The western Mediterranean Basin (WMB) is characterized by warm summers and temperate winters with irregular precipitation rates throughout the year. In winter the location of the Azores high-pressure system favours the entry of Atlantic advections that clear the atmosphere of pollutants. In summer, atmospheric dynamics coupled to local orography result in local/regional circulations with a consequent accumulation of pollutants (Millán et al., 1997). Recirculation and aging of pollutants is favoured by weak gradient atmospheric conditions, scarce precipitation and continuous exposure to solar radiation driving photochemical reactions (Rodríguez et al., 2002; Pérez et al., 2004). Additionally, large mineral dust contributions from Saharan dust events may exceed air quality standards (Escudero et al., 2007; Querol et al., 2009). The conjunction of all these processes surrounding the WMB lead to a radiative forcing among the highest in the word (Jacobson, 2001a).

PM chemical and optical measurements were performed at three sampling sites located in NE Spain (Fig. 1). The large coastal Barcelona urban area (BCN; 41∘23′ N, 02∘6′ E, 80 m a.s.l.) is one of the most populated areas in the NW Mediterranean, resulting in a very high road traffic density. Additionally, the metropolitan area is surrounded by a broad industrial area and is home to one of the major harbours in the Mediterranean Basin, with a large number of cruise ships (Pey et al., 2013). The conjunction of these emission sources contribute greatly to the air quality degradation in the area (Querol et al., 2001; Pey et al., 2008; Amato et al., 2009; Reche et al., 2011; Dall'Osto et al., 2013).

The Montseny regional background station (MSY; 41∘19′ N, 02∘21′ E, 720 m a.s.l.) is located in the Montseny natural park in a densely forested area, 50 km to the N–NE of the Barcelona urban area and 25 km from the Mediterranean coast. Despite the site being far enough from the industrialized and populated Barcelona metropolitan region, it can be affected by anthropogenic emissions transported to regional inland areas (Pérez et al., 2008).

The Montsec continental background site is a remote high altitude location (MSA; 42∘3′ N, 0∘44′ E, 1570 m a.s.l.) on the southern side of the Pre-Pyrenees at the Montsec d'Ares mountain range, located 140 km to the NW of Barcelona and 140 km to the WNW of MSY. Despite the high-altitude location and the frequent free-troposphere conditions during the cold season, the station can be slightly influenced by anthropogenic emissions during the warmer period, when it is positioned within the planetary boundary layer (PBL) (Ripoll et al., 2014).

The three sites are members of the Catalonian air quality monitoring network. Additionally, MSY and MSA are part of the ACTRIS (Aerosol, Clouds and Trace gases Research InfraStructure) and GAW (Global Atmosphere Watch) networks. Aerosol optical properties at the sites are measured following standard network protocols (WMO/GAW, 2016). Further information characterizing physical, chemical and optical properties of atmospheric aerosols detailing the prevailing atmospheric dynamics at the three stations can be found in Querol et al. (2001), Pey et al. (2009, 2010), Reche et al. (2011), Pandolfi et al. (2011, 2014a), Cusack et al. (2012) and Ealo et al. (2016).

Aerosol light scattering coefficients were measured every 5 min at three wavelengths (450, 525 and 635 nm) with a LED-based integrating nephelometer (model Aurora 3000, ECOTECH Pty, Ltd, Knoxfield, Australia). σsp measurements were collected at MSY for the period 2010–2014 (Table S1 in the Supplement) using a PM10 cut-off inlet. Measurements at MSA were carried out using a PM2.5 cut-off inlet from 2011 to March 2014 and then replaced with a PM10 cut-off inlet. σsp measurements at BCN are not available. Calibration of the two nephelometers was performed three times per year using CO2 as span gas, while zero adjusts were performed once per day using internally filtered particle-free air. The relative humidity (RH) threshold was set by using a processor-controlled automatic heater inside the Aurora 3000 nephelometer to ensure a sampling RH of less than 40 % (GAW report 226). σsp coefficients were corrected for non-ideal illumination of the light source and for truncation of the sensing volumes following the procedure described in Müller et al. (2011a).

Aerosol light absorption coefficient at 637 nm (Müller et al., 2011b) was measured at 1 min resolution with a Multi-Angle Absorption Photometer (MAAP, model 5012, Thermo) operated in the heated sampling mode and connected to a PM10 cut-off inlet. σap measurements were collected at BCN, MSY and MSA for the periods 2009–2014, 2010–2014 and 2011–2014 respectively.

Gravimetric PM10 mass concentrations were determined by standard gravimetric procedures, according to UNE-EN 12341, 1999 protocol (Alastuey et al., 2011). Samples were collected every 3 to 4 days on 150 mm quartz micro-fibre filters (Pallflex 2500 QAT-UP and Whatman QMH) using high-volume samplers (DIGITEL DH80 and/or MCV CAV-A/MSb at 30 m3 h-1) for the periods 2004–2014 at BCN and MSY, and for the period 2010–2014 at MSA.

Chemical offline filter analyses were carried out at the three sites following the procedure proposed by Querol et al. (2001). A quarter of the filter was acid digested (HNO3 : HF : HClO4). The resulting solution was analysed by inductively coupled plasma atomic emission spectroscopy (ICP-AES; IRIS Advantage TJA Solutions THERMO) for the determination of major elements (Al, Ca, Fe, K, Na, Mg, S, Ti, P) and by inductively coupled plasma mass spectrometry (ICP-MS; X Series II, THERMO) for the trace elements (Li, Ti, V, Cr, Mn, Co, Ni, Cu, Zn, As, Se, Rb, Sr, Cd, Sn, Sb, Ba, rare earths, Pb, Bi, Th, U). In order to examine the accuracy of the acid digestion, a few milligrams of the reference material NIST 1633b were added to a quarter of the blank filter. Another quarter of each filter was water extracted to determine soluble anions. The nitrate, sulfate and chloride concentrations were resolved by ion high-performance liquid chromatography (HPLC) using a WATERS ICpakTM anion column with a WATERS 432 conductivity detector and the ammonium concentrations with an ion selective electrode (MODEL 710 A+, THERMO Orion). Organic carbon (OC) and EC concentrations were determined by a thermal-optical carbon analyser (SUNSET) following the EUSAAR2 thermal protocol (Cavalli et al., 2010). Blank filters were analysed together with the samples, and concentrations were subtracted from those found in the samples in order to calculate the ambient concentrations.

The positive matrix factorization (PMF) model (PMFv5.0, EPA) was individually applied to the daily chemical speciated data collected at BCN, MSY and MSA for source identification and apportionment to PM10. Source contributions obtained for BCN and MSY can be found in Pandolfi et al. (2016), whereas sources identified at MSA are presented in this study. Detailed information describing the PMF model can be found in the literature (Paatero and Tapper, 1994; Paatero, 1997; Paatero and Hopke, 2003; Paatero et al., 2005).

The PMF model is a factor analytical tool based on the weighted least-squares method, which reduces the dimension of the input matrix (i.e. the daily chemical speciated data) to a limited number of factors (or sources). Calculation of individual uncertainties and detection limits were based on the approach by Escrig et al. (2009) and Amato et al. (2009), wherein both the analytical uncertainties and the standard deviations of species concentrations in the blank filters were considered in the uncertainty calculations. This procedure provides a criterion with which to separate the species which retain a significant signal from the ones dominated by noise, based on the signal-to-noise S/N ratio defined by Paatero and Hopke (2003). Species with S/N greater than 1 may often indicate a good signal, though this depends on how uncertainties are determined. In order to avoid any bias in the PMF results, the data matrix was uncensored; i.e. negative, zero and below detection limit values were included in the analyses.

The PMF was run in robust mode (Paatero, 1997) and rotational ambiguity was handled by means of the FPEAK parameter (Paatero et al., 2005). The final number of sources was selected based on several criteria: (i) investigating the variation of the objective function Q (defined as the ratio between residuals and errors in each data value) depending on the number of sources (i.e. Paatero et al., 2002), (ii) studying the physical meaningfulness of factor profiles and contributions and (iii) analysing the scaled residuals and the G space plots.

Previous studies based on the IMPROVE algorithm have applied the multilinear regression (MLR) method to estimate the mass scattering and extinction efficiencies (MSEs and MAEs) of chemical species (White, 1986; Vasconcelos et al., 2001; Hand and Malm, 2007). This kind of regression model between chemical species mass concentration and aerosol particle scattering or extinction coefficients assumes an externally mixed aerosol. However, the apportionment of scattering by more than one species to the total scattering depends on the assumption of the internal or external mixing state of atmospheric aerosols, as already noted previous studies (White, 1986). The assumption of internal mixing among chemical species that form a single variable in the regression equation will reduce the possible collinearity among the dependent variables of the MLR model, making the regression coefficients less sensitive to data uncertainties at the same time (Hand and Malm, 2007). As shown in the matrix correlation in Fig. S1, a very low correlation was observed between pairs of aerosol sources identified with the PMF model at the three sites considered here.

In this study, we used the PM10 source contributions (µg m-3) as dependent variables in the MLR and the measured σsp and σap coefficients (Mm-1) as independent ones. Thus, the resulting regression coefficients of the model represent the MSEs and MAEs (m2 g-1) of mixed aerosol modes, given that the sources from PMF take into account the possible internal mixing among chemical species. Moreover, the MLR method assumes that all the species contributing to σsp and σsp are included in the equation. Thus, a better model performance is achieved here given that we used the full PM10 chemical speciation in the PMF model for source identification and apportionment. Following Eqs. (1) and (2) (as example for MSY), the partial σsp and σap contribution of each source can be computed as the product between the PM10 source contributions and the corresponding MSEs/MAEs. Then, total aerosol light σsp and σap can be modelled as the sum of the scattering or absorption source contributions. σsp,PM10λ=MSESecondarysulfateλ⋅Secondarysulfate+MSESecondarynitrateλ⋅Secondarynitrate+MSEV-Niλ⋅V-Ni+MSEAgedorganicsλ⋅Agedorganics+MSEMineralλ⋅Mineral+MSEAgedmarineλ⋅Agedmarine+MSEIndustrial/trafficλ⋅Industrial/trafficσap,PM10λ=MAESecondarysulfateλ⋅Secondarysulfate+MAESecondarynitrateλ⋅Secondarynitrate+MAEV-Niλ⋅V-Ni+MAEAgedorganicsλ⋅Agedorganics+MAEMineralλ⋅Mineral+MAEAgedmarineλ⋅Agedmarine+MAEIndustrial/trafficλ⋅Industrial/traffic It should be considered that changes in the sampling conditions (i.e. RH or size cut-off) or differences in the chemical analysis methods used on sampled filters can affect the intensive particle optical properties (Delene and Ogren, 2002) and consequently the comparison among the computed MSEs and MAEs. In fact, the resulting efficiencies can be biased by the cut-off inlet, given that absorbing aerosols tend to be predominately in the submicron fraction (Andrews et al., 2011). In this study both σsp and σap were collected using a PM10 cut-off inlet, thus guaranteeing uniformity among the performed optical measurements. An exception occurs at MSA, where a PM2.5 cut-off inlet was used until March 2014 and then replaced by a PM10 inlet. Therefore, a slight overestimation of the MSEs obtained for aged marine and mineral sources at MSA might be expected when sampling was performed through the PM2.5 inlet, given that particles contained in these sources are mainly present in the coarse fraction and significantly contribute to PM1–10 mass concentration (Ripoll et al., 2015a). However, an estimation of the influence of the inlet change on the resulting MSEs and MAEs at MSA is difficult to achieve, given the relatively short σsp and σap time series available, thus preventing performing two different MLR analyses for the two fractions. Moreover, scattering RH was controlled below 40 % at MSY and MSA in order to minimize the hygroscopic growth of the particles and then prevent a significant enhancement in the scattering efficiencies. An overestimation of the scattering or absorption efficiencies can also be due to the fact that the MLR method tends to give more weight to those variables that are more accurately measured (such as sulfate), and conversely, underestimates the regression coefficients for species with larger uncertainty (i.e. organic matter) (White and Macias, 1987). In the present study, a comparison between modelled and measured coefficients was performed using quantitative statistics. With this aim, the root mean square error (RMSE) and fractional bias (FB) were computed for modelling evaluation. FB is described in Eq. (3) (Ryan et al., 2005), where σspsim is the modelled scattering coefficient and σsp is the measured value for each daily data point. FB=σspsim-σspσsp A total of 303, 379 and 503 daily data points were used in the MLR analysis for source apportionment analysis of absorption at MSA, MSY and BCN respectively, whereas 222 and 307 daily data points were considered for MSE calculation at MSA and MSY.

The Theil-Sen slope estimate (TS) (Theil, 1950; Sen, 1968) is a non-parametric test which was investigated for the monthly averages of light scattering and absorption in order to test for the occurrence of a non-null slope in the data series during the period 2004–2014 at MSY. The total and annual reduction of these optical parameters was investigated using bootstrap resampling for the monthly deseasonalized time series, reducing the possible influence of outliers on trend estimates and obtaining robust slope p values.

A multi-exponential fit aiming to study temporal trends in the multi-exponential form (Shatalov et al., 2015) was used to represent the decomposed modelled monthly temporal series in the main component, seasonal component and residual component. Additionally, this technique allowed us to estimate the non-linearity (NL) parameter for the trends. An NL of 10 % was used as threshold to define a linear trend (NL < 10 %).

Seven aerosol particle sources were identified at MSA in the PM10 fraction by performing a PMF analysis for the period 2010–2014. The chemical profiles and source contributions to the measured PM10 mass are shown in Fig. 2 and Table 1. These results will be studied together with the chemical profiles (Fig. S2) and source contributions (Table 1) previously quantified by Pandolfi et al. (2016) for BCN and MSY for the period 2004–2014. The highest PM10 average concentration was found at the BCN urban station, followed by the regional (MSY) and remote (MSA) background sites (34.0 ± 17.1, 16.7 ± 9.3 and 9.6 ± 8.2 µg m-3 respectively), consistent with the progressive distance between the three stations and important emission sources. On average, the most abundant sources contributing to PM10 mass concentration at MSA were aged organics, followed, in this order, by mineral, industrial/traffic, aged marine, secondary sulfate, V-Ni bearing and secondary nitrate. Aged organics sources was mainly traced by OC and EC with maxima in summer, pointing to a large contribution from biogenic emission sources, and accounted for 2.8 ± 2.0 µg m-3 (29 %) of the PM10 mass concentration. The internal mixing with EC suggests a contribution from combustion sources to this source. However, the aged organics source at MSA can be considered to be dominated by secondary organic aerosols (SOAs) arising from biogenic volatile organic compounds (VOCs) due to the predominance of OC in the chemical profile. Furthermore, the higher summer VOC oxidative potential occurring in the Mediterranean should be considered, which enhances SOA concentrations due to both higher insolation and tropospheric ozone concentration (Fuzzi et al., 2006). This assertion is in agreement with previous studies deployed at MSA where SOA was found to be the foremost constituent of PM1 organic aerosols (OAs), especially in summer (90 %) (Ripoll et al., 2015a). The mineral source, traced by typical crustal elements such as Al, Ca, Mg, Fe, Ti, Rb and Sr, was related to both Saharan dust events and regional/local mineral contribution and accounted for an average PM10 contribution of 2.3 ± 5.2 µg m-3 (24 %). The industrial/traffic source, primarily traced by Pb, Zn, As, Sb, Cu and Ni, contributed 1.1 ± 1.0 µg m-3 (11 %). Aged marine sources, mainly traced by Na and Cl, and in a minor proportion by Mg, SO42- and NO3-, contributed 1.1 ± 1.3 µg m-3 (11 %). Secondary sulfate, mainly traced by SO42- and NH4+, and secondary nitrate, traced by NO3- and NH4+ but also enriched in EC, contributed 0.9 ± 1.0 µg m-3 (9 %) and 0.7 ± 1.0 µg m-3 (8 %) respectively. The V-Ni bearing source, traced by V, Ni and SO42-, represented the direct emissions from heavy oil combustion, mainly shipping in the study area, and contributed 0.8 ± 1.0 µg m-3 (8 %). In contrast to BCN and MSY, the V-Ni bearing source at MSA was not enriched in EC, possibly because of the high altitude of this station and its position far from the NW Mediterranean coastline and shipping emissions.

Common sources identified at the three stations were mineral, aged marine, secondary nitrate, secondary sulfate and V-Ni. The sources identified in BCN showed similar contributions, ranging from 10 to 17 % of the total PM10 mass concentration, except for the industrial source (3 %), given that most of the secondary industrial aerosols are apportioned to other secondary sources presented in this study. At BCN, sources traced by pollutants from anthropogenic activities were mostly related to fresh emissions from the Barcelona metropolitan area (i.e. traffic and road dust resuspension), from the surrounding industrial zone (industrial) and from vessel traffic (V-Ni). However at MSY and MSA, which are representative of regional and remote backgrounds, pollutants were transported together from the urban and industrial areas of Barcelona, thus resulting in an aged aerosol mixed with local pollutants. Larger relative contributions of mineral and aged organics sources were observed at the MSA high-altitude site due to a less direct exposure to anthropogenic emissions (Fig. 3c). In agreement with previous studies (Ripoll et al., 2015b; Ealo et al., 2016), a higher relative mineral contribution was found at MSA (23 %) compared to MSY (16 %) and BCN (14 %). However, a higher absolute mineral contribution mainly originating from local sources was observed at BCN (4.6 ± 5.3 µg m-3). The aged organics source also presented a higher relative contribution at MSA (29 %) compared to MSY (23 %). However, this source was not identified at BCN, where the traffic source explained the majority of the measured OC. The aged marine source in Barcelona showed higher absolute and relative contributions (5.7 ± 5.2 µg m-3; 17 %) due to its proximity to the coast compared to MSY (1.8 ± 1.8 µg m-3; 11 %) and MSA (1.1 ± 1.3 µg m-3; 11 %). Higher relative contributions of secondary sulfate and secondary nitrate were found at MSY (24 and 8 %) compared to MSA (9 and 7 %), likely because of the longer distance from MSA to the Barcelona metropolitan area. Moreover, the free-troposphere conditions typically occurring in MSA during the colder months prevented the direct transport of aerosol particles from anthropogenic sources to the station. The V-Ni bearing source showed similar absolute contributions at MSY (0.7 ± 0.7 µg m-3; 4 %) and MSA (0.8 ± 1.0 µg m-3; 8 %) despite the longer distance from MSA to the Mediterranean coast, pointing to a possible influence of long-range transport affecting the mountain-top site. It should be noted that the current increasing shipping emissions contribute greatly to air quality degradation in coastal areas (Viana et al., 2014), but also in regional and remote environments as consequence of atmospheric transport processes.

Overall, the impact of the identified aerosol sources at the different background sites depended on the distance to important emission sources and on the aging and transport of aerosol particles to regional and remote inland areas driven by orography and meteorology, thus mostly explaining the differences in the chemical profiles of the sources identified at the three sites.

Monthly average source contributions to PM10 obtained at the three stations are shown in Fig. 3. MSY and MSA were characterized by a marked PM10 seasonal variation with higher concentrations in summer (June, July and August) and lower in winter (December, January and February), in agreement with previous studies (Pérez et al., 2008; Ripoll et al., 2014). The summer increase is related to the higher frequency of Saharan dust events, the recirculation of air masses that prevent air renovation, the resuspension processes due to the dryness of soils, the low precipitation and the formation of secondary aerosols (Rodriguez et al., 2002). The lower winter concentrations can be explained by the high frequency of Atlantic advections leading to a higher dispersion of pollutants and to higher precipitation rates compared to summer. Moreover, the reduced contribution from the PBL in winter due to frequent thermal inversions also contributed to the relatively low PM10 mass concentration observed at MSY, and especially at MSA (Pandolfi et al., 2014a). The PM10 concentration peak observed in February and March at MSY is remarkable and might be attributed to the winter regional pollution episodes typical of the WMB (Pandolfi et al., 2014b). Such scenarios are characterized by anticyclonic conditions which favour the accumulation of pollutants close to the emission sources, and the subsequent transport of pollutants towards the station with the daily increase in the PBL. Pandolfi et al. (2014b) and Pey et al. (2010) reported high nitrate concentrations during these atmospheric conditions at MSY, in agreement with the increased contributions of secondary nitrate shown in Fig. 3b during this time of the year. The relatively low PM10 concentration observed in August at BCN and MSY could be partially explained by reduced anthropogenic activities in the Barcelona metropolitan and industrial areas as a result of the holiday period in Spain. This result is supported by the minima of absolute contributions observed in August for industrial and traffic sources at BCN (0.6 ± 0.6 and 2.7 ± 1.6 µg m-3 respectively) and for the industrial/traffic source at MSY (0.9 ± 0.7 µg m-3). The higher precipitation rates observed in August compared to June–July (Perez et al., 2008) might also contribute to reducing PM10 concentrations at MSY. Conversely at MSA, the highest PM10 concentration was observed in August, probably due to the frequent Saharan dust events affecting the mountain top site, in accordance with the highest absolute contribution found for the mineral source in August (4.8 ± 4.8 µg m-3).

Higher relative contributions of aged marine (23 %), mineral (18 %), secondary sulfate (16 %) and V-Ni bearing (13 %) sources were observed on average in summer at BCN. By contrast, traffic (23 %), secondary nitrate (21 %) and industrial (4 %) sources maximized in winter (Fig. 3a). The seasonal variation of secondary sulfate and secondary nitrate can be attributed to a higher SOA contribution, the favoured formation of sulfate and the nitrate gas–aerosol partitioning, leading to the thermal instability of secondary nitrate during the warmer period, as was already observed in the area under study using offline filter sampling (Pey et al., 2009; Ripoll et al., 2015b) and online measurements (Ripoll et al., 2015a). In contrast to BCN, a higher relative contribution of secondary sources, some of them related to natural processes, was observed at MSY and MSA (3b and 3c). Increased contributions of secondary sulfate were observed in summer (29 and 8 % at MSY and MSA), whereas secondary nitrate maximized in winter (17 and 11 %). Aged organics showed the highest contribution in relative terms in winter (30 and 45 % at MSY and MSA); however the highest absolute contributions were observed in summer (4.8 ± 2.8 and 4.1 ± 1.9 µg m-3). This result is in agreement with the higher SOA formation found at MSA (Ripoll et al., 2015a) and MSY (Minguillón et al., 2015) during the warm period. The mineral source (19 and 27 % at MSY and MSA) maximized in summer, although high contributions were also observed in spring. Similarly to BCN, aged marine (14 and 13 % for MSY and MSA) and V-Ni bearing (5 and 11 %) sources showed the highest contributions in summer, whereas the industrial/traffic source maximized in winter (11 and 17 %).

Source-dependent mass scattering (at 450, 525 and 635 nm) and absorption (at 637 nm) efficiencies obtained at the different sites are shown in Table 2. The MSEs and MAEs for some of the sources reported in this study cannot be directly compared to MSEs and MAEs published in the literature for specific chemical species, given that the sources identified from PMF take into account the possible particle internal mixing. Similar MSEs were observed for secondary nitrate at MSY and MSA (8.8 ± 0.4 and 7.8 ± 0.8 m2 g-1 respectively at 525 nm). These values are in the upper range when compared to MSEs reported in the literature for the ammonium nitrate specie calculated through stoichiometry. Hand and Malm (2007) determined MSEs of 3.2 ± 1.2 m2 g-1 for dry PM2.5 ammonium nitrate, Cheng et al. (2015) obtained values of 4.3 ± 0.6 m2 g-1 under high mass loading in Shanghai, Tao et al. (2014) found MSEs ranging from 1.7 ± 0.8 in summer to 6.7 ± 1.8 m2 g-1 in winter in Chengdu (China) and Titos et al. (2012) observed a coefficient of 5 ± 2 m2 g-1 for nitrate ion in an urban area in southern Spain. MSEs for secondary sulfate were quite different between MSY and MSA (4.5 ± 0.2 and 10.7 ± 0.5 m2 g-1), probably due to differences in the source origin and the related particle size. Hand and Malm (2007) published lower values for the total mode of dry ammonium sulfate ranging between 0.8 and 2.4 m2 g-1, whereas a MSE of 3.5 ± 0.5 m2 g-1 was found by Cheng et al. (2015) in a polluted environment. Tao et al. (2014) showed MSEs of 4.4 ± 0.7 and 5.7 ± 0.2 m2 g-1 in winter and summer respectively for the PM2.5 fraction in Chengdu. MSEs for non-sea-salt (nss) sulfate ion were calculated at Finokalia and Erdemli from the slope between total scattering and nss sulfate concentration, showing values of 5.9 ± 1.8 and 5.7 ± 1.4 m2 g-1 (Vrekoussis et al., 2005). Higher MSEs were found in an urban background in the south of Spain (Titos et al., 2012) and in the Negev desert (Formenti et al., 2001) at 7 ± 1 and 7 ± 2 m2 g-1. Given that in our study sulfate concentrations were mainly explained by secondary sulfate and V-Ni sources, significant differences were also observed for the MSE of the V-Ni bearing source at MSY and MSA (8.0 ± 1.5 and 3.5 ± 0.5 m2 g-1). The V-Ni bearing source at MSY originated mainly from shipping emissions at regional (vessel traffic in the Mediterranean) and local (Barcelona harbour) scales. Conversely at MSA, located at a higher altitude, this source might also be influenced by continental transboundary transport and then internally mixed with different chemical species. In fact, as shown in Fig. 2a for MSA and in Fig. S2 for MSY, the V-Ni bearing source profile at MSA is enriched in OC, which is not observed at MSY. The aged marine source at MSA showed negative MSEs at 525 and 635 nm. This was likely due to the larger distance from the coast of MSA, thus preventing a strong signal from the aged marine source at this site, and/or due to the PM2.5 cut-off inlet used at the beginning of the measurement period which prevented the sampling of coarse particles. However, MSEs for the aged marine source at MSY (1.2 ± 0.3 m2 g-1) exhibited values within the same range as those reported by Hand and Malm (2007) for coarse-mode sea salt (1.0 m2 g-1). The MSE for the mineral source (1.3 ± 0.1 and 1.1 ± 0.1 m2 g-1) was similar at MSY and MSA. This similarity could be explained by the low reactivity of mineral dust particles, which were mostly externally mixed with other chemical species. Thus, less chemical transformation can be expected for mineral particles during the transport towards the stations. Lower MSEs were found for mineral matter by Hand and Malm (2007) (0.7 ± 0.2 m2 g-1), by Titos et al. (2012) in the urban background of Granada (0.2 ± 0.3 m2 g-1) and by Vrekoussis et al. (2005) in Erdemli (0.2 m2 g-1). Similar coefficients were obtained by Vrekuossis et al. (2005) in Finokalia (1 m2 g-1), and by Pereira et al. (2008) and Wagner et al. (2009) for mineral dust in Portugal, 1 ± 0.1 and 0.9 m2 g-1 respectively. The MSE for the aged organics source (1.4 ± 0.2 and 1.3 ± 0.3 m2 g-1) was also quite similar at MSY and MSA, probably due to similarities in the processes that govern the OA formation at both sites, which originated mainly from local/regional biogenic emissions and SOA formation (Minguillón et al., 2015; Ripoll et al., 2015a). A similar MSE (1.4 m2 g-1) was reported by Hand and Malm (2007) for the total mode of primary organic matter (POM). Higher MSEs were found by Cheng et al. (2015) during a pollution episode (4.5 ± 0.7 m2 g-1) and by Tao et al. (2014) in China (4.8 ± 0.8 and 6.5 ± 0.5 m2 g-1 in summer and winter). The industrial/traffic source showed similar MSEs at MSY (2.1 ± 0.8 m2 g-1) and MSA (2.3 ± 0.5 m2 g-1). This similarity was related to the common origin of this source at both sites (i.e. emission from the traffic and industrial activities). It is remarkable that MSEs for some of the sources identified in this work which contribute greatly to air quality degradation, such as industrial/traffic or V-Ni, are not available in the literature.

Prior studies dealing with the absorption efficiency of aerosol particles referred mainly to BC particles and to the possible effect of coating with non-absorbing material (Bond et al., 2013; Ramana et al., 2010). Other studies have reported the MAE of mineral matter (Linke et al., 2006) and OA (Lu et al., 2015; Updyke et al., 2012) due to the significant contribution of BrC to UV light absorption. However, to the author's knowledge, this is the first time that absorption efficiencies, as well as scattering efficiencies, are computed for aerosol particle sources. MAE values at 637 nm for the three sites are summarized in Table 2. The highest absorption efficiencies were observed for the traffic source identified at BCN (1.672 ± 0.050 m2 g-1) and for the industrial/traffic source at MSY (0.867 ± 0.047 m2 g-1) and MSA (0.206 ± 0.02 m2 g-1) due to their content in BC particles from fossil fuel combustion. The V-Ni bearing source, which contributed greatly to light scattering, also exhibited high MAE at BCN (0.928 ± 0.058 m2 g-1) with decreasing coefficients at MSY (0.526 ± 0.065 m2 g-1) and MSA (0.165 ± 0.017 m2 g-1).We have shown here that the V-Ni bearing source, which is progressively becoming more relevant for air quality degradation due to the increased shipping emissions in recent years (Viana et al., 2014), also has an important effect on light absorption as consequence of the internal mixing with combustion aerosols. The large MAE observed for secondary nitrate at MSA (0.364 ± 0.023 m2 g-1) was due to the fact that this source explained around 20 % of the measured EC concentration (Fig. 2a). Recently, Ripoll et al. (2015b) have shown the increased concentration of nitrate, ammonium, EC and traffic/industrial tracers at MSA under European scenarios. Such scenarios are characterized by the transport of polluted air masses at high altitude from central and eastern Europe to the MSA site. This fact may explain the internal mixing of BC particles in the chemical profile of secondary nitrate, and consequently the high MAE values found for this source at MSA. Lower MAEs for the secondary nitrate source were observed at BCN and MSY (0.28 ± 0.040 and 0.234 ± 0.028 m2 g-1) compared to MSA. Lower MAEs were observed for secondary sulfate (0.359 ± 0.035, 0.122 ± 0.010 and 0.173 ± 0.021 m2g-1) at BCN, MSY and MSA. Overall, higher absorption efficiencies were observed for the main anthropogenic sources at BCN, where fresh primary pollutants, mostly composed of darker particles, are emitted within the metropolitan, industrial and harbour areas. However, lower MAEs were found for the same pollutant sources at MSY and MSA. This result points to a decrease in the absorption efficiency towards inland areas, as a consequence of the different mixing and aging of pollutants during the transport towards the stations. Aerosol sources dominated by natural contributions, such as aged marine and mineral sources, showed the lowest MAE at MSY and MSA. The road dust resuspension source, which was partially composed of mineral matter, exhibited the lowest MAE at BCN (0.062 ± 0.084 m2 g-1). The mineral source presented MAE values of 0.09 ± 0.05, 0.005 ± 0.007 and 0.03 ± 0.003 m2 g-1 at BCN, MSY and MSA. Coefficients in the same order of magnitude at 660 nm were found for the Sahara–Sahel and Gobi deserts, ranging between 0.01 and 0.02 m2 g-1 (Alfaro et al., 2004), and for El Cairo and Morocco, 0.02 ± 0.004 and 0.06 ± 0.014 m2 g-1 (Linke et al., 2006). Aged marine also exhibited low absorption efficiencies at BCN, MSY and MSA (0.108 ± 0.021, 0.027 ± 0.018 and 0.015 ± 0.010 m2 g-1), but was higher at BCN due to a possible mixing with darker particles in the urban area. Similarly to the results observed for MSEs, aged organics showed similar MAEs at MSY and MSA (0.169 ± 0.011 and 0.140 ± 0.009 m2 g-1) due to the local/regional origin of this source with a similar composition at both sites. The absorption efficiency of this latter source was mainly explained by the EC contained within the source chemical profile, but also might be partially due to the presence of light-absorbing material detected as OC, such as BrC (Putaud et al., 2014).

Monthly source contributions to the total scattering and absorption coefficients are shown in Fig. 3. The partial σsp and σap apportioned to each source was calculated as the product between the aerosol source contribution and the corresponding MSEs or MAEs (Eqs. 1 and 2). According to the scattering efficiencies previously reported in Table 2, average scattering for the whole period was mainly dominated by secondary sulfate (35 and 33 % at MSY and MSA) and secondary nitrate (24 and 21 %) (Fig. 3g and h). The annual cycle of secondary sulfate and secondary nitrate scattering coefficients followed those of the PM10 mass concentration, with maxima in summer (46 and 35 % at MSY and MSA) and winter (42 and 29 %) respectively. The scattering contribution from aged organics accounted for 11 and 16 % of the total σsp at MSY and MSA. The V-Ni bearing source exhibited substantial contribution to σsp in summer (16 %), despite the relatively low contribution to PM10 mass concentration (5 % and 10 % at MSY and MSA). Less relevant were the scattering contributions from industrial/traffic (6 and 11 % at MSY and MSA) and mineral (7 %) sources, peaking in winter and summer respectively.

The traffic source at BCN and the industrial/traffic source at MSY clearly exerted a major influence on light absorption contributing 54 and 41 % to σap respectively, despite the relatively low PM10 contributions (16 and 10 %). Maxima contributions were observed in winter at BCN for the traffic source (65 %) and in October–January at MSY for the industrial/traffic source (46 %), whereas a lower influence of industrial/traffic was observed on average at MSA (18 %) (Fig. 3 d, e, f). Interestingly, the V-Ni bearing source also played an important role in light absorption, especially in summer as a consequence of the increased vessel traffic in the Mediterranean but also because of the more intense sea breeze circulations, transporting pollutants to inland regions. Average contributions to σap in summer were 31 % at BCN, 17 % at MSY and 16 % at MSA. Therefore, traffic, industrial/traffic and V-Ni bearing sources, which highly influenced air quality, also significantly contributed to σap, especially in those sites closer to the emission sources. Aged organics became a relevant source in the absorption process at the regional and remote background sites, contributing on average 20 and 32 % due to both its large contribution to PM10 and its relatively large MAE compared to other sources. Secondary sulfate contributed on average 10, 16 and 12 % to the total σap at BCN, MSY and MSA, whereas secondary nitrate showed increasing contributions to σap towards inland areas (8, 10 and 21 %), markedly maximizing during the colder months.

Scattering (σsp) and absorption (σap) time series were reconstructed by means of the sum of the partial scattering and absorption contributions determined for each source (Eqs. 1 and 2). Strong correlations were found between the measured and modelled extensive optical parameters at the three sites (Fig. 4). Results showed good agreement for σsp at 525 nm at MSY (R2 = 0.88) and MSA (R2 = 0.92). σap at 637 nm also exhibited a good correlation when comparing measured and predicted coefficients at BCN (R2 = 0.81), MSY (R2 = 0.80) and MSA (R2 = 0.93). Slopes were close to one in all cases and ranged between 0.96 and 0.98. These results are consistent with the good agreement obtained in the MLR model for MSE and MAE calculation. A R2 of 0.96 was obtained for all the cases ensuring the accuracy of the regression coefficients computed for each site. The root mean square error (RMSE) was calculated for the observed–modelled data sets, showing low dispersion and high accuracy in the modelled values. Scattering and absorption coefficients were well reproduced by the model, showing RMSE values of 8.76 and 6.06 Mm-1 for σsp at MSY and MSA, and values of 2.61, 0.55 and 0.23 Mm-1 for σap at BCN, MSY and MSA. The fractional bias (FB) between measured and predicted coefficients was calculated for each sampling site following Eq. (3). Results are shown in Fig. 5, where the FB is broken down by quintile from lowest to highest σsp and σap values. According to published results (Ryan et al., 2005 and references therein), a consistent overestimation was observed for all the modelled coefficients in the lower ranges of σsp and σap, showing the highest bias in the 1st quintile. Biases were substantially reduced in the median range values, whereas a minor underestimation was observed for the highest σsp and σap values, 4th and 5th quintiles, with negative FB. On average, a 3.8 and 5.6 % overprediction was obtained for the modelled σap coefficients at MSY and BCN, using 503 and 375 daily data points in the analysis. A σsp overprediction at MSY pointed to 4 % using 307 daily points. However at MSA, σsp and σap coefficients biased the observed values by 30.9 and 19.9 %, considering 220 and 303 data points in the analysis. A larger overestimation of the measured coefficients at MSA might be mainly explained by the lower number of daily chemical data used in the PMF model for the quantification of source contributions, but also because of the lower number of scattering and absorption data points available for the MLR analysis.

An independent subset of the study period was considered in order to further evaluate the PMF–MLR technique and the accuracy of the method to simulate optical properties when chemical source contributions were available. Therefore, a new PMF was performed in order to obtain the source contributions for the period 2004–2015 at MSY. With this aim, the simulation of σsp and σap coefficients for the period January–December 2015 was carried out by means of the source-specific MSEs and MAEs previously obtained in the MLR analysis for the period 2010–2014. Good agreement was found between modelled and measured σsp (R2 = 0.85) and σap (R2 = 0.76) coefficients, at 525 and 637 nm respectively, showing slopes close to one for the year 2015 at MSY (Fig. 6). This analysis confirms the confidence of the PMF–MLR technique to accurately estimate σsp and σap coefficients when chemical data are available.

As a result of the aforementioned sensitivity test, long-term time series of σsp and σap were satisfactory reconstructed for the period 2004–2014 at BCN and MSY and for the period 2011–2014 at MSA (Fig. 7), when PM10 chemical speciated data were available.

Long-term trends of σsp and σap and their relationship with the trends of PM10 source contributions were investigated for an 11-year period at MSY (2004–2014). The trend of σap at BCN was not studied due to the change in the location of the BCN sampling station in 2009 (Pandolfi et al., 2016), which affected mainly the contribution from the traffic source. The short time series available for chemical species concentration at MSA made the analysis of σsp and σap trends unfeasible at this station.

Temporal trends of the deseasonalized monthly averages for the modelled σsp and σap at MSY for the period 2004–2014 are shown in Table 3. The multi-exponential (ME) approach allowed σsp and σap time series to decompose in main, seasonal and residual components (Fig. 8). Linear trends were identified for σsp and σap, given that the non-linearity (NL) parameter was less than 10 % (Shatalov et al., 2015). Statistically significant decreasing trends were found for both σsp and σap at MSY (Table 3). σap decreased by -4.1 % yr-1 (-0.16 Mm-1 yr-1), whereas a reduction of -4.6 % yr-1 (-2.14 Mm-1 yr-1) was obtained for σsp at 635 nm. Very similar trends were observed for σsp at 450 (-4.4 % yr-1) and 525 nm (-4.5 % yr-1).

According to these results, decreasing trends were also observed for the majority of the PM10 source contributions identified at MSY for the period 2004–2014, except for aged organics and aged marine sources (Pandolfi et al., 2016). A reduction in the absorption coefficient was directly related to the significant decreasing trends found by Pandolfi et al. (2016) for strong light-absorbing sources, such as industrial/traffic (-5.09 % yr-1) and V-Ni bearing source (-5.82 % yr-1). The observed scattering decreasing trend could be mainly associated with a reduction in the contributions from those sources which scattered light more efficiently, i.e. secondary nitrate and secondary sulfate. In Pandolfi et al. (2016) these sources showed reduction rates of -6.27 and -4.82 % yr-1 respectively. A marked decline has also been observed for nitrate and sulfate particles in other European monitoring sites since 1990, as outlined in the EMEP report 1/2016 (Colette et al., 2016). Other studies have been published in recent years, clearly showing that the concentrations of PM and other air pollutants, such as SO2 and NO2, have markedly decreased during the last 15 years in many European countries (EEA, 2013; Barmpadimos et al., 2012; Cusack et al., 2012; Querol et al., 2014; among others). Querol et al. (2014) and Pandolfi et al. (2016) investigated trends of PM chemical components and aerosol sources at MSY, providing further explanation on the causes leading to the reduction of the atmospheric pollutants in the area. The financial crisis affecting Spain from 2008 contributed a reduction in the ambient PM concentrations. A decrease in secondary nitrate can be explained by the reduction of ambient NOX and NH3 concentrations (Querol et al., 2014). The decreasing trend of the secondary sulfate source may be supported by the reduction of sulfate particles, mainly attributed to the gas desulfurization at several facilities (Pandolfi et al., 2016). A decrease in secondary sulfate may be also explained by the 75 % reduction of SO2 concentration in the Barcelona harbour, supported by the regulation of sulfur content in shipping emissions in EU harbours from 2010 (Schembari et al., 2012). This regulation together with the 2007 ban around Barcelona on the use of heavy oils and petroleum coke for power generation, which contributed to a drastic decrease in V and Ni concentrations (Querol et al., 2014), were the main causes of the observed reduction of the contribution of the V-Ni bearing source. This source was enriched in sulfate and combustion particles as shown in the source chemical profile, and, as was already observed, contributed simultaneously to both σsp and σap. Thus, the abatement strategies adopted in the recent years might have caused changes in the internal mixture of particles emitted from the V-Ni bearing source, and consequently in the contribution of this source to light extinction.

Only a few studies have been published in Europe aiming to study trends of particle optical properties. Statistically significant downward trends of PM mass concentration, σsp, σap and SSA were found in the Po valley (Italy) for the period 2004–2010 (Putaud et al., 2014). A higher decreasing rate was observed for σsp (-2.8 % yr-1) compared to σap (-1.1 % yr-1), likely due to the increasing contribution of light-absorbing organic matter to light absorption during cold months in the Po Valley (Putaud et al., 2014). In the present study, smaller differences between σsp and σap were observed at MSY, accounting for the total reduction trends of -50 and -45 % respectively. This fact might be explained by the different background sites considered, whereas the Po Valley is a highly polluted area, MSY is representative of a cleaner environment where biomass burning emissions, which contribute greatly to light absorption, are considerably lower (Minguillón et al., 2015; Ealo et al., 2016).

Further research on light scattering and absorption long-term trends and their relation to changes in atmospheric composition is needed to better understand the role of aerosols on optical properties and on the climate system. Based on the published studies and the present results, further efforts focusing on the reduction of atmospheric pollutants containing BC particles (mainly emitted from fossil fuel combustion and biomass burning sources) need to be addressed. Given the toxicity of their chemical tracers, as well as their large contribution to light absorption, industrial/traffic and V-Ni bearing sources must be reduced through the implementation of win–win policies which aim to improve air quality and public health and mitigate climate warming.