The ChArMEx SOP2 (short observation period 2) field campaign took place from 15 July to 5 August 2013. The measurement site is located in Ersa at Cape Corse (42.969∘ N, 9.380∘ E) at the top of a hill (533 m a.s.l.; meters above sea level) a few kilometers from the sea in all directions (6, 4.5 and 2.5 km from the east, north and west sides, respectively; see Fig. 1). The measurement site is surrounded by widespread vegetation, such as the “maquis” shrubland typical of Mediterranean areas (Zannoni et al., 2015). The closest city, Bastia, is located approximately 30 km south of the site. It is the second largest city in Corsica (44 121 inhabitants; census 2012), which hosts the main harbor of the island with about 413 000 and 614 000 passengers in July and August 2013, respectively (CCI Territorial Bastia Haute Corse, 2013). However, the Cape Corse peninsula is characterized by a mountain range (peaking between 1000 and 1500 m a.s.l.), which acts as a natural barrier isolating the measurement site from any atmospheric flow originating from Bastia.

During the ChArMEx SOP2 field campaign, more than 80 VOCs, including non-methane hydrocarbons (NMHCs) and oxygenated (O) VOCs, were measured using complementary online and offline techniques with sampling inlets located approximately 1.5 m above the roof of a trailer in which the instruments were housed. Table 1 summarizes the VOC measurements performed during the campaign.

Sixteen protonated masses were extracted from a proton transfer reaction time-of-flight mass spectrometer (PTR-TOF-MS; KORE Technology^®, second generation), leading to the measurements of OVOCs (alcohols, such as methanol m/z 33.03, aldehydes, ketones and carboxylic acids), aromatics (sum of both C–8 and C–9 aromatics; m/z 107.09 and 121.10, respectively) and biogenic VOCs (BVOCs, such as isoprene, m/z 69.07, and the sum of monoterpenes, m/z 137.13). Ambient air was sampled through a 5 m long PFA line (perfluoroalkoxy, 1/4′′ OD) held at 50 ∘C using a constant flow rate of 1.2 L min-1 to minimize the residence time to 4 s. The PTR-TOF-MS sampling flow rate was set at 150 mL min-1 and an additional pump was used to raise the flow rate to the required 1.2 L min-1 in the sampling line. The instrument was operated at a reactor pressure and temperature of 1.33 mbar and 40 ∘C, respectively, leading to an E/N ratio of 135 Td.

An automated zero procedure was performed every hour for 10 min. Humid zero air was generated by passing ambient air through a catalytic converter (stainless steel tubing filled with Pt wool held at 350 ∘C) allowing for the same relative humidity as in the ambient air. During the campaign, the PTR-TOF-MS was calibrated every 3 days using a gas calibration unit (Ionicon^®) and various standards, including a mix of 15 VOCs (NMHCs, OVOCs and chlorinated VOCs) in a canister (Restek^®), a mix of 9 NMHCs in a second cylinder (Praxair^®) and a mix of 9 OVOCs in a cylinder (Praxair^®; see Table S1 in the Supplement). Additional calibrations were performed before and after the campaign using permeation tubes (Kin-Tek Analytical^®) for carboxylic acids and a liquid calibration unit (Ionicon^®) with a certified solution for methyl glyoxal. To account for a possible drift of the PTR-TOF-MS sensitivity during the campaign, relative calibration factors were determined for the carboxylic acids and methyl glyoxal using a specific VOC as a reference (present in the standard mixtures used to calibrate the PTR-TOF-MS during the campaign with an m/z value as close as possible for each compound; e.g., acetaldehyde, acetone and methyl ethyl ketone for formic acid, acetic acid and methyl glyoxal, respectively).

The signal of every unit mass is accumulated over 10 min and normalized by the signals of H3O+ and the first water cluster H3O+(H2O) as proposed by de Gouw and Warneke (2007). Concentrations are calculated using Eq. (1): [X]=iX_netiH3O++Xr⋅iH3O+H2O⋅150000Rf,X, where [X] represents the mixing ratio of a given VOC, iX_net is the net signal recorded for this VOC and iH3O+ and iH3O+(H2O) are the signals of H3O+ and H3O+(H2O) at m/z 19 and 37 respectively recorded at m/z 21 and 39 to avoid any saturation of the detector at m/z 19 and 37 and recalculated using the isotopic ratio between 16O and 18O. Xr is a factor introduced to account for the effect of humidity on the PTR-TOF-MS sensitivity (de Gouw and Warneke, 2007) and determined experimentally through calibrations performed at various relative humidities. Rf,X is the sensitivity determined during calibration experiments (in ncts ppt-1) and normalized to 150 000 counts s-1 of H3O+ ions. The latter is the number of counts of reagent ions observed in our PTR-TOF-MS instrument.

Forty-three C2–C12 NMHCs, including alkanes, alkenes, alkynes and aromatics, were measured using an online gas chromatograph (GC) equipped with two columns and a dual flame ionization detection (FID-FID) system (Perkin Elmer^®). This instrument has been previously described in detail by Badol et al. (2004). Air is sampled via a 5 m PFA line (1/8′′ OD) at a flow rate of 15 mL min-1. Ambient air passes through a Nafion membrane to dry it and is then pre-concentrated for 40 min onto a sorbent trap made of Carbopack B and Carbosieve SIII and held at -30 ∘C by a Peltier cooling system. The trap is then heated to 300 ∘C (40 ∘C s-1) to desorb and inject VOCs in a Perkin Elmer GC system. The chromatographic separation is performed using two capillary columns with a switching facility. This approach allows for a better separation and reduces co-elution issues (Badol et al., 2004). The first column designed for C6–C12 compounds is a CP-Sil 5 CB (50 m × 0.25 mm × 1 µm), while the second column designed for the C2–C5 compounds is a plot Al2O3/Na2SO4 (50 m × 0.32 mm × 5 µm). The separation step lasts 50 min, leading to a total time of 1 h 30 min. Finally, eluted compounds are detected using the two FID detectors. Calibrations were performed at the beginning, middle and end of the campaign using a standard mixture containing 32 compounds (NPL^®; see Table S2).

Sixteen C3–C7 OVOCs, including aldehydes, ketones, alcohols, ethers, esters and six NMHCs, including BVOCs and aromatics, were measured using an online GC-FID mass spectrometer (MS). This instrument has been described in detail by Roukos et al. (2009). Ambient air is sampled via a 5 m PFA line (1/8′′) at a flow rate of 15 mL min-1 by an air server unit (Markes International^®; Unity 1) and passes through a KI ozone scrubber. The sampled air is pre-diluted (50 % dilution) with dry zero air to keep the relative humidity below 50 %. A sample is then collected into an internal trap cooled by a Peltier system at 12.5 ∘C and consists of a 1.9 mm i.d. quartz tube filled with two different sorbents (5  mg of Carbopack B and 75 mg of Carbopack X; Supelco^®). Compounds trapped onto the sorbents are then thermally desorbed at 280 ∘C and injected into the column and analyzed by a GC (Agilent^®) equipped with an FID for quantification and a mass spectrometer (MS) to help with the identification. The thermally desorbed compounds are passed through a highly polar CP-lowox column (30 m × 0.53 mm × 10 µm; Varian^®) for separation. The sampling and analysis steps last 40 and 50 min, respectively, for a total time of 1 h 30 min. Calibrations were performed several times during the campaign using a standard mixture containing 29 compounds (Praxair^®; see Table S2).

Thirty-five C5–C16 NMHCs, including alkanes, alkenes, aromatics and BVOCs, as well as five C6–C12 n-aldehydes were collected by active sampling into sorbent cartridges using an automatic clean room sampling system (Tera Environment^®) and later analyzed by GC-FID. This technique has already been described by Detournay et al. (2011) and its setup in the field was discussed by Ait-Helal et al. (2014). Air was sampled via a 3 m PFA line (1/4′′ OD) at 200 mL min-1 and passed through an MnO2 ozone scrubber and a stainless steel particle filter (2 µm pore size diameter). VOCs are collected over 3 h in cartridges filled with Carbopack C (200 mg) and Carbopack B (200 mg), formerly conditioned with purified air at 250 ∘C over 24 h.

Finally, sixteen C1–C8 carbonyl compounds were collected offline over 3 h using the same sampling device as for solid sorbent cartridges by active sampling on dinitrophenylhydrazine (DNPH) cartridges (Waters^®). These compounds were later analyzed by high-performance liquid chromatography (HPLC) with UV detection. Air was sampled via a 3 m PFA line (1/4′′ OD) at 1.5 L min-1 and passed through a KI ozone scrubber and a stainless steel particle filter (2 µm pore size diameter). Data are available only for the first 10 days of the campaign (15–25 July) due to unresolved leakage issues for the rest of the campaign; hence contamination of the cartridges with indoor air from the trailer was suspected.

The detection limits for each species measured by all five techniques were determined as 3σ of the blank variation for PTR-TOF-MS and offline sampling methods and as 3σ of the baseline fluctuations for online GCs. The uncertainties for each species were estimated following the “Aerosols, Clouds, and Trace gases Research InfraStructure network” (ACTRIS) guidelines for uncertainty evaluation (ACTRIS Measurement Guideline VOC, 2014) taking into account precision, detection limits and systematic errors in the measurements. The range of uncertainties and detection limits for each technique are given in Table 1. Furthermore, systematic intercomparisons for compounds measured by different techniques (e.g., isoprene, monoterpenes, acetone, n-pentane and benzene) were performed to validate the database (not shown).

During the campaign, measurements of other trace gases (NO, NO2, O3, CO, CO2, CH4, H2O and SO2) were additionally performed at the same measurement site.

NO and NO2 were measured by a commercial analyzer (Cranox II; Eco Physics^®) using ozone chemiluminescence with a time resolution of 5 min. Since this technique allows for the direct measurement of NO only, NO2 was converted into NO using a photolytic converter incorporated in the analyzer.

O3 was measured using a UV absorption analyzer (TEI 49i; Thermo Environmental Instruments Inc^®) at a time resolution of 5 min. CO, CO2, CH4 and H2O were simultaneously measured by a commercial analyzer (G2401; Picarro^®) based on cavity ring-down spectroscopy (CRDS). Finally, SO2 was measured by a commercial analyzer (TEI 43i; Thermo Environmental Instruments Inc^®) using fluorescence spectroscopy at a time resolution of 5 min.

Online measurements of organic aerosols (PILS-IC, PILS-TOC, OCEC Sunset field instruments and Q-ACSM) have been available since the beginning of June 2013, but the data reported here are restricted to the ChArMEx SOP2 period (15 July to 5 August 2013) for which VOC measurements have been performed.

In addition, black carbon (BC) was continuously monitored during the same extended period using a seven-wavelength Aethalometer (model AE-31; Magee Scientific^®) at a time resolution of 15 min.

A study of back trajectories was performed to identify and classify the origin and typology of the different air masses reaching Cape Corse during the campaign and to support interpretation of the results. Back trajectories of 48 h were calculated every 6 h during the whole campaign with an ending point at the measurement site (42.969∘ N, 9.380∘ E; 600 m a.s.l.) using the online version of the HYSPLIT model (HYbrid Single-Particle Lagrangian Integrated Trajectory) developed by the National Oceanic and Atmosphere Administration (NOAA) Air Resources Laboratory (ARL; Draxler and Hess, 1998; Stein et al., 2015). This model was chosen for its easy and quick visualization facility.

A visual classification of these back trajectories was performed as a function of their origin, altitude and wind speed and segregated into five clusters (Fig. 2). A description of the five clusters is provided in Table 2. Four clusters correspond to different wind sectors defined by the origin of the air masses reaching the measurement site (west, northeast, south and northwest). These clusters are characterized by different transit times since the last potential anthropogenic contamination (i.e., since the air mass left the continental coasts). The air masses from the “marine west” cluster have spent 36 to more than 48 h above the sea, while they have spent 10–20 and 12–18 h for the “Europe northeast” and the “France northwest” clusters, respectively (Table 2). For the “Corsica south” cluster, the indicated transit time (12–24 h) considers the time spent by air masses above land (the islands of Corsica and Sardinia) before passing over the sea. These different transit times potentially indicate different atmospheric processing times for the air masses, the longest being for the “marine west” cluster.

The last cluster gathers air masses transported over short distances over 48 h and therefore during calm situations with low wind speeds (Fig. 2). The “calm low wind” cluster and the “marine west” cluster are the two most representative clusters, each representing 30 % of the air mass origin. They are followed by the “Europe northeast” cluster representing 26 %, and then by the “Corsica south” and “France northwest” clusters representing 8 and 6 % of the air mass origins, respectively.

Regarding the relatively long transit time of air masses traveling from continental source areas to the measurement site (from 10 to more than 48 h; see Sect. 2.5), an assessment of the photochemical age using field observations can be performed with specific ratios of long-lived VOCs measured at significant levels at the site. The use of graphic representations of the ratios for three different alkanes, such as ln(butane / ethane) versus ln(propane / ethane), is well suited to assessing the photochemical age of air masses that experienced long-range transport (Rudolph and Johnen, 1990; Jobson et al., 1994; Parrish et al., 2007). Considering an air parcel isolated from any new emissions or mixing with other air parcels and also considering that the main loss of alkanes is their oxidation by the OH radical, the relation of the three alkanes can be estimated as described by Eq. (2) (Jobson et al., 1994): ln⁡[butane][ethane]=kbutane-kethanekpropane-kethaneln⁡[propane][ethane]+β, where ki is the bimolecular reaction rate constant of the reaction between the species i and OH. The β parameter depends on the emission ratios of these three species and the reaction rate constants.

Since ethane is the least reactive of these compounds, the ratios will tend to decrease with increasing photochemical age. The evolution of ln(butane / ethane) as a function of ln(propane / ethane) during the ChArMEx SOP2 field campaign in Cape Corse is presented in Fig. 3. The points in Fig. 3 have been color coded as a function of the back-trajectory clusters described in the previous section.

Figure 3 reveals that the air masses of the marine west (light blue) cluster present higher photochemical ages (lower alkane ratios) relative to the air masses of the Europe northeast (purple) cluster, which is consistent with the analysis of back trajectories (Sect. 2.5). Moreover, the good linearity observed in the evolution of the ratios allows for a qualitative comparison of the photochemical age of air masses from the different wind clusters.

These ratios have been compared to ratios observed at measurement sites of different types (see Fig. S4). The ratios obtained during the campaign cover a large range of values with particularly low values for the marine west cluster, which is typical of relatively aged air masses sampled at very remote sites. It indicates that air masses can spend several days over the sea before reaching the measurement site, especially for the marine west cluster. In general, ratios representative of remote locations are observed all along the campaign, confirming the remote nature of the Cape Corse station.

It is noteworthy that the slope observed for our dataset (0.65; see Fig. 3) is significantly lower than the theoretical ones calculated for an isolated air mass experiencing selective oxidation by OH (2.50) or Cl (1.97). The lack of concordance with theoretical slopes has often been observed (e.g., Parrish et al., 1992; McKeen et al., 1996) and has been attributed to the mixing between air parcels of different histories and origins during long-range transport (Parrish et al., 2007 and references therein). A deviation from the theoretical slope could also occur if the sampled air masses were enriched in new emissions from different sources, such as ship or marine emissions, during transport.

In this study, US EPA PMF 3.0 was used to perform the factor analysis. For a detailed presentation of the PMF principle, the reader can refer to the first description made by Paatero and Tapper (1994) and to the user guide written by Hopke (2000). A specific dataset at a receptor site can be viewed as a data matrix X containing i samples and j measured chemical species. The PMF identifies the number of factors p, i.e., the number of emission sources and/or chemical processes driving the ambient concentrations of the measured species. It therefore allows for the decomposition of the matrix X into a product of two matrices: the species profile (f) of each source with dimensions p × j (representing the repartition of each measured chemical species in the factors) and the contribution (g) of each factor to each sample with dimensions i × p (representing the time evolution of each factor), allowing for the minimization of the residual error e. This is summarized in Eq. (3): Xij=∑k=1pgik×fkj+eij. The minimization of the residual sum of squares Q is performed using Eq. (4) to derive the solution for Eq. (3): Q=∑i=1n∑j=1meij2Sij2=∑i=1n∑j=1mXij-∑k=1pgik×fkjSij2, where Sij is the uncertainty matrix associated with the data matrix Xij, estimated as described in Sect. 2.2.

The PMF analysis was conducted on a dataset of 42 species, including NMHCs and OVOCs measured by the two online GCs and the PTR-TOF-MS (see Sect. S5), and 329 observations with a time of 1 h 30 min (time resolution of the GCs). Measurements taken by active sampling on sorbent and DNPH cartridges were not included in this dataset due to their low time resolution (3 h), which would have resulted in too few observations. Furthermore, compounds were not considered when missing, when more than half of the observations were below the detection limit or when associated with a low signal-to-noise ratio (< 1 in our case). Missing values and values below the detection limit in the selected dataset were replaced by the geometric mean and half of the detection limit, respectively, following the method used by Sauvage et al. (2009). To minimize the weight of these observations in the PMF results, the uncertainties in the missing values and values below the detection limit were set to 4 times the geometric mean and 5/6 of the detection limit, respectively. PMF also allows for the minimization of the species contributions with low signal-to-noise ratios (< 1.5 in our case) by declaring these species as “weak” and hence tripling their original uncertainties. Fourteen species have been declared as “weak” in this work.

Ethane, methanol and acetone are characterized by high background concentrations at the measurement site. To minimize the weight of these three species in the PMF results, their estimated background concentrations (500, 1000 and 1200 ppt for ethane, methanol and acetone, respectively) were subtracted from the measured concentrations in the data matrix X. These values were chosen arbitrarily to subtract the background concentrations of these species, thereby keeping their variability and avoiding near-zero values.

The PMF was run following the protocol proposed by Sauvage et al. (2009) and relying on several statistical indicators (unexplained part for each factor, correlation between the sum of the factor contributions and the sum of the measured concentration, the parameter Q (see above) and the mean and standard deviation of scaled residuals) to determine the optimal model parameters (number of factors, rotational parameter Fpeak) leading to the best solution. Based on this approach, we have derived a final solution with six factors for an Fpeak of -0.5.

Moreover, the homogeneity of the database built using measurements from different techniques was studied to ensure that all instruments are well represented in the solutions. This was done by ensuring that no substantial differences are observed between the scaled residuals of the different instruments. We therefore calculated the mean of the absolute values of the scaled residuals for the three instruments eijsij‾ (0.73, 0.67 and 0.75 for the PTR-TOF-MS, GC-FID-FID and GC-FID-MS, respectively). The differences observed between these parameters calculated for the three instruments are lower than 0.08. This indicates reasonable homogeneity among the instrument databases (concentrations, uncertainties) since absolute differences below 0.25 have been determined to be satisfactory to avoid overweighting the measurements of a particular instrument in PMF solutions (Crippa et al., 2013a). Therefore, no scaling procedure was performed on the database used in our PMF analysis.

Furthermore, 100 bootstrap runs were performed for the six-factor solution to estimate the stability and uncertainty of this solution. This operation consisted of performing additional PMF runs using new input data files built by randomly selecting nonoverlapping blocks of the original data matrix; the contribution of each factor was derived from these runs and then compared to the original solution. The lowest correlation coefficient between bootstrap solutions and base run solutions was 0.6. The six-factor solution appeared to be well mapped in the base run with a mapping of bootstrap factors to base run factors higher than 86 % for all factors (see Sect. S6).

The source apportionment of organic aerosol components from Q-ACSM was performed using positive matrix factorization (PMF; Paatero, 1997; Paatero and Tapper, 1994) via the ME-2 solver (Paatero, 1999). An extended Q-ACSM dataset of 2 months (from 5 June to 5 August 2013) was used here in order to obtain a wider range of atmospheric variability and improve the PMF output results. The extraction of OA data and error matrices as mass concentrations in µg m-3 over time and their preparation for PMF and ME-2 according to Ulbrich et al. (2009) was done within the ACSM software; the down-weighting procedure of mass fragments, however, was performed using the interface source finder (SoFi; Canonaco et al., 2013) version 6.1. Only m/z up to 100 were considered here since they represented nearly the whole OA mass (around 98 %) and did not interfere with ion fragments originating from naphthalene. The interface SoFI was used to control ME-2 for the PMF runs of the ACSM OA data. Unconstrained PMF runs were investigated here with one to six factors and a moderate number of seeds (10) for each factor number with no conclusive results on the consistency of the mass spectra profiles obtained for the different factors. Constrained PMF runs have been investigated for that purpose with fixed factors for HOA (hydrogen-like OA) with more conclusive results and significant improvements compared to the unconstrained PMF. The results presented here were obtained for constrained PMF using an averaged HOA profile taken from Ng et al. (2010b) and constrained with a value of 0.1. The properly constrained PMF solution was selected based on the recommendations from Canonaco et al. (2013). These include consistency of the factor profile mass spectra, consistency of times series with external tracers and a low Q/Qexp value. The criteria are presented and discussed hereafter.

In this study, we therefore applied separate factorization analysis to both VOCs and aerosol databases. Another approach consists of a factorization analysis of combined aerosol and gaseous databases (Slowik et al., 2010; Crippa et al., 2013a). Thus, an attempt to perform such PMF analysis was conducted using the gaseous database (42 VOCs) described above and full ACSM spectra as inputs; the homogeneity of the different inputs was taken into account by applying a scaling procedure as proposed by Slowik et al. (2010) and Crippa et al. (2013a). However, it did not allow for the satisfactory apportionment of aerosol measurements and led to weaker solutions than the ME-2 analysis. It was therefore decided to keep separate solutions for gas- and aerosol-phase organics.

Receptor-oriented models have been developed to identify, localize and quantify potential source areas that impact the concentrations of a variable measured at a receptor site in the form of a contribution quantity map. In this study we have used the concentration field (CF) approach developed by Seibert et al. (1994). This method consists of redistributing concentrations of a variable observed at a receptor site along the back trajectories, ending at this site inside a predefined grid (0.5∘ × 0.5∘ for this study). The calculated concentrations in each grid cell are weighted by the residence time that air parcels spent in each cell following Eq. (5): log⁡C‾ij=∑L=1MnijL×log⁡CL∑L=1MnijL, where Cij is the calculated concentration of the ij th grid cell, L is the back-trajectory index, M is the total number of back trajectories, CL is the concentration measured at the site when the back trajectory L reached it and nijL is the number of points of the back trajectory L that fall in the ijth grid cell. The latter is representative of the time spent by the back trajectories in the ijth grid cell since a constant time step of 1 h is used between each point of a back trajectory.

The 3-day back trajectories (selected to account for distant potential source areas of species with long lifetimes) used in the CF analysis were calculated by the British Atmospheric Data Centre (BADC) model every hour. This model uses the wind fields calculated by the European Centre for Medium-Range Weather Forecasts (ECMWF) to determine the trajectories of air masses. This model was selected here instead of HYSPLIT for convenience, since the format of the output files matches that needed for our CF model. Comparisons of randomly selected back trajectories in each identified cluster (see Sect. 2.5) calculated by both models (BADC and HYSPLIT) have revealed satisfactory agreement in terms of origin and areas overflown. The BADC back trajectories were interrupted when the altitude of the air mass exceeded 1500 m a.s.l. to get rid of the important dilution affecting air masses in the free troposphere (the boundary layer height has been arbitrarily set here to 1500 m a.s.l. for all trajectories). Furthermore, the grid cells containing fewer than five trajectory points were not considered for robustness purposes.

To take into account the uncertainties associated with the back trajectories, a smoothing of concentrations was applied to all the grid cell values as recommended by Charron et al. (2000) and using Eq. (6): Cij-l=∑p=18Cp+Cij9, where Cij-l is the calculated concentration of the ijth grid cell after smoothing, Cij is the calculated concentration of the grid cell before smoothing and Cp (1 < p < 8) is the concentration before smoothing of the eight neighbor grid cells.

Source-receptor models, such as PMF, usually aim at identifying and quantifying the contributions of sources of pollutants impacting a measurement site. In our case, the remote location of the site combined with the reactivity of the selected species does not allow for the proper identification and quantification of primary sources. Our main objective here is the identification of covariation factors of species that could be representative of aged or fresh primary emission and also of photochemical processes occurring during long-range transport or occurring locally. For this purpose, PMF was applied to a large dataset (42 different species), including primary VOCs from anthropogenic or biogenic origins, and also secondary products measured by three different techniques (PTR-TOF-MS, GC-FID-FID and GC-FID-MS; see Sect. 2.2).

Figure 6 shows the time series of the six factors obtained by the PMF analysis. Figure 7 shows the contributions of each factor to the species selected as inputs for the PMF model (in %) and the absolute averaged contribution of each species to the six factors determined by the PMF analysis (in ppt). Finally, Fig. 8 presents the maps of simulated contributions (in ppt) using the CF model for four of the six PMF factors. The relative contributions of the different PMF factors to the sum of species used as inputs are presented in Fig. S8.

Based on the two available months of ACSM data, a three-factor solution was selected here, corresponding to a minimum of the quality parameter Q/Qexp. The mass spectra reported in Fig. 10 show a typical HOA (hydrogen-like OA) profile for the first factor with n-alkanes, branched alkanes, cycloalkanes and aromatics, leading to high signals in the ion series CnH2n+1+ and CnH2n-1+ (m/z 27, 29, 41, 43, 55, 57, 69, 71) that are typical for fossil fuel combustion (Canagaratna et al., 2004; Chirico et al., 2010). We have also used the terms “SV-OOA” (semi-volatile oxygenated organic aerosol) and “LV-OOA” (low volatility OOA) as introduced by Jimenez et al. (2009) to describe the two remaining factors. In these two factors, m/z 43 and 44 are the most prominent peaks, which is consistent with OOA (oxygenated OA) spectra and the m/z 44 to 43 ratio that increases with aging (Ng et al., 2010a). The signal at m/z 43 is dominant for the second factor and mainly comes from the fragmentation of either hydrocarbon chains to form C3H7+ or carbonyls to form C2H3O+; therefore this factor appears to be less oxidized and was consequently named SV-OOA.

The consistency of the different OA factors was further checked with the external tracers in Fig. 11; HOA with BC (fossil fuel tracer), SV-OOA with WSOC and LV-OOA with oxalate. The good agreement of SV-OOA with WSOC is consistent with freshly formed SOA being semi-volatile and water soluble as reported, for instance, by Hennigan et al. (2008a) who observed strong similarities between semi-volatile NH4NO3 and (PILS-TOC-based) WSOC. The good agreement between LV-OOA with oxalate is consistent with the fact that both are mostly composed of carboxylic acid COO chains and the use of oxalate as a proxy for highly oxidized OA, as stated before. Also note that good correlation is obtained between the averaged OOA mass spectra taken from Ng et al. (2010b) and our two factors with correlation coefficients (r2) of 0.96 and 0.81 for our SV-OOA and LV-OOA factors, respectively.

The different OA factors obtained here are mainly of continental origin, and therefore their temporal variability is mostly related to the amount and frequency of continental air masses reaching the sampling site. Nevertheless, the diurnal variation in SV-OOA and LV-OOA (Fig. S11) suggests that local photochemical processes have also occurred, with local formation of fresh SV-OOA in the morning followed by rapid oxidation, which could explain the enhancement of LV-OAA in the afternoon.

Average mass concentrations are 0.13, 1.59 and 1.92 µg m-3 for the three determined factors HOA, SV-OOA and LV-OOA, respectively, for a total average OA concentration of 3.63 µg m-3 and a contribution of OA to NR-PM1 of 52 %. As a result, secondary OA represents about 96 % of OA with aged (LV-) OOA contributing approximately 55 % of this secondary OA fraction. In recent years, increasing background OA observations have become available in the Mediterranean, mostly at coastal sites located in the northern part of the basin (Spain, France, Italy and Greece). For instance, long-term (13-month) ACSM measurements were performed at a regional background site in the western Mediterranean (Spain) located in Montseny Natural Park 50 km north-northeast of Barcelona, approximately 500 km west of Cape Corse. Reported observations are similar to ours with an OA contribution to NR-PM1 of ca. 60 % (Minguillón et al., 2015), three major OA sources (HOA, SV-OOA and LV-OOA) during summer with a very prominent secondary fraction (85 % of OA) and OA profiles very similar to those obtained here.

First, the gas-phase “oxygenated factor” (factor 4) is correlated with the organic fraction of the aerosol measured by ACSM (R2 = 0.58, n = 498; see Sect. S9). This fair correlation likely highlights the close link between gaseous oxidation products observed at the site and measured organic aerosol (OA) since they stem from similar processes. During the campaign, very low levels of primary organic aerosols were observed (HOA determined by ACSM measurements below 0.4 µg m-3; see Fig. 11a, top panel). Thus, this correlation is most likely due to the secondary fraction of OA, representing 96 % of OA (see Sect. 4.3), which can come from the oxidation of both biogenic and anthropogenic gaseous precursors; this explains the similar behavior of factor 4.

Figure 12 shows stacked time series of the different fractions (inorganic and organic) of aerosol measured by ACSM (top panel) and stacked time series of contributions of PMF factors (middle panel) for the VOCs (see Sect. 4.2). This figure aims to draw a parallel between aerosol- and gas-phase compositions to highlight the link between the two phases.

From these graphs and from the back-trajectory clusters (also shown in Fig. 12), it is possible to distinguish two periods during which processed anthropogenic continental air masses reached the site (between 19 and 24 July and between 30 July and 3 August 2013). The first period is characterized by high contributions of anthropogenic and oxygenated gas PMF factors (middle panel of Fig. 12) and an aerosol with inorganic (ammonium sulfate) and organic fractions in approximately similar proportions (top panel of Fig. 12). This period also corresponds to the highest values of ln(propane / ethane) of -1.4 on average and up to -0.8 (see bottom panel of Fig. 12); hence the less-aged air masses coinciding with the Europe northeast sector. The evolution of aerosol components and VOC factors during this period is also similar to that observed for the calm low wind conditions at the beginning of the campaign. These similarities could be related to the recirculation of air masses already observed in the western Mediterranean Basin, causing the formation of reservoir layers at high altitudes described in several studies (Pey et al., 2009; Minguillón et al., 2015; Ripoll et al., 2015).

The second period of long-range transported anthropogenic continental emissions is characterized by less intense anthropogenic gas-phase PMF factors, especially for the long-lived anthropogenic factor, and a clear predominance of the organic fraction for aerosols. Aerosol mass concentrations are also lower by approximately 50 % compared to the first period. During both periods, a non-negligible biogenic influence is also observed from primary and secondary biogenic PMF VOC factors. This is even more pronounced for the second “anthropogenic” period. During these periods, it is therefore likely that oxygenated VOCs and OOAs have both biogenic and anthropogenic origins in variable proportions.

A period of intense biogenic influence without significant long-range transport of anthropogenic continental emissions can also be distinguished (between 26 and 28 July) with elevated contributions of the primary and secondary biogenic gas-phase PMF factors (Fig. 12). The oxygenated gas-phase PMF factor also rose during this period, and the aerosol composition is dominated by OA with low levels of inorganic aerosols. The inorganic fraction of aerosols decreases to reach less than 10 % of the aerosol composition on 27 July. This strong decrease occurred at the same time as a change in air mass origin from marine west to Corsica south. This is consistent with the lack of anthropogenic influence during this period, confirmed by lower ln(propane / ethane) of -1.8 on average up to -2.3 (see bottom panel of Fig. 12). It is therefore likely that the oxygenated VOCs and the organic fraction of aerosols during these days are mainly influenced by biogenic sources.

Finally, very low contributions of HOA were observed during the whole campaign from the PMF analysis of ACSM measurements (typically below 0.3  µg m-3). This illustrates the weak influence of freshly emitted primary anthropogenic sources of OA at the site. This is also confirmed by low levels of black carbon (BC < 0.9 µg m-3 for the whole campaign; see Fig. 11a).

An analysis of the isotopic ratio of 14C in aerosol sampled at Cape Corse reveals that the organic fraction of the aerosol measured during the ChArMEx SOP2 field campaign mainly came from biogenic sources and the oxidation of biogenic VOCs with a measured nonfossil OC of 2.42 ± 0.86 µgC m-3 on average (76 ± 3 % of OC on average). The secondary and primary anthropogenic sources to OC represented by measured fossil OC was 0.44 ± 0.22 µgC m-3 on average with a contribution to OC of 14 ± 3 % for OC on average. Elementary carbon contributed only 10 % of total carbon during the campaign with an average measured biomass EC and fossil EC of 0.16 ± 0.06 and 0.17 ± 0.06 µgC m-3, respectively. The results from this analysis will be presented in more detail in a forthcoming paper (Pey et al., 2017).

Given the good correlation observed between OA and the gas-phase oxygenated factor (R2 = 0.58, n = 498), a common origin can be attributed to both OA and OVOCs observed at Cape Corse. A predominance of secondary biogenic origin during the whole campaign is likely for OVOCs, such as acetone, methanol and carboxylic acids, which composed the oxygenated PMF factor. As stated previously, this is also consistent with the large fraction of WSOC in OA, the fraction of which usually refers to biogenic SOA. A less important but still significant secondary anthropogenic origin is also likely for OVOCs.