Evidence for heavy fuel oil combustion aerosols from chemical analyses at the island of Lampedusa : a possible large role of ships emissions in the Mediterranean

Measurements of aerosol chemical composition made on the island of Lampedusa, south of the Sicily channel, during years 2004–2008, are used to identify the influence of heavy fuel oil (HFO) combustion emissions on aerosol particles in the Central Mediterranean. Aerosol samples influenced by HFO are characterized by elevated Ni and V soluble fraction (about 80 % for aerosol from HFO combustion, versus about 40 % for crustal particles), high V and Ni to Si ratios, and values of V sol > 6 ng m−3. Evidence of HFO combustion influence is found in 17 % of the daily samples. Back trajectories analysis on the selected events show that air masses prevalently come from the Sicily channel region, where an intense ship traffic occurs. This behavior suggests that single fixed sources like refineries are not the main responsible for the elevated V and Ni events, which are probably mainly due to ships emissions. Vsol, Nisol, and non-sea salt SO 2− 4 (nssSO 2− 4 ) show a marked seasonal behaviour, with an evident summer maximum. Such a pattern can be explained by several processes: (i) increased photochemical activity in summer, leading to a faster production of secondary aerosols, mainly nssSO 2− 4 , from the oxidation of SO2 (ii) stronger marine boundary layer (MBL) stability in summer, leading to higher concentration of emitted compounds in the lowest atmospheric layers. A very intense event in spring 2008 was studied in detail, also using size segregated chemical measurements. These data show that elements arising from heavy oil combustion (V, Ni, Al, Fe) are distributed in the sub-micrometric fraction of the aerosol, and the metals are present as free metals, carbonates, oxides hydrates or labile complex with organic ligands, so that they are dissolved in mild condition (HNO 3, pH1.5). Data suggest a characteristic nssSO 2− 4 /V ratio in the range 200–400 for HFO combustion aerosols in summer at Lampedusa. By using the value of 200 a lower limit for the HFO contribution to total sulphates is estimated. HFO combustion emissions account, as a summer average, at least for 1.2 μg m−3, representing about 30 % of the total nssSO 2− 4 , 3.9 % of PM10, 8 % of PM2.5, and 11 % of PM1. Within the used dataset, sulphate from HFO combustion emissions reached the peak value of 6.1 μg m −3 on 26 June 2008, when it contributed by 47 % to nssSO 2− 4 , and by 15 % to PM10.


Introduction
The lowest grade fuels (the so-called residual fuel oils or heavy fuel oils -HFO) are largely used in power plants and in marine diesel engines.They contain large concentrations of sulphur and different ash forming metals, which contribute Published by Copernicus Publications on behalf of the European Geosciences Union.

S. Becagli et al.: A possible large role of ships emissions in the Mediterranean
strongly to particle emissions (Agrawal et al., 2010).It is difficult to distinguish atmospheric particles produced by power plants or refineries and by ship engines because of the mixing of sources having similar tracers and ratios between them (Viana et al., 2009).These sources are often grouped together in various studies aimed at source apportionment of atmospheric aerosols (e.g.Amato et al., 2009;Mamanea et al., 2008).The simultaneous presence of elevated anthropic (including refineries, power plants, and an intense ship traffic) and natural emissions in the Mediterranean make this region one of the most polluted in the world (e.g., Kouvarakis et al., 2000;Lelieveld and Dentener, 2000;Marmer and Langmann, 2005).Sudies on HFO combustion and ship emissions in the Mediterranean have been carried out mainly in harbour areas and coastal cities by aerosol chemical analysis and source apportionment (Pandolfi et al., 2011;Viana et al., 2009;Mazzei et al., 2008).Assessement studies on the role of ship emissions in the Mediterranean basin are so far based on inventories and model analyses (e.g., Marmer et al., 2009;Olivier et al., 2005;Eyring et al., 2005;Vestreng et al., 2007) and show that the contribution of ships to air pollution in the Mediterranean atmosphere is significant, although its quantification strongly dependent on the used emission inventory.The verification of ship emission inventories, and particularly those in the open sea, against observations is a difficult task due to lack of continuous measurements over the open sea and to the complex involved atmospheric processes.
In this context, studies performed in open sea far from source of HFO combustion other than ships, are relevant in order to to investigate the current impact of the ship emissions on the formations of primary and secondary aerosol, and how the predicted future growth of ship traffic and the geographical expansion of waterways and ports, possibly combined with international regulations, are going to affect the atmospheric composition.
Although shipping contributes significantly to the international transportation sector, its emissions are not well quantified and are one of the least regulated anthropogenic sources (IMO, 2008;Cofala et al., 2007).Several studies point towards the need of international regulations on ship emissions, as those active in Europe, where the land based emissions of sulphur have been successfully reduced since 1980's.
Recently, several studies (Petzold et al., 2008;Moldanova et al., 2009;Lack et al., 2009;Agrawal et al. 2008aAgrawal et al. , b, 2010) ) reported chemical, physical, and optical properties of emitted particles and gases by analyzing plumes of a large number of commercial shipping vessel, also considering different engines load condition.These studies provide the emission factors of various gases (carbon monoxide, nitrogen oxides, sulphur dioxide, carbon dioxide), chemical compounds in the particulates (S, metals), and aerosol mass emitted from marine engines fed with heavy fuel oil.
Gases and particles emitted by ship and other HFO combustion impact human health (Corbett et al., 2007), influence acidification and eutrofization of water and soil in coastal regions due to deposition of sulphur and nitrogen compounds (Derwent at al., 2005), and affect climate through sulphur containing particles (Devasthale et al., 2006;Lauer et al., 2007), greenhouse gases (Stern, 2007), and absorbing black carbon aerosols (Lack et al., 2009).Moreover, sulphate aerosol has an indirect climate effect influencing cloud structure and properties (e.g., Conover, 1966;Coakley and Walsh, 2002) The present paper presents the first experimental identification and quantification of SO 2− 4 from HFO combustion aerosol based on chemical analyses of PM 10 samples collected at the island of Lampedusa in the central Mediterranean Sea.

Measurements and methods
The measurements were carried out at the Station for Climate Observations, maintained by ENEA (the national Agency for New Technologies, Energy, and Sustainable Economic Development of Italy) at Lampedusa (35.5 • N, 12.6 • E).Lampedusa is a small island in the Central Mediterranean sea far at least 100 km from the nearest Tunisian coast.At the Station for Climate Observations, which is located on a 45 m a.s.l.plateau on the North-Eastern coast of Lampedusa, continuous observations of greenhouse gases concentration (Artuso et al., 2009, 2010), aerosol properties (di Sarra et al., 2011;Meloni et al., 2006;Pace et al., 2006), total ozone, ultraviolet irradiance (di Sarra et al., 2002;Meloni et al., 2004;Di Biagio et al., 2010), and other climatic parameters are carried out.
In this study we will use aerosol optical properties measured with a multi-filter rotating shadowband radiometer (MFRSR), and chemical analyses of sampled aerosols on filters.
The MFRSR (Harrison et al., 1994) is a seven-channel radiometer which measures global and diffuse irradiances, and allows the determination of column aerosol optical depth at 5 wavelengths (416, 496, 615, 671, and 869 nm).The measurement details and data retrieval is described by Pace et al. (2006).
The aerosol sampling was started in June 2004 at daily resolution, alternating in sequence sampling of PM 10 , PM 2.5 , and PM 1.0 (particulate matter with aerodynamic diameter lower than 10, 2.5 and 1.0 µm respectively).Only the PM 10 sampling head at daily resolution was used since 2007.During the sampling period some interruptions occurred due to technical failures.Here results on the chemical composition of PM 10 are reported.The sampler was loaded with 47 mm diameter 2 µm nominal porosity Teflon filters.The filters were weighted before and after sampling in order to obtain the mass of the collected atmospheric particulate.All filters were conditioned for at least 24 h prior to weighing at a relative humidity of 35-45 % and temperature of 25 • C.
Another quarter was extracted in ultrasonic bath for 15 min with MilliQ water acidified at pH 1.5-2 with ultrapure nitric acid obtained by sub-boiling distillation.This extract was used for determination of the soluble part of metals by Inductively Coupled Plasma Atomic Emission Spectrometer (ICP-AES, Varian 720-ES) equipped with an ultrasonic nebulizer (U5000 AT + , Cetac Technologies Inc.).Samples have been spiked with 100 ppb of Ge used as internal standard (λ = 209.4nm), and calibration standards were prepared by gravimetric serial dilution from mono standards at 1000 mg L −1 .The value of pH was chosen because it is the lowest values found in rainwater (Li and Aneja, 1992) and leads to the determination of the metals fraction more available for biological organisms.Filters field blancks show V and Ni soluble fractions (hereafter V sol and Ni sol ) concentrations below the detection limits (0.04 ng m −3 and 0.09 ng m −3 , respectively) in working conditions.
The remaining half filters were analysed for the total (soluble + insoluble) elemental composition by the proton induced X ray emission (PIXE) technique (Calzolai et al., 2006;Lucarelli et al., 2011).The PIXE analysis was carried out for a reduced number of samples, therefore, the elemental composition is available for a restricted data set.In addition, the amount of V was in several cases below the minimum detection limit (MDL) of the PIXE technique.
Additional sampling with an 8 stage impactor equipped with Teflon back-up filter was performed in Lampedusa in the period 17 May-20 June 2008.Half of each filter was used to determine the ionic composition, while the other half was used to measure the metals soluble fraction content with the same methodology above described.

Vanadium and nickel
Sulphur is the dominant element in the exhausts of heavy fuel oil combustion, followed by V and Ni (Agrawal et al., 2008a).These metals are also present as minor constituents in the Earth crust, whose main component is Si (66.6 % as SiO 2 in the upper continental crust; Henderson and Henderson, 2009); in the upper continental crust the V/Si and Ni/Si ratio are 3.1×10 −4 w/w and 1.5×10 −4 w/w, respec-tively, while the mean V/Ni ratio is 2.06 w/w (Henderson and Henderson, 2009).Heavy oil is enriched in V and Ni content with respect to the crust, and these metals are generally used as markers of HFO combustion in all size fractions of the atmospheric particulate.Different authors (Mazzei et al., 2008, Viana et al., 2009;Pandolfi et al., 2011) report characteristic values of V/Ni between 2.5 and 3.5 for ships emmissions.These values are obtained by applying statistical approaches (Positive Matrix Factorization) to an extensive chemical data set of aerosol sampled near harbours in the Mediterranean Sea.A relatively wide range of V/Ni ratio (2.3-4.5) was measured by direct sampling at the exhausts of different auxiliary ship engines fed by different fuels (Nigam et al., 2006), and from the main propulsor ship engine at different speed mode (Agrawal et al., 2008a and b).V/Ni = 3 was found from power plant and oil refinery emissions (Pandolfi et al., 2011).
Given the small difference between the typical V/Ni ratios for HFO combustion aerosol and for dust, these two sources cannot be easily separated on the basis of this ratio.

V and Ni enrichment with respect to crustal sources
Since the V/Ni ratio does not allow an unequivocal attribution to the HFO combustion source, we use Si, which is used, when analytically determined, as the main marker for crust to identify and remove from the dataset samples with a significant crustal contribution.Elevated Si levels might be associated, in addition to crustal particles, also to fly ash from high temperature combustion processes, mainly from coal electric power plants (Gaffney and Marley, 2009).However, the contribution of fly ash to PM 10 is negligible (Gaffney and Marley, 2009) and Si, Al, and Fe elements are commonly used to identify and quantify the crustal content, particularly in the Mediterranean region (e.g.Remoundaki et al., 2011;Koc ¸ak et al., 2007;Nicolás et al., 2008).
Peaks of V and Ni are expected when HFO combustion or crustal particles are sampled; the two sources are characterized by different Si content, low for the anthropic (HFO combustion), high for the natural (crustal).Thus, we used the Ni/Si and V/Si ratios to distinguish between V and Ni due to heavy oil combustion and to crustal sources.Ni and V determined by PIXE (i.e.their total content) was used for Ni/Si and V/Si ratios calculation.The ratios between the measured Ni/Si or V/Si and the typical corresponding crustal ratios, defined as enrichment factors, are used to identify crustal samples.Following previous studies (e.g., Adams et al., 1980;Chester et al., 2000), samples with an enrichment value <10 are assumed to correspond with cases influenced by the crustal source, while an enrichment value >10 identifies cases dominated by non-crustal sources (anomalously enriched elements), and in particular heavy oil combustion.
Table 1.Correlation parameters between V and Ni soluble and total fractions for ship aerosol cases (V sol > 6 ng m −3 ) and for Saharan dust cases (Si > 925 ng m −3 and V sol < 6 ng m −3 ).Slope (± error) R 2 n.
V sol /V tot ship events 0.80 (±0.02) 0.932 113 V sol /V tot pure Saharan dust events 0.36 (±0. Figure 1 shows the daily values of Ni/Si derived from PIXE measurements (Ni tot /Si tot ).The Ni tot /Si tot ratio is 10 times higher than the value for crustal particles (dashed line in Fig. 1) in 79 % of the measured samples.A higher fraction of enriched samples is obtained for the V tot /Si tot ratio (88 % of the samples are enriched in V).The V total content was determined in a minor number of samples than Ni because there are cases with V below the PIXE detection.As a consequence, the fraction of data with elevated ratio is higher for V/Si than for Ni/Si.
Most of the not enriched samples belongs to days in with Si > 925 ng m −3 ; this value corresponds to the 75th percentile of the distribution of occurrence of Si concentration.Previous studies carried out on the basis of aerosol optical depth measurements show that dust is present at Lampedusa in about 26 % of the days (Meloni et al., 2007).Cases with Si > 925 ng m −3 correspond to moderate and high dust conditions at Lampedusa, and identify cases which are evidently affected by the crustal component.Back trajectories (not shown for brevity) confirm that the air masses originate from or pass over the Sahara desert before reaching Lampedusa in days with concentration of Si > 925 ng m −3 .Figure 2 shows the scatterplot of daily total content of V versus Si (blue dots).In general, dust-laden samples show a relatively small amount of V, suggesting that cases with elevated values of V have a different source with respect to Si.

V and Ni solubility
The measured concentrations of V sol and Ni sol for the nonenriched events are lower than for the enriched cases, and are always below 8 and 2.6 ng m −3 , respectively.Only 6 events (less than 5 % of the dataset) are in the range 6-8 ng m −3 for V, and 2.3-2.6 ng m −3 for Ni.The threshold of 6 ng m −3 (hereafter V st ) for V sol is chosen to identify HFO combustion enriched samples on the basis of measurements of V sol .
The V and Ni solubility appears to depend on the originating source.By plotting V sol and Ni sol versus the total V and Ni, respectively, we obtain two different behaviours: the HFO combustion enriched cases (V sol > 6 ng m −3 ) display a larger slope, thus a higher solubility than during Saharan dust events (Si > 925 ng m −3 and V sol < 6 ng m −3 ).Table 1 reports slopes and regression coefficients for the various plots.In the events classified as influenced by the anthropic source both V and Ni are easily dissolved (the soluble fraction is 80 % and 77 % for V and Ni, respectively) in the mild extraction condition (HNO 3 -pH 1.5) because they are mainly present as free metals, oxides hydrates, or organo-metal compounds.Sippula et al. (2009) report that several soluble compounds such as sodium vanadates and nickel idroxides are formed during HFO combustion.On the contrary, the presence of V and Ni in the silica matrix or as oxides in samples with high dust content is expected to produce a lower solubility, as it is found in our samples (36 % and 45 % for V and Ni, respectively).Consequently, samples influenced by heavy oil combustion have a higher content of V sol and Ni sol , which are present in the available form for biological organisms, and are more capable to exert toxicity.

Sulphate and PM 10 concentration
Figure 3 shows the temporal evolution of the daily average aerosol optical depth (AOD) at 500 nm, of the daily PM 10 mass concentration, and of the main markers of heavy fuel oil: non sea salt sulphate (nssSO 2− 4 ), V sol and Ni sol from PM 10 samples collected at Lampedusa in the period June 2004-December 2008.Three-month averages of the measured parameters are also shown to emphasize the seasonal evolution.Cases in which the measured AOD is strongly influenced by Saharan dust are also reported in the figure (red dots in Fig. 3a).The method by Pace et al. (2006), based on the measured values of AOD and Ångström exponent and on the analysis of the backward trajectories, was used for the identification of cases with large Saharan dust contribution to the AOD.Since the AOD provides information on the entire air column, the identification of a Saharan dust case does not necessarily imply that mineral particles are present close to the surface.
Figure 3 shows a marked seasonal pattern with springsummer maxima of the AOD and of the tree chemical markers.As discussed by Di Iorio et al. (2009), the dust optical depth and vertical distribution show a large seasonal cycle, with elevated AOD and a wider vertical extension in spring and summer; the seasonal change is mainly controlled by dust transport occurring over the boundary layer.On the contrary, non dust cases and boundary layer aerosols display a very limited seasonal change (black dots and blue line in Fig. 3a), as it is also confirmed by the PM 10 measurements (Fig. 3b).The PM 10 concentration reaches elevated values, up to about 140 µg m −3 , mainly in spring; also the highest peaks in AOD in 2007 and 2008, close to 1, occur in spring.The highest peaks in PM 10 and AOD are due to Saharan dust events, and will not be discussed in detail here.Elevated values and isolated peaks of V sol , Ni sol , and nssSO 2− 4 are observed throughout spring -summer and especially in June-July.A scatter plot of Ni sol versus V sol , and of nssSO 2− 4 versus V sol for cases in wich the antropic contribution is relevant (i.e.V sol >6) is shown in Fig. 4. V sol and Ni sol are closely related, suggesting that the two species originate from the same source.The slope of the linear regression between V sol and Ni sol for the cases classified as influenced by HFO combustion (R 2 = 0.978) is 2.98 ± 0.04, in agreement with previous studies (Mazzei et al., 2008;Agrawal et al.,  represent the three months mean.Red dots in the plot a are related to Saharan dust events chosen as AOD> 0.15 and a < 0.5 (Pace et al., 2006).Black dots represent the remaining days.In this plot the dark line represent the AOD three month average, the blue one the three month mean excluding the Saharan dust event (i.e.calculated over the black dot).
2008a and b, Viana et al. 2009;Pandolfi et al., 2011).A very good correlation (R 2 = 0.945) is also found for events influenced by crustal particles; the slope of the linear fit is lower (2.54 ± 0.05) than for the HFO combustion cases, and somewhat higher than the mean V/Ni crustal ratio (2.06).Cases of mixing between crustal and anthropogenic particles might explain this behaviour.It must be pointed out that Viana et al. (2009) found in some Saharan cases very high V/Ni ratios, up to about 12; although we do not find such high values, mixing of crustal particles of different V/Ni ratio may also play a role.The limited difference found between the slopes for HFO combustion and crustal samples confirms that the V/Ni ratio alone does not allow to separate the two sources.
The behaviour of non-sea salt sulphate does not appear to be closely linked to V sol and Ni sol .Beside heavy fuel oil combustion, several other sources (anthropic origin from long range transport, marine biogenic, crustal, volcanic) contribute to the non sea salt sulphate in the Central Mediterranean Sea.For this reason V sol and nssSO 2− 4 are not directly coupled, and the quantification of SO 2− 4 contribution from HFO combustion emissions to the nssSO 2− 4 total budget is a difficult task.

Identification of heavy oil combustion aerosols: trajectory analysis
The chemical aerosol composition was related to airmass trajectories with the aim of identifying main source regions.Back-trajectories were calculated using local wind data measured at the Lampedusa Station for Climate Observations on a 10 m tall mast with a time resolution of 10 min, assuming that the wind along the back trajectory is at each time step equal to the one measured at the same time at Station.Lampedusa is a small island, far more than 100 km from the nearest coast, and we assume that the wind measured at Lampedusa is representative for a relatively wide region.The main advantage of this method is the direct use of high temporal resolution measurements.The objective of the trajectory analysis is to derive information on the airmass origin in the surrounding of Lampedusa and in a short time interval (generally up to 12 h).We expect that the reliability of the method rapidly decreases with time, and in proximity of the coasts or over land.
Backtrajectories were calculated also with the Hysplit model (Draxler and Rolph, 2012;Rolph, 2009) using the NCEP reanalysis data (2.5 • × 2.5 • spatial and 6 h temporal resolution) during selected periods The two approaches show very similar results.However, we believe that in the proximity of Lampedusa and for the considered limited time interval our methodology produces more detailed results, especially for cases of low wind velocity and fast changing conditions, because of the difficulty of models in reproducing the wind field close to the surface and the low temporal resolution of the reanalysis.Twelve-hour long trajectories are in most cases suited to investigate the area surrounding Lampedusa, and a broad temporal interpolation between successive model analyses is required for the HYSPLIT trajectory calculation.
Figure 5 shows 18-h long trajectories arriving at Lampedusa at the middle of the PM sampling interval.Trajectories are grouped in three-month periods, and plotted separately for cases with V sol < 6 ng m −3 and V sol > 6 ng m −3 .Similar results are found by using 12-h or 24-h trajectories.Trajectories were also calculated for different arrival times, at the beginning or at the end of the PM sampling interval.Except than in few cases, no substantial changes in the trajectory pattern are found.
The origin of the air masses corresponding to V sol <V st reflects the overall wind direction statistics at Lampedusa, without specific preferred directions (22 % of cases wind originates from N-NW; all the other directions show frequencies of occurrence between 2 % and 5 %).The air masses showing evidence of HFO combustion aerosol influence display a strong dependency on the originating direction: they are mostly of Northern origin, and passed over the Strait of Sicily, i.e. in correspondence with the main maritime route crossing the Mediterranean sea from the Strait of Gibraltar to the Suez Canal.26,5 % (35) of the cases with V sol >V st correspond to air masses coming from a narrow direction (345 • ± 7.5 • ); 63.6 % (84) of the cases correspond to air masses originating between 322.5 • and 37.5 • .The same region also includes 80.8 % of the cases with V sol >10 (52 cases).Trajectories with V sol > V st do not show any significant dependency on the wind intensity (the mean wind intensity along the back trajectory ranges from 1.5 to 11 m s −1 ).
The main shipping route through the Sicily channel runs about 180 km North of Lampedusa and the good correspondence between the air mass origin and the Mediterranean Table 2.For the two periods having different resolution time (June 2004 andAugust 2007) are reported: the number of PM samples (measurements of V sol ), the number (frequency occurrence on the total number of annual V sol measurements) of measurements with V sol >6 ng m −3 , the mean and standard deviation values.The number of episodes of single and consecutive days (frequency occurrence on the number of annual V sol measurements affected by ship emission) of V sol >6 ng m −3 , the correspondent periods mean and standard deviation values of the single and multiple day episodes are also presented.

Year
Number main route of large vessels suggests that most of the events characterised by V sol > 6 ng m −3 are directly influenced by ship emissions.Figure 6 shows a statistic of the number of cases with V sol < 6 ng m −3 and V sol > 6 ng m −3 for different originating trajectory directions.Data are displayed for April-May-June, and July-August-September, separately; only two quarterly statistics are shown because the few HFO combustion aerosols events measured in autumn and winter.Evidently, as also appears in Fig. 5, air masses coming from the Sicily Channel during autumn and winter do not neces-sarily show elevated values of V sol , suggesting that transport from north north-west is required, but is not sufficient to determine high concentration of HFO combustion aerosol; different processes probably play a role in the observed seasonal evolution.

Seasonal evolution
As shown in Fig. 3, V sol , Ni sol , and nssSO 2− 4 display a large seasonal cycle, with a marked maximum in spring-summer.Three-month averages of the different parameters displayed  in Fig. 3 were drawn to highlight their seasonal evolution.
The yearly evolution of the three-month averages of V sol and nssSO 2− 4 appear remarkably similar, suggesting that, although the sources may not be coupled, they may partly respond to similar mechanisms.As discussed above, the seasonal behaviour is not due to a significant change in the dynamical patterns and to more frequent trajectories from HFO combustion sources.Different processes thus contribute to produce this seasonal cycle.
Production of secondary aerosol is influenced by solar radiation.This effect is well known for the formation of nssSO 2− 4 , which is derived from oxidation of SO 2 by OH (see, e.g., Barbu et al., 2009).In its turn, the OH production is linked to elevated levels of ultraviolet radiation.The central Mediterranean is characterized by elevated levels of UV radiation in summer, and high ozone photolysis and OH production (e.g., Casasanta et al., 2011).Consequently, SO 2 conversion into SO 2− 4 is expected to be faster in summer.Ault et al. (2010), in a recent study found V and SO 2− 4 in single aerosol particles from fuel combustion; they found that SO 2− 4 is in higher concentration than in particles not containing V and suggest a catalytic effect of V on the oxidation of SO 2 to SO 2− 4 .In this process the metals would be involved in the secondary aerosol formation, favoured by high level of solar radiation.
The structure of the planetary boundary layer is expected to play a role in producing the observed seasonal cycle.Song at al. ( 2003) used a Lagrangian photochemical box model for an air parcel emitted from ship in the marine boundary layer.They found that conditions favouring the stability of the marine boundary layer produce a larger increase in the SO 2 and sulphate concentrations than an increased emission of SO 2 .The marine boundary layer is generally more stable in summer than in winter (e.g., Dayan et al., 1989), and this effect may produce elevated concentrations of sulphate and metals during this season.
Also the variability of the traffic ship may play a role in determining the observed annual cycle.Recent modeling studies which used ship emission inventories assumed a constant source throughout the year (e.g., Marmer and Langmann, 2005;Jonson et al., 2009), explicitly because of lacking information.For instance, Marmer and Langmann (2005) found mean concentrations of sulphate aerosol almost four times higher in summer than in winter in the lowest level of their model.A limited information is available on the seasonal variability of the ship traffic in the Mediterranean, and only for selected regions (e.g., Tzannatos, 2010;ORTC, 2011).Seasonal changes in the ship traffic appear to be primarily connected with touristic activities/passenger transport, and mainly interest specific coastal areas (e.g., the Ligurian sea, several routes in Greece, etc.).The emissions from the passenger ships are however estimated to be small (Jonson et al., 2009), and probably produce a limited effect on the seasonal distribution of the emissions.
Table 2 presents some statistics on V sol measurements and the cases of consecutive days with V sol >V st .The whole database extends over 5 yr, but only the last 2 yr are characterised by regular daily samples.As shown in Table 2, from 2004 to 2006 there are a total of 235 V sol measurements (i.e.yearly number of measurements varying between 66 and 91), and only about 12 % of the cases presents V sol >V st .The number of HFO combustion aerosol events with duration of more than one day is largest in summer.
The representativeness of 2007 and 2008 is definitely larger than in previous years, with a total of 549 daily measurements of V sol .Thus we believe that results for 2007-2008, with about 19 % of cases with V sol >V st is more representative for the normal situation at the Lampedusa.
A simple statistics on HFO combustion events lasting more than one day is reported in Table 2.The number of cases lasting more than one day is larger in 2007-2008 (70.2 % of the cases last between 2 and 5 consecutive days) than in the 2004-2005 period (28.6 %); however, results for 2004-2006 are biased, since PM 10 samples were not taken continuously.This effect tends to produce an underestimate of the influence of HFO combustion events in 2004-2006.Moreover, most of the episodes lasting more than one day show mean V sol values larger than the mean V sol of the single day cases, suggesting a progressive increase of the HFO combustion aerosol load in specific situations, leading to a strong intensity of episodes lasting for more than one day.

The June-July 2008 event: trajectory analysis
Figure 7 shows the back trajectories of a particular long lasting (20 days) and interesting event in which the HFO combustion influence on the aerosol chemical composition was observed to occur almost continuously from 20 June to 9 July, 2008.Elevated values of V sol were measured throughout the period; V sol decreased below V st during only 3 days.At the beginning of the event (20 and 21 June) the wind intensity decreased and advected air from the middle of the Sicily channel.From 22 to 24 June there is a notable reduction of the wind intensity, and easterly winds.On 25 June the wind was again coming from North-West, and values of V sol <V st were measured.From 26 June to 2 July V sol reached very high values, including the maximum value of the whole dataset (30.6 ng m −3 ) on 26 June; 4 out of the 7 values of V sol > 20 ng m −3 of the dataset occurred between 26 June and 1 July.In this period the air mass came from North, progressively shifting from Northwest to Northeast.A change of the wind direction led to Southern local air masses and V sol <V st on 3 July.With the exception of the back trajectory of 7 July, when southern air mass induced a low value of V sol , the period ends with a recovery of the Northerly air mass flow, still inducing high values of V sol .
This long extended period with strong influence from HFO combustion, and very likely ship emissions, confirms that the origin of the air masses plays a central role.In particular, the strait of Sicily seems to be the main source region.Moreover, very low winds connected with stagnant conditions are not responsible for elevated values of V sol , confirming that high V sol values are not of local origin (i.e., from harbour activities).As expected, V and Ni in the HFO combustion event display a maximum in the finest mode (diameter <0.4 µm).Conversely, their concentrations peak at larger size (1.1-2.1 µm for Ni, and 0.4-0.7 µm for V) during the Saharan dust event.

The
The sulphate distribution shows a monomodal distribution peaked in the 0.4-0.7 µm fraction during the HFO combustion event, due to the main secondary source.On 17 May sulphate displays a bimodal distribution with a second mode (1.1-2.1 µm) related to its primary sources from Saharan dust and sea spray.
Al and Fe are considered typical markers of the crustal source and are mainly present in the coarse mode on 17 May.During the HFO combustion event their distributions are shifted towards fine particles, suggesting that also the anthropic source may contribute to the occurrence of these elements.The presence of a fine mode also on 17 May suggest the presence of mixed source in the selected event.
It must be noticed that even if for Ni and V we report here the pH 1.5 soluble fraction, and the solubility is different for the two aerosol sources, their concentrations are higher in the HFO combustion event than during the Saharan dust case.The situation is different for the crustal markers.Al shows higher concentrations in the Saharan dust event, while Fe concentrations are higher in the HFO combustion case.Part of this behaviour is due to the different solubility of Fe of different origin.In Saharan dust aerosol Fe is associated with the silicate matrix, or present as oxides, and these compounds are not soluble in the applied conditions (HNO 3 -pH 1.5).In PM 10 samples collected at Lampedusa Fe sol is generally less than 10 % of the total Fe.On the contrary, the Fe solubility is high for anthropic sources, due to the presence of Fe 0 or Fe-idroxides, both soluble in HNO 3 .

Heavy oil combustion contribution to the total aerosol load
By using specific markers of different sources and applying receptor models it is possible to quantify the mass contribution of a specific source to the total PM (Gordon, 1988).Emission factors from different marine vessels and engines and other heavy oil combustion sources can provide insightful information to aid source apportionment studies even if the ratios between the selected markers change from the used to discriminate between crustal and ship sources.The Ni and V soluble fractions are consistently higher for the cases influenced by HFO combustion than for mineral particles.The V soluble fraction was chosen to identify days affected by high anthropic aerosol content.A threshold of 6 ng m −3 was established to select heavy oil combustion aerosol events.
A progressive vector analysis based on wind measurements show that the selected events are very likely mainly affected by sea going ship, mainly transiting in the Sicily Chanel.
A marked seasonal behaviour, with summer maxima, is observed for all the main HFO combustion aerosol markers (V sol , Ni sol and nssSO 2− 4 ).The following processes occurring in summer are believed to play an important role in determining the observed seasonal evolution: 1-Increased photochemical activity of atmosphere in summer, leading to a larger production of secondary aerosols, mainly nssSO 2− 4 2-MBL stability, higher in summer than winter, that prevents the accumulation of HFO combustion emission compounds in the lowest atmospheric levels in winter.
The higher occurrence of late spring-summer episodes lasting more than one day in correspondence with dominant wind direction from the Sicily Channel, suggests that ship emissions contribute significantly to the heavy oil combustion aerosols.
A very intense event in spring 2008 was studies.Size segregated chemical analyses show that elements arising from heavy oil combustion (V and Ni, but also Al and Fe) are mainly distributed in the sub-micrometric fraction of the aerosol.The metals are present as free metals, carbonates, oxides hydrates or labile complex with organic ligands, so that they are dissolved in mild condition (HNO 3 , pH1.5).
The SO 2− 4 /V ratio of 200 is proposed as lower limit characteristic ratio for HFO combustion in summer at Lampedusa.The experimental determination of a characteristic ratio of HFO combustion markers in a receptor site is particularly useful for understanding and quantifying the profile emission.
Between the components emitted by HFO combustion, SO 2− 4 contributes significantly to the total PM mass.In summer the estimated average mass is 1.2 µg m −3 , about 30 % of the total nssSO 2− 4 , 3.9 % of PM 10 , 8 % PM 2.5 , and 11 % PM1.Sulphate from HFO combustion reach a peak value of 6.1 µg m −3 on 26 June 2008, corresponding to 47 % of nssSO 2− 4 , and 15 % of PM 10 .It has to be emphasized we used a conservative value for the SO 2− 4 /V ratio which is expected to produce a lower limit of the HFO combustionderived sulphates, whose contribution may be higher than derived here.
For future studies, more data on size resolved chemical composition (including other pollutants emitted in the ship plume, e.g.organic and elemental carbon) from sites located in the open sea can provide a better evaluation of the regional and global impact of ship aerosols on the total aerosol budget and on climate.

Fig. 1 .
Fig. 1.Temporal evolution of Ni tot /Si tot in PM 10 samples collected at Lampedusa island.Black line represent the Ni/Si mean ratio in upper continental crust(Henderson and Henderson, 2009), dashed line represent the threshold values for sharing samples enriched (one order of magnitude higher than black line).Red dots are samples with high dust content (Si> 925 ng m −3 ).

Fig. 2 .
Fig. 2. Scatter plot of V total content obtained by PIXE versus Si.

Fig. 3 .
Fig. 3. Temporal evolution of AOD, PM 10 , nssSO 2− 4 V sol and Ni sol , at Lampedusa island.Blue lines in plot (b), (c), (d), and (e)represent the three months mean.Red dots in the plot a are related to Saharan dust events chosen as AOD> 0.15 and a < 0.5(Pace et  al., 2006).Black dots represent the remaining days.In this plot the dark line represent the AOD three month average, the blue one the three month mean excluding the Saharan dust event (i.e.calculated over the black dot).

Fig. 5 .
Fig. 5. Panels (a) and (b)show the 18 h back trajectory relatives to the measurements of V sol less than or great/equal to 6 m −3 (i.e.V st ) respectively.The trajectories are grouped in quarters, each one representing three months.

Fig. 6 .
Fig. 6.Number of cases with V sol < 6 ng m −3 (blu regions) and V sol >6 ng m −3 (red regions) for different originating directions of the corresponding airmass trajectories for April-May-June (left), and July-August-September (right).

Fig. 7 .
Fig.7.18-h back trajectories for the 20 June-9 July 2008 period, when only 3 daily V sol measurements (25 June, 3 and 7 July) were below V st .The 24-h back trajectory of 17 May is also presented.The number at the end of each back-trajectories indicates the corresponding day of June and July.
Figure8shows the particle size distribution derived from multi-stage impactor sampling.The size distribution of the main markers of HFO combustion emissions and crustal elements for the first day of the described event (20 June 2008) and for 17 May 2008, characterized by a high crustal content, are shown.The backward trajectories of 17 May computed over 24 h, shows that the airmass originated from North Africa.As expected, V and Ni in the HFO combustion event display a maximum in the finest mode (diameter <0.4 µm).Conversely, their concentrations peak at larger size (1.1-2.1 µm for Ni, and 0.4-0.7 µm for V) during the Saharan dust event.The sulphate distribution shows a monomodal distribution peaked in the 0.4-0.7 µm fraction during the HFO combustion event, due to the main secondary source.On 17 May sulphate displays a bimodal distribution with a second mode (1.1-2.1 µm) related to its primary sources from Saharan dust and sea spray.Al and Fe are considered typical markers of the crustal source and are mainly present in the coarse mode on 17 May.During the HFO combustion event their distributions are shifted towards fine particles, suggesting that also the anthropic source may contribute to the occurrence of these elements.The presence of a fine mode also on 17 May suggest the presence of mixed source in the selected event.It must be noticed that even if for Ni and V we report here the pH 1.5 soluble fraction, and the solubility is different for the two aerosol sources, their concentrations are higher in the HFO combustion event than during the Saharan dust case.The situation is different for the crustal markers.Al shows higher concentrations in the Saharan dust event, while Fe concentrations are higher in the HFO combustion case.Part of this behaviour is due to the different solubility of Fe of different origin.In Saharan dust aerosol Fe is associated with the silicate matrix, or present as oxides, and these compounds are not soluble in the applied conditions (HNO 3 -pH 1.5).In PM 10 samples collected at Lampedusa Fe sol is generally less than 10 % of the total Fe.On the contrary, the Fe solubility is high for anthropic sources, due to the presence of Fe 0 or Fe-idroxides, both soluble in HNO 3 .