Radiocarbon analysis of elemental and organic carbon in Switzerland during winter-smog episodes from 2008 to 2012 – Part 1: Source apportionment and spatial variability

. While several studies have investigated wintertime air pollution with a wide range of concentration levels, hardly any results are available for longer time periods covering several winter-smog episodes at various locations; e.g., often only a few weeks from a single winter are investigated. Here, we present source apportionment results of winter-smog episodes from 16 air pollution monitoring stations across Switzerland from ﬁve consecutive winters.


Introduction
Ambient particulate matter (PM) influences the Earth's climate directly by scattering and absorbing solar radiation and indirectly by modifying cloud microphysics (Pöschl, 2005;IPCC, 2013). In addition, aerosol particles also adversely affect human health as they can cause respiratory and cardiovascular diseases which can lead to increased mortality (Pope and Dockery, 2006;WHO, 2006). In Alpine regions and most parts of Switzerland elevated PM concentrations are often found during wintertime since topography (e.g., alpine valleys) and frequent thermal inversions favor the accumulation of pollutants (Gehrig and Buchmann, 2003;Ruffieux et al., 2006). Environmental pollution control strategies and policies have focused mainly on emissions from fossil fuel combustion so far (e.g., road traffic and industry). However, many recent studies have shown that wood burning emissions from domestic heating can be the dominating source of carbonaceous aerosols during the cold season, in Europe (e.g., Szidat et al., 2006Szidat et al., , 2007Lanz et al., 2008Lanz et al., , 2010Favez et al., 2010;Gilardoni et al., 2011;Harrison et al., 2012;Herich et al., 2014 and references therein). Therefore, the quantification of the fossil and non-fossil, especially wood burning, contributions to PM, particularly for days with high PM concentrations, is crucial for establishing effective mitigation strategies.
Carbonaceous particles are a major fraction of the fine aerosol (PM 2.5 , PM < 2.5 µm), contributing from 10 up to 90 % of the PM mass (Gelencsér, 2004;Putaud et al., 2004;Jimenez et al., 2009). Carbonaceous aerosols are further classified into two sub-fractions: elemental carbon (EC) and organic carbon (OC) (Jacobson et al., 2000). EC originates from incomplete combustion of fossil and non-fossil fuels (e.g., coal, gasoline, diesel, oil and biomass), exclusively emitted directly as primary aerosol into the atmosphere. In contrast, OC may be either primary OC (POC) directly emitted into the atmosphere or secondary OC (SOC) formed in the atmosphere through the oxidation of volatile organic compounds (VOCs) from both fossil (coal combustion, industrial and vehicle emissions) and non-fossil (e.g., wood burning and biogenic emissions as well as cooking) sources (Jacobson et al., 2000;Pöschl, 2005;). Among several techniques applied to identify and quantify carbonaceous aerosol sources, radiocarbon ( 14 C, halflife = 5730 years) analysis is a quantitative tool for unambiguously distinguishing fossil and non-fossil sources. 14 C is completely depleted in emissions from fossil-fuel combustion, which can therefore be separated from non-fossil car-bon sources which have a similar 14 C signal as atmospheric carbon dioxide (CO 2 ) (Szidat, 2009;Heal, 2014). The most detailed information about different sources can be achieved when 14 C measurements are performed on OC and EC separately, since EC originates exclusively from biomass burning and fossil fuel combustion. By contrast, the apportionment of OC into these two sources using this methodology is less straightforward due to the complex primary and secondary sources of this fraction.
Radiocarbon-based source apportionment results available in the literature are often reported from measurement campaigns covering rather short periods (e.g., several days or a few months, see Hodzic et al. (2010), Minguillón et al. (2011) andHeal (2014) and references therein for a summary of several publications). Very few studies present annual or seasonal results from a full year or several seasons. For example, only two 14 C data sets are available covering a time period of 2 full years Larsen et al., 2012), while only a few studies present a yearly cycle (e.g., Huang et al., 2010;Ceburnis et al., 2011;Genberg et al., 2011;Gilardoni et al., 2011;Zhang et al., 2014) or data from two consecutive summers (Tanner et al., 2004) or winters (Glasius et al., 2011). In addition, 14 C results from the same time period are available simultaneously only for a limited number of stations (usually less than five, see Heal (2014) and references therein). Furthermore, only a few groups worldwide perform 14 C measurements of the EC fraction, since such analyses are still challenging and since there are still open questions concerning the optimal approach for the EC isolation for 14 C analysis Bernardoni et al., 2013;Szidat et al., 2013;Dusek et al., 2014). As a consequence, results of 14 C measurements carried out separately on EC and OC are still very scarce (see Minguillón et al. (2011) andHeal (2014) and references therein).
In this study, we present, to the best of our knowledge, for the first time 14 C measurements covering a time period of five years. Aerosol filter samples were collected during winter-smog episodes (days exceeding the Swiss daily PM 10 limit of 50 µg m −3 ), at 16 air pollution monitoring stations across Switzerland to provide a good spatial resolution as well as different source characteristics in various area types (e.g., urban, suburban, rural, alpine valley, traffic, background, etc.). These samples were analyzed for the 14 C content in EC and OC, levoglucosan, and major water soluble ionic species. The duration of this project together with the large number of stations results in one of the world's largest aerosol 14 C data sets available. This paper is the first paper of a two-part series investigating the spatial and temporal variability in the fossil and non-fossil sources of the organic and elemental carbon during high pollution events in Switzerland. This paper presents the 14 C-based source apportionment results of carbonaceous aerosols and investigates their spatial variability. The second paper will explore the influence of meteorological parameters on the dif-ferent carbonaceous components, their temporal variability and their possible trends in the last years .

Aerosol sampling
The filter samples analyzed in this study were collected at four stations of the Swiss National Air Pollution Monitoring Network NABEL and 12 stations of the Swiss Cantonal air pollution monitoring networks (EMPA, 2013;Cercl'Air, 2012). These were selected such that a good spatial distribution across Switzerland is achieved (see Fig. 1). In detail, eight stations (PAY, SOL, SIS, BAS, REI, BER, ZUR and STG) are located on the Swiss Plateau, one station each in the Rhine and Rhone valley (VAD and MAS, respectively) and one station (SCH) in a small alpine valley in central Switzerland. Those 11 stations will be further referred to as stations "north of the Alps". In addition, five sites "south of the Alps" were selected. These include stations at the Italian border where the terrain is more open (e.g., station CHI), plus other stations enclosed within narrow valleys (e.g., stations SVI and ROV). The locations of the stations are shown in Fig. 1 and related details are listed in Table 1. Furthermore, the selection of the stations was also carried out such that the full range of different station characteristics (from urban/traffic to rural background, see Table 1) was covered.
At the selected sites, aerosols samples were collected onto quartz fiber filters (Pallflex 2500QAT-UP) for 24 h on a regular basis (every 2nd or 4th day or daily depending on the station) using high-volume samplers (Digitel DHA-80, Switzerland) operating at a flow rate of 500 L min −1 and equipped with PM 10 inlets. After the sampling, filters were wrapped in aluminum foil or lint free paper, sealed in plastic bags, and stored at −20 • C until analysis. Filter sampling has been widely used, but well-known non-systematic artifacts due to adsorption and volatilization of semi-volatile compounds exist (Viana et al., 2006;Jacobson et al., 2000). Since a more complex sampling (e.g., using two sampling lines in parallel, one with and the other without a denuder system for volatile OC removal or using two filters in series) is not carried out at regular air pollution monitoring stations, artifacts could not be quantified. However, due to the high filter loadings in winter such sampling artifacts are not expected to have a large contribution (e.g., Viana et al. (2007) found a 5 and 7 % contribution of OC from positive sampling artifacts for winter samples in Amsterdam and Ghent), and we assume that they will not significantly influence the results presented in this study. It should be noted that on some filters PM 10 mass was measured gravimetrically which includes weighting before and after the sampling at a relative humidity (RH) of 50 ± 2 % and a temperature (T ) of 20 ± 2 • C after conditioning for 48 h. Since these handling steps may introduce additional artifacts and none of the samples were pre-heated to remove any OC or EC present on the filters prior to sampling, the analysis of blank filters which were treated exactly the same way as the samples is very important. Therefore, ∼ 50 field blank filters were collected and 34 of them were analyzed for 14 C in OC, 45 for major water-soluble ionic species and 47 for OC and EC mass loading.
Every winter, 5 days with high PM 10 concentrations were investigated and therefore, most of the results presented below are considered as representative for wintersmog episodes, which were the objective of our study. Winter-smog episodes in Switzerland frequently occur on days with inversions, and hence relatively shallow boundary layer heights. The days were selected such that ideally PM 10 concentrations at all stations exceeded the daily limit value of 50 µg m −3 . However, since meteorological conditions in Switzerland north and south of the Alps can differ strongly in winter, it was not possible to find enough days where the selection criterion was fulfilled at all stations simultaneously. Therefore, 5 identical days were chosen separately for stations south and north of the Alps. This ensures similar meteorology and the interpretability of the results in terms of spatial variations within the two regions. In addition, two to three filters per month from August 2008 to July 2009 of the urban background station ZUR were selected to cover a full yearly cycle. In total 320 aerosol filter samples were analyzed for this study. The detailed selection of all analyzed days and the distribution of PM 10 concentrations on those days for every station are shown in Table S1 in the Supplement and Fig. 2, respectively.

EC / OC measurements
The EC and OC concentrations were measured on all samples (n = 320) and blanks (n = 47) using a thermo-optical OC / EC analyzer (Model 4L, Sunset Laboratory Inc., USA), which is equipped with a non-dispersive infrared (NDIR) detector. All samples were combusted following the thermaloptical transmittance method (TOT) using the EUSAAR2 temperature protocol (Cavalli et al., 2010). It should be noted here that the OC / EC determination with TOT instruments is not standardized yet and that measurements with different thermal protocols (e.g., NIOSH NIOSH, 1999;Peterson and Richards, 2002), IMPROVE (Chow et al., 1993), EUSAAR2 (Cavalli et al., 2010) may lead to discrepancies. Typically, total carbon (TC) measured with different protocols shows good agreement (within 10 %), whereas EC can differ significantly from method to method, up to 25 %, and for highly polluted winter samples even up to 60 % , Schmid et al., 2001Piazzalunga et al., 2011a). We chose the EUSAAR2 protocol since this protocol is also used by the NABEL and Cantonal air quality monitoring programs to measure OC / EC concentrations for some stations on a regular basis. Repeated measurements were carried out for 150 samples out of the 320. A blank correction was performed using the average TC filter loading (2.5 ± 0.8 µg C cm −2 ) of all measured blank filters (n = 47) since no systematic differences between the different stations or throughout the years were found (see Fig. S1 in the Supplement). Since EC was not detectable in any of the blank samples, the mean TC blank concentration was also used for the blank correction of OC. The average contribution of the blanks to the total filter loading was 5 ± 2 and 4 ± 2 % for OC and TC, respectively. The mean measurement uncertainty for OC and TC was estimated to be 7.7 and 8.1 %, respectively, using the variability of all samples (n = 8) that were measured three or four times and the variability of the blanks. The uncertainty for EC was assumed to be 25 % to account for possible differences between different TOT protocols (Schmid et al., 2001).

14 C analysis
2.3.1 Separation of carbonaceous particle fractions and 14 C analysis 14 C analysis of EC and OC was carried out on all samples. 14 C content in the blanks was only measured for TC, since an EC loading was not found on those filters (see Sect. 2.2). In the following, we will describe the techniques and procedures of the separation of OC and EC for subsequent 14 C measurements. OC was separated for 14 C analysis using the THEODORE system and the Sunset analyzer (see Szidat et al., 2004 and, respectively for more details). In brief, in THEODORE filter punches with a diameter of 11 to 16 mm were combusted at 340 • C for 10 min in a pure oxygen (O 2 ) stream. The Sunset analyzer, connected to the trapping part of the THEODORE system, was modified such that it could be operated with pure oxygen as a carrier gas in addition to the conventionally used He and He/O 2 . The temperatures and combustion times for the oxidation of OC to CO 2 from filter punches with 0.8-1.5 cm 2 in the Sunset analyzer were set to the same values as those used in the THEODORE pro-tocol. The evolving CO 2 , from the THEODORE and the Sunset analyzer, was separated from interfering reaction gases, cryo-trapped and sealed in glass ampoules for 14 C measurements.
The separation of EC for the 14 C measurement was carried out following the Swiss 4S protocol as described by . First, water-soluble OC (WSOC) and other water-soluble components were removed by water extraction in order to minimize positive artifacts from OC charring (Piazzalunga et al., 2011a;. The remaining water-insoluble OC (WINSOC) was then removed by a thermal treatment in three steps. In the first two steps, OC was oxidized in O 2 at 375 • C for 150 s and then at 475 • C for 180 s. In the third step, OC was then evaporated in an inert atmosphere in helium at 450 • C for 180 s followed by 180 s at 650 • C. In the end (step four), EC was isolated by the combustion of the remaining carbonaceous material at 760 • C for 150 s in O 2 . This method was optimized to reduce biases in 14 C measurements of EC related to OC charring (leading to higher non-fossil EC (EC NF ) values) or losses of the least refractory EC (mostly from wood burning) during the WINSOC removal (in the steps one to three) as those would lead to lower EC NF fractions. Furthermore, using the Sunset analyzer for the combustion made it possible to quantify those artifacts online, since this instrument monitors the filters during the combustion with a laser. As proposed by  we tested the effect of different temperatures in step two and three of the thermal protocol on the EC yields and the OC charring for some samples from stations with contrasting sources and filter loadings (e.g., filters with high and low loading from stations with a large wood burning contribution vs. more traffic influenced stations). Charring of OC most likely occurred only at lower temperature in the steps one and two and was quantified as the difference of the maximum attenuation (ATN) and the initial ATN normalized to the initial ATN of the given thermal step. The EC yield denotes the fraction of EC remaining on the filter samples after the first three OC removal steps before the last step (step four) starts, which was used for the EC recovery for 14 C analysis, and is defined as the ratio between the initial ATN of the laser signal through the filter before step one of the thermal treatment and the ATN before step four. We found that the EC yield and charring did not vary significantly due to different temperatures (550-700 • C) in step three and therefore this temperature was set to 650 • C as suggested by . In contrast, varying the temperature in step two we found 525 and 500 • C as optimal values for SVI and BER, respectively, which exhibited very high filter loadings. Higher temperatures for these two stations were necessary to assure complete removal of OC and possibly charred OC before the EC step (step four). For the samples from the other stations 475 • C, as suggested by , was found to be the optimal setting. On average 74 ± 11 % of the EC was recovered for the 14 C mea-surement for all samples and charred OC only contributed 5.3 ± 4.5 % to EC recovered in step four.
The 14 C measurement of the collected CO 2 from the separated carbonaceous fractions was performed with the MIni radioCArbon DAting System, MICADAS (Synal et al., 2007) at the Swiss Federal Institute of Technology (ETH) Zürich and the Laboratory for the Analysis of Radiocarbon with AMS (LARA), University of Bern , Switzerland, using a gas ion source (Ruff et al., 2007;Wacker et al., 2013), which allows direct CO 2 injection after dilution with He . All 14 C results are expressed as fraction of modern (f M ) representing the ratio of the 14 C / 12 C content of the sample related to the isotopic ratio of the reference year 1950 (Stuiver and Polach, 1977). The f M values were corrected for δ 13 C fractionation  and for the 14 C decay between 1950 and the year of measurement. The uncertainty of the measured f M values for OC and EC (f M,OC and f M,EC , respectively) is on average ∼ 2 % for the samples presented here.

Data correction and presentation
As discussed in the following, several corrections have to be applied to the f M values obtained from the 14 C measurement (see also Table 2 for a summary).

Blank correction: a mass-dependent blank correction is
applied to the measured f M values following an isotopic mass balance approach (Zapf et al., 2013): where f M,corr is the blank corrected f M , and f M,sample and f M,blk are the f M measured for samples and blanks, respectively. mC sample and mC blk denote the carbon mass in the samples and the blanks, respectively. Since blank filters are not available for all stations and years and since the 14 C results of the blanks were not systematically different (between different stations or years, see Fig. S1), the average f M and TC values of the blanks, 0.53 ± 0.12 (n = 34) and 2.5 ± 0.8 µg C cm −2 (n = 47), respectively, were considered for the correction of f M,OC (f M,OC,corr ). The blank correction increases the f M,OC,corr values by ∼ 3 % and the uncertainty (error propagation of Eq. (1)) rises to ∼ 3 %.
No EC was detected on the blank filters (see Sect. 2.2 above); therefore no blank correction was carried out for f M,EC .
2. EC yield correction: the fraction of EC, which was isolated for the 14 C measurement (EC yield) was on average 74 ± 11 % as shown in Sect. 2.3.1. However,  showed that f M,EC changes with different EC recoveries. They found a linear relationship between  f M,EC and the EC yield, which they used to extrapolate f M,EC to 100 % EC yield using the average slope (0.31 ± 0.1) from several samples (n = 5) in order to account for the slight underestimation of biomass burning EC caused by the EC loss during EC isolation for 14 C measurement (see Sect. 2.3.1 above). In this study, we also measured f M,EC from 11 samples at different EC yields. As shown in Fig. S2 there is also a linear relationship between the EC yield and f M,EC for the samples from this study. Even though the slopes exhibit a larger variability compared to the ones presented in  the average slope of all winter samples is very similar. In contrast, the slopes for the summer filters show only a very weak relationship between f M,EC and the EC yield due to the smaller fraction of less refractory EC (mainly from biomass burning) which is removed before the EC isolation for the 14 C analysis. Beside the clear difference between samples from summer and winter, no systematic differences between different stations or years were found. Therefore, average slopes of 0.35 ± 0.11 and 0.07 ± 0.03 for winter and summer samples, respectively, were taken to correct all f M,EC values to 100 % EC yield (f M,EC,total ) using the following equation : The uncertainty of f M,EC,total was obtained by an error propagation of Eq.
(2) using the variability of the average slopes, the measurement uncertainty of f M,EC and an assigned uncertainty of 10 % for the EC yield and is on average 4.2 %.
3. Charring correction: approximately 50 samples exhibited OC charring contributing > 10 % to EC even though the method used here for EC isolation is optimized to minimize OC charring. Therefore, the f M,EC,total values were corrected for charring (f M,EC,final ) using the same isotopic mass balance approach as described in Eq. (1) in which the f M and mC values of the samples and blanks were replaced by f M,EC,total and EC as well as the fraction (f charr , formed in step one and two of the thermal treatment as described in Sect. 2.3.1) and f M of charred OC (f M,charr ). We assumed that only 50 % of the charred OC contributed to the 14 C result of EC since some charred material was most likely removed in step three. However, since some EC could be lost in step three as well, the charred OC evaporated in step three cannot be quantified. Therefore, a high uncertainty of 33 % is assigned to the fraction of charred OC which should in addition account for possible differences and variability between samples and stations. The f M,charr was obtained from 14 C measurements (n = 11) of WINSOC from water-extracted filters released in step one and was found to be on average 0.78. To account for possible sample-to-sample differences and variability between samples and stations, we assigned an uncertainty of 0.10 for f M,charr . The uncertainty of f M,EC,final was on average 4.4 %, which is only slightly higher than for f M,EC,total (4.2 %). whereas it includes additionally f M,bio and the fraction of biogenic sources to the total non-fossil sources (p bio ) for the calculation of f NF,OC . f M,bio was taken from long-term 14 CO 2 measurements at the Schauinsland background station (Levin et al., 2010) and f M,bb was estimated using a tree growth model as described in Mohn et al. (2008). p bio was set to 0.

Analyses of water-soluble major ionic species and levoglucosan
The concentrations of major water-soluble ionic species (cations: K + , Na + , Mg 2+ , Ca 2+ and NH + 4 ; anions: methanesulfonate (MSA), oxalate (Ox 2− ), SO 2− 4 , NO − 3 and Cl − ) were analyzed on all filters (n = 320) and field blanks (n = 45) with an ion chromatographic system (850 Professional, Metrohm, Switzerland) equipped with a Metrosep C4 cation column and a Metrosep A anion column, respectively. Prior to the measurement a water extraction (15 and 50 mL for samples from 2008-2010 and 2011-2012, respectively) with ultrapure water (18.2 M cm −1 ) for 30 min at 40 • C in an ultrasonic bath of filter punches with a diameter of 11 mm was carried out. The measurement uncertainty for most of the water-soluble ions was estimated to be 10 %. An uncertainty of 30 % was assigned for all cations as well as for Ox 2− and Cl − with concentrations < 5 ppb in solution. A blank correction was carried out subtracting an average value of each ionic species from the concentrations in the samples. In contrast to the blank correction of the OC and TC concentrations as well as f NF,OC , where an average value of all blanks (different stations and years) was used, the average of all blanks from the different stations from each winter was taken separately. It should be noted here that not all ionic species were detected in all blanks (see Fig. S1 and Table S2). The overall uncertainty of the major water-soluble ionic species was derived from the error propagation of the measurement uncertainty and the blank variability.
Levoglucosan was measured following the procedures described in Piazzalunga et al. (2010) and(2013a). In brief, levoglucosan was measured by a high-performance anionexchange chromatography (HPAEC) with pulsed amperometric detection (PAD) using an ion chromatograph (Dionex ICS1000) equipped with an isocratic pump and a sample injection valve with a 100 µL sample loop. Prior to the analysis, a water extraction was carried out by three subsequent extractions of ∼ 2 cm 2 filter punches by 20 min sonication using 2 mL Millipore-MilliQ water (18.2 M cm −1 ). Levoglucosan was then separated from other compounds by a Carbopac PA-10 guard column (50 mm × 4 mm) and a Carbopac PA-10 anion exchange analytical column (250 mm × 4 mm) using 18 mM NaOH as an eluent. The analytical system comprised an amperometric detector (Dionex ED50) equipped with an electrochemical cell. The detector cell had a disposable gold electrode and a pH electrode as reference (both from Dionex) and was operated in the pulsed amperometric detection (PAD) mode. The measurement uncertainty was estimated to be ∼ 5 % using the average repeatability of several standards and the limit of detection in solution is 2 ppb. The levoglucosan concentrations were also analyzed for blank filters but were below the detection limit; therefore no blank correction was performed.

Additional data
Since all sampling sites in this project are part of the Swiss national (NABEL) or cantonal air pollution monitoring networks, additional parameters (e.g., gas phase pollutants, particle mass and meteorology) are routinely measured. PM 10 and nitrogen oxides (NO x = NO and NO 2 ) data are available from all stations (except SCH), whereas ozone (O 3 ), sulfur dioxide (SO 2 ) and carbon monoxide (CO) measurements are only performed at some stations. Reference instrumentation according to the valid European standards was used. PM 10 is measured online with beta attenuation monitors (FH62-IR, Thermo ESM Andersen) and by TEOM-FDMS (Thermo Environmental) instruments and an approach presented in Gehrig et al. (2005) to correct/harmonize online and gravimetric PM 10 measurements is routinely applied to data from all stations. It should be noted that NO x measurements using molybdenum converters suffer from interference of oxidation products of NO x which is however not crucial for winter-time conditions (Steinbacher et al., 2007). The meteorological parameters wind-speed, wind-direction, temperature (T ), relative humidity (RH), precipitation and global radiation were also only measured at some of the sites. For the remaining sampling locations meteorological data were taken from nearby stations operated by the Swiss weather service (Me-teoSwiss, 2014). In all networks (NABEL, Cantons and Me-teoSwiss) data sets (except results obtained offline from filter samples, i.e., EC / OC and levoglucosan concentrations as well as 14 C data) undergo an automatic and a manual quality check; data should be (1) within a plausible range, (2) show plausible variability, (3) reproduce to a reasonable extent the expected daily, monthly and yearly variations, and (4) whenever possible measurements are compared to nearby or similar stations with the expectation of similar values (Barmpadimos et al., 2011).

Composition of PM 10
As we were interested in winter-smog episodes, only days with high PM 10 concentrations at all stations were analyzed. As shown in Fig. 2a the selected days from almost all locations exhibited on average values ∼ 50 µg m −3 (European and Swiss daily limit) or above. While not exactly the same days were chosen for stations north and south of the main chain of the Alps, it is nevertheless evident that the PM 10 burden during winter-smog episodes in Switzerland is higher south of the Alps (73 ± 27 µg m −3 in the south compared to 55 ± 16 µg m −3 in the north). These episodes often occur in winter during stable meteorological conditions including periods with high pressure, rather low temperatures and weak winds (typically less than 2 m s −1 ). Such conditions often lead to inversions with low mixing layer heights, thereby favoring the accumulation of pollutants and consequently causing high PM 10 concentrations. The reason for the higher PM 10 values at stations south of the Alps is most likely due to a combination of topography (e.g., several stations are located in alpine valleys), local meteorology (e.g., more persistent inversions with rather low mixing heights compared to the north) and emissions (strong local wood burning influence, see Sects. 3. 2.1 and 3.2.2 below).
As only 5 winter-smog-episode days from each of the five winter seasons were selected and to account for possible differences in the concentration levels between the stations (especially locations north vs. south of the Alps), we will mainly focus here on the fractional contributions of the individual compounds to total PM 10 . The major water-soluble ions, EC and organic matter (OM) measured here explain 82 ± 11 % of the total PM 10 mass. The missing fraction could mostly be attributed to aerosol water content, the water insoluble fraction (e.g., dust particles), and/or to the uncertainties of the different measurement methods and OM : OC ratio used to convert OC to OM. The major contributors to PM 10 during winter-smog episodes in Switzerland were on average the organic matter (OM = OC ×1.8, Turpin and Lim, 2001), with 29 ± 7 and 46 ± 17 % followed by the secondary inorganic aerosol (SIA) constituents nitrate (NO − 3 , 25 ± 9 and 20 ± 11 %), sulfate (SO 2− 4 , 10 ± 4 and 6 ± 3 %) and ammonium (NH + 4 , 9 ± 3 and 7 ± 4 %) for stations north and south of the Alps, respectively (see Fig. 2). Differences observed in the chemical composition of the aerosol between south and north are a first indication that different emission sources may dominate the aerosol burden at these locations. The EC shares of PM 10 were on average 4 ± 2 % in the north and 6 ± 3 % in the south.
For stations north of the Alps, the range of OM contribution is rather stable (station averages 23-32 %), whereas south of the Alps, the OM fraction spans a wider range (station averages 35-52 %), with values statistically significantly higher than in the north (t test significant at 95 %, in general throughout the manuscript we always used a t test with p = 0.05 to test the statistical significance of differences between stations north and south of the Alps). Furthermore, a clear trend towards larger OM contributions at more rural stations is evident in the south. The EC shares of PM 10 are on average slightly lower in the north compared to the south but show similar variations among the different stations (averages range between 3-5 % in the north and 5-7 % in the south). As already shown above the contributions of the different SIA components to PM 10 are larger in the north. In addition, they also show larger station-to-station differences (averages range from 9-30 % for NO − 3 , 5-11 % for NH + 4 and 7-12 % for SO 2− 4 in the north compared to 14-24 % for NO − 3 , 5-8 % for NH + 4 and 5-6 % for SO 2− 4 in the south). While almost all constituents of PM 10 (OM, EC and NO − 3 ) exhibit on average larger concentrations in the south (mainly due to the selection of days with higher PM 10 concentrations  Table 1) to the most rural one. Data from the yearly cycle in ZUR are excluded. Only averages ± standard deviations are displayed for stations from which only filters from one winter were analyzed. The whisker box plots showing the absolute concentrations are presented in Fig. S3. compared to the ones selected in the north), NH + 4 shows on average very similar levels in both regions and SO 2− 4 even higher ones in the north (see Fig. S3). The higher SO 2− 4 fractions and levels observed north of the Alps indicate a higher background of this species possibly caused by occasional long-range transport of SO 2 emissions from Eastern Europe. Another interesting feature is evident for the stations south of the Alps. The relative contributions of NO − 3 and NH + 4 exhibit a trend towards lower values at rural stations, as opposed to the OM fraction (see Fig. 2), which may be due to the influence of the stations in the south by air masses advected from the Po Valley, where emissions from fossil fuel combustion (e.g., NO x ) are elevated (Piazzalunga et al., 2011b;Larsen et al., 2012) compared to the southern part of Switzerland. More details about the influence of air masses originating from other regions outside Switzerland will be discussed in .  Table 2. The use of whisker box plots enables the identification of the variability of the results for each station as well as the station-to-station differences. Several filters from BAS showed clearly elevated f NF,OC values (larger than one and up to five) indicating that BAS is influenced by sources emitting anthropogenic 14 C (e.g., from nuclear power plants, pharmaceutical industry and biochemical laboratories working with labeled 14 C, incinerators for medical waste). BAS is the base for two of the world's largest pharmaceutical enterprises, Roche and Novartis, and in addition an incinerator for medical waste is located in the vicinity of the station. Furthermore, 14 C measurements on leaf samples across the city of Basel also showed partially highly elevated results (BAG, 2008), indicating 14 C-enriched CO 2 . Therefore, f NF,OC values from BAS were not considered for the further analysis. This artifact is however restricted to OC; the f NF,EC results did not show such an influence (see Fig. 3b) and are included and discussed throughout this study. The data from the yearly cycle in ZUR is also excluded here but will be investigated in part II .

Relative fossil and non-fossil contributions of OC and EC
The range of all f NF,OC values (except BAS) as displayed in Fig. 3a is 0.59-0.95 and 0.62-1.02 for stations north and south of the Alps, respectively. A few samples (n = 4) with f NF,OC values slightly above one were found in SVI and are within the uncertainty (∼ 3 %) of f NF,OC . They can be explained on the one hand with very high local woodburning contributions and on the other hand with the un-  Table 1) to the most rural one. Only averages ± standard deviations are displayed for stations from which only filters from one winter were analyzed. f NF,OC values for BAS are not included, since several values above one were found (see Sect. 3.2). Data from the yearly cycle in ZUR are excluded as well. certainties in the reference value f NF,ref used for the correction of the still elevated 14 C concentrations due to the above-ground thermo-nuclear bomb tests (see Sect. 2.3.2). The average f NF,OC values for stations north and south of the alps are 0.78 ± 0.08 (median = 0.78) and 0.82 ± 0.07 (median = 0.83), respectively, showing that on average locations south of the Alps are more impacted by non-fossil sources. As discussed above, non-fossil OC may include, POC and SOC from wood burning and cooking emissions, as well as primary biological particles and biogenic SOC. Cooking was estimated to contribute on average only 7.5 % to organic aerosol (OA) during winter in ZUR which is the largest city of Switzerland (Canonaco et al., 2013), and is therefore expected to contribute less at the other stations. Furthermore, large inputs from biological and biogenic sources are also not expected under Swiss winter conditions, characterized by low biological activity. Therefore, the high f NF,OC values indicate that wood burning POC and SOC are most proba-bly the main source of OC during winter-smog episodes in Switzerland. The highest f NF,OC values north and south of the Alps were found at the rural stations SCH (0.85 ± 0.04) and SVI (0.95 ± 0.05), which are located in narrow Alpine valleys. The lowest non-fossil contributions to OC were observed in BER, STG, VAD and ZUR north of the Alps as well as in MOL and CHI south of the Alps, but were on average never below 70 % showing that sources of fossil carbon only account for a small fraction of OC during winter-smog episodes in Switzerland, even at urban and traffic-influenced stations. Furthermore, the variability of all f NF,OC values for the individual stations and the station to station differences (with the exception of SVI and BER which present the highest and lowest values, respectively) are low as displayed by the small interquartile ranges (IQR = 3rd − 1st quartile; 0.10 ± 0.02 in the north and 0.08 ± 0.02 south of the Alps) and the small range of the station averages (0.75-0.85 and 0.80-0.86 for stations north and south of the Alps, respectively). This suggests that the relative source contributions to OC are very consistent within Switzerland during wintersmog episodes.
Similar high non-fossil contributions to OC were also found in previous studies in Switzerland. The f NF,OC values for ZUR, ROV, MOL, REI and Sedel as well as MAS, Saxon, Sion and Brigerbad ranged on average from 61-76 % with values above 90 % in ROV Sandradewi et al., 2008a, b;Perron et al., 2010). Results previously reported for other regions in Europe show lower biomass burning contributions to OC: e.g., biomass burning OC (OC BB ) to the total OC fraction of 35-54 % at three Austrian cities (Vienna, Graz and Salzburg, Caseiro et al., 2009), 28-65 % at three locations in the Po Valley (Milan, Sondrio and Ispra, Gilardoni et al., 2011;Piazzalunga et al., 2011b) and 60 % in Grenoble .
The non-fossil fraction of EC relates more unambiguously to wood burning. For most stations the wood burning contribution was found to be < 50 % and thus the contribution from fossil fuel combustion, mostly due to traffic, was > 50 % (see Fig. 3b). However, since the average f NF,EC values, except for BER, REI and MOL, never decrease below 0.4, it is evident that wood burning emissions exceptionally account for a large fraction of EC during winter-smog episodes in Switzerland. The individual f NF,EC values range from 0.12-0.79 (on average 0.42 ± 0.13) and 0.25-0.87 (on average 0.49 ± 0.15) for all stations north and south of the Alps, respectively, showing that for EC the contributions from wood burning are higher for locations south of the Alps. The lowest f NF,EC values were found at the stations BER (0.22 ± 0.06), MOL (0.28 ± 0.06) and REI (0.35 ± 0.05), which are directly exposed to traffic emissions from nearby roads with a high traffic flow. Extremely high non-fossil contributions to EC up to 87 and 79 % were observed in SVI (66 ± 11 %) and SCH (69 ± 9 %). Both stations are located in narrow Alpine valleys characterized by frequent winter-time inversions and are strongly influenced by local emissions from wood combus-tion, which is the main source for residential heating in such areas in Switzerland.
The most prominent feature in Fig. 3 is the clear non-fossil increase south of the Alps from MOL to SVI for OC and EC. With the exception of MOL, which is directly located next to a highway, these stations are not only ordered from the most urban and traffic influenced to the most rural, but also geographically from south to north. CHI is located in a more open terrain at the Swiss/Italian border, whereas further north, towards the main Alpine chain, narrower alpine valleys dominate and the region is consequently more rural and wood burning for wintertime residential heating becomes more important. The observation that the non-fossil contributions for both, OC and EC, are on average higher at locations south of the Alps can thus be mainly attributed to the fact that there are more rural stations in the south whereas urban and suburban stations dominate north of the Alps (see Fig. 1).

Total fossil and non-fossil contributions
Next we will discuss the fossil and non-fossil concentrations of OC and EC and their contributions to TC. The fraction of TC in PM 10 is on average 19-25 % for stations north of the Alps and is slightly higher for locations in the south (27-30 %). Fig. 4 shows the average EC F , EC NF , OC NF and OC F concentrations as well as their relative contributions to TC for all analyzed winter samples for each station. It is evident that sources of non-fossil carbon dominate TC at locations north and south of the Alps with contributions around 70 ± 18 and 79 ± 10 % (sum of EC NF and OC NF ), respectively. Compared to other winter measurements across Europe this is rather at the higher end of the reported range and higher than reported for urban sites around the world but similar to values found for suburban and rural locations in the US and India (Hodzic et al., 2010;Heal, 2014).
OC NF is the largest fraction of TC, accounting on average for 61 ± 8 and 69 ± 9 % for stations north and south of the Alps, respectively, whereas EC NF contributes on average ∼ 9 % to TC in both regions of Switzerland. The fossil shares in the north of OC (18 ± 6 %) and EC (13 ± 6 %) are higher compared to those in the south (OC F /TC = 12 ± 6 % and EC F /TC 10 ± 5 %). The lowest and highest fossil contributions to TC (sum of EC F and OC F ) were found in SVI (10 ± 6 %) and BER (43 ± 7 %), respectively. For the stations south of the Alps, a clear decreasing trend in the relative contribution of fossil OC and EC from more traffic to more rurally influenced stations is found (see Fig. 4 and Fig. S4). North of the Alps, such a trend is only evident for EC F . Relative and absolute non-fossil OC and EC contributions in the north (except BER and SCH which present the highest and lowest values) only show low station-to-station differences (station averages range from 58-71 % and 1.5-2.5 µg C m −3 for OC NF as well as 8-11 % and 0.9-1.9 µg C m −3 for EC NF , see Fig. 4 and Fig. S4). In addition, also the variability of the relative and absolute OC NF and EC NF contributions at the individual stations north of the Alps is rather small as evidenced by low IQRs (2.8 ± 0.9 µg C m −3 and 7 ± 2 % for OC NF as well as 0.4 ± 0.1 µg C m −3 and 3 ± 1 % for EC NF ). Together with the low station-to-station differences, this suggests that non-fossil sources very consistently influence stations on the Swiss Plateau. Furthermore, as discussed above, OC NF can be influenced by SOC formation which can be highly variable. However, the low OC NF station-to-station and day-to-day variability points to a similar degree of atmospheric processing and SOC formation for the chosen days in this region of Switzerland. Last, the low absolute and relative EC NF and OC NF IQRs at the individual stations and station-to-station differences also indicate that locations on the Swiss Plateau are rather influenced by regional (still mainly within Switzerland) air pollution. This is confirmed by high correlations (r = 0.7 ± 0.2, 0.5 ± 0.3, 0.9 ± 0.1 and 0.7 ± 0.1) between the concentrations of EC F , EC NF , OC F and OC NF for all measured values from each station located on the Swiss Plateau (see Table 1) against ZUR which was chosen as a reference for this region. Furthermore, this is in agreement with Gehrig and Buchmann, 2003) who previously found that (1) under high pressure conditions inversions can extend over the entire Swiss Plateau and typically last several days possibly causing smog formation and (2) that PM concentrations were strongly influenced by meteorology (dilution with clean air or precipitation) rather than by variation of source activities. In contrast, correlating the fossil and non-fossil concentrations of OC and EC from stations south of the Alps against the ones from MAG shows lower values (r = 0.3 ± 0.2, 0.6 ± 0.3, 0.4 ± 0.3 and 0.3 ± 0.3 for OC NF , EC NF , EC F and OC F , respectively) indicating that local sources are more important for stations south of the Alps. Figure 5 presents the comparison between EC F , OC F and NO x , which are expected to be associated with traffic emissions, in Switzerland. EC F , which is emitted as primary aerosol from vehicles, exhibits a high correlation with NO x for the stations north (r = 0.79) and south (r = 0.75) of the Alps, with similar slopes and axis intercepts for both regions (0.021 and 0.015 µg C m −3 ppb −1 and 0.35 and 0.89 µg C m −3 for north and south of the Alps, respectively (see Fig. 5c), indicating a rather similar fleet composition in the two areas. Similar slopes (0.05, 0.03 and 0.02 µ g C m −3 ppb −1 ) have been reported previously for three locations in Switzerland (MAG, ZUR and PAY, Herich et al., 2011), Grenoble  and London (Liu et al., 2014). In contrast, no correlation is found between OC F and the primary vehicular markers, EC F and NO x (r<0.5, see Fig. 5b) for stations both north and south of the Alps. Further, the amounts of fossil organic carbon measured are significantly higher than amounts expected for traffic emissions; i.e., observed average OC F /EC F = 1.54 ± 0.83 vs. traffic OC / EC = 0.25-0.80 (El Haddad et al., 2013 and references therein). Taken together these observations indicate that a considerable amount of OC F is associated with emissions or atmospheric pathways that yield organic aerosol with little or no EC F and NO x . These processes may in-clude primary emissions from non-mobile fossil fuel combustion sources, e.g., heavy fuel combustion (e.g., crude oil, not widely used in Switzerland), or secondary organic carbon formed from fossil VOCs emitted from traffic.

Non-fossil fraction
As mentioned above a significant fraction of non-fossil carbon during winter-smog episodes originates from wood burning. The use of a single or a set of source specific compound markers from wood burning emissions is often applied to estimate the contribution of this source to ambient aerosol  and references therein). The most widely used tracer compound for biomass-burning emissions is levoglucosan , a product of cellulose combustion. Another wood burning tracer is water-soluble potassium (K + ), which is an inorganic compound mainly present in ash. The wide variability of levoglucosan emission ratios results in significant uncertainties in estimating wood burning contributions. For example, ratios of OC and EC to levoglucosan for alpine regions were reported in Schmidl et al. (2008) to range from 3.7 to 12.5 and from 0.7 to 4.7, respectively, dependent on the combustion conditions and fuel type used (Engling et al., 2006;Lee et al., 2010). Here, we examine the relationship between different measured wood burning markers and the measured OC NF , to investigate the main emission sources and chemical characteristics of this fraction. The comparison of EC NF and OC NF with levoglucosan (see Fig. 6) shows a high correlation for both species with the latter. The small intercept (1.3 and 2.3 µg C m −3 for stations north and south of the Alps, respectively) and the high correlation (r>0.87) between OC NF and levoglucosan suggests that the majority of OC NF originates from wood combustion; i.e., cooking and, biogenic emissions seem to be minor contributors (see Sect. 3.2.1 above). Furthermore, this indicates that OC NF is to a large extent emitted as primary aerosol, however, with the data presented in this study it is not possible to quantify a primary vs. secondary fraction of woodburning OC. OC NF also exhibits a high correlation with K + as well (r = 0.62 in the north and r = 0.87 in the south, see Fig. 8a). However, K + is also found in soil dust and sea salt or can be formed in incinerators and during meat cooking (Schauer et al., , 2001, and therefore cannot be used as unambiguous tracer for wood burning, although none of these sources are expected to have a large influence in Switzerland during winter. Another indication for OC NF originating to a large extent from wood combustion is its high correlation (r = 0.77, see Fig. 7) with EC NF , which can be almost exclusively attributed to this source.
A high correlation is also found between levoglucosan and K + (r>0.6). However, clearly different slopes (0.6 and 5.4) are observed for stations north and south of the Alps, respectively. Furthermore, also the comparison of OC NF and EC NF with levoglucosan as well as OC NF with K + shows signifi-   Table 2). An orthogonal distance regression was used to fit the data. cantly different ratios for stations located in the north and the south. These discrepancies between the two Swiss regions could originate from different wood types used (e.g., soft and hard wood), burning conditions, and atmospheric processing. Different ratios of OC NF and EC NF to levoglucosan indicate differences in SOC formation and/or photochemi-cal degradation of the latter which was recently reported by Kessler et al. (2010) and Hennigan et al. (2011). However, under winter-smog conditions in Switzerland (low temperatures and photochemical activity) rapid levoglucosan degradation is not expected and no large systematic differences in the photochemical activity and SOC formation between locations south and north of the Alps were found as evidenced by very similar OC NF to EC NF ratios (7.7 ± 2.1 and 8.6 ± 2.9, see Table 3 and Fig. 7) for these two regions in Switzerland. However, with our data we cannot completely rule out different wood burning OC / EC emission ratios in both regions of Switzerland as higher primary wood burning OC emissions in the south could be compensated by a larger non-fossil SOC fraction in the north. Higher ratios of OC NF and levoglucosan to K + in the south show that wood burning emissions contain a higher fraction of OC compared to the north. Data from the Swiss forest inventory (Swiss Federal Statistical Office, 2014) show that the fraction of soft (25 %) and hard woods (75 %) in the energy wood production is very similar between the Swiss Plateau and the regions south of the Alps (max. 16 % difference for the years 2008-2012) suggesting that households in both regions have similar access to soft and hard woods. Therefore, the different ratios between OC NF and K + as well as levoglucosan and K + are most likely due to different burning conditions. Previous studies demonstrated that particulate emissions from biomass combustion with high temperatures (e.g., in large combustion units, modern stoves and boilers) consist predominantly of inorganic material (K-salts) and contain little OC (Valmari et al., 1998;Johansson et al., 2003;Khalil and Rasmussen, 2003;Heringa et al., 2011;Schmidl et al., 2011). Consequently, dissimilar levoglucosan to K + ratios measured at different locations have already been used as indication for different burning conditions in recent studies (Sandradewi et al., 2008b;Caseiro et al., 2009;Piazzalunga et al., 2013b). The lower levoglucosan to K + ratios found in this study for locations north of the Alps therefore suggest a larger fraction of more efficient wood burners (e.g., pellet and wood chip burners) in this region compared to the south where wood stoves seem to be operated at rather poor combustion conditions with high carbonaceous and thus lower relative K + emissions.
The discussions above clearly showed the differences in wood burning marker ratios at locations north and south of the Alps. However, a closer inspection of the results of Table 3 reveals that most wood burning marker ratios at the stations PAY and MAS (both north of the Alps) are rather similar to the average over all locations south of the Alps and the urban station CHI exhibits values more similar to the average in the north than to the other southern locations. Since in the north mainly urban and suburban stations and south of the Alps mostly rural and/or background sites were chosen (see Table 1 and Fig. 1), this suggests that the differences in the wood burning marker ratios between these two Swiss regions are most likely associated with the different station characteristics (e.g., rural and/or background with high wood burning influence vs. urban, suburban and more traffic influenced stations) rather than due to their geographical location within Switzerland.

Comparison of wood burning marker ratios with
other studies Herich et al. (2014) presented an overview about previous studies carried out during winter in Switzerland and other alpine regions in Europe. Several source apportionment methods (including 14 C analysis, aethalometer model, positive matrix factorization, chemical mass balance, macrotracer approach -see Gianini et al. (2013) and  for a discussion about possible differences in the biomass burning marker ratios due to different approaches) were used in these studies to estimate the wood burning frac- Figure 8. OC NF (a) and levoglucosan (b) as a function of the K + concentrations combined with whisker box plots of their ratios for all measured winter samples (red circles denote the mean). OC NF values from BAS and all data from the yearly cycle in ZUR are excluded (see Sect. 3.2.1 and Sect. 1). Levoglucosan data is only available for the first two winter seasons (see Table 2). An orthogonal distance regression was used to fit the data.
and OC NF to EC NF ratios presented here (see Table 3). The differences in the ratios most likely originate from (1) uncertainties in the OC BB determination (e.g., OC / levoglucosan emission ratios have to be assumed which can be highly variable) (2) SOC from wood burning is not taken into account in the OC BB values as presented in Herich et al. (2014) but is included in OC NF as obtained by the 14 C measurement and (3) a contribution of other non-fossil sources (e.g., cooking or biogenic aerosol) to OC NF as apportioned with the 14 C analysis cannot be completely ruled out although they are expected to have no large influence during winter-smog episodes in Switzerland (see discussion in Sect. 3.2.1 above). The differences in the wood burning marker ratios between locations north and south of the Alps is also evident for the results presented in Herich et al. (2014). OC BB and EC NF to levoglucosan ratios are higher in the north which was also shown for the same ratios obtained here. In addition, OC BB / EC NF previously found for stations north and south of the Alps in Switzerland are very similar confirming the findings from above (see Sect. 3.3.2) that there is no significant difference in the non-fossil SOC formation between these two regions.

Conclusions
In this study we present source apportionment results of winter-smog episodes in Switzerland (days exceeding the Swiss and European daily PM 10 limit of 50 µg m −3 ) using radiocarbon ( 14 C) analysis separated for the elemental (EC) and organic (OC) carbon fraction together with levoglucosan, major water-soluble ionic species and gas phase pollutant measurements. Overall, ∼ 300 filter samples from five winter seasons (2008-2012) from 16 air pollution monitoring stations across Switzerland with different characteristic (e.g., urban, suburban, rural, alpine valley, traffic, background, etc.) were analyzed providing one of the world's largest aerosol 14 C data sets.
The most important contributions to PM 10 during wintersmog episodes in Switzerland were on average the organic matter OM (29 ± 7 and 46 ± 17 %), followed by the secondary inorganic constituents nitrate (NO − 3 , 25 ± 9 and 20 ± 11 %), sulfate (SO 2− 4 , 10 ± 4 and 6 ± 3 %) and ammonium (NH + 4 , 9 ± 3 and 7 ± 4 %) for stations north and south of the Alps, respectively. The EC shares of PM 10 were on average 3-5 % north of the Alps and 5-7 % south of the Alps. PM 10 and OM concentrations during winter-smog episodes in Switzerland were significantly higher for stations south of the Alps, which is most likely due to a combination of topography (e.g., several stations are located in alpine valleys), local meteorology (e.g., more persistent inversions with rather low mixing heights compared to the north) and emissions (strong local wood burning influence).
The fractional non-fossil contribution of organic carbon (f NF,OC ) determined with the 14 C analysis ranges on average between 0.69-0.85 and 0.80-0.95 for stations north and south of the Alps, respectively, showing that traffic contributes on average only up to 30 % to OC. Furthermore, the elevated f NF,OC values together with high correlations with other wood burning markers (non-fossil EC, levoglucosan and water soluble potassium) indicate that residential wood burning is the major source of OC during winter smog episodes in Switzerland. The station-to-station differences and the variability at each individual location north of the Alps is small suggesting that on the one hand the relative source contributions, meteorological conditions, as well as the degree of atmospheric processing and secondary OC formation for the chosen days were very similar and on the other hand that different stations, especially those on the Swiss Plateau, are rather influenced by regional air pollution than from local sources. The relative non-fossil contributions of EC (f NF,EC ), which can be exclusively attributed to wood burning, are on average 0.42 ± 0.13 and 0.49 ± 0.15 for stations north and south of the Alps, respectively. Since f NF,EC values are often close to 0.5 (even slightly higher for some stations) this shows that also residential wood combustion contributes to a large extent to EC during winter-smog episodes in Switzerland. The sum of non-fossil OC and EC contributes on average 70 ± 18 and 79 ± 10 % to total carbon at stations north and south of the Alps, respectively, highlighting the importance of wood burning emissions from residential heating in Switzerland during winter-smog episodes. This is in agreement with recent studies which have shown that residential wood burning can be the dominating source of carbonaceous aerosols during the cold season in Europe.
The comparison between fossil EC (EC F , only emitted as primary aerosol) and nitrogen oxides (NO x ), which are mainly associated with traffic emissions, showed a good agreement whereas no correlation was observed between fossil OC (OC F ) and the two latter components, indicating that a considerable amount of OC F is secondary OC (SOC) formed from fossil precursors mainly emitted from traffic. Correlations between non-fossil OC (OC NF ) and EC (EC NF ) and the wood burning markers levoglucosan and water soluble potassium (K + ) clearly show different slopes for stations north and south of the Alps suggesting different burning technologies in both regions.
The Supplement related to this article is available online at doi:10.5194/acp-14-13551-2014-supplement.