Measurements were conducted at the European Commission – Joint Research Centre (EC-JRC) Ispra site (45∘48′ N, 8∘38′ E; 217 m a.s.l.; Fig. S1 in the Supplement), which is part of the European Monitoring and Evaluation Programme (EMEP) measurement network (http://www.nilu.no/projects/ccc/sitedescriptions/it/index.html) and the Global Atmosphere Watch (GAW) regional stations (http://www.wmo.int/pages/prog/arep/gaw/measurements.html). It is located on the northwest edge of the Po Valley region, 60 km northwest of the Milan urban area. It can be regarded as a “regional/rural background” site following the criteria recommended by the European Environment Agency (Larssen et al., 1999). For simplicity, the term “regional background site” will be used in the following although comparisons with rural background sites from other studies will also be reported. Further information on the study site can be found in Putaud et al. (2014b).

The recently developed ACSM (Aerodyne Research Inc., ARI) was used to measure the nonrefractory chemical composition (organics, nitrate, sulfate, ammonium, chloride) of submicron particles (PM1) with a 30 min time resolution. The operating principle of the ACSM is similar to the widespread Aerodyne aerosol mass spectrometer (Canagaratna et al., 2007; Jayne et al., 2000), with the difference that the former does not inform on the size distribution of the chemical composition of NR-PM1. A full description of the ACSM can be found in Ng et al. (2011a). Briefly, an aerodynamic lens is used to focus submicron particles (50 % transmission range of 75–650 nm; Liu et al., 2007), which are then vaporized in high vacuum, ionized by electron ionization (at 70 eV) and detected by a quadrupole mass spectrometer (Pfeiffer Vacuum Prisma Plus RGA). Two different quadrupole ACSMs (Q-ACSMs) were used in this study (from March 2013 to February 2014): Q-ACSM1 from 1 March to 18 August 2013 and Q-ACSM2 from 20 June 2013 to 28 February 2014. Note that Q-ACSM2 was not running from 3 November to 18 December due to its participation in the first inter-ACSM comparison exercise (Crenn et al., 2015). The reproducibility and consistency with independent measurements are discussed in Sect. 3.1. In the following, orthogonal regressions are reported unless otherwise stated.

Both ACSMs were operated with the latest Data Acquisition (DAQ 1.4.3.8 to 1.4.4.5) and Data Analysis (DAS 1.5.3.0 to 1.5.3.2) software (ARI, https://sites.google.com/site/ariacsm/mytemplate-sw) available at the time of use, which are developed within Igor Pro 6.32A (Wavemetrics). Recommendations provided by Aerodyne (2010a, b) and Ng et al. (2011a) were followed for the operation, calibration and data analysis of the ACSMs. Ammonium nitrate calibrations were performed seasonally and used for the determination of experimental nitrate response factors (RFs) and ammonium relative ionization efficiencies (RIEs; see Sect. S1 in the Supplement for further details). Annual average and season-dependent experimental RF and RIE values were alternatively applied to assess whether the ACSM is stable over multi-seasonal periods (see Sect. 3.1 for results). Seasons are defined as spring (MAM), summer (JJA), autumn (SON) and winter (DJF). RIEs for organics, nitrate and chloride (1.4, 1.1 and 1.3, respectively) were taken from the literature (Canagaratna et al., 2007; Takegawa et al., 2005). RIE for sulfate was experimentally determined based on ammonium sulfate calibrations for ACSM2 and was taken from the literature for ACSM1 (see Sect. S1). Collection efficiencies (CEs) set as (i) a fixed 0.5 value (e.g. Budisulistiorini et al., 2013) or (ii) following the composition-dependent CE algorithm introduced by Middlebrook et al. (2012) were compared in order to determine the most appropriate CEs (see Sect. 3.1 for results).

Additional measurements routinely performed at the JRC-Ispra site are used in this study (see Putaud et al., 2014a, for a full description). PM2.5 was sampled on quartz fibre filters (Pall, 2500 QAT-UP) with a Partisol PLUS 2025 sampler equipped with a carbon honeycomb denuder operating at 16.7 L min-1 from 1 March 2013 to 28 February 2014 with daily filter changes at 08:00 UTC. Major ions (NH4+, K+, NO3-, SO42-, etc.) are analysed by ion chromatography (Dionex DX 120 with electrochemical eluent suppression) after extraction in Milli-Q water (Millipore). Organic and elemental carbon (OC and EC, respectively) are quantified by a thermal–optical method (Sunset Laboratory Inc. dual-optical lab thermal–optical carbon aerosol analyzer) using the EUSAAR-2 protocol (Cavalli et al., 2010). Equivalent black carbon (BC) is measured by a multi-angle absorption photometer (MAAP, Thermo Scientific, model 5012) applying an absorption cross section of 6.6 m2 g-1 of equivalent BC at the operation wavelength of 670 nm. Particle volume concentrations are determined with a homemade differential mobility particle sizer (DMPS) combining a Vienna-type differential mobility analyser (DMA) and a condensation particle counter (CPC, TSI 3010), following the European Supersites for Atmospheric Aerosol Research (EUSAAR) specifications for DMPS systems (Wiedensohler et al., 2012). Meteorological variables (temperature, pressure, relative humidity, precipitation, wind speed and direction) are determined from a weather transmitter WXT510 (Vaisala, Finland). Solar radiation is measured by a CM11 pyranometer (Kipp and Zonen, the Netherlands).

The organic fraction was apportioned using the positive matrix factorization approach (PMF; Lanz et al., 2007; Paatero and Tapper, 1994; Ulbrich et al., 2009; Zhang et al., 2011) by applying the Multilinear Engine 2 algorithm (ME-2, Paatero, 2000) implemented in the SoFi tool (v4.8, Canonaco et al., 2013; Crippa et al., 2014). Details on the theory and application of PMF and ME-2 can be found in the aforementioned studies. Briefly, PMF aims at factorizing an initial X matrix (representing the temporal variation of m/z signals here) into two F and G matrices (representing factor profiles (FPs) and contributions, respectively), putting a constraint of non-negativity on F and G matrices. Contrary to the classical program used to resolve PMF (e.g. PMF2, PMF3), ME-2 allows any element of the F and G matrices to be constrained with a certain degree of freedom. This ME-2 approach has been typically used to constrain full FPs (e.g. Amato et al., 2009; Crippa et al., 2014), specific elemental ratios (e.g. Sturtz et al., 2014) or specific species contribution (e.g. Crawford et al., 2005) in a given FP.

In our study, ME-2 is applied with and without constraining FPs, using the so-called a value approach (Canonaco et al., 2013) in the former case, which can be described as follows: (fk,j)solution=(fk,j)reference±a(fk,j)reference , where k and j are the indexes for the factors and the species, respectively, fk,j is the element (k, j) of the F matrix, the index “solution” stands for the PMF user solution, “reference” is the reference profile and “a” is a scalar defined between 0 and 1 (e.g. applying an a value of 0.10 lets ±10 % variability to our FP solution with respect to the reference FP). Following Crippa et al. (2014), we perform a sequence of runs with (i) unconstrained PMF, (ii) fixed HOA, (iii) fixed HOA and BBOA, and (iv) fixed HOA, BBOA and cooking OA (COA) factors before selecting the most appropriate solution. Uncertainties are calculated using the DAS 1.5.3.0 version following the methodologies of Allan et al., (2003a) and Ulbrich et al. (2009). m/z 12 and 13 are removed for SA analysis since negative signals are observed most of the time. Reference factor profiles (RFPs) are taken from ambient deconvolved spectra from the aerosol mass spectrometry (AMS) spectral database (Ulbrich et al., 2015). HOA and BBOA profiles are taken from Ng et al. (2011c) (average of profiles from multiple studies) and COA from Crippa et al. (2013). Different a values are tested (see Sect. 3.2) applying (i) relative standard deviations of averaged RFPs defined for every m/z (i.e. assuming that the chosen averaged RFPs are representative of our dataset), (ii) recommendations of Crippa et al. (2014) based on the SA of 25 European AMS datasets and (iii) comparison with independent measurements (e.g. NOx, CO, BC). Solutions from two to eight factors are investigated in order to choose the appropriate number of factors (see Sects. S2 and 3.2).

Ammonium nitrate calibrations performed on each ACSM are shown in Fig. S2. RFNO3 and RIENH4 do not present significant seasonal variability – e.g. for ACSM2, RFNO3= 4.7E-11 ± 0.2E-11A. µg-1 m3 – suggesting constant calibration factors may be used throughout the campaign. However, calibration factors exhibit substantial discrepancies between both ACSMs (e.g. RFNO3 of 2.5E-11 and 4.7E-11A. µg-1 m3 for ACSMs 1 and 2, respectively), suggesting that instrument-specific factors are necessary. Applying constant and composition-dependent CEs does not lead to noticeable differences (e.g. for NR-PM1: r2= 0.97, slope = 1.00 ± 0.00, y intercept = 0.10 ± 0.03 µg m-3, n= 14842) due to (i) low sampling line RH (e.g. typically below 30 % for ACSM2) and (ii) few high-nitrate-content events (only 5 % of data exhibits ammonium nitrate mass fractions > 40 %, defined as high by Middlebrook et al., 2012). The Middlebrook et al. (2012) algorithm is, however, preferred since slightly acidic aerosols are observed at the study site (on average sulfate plus nitrate against ammonium in µeq m-3: r2= 0.96, slope = 1.21 ± 0.00, intercept = 0.01 ± 0.00 µeq m-3, n= 14842).

A comparison performed between the two ACSMs used in this study during a 2-month summer period is shown in Fig. S3. Very good correlations are observed for every chemical component (0.91 < r2 < 0.98, n= 1402, hourly average) – chloride excluded – with slopes relatively close to one (0.87 < slopes < 1.42), indicating a fairly good comparability between both instruments. One of the two ACSMs also participated in the first-ever inter-ACSM comparison exercise performed between 13 different European Q-ACSMs during 3 weeks in Paris, France (Crenn et al., 2015). Satisfactory performances – defined by z scores < 2 – are reported for our instrument regarding every chemical component and NR-PM1 mass, attesting the consistency of our measurements with other European sites.

Measurements performed by the ACSM and independent offline and online analytical techniques are compared in Fig. 1 and Table 1. An overall good agreement is found for every major components throughout the year (typically r2 > 0.8), although discrepancies are observable for specific species and seasons. On the annual scale, a good agreement (r2= 0.77, n= 317) is found between organics from ACSM and OC from filter measurements in spite of expected filter sampling artefacts (Maimone et al., 2011; Turpin et al., 2000; Watson et al., 2009). Even better agreements are observed on a seasonal basis (r2 ∼ 0.9), with steeper slopes in summer compared with winter, which likely reflects the different degrees of oxygenation of organics among seasons (leading to different ratios of organic matter (OM) to OC). However, these slopes cannot be directly regarded as the OM-to-OC ratios due to (i) differences in size fractions between both methods (PM1 for ACSM and PM2.5 for filter measurements) and (ii) uncertainties related to RIEOrg for ACSM measurements (Budisulistiorini et al., 2014; Ripoll et al., 2015). An estimation of the OM-to-OC ratio for submicron organics applying the methodology described by Canagaratna et al. (2015) is discussed in Sect. 4.2. Good correlations are observed for nitrate during all seasons (r2 > 0.9) but summer (r2= 0.5), which is most likely related to enhanced evaporative losses of ammonium nitrate from filter during the latter season (Chow et al., 2005; Schaap et al., 2004). Slopes range from 0.9 to 1.4 – summer excluded – which is comparable to what is reported elsewhere (Budisulistiorini et al., 2014; Crenn et al., 2015; Ripoll et al., 2015). Very good correlations are observed for sulfate in every season (r2= 0.9–1.0) with slopes close to 1 (0.9–1.1, winter excluded), consistent with its presence in the submicronic size fraction and its low volatility leading to the minimization of sampling artefacts. Note that discrepancies have been reported when comparing sulfate measured by the ACSM (Petit et al., 2015) or the AMS (Zhang, 2005) with independent measurements. Our results suggest that ammonium sulfate calibrations should be performed to experimentally determine sulfate RIEs, which appear to be instrument specific but stable over several months. Although aerosols are slightly acidic on average at the study site, ammonium mostly neutralizes nitrate and sulfate throughout the campaign and thus exhibits behaviour in between the two latter compounds. Higher uncertainties are associated with chloride from filter quantification, resulting in no agreement with ACSM measurements in summer when the concentrations are the lowest (r2= 0.00) and fairly good agreement during the other seasons (r2= 0.64–0.77). The high slope observed for ACSM1 (e.g. 2.1 during spring) compared to the fairly good slopes observed for ACSM2 (0.7–1.1) suggests that chloride RIE might be instrument specific and require appropriate calibrations for its accurate quantification (see also Riffault et al., 2013, on this topic).

The sum of NR-PM1 components and BC has been compared to the volume concentration of PM1. Good agreement is found between both variables (r2 > 0.8), giving further confidence on the consistency of our ACSM measurements. The annual average particle density estimated from this comparison (i.e. slope) is 1.6, which is typical of ambient aerosol particles densities (1.5–1.9 in Hand and Kreidenweis, 2002; Hu et al., 2012; McMurry et al., 2002; Pitz et al., 2003, 2008). The higher densities observed during spring and summer (1.9–2.0) than autumn and winter (1.3–1.5) are likely due to the enhanced contribution of secondary aerosol and aged particles during the former period (Pitz et al., 2008).

First, during the aforementioned inter-ACSM comparison study (Crenn et al., 2015), source apportionment of organics was performed based on data from 13 Q-ACSMs (Fröhlich et al., 2015), including one ACSM used in the present study. Satisfactory performances (z scores < 2) are reported for our ACSM using a similar approach as adopted in this study. This result demonstrates that our instrument and the associated data treatment, including the source apportionment modelling, are capable of accurately identifying and quantifying OA sources.

An overview of the chemical composition of NR-PM1 retrieved during this campaign is shown in Fig. 3. The annual averaged NR-PM1 mass reported here (14.2 µg m-3) ranges amongst the highest NR-PM1 levels (7th out of 41 sites) reported at rural and urban downwind sites in Europe (Crippa et al., 2014) and worldwide (Jimenez et al., 2009; Zhang et al., 2007, 2011). Please note that these previous studies are based on typically 1 month of measurements in different seasons. It is comparable to NR-PM1 levels reported during specific campaigns in the urban areas of New York City (USA, 12 µg m-3; Weimer et al., 2006), Tokyo (Japan, 12–15 µg m-3; Takegawa et al., 2006) or Manchester (UK, 14 µg m-3; Allan et al., 2003a, b). Our annual average NR-PM1 mass is higher than the 10 µg m-3 guideline given by the World Health Organization for the annual average PM2.5 mass (including refractory and nonrefractory compounds; WHO, 2006). After similar conclusions have been drawn for PM2.5 and PM10 size fractions (Putaud et al., 2010), the Po Valley appears to be one of the most polluted regions in Europe with regard to NR-PM1 levels this time. Submicron aerosol particles are mostly made of organics (58 %), nitrate (21 %), sulfate (12 %) and ammonium (8 %; Fig. 3). The predominance of organics is typical of urban downwind sites (e.g. average of 52 % reported in Zhang et al., 2011). However, the noticeable proportion of nitrate is characteristic of urban sites (18 % in Zhang et al., 2011), which likely reflects the substantial influence of anthropogenic activities emissions at our regional site. As a result, sulfate exhibits particularly low contributions at the study site compared with other locations (generally > 20 % in Zhang et al., 2011).

NR-PM1 levels present a clear seasonality with higher levels during spring (∼ 18 µg m-3) and winter (∼ 15 µg m-3) compared with summer and autumn (∼ 12 µg m-3). Higher levels were expected during cold months due to enhanced biomass burning emissions and lower boundary layer heights (BLHs), as previously observed at the study site (Putaud et al., 2013). Expected seasonal variations of the chemical composition of NR-PM1 are observed, with (i) higher nitrate contributions during the cold season which favours its partitioning in the condensed phase (Clegg et al., 1998), (ii) higher sulfate contributions during summer, which can e.g. be associated with enhanced photochemical production (Seinfeld and Pandis, 2006) and lower amount of rainout (Fig. S7), and (iii) relatively stable contributions for ammonium (mainly neutralizing the two previous species) and organics (discussed later on).

A focus will now be made on daily cycles of the chemical composition of NR-PM1 (Fig. 4), displayed for the first time during the four seasons in the Po Valley, thanks to the high time resolution and stability of the ACSM. On the annual scale, daily cycles of NR-PM1 levels are characterized by significantly higher concentrations during night-time than daytime, likely due to lower BLH, higher wood burning emissions (during cold seasons) and lower temperatures favouring the partitioning of semi-volatile inorganic (mainly ammonium nitrate) and organic material in the condensed phase, to name a few. A distinct peak is, however, observed around noon, probably caused by enhanced photochemical production of secondary organic compounds and increased BLH favouring downward mixing of advected pollution, especially during summer (Fig. 4; Decesari et al., 2014). Note that this annual daily pattern is the combination of distinct daily cycles varying with the season studied (Fig. 4). In terms of relative chemical composition, organics dominate NR-PM1 mass independently of the time of the day, with median contributions ranging from ∼ 60 to 70 %. Nitrate exhibits higher contribution during night-time due to its abovementioned semi-volatile nature. Sulfate shows unexpected daily cycles with significantly different (99.99 % confidence level) relative contributions – and absolute concentrations – during daytime (∼ 15 % of NR-PM1 mass around noon) compared to night-time (∼ 10 % around midnight, Fig. 4), although its formation was expected to occur mainly over longer timescales (i.e. days) in cloud droplets (Ervens et al., 2011). This observation could be due to (i) local production of sulfate with increased photochemical production around noon at the study site and/or (ii) diurnal changes of the atmospheric stratification in the Po Valley as described by Saarikoski et al. (2012) and Decesari et al. (2014), enhancing aged particle contribution during the middle of the day and the afternoon. Nonrefractory chloride (mostly NH4Cl, Huang et al., 2010) exhibits very low contributions independently of the hour of the day (medians below 0.5 % of NR-PM1 mass) but with a slight increase at night, which is likely due to its presumable semi-volatile nature here.

An overview of the contribution of HOA, BBOA and OOA to OA is shown in Fig. 5. On the annual average, the organic fraction is dominated by the secondary component (OOA, 66 %). Although this OOA contribution is substantial, higher proportions are generally reported at rural and urban downwind sites worldwide (90 and 82 % of OA on average, respectively; Zhang et al., 2011). This lower relative contribution of OOA is related to the higher contribution of (primary) BBOA in our study (23 % of OA on the annual average) compared to the previous ones. Considerable contributions of BBOA are explained by the specific location of the study site in the vicinity of the Alps, where biomass burning is a major contributor to OA (Belis et al., 2011; Herich et al., 2014; Lanz et al., 2010). Biomass burning emissions hence substantially affect OA levels on the annual scale here. The contribution of HOA is comparatively smaller (11 %), indicating that despite the expected large contributions of fossil fuel emissions (i.e. traffic and industrial emissions), those are not the major sources of primary OA at the study site. However, it is likely that fossil fuel emissions of volatile organic compounds (VOCs) – which are OOA precursors – contribute to our OOA levels as reported elsewhere (Gentner et al., 2012; Volkamer et al., 2006). At the study site, Gilardoni et al. (2011) previously estimated on the basis of 14C analyses that secondary organic carbon stemming from fossil emissions might represent 12 % of OC on the annual average. In other words, fossil fuel emissions could represent approximately a quarter (12 + 11= 23 %) of total OA mass when both primary and secondary OA fractions are accounted for. The analysis of the components' seasonal variations shows relatively stable HOA contributions (9–14 %), higher contributions of BBOA during cold seasons due to residential heating (up to 36 % of OA on average during winter) and higher OOA contributions during summer related to enhanced photochemical production (86 % of OA on average).

OA can be further characterized investigating specific organic fragments. m/z 44 (mainly CO2+) and 43 (mainly C2H3O+) signals give insights on the nature of OA, as the former is primarily related to acids or acid-derived species whereas the latter is mostly associated with non-acid oxygenates (Duplissy et al., 2011; Ng et al., 2011b). Daily variations of both f44 and f43 are shown in Fig. 6, along with other major organic fragments. On average, f44 is predominant with respect to f43 (15 and 7 %, respectively), which indicates that acid species dominate the OA composition with respect to non-acid oxygenates. Both fragments present different daily patterns underlying distinct mechanisms of formation. Acids' contributions are enhanced during daytime, which could be explained by photochemical processes and/or daily BLH variations as already discussed for sulfate. Non-acid oxygenates exhibit higher contributions during night-time than daytime. This pattern could be due to (i) the formation of semi-volatile non-acids during night-time by e.g. condensation (Lanz et al., 2007), (ii) their degradation during daytime by e.g. fragmentation reactions (Daumit et al., 2013) and/or (iii) their conversion into acid-related species during daytime e.g. by functionalization or oligomerization reactions (Daumit et al., 2013). It should be specified that the enhancement of f44 during daytime and the increasing of f43 during night-time only represent a small fraction of their total contributions to OA (Fig. 6), suggesting that most acid and non-acid oxygenates have been formed before reaching our sampling site, i.e. have been imported from other regions. The other major OA fragments (m/z 29, 55, 57 and 60) present (i) constant contributions for f29 due its various emission sources (HOA, BBOA, OOA; Ng et al., 2011c), (ii) the absence of lunch peak for f55 (and also for the absolute contributions of m/z 55) consistent with the presumable low influence of cooking emissions, (iii) morning and evening peaks for f57 characteristic of fossil fuel emissions and (iv) higher contributions during night-time for f60 in agreement with its biomass burning origin.

Using f43 and f44, the oxygen-to-carbon (O / C), OM-to-OC (OM / OC) and hydrogen-to-carbon (H / C) ratios and the carbon oxidation state (OSc) have been estimated for total OA based on the methodologies described by Aiken et al. (2008), Kroll et al. (2011) and Ng et al. (2011b) and applying the parameterization defined in Canagaratna et al. (2015), which can be summarized as follows: O/C=4.31f44+0.079,OM/OC=1.28O/C+1.17,H/C=1.12+6.74f43-17.77f432,OSc=2⋅O/C-H/C, with H / C (and therefore OSc) being estimated only when f44 > 0.05 and f43 > 0.04 (Canagaratna et al., 2015). Uncertainties associated with these estimates – in particular based on ACSM measurements – are discussed in Sect. S3 (see also Fig. S8). Comparisons with studies using (high-resolution time-of-flight) AMS instruments will not be reported and only variations within this dataset will be discussed (see Sect. S3). Seasonal and annual O / C, OM / OC, H / C and OSc are shown in Fig. 7. High O / C, OM / OC and OSc are found on the annual scale (medians of 0.7, 2.1 and -0.2, respectively), reflecting once more the aged, oxidized properties of organic matter at the study site, consistent with the predominance of the OOA component. Little seasonal variations are observed for the aforementioned variables, hence highlighting the high degree of oxidation of OA throughout the year (Fig. 7). The unexpectedly high degree of oxygenation of OA observed during cold seasons despite the increased contribution of primary BBOA (with OM / OC ratios of 1.4–1.6) could be explained by the contribution of secondary BBOA in our OOA factor during these cold seasons, which could be associated with the enhancement of e.g. dicarboxylic and ketocarboxylic acid contents (Kundu et al., 2010) that have extremely high OM / OC ratios (up to 3.8 and 3.1, respectively; Turpin and Lim, 2001). This assumption is supported by the higher proportion of f60 in our OOA factor (discussed in Sects. 3.2 and S2), as well as the surprisingly high OM / OC ratio observed for OOA during winter (2.5 compared to 2.2–2.4 during the other seasons). Note that Canonaco et al. (2015) also report a higher f44 in (LV-) OOA in winter compared to summer in Zurich (Switzerland). According to these authors, this could be due to enhanced aqueous-phase production of (LV-) OOA in clouds or hygroscopic aerosols in winter, which would lead to higher levels of oxygenation compared to gas-phase oxidation mechanisms typically occurring during summer.

In order to investigate the characteristics of fine aerosol pollution events, the variations of NR-PM1 chemical composition and OA factors' contributions as a function of total NR-PM1 mass are examined. This investigation is made on the annual (Fig. 8, discussion below) and seasonal scales (Fig. S9, discussion in Sect. S4). Distinct trends are observed depending on the chemical species and OA components studied. The proportion of nitrate is clearly enhanced with increasing NR-PM1 levels (from ∼ 10 to > 30 % when [NR-PM1] > 30 µg m-3), indicating that nitrate – or NOx – abatement policies should be highly effective when attempting to limit PM1 pollution events in the Upper Po Valley. Sulfate shows an opposite trend with decreasing relative contribution when NR-PM1 mass increases (e.g. < 5 % when [NR-PM1] > 50 µg m-3), likely due to the lower concentrations of sulfate during cold seasons, when the highest number of pollution events is observed. The proportion of organics is substantial (48–66 %) independently of NR-PM1 mass, justifying once again the importance of determining its sources to design adequate abatement policies. When focusing on the organic fraction, BBOA is the only OA factor exhibiting increased contributions (from ∼ 10 to > 40 %) with increased NR-PM1 mass (from < 10 to > 60 µg m-3), which points out the PM abatement potential of effective biomass burning emission reductions. HOA levels are rather constant throughout the year and therefore their proportions steadily decrease when NR-PM1 levels increase, implying that local fossil-fuel-related emissions of primary OC are not the main responsible for submicron pollution events observed at the study site on the annual scale. Although OOA always represents a major fraction of OA (41–75 % depending on the mass bin studied), its contribution steadily decreases with increasing NR-PM1 mass. This unexpected result signifies that even though aged secondary oxidized organics are the main contributor to OA on the annual average (66 %), they do not play a prominent role in fine PM acute pollution events.

Current European legislation sets daily and/or annual PM limit values depending on the size fraction addressed (Directive 2008/50/EC). Volume size distributions suggest that approximately 90 % of the PM2.5 mass concentration is borne by particles below an aerodynamic diameter of 1 µm at the study site (Putaud et al., 2014a). Therefore, measures tackling the main constituents of the submicron aerosol fraction would be efficient for complying with PM2.5 legislations. Based on the chemical characterization of NR-PM1 and SA of its organic fraction with a time resolution of 30 min over 1 year, this study provides new evidence which could orient PM abatement strategies also at similar regional background sites of the Po Valley. On the annual scale, OA and especially OOA should be of main concern given their predominance in NR-PM1 chemical composition (Fig. 3). On the seasonal scale, efforts should be directed towards the cold seasons (winter and early spring), for which the highest NR-PM1 levels are observed due to specific meteorological conditions (e.g. low BLH, low temperatures) and emission sources (e.g. biomass burning, Figs. 3 and 5). In particular, measures addressing emissions of NOx and BBOA would be the most efficient for reducing the magnitude and frequency of PM pollution events (Fig. 8).

Recommendations for PM abatement strategies are formulated here from a legislative perspective, which aims at decreasing PM levels. Although diminishing PM levels should help reducing PM impacts, the existence of a direct causal relationship can be debatable since each chemical component has a specific effect on human health (WHO, 2013), the radiative forcing (Boucher et al., 2013) or ecosystems (e.g. Carslaw et al., 2010). For instance, implementing policies aiming at mitigating nitrate concentrations – as suggested previously in this section – would likely have limited health benefits according to toxicological studies (Reiss et al., 2007; Schlesinger and Cassee, 2003) and should lead to an increased global warming (Boucher et al., 2013). However, measures reducing BBOA levels should be beneficial, since the cardiovascular effects of biomass burning particles have been widely reported in the literature (Bølling et al., 2009; Miljevic et al., 2010; Naeher et al., 2007) and could be similar to those of traffic-emitted particles (WHO, 2013, and references therein), whereas their impacts on the radiative forcing could be null (Boucher et al., 2013). Strategies aiming at reducing solely PM mass are therefore limited, and an assessment of their impacts – e.g. using integrated assessment models (Carnevale et al., 2012; Janssen et al., 2009) with appropriate parameterizations of fundamental processes – would be beneficial.