PM10 was sampled at two sites: at the station for climate Observations, maintained by ENEA (the Italian Agency for New Technologies, Energy, and Sustainable Economic Development) on the island of Lampedusa (35.5∘ N, 12.6∘ E), and at the Italian CNR (National Research Council) Research Centre at Capo Granitola (36.6∘ N, 12.6∘ E).

Lampedusa is a small island in the Central Mediterranean sea, more than 100 km far from the nearest Tunisian coast. At the station for climate Observations, which is located on a  plateau 45  m a.s.l on the northeastern coast of Lampedusa, continuous observations of aerosol properties (di Sarra et al., 2011, 2015; Becagli et al., 2013; Marconi et al., 2014; Calzolai et al., 2015; Sellitto et al., 2017), aerosol radiative effects (e.g. Casasanta et al., 2011; di Sarra et al., 2011; Meloni et al., 2015) and other climatic parameters are carried out. Figure 1 shows the map of the central Mediterranean with the measurement stations.

PM10 is routinely sampled on a daily basis at LMP (Becagli et al., 2013; Marconi et al., 2014; Calzolai et al., 2015) by using a low-volume dual-channel sequential sampler (HYDRA FAI Instruments) equipped with two PM10 sampling heads, operating in accord with UNI EN12341. For the intensive ChArMEx campaign, samples were collected at 12 h resolution (08:00–20:00 and 20:00–08:00 LT–local time) from 1 June to 3 August 2013.

The two channels operated in parallel and were loaded with different types of filters: the first one with 47 mm diameter, 2 µm-nominal porosity Teflon filters, and the second one with 47 mm pre-fired, 2 µm-nominal porosity quartz filters. Ion chromatographic analysis of soluble ions, atomic emission spectroscopy for soluble metals and proton-induced X-ray emission (PIXE) for the total (soluble + insoluble) elemental composition were carried out on the Teflon filters. Elemental carbon (EC) and organic carbon (OC) were measured by analysing the quartz filters.

The sampling site at CGR is located at Torretta Granitola, a Research Center of the Italian National Research Council, in southwestern Sicily (12 km from Mazara del Vallo). The sampler was installed on the roof of one of the research centre buildings at about 20 m a.s.l., directly on the coastline, facing the Straight of Sicily.

At CGR PM10 samples were collected at 12 h resolution (08:00–20:00 and 20:00–08:00 LT) with a TECORA Skypost sequential sampler on 47 mm pre-fired 2 µm-nominal porosity quartz filters, which were used to determine ions, metals, EC and OC on different fractions of the filter. Due to technical problems, some daytime (08:00–20:00 LT) samplings were lost at CGR.

The PM10 mass was determined by weighting the filters before and after sampling with an analytical balance in controlled conditions of temperature (20 ± 1 ∘C) and relative humidity (50 ± 5 %). The estimated error on PM10 mass is around 1 % at 30 µg m-3 in the applied sampling conditions.

A quarter of each Teflon filter from LMP and a 1.5 cm2 punch of the quartz filter from CGR were analysed by Ion Chromatography (IC) in the analytical conditions described in Marconi et al., (2014). The estimated uncertainty for IC measurements is 5 % for all the considered ions.

Blank values were negligible with respect to the concentration in the samples for Teflon filters. Blank values for quartz filters were negligible for most of the analysed species. For some species characterized by high blank values, always lower than the 25th percentile value, they were subtracted from the measured concentrations.

Another quarter of the Teflon filter from LMP and another 1.5 cm2 punch of the quartz filter from CGR were extracted in an ultrasonic bath for 15 min with MilliQ water acidified at pH 1.5–2 with ultra pure HNO3 obtained by sub-boiling distillation. This extract was used for the metal's soluble part determination by means of an Inductively Coupled Plasma Atomic Emission Spectrometer (ICP-AES, Varian 720-ES) equipped with an ultrasonic nebulizer (U5000 AT+, Cetac Technologies Inc.). The pH chosen value is the lowest found in rainwater (Li and Aneja, 1992) and leads to the determination of the metals fraction available to biological organisms and, for some metals (e.g. V and Ni), related to the anthropic source (Becagli et al., 2012).

The remaining half Teflon filter from Lampedusa, another punch of the quartz filter from CGR, was used for the determination of metals by ICP-AES through the solubilisation procedure reported in the EU EN14902 (2005) rule, by concentrating sub-boiling distilled HNO3 and 30 % ultra pure H2O2 in a microwave oven at 220 ∘C for 25 min (P = 55 bar). This solubilisation procedure is not able to completely dissolve the silicate species. However, this procedure allows to the recovery at least 70 % of the same elements measured by a proton induced X-ray emission technique also for elements with dominant crustal source (unpublished data) due to the low crustal aerosol load in these sampling periods (e.g. Mailler et al., 2016). La and Ce presented very low concentrations in the collected aerosol samples. Thus, particular attention was devoted to the minimization of the La and Ce detection limit. In the used sampling and analytical conditions of LMP samples the detection limits for La and Ce are 0.02 and 0.08 ng m-3, respectively, and are about 4 times higher for the CGR samples, due to the smaller filter portion used for the analysis. The OC and EC measurements were carried out on a 1.5 cm2 punch of the quartz filters from Lampedusa and Capo Granitola by means of a Sunset thermo-optical transmittance analyser, following the NIOSH protocol (Wu et al., 2016).

The overall aerosol sampling and analytical strategy for the two sites are reported in Table 1.

Numerical simulations with a non-hydrostatic mesoscale atmospheric model were used to characterize the meteorological conditions in the Straight of Sicily during the campaign and to support the interpretation of the experimental results. The Weather Research and Forecasting (WRF) model (Skamarock et al., 2008) outputs, provided by the Department of Physics of the University of Genoa, Italy, were used, covering the entire Mediterranean with a grid spacing of 10 km and hourly temporal resolution. Initial and boundary conditions to drive WRF simulations were obtained from the Global Forecast System operational global model (Environmental Modeling Center, 2003) outputs (0.5 × 0.5 squared degrees). Some recent applications of the modelling chain are described in Mentaschi et al., (2015) and Cassola et al., (2016), where full details on the model configuration can also be found.

In particular, the WRF 3-D hourly meteorological fields were used to calculate backward trajectories with the NOAA HYbrid Single-Particle Lagrangian Integrated Trajectory Model (HYSPLIT; Stein et al., 2015). The trajectories were used to assess the origin of the air masses impacting the monitoring sites and to support the source attribution suggested by the analysis of specific markers (see Sect. 3.2.2). The use of a high-resolution regional atmospheric model for trajectory calculations allows for a better representation of boundary layer properties and mesoscale phenomena such as land and sea breezes, which can have a relevant impact especially in complex topography coastal sites like CGR.

Also, the high temporal resolution of meteorological data (hourly instead of three- or six-hourly products typically available from global models) permits a better description of diurnal cycles and a more accurate trajectory computation without time interpolation between subsequent atmospheric fields (Solomos et al., 2015).

Specifically, 48 h-long back trajectories arriving at LMP and CGR were computed from a reference height of 10 m above the ground level, starting every six hours for the whole period of the campaign, from 10 June to 31 July 2013.

Position and main characteristics of the ships travelling in the central Mediterranean were derived from the MarineTraffic database (http://www.marinetraffic.com/), which provides data with a high temporal resolution (about 3–5 min) by means of the Automatic Identification System (AIS).

Three classes of ships defined by the AIS classification were considered: all the ships, the merchant ships (i.e. cargo and tanker) and the fishing vessels. Merchant and fishing vessels are the most frequent ships in the Straight of Sicily; merchant ships are expected to produce the highest impact due to their higher emissions (http://ec.europa.eu/environment/archives/air/pdf/chapter2_ship_emissions.pdf).

The sea salt aerosol (SSA) component of PM10 was estimated as the sum of Na+, Mg2+, Ca2+, K+, sulfate and chloride sea salt (ss) fractions. Details on the calculation of sea salt Na+ and Ca2+, and non-sea salt (nss) fractions are reported in Marconi et al., (2014). The Mg2+, Ca2+, K+ and sulphate sea salt fractions were calculated from sea salt Na+ (ssNa+) by using the ratio of each component to Na+ in bulk sea water: Mg2+ / Na+ = 0.129, Ca2+ / Na+ = 0.038, K+ / Na+ = 0.036, SO42- / Na+ = 0.253 (Bowen, 1979). Chloride undergoes depletion processes during the aging of sea spray, mainly due to exchange reactions with anthropogenic H2SO4 and HNO3, leading to the re-emission of gaseous HCl in the atmosphere. Thus, for chloride we use the measured chloride concentration instead of the one calculated from ssNa+. Thus, SSA=1.46⋅ssNa++Cl-. The crustal component is calculated from Al, which represents 8.2 % of the upper continental crust, UCC (Henderson and Henderson, 2009). A previous study using an extensive data set at Lampedusa showed that the crustal content determined from the total Al was in very good agreement with calculations made from the sum of the metal oxides (Marconi et al., 2014). However, in this study we use measurements of the soluble Al concentration obtained by ICP-AES on the solution obtained with H2O2 and HNO3 in a microwave oven instead of the total Al content. Therefore, in this work we underestimate the crustal contribution by about 30 % (unpublished results). However, it must be emphasized that the crustal aerosol contribution has been very low throughout the measurement campaign.

Figure 2 shows the time series of the main PM10 components at LMP and CGR. An intense mistral event occurred from 22 June to 1 July. Mistral events are characterized by strong winds from the northwesterly sector and often by subsiding air masses originating from the free troposphere (Jiang et al., 2003). Thus, elevated values of SSA and low concentrations of other compounds are generally found during mistral at Lampedusa.

Average concentrations of PM10 and of the different aerosol components for the whole measurement campaign and for the non-mistral conditions are reported in Table 2. The averages were calculated over a homogeneous data set, i.e. when measurements are available at both sites.

The largest PM10 values were associated with elevated SSA during the mistral event at both sites. PM10 is about two times larger at Capo Granitola than at Lampedusa. The PM10 measured during the campaign at Lampedusa was significantly lower than its long-term average (31.5 µg m-3; Marconi et al., 2014). No Saharan dust transport events occurred at low altitude in this period (e.g. Mailler et al., 2016), and the crustal aerosol contribution remained very low and almost constant at both sites (average < 1 µg m-3 at LMP and around 3 µg m-3 at CGR, corresponding to 4.6 and 8.2 % of the PM10 at LMP and CGR respectively).

SSA accounted for about 26 and 24 % of PM10 at LMP and CGR, respectively. The SSA contribution was about 14 % of PM10 at LMP and 8 % at CGR during the periods not influenced by the mistral. Non-sea salt SO42- was the most abundant among the secondary inorganic species.

Nitrate concentrations, although relatively high at both sites, are in agreement with the long term measurements performed at Lampedusa (e.g. Calzolai et al., 2015) and with data from other remote sites in the western (Mallorca; e.g. Simo et al., 1991) and eastern Mediterranean (Finokalia; e.g. Mihalopoulos et al., 1997).

Organic aerosol (OA) was the most abundant component at CGR, where its mean concentration was > 9 µg m-3 and represented 35 % of PM10 in the days not characterized by the mistral.

Elemental carbon and organic carbon show higher values at CGR than LMP. At CGR, moderate and elevated values of OC and EC appear correlated (R2 = 0.60; n = 59; Fig. 3b), suggesting a strong influence of carbon species from primary sources, characterized by the simultaneous EC and OC emission. The influence from primary sources is apparent at EC > 0.4 µg m-3. At LMP, on the contrary, OC was not correlated with EC (Fig. 3a), indicating a strong impact of OC secondary and/or natural sources. This confirms that Lampedusa may be considered a background site in the central Mediterranean (see e.g. Artuso et al., 2009; Henne et al., 2010), and the observations there may be taken as representative for a relatively wide open sea region.

Thus, we used a conversion factor of 1.8 (typical for urban background sites, Turpin and Lim, 2001) at CGR, and of 2.1 (typical for remote sites characterized by the high impact of secondary sources, Turpin and Lim, 2001) at LMP to estimate the total organic aerosol amount from the OC measured values. Once we estimated OA with this method, the sum of the various species accounted for more than 85 % of the measured mass at both sites. The unreconstructed mass could be due to an underestimation of OA from OC, or to the presence of bound water not removed by the desiccation procedure at 50 % relative humidity (Tsyro, 2005; Canepari et al., 2013).

SO2 is one of the main species emitted in the ship plume in the gas phase (Agrawal et al., 2008a, b). SO2 is produced through oxidation of the S contained as impurity in heavy fuel oil and is an aerosol precursor.

A previous study based on five years of data from Lampedusa (Becagli et al., 2012) has shown that the non-sea salt sulfate behaviour is not directly correlated with V and Ni because several other SO42- sources (anthropogenic, marine biogenic, crustal, volcanic) contribute to the non-sea salt sulfate in the Central Mediterranean Sea.

The same study suggested a lower limit of about 200 for the nssSO42- / V ratio for particles originating from heavy oil combustion at Lampedusa.

Figure 8a and 8b shows nssSO42- / V vs. V at LMP and CGR. At both sites, nssSO42- / V decreases with increasing V and reaches a lower limit at elevated values of V (> 15 ng m-3). The analysis on REE, trajectories and ship traffic has shown that all samples with V > 15 ng m-3 are strongly influenced by ships and we assume that ship emission is the dominant source of the sampled particles for these cases. This implies that in these cases virtually all sulfate originated from the ship source and the observed lower limit for nssSO42- / V can be considered the lower limit for the sulfate to V ratio in the ship plume. Thus, to derive a lower limit for this ratio we calculate the mean and standard deviation of nssSO42- / V for V > 15 ng m-3. The mean ratio and the mean ratio minus one standard deviation are shown in Fig. 8.

The nssSO42- to V ratio may still be decreasing for V by around 15 ng m-3, and we used a limit value equal to the average minus one standard deviation (solid red lines in Fig. 8) to estimate the minimum expected contribution from ships to the total sulfate amount.

The calculated lower limit of the sulfate to V ratio at LMP is 207, in agreement with the values of 200 estimated by Becagli et al., (2012). The nssSO42- / V limit value at CGR, 323, is larger than at LMP. This difference may be due to the contribution of other sulfate sources, which may contribute to the nssSO42-, even at high V concentration, and to the smaller distance from the ship source with respect to LMP. This result highlights the importance of remote sites like LMP to obtain information on the open Mediterranean.

NOx are among the main compounds emitted in the gas phase acting as aerosol precursors. The photochemistry of NOx leading to NO3- formation in the particulate phase is complex, especially in summer, due to the presence of high amounts of OH radical (see, for example, Chen et al., 2005), and the NOx contribution to the particulate phase is not easy to be quantified.

Here we try to use the same approach used for sulfate for the determination of a lower limit for the NO3- / V ratio in the ship plume.

Figure 8c and d show the NO3- / V ratio vs. V at the two sites. Similarly to sulfate, the average value of NO3- / V for V > 15 ng m-3 is larger at CGR than at LMP. However, the standard deviation at CGR is significantly larger at CGR. The NOx concentration in the ship plume close to the source is larger than that of SO2 and is strongly dependent on the engine operating conditions (Agrawal et al., 2008b). The NOx lifetime is extremely low (1.8 h during the daytime and 6.5 h during the nighttime, Chen et al., 2005). However, the NO3- / V limit ratio values is low compared to the limit ratio for SO42-. It has to be considered that NO3- takes part in other photochemical atmospheric reactions that lead to its removal. In addition, the presence of HNO3 in the gas phase, not neutralized by NH3 or by sea salt, could explain the low NO3- / nssSO42- ratio in the aerosol. Indeed, the NO3- concentration measured at LMP and CGR is four to six times lower than that of nssSO42- (Table 1). Low amounts of NO3-, with respect to SO42- from ship emissions, are found in model simulations in southern California (Dabdub, 2008). Indeed, Dabdub (2008) shows that the aerosol contribution from ship emissions is 0.05 % for NO3- and 44 % for SO42-.

Elemental and Organic Carbon are also present in the ship plume (Shah et al., 2004). In particular, OC constitutes about 15–25 % and EC is generally lower than 1 % of the PM sampled at the plume of main ship engine powered by heavy fuel oil (Agrawal et al., 2008b).

Figure 8 shows EC / V and OC / V vs. V at LMP and CGR. Similarly to sulfate and nitrate, OC / V decreases with increasing V and reaches a minimum value for V > 15 ng m-3 (43.1 and 179 at LMP and CGR, respectively) As discussed in Sect. 3.1, other OC sources in addition to ships are present at CGR even at high values of V.

The pattern of the ratio EC / V vs. V is less clear; in particular, several very low values of EC / V also appear at small values of V. This result is unexpected because V and EC are both markers of the primary ship aerosol, but the data here presented seem to suggest that non negligible EC contributions from other sources were present at CGR and that different fractionating effects acted during the transport. Also in this case the limit value is lower at LMP than at CGR.

Finally, as the limit ratios at CGR are likely affected by other sources than shipping, we assume that the limit ratios obtained at Lampedusa for V > 15 ng m-3 are more representative of cases dominated by ship emissions during summer in a wide region. For this reason, the retrieved lower limits at LMP are also used to quantify the ship contribution at CGR.

With all the limitations described above, by using the lower limits for the ratios nssSO42- / V, NO3- / V and OC / V (representative for ship aerosol) it is possible to estimate the minimum contribution of nssSO42-, NO3- and OC emitted by ships to the total budget of these components and also to the total PM10 mass. It has to be noticed that the aerosol quantification obtained by this method is a rough estimate useful to constrain the ship aerosol contribution. In addition, due to possibly different meteorological conditions and photochemical activity, these values may vary spatially and seasonally.

The minimum ratio of each species with respect to V and the minimum estimated contribution of ship emissions, for the average amount and for the maxima, of the total concentration of these species and of PM10 are reported in Table 4. As previously discussed, the measured OC contribution is multiplied by 2.1 at LMP and by 1.8 at CGR to obtain the total organic aerosol contribution.

The estimated minimum concentration of non-sea-salt sulfate from ship emissions was 1.35 µg m-3 on average during this campaign at LMP. This value is lower than in the previous study by Becagli et al., (2012) obtained over a longer period (2004–2008). The relative contribution to the total sulfate is, however, similar here and in Becagli et al., (2012), suggests a similar role of nssSO42- from ship emissions to the total nssSO42- budget. The study by Becagli et al., (2012) covered an extended time period (2004–2008); the consistency with that study suggests that the results obtained during ChArMEx are not specific of summer 2013, but are representative for a wider temporal and spatial range.

At CGR the minimum ship contribution to sulfate, averaged over the same time period, is higher than at LMP (2.1 µg m-3), but this higher value corresponds to a lower contribution to the total nssSO42-, confirming that other nssSO42- sources are important at CGR.

Marmer and Langmann (2005) estimate that ship emissions contribute by 50 % to the total amount of nssSO42- in the Mediterranean. This value is, as expected, larger than the estimated minimum contribution we derive (about 30 %).

The estimated minimum contribution by ships to the total nssSO42- for cases with the largest ship impact (i.e. highest V concentration) is 69 and 77 % at LMP and CGR, respectively.

Ships appear to contribute, by small fractions, to the total budget of NO3-. As previously mentioned, the NO3- atmospheric chemistry is complex and the contribution of nitrate from ship emission could be highly variable, especially in the Mediterranean region where high amounts of UV radiation and highly reactive radical species are present.

Organic aerosol from ships also contributes significantly to the total OA amount and to the total PM; in particular, at LMP virtually all the OA present in cases with maximum ship impact may be attributed to the ship source.

By summing these three contributions, it is possible to estimate the total aerosol mass due to ship emissions and its contribution to the total mass of PM10. The lower limit for the ship contribution was 2.0 and 3.0 µg m-3, corresponding to 11 and 8.6 % of PM10 at LMP and CGR, respectively.

These percent contributions are higher than the annual average for the Mediterranean region estimated by Viana et al., (2014). It has to be considered that these authors used data from harbour or coastal sites, which are highly affected by other sources in addition to ships, and where gas-to-particle conversion is still at its initial phase. Moreover, the percentages reported in this study are relative to the summer season, when the ship contribution in the Mediterranean region is highest (Becagli et al., 2012).

The estimated lower limit for the ship contribution in cases with maximum ship impact was between 42 and 50 % of the total PM10.