The Montseny measurement station (MSY, 41∘46′45.63′′ N, 02∘21′28.92′′ E; 720 m a.s.l.) is a regional background site in the NE of Spain, located in a regional natural park about 50 km to the NNE of the city of Barcelona (BCN) and 25 km from the Mediterranean coast (Fig. 1). This site is representative of the typical regional background conditions of the western Mediterranean Basin (WMB), characterized by severe pollution episodes affecting not only the coastal sites closest to the emission sources, but also the more elevated rural and remote areas inland (i.e., Pérez et al., 2008; Pey et al., 2010; Pandolfi et al., 2011, 2014). This station is part of the ACTRIS (www.actris.net) and GAW (www.wmo.int/gaw) networks, of the EMEP Program (http://www.emep.int/) and of the measuring network of the government of Catalonia.

The Barcelona measurement station (BCN, 41∘23′24.01′′ N, 02∘06′58.06′′ E; 68 m a.s.l.) is an urban background measurement site influenced by vehicular emissions from one of the main avenues of the city (Diagonal Avenue) located at a distance of around 300 m (cf. Fig. 1). The BCN measurement site is part of the air quality measuring network of the government of Catalonia. The metropolitan area of Barcelona (BMA), with nearly 4.5 million inhabitants, covers a strip 8 km wide between the Mediterranean Sea and the coastal mountain range. Several industrial zones, power plants, and highways are located in the area, making this region one of the most polluted in the WMB (i.e., Querol et al., 2008; Amato et al., 2009; Pandolfi et al., 2012, 2013, 2014). At BCN the location of the measuring station changed in 2009 when it was moved by around 500 m (cf. Fig. 1). The effect of this change on PM measurements performed at BCN will be discussed later.

Real-time PM concentrations were continuously measured at 1h resolution by optical particle counters (OPCs) using GRIMM spectrometers (GRIMM 180 at MSY, and GRIMM 1107, 1129, and 180 at BCN). Hourly PM concentrations were corrected by comparison with 24 h gravimetric mass measurements of PMx (Alastuey et al., 2011).

For gravimetric measurements, 24 h PMx samples were collected at both stations every 3–4 days on 150 mm quartz micro-fiber filters (Pallflex QAT and Whatman) with high-volume samplers (DIGITEL DH80 and/or MCV CAV-A/MSb at 30 m3 h-1). The mass of PM10 and PM2.5 samples collected on filters was determined using the EN 12341 and the EN14907 gravimetric procedures, respectively.

The PMF model (PMFv5.0, EPA) was applied to the collected daily PM10 speciated data for source identification and apportionment at both sites. Detailed information about the PMF model can be found in literature (Paatero and Tapper, 1994; Paatero, 1997; Paatero and Hopke, 2003; Paatero et al., 2005). The PMF model is a factor analytical tool, reducing the dimension of the input matrix to a limited number of factors (or sources), and it is based on the weighted least-squares method. Thus, most important in PMF applications is the estimation of uncertainties of the chemical species included in the input matrix. In the present study, individual uncertainties and detection limits were calculated as in Escrig et al. (2009) and Amato et al. (2009). Thus, both the analytical uncertainties and the standard deviations of species concentrations in the blank filters were considered in the uncertainties calculations. The signal-to-noise ratio (S/N) was estimated starting from the calculated uncertainties and used as a criterion (S/N > 2) for selecting the species used within the PMF model. In order to avoid any bias in the PMF results, the data matrix was uncensored (Paatero, 2004). 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 optimal number of sources was selected by inspecting the variation of the objective function Q (defined as the ratio between residuals and errors in each data value) with a varying number of sources (i.e., Paatero et al., 2002) and by studying the physical meaningfulness of the calculated factors.

The purpose of the Mann–Kendall (MK) test (Mann, 1945; Kendall, 1975; Gilbert, 1987) is to statistically assess whether there is a monotonic upward or downward trend of the variable of interest with time. A monotonic upward (downward) trend means that the variable consistently increases (decreases) through time. The Mann–Kendall test tests the null hypothesis H0 of no trend, i.e., the observations are randomly ordered in time, against the alternative hypothesis, H1, where there is an increasing or decreasing monotonic trend. The main advantage of the Mann–Kendall test is that data need not conform to any particular distribution and missing data are allowed. To estimate the slope of the trend, Sen's method was used (Salmi et al., 2002).

A program aiming at studying trends by means of multi-exponential fit of the data was developed within the Task Force on Measurements and Modelling (TFMM) by the Meteorological Synthesizing Centre – East (MSC-E; http://www.msceast.org/) group (Shatalov et al., 2015). The TFMM together with the Task Force on Emission Inventories and Projections (TFEIP), the Task Force on Integrated Assessment Modelling (TFIAM), and Task Force on Hemispheric Transport of Air Pollution (TFHTAP) provide fora for discussion and scientific exchange in support of the EMEP (European Monitoring and Evaluation Programme; http://www.emep.int/) work plan, which is a scientifically based and policy-driven program under the Convention on Long-range Transboundary Air Pollution (CLRTAP; http://www.unece.org/env/lrtap/lrtap_h1.html) to promote international co-operation and to solve transboundary air pollution problems. The TFMM was established in 2000 to evaluate measurements and modeling and to further develop working methods and tools. In this context, five EMEP centers are undertaking efforts in support of the EMEP work plan, namely the MSC-E, the Centre on Emission Inventories and Projections (CEIP; http://www.ceip.at/), the Chemical Coordinating Centre (CCC; http://www.nilu.no/projects/ccc/), the Meteorological Synthesizing Centre – West (MSC-W; http://emep.int/mscw/index_mscw.html), and the Centre for Integrated Assessment Modelling (CIAM; http://www.iiasa.ac.at/~rains/ciam.html). In 2014 the TFMM initiated an exercise dedicated to assessing the efficiency of CLRTAP air pollution mitigation strategies over the past 20 years, evaluating the benefit of its main policy instruments. Within this exercise a piece of software was made available by EMEP/MSC-E, with the aim of studying nonlinear trends. Annual, monthly, and daily resolution data can be analyzed with the help of this program. Since in this paper we will apply the program to annual averages of species concentrations and source contributions, we restrict the description of the multi-exponential approximations to this case. In particular, seasonal variations are not taken into consideration. The basic equations solved by the program for this particular case (annual averages) are reported below: Ct=a1⋅exp-tτ1+a2⋅exp⁡-tτ2+…+an⋅exp⁡-tτn+ωt, where Ct are the values of the considered time series, with t= 1, …, N, N being the length of the series (years), τn are the characteristic times of the considered exponential, an are constants, and ωt are the residue values. In the case of single exponential decay (n= 1) the characteristic time τ is the time at which the pollutant concentration is reduced to 1/e (i.e., 0.3678) times its initial value. The main difference between linear and exponential fit is that in the latter case, the trend is not gradual and constant over time. For an exponential trend the absolute (µg m-3) reduction per year decreases with time, being the highest at the beginning of the period. Conversely, for a linear fit, the absolute reduction is constant over time. For an exponential fit, the lower the characteristic time τ, the more rapidly the considered quantity vanishes. Deviations from single exponential fit can be taken into account introducing more exponential terms. In this work, for example, two exponential terms were sometimes used. In this case two characteristic times are calculated by the software. If the decrease of the considered quantity is very sharp at the beginning of the period (more than single exponential), then both τ1 and τ2 are positive. Conversely, one exponential term with negative τ takes possible increase of the quantity at the end of the period into account. Both τn and an are calculated by the program by means of the least-squares method, minimizing the residue ω, and the statistically significance of the exponential fit is provided by means of the p value. The number of exponential terms that should be included into the approximation can be evaluated using F-statistics (i.e., Smith, 2002). For example, the F-statistics for the evaluation of the statistical significance of the second term in equation (1) for n= 2 can be calculated as F=SS1-SS22⋅s, where SS1 and SS2 are sums of squares of residual component for approximations with one and two exponential terms, respectively, and s is the estimate of standard deviation of residual component. This statistic approximately follows the Fisher distribution, with 2 and N -2∘ of freedom. The second exponential is considered to be significant if F exceeds the corresponding threshold value at the chosen significance level.

The following parameters can be calculated from Eq. (1): total reduction: TR=Cbeg-CendCbeg=1-CendCbegannual reduction for yeari:Ri=ΔCiCi=1-Ci+1Ciaverage annual reduction:Rav=1-CendCbeg1N-1, where Cbeg and Cend are the first and the last points, respectively, of the exponential fit. The formula for the calculation of the average annual reduction takes into account that the ratio Ci+1/Ci is a multiplicative quantity, so that geometrical mean of ratios should be used. The relative contribution of residues (residual component: RC) is calculated as the standard deviation of the ratios between the residue values of the fit ωt (cf. Eq. 1) and the main component of the fit.

The MSC-E also proposed a statistic that measures the deviation of the obtained trend from the linear one (nonlinearity parameter: NL). A trend is defined as linear if the NL parameter is lower than 10 %, indicating a small difference between ME and MK fits (Shatalov et al., 2015). In the following, the reported trends were analyzed using the MK test for NL < 10 % and the ME test for NL > 10 %. A more detailed description of the multi-exponential approach is available in the TFMM wiki and in the MSC-E technical report 2015 (Shatalov et al., 2015).

Annual data coverage is an important factor to take into account in order to study trends of a given parameter. The gravimetric PM measurements, from which chemical speciated data are obtained, are typically performed with rather low frequency over 1 year. In our case, the annual data coverage of gravimetric measurements was around 20–30 % at both Barcelona and Montseny. In this section, we compare the trends of PM concentrations from gravimetric and real-time optical measurements (Table 1). Given that the trends of the considered PMx fractions were linear at both sites (NL < 10 %), only results from MK test were reported in Table 1. However, we will show later (Sect. 4.1) that the contributions from specific PM10 pollutant sources, mainly those related with anthropogenic activities, showed nonlinear (i.e., exponential) decreasing trends, thus mirroring the different effectiveness of the mitigation strategies depending on the source of pollutants considered.

As reported in Table 1, statistically significant decreasing trends were observed for the considered PM size fractions at BCN (-2.20 µg m-3 yr-1 with p < 0.001 for PM10 and -1.55 µg m-3 yr-1 with p < 0.01 for PM2.5 from OPC measurements), whereas at Montseny, only the PM2.5 fraction showed a slightly significant decreasing trend (-0.26 µg m-3 yr-1; p < 0.1 from OPC measurements). Total reduction (TR) ranged between 50.4 % (OPC PM10 at BCN) and 7.8 % (OPC PM10 at MSY), and the residual component (RC) was lower than 18 %, reflecting the goodness of the linear (MK) fit used. It must be noted that the higher p-values, magnitude of the trends, and TR observed at BCN compared to MSY was likely due to the change of the measuring station in 2009 in BCN (cf. Fig. 1). Based on the comparison between simultaneous PMx chemical speciated data collected at both BCN measurement sites during 1 month (not shown), we concluded that after 2009, the BCN measuring site was less affected by mineral matter and, to a lesser extent, by road traffic emissions, both being important sources of PM in Barcelona. In Fig. 1 we highlight the proximity of the BCN measuring station before 2009 to an unpaved parking area and different construction works. The effect of the change of the station in BCN in 2009 on PM10 gravimetric measurements is reported in the Supplement (Fig. S1). However, despite the change of the station, the comparison between BCN and MSY for specific chemical species and pollutant sources not linked with mineral matter and road traffic emissions was possible.

Table 1 shows that the p-values calculated using gravimetric and OPC measurements were the same despite the different annual data coverage. The differences in the magnitude of the trends were 22 and 24 % between gravimetric and OPC PM10 and PM2.5 measurements, respectively, at BCN and 24 and 21 %, respectively, at MSY. Relative differences of total reduction (TR) ranged between 24 % for gravimetric and OPC PM10 measurements at MSY and 15 % for PM10 at BCN. Thus, despite the different data coverage, the magnitude of the trends and TR calculated from OPCs and gravimetric measurements were rather similar. Other PM mass fractions (PM1-10 and PM2.5-10) and PM ratios (PM1 / PM10 and PM2.5 / PM10) at MSY showed nonstatistically significant trends.

In this study we used the period 2004–2014 for trends analysis, given that gravimetric PM2.5 measurements at BCN have been available since 2004. Conversely, at MSY PM2.5, gravimetric measurements started in 2002. Figure 2 shows the trends of PM2.5 concentrations at MSY, calculated using the MK test for the three different periods. The ME test was not used here given that the observed trends were linear (NL < 10 %). The period 2002–2010 was the period considered in the paper from Cusack et al. (2012) presenting the trends of PM2.5 gravimetric mass and chemical species at MSY. The period 2002–2014 is the largest period, with PM2.5 filter measurements available at the time of writing. The trend observed at MSY for the PM2.5 fraction during 2004–2014 confirmed what was already observed by Cusack et al. (2012) at the same station for the period 2002–2010. In Cusack et al. (2012), the MK test provided a decreasing trend of around -0.66 µg m-3 yr-1 at 0.01 significance level (TR of 35 %). During the periods 2004–2014 and 2002–2014, decreasing trends of -0.33 µg m-3 yr-1 (p < 0.1; TR of 26 %) and -0.37 µg m-3 yr-1 (p < 0.05; TR of 31 %), respectively, were observed. Thus, a statistically significant trend for PM2.5 mass at regional level can be observed, even when considering different periods, thus confirming the effectiveness of mitigation measures together with the effect of the economic crisis in Spain from 2008. However, it should be noted that the statistical significance of the trends observed for the larger periods was lower compared to Cusack et al. (2012). The difference observed in the magnitude of the trends during 2004–2014 compared to the results provided by Cusack et al. (2012) was mainly due to the increase of PM2.5 mass concentration in 2012 (cf. Fig. 2). Chemical PM2.5 speciated data revealed that this increase was partly driven by organic matter showing a mean annual concentration in 2012 higher by around 20 % compared to the 2004–2014 average.

The trends of the annual mean concentrations of chemical species at BCN and MSY are reported in Table 2 (for PM10) and Table 3 (for PM2.5). Figure 3 (for BCN) and Fig. 4 (for MSY) show the trends of chemical species in PM10. In Tables 2 and 3 and Figs. 3 and 4, only the species that have statistically significant trends were reported.

As already noted we assume that the change of the station in BCN in 2009 affected the trends of the concentrations of OC, EC, Cu, Sn, Sb, and Zn (mainly traffic tracers), Al2O3, Ca, Mg, Ti, Rb, Sr (crustal elements related with both natural and anthropogenic sources), and Fe (traffic and crustal tracer). These chemical species at BCN were removed from Tables 2 and 3 and from Fig. 3.

Other species measured in BCN were not affected by the change of the station. These are SO42-, NH4+, V, Ni (related with heavy oil combustion in the study area according to source apportionment results, cf. Par. 4), Pb, Cd, and As (related with industrial/metallurgy activities), Na and Cl (sea spray), and NO3-. Although nitrate particles in Barcelona were mainly from traffic, the concentrations of these particles were not strongly affected by the change of the station due to their secondary origin. The MSY station will be considered as reference station given that no location change occurred at this monitoring site during the study period.

Statistically significant exponential trends (p < 0.01 or 0.001) were mainly observed for the industrial tracers (Pb, Cd, As) in both PM10 and PM2.5. For these elements TR was high and around 50–80 % in PM10 and 67–81 % in PM2.5. The RCs were lower than 20 %, thus suggesting the goodness of the exponential fits used to study the trends of these species. Exponential fits were needed on average, indicating that the trends were not gradual and consistent over time and that the effectiveness of the control measures for these pollutants was stronger at the beginning of the period under study (2004–2009 approximately) compared to the end of the period (Figs. 3 and 4). This is also evident by comparing the linear MK fit (dashed black line) with the ME fit (red line) in Figs. 3 and 4. In PM10 the magnitudes of the trends ranged from -0.00222 (Pb; p < 0.001) to -3.10 × 10-5 µg m-3 yr-1 (Cd; p < 0.001) at BCN and from -0.00031 (Pb; p < 0.01) to -1.12 × 10-5 µg m-3 yr-1 (Cd; p < 0.01) at MSY. In PM2.5 the magnitudes of the trends were similar and ranged between -0.00163 µg m-3 yr-1 (Pb; p <  0.001) and -3.11 × 10-5 µg m-3 yr-1 (Cd; p < 0.001) at BCN and between -0.00049 µg m-3 yr-1 (Pb; p < 0.001) and -1.35 × 10-5 µg m-3 yr-1 (Cd; p < 0.001) at MSY. Similar magnitude of the trends for these species in both PM fractions at both sites confirmed the common origin of these elements and the impact at regional scale of industrial sources. For Pb and Cd, the characteristic time (τ) of the exponential trends was similar at both sites, whereas for As it was higher due to the slightly less intense exponential downward trend observed for As compared to Cd and Pb. Note that the PMF analysis (cf. Sect. 4) revealed that the concentrations of As were explained by multiple sources (especially at BCN), whereas the industrial/metallurgy source alone explained more than around 70 % of Pb and Cd concentrations (not shown). The implementation of the IPPC Directive in 2008 in Spain is the most probable cause for this downward trend. The decrease observed for Pb, Cd, and As may also be attributed to a decrease in the emissions from industrial production (smelters; Querol et al., 2007) at a regional scale around Barcelona.

The concentrations of V and Ni in Barcelona in both PM10 and PM2.5 fractions showed very similar exponential decreasing trends. Similar characteristic times (around 10–11 years), TR (around 59–63 %), and RC (15–17 %) in both fractions suggested the common and mainly fine origin of these two elements. At MSY, V and Ni showed linear trends likely because of the higher distance of the MSY station to the sources of V and Ni (shipping and, before 2008, energy production) compared to BCN. Note also that the NL parameter for BCN V and Ni was around 10–12 %, indicating that in this case the exponential fit did not differ very much from the linear one. Total reduction for V and Ni at MSY was around 59–64 and 42–43 %, respectively, and RCs were lower than 24 %.

Sn and Cu in PM10 at MSY showed very similar behavior by decreasing linearly with time, with TR around 36–39 % and RC around 16–20 %. Decreasing rates of -3.65 × 10-5 (p < 0.05) and -0.00014 µg m-3 yr-1 (p < 0.05) were observed in PM10 for Sn and Cu, respectively. In PM2.5, the concentrations of Sn and Cu decreased markedly compared to PM10 at the rate of -0.00084 (p < 0.001) and -0.00026 µg m-3 yr-1 (p < 0.01), respectively. This difference could be explained by possible sources of coarser Sn and Cu, which reduced the magnitude of the trends in PM10 mass fraction. Sb showed marked decreasing trends in both PM mass fractions compared to Sn and Cu, with TR around 62–70 %. The magnitude of the trend for Sb was similar in both fractions and around -3.57 ÷ -3.86 × 10-5 µg m-3 yr-1. The concentrations of Sb were better fitted with exponential curves (SE with p < 0.01 in PM10 and DE with p < 0.01 in PM2.5). The DE fit for Sb in PM2.5 had one positive and one negative characteristic time; the latter was needed to explain the slight increase in Sb concentrations at the end of the considered period. The marked decreasing trend observed for Sb compared to other traffic tracers could be explained by a progressive reduction of Sb contained in the vehicle brakes. Cr did not show a statistically significant trend in both PM fractions.

Sulfate (SO42-) and ammonium (NH4+) particles' concentrations showed very similar behavior in PM2.5 and PM10 size fractions due to their fine nature. In BCN, the magnitudes of the trends were -0.37868 (p < 0.001) and -0.11095 µg m-3 yr-1 (p < 0.001) for SO42- and NH4+, respectively, in PM10 and -0.32778 (p < 0.001) and -0.12701 µg m-3 yr-1 (p < 0.001), respectively, in PM2.5. The trends were SE with very similar characteristic times (9.64–9.81 years in PM10 and 9.69–10.53 years in PM2.5), TR (64–65 % in PM10 and 61–64 % in PM2.5), and RC (12–14 % in PM10 and 9–15 % in PM2.5). At MSY, the magnitudes of the trends of SO42- and NH4+ and their statistical significance were lower compared to BCN in both fractions. Moreover, at MSY, the trends were linear for SO42- in both fractions (as for V and Ni). These differences could be explained by the distance of MSY to direct specific sources of sulfate, such as shipping, compared to BCN, thus slightly reducing the magnitude and the statistical significance of the trend of SO42- at regional level. It is also interesting to note the similitude between the characteristic times of the exponential fits for V and Ni and SO42- in both PM fractions at BCN, suggesting the main common origin of these chemical species. Possible reasons for the observed reduction in the concentrations of ambient sulfate in and around Barcelona will be discussed later.

Fine NO3- (Table 3) showed statistically significant SE trends similar at both sites with p < 0.001, TR around 73–82 %, RC around 16–21 %, and characteristic times around 5.8–7.6 years. In PM10 the TRs were lower and around 54–64 % and the fits were linear at MSY and SE at BCN. The SE fit in PM10 at BCN provided a characteristic time around 9.8 years, which was higher compared to the τ obtained for the fine mode (7.6 years), thus indicating that fine NO3- had a more pronounced downward trend compared to PM10 NO3-.

For the mineral species (Al2O3, Ca, Fe) linear (with the exception of Al2O3 in PM2.5 which was SE) and statistically significant decreasing trends were detected at MSY. On average the TR was higher in the fine fraction, ranging from 50 for Ca to 66 % for Al2O3, compared to PM10 (6–38 % cf. Table 2). A downward decreasing trend for crustal material in PM2.5 at MSY was also reported by Cusack et al. (2012) for the period 2002–2010 and by Querol et al. (2014) for the period 2001–2012. These trends were probably driven by weather conditions associated with negative North Atlantic Oscillation (NAO) index (iNAO) that could be the cause for this slight reduction observed in crustal material. Pey et al. (2013) found a correlation between iNAO (calculated between June and September) and the contribution of Saharan dust to PM10 mass in the NE of Spain, and showed that the more negative the iNAO is, the lower the dust contribution to PM. The iNAO was unusually negative during the period 2008–2012 (http://www.cpc.ncep.noaa.gov/products/precip/CWlink/pna/norm.nao.monthly.b5001.current.ascii), thus likely contributing to the explanation of the observed trends of crustal elements. Moreover, negative NAO can favor the presence of fronts that can sweep the Iberian Peninsula from west to east, causing stronger winds and less stagnant conditions, thus favoring the dispersion of pollutants. In addition, as suggested by Cusack et al. (2012), it could also be hypothesized that some part of the crustal material measured at MSY is a product of the construction industry. The construction industry in Spain has been especially affected by the current economic recession, and crustal material produced by this industry may have contributed to the crustal load in PM2.5. For example the number of home construction works in Barcelona during 2008–2014 (from the beginning of the economic crisis; mean number of works of 1281) reduced by around 75 % compared to the period 2000–2007 (mean number of works of 5187; http://www.bcn.cat/estadistica/castella/dades/timm/construccio/index.htm). The fact that the total reduction calculated for mineral elements reported in Tables 2 and 3 was higher in PM2.5 compared to PM10 could corroborate the hypothesis of an anthropogenic contribution to the mineral matter measured at MSY.

Finally, Na concentrations showed linear decreasing trends at both sites, with the exception of PM10 Na at MSY. Other species at MSY such as OC and EC did not show statistically significant trends. Consider that the concentrations of EC at MSY are very low and around an annual mean of 0.2–0.3 µg m-3. Both anthropogenic activity and biomass burning were expected to contribute to this chemical species. Concerning OC, the lack of trend was probably due to the contribution from biogenic sources to the concentration of this species at regional level.

Figures 6 and 7 and Table 4 show the results from MK or ME tests applied to the annual averages of PM10 source contributions at BCN and MSY. As already noted we cannot study trends for traffic, road/work resuspension, and mineral source contributions at BCN because of the change of the station location in 2009. The contributions that showed statistically significant downward trends at both stations were from secondary sulfate, secondary nitrate, and V–Ni-bearing sources (p < 0.001 or p < 0.01). Moreover, statistically significant decreasing trends were observed for the industrial/traffic (p < 0.01) and mineral (p < 0.1) source contributions at MSY and for the industrial/metallurgy source contribution (p < 0.001) at BCN. These sources were mostly linked with anthropogenic activities, and the observed decreasing trends confirmed the effectiveness of pollution control measures together with the possible effect of the economic crisis in Spain from 2008. Conversely, the contributions from sources mostly linked with natural processes such as aged marine (at both BCN and MSY) and aged organic (at MSY) did not show statistically significant trends.

The trends of the secondary sulfate source contributions were DE and SE at BCN and MSY, respectively, thus indicating that the decrease over time of this source contribution was not gradual and monotonic. Overall the observed decreasing trends at both stations may be attributed to the legislation that came into force in 2007–2008 in Spain (the EC Directive on Large Combustion Plants), which resulted in the application of flue gas desulfurization (FGD) systems in a number of large facilities. Figure 8 shows the sharp decreases after 2007 observed for the national SO2 and NOx emissions mostly from power generation (MAGRAMA, 2013; Querol et al., 2014). In BCN the two characteristic times (one low and the other high, cf. Table 4) of the DE fit indicated a strong decrease of the secondary sulfate source contribution at the beginning of the period. This decrease was sharper compared to MSY where the SE fit was used. This difference was mostly due to the ban of heavy oils and petroleum coke for power generation around Barcelona from 2007. The effects of this AQ regional plan were likely more visible in BCN compared to MSY, thus explaining the two different exponential fits used. Overall, for the secondary sulfate source contributions, the TRs were rather high around 53 at MSY and 67 % at BCN, with RC ranging from 16 (BCN) to 21 % (MSY). The fact that the trend of the secondary sulfate source contribution was exponential likely suggested the attainment of a lower limit, and indicated a limited scope for further reduction of SO2 emissions in our region. In fact, it has been estimated that the maximum in EU will be a further 20 % reduction through measures in industry, residential and commercial heating, and reduced agricultural waste burning (UNECE, 2016). Conversely, in eastern European countries the scope for reduction is much greater and around 60 % (UNECE, 2016).

The trends of the secondary nitrate source contributions were SE at both stations with very similar τ (8.96–8.59 years), TR (67–69 %), and RC (13–17 %). The decrease observed for the contribution from the secondary nitrate source was related to the reduction in ambient NOx concentrations (Figs. 8 and 9). Figure 9 shows the levels of tropospheric NO2 column from 2005 to 2014 in southern Europe from NASA NO2 Ozone Monitoring Instrument (OMI) Level 3 plotted using the Giovanni online data system (Acker and Leptoukh, 2007). In Spain, a general decrease of the concentrations of columnar NO2 at regional level can be observed. Overall, the implementation of European directives affecting industrial and power generation emissions as well as the increase of the proportion of energy produced from renewable sources (cf. Fig. 10 for Spain), among others, produced a significant reduction of SO2 and NOx emissions. Around Barcelona the observed decreases were also attributed to the decrease of NOx emissions, mainly from the five power generation plants around the city. Moreover, the implementation of the regional AQ Plan for SCRT (continuously regenerating PM traps with selective catalytic reduction for NO2) and the hybridization and shift to natural gas engines of Barcelona's bus fleet may have had an influence in the observed reductions.

The decreasing trends (p < 0.01) of the V–Ni-bearing source contributions were SE and L at BCN and MSY, respectively, reflecting the trends observed at both stations for the concentrations of V and Ni (cf. Table 2). At BCN the characteristic times (τ) were very similar to the characteristic times calculated for PM10 V and Ni (cf. Table 2), which were the main tracers of this source. TRs were around 61 % at BCN and 64 % at MSY, and RCs were similar (19–25  %). The observed decrease in the V–Ni-bearing source contribution was mainly attributed to the ban of the use of heavy oils and petroleum coke for power generation in Spain from 2008.

The industrial/metallurgy source contribution at BCN decreased exponentially (SE) at the rate of -0.10 µg m-3 yr-1 (p < 0.001), reflecting the SE decreasing trends observed for the main tracers of this pollutant source (Pb, Cd, and As; cf. Table 2). The decrease of industrial emissions was mainly attributed to the implementation of the IPPC (Integrated Pollution Prevention and Control) Directive. Moreover, the observed decrease may be attributed to a decrease in the emissions from industrial production (smelters; Querol et al., 2007) at a regional scale around Barcelona. Also, the financial crisis, whose impact on industrial production and use of fuels has been evident since October 2008 also contributed to the observed trend. TR and RC for the industrial source contributions at BCN were 65 and 16 %, respectively. As for the contributions from secondary sulfate and nitrate sources, the exponential trend observed for the industrial/metallurgy source contribution suggested the attainment of a lower limit. As evidenced in Fig. 6, the contribution from this source from 2010 was quite low and rather constant.

The contribution of the industrial/traffic source at MSY showed similar magnitude of the trend (-0.11 µg m-3 yr-1 with p < 0.01) compared to the BCN industrial/metallurgy contribution, both sources being traced mostly by the same industrial tracers. However, this source at MSY was also traced by traffic tracers (i.e., Cu and Sn), which decreased linearly with time (cf. Table 2), thus likely explaining the linear trend observed for the contribution of this source at MSY. TR and RC were 56 and 13 %, respectively, similar to those calculated for the industrial source contribution at BCN.

Finally, the mineral source contribution at MSY showed linear slightly significant decreasing trend (p < 0.1) in agreement with that observed at the same station by Cusack et al. (2012). As already noted in Sect. 3.3, this negative trend could be due to both a possible decrease of the emissions of finer anthropogenic mineral species from specific sources, such as cement and concrete production and construction works, and unusual weather conditions, reducing the Saharan dust contribution to PM and resuspension of dust.

In order to further interpret the observed trends, annual data on the annual National Energy Consumption (NECo) from different energy sources (MINETUR, 2013) were also evaluated (Fig. 10). Overall, the primary energy consumption in Spain (NECo statistical data for Spain-MINETUR, 2013) increased from 2004 to 2007 and decreased from 2007, with a marked decrease in 2009. From 2009, the energy consumption indicator remained rather low and constant until 2012, when an additional decrease in 2013 and 2014 was observed. Oil consumption was fairly constant during 2004–2007, showing an important decrease during 2008–2014. This trend was probably governed by the fuel consumption for traffic road. Coal consumption remained constantly high from 2004 to 2007, whereas, as for the emissions of SO2 (Fig. 8), a sharp decrease occurred from 2007. However, in the period 2011–2014, there was an important increase of coal consumption leading to an average consumption similar to the year 2008. However, the implementation of FGD systems contributed to maintain SO2 at low concentrations, even in the coal production regions in Spain (cf. Querol et al., 2014). The hydroelectric generation was rather specular to coal consumption. For example, the increase in 2010 of hydroelectric consumption, due to high rainfall rate, mirrored the decrease in the coal consumption observed the same year. Finally, renewable energy consumption increased by 440 % from 2004 to 2014, with a gradual growth in the NECo.