studied by comparing the diurnal trends of the OOA factors with the diurnal trends of the modeled OH and NO3 radical concentrations (Fig. 10). LV-OOA had only a small variation throughout the day; however, the largest concentrations were measured in daytime, suggesting that its source is likely to be OH radical reactions with aVOCs. LV-OOA also had a peak during the morning rush hour from 06:30 to 09:00 LT that was not seen for OH; however, the relative contribution of LV-OOA to OA was smallest in the morning, suggesting that the increase was probably due to the low mixing layer height in the morning. Also the advanced exhaust aftertreatment in vehicles can possibly increase the direct emissions of LV-OOA. For instance, the study of Arnold et al. (2012) indicated elevated exhaust concentrations of organic acids as a consequence of oxidation processes in the exhaust aftertreatment devices. Thus, the distinctive peak during the morning rush hours can also be caused partly by direct emissions of low-volatility organic compounds and their condensation to the particulate phase immediately after the emission. The diurnal trend of SV-OOA differed from that of radical concentrations, being smaller in daytime. This indicates that the main factor behind the diurnal trend of SV-OOA was ambient temperature as SV-OOA is likely to be semi-volatile. LV-OOA-LRT had a slightly larger concentration in daytime than in the early morning hours, but this was likely to be due to meteorological parameters such as wind direction as LV-OOA-LRT was already highly oxygenated when it arrived in Helsinki. On the other hand, the diurnal variation in LV-OOA-LRT had a correlation with OxPRO3 (R=0.71; Fig. S18). Ozone is also strongly related to long-range transport in Finland, but due to the short lifetime of terpenes, this production would be quite local.

Nopinone, an oxidation product of the monoterpene β-pinene, correlated with SV-OOA when the intense long-range-transported episode at the end of the measurement period was excluded (R=0.71). Nopinone is produced in the air through the oxidation of β-pinene by OH (Calvert et al., 2011; Kaminski et al., 2017) and ozone (Grosjean et al., 1993; Hakola et al., 1994; Winterhalter et al., 2000). The concentration of nopinone is a balance between the production from these reactions and its own oxidation by OH (Hellén et al., 2018). The main source of β-pinene is expected to be vegetation, but also some anthropogenic sources are possible (e.g., personal care products or cleaning agents).

The high correlation of the nopinone concentration with the SV-OOA factor supports the fact that SV-OOA is at least partly related to biogenic emissions. Both nopinone and SV-OOA had maximum concentrations just before midday (Fig. 11). This can be at least partly explained by the mixing layer height and oxidation rates. During the night, monoterpenes (and other VOCs) often accumulate in the air due to low mixing layer and lower reaction sinks (Hellén et al., 2018), as is also seen here by high early-morning concentrations including nopinone precursor β-pinene (Fig. 3d). After sunrise, when OH radical production starts, oxidation rates increase, and more oxidation products (including nopinone) are formed. The mixing layer height was still relatively low during the morning hours, and nopinone and SV-OOA concentrations increased. Later during the day, when high production of semi-volatile compounds continued, dilution started to play a role due to the higher mixing layer, and the concentrations decreased. This also indicated the local origin of SV-OOA. Nonetheless, the diurnal variation in SV-OOA and nopinone was relatively small. This could be explained by the production of semi-volatile compounds from O3 and NO3 reactions of primary VOCs also during the night. In terms of ambient temperature, OA was located close to the SV-OOA factor in the Van Krevelen diagram when the temperature was high (Fig. 8), which agrees with the larger VOC emissions from vegetation at higher temperatures.

SOA formation was also studied in terms of organonitrates. In order to investigate secondary organonitrates in the mass spectra of the PMF factors, NO+ and NO2+ ions were added to the PMF input data matrix. Both inorganic nitrate and organic nitrate consist dominantly of the ions NO+ and NO2+; however, the ratio of NO+ to NO2+ (NO+/NO2+) is different for organonitrates and ammonium nitrate and therefore allows the determination of organonitrates in OA (Farmer et al., 2010). Secondary organonitrates are formed mainly during dark aging, via gas-phase NO3 reactions and RO2-NO reactions (Atkinson, 2000; Kiendler-Scharr et al., 2016). Fry et al. (2009) have published a NO+/NO2+ value of ∼10 for the particles produced in the reaction of α-pinene with nitrate radicals, whereas Bruns et al. (2010) have measured NO+/NO2+ values of 10–15 and ∼5 for the reactions of nitrate radicals with monoterpenes and isoprene, respectively. In this study, SV-OOA and LV-OOA had NO+/NO2+ ratios of ∼6 and ∼63 (Table 1), suggesting the presence of organonitrates in the SV-OOA and LV-OOA factors. For comparison, for pure ammonium nitrate salt and PMF factors consisting almost solely of ammonium nitrate fragments the ratio was much lower, 0.8–1.0 and 1.2, respectively. Of the local production of the oxidation products, 10 % was estimated to be from nitrate radical reactions of bVOCs (Fig. 9).

LV-OOA-LRT had an extremely low NO+/NO2+, which can be explained by the fact that the oxidation products of nitrate radicals can be further oxidized with daytime oxidants (Tiitta et al., 2016). In terms of the POA factors, HOA-2 had a large NO+/NO2+, so it can be speculated that HOA-2 has been oxidized with a nitrate radical during the night; however, the contributions of NO+ and NO2+ were small in the HOA-2 factor, causing a high uncertainty in the ratio. Furthermore, HOA-2 also had the second-largest nitrogen-to-carbon ratio (N:C; 0.014) after the CoOA factor (Table 1), which included several N-containing ions related to the caffeine (e.g., C3H5N+, C3H3N2+, C4H6N2+, and C5H7N3+).