AIRUSE-LIFE+: a harmonized PM speciation and source apportionment in five southern European cities

Abstract. The AIRUSE-LIFE+ project aims at characterizing similarities and heterogeneities in particulate matter (PM) sources and contributions in urban areas from southern Europe. Once the main PMx sources are identified, AIRUSE aims at developing and testing the efficiency of specific and non-specific measures to improve urban air quality. This article reports the results of the source apportionment of PM10 and PM2.5 conducted at three urban background sites (Barcelona, Florence and Milan, BCN-UB, FI-UB and MLN-UB), one suburban background site (Athens, ATH-SUB) and one traffic site (Porto, POR-TR). After collecting 1047 PM10 and 1116 PM2.5 24 h samples during 12 months (from January 2013 on) simultaneously at the five cities, these were analysed for the contents of OC, EC, anions, cations, major and trace elements and levoglucosan. The USEPA PMF5 receptor model was applied to these data sets in a harmonized way for each city. The sum of vehicle exhaust (VEX) and non-exhaust (NEX) contributes between 3.9 and 10.8 µg m−3 (16–32 %) to PM10 and 2.3 and 9.4 µg m−3 (15–36 %) to PM2.5, although a fraction of secondary nitrate is also traffic-related but could not be estimated. Important contributions arise from secondary particles (nitrate, sulfate and organics) in PM2.5 (37–82 %) but also in PM10 (40–71 %), mostly at background sites, revealing the importance of abating gaseous precursors in designing air quality plans. Biomass burning (BB) contributions vary widely, from 14–24 % of PM10 in POR-TR, MLN-UB and FI-UB, 7 % in ATH-SUB, to – Local dust, 7–12 % of PM10 at SUB and UB sites and 19 % at the TR site, revealing a contribution from road dust resuspension. In PM2.5 percentages decrease to 2–7 % at SUB-UB sites and 15 % at the TR site. – Industry, mainly metallurgy, contributing 4–11 % of PM10 (5–12 % in PM2.5), but only at BCN-UB, POR-TR and MLN-UB. No clear impact of industrial emissions was found in FI-UB and ATH-SUB. – Natural contributions from sea salt (13 % of PM10 in POR-TR, but only 2–7 % in the other cities) and Saharan dust (14 % in ATH-SUB, but less than 4 % in the other cities). During high pollution days, the largest sources (i.e. excluding secondary aerosol factors) of PM10 and PM2.5 are VEX + NEX in BCN-UB (27–22 %) and POR-TR (31–33 %), BB in FI-UB (30–33 %) and MLN-UB (35–26 %) and Saharan dust in ATH-SUB (52–45 %). During those days, there are also quite important industrial contributions in BCN-UB (17–18 %) and local dust in POR-TR (28–20 %).


Introduction
Atmospheric particulate matter (PM) concentrations can vary widely across Europe due to different climatic conditions and local features such as anthropogenic source types, emission rates and dispersion patterns. Urban PM 10 concentrations show significant variability across Europe as reported by routine monitoring networks (EEA, 2015) and research studies (Querol et al., 2004;Putaud et al., 2010;Lianou et al., 2011;Kukkonen et al., 2005;Karanasiou et al., 2014). For PM 2.5 the spatial variability across southern Europe is less known because in most air quality zones it is not as widely measured as PM 10 . As a consequence, there is limited information on the geographical variability of the coarse fraction (PM 10 -PM 2.5 ), which is often linked to local sources and whose evidence of health concern is increasing (Brunekreef and Forsberg, 2005).
Comparability of data is also hampered by the fact that most research studies analysed PM data from different periods or with different sampling calendars. An example of this is given by the multi-city studies aimed at investigating the short-and long-term health effects of exposure to PM x mass concentrations, NO x and SO 2 (Eeftens et al., 2012;Medina et al., 2004;Meng et al., 2013;Romieu et al. 2012).
Moreover, the comparison of bulk PM concentrations only, without the necessary chemical characterization of collected samples and source apportionment analysis, does not allow for an in-depth investigation of sources, limiting the scope for air quality management purposes.
Recent studies have reported the PM 10 and PM 2.5 concentrations across Europe, but no information on PM composition and sources was provided (Karanasiou et al., 2014;Lianou et al., 2011). In another study, Querol et al. (2004) compared the PM 10 and PM 2.5 levels and chemistry of seven selected EU regions, but data sets were not simultaneous from 1998 to 2002. The ESCAPE project (Eeftens et al., 2012) investigated the health effects of long-term exposure to ambient air pollution across Europe. PM 2.5 , PM 10 and particle composition were compared at 20 sites across 2008-2011, but measurements were done three times for 14 days in different seasons without covering the full year period.
The AIRUSE LIFE+ project generated the first harmonized data set of southern European cities for PM 10 and PM 2.5 levels and composition, following the same sampling protocol and 12-month calendar in Barcelona (Spain), Porto (Portugal), Florence and Milan (Italy) and Athens (Greece). The goal is to characterize the similarities and heterogeneities in PM sources and contributions across the Mediterranean region. This article describes the source apportionment results for PM 10 and PM 2.5 , highlighting common features and dissimilarities across southern Europe. Once the main sources of PM 10 and PM 2.5 are identified, the strategic goal of the AIRUSE project is to test and develop specific measures to improve air quality in southern Europe, targeted to meet air quality standards and to approach as closely as possible the WHO guidelines.

PM sampling
PM sampling was carried out from January 2013 during 12 months simultaneously at five sites in Barcelona, Porto, Florence, Milan and Athens. PM 10 and PM 2.5 24 h samples were collected simultaneously every third day. Furthermore, in order to evaluate the chemical fingerprint of Saharan dust, additional PM 10 and PM 2.5 sampling was performed at each city under selected Saharan dust intrusions after forecasting the occurrence of this phenomenon. The forecast was based on the interpretation of (i) air mass back trajectories calculated with the HYSPLIT4 model from NOAA (Draxler and Rolph, 2003) and (ii) predictions of dust concentrations by the SKIRON model, University of Athens (http://forecast.uoa.gr.html) and Barcelona Supercomputing Center (NMMB/BSC-Dust forecasts) prediction models.
The details of each monitoring site and instrumentation used are described below. -POR-TR: Porto traffic (Portugal). The urban traffic (TR) station is located in Praça Francisco Sá Carneiro (41 • 09 46.10 N, 8 • 35 26.95 W) and is part of the national air quality network, QualAr. It is located on the eastern side of Porto, next to Fernão de Magalhães Avenue and 600 m from the inner circular motorway. Two low-volume samplers (TECORA) operating at 2.3 m 3 h −1 collected PM 10 and PM 2.5 onto 47 mm diameter Teflon filters. A high-volume sampler operating at a flow of 1113 L min −1 with impaction plates from Sierra-Andersen, which enabled the simultaneous collection of PM 2.5 and PM 2.5-10 onto quartz fibre filters, was used in parallel.
-FI-UB: Florence urban background (Italy). The urban site of Bassi is an air quality UB monitoring station (43 • 47 8.33 N, 11 • 17 13.19 E) of the Environmental Protection Agency of Tuscany. PM 10 and PM 2.5 samples were collected by means of two low-volume (2.3 m 3 h −1 ) CEN equivalent sequential samplers (HY-DRA Dual Sampler); each sampler is equipped with two inlets so that aerosol can be simultaneously collected on Teflon membrane and quartz microfibre filters (47 mm diameter, Pall R2PJ047 and Aquaria QF1, respectively).
-ATH-SUB: Athens suburban background (Greece). This station is part of the Global Atmosphere Watch network (GAW-DEM) and is located on the NCSR Demokritos campus (37 • 99 50 N, 23 • 81 60 E), at the north-eastern corner of the Greater Athens Metropolitan Area and at an altitude of 270 m a.s.l. The suburban site is away from direct emission sources in a vegetated area (pine). PM 10 and PM 2.5 samples were collected on Teflon filters by means of low-volume (2.3 m 3 h −1 ) samplers (Sequential 47/50-CD with Peltier cooler, Sven Leckel GmbH, a Tecora Echo PM sampler and a Demokritos EN12341 sampler). PM 10 and PM 2.5 samples were also collected on quartz microfibre filters by means of high-volume samplers (Sequential High-Volume Sampler CAV-A/MSb, MCV, SA).

Sample treatment and analysis
Before sampling, quartz or PTFE microfibre filters were dried for 5 h and conditioned for 48 h at 20 • C and 50 % of relative humidity. Weights of blank filters were measured three times every 24 h by means of a microbalance (1 or 10 µg sensitivity). After weighing, Ø 47 mm filters were kept in Petri holders, while Ø 15 cm and large rectangular filters were kept in aluminium foils. After sampling, filters were brought back to the laboratory to be weighed two more times every 24 h of conditioning at the same temperature and relative humidity as the first weighing.
Once the weights of samples were determined, filters were destined for several analytical determinations. These procedures are briefly listed below, according to the different species analysed.
-Major and trace elements were determined -in Teflon filters by different techniques: PIXE (particle-induced X-ray emission) without any pretreatment ; after acid digestion (5 mL HF, 2.5 mL HNO 3 , 2.5 mL HClO 4 ) of 1/2 of each filter, consecutively by ICP-MS (inductively coupled plasma mass spectrometry) and ICP-AES (inductively coupled plasma atomic emission spectroscopy) (Querol et al., 2001) to assure comparability between the two techniques (only for Porto samples, Fig. S1 in the Supplement); by ICP-AES after digestion with HNO 3 -H 2 O 2 in a microwave oven according to EU method EN14902:2005 , only for Florence samples (Fig. S1); by XRF (X-ray fluorescence) with a polarized primary X-ray beam and several secondary targets, after an intercomparison with PIXE ( Fig. S2), only for Milan samples (Table S1 in the Supplement); -in quartz filters by ICP-MS and ICP-AES after acid digestion (5 mL HF, 2.5 mL HNO 3 , 2.5 mL HClO 4 ) of 1/4 of each filter (Querol et al., 2001) (Table S1); and -in Teflon and MCE filters by XRF (X-ray fluorescence), only for Milan samples.
-Water-soluble ions by IC (ion chromatography), after extraction in 20 mL of MilliQ water (with an ultrasonic bath for 30 min) of 1/2 or 1/4 of filter -On quartz filters, organic carbon (OC) and elemental carbon (EC) by thermal-optical analysis with the EU-SAAR2 temperature program by means of Sunset analysers. Filters from POR-TR were analysed in a thermaloptical transmission system described in detail elsewhere, following a similar protocol of EUSAAR2 (Pio et al., 1998(Pio et al., , 2011.

3292
F. Amato et al.: AIRUSE-LIFE+: a harmonized PM speciation -On the PM 10 quartz microfibre filters, carbonate carbon (CC), by means of the procedure described by Pio et al. (1994): briefly, the carbonate measurement set-up comprises four components: a mass flow meter, a reaction cell, an infrared CO 2 analyser, and a computer terminal for data acquisition. A portion of each filter (9 mm punches) is acidified with orthophosphoric acid (20 %) in a CO 2 free gas stream to convert the carbonate carbon to CO 2 , which is then detected by an infrared analyser.
-On the PM 2.5 quartz filters, levoglucosan was determined by means of IC after extraction in 10 mL of MilliQ water (with an ultrasonic bath for 30 min) on a 1.5 cm 2 punch of filter (Table S1).
The ICP-MS, PIXE, GC-MS, CC and XRF analyses were carried out by one laboratory for each technique, while IC and ECOC analyses were separated among the five laboratories.

Source apportionment
Source apportionment studies of atmospheric particulate matter are often performed by means of receptor models that are based on the mass conservation principle: where x ij is the concentration of the species j in the ith sample, g ik is the contribution of the kth source in the ith sample, f j k is the concentration of the species j in source k, and e ij is the error of each individual datum. Equation (1) can also be expressed in matrix form as X = G F T . If f j k is known for all the sources, then the chemical mass balance (CMB) can be applied (Watson et al., 1984), as for this model the experimental profiles of all major sources are needed. When both g ik and f j k are unknown, factor analysis (FA) techniques such as principal component analysis (PCA) (Thurston and Spengler, 1985;Henry and Hidy, 1979) and positive matrix factorization (PMF) (Paatero and Tapper, 1994) are used for solving Eq. (1). PMF can be solved with the Multilinear Engine (ME-2) developed by Paatero (1999) and implemented in version 5 of the US EPA PMF (http://www.epa.gov/heasd/research/pmf.html). In this study, the US EPA PMF v5 was applied to the five data sets obtained at BCN-UB, FI-UB, ATH-SUB, MLN-UB and POR-TR. Since PMF is a weighted least-squares method, individual estimates of the uncertainty in each data value are needed. The uncertainty estimates were based on the approaches by Polissar et al. (1998), Amato et al. (2009) and Escrig Vidal et al. (2009). Species which retain a significant signal were separated from the ones dominated by noise, following the signal-to-noise (S / N ) criterion defined by Paatero and Hopke (2003). Species with S / N < 0.2 are generally defined as bad variables and removed from the analysis, and species with 0.2 < S / N < 2 are generally defined as weak variables and down-weighted (increasing uncertainty by a factor of 3). Nevertheless, since S / N is very sensitive to sporadic values much higher than the level of noise, the percentage of data above the detection limit was used as a complementary criterion.
In those cases where the PMF model was not able to resolve a Saharan dust source, the Saharan dust contribution was estimated according to the methodology proposed by SEC (2011).

Results and discussion
3.1 PM 10 and PM 2.5 levels As shown in Table 1, the observed mean PM levels during the study period were as follows. . For the UB sites mean PM 10 levels are similar, with the exception of MLN with relatively higher levels, due to intense local and regional PM source contributions and to the specific meteorology of the Po Valley, with frequent and intensive atmospheric thermal inversions that induce regional accumulation of pollutants. For PM 2.5 , similar and relatively lower values (as compared with PM 10 ) were recorded at UB and SUB sites (11-15 µg m −3 ), with the exception of MLN (30 µg m −3 ). In the case of the TR site, the mean level reached 27 µg m −3 . UB and SUB PM 2.5 / PM 10 ratios were close to 0.7 in the case of FI and BCN, 0.5 for ATH and 0.8 for MLN. At the POR-TR site it increased to 0.8. Although at the TR site levels of resuspension of road dust (mainly in the coarse size) are expected to be higher than at UB sites, the proximity to the exhaust emissions (dominated by fine PM) results in a higher PM 2.5 / PM 10 ratio.
When considering the stricter WHO guidelines, all cities exceeded both the PM 10 and PM 2.5 annual mean thresholds, with the exception of PM 10 in FI-UB. It is worth noting that the sampling year was particularly rainy at most of the study sites, including Florence. The daily PM 10 and PM 2.5 WHO threshold (50 and 25 µg m −3 , respectively) was exceeded only in MLN-UB and POR-TR. As shown by Fig. S3, seasonal trends for PM 10 at the five AIRUSE cities were very different. ATH-SUB was characterized by higher spring-summer and lower autumnwinter PM 10 levels, probably due to the higher African dust influence in the warm seasons. A similar but much smoother trend was observed for BCN-UB. For POR-TR, the highest levels were recorded in summer and the lowest in spring. At MLN-UB, lower levels were recorded in spring and summer, when PM 10 levels reached similar levels to BCN-UB, FI-UB or ATH-SUB; however, in autumn and winter, levels at MLN-UB were higher by a factor of more than 2 with respect to the other seasons and most of the other cities. This is due to the aforementioned atmospheric stagnation features of the Po Valley that favoured the accumulation of intensive winter PM emissions. At FI-UB a similar trend was observed, but with a less pronounced winter increase than that described for MLN-UB. In fact, Florence is also located in a closed basin (the Arno River valley), which is characterized by stagnant conditions during the cold season. PM 2.5 levels followed similar seasonal patterns to those described above for PM 10 at each city.

EC, OC and OM
EC is a tracer of emissions from road traffic, mainly from diesel vehicles (Querol et al., 2013); it is a carrier of highly health-relevant organic species (WHO, 2012). Mean EC concentrations reached 1.1 µg m −3 in PM 10 at the UB sites of BCN and FI and 1.9 µg m −3 at the UB site of MLN ( Fig. 1). At the SUB site of ATH, EC levels were lower (0.4 µg m −3 in PM 10 , Fig. 1) due to the higher distance from avenues with high traffic load, but also due to the lower proportion of diesel vehicles in the fleet of ATH compared to the other three cities. Until 2012, the use of diesel for private cars was not allowed in this city.
As expected, levels of EC at traffic sites were higher than at UB sites by a factor ranging from 3.6 to 4.9. EC levels recorded for PM 10 are very close to those of PM 2.5 (PM 2.5 / PM 10 ratios within 0.9 to 1.0), pointing to the very fine size of the EC-bearing particles.
OM was calculated by multiplying OC by a factor accounting for non-C atoms which may change according to location and existing sources (Turpin and Huntzicker, 1995). This factor ranged from 1.2 (at the POR-TR site due its proximity to traffic emissions, Pio et al., 2011) to 1.7-1.8 (rest of the sites, Minguillón et al., 2011, andMohr et al., 2012). Mean levels of OM ( Fig. 1) reached 5.0 and 5.5 µg m −3 in PM 10 at the BCN-UB and ATH-SUB sites, respectively, but increased up to 9.6 and 15.4 µg m −3 at FI-UB and MLN-UB. At POR-TR, the mean level of OM was 7.9 µg m −3 . OM also has a fine grain size since PM 2.5 / PM 10 ratios for this component ranged from 0.7 to 1.0. OM / EC ratios for PM 10 ranged from 1.6 at POR-TR, reaching values of 4.4 at the UB site of BCN, 8.2 and 8.7 at the UB sites of MLN and FI, and 13 at the SUB site of ATH. This ratio results from the influence of both the proximity to the emission sources and the biomass burning. Thus, primary diesel soot is characterized by an OM / EC ratio close to 1.0, whereas a higher ratio indicates a major relative abundance of secondary OM, due to a longer distance from emission hotspots (from TR to UB to SUB) and/or a higher contribution from biomass burning to PMx levels, since OM / EC from biomass burning is much higher (Puxbaum et al., 2007). Also, biogenic emissions can increase the OM / EC ratio.
Levels of OM + EC accounted for around 27-30 % of the PM 10 load at the UB and SUB sites, with the exception of FI and MLN (45-58 %). At POR-TR this contribution increased to 37 %. In PM 2.5 the OM + EC load increased as compared with PM 10 and ranged from 37 to 69 % of the PM 2.5 mass. This is due mainly to the decrease in mineral dust and sea salt in PM 2.5 when compared with PM 10 .
The higher levels of OM recorded at MLN, FI and, partially, at POR, are probably the result of a high contribution to PM 10 levels from biomass burning (BB). This is also confirmed by the high OM / EC ratios. In the case of MLN the frequent atmospheric stagnation episodes also favour the formation of secondary OM as also described later for nitrate. To a lesser extent this also applies for FI, where stagnant conditions are common during winter.
As expected, due to the coarse mass size distribution of sea salt, levels were reduced by 71-86 % in PM 2.5 with respect to PM 10 , with the exception of MLN, where levels of sea salt were only reduced by around 38 % in PM 2.5 .

Mineral dust
Mineral dust (or soil dust) was calculated based on average crust composition (Nava et al., 2012) as 1.15 · (3.79 · Al + 2.14 · Si + 1.67 · Ti) − . In the case of the quartz filter, a value of 3 for ratio SiO 2 / Al 2 O 3 was used. Mineral dust at UB sites reached around 2.3 µg m −3 as the annual mean in FI-UB and 3.8-4.2 µg m −3 at the UB and SUB sites of BCN, MLN and ATH (Fig. 1). These levels account for 18 and 25 % of the PM 10 mass at BCN and ATH and 12 and 10 % at FI and MLN, respectively. At the POR-TR site, mineral dust accounts for 12 % of the PM 10 .
As also expected from the coarse mode of occurrence of mineral dust, levels of this PM component were much lower in PM 2.5 , down to 20-49 % of the PM 10 levels. The contri-bution of mineral dust reached from 4 to 10 % of PM 2.5 for all sites.

Sulfate and nitrate
There is a marked geographical variability, with increased levels of sulfate from FI-BCN to POR-MLN to ATH (1.7-2.0 to 2.4-3.0 to 3.7 µg m −3 in PM 10 and 1.5-1.7 to 1.9-2.0 to 2.9 µg m −3 in PM 2.5 , Fig. 1). This is probably due to the influence of the use of coal and petroleum coke/fuel oil for power generation in and/or around the high sulfate regions of this study, but it may also be caused by the influence of SO 2 emissions from petrochemical plants. In the case of POR, when the site is under the influence of north-westerly winds, one of the major sources of sulfate could be the Porto Refinery (10 km away from the sampling site), which began operating in 1970. It is a crude oil industrial processing plant that has an annual installed capacity of 4.5 million ton and produces a wide range of products including fuels, lubricants, aromatics (BTX) for the petrochemical industry, industrial solvents and petroleum waxes. Emissions from shipping in the harbour may represent another possible source. As expected from the fine mode of occurrence of ammonium sulfate ((NH 4 ) 2 SO 4 or NH 4 HSO 4 ), 76 to 85 % of the sulfate in PM 10 is present in PM 2.5 .
Levels of nitrate in PM 10 show a less marked spatial variability, with the exception of MLN, with mean annual levels reaching 1.2 (FI-ATH) to 2.0 (POR-BCN) µg m −3 and 6.0 µg m −3 in MLN. This marked difference between MLN and the rest of the other AIRUSE regions is mainly due to the specific meteorological and emission patterns of the Po Valley. In this case, it coincides with a large urban and industrial agglomeration (with the associated road traffic), with the consequent elevated atmospheric emissions and a peculiar meteorology favouring frequent and marked thermal inversions that cause the accumulation of pollutants and the formation of high levels of ammonium nitrate (NH 4 NO 3 ) from the high anthropogenic NO x and NH 3 emissions and 3296 F. Amato et al.: AIRUSE-LIFE+: a harmonized PM speciation the coldest temperature in winter among AIRUSE cities. In particular, high NH 3 levels, emitted from agricultural and animal husbandry activities, can be transported from the southern part of the Po Valley to the urbanized northern part, inducing ammonium nitrate formation. On the other hand, the relatively coarse mode of nitrate is noticeable in all cases, since levels in PM 2.5 are generally 50-60 % lower than in PM 10 , with the exception of MLN by 40 % and the ATH-SUB area by 85 %. These large fractions of coarse nitrate are probably due to the high temperatures and dry conditions reached in summer in the study regions. Especially at higher temperatures, NH 4 NO 3 dissociate into gaseous HNO 3 and NH 3 , and in turn a fraction of this HNO 3 may react with NaCl or CaCO 3 to give coarse NaNO 3 or Ca(NO 3 ) 2 . This accounts for the large differences observed for nitrate in the coarse and fine size fractions at MLN and ATH, with 60 % of nitrate residing in the fine aerosol at the first site and only 15 % at the latter. The hypothesis of a temperature-driven behaviour of nitrate is reinforced by the seasonal trends of the ratios between fine (PM 2.5 ) nitrate and PM 10 nitrate: most of the nitrate, i.e. 60-80 % of it, is in PM 2.5 during autumn and winter, while the percentage drastically decreases during spring-summer, down to roughly 15 %. Seasonal trends are less and less pronounced when passing from cities with a continental climate such as FI (and MLN) to coastal ones with a mild climate such as POR and BCN, and are not visible in ATH, with a warmer climate.
The contribution of sulfate, nitrate and ammonium (secondary inorganic aerosols, SIA) to the PM 10 load reaches 15-21 % of PM 10 in BCN-UB, POR-TR and FI-UB, but 30-31 % at MLN-UB and ATH-SUB. In spite of the prevailing fine mode of these PM components, the contribution of SIA to the PM 2.5 load remains similar to that of PM 10 , with 13-23 and 31-37 % for the above groups of sites. The latter two cities (MLN and ATH) are both characterized by similarly high SIA loads, but due to different causes, whereas at MLN SIA is dominated by ammonium nitrate; in ATH ammonium sulfate prevails.
The ion balances between cations and anions of the daily PM 10 and PM 2.5 samples collected at the five AIRUSE cities are reported in Fig. S4. The results showed that there was a neutralization of the acidic components of the PM 10 and PM 2.5 in all cases, with the exception of FI in the PM 2.5 fraction. Thus, the balances range from anions = 0.82 · cations (BCN, R 2 = 0.72) to anions = 1.0 · cations (POR, R 2 = 0.96). A slope lower than 1.0 means that there is a cation load in excess, probably due to the occurrence of Ca-and Mg-bearing carbonate minerals (Alastuey et al., 2005). This is expected to occur specially during African dust episodes (see low anion/cation samples present in the balance of BCN in Fig. S4). In the case of FI PM 2.5 samples, the regression fits with anions = 1.12 · cations (R 2 = 0.95), which points to a slightly acidic character of the PM samples with a fraction of 12 % of the anionic species being present as acids. However, con-clusions on aerosol pH cannot be conclusive given that ionic balance and molar ratio methods have been recently criticized (Hennigan et al., 2015).

Secondary vs. primary PM
The amount of secondary aerosol was roughly estimated at each site as where a expresses the primary OC / EC ratio (averaged among existing sources, and varying between 0.7 and 2.2; Pio et al., 2011) and b accounts for the non-C atoms in secondary aerosol mass (varying between 1.6 and 2.1; Aiken et al., 2008). Figure 2 shows that for PM 10 , at the UB and SUB sites the secondary fraction dominated the PM 10 mass by contributing from 53 % (BCN) to 71 % (MLN), whereas at POR-TR the primary contribution prevailed (60-63 %). For the calculation of the secondary and primary fractions, all values are normalized by the sum of reconstructed PM mass. It is very important to be taken into account this high load of secondary PM mass in designing air quality plans, since these components are formed in the atmosphere from organic and inorganic gaseous precursors. The organic and inorganic secondary contributions to PM 10 are very well balanced (close to 50-50 % in most cases), but not in the case of FI-UB, where 65 % of the secondary PM 10 load is from organic aerosols (Fig. 2), probably indicating a relative contribution from biomass burning compared with the other sources. The contributions of secondary aerosols to PM 2.5 increase when compared with PM 10 due to the lower contributions of dust and sea salt to the fine fraction. This secondary contribution then ranged from 66 % (BCN) to 82 % (ATH) at the UB and SUB sites and 37 % at POR-TR (Fig. 2).

Biomass burning tracers
Levels of levoglucosan and K are considered tracers for biomass burning (Fine et al., 2001;Gonçalves et al., 2010;Nava et al., 2014). In fact, levoglucosan is a sugar emitted into the atmosphere exclusively by breaking cellulose chains during biomass burning. Levoglucosan levels varied by 1 order of magnitude: 20 (BCN), 37 (ATH), 255 (FI), 274 (MLN) and 387 (POR) ng m −3 in PM 2.5 (Table 1), reflecting a very clear difference in the impact of biomass burning on air quality across southern European cities. The values obtained should be considered minimum levels because it is well known that levoglucosan may be degraded in highly oxidizing environments (Simoneit et al., 1999). This impact is mostly due to differences in using biomass burning for domestic purposes, but in POR and MLN, peak events were also detected in summer as a consequence of the impact of the emissions of forest fires and/or agricultural fires. passing from MLN to FI to POR to ATH and to BCN is clearly demonstrated by the time series of daily levels of levoglucosan (Table 1). The lower values for Athens can also be attributed to the suburban character of the site. On the other hand, levels of K in the fine fraction (PM 2.5 ) may also increase as a consequence of the influence of inorganic biomass burning ash contributions to the PM load. In this case, the differences are also evident but less marked than for levoglucosan: 87 (BCN), 115 (ATH), 172 (FI), 345 (POR) and 303 (MLN) ng m −3 in PM 2.5 (Table 1). This lower differentiation is due to the fact that K may be partially supplied by mineral dust (clay minerals and feldspars) and probably by specific industrial sources.

Trace elements
Levels of Ti, Mn, Sr, Cu, Zn, Sn, Ni, Ba and Pb are higher in MLN, and in some cases in POR (Zn, Ba, Pb, Cd), when compared with the other sites. Most of these elements are usually related to metallurgy and heavy industry emissions. Levels of As and Se (generally used as tracers of coal combustion) were relatively higher in POR and ATH, pointing to a possible local, regional or long-range transport influence of emissions from coal combustion sources on air quality. Northern Greece has a number of large coal-fired power plants that may partly account for these concentration levels. Transport of emissions from coal-fired power plants from the Balkans, eastern Europe and Turkey may also contribute. In the case of POR, the coal power plant located 15 km southeast of Porto was deactivated in December 2004, and there was no coal combustion activity in the area during the campaign. The closest coal-fired power plant is located in northwestern Spain, at around 300 km from Porto. It is possible that other industrial sources of As and Se (ceramic, glass, and cement production, among others) may also contribute to increased levels of these elements. Levels of Cu, Sb, Ba and Sn were relatively higher in POR-TR as compared with UB sites, due to their association with non-exhaust vehicle emissions including brake and tyre wear (Gietl et al., 2010).
Levels of V were 4 to 2 times higher in the cities with a harbour (POR, BCN and ATH), pointing to the fact that, in spite of the possible use of fuel oil or petroleum coke for power generation or industrial processes, shipping emissions are the main sources of these heavy oil combustion tracers in these AIRUSE cities.
Rare earth elements (REEs) usually occur in mineral dust, and the ratio Ce / La in the earth's crust is close to 2. The marked decrease in this ratio may be used as a tracer of the influence of emission from La-based oil cracking in petrochemical plants. In this study most of the cities have a ratio Ce / La close to 2.0, with the exception of FI and ATH (1.0 and 1.3), this indicating the possible influence of this type of emission at both sites, although other sources of La cannot be discounted. In Athens the refinery is located 15 km northwest of the city centre, while for Florence, about 90 km away to the west, in the province of Livorno.
The unaccounted mass of PM is that resulting from the difference in the gravimetric measurements of the filters and the sum of all the components determined by chemical analysis. This unaccounted mass is usually attributed to water molecules contained in potential remaining moisture, and crystallization and formation water (water molecules in the structure of specific chemical species). In this study the unaccounted mass was higher at BCN-UB and POR-TR (23 and 31 % for PM 10 and PM 2.5 , respectively) and similarly lower (5-14 %) at all the other sites.

Source apportionment results
For each AIRUSE station, the best PMF solution was found when combining PM 10 and PM 2.5 samples in a single input   (Paatero and Hopke, 2008), in order to reduce rotational ambiguity and drive the iterations towards a more realistic solution. Auxiliary equations were therefore introduced by means of the USEPA software, and included both physical and chemical constraints. Table 2 summarizes the constraints used in each case, as well as the increment of dQ , i.e. the increase in the object function due to the use of auxiliary equations, which in all cases were accomplished with a total increase in dQ within 3-19 %.
The distribution of residuals, G-space plots, Fpeak values and Q values was explored for solutions with a number of factors varying between 6 and 11 (US EPA, 2014). Table 2 also lists the factors/source finally identified in each case, which vary between 8 and 10. Factor labelling was kept as generic as possible since perfect decomposition is difficult to achieve and factors likely represent a mixture of different sources and processes (e.g. vehicle non-exhaust). in BCN-UB, FI-UB and ATH-SUB, while in MLN-UB and POR-TR, these two factors were combined in a composite heavy oil and secondary (HOS) factor. In addition, the IND source was identified at BCN-UB, POR-TR and MLN-UB, while possible industrial contributions in FI-UB and ATH-SUB would be included in the HOC factor. Finally, in FI-UB, two sea salt factors (aged, with the influence of the refinery located at the coast, and fresh) and two mineral dust factors (local and Saharan dust) were separated.

Traffic sources
The NEX source generally shows a mixed composition (Fig. 3), including crustal species (Fe, Ca, Al and Si, among others), carbonaceous compounds (EC generally higher than OC) and trace elements (Cu, Zn, Mn and occasionally Sn, Sb and Ba) from brake wear (see the percentage of species in Tables S2-S6). However, rather dissimilar chemical profiles were found when comparing different cities (Fig. 4). Although the enrichment in Fe is common to all the cities, the main component of NEX can be either Ca (in BCN-UB), EC (in POR-TR and MLN-UB), OC (in FI-UB) or S (in Athens-SUB). These differences can be due to several factors: -the proximity to the source: at the traffic site (POR-TR) 28 % of the NEX source is made up by EC, which is probably related to brake particles; -the climatic conditions: Ca is higher in drier regions (BCN-UB and ATH-SUB) due to the enhanced resuspension when compared to POR-TR and MLN-UB; and -the type of materials used for brakes and road pavement (the higher OC in Florence might be due to higher road wear compared to other cities).
Other important differences are the absence of OC in POR-TR (again likely due to the dominance of brake wear particles) and in ATH-SUB, and the high abundance of NO − 3 in POR-TR and NH + 4 in Athens. Note that some elements are absent in some cities since they were not used for the source apportionment study (see Table 2).
The VEX source at all cities is almost totally composed of carbonaceous aerosol (Fig. 3), with the sum of OC + EC approximately 90-98 % of the mass. The ratio OC / EC varies widely among different cities. The lowest OC / EC value was found at the traffic site of Porto (POR-TR) due to its proximity to the source and the consequent lower proportion of secondary OC. The value varies between 1.8 and 3.7 at the UB sites, probably linked to the distance from main roads (BCN-  UB, MLN-UB and FI-UB). A much higher value (16.4) is observed in ATH-SUB due to the reduced share of diesel vehicles in the local fleet. Besides EC and OC, other components rarely reach 1 % of the mass: Zn in MLN-UB, S in BCN-UB and POR-TR due to their relatively higher content in diesel fuel and/or to engine oil anti-wear additive ZDDP (zinc dialkyldithiophosphate). Potassium is present in all unleaded fuels (Spencer et al., 2006), and it is also used as an antifreeze inhibitor and as an additive in some oil types.
The annual mean contributions for VEX and NEX sources and associated errors for each city are shown in Figs. 5 and 6. The error estimates were calculated based on the standard error of the coefficients of a multiple regression between the daily PM concentration (independent variable) and the source contributions (dependent variables). As a result we obtained the symmetric error for each source contribution (Fig. 6).
The sum of VEX and NEX contributions to PM 10 varies significantly in absolute terms (3.9-10.8 µg m −3 ), with the maximum found at POR-TR; the fractional contribution to PM 10 is within 18-31 %. Similarly in PM 2.5 , absolute annual contributions vary within the range 2.3-9.4 µg m −3 , but the percentage is quite constant within 14-37 % (Fig. 5). For the total contribution one should add the fraction of secondary nitrate attributable to traffic, but the lack of linearity between emissions and concentrations and the uncertainty in emission inventorying do not allow for a robust estimate. Although the importance of NEX sources has been increasing during the last decade due to the lack of mitigation measures, VEX contribution is still generally higher than NEX. However, the difference (in PM 10 ) between the VEX and NEX contribution is not significant at the BCN-UB, FI-UB and ATH-SUB sites, which can be generally interpreted as an equal contribution from the two sub-sources. At the TR site (POR-TR) the VEX contribution is significantly higher (by a factor > 2). . Phys., 16, 3289-3309  In PM 2.5 the share between VEX and NEX varies considerably, depending on the site (Fig. 6). At POR-TR, BCN-UB, FI-UB and ATH-SUB, the emissions from VEX have a higher contribution, but not in MLN-UB. The NEX contribution is significantly lower in PM 2.5 due to the coarser size distribution; only in MLN-UB are the VEX and NEX contributions in PM 2.5 similar. The daily variation of source contributions at all sites can be seen in Fig. S5.

Atmos. Chem
Individual source contributions were also averaged during days only of exceedances of the PM 10 daily limit value (50 µg m −3 ), except in BCN-UB, where this threshold was never reached and 40 µg m −3 was used. During exceedance days (or high pollution days), the sum of VEX and NEX contributions changes in PM 10 and PM 2.5 , respectively, from 25 and 21 % to 27 and 22 % in BCN-UB, from 31 and 31 % to 43 and 32 % in FI-UB, from 16 and 15 % to 19 and 13 % in MLN-UB, from 18 and 20 % to 7 and 10 % in ATH-SUB and from 31 and 37 % to 31 and 33 % in POR-TR (Table 3).

Secondary processes
The secondary nitrate (SNI) factor reproduces nearly the totality of the mass of ammonium nitrate formed from NO x precursors emitted from traffic, biomass burning and industries. The contribution of road traffic to this factor is unknown. The composition of this factor (Fig. 3) is very simi-lar among the different cities, NO − 3 being the main component (28-50 %). Nitrate is usually neutralized by ammonium (5-16 %) or, in the case of ATH-SUB, by Na (11 %). Another important component is OC, representing the condensation of semi-volatile organics on the high specific surface area of the ammonium nitrate particles. EC is also present in significant concentrations (> 1 %), although not in the case of POR-TR, where other primary elements can be observed in this profile (Si, Al, Ca, Fe and K). As previously explained, the traffic-related share of secondary nitrate is unknown, but at most of the sites it is likely to be very important: as an example, one should consider that the traffic share of the NO x emission inventory for the study areas is between 30 and 80 %. However, the high uncertainty in the emission inventory (mostly for the lack of biomass combustion emissions) does not allow drawing of quantitative estimates. The secondary nitrate annual average contributions to PM 10 range from 2.2 µg m −3 in FI-UB, 3.0-3.2 µg m −3 in POR-TR, BCN-UB and ATH-SUB, to 11.0 µg m −3 in MLN-UB. For PM 2.5 the annual contributions are in the range between 0.7 µg m −3 (ATH-SUB) and 8.9 µg m −3 in MLN-UB. As already mentioned, these high nitrate levels in the Po Valley are due to several factors: (i) a large urban and industrial agglomeration and associated road traffic; (ii) an intensive use of biomass burning in the basin; (iii) high NH 3 levels, emitted from agricultural and animal husbandry activities; and (iv) the peculiar meteorology favouring frequent and marked thermal inversions.
Factor SSO was found at all cities with a rather constant composition, revealing a mainly regional origin of these aerosols across the Mediterranean (Fig. 3). The main components are S, NH + 4 and OC. EC can also be found at 4-5 % concentration in BCN-UB and MLN-UB, indicating a mixing with other (primary) sources. At two sites (POR-TR and MLN-UB), this factor was combined with the heavy oil combustion, thus adding NO − 3 , V and Ni, which in the other cities (BCN-UB, FI-UB and ATH-SUB) appears as an independent source. Other elements that can appear in this source are Si, Na and K, although in concentrations below 1 %. SSO aerosols are at all sites in the fine mode (ratio PM 2.5 / PM 10 close to 1), but the variability of contributions across the Mediterranean does not show the same pattern observed for the concentrations of sulfate. The contributions were progressively increasing from POR-TR (3.3 µg m −3 in PM 2.5 ), ATH-SUB (3.8 µg m −3 ), and FI-UB (4.2 µg m −3 ) to MLN-UB and BCN-UB (5.7 µg m −3 ) (Fig. 5). The reason for this higher contribution might be related to the inclusion of some local sources, as suggested by the presence of EC primary combustion particles only at these two sites. The seasonal trend is clear with maxima in the warmer months, probably due to the enhanced photochemical activity; this is less clear in MLN-UB, where high concentrations in autumn and winter may be due to aqueous phase formation in the fogs which occur frequently at this time of year (Fig. S5). POR-TR registered the highest daily peaks, when the site is under the influence of NW winds, probably due to the emissions from the Porto Refinery, a plant carrying out crude oil industrial processing for the petrochemical industry. During exceedance days (or high pollution days), the SSO contribution in PM 10 -PM 2.5 always decreases from 27-38 to 19-22 % in BCN-UB, from 10-13 to 5-2 % in POR-TR, from 21-30 to 6-6 % in FI-UB, from 13-19 to 9-11 % in MLN-UB, and from 19-32 to 2-5 % in ATH-SUB (Table 3).

Marine aerosols
Depending on the geography of each site, sea salt can reach the receptor in a "fresh" or "aged" form, or both. The main chemical difference is the absence of chlorine in the aged sea salt coupled with the presence of the nitrates, which means that sodium is in the form of sodium nitrate. Fresh sea salt was identified at POR-TR, FI-UB and ATH-SUB, while aged sea salt was found at MLN-UB, FI-UB and BCN-UB. As shown in Fig. 3, the mass fraction of nitrate (µg µg −1 ) is considerably higher in the aged profiles, by 1 order of magnitude when compared to the fresh profile. In FI-UB, where both fresh and aged factors were found, nitrate is absent in the fresh sea salt and chlorine is absent in the aged factor, which also shows a particularly high level of OC, indicating the mixing of the aged sea salt with anthropogenic plumes, likely due to the refinery emissions at the coast of Livorno. In all factors the absence of ammonium indicates clearly the neutralization of nitrate by sodium. Mean contributions of sea salt (Fig. 5) in PM 10 were revealed to be generally higher (by 25 %) than what was found with the chemical speciation (2.5 vs. 1.5 µg m −3 in BCN-UB, 5.5 vs. 4.4 µg m −3 in POR-TR, 1.8 vs. 0.6 µg m −3 in FI-UB, 1.1 vs. 0.7 µg m −3 in MLN-UB, and 1.0 vs. 1.6 µg m −3 in ATH-SUB) due to the involvement of coarse nitrate and water. In PM 2.5 the PMF contribution matches with the chemical data (slope = 1, r 2 = 0.95).
Consequently, the comparison of PMF contributions reflects the same conclusions drawn in the PM speciation section, with lower levels of sea salt at the inland Italian cities (FI and MLN) and higher at the Mediterranean coastal sites, with the highest contribution observed at the Atlantic site (POR). Daily source contributions can be seen in Fig. S5. During exceedance days (or high pollution days), the sea salt contribution in PM 10 -PM 2.5 generally decreases from 11-3 to 2-1 % in BCN-UB, from 16-4 to 3-0 % in POR-TR, from 9-2 to 0-1 % in FI-UB, or does not change in MLN-UB and ATH-SUB (Table 3).

Mineral sources
Saharan dust was separated from local dust only at FI-UB, BCN-UB and ATH-SUB (by means of PMF in FI and by the percentile method for BCN and ATH). In MLN-UB and POR-TR the contribution of Saharan dust was estimated as negligible. The chemical profile of Saharan dust (as estimated in FI-UB by PMF) shows a composition very close to the average of the earth's crust, with enrichment factors (EF) calculated with respect to Al using the average continental crust composition reported by Mason (1966) and Rahn (1976) that are all close to 1: 0.99 (Mg), 0.83 (Si), 1.08 (K), 0.92 (Ca), 1.0 (Ti) and 1.1 (Fe) (Fig. 3). The annual mean Saharan dust contribution during AIRUSE sampling days was estimated as 0.3 µg m −3 (1 %) in BCN-UB, 0.7 µg m −3 (4 %) in FI-UB, and 3.0 µg m −3 (14 %) in ATH-SUB (Fig. 5). This large difference is due to the southern location of Athens and the severity of some Saharan dust episodes in the eastern part of the basin with a higher frequency than average during the period of this study. As mentioned by previous studies, Saharan dust transport occurs in different seasons on the western and eastern sides of the Mediterranean  , with a PM 2.5 / PM 10 ratio equal to 0.2 in both cases. Concerning PM exceedances, the relative burden of Saharan dust increases during exceedance days only in ATH-SUB, where it is on average the main source of PM 10 (52 %) and PM 2.5 (45 %) when the daily limit value of 50 µg m −3 is exceeded. No contributions from SAH were found in FI-UB and BCN-UB during days with PM 10 above 50 and 40 µg m −3 , respectively. Besides the long-range transported mineral dust from the Sahara, a significant part of mineral dust was found to be locally emitted in all cities (local dust, LDU). The ratio SAH / LDU is usually very low, < 0.1 in POR-TR and MLN-UB, 0.1 in BCN-UB and 0.3 in FI-UB, indicating that at these sites Saharan dust is 0-23 % of measured mineral dust, with the rest emitted by human activities or of local origin. Only in ATH-SUB was the SAH contribution higher (ratio SAH / LDU = 1.4) than the local dust due both to the geographical position of Athens and the suburban location of the measurement site (i.e. a lower anthropogenic contribution than at UB sites). The chemical profile of LDU is the one shown in Fig. 3 (the MIN profile is used). Although the major components are similar at all sites (Si, Al, Ca, Fe, OC and K), some differences can be observed. The highest ratio Ca / Al is found in FI-UB (6.3) due to the clear PMF separation from the SAH contribution (ratio Ca / Al = 0.4, similar to the earth's crust). The other cities show Ca / Al ratios in the mineral dust factors varying from 0.5 (POR-TR) to 2.2 (BCN-UB), which are influenced by the SAH contribution as well as the local geology and share of Ca-rich local emissions. The ratio Si / Al varies between 2.0 and 3.8. The ratio OC / Ca is generally well above the stoichiometric CC / Ca ratio in calcite (0.3), revealing additional sources of organic carbon such as biogenic OC and/or road dust, mostly in MLN-UB, where the complex atmospheric dynamics impede the separation of sources. Other common components of LDU are NO − 3 (except in POR-TR), S likely from gypsum, Cu, Zn and Ba, suggesting possible interference from road dust. LDU contributions to PM 10 range between 8 and 12 % at SUB and UB sites (2.4-3.3 µg m −3 ) and increase to 6.3 µg m −3 (18 %) at the TR site, revealing a contribution from road dust resuspension. In PM 2.5 the SUB-UB range was 0.3-1.5 µg m −3 (2-7 %) and 3.8 µg m −3 (15 %) at the TR site. The relative contribution (%) does not increase during exceedance days, with the exception of POR-TR, where it rises to 28 and 20 % for PM 10 and PM 2.5 , respectively. The daily variation of LDU contributions shows generally higher values from spring to autumn at all cities (Fig. S5). Above the background contributions, sporadic peaks are also found, mostly at POR-TR, probably related to road dust emissions not completely included in the NEX factor.

Heavy oil combustion
As already mentioned, the contribution of heavy oil combustion was separated only at the cities nearest to the Mediterranean shipping routes (BCN, FI and ATH) in spite of the fact that V and Ni concentrations in POR-TR were twice as high as in FI-UB. In POR-TR and MLN-UB, the HOC source is mixed with the SSO in the combined source HOS. HOC particles are commonly composed of EC, OC and S, explaining the high variance of V and Ni. The difference between AIRUSE sites concerns only specific trace elements such as Zn, Sn and Ba (in BCN-UB), Ba and Se (in FI-UB), or Sr and Sb (in ATH-SUB), although only a small amount of variance of these elements is explained by HOC (Tables S2-S6). The annual contribution of HOC is practically the same in BCN-UB, FI-UB and ATH-SUB (0.8-1.0 µg m −3 in PM 10 and 0.7-0.8 µg m −3 in PM 2.5 ). The contribution in POR-TR and MLN-UB could not be separated from the composite HOS. The daily variation of HOC contributions, as estimated by PMF, is shown in Fig. S5. In FI-UB and BCN-UB higher contributions are observed in summer due to the higher air circulation, which favours their transport and distribution across the regional area, while in ATH-SUB no clear seasonal trend is observed for HOC probably because of the variety of the sources that may contribute to this factor (residential heating and shipping). During exceedance days (or high pollution days), the HOC contribution in PM 10 -PM 2.5 does not change significantly.

Industrial emissions
The impact of industrial emissions was identified in only three cities: BCN-UB, POR-TR and MLN-UB. Florence and Athens are in fact the less industrialized cities among the AIRUSE consortium. In the three industrialized cities, OC and Fe are commonly present as the main components (Fig. 3). Besides OC and Fe, in BCN-UB and POR-TR the presence of Zn, Pb, S, Cu, Cd, Sb and Mn indicates hightemperature metal processing, pointing to the smelters located south-west of BCN-UB and east of POR-TR. In MLN-UB the industrial source presents a different chemical profile with NO − 3 , EC, Ca and Cl as the main components (together with OC and Fe) and a high variance of Cr, Ni, Cu and Mn explained, suggesting a more mixed origin, including metallurgy and construction activities (Tables S2-S6). The impact of industrial emissions upon PM 10 and PM 2.5 is similarly low at MLN, POR and BCN, ranging between 1.2 and 3.3 µg m −3 as annual means and with a PM 2.5 / PM 10 ratio close to 1. No typical seasonal trend is observable at any site (Fig. S5). The contribution was rather constant throughout the year (around 1 µg m −3 daily) in MLN-UB and BCN-UB, while elevated As previously shown, concentrations of levoglucosan varied by over 1 order of magnitude among the AIRUSE cities, indicating very contrasting scenarios across the Mediterranean for emissions from biomass burning. Similarly, the PMF identified a biomass burning source only in four of the five cities due to the low levels of levoglucosan in BCN-UB (20 ng m −3 as annual mean). In all other cities, levoglucosan is the main tracer in the chemical profile of the biomass burning source identified by PMF (except in ATH-SUB, where levoglucosan was not used as input species for PMF due to the high S / N ratio). Levoglucosan represents 4-8 % of PM mass in the biomass burning factor profile; OC and EC are the major components in the BB profile (12-65 and 4-14 %, respectively). In spite of these quite large ranges, the OC / EC ratio can be used as a more robust diagnostic of BB composition. The OC / EC ratio in BB aerosols varies from 2.6 (ATH-SUB), 2.9 (POR-TR), 4.6 (MLN-UB) to 6.1 (FI-UB), which may be explained by a higher proportion of secondary organic aerosols in MLN-UB and FI-UB or by different wood types and combustion appliances. Also, K (probably the soluble fraction) tracks BB aerosols, representing 2-4 % of the mass. Other components can be observed, although more sporadically, such as Cl, S, Zn, Pb, NH + 4 and NO − 3 ( Fig. 3 and Tables S2-S6). BB contributions reproduce quite well the gradients found for levoglucosan among the AIRUSE cities. Although levoglucosan has been detected in some samples from BCN, biomass burning could not be assigned as a significant contributor to PM. In the other cities, an annual mean of 1.2-1.4 gm −3 (7-10 %) is estimated in ATH-SUB, 2.9-2.8 µg m −3 (15-20 %) in FI-UB, 4.2-4.4 µg m −3 (12-17 %) in POR-TR, and up to 7.3-5.1 µg m −3 (17 %) in MLN-UB. Therefore, this reveals quite a contrasting impact of BB emissions across the Mediterranean, depending on the type of fuel and combustion device used in each region for residential heating. Differently from other cities, Barcelona is well supplied with natural gas for residential heating; Florence is also well supplied with natural gas, but in the suburbs on the hill there are commonly chimneys. Even in Milan the use of natural gas for heating is very extensive; however, also due to the current economic crisis, many citizens are equipped with small pellet stoves. In ATH the BB source is also associated with tracers of waste combustion, such as As, Cd, Sb and Pb, with explained variance ranging between 12 and 72 %, as citizens of Athens have turned to alternative heating fuels, such as wood, due to the economic crisis and the increased prices of diesel oil, which has been the regular means of residential heating in Greece. In many cases, treated wood or even combustible wastes are now used as fuel. As previously shown for the traffic source, another factor identified by PMF was secondary nitrate (SNI). Although in urban environments nitrate mainly arises from NO x from traffic, a substantial fraction can also be derived from biomass burning emissions. Therefore for each city, the corresponding share of NO x due to biomass burning can also be applied to SNI. Based on this approach, percentages of 16 and 13 were adopted in POR-TR and MLN-UB, respectively, to account for SNI from biomass burning. In FI-UB, on the basis of the emission inventory, about 10 % of NO x emissions are due to domestic heating, with only 2 % attributable to stoves and chimneys (http://servizi2.regione.toscana.it/aria/); it is however suspected that these data underestimate the contribution of domestic heating BB to NO x . Thus, in POR-TR, the total contribution from BB was estimated to be 4.7 µg m −3 (13 % of PM 10 ) and 4.6 µg m −3 (18 % of PM 2.5 ). In MLN-UB, the total contribution from BB represented 8.7 µg m −3 in PM 10 (23 %) and 6.3 µg m −3 in PM 2.5 (21 %).
The impact of BB emissions is especially high in the winter months (Fig. S5), due to the generalized use of wood for residential heating (Gonçalves et al., 2012). The contribution of BB to PM in POR-TR was also higher in September. Several wildfires were registered in the Porto district in this particularly hot and dry month. In MLN-UB the stagnant conditions and reduced boundary layer height induced by the typical meteorology of the Po Valley also enhance BB contributions during winter months. The relative contribution from biomass burning to PM 10 and PM 2.5 generally increases on exceedance days in POR-TR, MLN-UB and FI-UB. The percentages increase in fact from 23-21 to 35-26 % in MLN-UB, from 13-18 to 25-36 % (POR-TR), and from 15-20 to 30-32 % (FI-UB). Conversely, in ATH-SUB, during exceedance days, the contribution from BB is substantially reduced (from 7-10 to 1-2 %, Table 3). This is again due to exceedances in Athens being caused by African dust intrusions and the relative suppression of other source contributions during exceedance days.
Recent research attention has been focused on the importance of other sources/processes contributing to the nonfossil OC in urban ambient air, namely food cooking (Crippa et al., 2013;Allan et al., 2010;Mohr et al., 2012) and enhanced biogenic secondary aerosols (Hoyle et al., 2011;Kroll et al., 2006). The quantification of these two sources/processes requires the use of a combination of measurement techniques: Aerosol Mass Spectrometer (AMS) and offline radiocarbon analysis on PM filters, which could not be performed within AIRUSE. For example, in summer in Barcelona, cooking was estimated to be responsible for 17 % of organic aerosols (OA being 60 % of PM1) (Minguillón et al., 2015), while in Athens, COA was not found as a separate source, but mixed with the traffic-related factor, contributing 17 % of OA (Kostenidou et al., 2015). The influence of road traffic emissions such as NO x on the forma-tion of biogenic secondary aerosols (Hoyle et al., 2011) has been observed in Barcelona, where an increase in non-fossil secondary organic aerosol was registered when comparing a period with low traffic emissions with a period with high traffic emissions, supported by the increase in individual organic compounds such as poly-acids (malic acid, 3-hydroglutaric acid, MBTCA and 2-methylglyceric acid) (Minguillón et al., 2015).

Conclusions
Based on the simultaneous chemical characterization of PM 10 and PM 2.5 samples collected every third day across 12 months at five southern European cities, important differences were observed. Thus, in FI-UB, MLN-UB and POR-TR, OM + EC was the main PM 10 constituent, accounting for 57, 45 and 37 % of the PM 10 load and 69, 45 and 48 % of the PM 2.5 . However, at BCN-UB and ATH-SUB these PM components were still dominant, but made up only 27 and 36 % of PM 10 and 30 and 45 % of PM 2.5 .
The PM fraction accounted for by secondary inorganic aerosols (sulfate, nitrate and ammonium) also showed significant variations with a load of 18-20 % of PM 10 for FI-UB and BCN-UB, 30 % for ATH-SUB and 35 % in MLN-UB, and 21-22 % of PM 2.5 for FI-UB and BCN-UB, 33 % in MLN-UB and 37 % for ATH-SUB. For the POR-TR site the SIA percentage was reduced due to the increase in contributions from OM + EC (mostly from exhaust emissions) and mineral dust (mostly from road dust), and made up 15 % of PM 10 and PM 2.5 .
Mineral dust was one of the main PM 10 components at all sites, ranging from 10 (MLN-UB) to 21 % (ATH-SUB) of the PM 10 load. In PM 2.5 , as expected from the coarser mode of occurrence of mineral dust, the contribution was reduced down to 4 (FI)-10 % (ATH) in all cases.
The sea salt contribution reached the maximum at the Atlantic site, 13 % of PM 10 and 5 % of PM 2.5 in POR, followed by the other two Mediterranean coastal sites, BCN and ATH (7 and 8 % of PM 10 and 3 and 2 % of PM 2.5 ); the minimum sea salt contributions were recorded in the inland Italian cities, MLN and FI (2-3 % of PM 10 and 1 % of PM 2.5 ).
The sum of vehicle exhaust and vehicle non-exhaust emissions is unequivocally the most important source of PM 10 (18-31 % at all sites, except in MLN-UB, where SNI and SSO dominate the mass), while for PM 2.5 it is clearly the largest source of POR-TR (37 %) and the second one at BCN-UB, FI-UB and ATH-SUB after secondary sulfate and organics (which likely receives significant transboundary contributions). In PM 2.5 MLN-UB secondary nitrate is the main contributor, although it does not identify one specific source.
Another important source of PM 10 is biomass burning (13 % in POR-TR, 15 % in FI-UB, and 23 % in MLN-UB), although it is only 7 % in ATH-SUB and negligible in BCN-UB. In PM 2.5 , BB is the second most important source in MLN-UB (21 %) and in POR-TR (18 %), the third in FI-UB (20 %) and ATH-SUB (10 %), but again negligible (< 2 %) in BCN-UB. This large discrepancy among cities is mostly due to the degree of penetration of wood (and its derivatives) as fuel for residential heating. In Barcelona natural gas is very well supplied across the city and used as fuel in 96 % of homes, while, in other cities, PM levels increase on an annual basis by 1-6 µg m −3 due to this source.
Other significant anthropogenic sources are the following.
-Local dust, 8-12 % of PM 10 at the SUB and UB sites and 18 % at the TR site, revealing a contribution from road dust resuspension. In PM 2.5 , percentages decrease to 2-7 % at the SUB-UB sites and 15 % at the TR site.
-Industry, mainly metallurgy, contributing 4-8 % of PM 10 (5-9 % in PM 2.5 ), but only at BCN-UB, POR-TR and MLN-UB. No clear impact of industrial emissions was found in FI-UB and ATH-SUB.
-Natural contributions consist of sea salt (16 % of PM 10 in POR-TR but only 2-11 % in the other cities) and Saharan dust (14 % in ATH-SUB but less than 4 % in the other cities).
-Other sources of non-fossil OC, such as food cooking and the formation of enhanced biogenic secondary organic aerosol, could not be separated due to the lack of specific techniques.
The Supplement related to this article is available online at doi:10.5194/acp-16-3289-2016-supplement.