Aerosol filter samples were collected at nine European rural background sites (Table 1, Fig. 1) for a fall period (17 September–15 October 2008; denoted fall) and a winter/spring period (25 February–25 March 2009; denoted winter/spring). For a description of the sampling sites, see Appendix A.

Ambient aerosol filter samples were obtained using various low volume filter samplers equipped with a PM10 inlet, collecting aerosol on pre-fired (850∘ C; 3 h) quartz fiber filters (Whatman QMA; 47 mm in diameter, batch number 11415138). The only exception was for samples collected at the Mace Head station, which used a high-volume sampler with a PM2.5 inlet. The samplers were operated at a flow rate ranging from 16.7 L min-1 to 1.71 m3 min-1, corresponding to a filter face velocity ranging from 20 to 69 cm s-1 (Table 1). The filter samples were collected according to the quartz fiber filter behind quartz fiber filter (QBQ) approach to provide a quantitative estimate of the positive sampling artefact of organic carbon (OC); thus the impact of the different filter face velocities at the various sites should be minimized. The sampling time was 1 week, and four samples were collected at each site for each of the two periods. At Mace Head, the collection of filter samples deviated slightly from the protocol in fall 2008, as the second week of sampling was divided into two to separate polluted air masses passing over the European continent for the first three days of the week and clean marine air masses for the last four days of the week. The sampling inlets were installed approximately 4 m above ground level, except at Mace Head (10 m). Post-exposure filter samples were placed in PetriSlides and stored in a freezer (-18 ∘C) to prevent degradation or evaporation of the analytes.

Total carbon (TC), elemental carbon (EC), and organic carbon (OC) were quantified using the Sunset Lab OC-EC Aerosol Analyzer (Birch and Cary, 1996), using transmission for charring correction and operated according to the EUSAAR-2 temperature program (Cavalli et al., 2010)

For the measurement of 14C(TCp) (14C of particulate TC), 0.2–2 cm2 punches, corresponding to 4–40 µg TC, were transferred into preheated quartz tubes (4 mm outer diameter) filled with ∼0.1 g cupric oxide. The tubes were connected to a vacuum line, cooled to -70 ∘C, evacuated to <10-3 hPa within 1 min, and then sealed. The sealed ampoules were heated to 850 ∘C for 4 h for oxidation of TC to carbon dioxide (Fahrni et al., 2010). 14C measurements were performed at the Laboratory of Ion Beam Physics of ETH Zurich, using the accelerator mass spectrometer MICADAS, equipped with a gas ion source (Ruff et al., 2007), which allowed a direct injection of the carbon dioxide after dilution with helium (Wacker et al., 2013). 14C results for the front filters were corrected for SVOC contributions using the TC mass of the corresponding back filters and the mean 14C result of the four back filters for the respective site and season. 14C(TCp) values are given as fractions modern (F14C), i.e., as the 14C/12C ratios of the samples related to the isotopic ratio of the reference year 1950 (Reimer et al., 2004). For determination of the non-fossil fraction of TCp (i.e., fnf(TCp) from 14C(TCp) determinations, a reference F14C value of pure non-fossil emissions of 1.08±0.04 was used to consider the different impacts of excess 14C from atmospheric nuclear bomb tests to fresh biomass and tree wood (Mohn et al., 2008). This is based on the assumptions that 50 % of non-fossil TC originates from fresh biomass and 50 % from burning of wood, whereof the latter includes 10-, 20-, 40-, 70-, and 85-year old trees with weights of 0.2, 0.2, 0.4, 0.1, and 0.1, respectively.

Quantification of the monosaccharide anhydrides (MAs) levoglucosan, mannosan, and galactosan was performed according to the method described by Dye and Yttri (2005), which has been successfully applied for aerosol samples ranging from the urban (e.g., Fuller et al., 2014) to the remote environment (e.g., Yttri et al., 2014).

For the analysis, punches (1.5 cm2) of the filter were spiked with 13C6 levoglucosan and 13C6 galactosan and extracted twice with 2 mL tetrahydrofuran under ultrasonic agitation (30 min). The filtered extracts (Teflon syringe filter, 0.45 µm) were evaporated to a total volume of 1 mL in a nitrogen atmosphere. Before analysis the sample solvent elution strength was adapted to the mobile phase by adding Milli-Q water (0.8 mL). The concentrations of the MAs were determined using high-performance liquid chromatography (HPLC) (Agilent model 1100) in combination with HRMS-TOF (high-resolution time-of-flight mass spectrometry; Micromass model LCT) operated in the negative ESI mode. Levoglucosan, mannosan, and galactosan were identified on the basis of retention time and mass spectra of authentic standards. Quantification was performed using isotope labeled standards of levoglucosan and galactosan. The mass traces at m/z 161.0455 and 167.0657 were used for quantification (approximately 50 mDa peak width).

The method described has been subject to intercomparison (Yttri et al., 2015).

An important use of the carbonaceous aerosol Latin hypercube sampling (LHS)-based source apportionment is to evaluate and constrain model systems for simulating particulate matter in the atmosphere. The EMEP MSC-W model (Simpson et al., 2012, 2017 and references therein) is an open-source chemical transport model widely used for research, within the EMEP and elsewhere (e.g., Simpson et al., 2007; Bergström et al., 2012, 2014; Dore et al., 2015; Ots et al., 2016; Vieno et al., 2016). In the present study, we run the EMEP model with a horizontal resolution of 50 km × 50 km across Europe, using 21 vertical levels, the lowest level being approximately 50 m thick. Meteorological data from the Integrated Forecast System model (IFS; Cycle 40r1) of the European Centre for Medium-Range Weather Forecasts (ECMWF) were used to drive the model. For this study, version rv4.15 of the model was used with some modifications: the OC emissions from all sources (except wildfires and open agricultural fires, which were treated as non-volatile in order to provide a tracer of these emissions but without adding the considerable uncertainties associated with aging of any assumed volatility basis set (VBS) components) were treated as semi-volatile and subject to evaporation and oxidation in the gas phase (aging), using a VBS approach, similar to the VBS PAA scheme in Bergström et al. (2012; the PAA scheme includes gas-particle partitioning of primary organic aerosol emissions and aging of all semi-volatile OA components in the gas phase). The model was run for the years 2008 and 2009, with two different emission setups (see Sect. 2.7.1) in order to evaluate model performance for biomass-burning-derived OC and EC with these inventories. Initial and lateral boundary conditions for the EMEP model are specified for most pollutants, as in Simpson et al. (2012). For organic matter (OM), the model assumes a background level of organic matter to represent OM transported into the modeling domain or otherwise not accounted for (e.g., marine aerosol, some primary biological aerosol particles, or very aged aerosol from outside the domain). In the initial setup of Bergström et al. (2012) and Simpson et al. (2012), we used 1.0 µg m-3 OM, but results presented in Bergström et al. (2012) and later studies suggested that this was too high. As in Bergström et al. (2014), we assume a background concentration of particulate OM of 0.4 µg m-3 (with an OM/OC ratio of 2.0) near the ground.

Concentrations of elemental carbon (EC), positive-artefact-corrected particulate organic carbon (OCp), organic carbon on back filters (OCB), positive-artefact-corrected particulate total carbon (TCp), and levoglucosan, as well as the EC/TCp ratio and the fnf(TCp) fraction observed during the fall 2008 and the winter/spring 2009 Intensive Measurement Periods, are presented in Table 2.

Fossil-fuel combustion was the major source of EC at all sites in fall, accounting for 6 % to 22 % of TCp, whereas EC from biomass burning was <8 % at all sites. The influence of ECff was particularly pronounced at the sites Montelibretti (22 %) and Lille Valby (21 %), which for Montelibretti could be due to the proximity of the Rome metropolitan area, with 3.7 million inhabitants. Lille Valby is a semi-rural site, and thus could be more influenced by, e.g., vehicular particulate emissions. Fossil-fuel combustion continued to be the most important source of EC in winter/spring for the five northernmost sites, whereas there was a shift towards biomass burning for the four southernmost sites. The relative contribution of ECbb and ECff to TCp in winter/spring was ≤10 %, except at the sites Lille Valby, Melpitz, and Birkenes, which experienced relative contributions of ECff exceeding 10 %. ECbb was a more abundant fraction of TCp in winter/spring compared to fall at all sites. The picture was less consistent for ECff, with a higher relative contribution in fall at the four southernmost sites and for Lille Valby and a higher fraction in winter/spring for the four other sites.

Biomass burning was the major anthropogenic source of OC at most sites in fall, accounting from 5 % to 36 % of TCp, whereas OC from fossil fuel ranged from 8 % to 21 %. The exceptions were Birkenes and Mace Head for which OCff dominated with 16 % and 21 %, respectively. At Montelibretti, OCbb and OCff made equally large contributions to TCp (18 % each).

In winter/spring, biomass burning was the major anthropogenic source of OC at all sites except at Mace Head, constituting 11 % to 46 % of TCp, whereas the range for OCff was 10 % to 23 %. OCbb was more abundant in winter/spring compared to fall for all sites but Mace Head, whereas there was no consistent pattern observed for OCff. There was a general tendency that OCbb became less abundant along a south–north transect, as seen for ECbb.

Biomass burning had a pronounced influence at most sites already in the first week of sampling in fall (17–24 September): ECbb and OCbb contributed a substantial 57 % of TCp at K-puszta and 54 % at Ispra and 34 % and 37 % at Melpitz and Payerne, respectively, whereas it ranged from 21 % to 29 % for the sites Mace Head, Košetice, and Lille Valby. Birkenes was the only site where wood burning made a minor contribution (6 %) in this first week. Model calculations suggest that wild and agricultural fires were of minor importance at all sites for the week, with the highest model-calculated concentration (0.02 µg C m-3) at Ispra and Lille Valby, corresponding to 3 % and 5 % of the modeled TCbb (see Sect. 5.2). Hence, residential wood burning appears to be the source of ECbb and OCbb, although given the uncertainties of emission estimates for wild and agricultural fires, such sources cannot be ruled out. The mean temperature during the first week of sampling was not noticeably lower than for the rest of the sampling period. Still, it was the week with the lowest mean temperature for the sites K-puszta, Payerne, and Košetice.

Wild and agricultural fires are major sources of carbonaceous aerosol (Bond et al., 2004) but with large regional, seasonal, and annual differences in emissions and occurrence (Hao et al., 2016; Korontzi et al., 2006). Agricultural waste burning is banned in most European countries; nevertheless, remote sensing data show such fire events in several countries, including those with a ban (Korontzi et al., 2006), and they appear to be particularly frequent in eastern Europe (e.g., Belarus and the Ukraine), in western parts of Russia, and in Central Asia. In most cases when natural vegetation catches fire in Europe, this is due to human activity (Winiwarter et al., 1999).

Incidences of wild and agricultural fires that severely deteriorate air quality in large parts of Europe are regularly reported, e.g., by Yttri et al. (2007a) for 2002, by Stohl et al. (2007) for 2006, and Diapouli et al. (2014) for 2010. The two periods discussed in the present study partly coincide with the time when concentrations from wild and agricultural fires peak in Europe (Korontzi et al., 2006). Levoglucosan by itself cannot differentiate between emissions from residential wood burning and wild and agricultural fires. Hence, we have used modeled concentrations to address the relative contribution of TC from wild fires and agricultural fires (TCwf) to the sum of TC from residential wood burning (TCbb) and TCwf for the two sampling periods.

There was an influence from wild and agricultural fires at all sites, with a higher mean contribution in fall (TCwf=0.05 µg C m-3), corresponding to 9 %–16 % (for base case or DT+IVOC) of modeled TCbb, than in winter/spring (TCwf=0.015 µg C m-3), corresponding to 2 %–4 % of modeled TCbb. TCwf was also typically low on a weekly basis, but for the last week of sampling in fall, a noticeable contribution was calculated for Ispra (34 %), K-puszta (31 %), and Montelibretti (16 %).

The major conclusion to be drawn from these results is that the model predicts that wild and agricultural fires make minor contributions to the biomass burning carbonaceous aerosol at the sites addressed and that residential wood burning is the major source.

Remaining non-fossil sources of OC (OCrnf) are typically associated with biogenic secondary organic aerosol (OCBSOA) and primary biological aerosol particles (OCPBAP); however there are anthropogenic sources of modern carbon as well, as discussed in detail by Yttri et al. (2011a). Here, we discuss the results obtained for OCrnf as if natural sources are dominating.

The OCrnf level varied more widely in winter (0.1–2.2 µg C m-3) than in fall (0.6–3.0 µg C m-3) (Fig. 2) and corresponds well with levels reported for the European rural background environment (Gelencsér et al., 2007; Genberg et al., 2011; Yttri et al., 2011a, b). The spatial distribution of OCrnf equaled that of OCp, with high concentrations at the southernmost sites and decreasing levels along a south–north transect.

OCrnf levels were higher in fall compared to winter/spring for all sites, but the difference varied between minor at most sites, moderate at the continental sites Košetice and Payerne, and substantial at the Norwegian site Birkenes. Studies consistently point towards BSOA as the major contributor to OCrnf in Europe (e.g., Simpson et al., 2007; Bessagnet et al., 2008; Yttri et al., 2011a); e.g., Gelencsér et al. (2007) showed that BSOA in PM2.5 was 1.6–12 times higher in summer than in winter for six European rural background sites. Hence, the observed pattern could partly be explained by a higher formation rate of BSOA in fall, propelled by larger emissions of BSOA precursors and a higher ambient temperature (see Table 1 ambient temperature values). In the present study, PM10 filter samples were collected (except at Mace Head, where PM2.5 was collected). Consequently, primary biological aerosol particles (PBAPs), typically residing in the coarse fraction of PM10 (e.g., Yttri et al., 2007b; Kourtchev et al., 2009; Bozzetti et al., 2016), could contribute to OCrnf as well. In Scandinavia, PBAPs peak in summer and fall, reflecting the vegetative season and the absence/presence of a snow cover (Yttri et al., 2007a, b, 2011a, b), and summertime OCPBAP concentrations (PM10), being 7–8 times higher than in winter, have been reported for two Norwegian sites (Yttri et al., 2011a). In continental Europe, the vegetative season is longer than in Scandinavia and a permanent snow cover is associated with high-altitude regions and rare occasions, lasting for short periods, in low-altitude regions. Hence, one could speculate that there is a PBAP emission flux in continental Europe in the heating season, which is comparatively larger than that observed in Scandinavia. We find support of this view in the study by Waked et al. (2014), which showed a tail of PBAPs and episodes with high PBAP concentrations in winter for an urban background site in northern France. Knowledge of PBAP concentrations in Europe is limited; thus we can only speculate about how much of OCrnf in the present study is due to PBAPs. A noticeable 20 %–32 % contribution of OCPBAP to TCp was found at four Nordic rural background sites in late summer (Yttri et al., 2011b). Similar figures (OC from primary biogenics constituting up to 33 % of OC in PM10) were reported for the densely populated region of Berlin in northeastern Germany (Wagener et al., 2012) in late summer and fall. Gelencsér et al. (2007) and Gilardoni et al. (2011) both reported levels of OC associated with PBAPs for an entire year for the European rural background environment, finding that the relative contribution to total carbon was <5 % in summer and <8 % in winter. However, both studies relied on PM2.5 samples, likely excluding the majority of PBAPs. Further, Gelencsér et al. (2007) accounted for plant debris only when measuring cellulose, whereas Gilardoni et al. (2011) only accounted for fungal spores, measuring arabitol and mannitol. Waked et al. (2014) found that 17 % of the OC was attributed to OCPBAP on an annual basis for an urban background site, with substantially higher concentrations in summer (37 %) and fall (20 %) compared to winter (7 %) and spring (6 %). At the rural background site Payerne, Bozzetti et al. (2016) found that PBAPs, mainly from plant debris, equaled the contribution of SOA to organic matter in PM10 in summer.

The non-fossil signal was typically most pronounced in fall, with the highest relative share (52 %–69 %) observed for the two low loading sites situated on the outskirts of Europe (Birkenes and Mace Head) and the lowest for the highest loading site, Ispra (23 %). Note that OCrnf obtained for Mace Head is a conservative estimate, as PBAPs typically residing in the coarse fraction are not accounted for, as PM2.5 filter samples were collected at this site. Nevertheless, OCrnf was the major fraction at Mace Head, regardless of season; hence, our conclusions would not change if the filter samples had PM10 cutoff size. A pronounced non-fossil signal (52 %–54 %) was seen for the continental sites Košetice and Payerne as well, whereas the relative share ranged between 38 % and 48 % for the remaining sites. Non-fossil OC was by far the major source of OC at all sites in fall, except at Ispra, for which biomass burning dominated. The non-fossil signal decreased, or remained unchanged, for all but one site going from fall to winter/spring, but the reduction was substantial only at the Norwegian site Birkenes (a factor of ∼2), at Payerne and Košetice (a factor of 1.5–1.7), and at Melpitz (a factor of 1.5). Still, non-fossil OC was the major source of OC at five sites, even in winter/spring, K-puszta, Košetice, Lille Valby, Mace Head, and Birkenes. It has been suggested that increased condensation due to lower temperatures could be an efficient way of forming BSOA, even in winter (Simpson et al., 2007). It is however difficult to argue for such a hypothesis only by looking at the observed ambient air temperatures during the winter/spring period. Another possibility is that some of the remaining non-fossil OC may be secondary organic aerosol formed from volatile or semi-volatile OC emitted from wood burning. OCbb determined based on levoglucosan may not include all SOA formed after aging of the gas-phase emissions, even if the emission ratios were derived from ambient measurements and likely include condensed vapors and secondary products.

In the current study, results obtained for OCrnf are discussed as if natural sources are dominating, despite the fact that anthropogenic sources can make a certain contribution, e.g., from cooking emissions and by anthropogenic enhancement of BSOA formation. EC and OC emitted from combustion of fossil fuel and biomass are considered entirely anthropogenic, as we define wild fires as anthropogenic.

In fall, the anthropogenic and natural influences were of comparable magnitude at most sites. Exceptions were Birkenes, with a clearly larger natural contribution (69 %), and Ispra, with a larger anthropogenic contribution (77 %), the latter affected by regional air pollution in the strongly polluted Po Valley region. For the other sites, the anthropogenic fraction ranged from 46 % to 62 % and from 38 % to 54 % for the natural fraction. Increased condensation due to lower temperatures can be an important source of BSOA in fall and winter, which could outweigh the effect of high temperature and increased terpene emissions in summer (Andersson-Sköld and Simpson, 2001; Simpson et al., 2007). Further, PBAPs can make a pronounced contribution in fall both in Scandinavia (Yttri et al., 2007a, b, 2011a, b) and in continental Europe (Waked et al., 2014; Bozzetti et al., 2016), and the fall peak of the northeastern Atlantic Ocean phytoalgal bloom takes place during the period in question, likely contributing with marine PBAPs at Mace Head (Ceburnis et al., 2011).

In winter/spring, anthropogenic sources dominated at all sites (60 %–78 % anthropogenic), except for Mace Head (37 %). Ispra also had the most pronounced anthropogenic contribution of all sites in winter/spring (78 %), and it was largely unchanged from that observed in fall. Three of the four sites experiencing a high natural influence in fall (Birkenes, Košetice, and Payerne) saw a major increase in the anthropogenic contribution going from fall to winter/spring. This was attributed to a substantial reduction in natural sources, accompanied by an increase in the anthropogenic sources, being primarily biomass burning at Payerne and Birkenes and fossil-fuel sources at Košetice. Residential wood burning is considered a decentralized source in Europe, and emissions from local sources can be substantial in winter (Szidat et al., 2007). A certain local contribution could also be speculated for Košetice, as small coal-fired ovens are still common in rural areas in eastern Europe (Spindler et al., 2012).

The EMEP MSC-W model was run with two different emission and SOA modeling setups (a base case and DT+IVOC) in order to reflect (to some extent) the very large uncertainties in both emissions and atmospheric processing of the primary organic aerosol (POA) (see Sect. 2.7). The model results were compared with that of the LHS analysis discussed above. In the following, model results that are within the 10th–90th percentile range of the LHS analysis are considered as being in agreement with the measurements. Results outside this (fairly wide) concentration range are considered as under- or overestimations.

Modeled OCbb and ECbb concentrations were compared to the LHS source apportionment results for each sample individually in Fig. 3 and as averages over the measurement periods in Table 4. The base-case model simulations underestimated OCbb severely at most sites (Fig. 3a). The only exception was Birkenes, for which the model slightly overestimated the LHS-derived estimates (the modeled OCbb were within the LHS 10th–90th percentile range for 3 out of 5 weeks, whereas 2 out of 5 weeks were overestimated). For the other sites, the mean underestimation of the LHS 10th percentile for OCbb ranged from -26 % at Lille Valby to -84 % at Payerne.

The model results for OCbb were clearly better with the DT+IVOC emission setup (Fig. 3b) than for the base case, at all sites except Birkenes and Lille Valby. For Košetice and Payerne, the modeled OCbb was within the LHS range for the majority of the samples, and the underestimation of OCbb was smaller than with the base case for Ispra, Montelibretti, K-puszta, and Melpitz. A few individual OCbb measurements were, however, clearly overestimated with the DT+IVOC setup (one sample each for Melpitz, K-puszta, and Lille Valby).

The results for ECbb roughly split in two groups for the base case (Fig. 3c): at Birkenes and Lille Valby, the ECbb concentrations were overestimated by the model most of the time; only for one sample at each site did the model ECbb fall within the LHS range. The average overestimation of the LHS 90th percentile was 69 % at Lille Valby and 43 % at Birkenes. At the other sites, ECbb was underestimated (with a few exceptions), with an average underestimation ranging from -34 % compared to the LHS 10th percentile at Melpitz to -84 % at Mace Head. For the two Italian sites the average underestimation was -38 %, whereas it was -39 % at K-puszta and Košetice and -60 % at Payerne.

The DT+IVOC model results were clearly better for ECbb, except for the Italian sites and K-puszta where the ECbb underestimation was larger due to lower emissions in the inventory of Denier van der Gon et al. (2015). ECbb was largely overestimated at the Scandinavian sites but not as much as for the base-case emissions. The modeled ECbb was within the 10th–90th percentile LHS range for five of the weeks at Košetice and Payerne using the DT+IVOC emissions, but there was still a tendency that levels were underestimated (one week was underestimated at Košetice, two at Payerne). For Melpitz the modeled ECbb was within the LHS range for 3 out of 6 weeks (2 weeks were underestimated and 1 overestimated).

The present comparison of modeled and LHS-derived biomass burning carbonaceous aerosol concentrations indicates that the base-case setup with the TNO MACC-III emission inventory, which is similar to official EMEP PM2.5 emissions estimates, likely underestimates emissions from residential wood burning substantially in large parts of Europe. This is in line with the findings of Denier van der Gon et al. (2015) and reflects that emissions are established following national practice that is inconsistent between countries. Note that the inventory POA emissions were distributed across different volatility classes for the DT+IVOC emissions, as for a typical VBS treatment, whereas we did not add IVOCs to the MACC-III emissions in our base case. Although the DT+IVOC emission setup with updated wood burning emissions and extra IVOCs improved the model results, large uncertainties still remain, and it cannot be excluded that wood burning emissions in some parts of Europe may be considerably larger than those estimated by Denier van der Gon et al. (2015).

The issue of long-range transport into Europe is important for some pollutants (especially ozone, e.g., Fiore et al., 2009, or carbon monoxide from forest fires, e.g., Forster et al., 2001). However, many years of measurements and modeling analyses support our assumption that the most likely sources of carbonaceous aerosols in our study are from Europe. For example, many years of analysis of aerosols at Mace Head on the west coast of Ireland give little evidence for aerosol transport from North America, with most organic matter (OM) assigned to marine or European sources (O'Dowd et al., 2014). Emissions from major wildfires in eastern Europe explained the highest OC and EC concentrations at Birkenes in 2001–2015 as did episodes of air pollution carrying the hallmark of long-range transport, i.e., elevated levels of secondary inorganic aerosol and air masses transported at low altitude over major emission regions in central and eastern Europe (Yttri et al., 2019). Meanwhile, elevated concentrations of equivalent black carbon (eBC) from fossil-fuel sources (eBCff) and from biomass burning (eBCff) at Birkenes were associated exclusively with source regions in continental Europe (Yttri et al., 2019). Consequently, long-range transport is of major importance for elevated concentrations of carbonaceous aerosol at Birkenes, but sources are confined to the European continent.

Further, modeling by Simpson et al. (2007) showed that observed levels of OC and EC could be reproduced quite well over a 2-year period (CARBOSOL study) at two sites on the western coast of Europe, Mace Head in Ireland, and Aveiro in Portugal, with no suggestion of missing background sources in the model. Tsyro et al. (2007) examined the EC concentrations for the same study and showed that European forest fires only had significant impacts for a few samples. We note that the modeling domain we use is rather large, covering all of Europe from approximately 40∘ W to 60∘ E and 30–90∘ N, such that we capture all major sources and air mass circulations within several days of transport. Global model results from the EMEP model (e.g., McFiggans et al., 2019) also suggest that OM generated over North America only makes a small contribution to European particulate matter levels.