Characterization of PM10 sources in the central Mediterranean

Abstract

The Mediterranean Basin atmosphere is influenced by both strong natural and anthropogenic aerosol emissions and is also subject to important climatic forcings. Several programs have addressed the study of the Mediterranean basin; nevertheless important pieces of information are still missing. In this framework, PM${}_{10}$ samples were collected on a daily basis on the island of Lampedusa (35.5${}^{\circ }$ N, 12.6${}^{\circ }$ E; 45 m a.s.l.), which is far from continental pollution sources (the nearest coast, in Tunisia, is more than 100 km away). After mass gravimetric measurements, different portions of the samples were analyzed to determine the ionic content by ion chromatography (IC), the soluble metals by inductively coupled plasma atomic emission spectrometry (ICP-AES), and the total (soluble $+$ insoluble) elemental composition by particle-induced x-ray emission (PIXE). Data from 2007 and 2008 are used in this study.

The Positive Matrix Factorization (PMF) model was applied to the 2-year long data set of PM${}_{10}$ mass concentration and chemical composition to assess the aerosol sources affecting the central Mediterranean basin. Seven sources were resolved: sea salt, mineral dust, biogenic emissions, primary particulate ship emissions, secondary sulfate, secondary nitrate, and combustion emissions. Source contributions to the total PM${}_{10}$ mass were estimated to be about 40 % for sea salt, around 25 % for mineral dust, 10 % each for secondary nitrate and secondary sulfate, and 5 % each for primary particulate ship emissions, biogenic emissions, and combustion emissions. Large variations in absolute and relative contributions are found and appear to depend on the season and on transport episodes. In addition, the secondary sulfate due to ship emissions was estimated and found to contribute by about one-third to the total sulfate mass. Results for the sea-salt and mineral dust sources were compared with estimates of the same contributions obtained from independent approaches, leading to an estimate of the water content bound to the sea salt in the marine source.

Introduction

Atmospheric aerosols are estimated to have negative effects on human health and to play a relevant role in climate, also affecting the hydrological cycle (IPCC, 2007). In this context, the Mediterranean Basin is a complex area due to the strong influences from both natural and anthropogenic emissions, to the peculiar meteorology of the basin, and to conditions (high solar irradiation, summer temperature, humidity) favoring secondary aerosol formation processes (Pey et al., 2009; Im et al., 2012). Information on the aerosol sources and properties in the Mediterranean area is essential for the determination of the impacts on the health of the population living in the countries facing the basin (among the world's most populated areas) and on the climate of the area. This is particularly important because future climate projections by IPCC consider the Mediterranean among the regions most sensitive to climate change (IPCC, 2007).

Owing to its importance, Mediterranean atmospheric aerosol and meteorology have been intensively studied (e.g., Salvador et al., 2014; Pey et al., 2013; Pace et al., 2006; Kopanakis et al., 2012; Dayan et al., 1989) and several international programs have focused on the assessment of chemistry and climate and their interactions; nevertheless, important pieces of information are still missing. For instance, the source contribution and the forcing of short-lived species (both of them highly space and time variable) remain uncertain and are not taken into account well in climate simulation.

Without aiming for a thorough review of the published information, it is worth recalling that PM${}_{10}$ concentration is observed to increase from the northwestern to southeastern Mediterranean due to increasing contributions from both natural and anthropogenic sources (Pey et al., 2013); the western and eastern Mediterranean basins have been observed to be characterized by different atmospheric dynamical and chemical processes (e.g., neutralization of secondary aerosol; Querol et al., 2009). The Mediterranean boundary layer has been observed to generally follow a clear annual behavior, with minimum vertical extension during summer and maximum altitudes in winter (with an opposite trend with respect to the one commonly observed in continental areas) (Pey et al., 2009; Dayan et al., 1989); therefore, stagnation of pollutants is favored in summer over the Mediterranean basin.

Most of the studies performed in the Mediterranean area are based on samplings carried out at coastal sites, with strong influences from the continental meteorology and from anthropogenic sources; thus, regional background and remote sites are essential for the study of natural sources and to obtain information on the Mediterranean basin (Querol et al., 2009; Pey et al., 2009; Koulouri et al., 2008; Pace et al., 2006). In this context, the island of Lampedusa represents an ideal site; in fact, Lampedusa's environment is poorly affected by anthropogenic pollution sources from local origin, due to its scarce population and industrialization, and from abroad, as it is more than 100 $\mathrm{km}$ far from the nearest coast. As discussed by Henne et al. (2010), Lampedusa is one of the most remote sites with respect to air quality measurements among 34 measurement stations throughout Europe. In addition, Lampedusa is located in the central Mediterranean Sea, in an area where few remote or background sampling sites exist despite the crucial position; in fact, as previously mentioned, western and eastern Mediterranean Basin areas are known to be prone to different atmospheric dynamics and pollution conditions.

In this work, an investigation aimed at assessing the aerosol sources affecting the central Mediterranean basin and at quantifying their contributions has been carried out. This study answers to one of the main scientific question of the ChArMEx (Chemistry-Aerosol Mediterranean Experiment) initiative. ChArMEx proposes to set up a coordinated experimental and modeling international effort based on most updated tools, for an assessment of the regional budgets of tropospheric trace species, their trends, and their impacts on air quality, marine biogeochemistry, and regional climate. The source apportionment here applied is based on the application of the Positive Matrix Factorization (PMF) model to a 2-year long data set of PM${}_{10}$ mass concentration and chemical composition of samples collected on the Lampedusa Island. It is worth recalling that information on aerosol sources and impact is among the data needed to constrain uncertainties in climate change models, and it is also important for local administrations and policy makers for the development of effective pollution abatement strategies. Studies on the source assessment/apportionment of aerosols, often focused on specific sources such as African dust or ship emissions, are available in literature for sites in the western Mediterranean (e.g., Montseny (Cusack et al., 2013) and Palma de Mallorca (Pey et al., 2009) in Spain) and in the eastern Mediterranean (e.g., the Greek islands of Crete (Koulouri et al., 2008; Hildebrandt et al., 2011) and Rhodes (Argyropoulos et al., 2012)). This work adds an important piece of information still missing in the central Mediterranean area.

Sampling site location.

[Figure omitted. See PDF]

Materials and methods

Sampling site and instrumentation

As previously mentioned, sampling was performed on Lampedusa (Italy), a small island located in the central Mediterranean, far from continental pollution sources (the nearest coast, in Tunisia, is more than 100 $\mathrm{km}$ away). More in detail, sampling was carried out on the plateau on the northeastern coast, at the Station for Climate Observations (35.5${}^{\circ }$ N, 12.6${}^{\circ }$ E, 45 $\mathrm{m}$ a.s.l.; see Fig. 1), maintained by ENEA (Italian Agency for New Technologies, Energy and Sustainable Economic Development), where continuous observations of greenhouse gases concentration (Artuso et al., 2009), total ozone (Gómez Amo et al., 2012), ultraviolet irradiance (Meloni et al., 2005), aerosol properties (Pace et al., 2005, 2006; Di Iorio et al., 2009; Marconi et al., 2014), and other climatic parameters (Di Biagio et al., 2009; di Sarra et al., 2011) are routinely carried out.

Daily PM${}_{10}$ samples have been collected at this station since 2004, every third day up to 2006 (Becagli et al., 2012; Marconi et al., 2014) and every day since 2007. Particulate was sampled by a low volume sequential sampler (2.3 ${\mathrm{m}}^{\mathrm{3}}{\mathrm{h}}^{-\mathrm{1}}$) equipped with a PM${}_{10}$ inlet, on 47 $\mathrm{mm}$ diameter 2 $\mathrm{\mu }\mathrm{m}$ nominal porosity Pall Gelman Teflon filters.

This paper focuses on the 2-year data set relative to the years 2007–2008, when PM${}_{10}$ was collected every day, accounting for a total of 562 samples (due to episodic technical failures, some short sampling interruptions occurred in the period).

Additional samplings of PM${}_{10}$ on quartz filters were performed in 2007, from 16 May to 15 August, using a low volume Tecora ECHO-PM sampler (2.3 ${\mathrm{m}}^{\mathrm{3}}{\mathrm{h}}^{-\mathrm{1}}$), with collection times ranging from 24 to 72 $\mathrm{h}$, for a total of 24 samples.

Analyses

The mass of the collected PM${}_{10}$ was obtained by gravimetric measurements of the Teflon filters before and after sampling. Before weighting, filters were conditioned for at least 24 $\mathrm{h}$ in an environmentally controlled room (35–45 % relative humidity, 25 ${}^{\circ }$C).

Afterwards, every Teflon filter was cut into three portions: a quarter was analyzed for the ionic content by ion chromatography (IC), a second quarter for soluble metals by inductively coupled plasma atomic emission spectrometry (ICP-AES), and the remaining half filter was analyzed for the total (soluble and insoluble) elemental composition by particle-induced x-ray emission (PIXE).

More in detail, the quarter of Teflon filter devoted to ion chromatography was extracted in about 10 $\mathrm{mL}$ (accurately weighted) of ultra-pure water (MilliQ water, resistivity $>18$ $\mathrm{M}\mathrm{\Omega }$) in ultrasonic bath for 15 $\min$ and analyzed for the ionic content by using three Dionex ion chromatographs fed by an auto-sampler. Cations (${\mathrm{Na}}^{+}$, ${\mathrm{NH}}_{\mathrm{4}}^{+}$, ${\mathrm{K}}^{+}$, ${\mathrm{Mg}}^{\mathrm{2}+}$, ${\mathrm{Ca}}^{\mathrm{2}+}$), inorganic anions (${\mathrm{F}}^{-}$, ${\mathrm{Cl}}^{-}$, ${\mathrm{NO}}_{\mathrm{3}}^{-}$, ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$), and some organic anions (methanesulfonate (MSA), acetate, formate, glycolate, oxalate) were determined for each sample as reported in Becagli et al. (2011). Detection limits (DLs) were in the range 0.08–0.16 $\mathrm{ng}{\mathrm{m}}^{-\mathrm{3}}$ and reproducibility was better than 5 %.

Another quarter was extracted in ultrasonic bath for 15 $\min$ with MilliQ water acidified at pH 1.5–2 with ultra-pure nitric acid obtained by sub-boiling distillation. This extract was used for the determination of the soluble part of selected metals (Al, As, Ba, Cd, Cr, Cu, Fe, Mn, Mo, Ni, Pb, V, Zn) with an ICP-AES spectrometer (Varian 720-ES) equipped with an ultrasonic nebulizer (U5000 AT$+$, Cetac Technologies, Inc.). The value of pH was chosen because it is the lowest value found in rainwater (Li and Aneja, 1992) and therefore leads to the determination of the metal fraction more representative of the anthropogenic sources (Becagli et al., 2012; Traversi et al., 2014).

PIXE measurements on the remaining half Teflon filters were performed at the INFN-LABEC laboratory, equipped with a 3MV Tandetron accelerator, where research is focused in the fields of cultural heritage (Castelli et al., 2013; Fedi et al., 2013), material science, and atmospheric aerosols (Moreno et al., 2013; Calzolai et al., 2011). The PIXE technique is based on the detection and analysis of the X-rays emitted by the sample after excitation by an accelerated particle beam and allows the quantification of $Z>10$ elements. Briefly, samples were bombarded by a 3.2 $\mathrm{MeV}$ proton beam ($i\sim \mathrm{5}$ $\mathrm{nA}$, $t\sim 300$ $\mathrm{s}$) using the setup widely described elsewhere (Calzolai et al., 2006; Lucarelli et al., 2011); spectra were fitted using the Gupix software package (Campbell et al., 2010) and elemental concentrations were obtained via calibration relative to thin reference standards. DLs are in the 1–20 $\mathrm{ng}{\mathrm{m}}^{-\mathrm{3}}$ range. Uncertainties were determined by a sum of independent uncertainties on certified standard sample thickness (5 %), peak areas (from 2 to 20 % or higher when concentrations approach DLs) and sampling parameters (of the order of few percent). The concentrations of lighter elements (Na, Mg, Al, and Si) were corrected for self-absorption effects, i.e., lower energy x-rays attenuation inside aerosol particles. Correction factors were estimated on the basis of PIGE (Particle Induced Gamma-ray Emission) measurements carried out for the quantification of $\mathrm{Na}$ (Calzolai et al., 2010), using an attenuation model for single spherical particles (Formenti et al., 2010); the uncertainty on the PIXE quantification of light elements also bears the uncertainty on these correction factors (of the order of few percent).

As regards the additional samplings carried out in May–August 2007, the 24 samples collected on quartz fibre filters were analyzed for total, elemental, and organic carbon fractions (TC, EC, and OC) by a Sunset thermo-optical transmittance analyzer following a proxy of the IMPROVE protocol (He-580 in Piazzalunga et al., 2011).

PMF analysis

The PMF is a receptor model allowing the determination of the chemical profiles of the aerosol sources affecting the receptor site and the estimation of their contributions to the aerosol load, using as input data the atmospheric concentrations of different chemical species.

Briefly, PMF is an advanced factor analysis technique (Paatero, 1997; Paatero and Hopke, 2003) that uses realistic error estimates to weigh data values and imposes non-negativity constraints in the computational process, which is based on a weighted least-squares approach. With more detail, PMF model may be written as $\mathbf{X}=\mathbf{G}\cdot \mathbf{F}+\mathbf{E}$, where $\mathbf{X}$ is the $n\times m$ matrix of the $m$ measured chemical species in $n$ samples and $\mathbf{G}$ and $\mathbf{F}$ are factor matrices to be determined: $\mathbf{G}$ is the $n\times p$ matrix of source contributions to the samples, $\mathbf{F}$ is the $p\times m$ matrix of factor composition (source profiles), and $\mathbf{E}$ is the residual matrix.

Input data were handled according to Polissar et al. (1998): measured data were weighted with their overall analytical uncertainty increased by one-third of the DL; data below the DL were substituted with a value corresponding to DL/2, and a DL $\cdot$ 5/6 uncertainty was associated to them. Missing data were handled by filling the gaps with the geometric mean value calculated over the days with data and associating to them a 400 % uncertainty.

Mass concentration was put in the PMF analysis as an internal variable, with very high (400 %) uncertainty (Norris et al., 2009), as it is suggested when portions of the mass (not directly correlated with the other input variables) are not measured/available: in our case, no data on the carbonaceous fraction are available for the whole considered period. As an additional test, PMF was also run with normalization of $\mathbf{G}$ and $\mathbf{F}$ matrices by multilinear regression analysis on the PM${}_{10}$ gravimetric mass concentrations and gave comparable results for both contributions and profiles of the detected sources with respect to PMF runs performed with internal mass, thus assuring that mass data did not influence the profiles.

PMF was run with the algorithm PMF2 (Paatero, 1997) in the robust mode, in order to reduce the influence of outliers. Solutions with different numbers of factors were examined; rotational ambiguity was tested by analyzing multiple FPEAK rotations; several seeds were investigated and values of $Q/Q_{\exp }$ were examined (where $Q$ is the weighted least-squares function to be minimized).

Daily and monthly mean PM${}_{10}$ concentrations in the years 2007–2008.

[Figure omitted. See PDF]

Results and discussion

PM${}_{10}$ concentration

The PM${}_{10}$ daily concentration is reported in Fig. 2: a great variability was observed, with values spanning 1 order of magnitude, as minima were found to be around 10 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ and several episodes reached more than 100 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$.

The average PM${}_{10}$ concentration for the whole period (2007–2008) was $32\pm 17$ $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ (median: 29 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$; 25th percentile: 21 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$; 75th percentile: 39 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$); no relevant difference was observed between the years 2007 and 2008, with mean PM${}_{10}$ concentrations of $33\pm 15$ $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ and $32\pm 19$ $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$, respectively. These values are similar to those measured in two different background/rural stations on the island of Crete, Greece (Kopanakis et al., 2012; Gerasopoulos et al., 2006), and in the range observed for urban background sites across southern Europe (Karanasiou et al., 2014).

Despite the distance of the sampling site from pollution sources, concentrations exceeding the daily European Union's limit value (50 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$) were registered in 24 days (out of 278 sampling days) in 2007 and in 32 (out of 284 sampling days) in 2008, i.e., for 10 % of the sampling period, with the highest concentrations recorded in correspondence to Saharan dust events as shown in Marconi et al. (2014). The EU legal term of not exceeding the daily limit value of 50 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ for the percentile 90.4 of the annual values was not respected in 2008 (when percentile 90.4 was 52.5 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$) and nearly reached in 2007 (when percentile 90.4 was 48.9 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$).

Average concentrations (Av) and standard deviation (Sd) of the main measured elements/species for all the sampling period and in the different seasons.

	Year			Winter (DJF)			Spring (MAM)			Summer (JJA)			Fall (SON)
$n$	562			136			156			142			128
	$\mathrm{ng}{\mathrm{m}}^{-\mathrm{3}}$			$\mathrm{ng}{\mathrm{m}}^{-\mathrm{3}}$			$\mathrm{ng}{\mathrm{m}}^{-\mathrm{3}}$			$\mathrm{ng}{\mathrm{m}}^{-\mathrm{3}}$			$\mathrm{ng}{\mathrm{m}}^{-\mathrm{3}}$
	Av	Sd		Av	Sd		Av	Sd		Av	Sd		Av	Sd
Na	3137	1996		3584	1972		3293	1937		2953	2108		2686	1860
Mg	374	187		377	171		405	201		345	190		367	177
Al	360	549		234	432		347	570		339	446		530	682
Si	896	1389		581	1106		889	1435		830	1097		1308	1756
S	1534	806		1051	387		1644	762		2098	904		1281	634
Cl	4708	3725		5727	3945		5263	3783		3592	3663		4212	3052
K	314	223		267	190		319	238		324	198		346	256
Ca	862	1120		694	916		922	1277		761	964		1077	1237
Ti	30	46		20	35		30	49		29	41		41	56
$V$	5.2	5.4		2.8	3.0		6.1	5.6		8.4	6.5		3.2	3.3
Mn	5.3	7.1		3.5	5.5		5.7	7.7		5.5	6.1		6.5	8.4
Fe	304	445		200	333		316	471		290	379		415	547
Ni	2.0	1.7		1.2	1.0		2.4	1.9		3.1	1.9		1.3	1.0
Cu	1.8	1.3		1.3	1.4		1.7	1.3		2.5	1.2		1.5	1.1
Br	15	11		20	13		15	10		13	11		12	9
Sr	7	8		6	5		8	10		6	7		7	8
Na${}^{+}$	3191	2191		3481	2230		3629	2352		2912	2211		2670	1751
NH${}_{\mathrm{4}}^{+}$	673	452		477	313		757	481		954	441		463	340
K${}^{+}$	161	88		153	84		172	82		183	103		131	69
Ca${}^{\mathrm{2}+}$	615	717		475	433		670	859		532	508		784	905
Cl${}^{-}$	4304	3743		5111	3807		4924	4019		3278	3646		3829	3072
NO${}_{\mathrm{3}}^{-}$	2128	1148		1866	958		2434	1232		2283	1218		1861	1028
SO${}_{\mathrm{4}}^{\mathrm{2}-}$	3652	2074		2478	1110		4011	2059		4961	2276		3010	1674
Ossal.	144	85		101	58		151	83		194	80		126	88
Glyc.	6	5		5	4		8	6		7	6		6	4
MSA	22	23		4	4		22	19		46	25		14	14
$V_{\mathrm{sol}}$	4.0	4.3		2.1	1.7		4.1	4.2		7.4	5.6		2.3	1.7
Mn${}_{\mathrm{sol}}$	3.1	3.9		1.9	3.9		3.3	4.2		3.4	3.0		3.9	4.2
Ni${}_{\mathrm{sol}}$	1.4	1.5		0.7	0.5		1.6	1.6		2.5	1.8		0.8	0.6
Cu${}_{\mathrm{sol}}$	1.2	1.1		0.7	1.1		1.3	1.3		1.8	0.9		0.9	0.8
As${}_{\mathrm{sol}}$	0.1	0.1		0.1	0.1		0.2	0.2		0.2	0.1		0.1	0.1
Pb${}_{\mathrm{sol}}$	2.9	2.9		2.3	2.9		3.5	2.4		3.4	2.5		2.5	3.4
	$\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$			$\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$			$\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$			$\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$			$\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$
Dust	5.8	8.2		4.0	6.5		5.8	8.8		5.3	6.8		8.1	10.1
SeaS.833	8.3	6.8		9.6	7.1		9.6	7.2		7.0	6.8		6.9	5.4

PM${}_{10}$ speciation

Average concentrations (and standard deviation) of the main measured elements/species for all the sampling period and in the different seasons are reported Table 1. The determination of several elements by different techniques allowed performing a quality assurance analysis of data and studying the solubility of specific elements (Becagli et al., 2012), by comparing results on the total content as obtained by PIXE and on the soluble content as obtained by IC or ICP-AES.

Sea-salt and mineral dust contributions were estimated assuming the average composition of seawater and Earth's soil (Mason, 1966). In particular, sea-salt concentration (i.e., [sea salt]) was evaluated, for every day, according to the expression $$\begin{array}{cc}\text{[sea salt]}& =\left[{\mathrm{Cl}}^{-}\right]+\left[{\mathrm{ssNa}}^{+}\right]\\ & +\left[\mathrm{ssCa}\right]+\left[\mathrm{ssMg}\right]+\left[\mathrm{ssK}\right]+\left[{\mathrm{ssSO}}_{\mathrm{4}}^{\mathrm{2}-}\right],\end{array}$$ where $\left[{\mathrm{Cl}}^{-}\right]$ is the $\mathrm{Cl}$ concentration measured by IC and $\mathrm{Na}$, $\mathrm{Mg}$, $\mathrm{K}$, $\mathrm{Ca}$, and ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ contributions due to sea salt ($\left[{\mathrm{ssNa}}^{+}\right]$, and $\left[\mathrm{ssMg}\right]$, $\left[\mathrm{ssK}\right]$, $\left[\mathrm{ssCa}\right]$, and $\left[{\mathrm{ssSO}}_{\mathrm{4}}^{\mathrm{2}-}\right]$) are estimated according to the following equations: $$\begin{array}{cc}\left[{\mathrm{ssNa}}^{+}\right]& =\left[{\mathrm{Na}}^{+}\right]-\left[{\mathrm{nssNa}}^{+}\right]=\left[{\mathrm{Na}}^{+}\right]-\left[\mathrm{nssCa}\right]\cdot (\mathrm{Na}/\mathrm{Ca}{)}_{\mathrm{crust}}\\ & =\left[{\mathrm{Na}}^{+}\right]-(\left[\mathrm{Ca}\right]-\left[\mathrm{ssCa}\right])\cdot (\mathrm{Na}/\mathrm{Ca}{)}_{\mathrm{crust}},\end{array}$$ $$\left[\mathrm{ssCa}\right]=\left[{\mathrm{ssNa}}^{+}\right]\cdot (\mathrm{Ca}/\mathrm{Na}{)}_{\mathrm{seawater}},$$ $$\left[\mathrm{ssMg}\right]=\left[{\mathrm{ssNa}}^{+}\right]\cdot (\mathrm{Mg}/\mathrm{Na}{)}_{\mathrm{seawater}},$$ $$\left[\mathrm{ssK}\right]=\left[{\mathrm{ssNa}}^{+}\right]\cdot (\mathrm{K}/\mathrm{Na}{)}_{\mathrm{seawater}},$$ $$\left[{\mathrm{ssSO}}_{\mathrm{4}}^{\mathrm{2}-}\right]=\left[{\mathrm{ssNa}}^{+}\right]\cdot ({\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}/\mathrm{Na}{)}_{\mathrm{seawater}},$$ where $\left[{\mathrm{Na}}^{+}\right]$ and $\left[\mathrm{Ca}\right]$ are the concentrations obtained, respectively, by IC and PIXE; $(\mathrm{Na}/\mathrm{Ca}{)}_{\mathrm{crust}}$ represents the mean $\mathrm{Na}/\mathrm{Ca}$ ratio in the Earth crust, assumed to be 0.56 according to literature (Henderson and Henderson, 2009); $(\mathrm{Ca}/\mathrm{Na}{)}_{\mathrm{seawater}}$, $(\mathrm{Mg}/\mathrm{Na}{)}_{\mathrm{seawater}}$, $(\mathrm{K}/\mathrm{Na}{)}_{\mathrm{seawater}}$, and $({\mathrm{SO}}_{\mathrm{4}}/\mathrm{Na}{)}_{\mathrm{seawater}}$ are the mean elemental ratios in bulk seawater and are set to 0.038, 0.119, 0.037, and 0.253, respectively (Henderson and Henderson, 2009).

The contribution of mineral dust ([mineral dust]) was calculated as sum of the metal oxides for every day based on stoichiometric ratios (e.g., Nava et al., 2012): $$\begin{array}{cc}\text{[mineral dust]}& =1.35\left[{\mathrm{nssNa}}^{+}\right]+1.66\left[\mathrm{nssMg}\right]+1.89\left[\mathrm{Al}\right]\\ & +2.14\left[\mathrm{Si}\right]+1.21\left[\mathrm{nssK}\right]+1.40\left[\mathrm{nssCa}\right]\\ & +1.67\left[\mathrm{Ti}\right]+1.43\left[\mathrm{Fe}\right],\end{array}$$ where “nss” stands for “non-sea salt” (see previous equations) and [Al], [Si], [Ti], [Fe] are the elemental concentrations of Al, Si, Ti, Fe as measured by PIXE; the numerical coefficients account for the weight of the oxygen atom in the main oxides of such elements in the average Earth crust.

According to this estimate, sea salt and mineral dust contribute to PM${}_{10}$ in Lampedusa with 8.3 and 5.8 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ (corresponding to 26 and 18 % of the PM${}_{10}$ mass) as average on the 2-year period, respectively. None of the components has a clear seasonality, and the temporal pattern of mineral dust shows several episodic peaks. Sea-salt levels are of the same order as observed in other coastal Mediterranean sites. Concerning mineral dust, the observed concentrations are in the range observed for the crustal component in Mediterranean background sites (Querol et al., 2009), as well as consistent with the mean African dust contributions observed in Mediterranean urban and suburban sites (Pey et al., 2013), due to the character prevalently of long-range transport of this component.

Despite the remoteness of the site, non-negligible secondary aerosol levels characterize the PM${}_{10}$ in Lampedusa. Nitrate in Lampedusa accounted for 2.1 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ as average on the 2-year period: this value appears to be slightly higher than in other Mediterranean Basin background sites (1.7–1.9 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$), in rural and urban background sites in Italy, and on the lower edge of those measured in central Europe and the UK (2–4 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$) (Querol et al., 2009). Sulfate levels (3.6 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ as average on the 2-year period, out of which 2.9 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ are nss-sulfate) are also in the range reported for central and southern Europe and UK rural background stations (2–4 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$), although lower than in background sites in the eastern Mediterranean Basin (5–6 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$), influenced by long-range transport episodes from eastern Europe (Querol et al., 2009).

MSA, a typical marker of biogenic activity, shows the typical seasonality with maxima during summer and minima in winter due to the seasonality in the emission of its precursor dimethyl sulfide (DMS; Becagli et al., 2013).

As concerns the carbonaceous fraction of aerosol, data are available only for a very limited time (24 samples averaging on 24 to 72 $\mathrm{h}$ in the period 16 May–15 August 2007): in such period, on average, TC and EC accounted for 1.9 and 0.4 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$, i.e., for 8.0 and 1.5 % of PM${}_{10}$ mass, respectively. Preliminary results on the 2011 campaign on 1-year-long data set estimate TC and EC levels as 9.6 and 0.1 % of PM${}_{10}$ mass, thus confirming the order of magnitude of carbonaceous aerosol levels in Lampedusa as measured during the short 2007 campaign. In particular, EC appears to be a negligible component of PM${}_{10}$ in Lampedusa, and carbonaceous aerosol appears to be mainly organic; taking into account for non-C atoms in particulate organic matter (POM) by multiplying OC for a factor 2.1 as suggested by Turpin and Lim (2001) for non-urban sites, POM mass appears to contribute for about 20 % of PM${}_{10}$ mass in Lampedusa.

The reconstructed mass, i.e., the sum of all the components determined by chemical analysis, corresponds to 61 % of the gravimetrical mass. When adding to this value the estimated 20 % of POM, the unaccounted mass usually attributed to water molecules is about 19 %, in agreement with the 10 to 20 % range reported in literature for coastal rural and Mediterranean background sites (Koçak et al., 2007; Pey et al., 2009; Querol et al., 2009). Nevertheless, some of the unaccounted mass may be partially attributed to carbonates, which were not assessed in this study.

PM${}_{10}$ sources identification by PMF

The PMF model was applied to the data set of mass concentration and chemical composition relative to the 562 samples collected in the years 2007–2008, considering 21 chemical species as measured by either PIXE (Mg, Al, Si, K, Ca, Ti, Mn, Fe, Cu, Br, Sr), ICP-AES ($V_{\mathrm{sol}}$, Ni${}_{\mathrm{sol}})$, or IC (${\mathrm{Na}}^{+}$, ${\mathrm{NH}}_{\mathrm{4}}^{-}$, ${\mathrm{Cl}}^{-}$, ${\mathrm{NO}}_{\mathrm{3}}^{-}$, ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$, oxalate, glycolate, MSA); all such variables are “good”, according to the signal-to-noise (S $/$ N) ratio criterion by Paatero and Hopke (2003). The soluble fraction of $V$ and Ni ($V_{\mathrm{sol}}$, Ni${}_{\mathrm{sol}})$ was used instead of their total fraction due to the lower DLs of ICP-AES compared to PIXE, and because it is a more representative marker for anthropogenic sources (Becagli et al., 2012). Missing data were less than 1 % of the total number of samples for all chemical species, except soluble elements (5 %) and Cu (9 %).

A seven-factor solution, with FPEAK value 0.1, was selected, on the basis of the $Q$ value, observed residual distributions, factor edges, physical sense of the source profiles, and temporal evolutions ($\mathbf{F}$ and $\mathbf{G}$ matrixes; Paatero, 1997); observed relative ratios and enrichment factors for the different sources are consistent with literature data, strengthening the results of the analysis. Briefly, the analysis allowed the identification of seven different sources which are identified as sea salt, mineral dust, biogenic emission aerosol, secondary nitrate particles, secondary sulfate particles, primary particulate ship emissions, and combustion emission aerosol (see the following sections). All together, these sources reconstruct the measured gravimetric mass within 4 %. The six-factor solution was excluded, as it did not separate the secondary nitrate source from the combustion emission one. The eight-factor solution was also discarded, as it was identifying an unlikely source related primarily to $\mathrm{Ca}$ and $\mathrm{Sr}$ (44 and 21 % explained variations, respectively): the hypothesis of facing a local dust resuspension source was discarded as the profile of this source did not contain any contributions from other crustal elements (such as $\mathrm{Ca}$, $\mathrm{Si}$, and $\mathrm{Al}$); further, the temporal profile of this factor is very similar to the mineral dust one (in particular, it shows the same peaks during Saharan dust advection episodes).

Profiles (fraction of the elemental mass concentration, blue columns) and explained variations (red squares, referring to the right $y$ axis) of the identified PM${}_{10}$ sources.

[Figure omitted. See PDF]

Temporal evolution of the PM${}_{10}$ contributions from the identified sources.

[Figure omitted. See PDF]

The seven sources are described in the following sections. Profiles are reported in Fig. 3: the left $y$ axis displays the fraction of the elemental mass concentration, reported in the graph as blue column, whereas the right $y$ axis refers to the explained variation of each element, depicted as red squares (the contribution of one element in all the profiles sum up to one, when the variation of the element is completely reconstructed by the model). Error bars on the fraction of the elemental mass concentration represent the computed errors as provided by the PMF2 model. These estimates take into account the uncertainties on the input data and the application of non-negativity criteria; nevertheless, they do not consider rotational ambiguity (Paatero et al., 2014), which was in any case investigated by systematically exploring solutions with different FPEAK values (see Supplement), according to a literature approach (Vecchi et al., 2008). The evolution of the daily contribution of each identified source is shown in Fig. 4.

Elemental ratios for the sea-salt source.

Ratio	Mg $/$ Na${}^{+}$	SO${}_{\mathrm{4}}$ $/$ Na${}^{+}$	Cl${}^{-}$ $/$ Na${}^{+}$	K $/$ Na${}^{+}$	Ca $/$ Na${}^{+}$	Br $/$ Na${}^{+}$
PMF profile	0.13	0.23	1.68	0.04	0.05	0.005
Henderson and Henderson (2009)	0.119	0.253	1.796	0.037	0.038	0.006

Enrichment factors for the mineral dust source.

EF(Mg)	EF(Si)	EF(K)	EF(Ca)	EF(Ti)	EF(Mn)	EF(Fe)	EF(Sr)
0.90	0.83	0.92	2.81	1.21	0.79	1.05	1.72

Sea salt

${\mathrm{Na}}^{+}$, ${\mathrm{Cl}}^{-}$, $\mathrm{Mg}$, $\mathrm{K}$, and $\mathrm{Br}$ are the characterizing elements for this factor (Fig. 3a); most of their mass concentration is explained by this source (explained variations: 64, 83, 48, 30, and 61 %, respectively). Some ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ is also present. The elemental ratios are in good agreement with those measured in marine water, suggesting a common marine origin for the species characterizing this factor. More in detail, the source profile elemental ratios are reported in Table 2 together with the sea-salt literature elemental ratio (Henderson and Henderson, 2009). Thus, this factor of the PMF is identified with the primary marine aerosol, sea-salt aerosol.

The temporal evolution shows no clear seasonality, as it can be expected for this source, since the sampling site is located directly on the coast and sea-salt aerosol is known to be strongly related to wind intensity (Kishcha et al., 2011).

HYSPLIT backward trajectories ending at Lampedusa at 50 $\mathrm{m}$ a.s.l. corresponding to the peaks in the PM${}_{10}$ contributions due to mineral dust source (see text for the details).

[Figure omitted. See PDF]

Mineral dust

Several crustal elements such as Al, Si, Ti, Mn, Fe, Sr, Ca, K, and Mg characterize this factor (Fig. 3b). This source reconstructs almost completely the mass concentration of some of these elements, such as Al, Si, and Ti (more than 80 % of their signal).

Enrichment factors (EFs) with respect to Al in this profile were calculated using the average Earth's soil composition reported by Mason (1966) and are reported in Table 3.

The values of these EFs, all close to 1 (with a moderate enrichment only for Ca and Sr which is usually associated with Ca in carbonates), strengthen the identification of this source as “mineral dust”.

The temporal evolution of this factor shows no clear seasonality, being dominated by many narrow peaks. Sixty-hour backward trajectories ending at Lampedusa at 50 $\mathrm{m}$ a.s.l. corresponding to the start, middle, and end time of the sampling interval were calculated for the peaks in PM${}_{10}$ concentrations due to mineral dust larger than the 95th percentile (about 32 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$), using the NOAA Air Resource Laboratory HYSPLIT transport model (Draxler and Rolph, 2012). The analysis showed that in about 86 % of the cases at least one of the three trajectories associated to the sampling overpassed northern Africa (see Fig. 5). This result should be read taking into account the uncertainties in the HYSPLIT trajectories due to the lack of meteorological measurements in the examined region, the low model horizontal resolution, and the possible rapid temporal variation of the air masses during the sampling interval. No relevant difference in the available crustal markers ratios was observed in correspondence of African dust transport episodes, suggesting a similar chemical composition for the Lampedusa soil and the African dust (limitedly to the measured elements) or a negligible impact of local dust. In fact, enrichments in $\mathrm{Ca}$ with respect to the average Earth's soil composition are reported for dust originating from different African areas, e.g., western Sahara and Moroccan Atlas, due to the high content of minerals such as calcite and dolomite (Avila et al., 1997; Coz et al., 2009). An enhancement of $\mathrm{Ca}$ amounts for dust originating from the Algeria–Tunisia sector was shown by Marconi et al. (2014).

The seasonality of African desert dust advection episodes in Lampedusa was discussed in a recent study (Marconi et al., 2014) based on a larger data set (June 2004–December 2010), in which ground level aerosol concentrations are compared with aerosol optical depth measurements routinely performed at the Station for Climate Observations of Lampedusa, providing information on the dust content in the entire air column above the sampling site. African desert dust transports often occur above the boundary layer (mainly in summer); in fact, despite the strong seasonality observed for the aerosol optical depth, at the ground level no evident seasonal trend was observed for PM${}_{10}$ or crustal markers, in agreement with the mineral dust temporal pattern obtained by PMF on the 2007–2008 data set. In particular, as shown by Marconi et al. (2014), while the aerosol optical depth and the frequency of occurrence of vertically integrated dust events peak in summer, a relative minimum of PM${}_{10}$ concentration and its frequency of occurrence is found in the months of June–July.

Biogenic emissions

This factor is mainly characterized by MSA and explains almost completely (87 %) its mass concentration. A small percentage (6 %) of the total mass of ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ is also accounted by this source and constitutes, however, the prevalent absolute mass contribution to this factor.

MSA is used as a tracer for biogenic sulphur emissions (from phytoplankton processes), as it is solely produced by oxidation of DMS. The evaluation of the ${\mathrm{nssSO}}_{\mathrm{4}}^{-}$ emitted by the biogenic source, (nss-${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$)${}_{\mathrm{bio}}$, suffers from the variability of the (nss-${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$)${}_{\mathrm{bio}}$ $/$ MSA emission ratio as this depends on the latitude, the NO${}_x$ levels, and the temperature (Seinfeld and Pandis, 1998; Bates et al., 1992); this is especially true for the Mediterranean basin, where only a maximum value for such ratio has been estimated in previous studies (in Crete) (Mihalopoulos et al., 1997; Kouvarakis and Mihalopoulos, 2002). The observed ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ $/$ MSA ratio in this source, 11.3, is only slightly higher than the aforementioned estimated maximum value (10) and it could be considered as representative for the (nss-${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$)${}_{\mathrm{bio}}$ $/$ MSA ratio for the biogenic emissions in the central Mediterranean basin. By using the value of 11.3 we found that in summer (JJA), when biogenic sulphur emission is at its maximum, 0.5 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ of ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$, corresponding to 11 % of total ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$, arise from a natural source. This result confirms the order of magnitude of the biogenic sulphur contribution during summer estimated as 17 % by Kouvarakis and Mihalopoulos (2002) by observations and modeling.

The strong seasonality observed for this factor, with minima during winter and maxima in summer, is also supporting its attribution to the biogenic emissions source. A recent study investigated the role of biogenic aerosols in the PM at Lampedusa (Becagli et al., 2013) with respect to the phytoplankton productivity, suggesting that the seasonal evolution of methanesulfonate, with a very evident summer maximum, is mainly driven by phytoplankton physiology and cell stress factors, such as high irradiance and shallow depth of the marine upper mixed layer.

Secondary nitrate

This factor is characterized by ${\mathrm{NO}}_{\mathrm{3}}^{-}$ and ${\mathrm{Na}}^{+}$, with the mass concentration of ${\mathrm{NO}}_{\mathrm{3}}^{-}$ being explained almost for the 80 % by it.

The ionic balance for this source is neutral, when taking into account ${\mathrm{NO}}_{\mathrm{3}}^{-}$ and the two cations ${\mathrm{Na}}^{+}$ and ${\mathrm{Mg}}^{\mathrm{2}+}$, as in a marine environment the neutralization of ${\mathrm{HNO}}_{\mathrm{3}}$ can occur to a large extent on sea-salt components (Bardouki et al., 2003; Metzger et al., 2006).

The temporal evolution shows a small seasonality, with somewhat larger values during summer. A similar seasonal behavior was observed by Querol et al. (2009) in eastern Mediterranean Basin (EMB) sites. The same study evidenced that in the EMB nitrate is mainly present in the coarse mode, chemically bound to alkaline ion species, while in the western Mediterranean Basin (WMB) it is largely in the fine mode, as it is most frequently present as ammonium nitrate. In Lampedusa, as in the EMB, the observed seasonality appears to be mainly regulated by the combination of nitrates with sea-salt cations (Metzger et al., 2006), while in the WMB nitrate is dominated by a strong opposite seasonality, with maxima during winter, due to both the instability of ammonium nitrate in warm environments and intense nitrate episodes in late winter.

Secondary sulfate

${\mathrm{NH}}_{\mathrm{4}}^{+}$, ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$, and oxalates are the characterizing elements for this source, with most of their signals explained by this factor (explained variations are 87, 54, and 45 %, respectively). The major absolute contribution to the profile is given by ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$; the ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ $/$ ${\mathrm{NH}}_{\mathrm{4}}^{+}$ ratio of 3.4 indicates that both the compounds $\left({\mathrm{NH}}_{\mathrm{4}}\right){\mathrm{HSO}}_{\mathrm{4}}$ and $({\mathrm{NH}}_{\mathrm{4}}{)}_{\mathrm{2}}{\mathrm{SO}}_{\mathrm{4}}$ are present.

The temporal evolution shows maxima during summer, when the marine boundary layer is more stable and the photochemistry leading to the production of secondary aerosols is enhanced. The same behavior was observed by Querol et al. (2009) for sulfate both at eastern and western Mediterranean sites. Further, it is worth noting that sulfate neutralization with ${\mathrm{NH}}_{\mathrm{4}}^{+}$ may leave very low amounts of ${\mathrm{NH}}_{\mathrm{3}}$ available for the neutralization of nitrate, which thus occurs mainly on sea-salt aerosol (see Sect. 3.2.4).

Primary particulate ship emissions

This factor explains most of the mass of $V_{\mathrm{sol}}$ and Ni${}_{\mathrm{sol}}$, even if the predominant mass contributions are from ${\mathrm{Na}}^{+}$ and ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$. $V$ and Ni have been clearly recognized as markers of heavy oil combustion, and characteristic $V$ $/$ Ni ratios ranging from 2.5 to 3.5 have been reported for ships emissions (Mazzei et al., 2008; Viana et al., 2009; Pandolfi et al., 2011). Further, both $V$ and Ni were found to have roughly the same solubility (80 and 77 %, respectively) when originated by heavy oil combustion, in the mild extraction condition (${\mathrm{HNO}}_{\mathrm{3}}$ – pH 1.5) used for ICP-AES extraction (Becagli et al., 2012). The $V_{\mathrm{sol}}/{\mathrm{Ni}}_{\mathrm{sol}}$ ratio for this source is 3.40, in good agreement with the aforementioned literature data. Moreover, the interpretation of this factor as “primary ship emissions” is strengthened by the fact that the observed ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ $/$ $V_{\mathrm{sol}}$ ratio (6.2 w/w; as $V$ has been found to have a 80 % solubility in ships exhausts, such value corresponds to a ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ $/$ $V$ ratio of 5.0) is of the same order of magnitude as the ones reported by Agrawal et al. (2008) for the particulate matter freshly emitted from the main engine of an ocean going container vessel (${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ $/$ $V$ ranging from 8.9 to 11.9 depending on the engine operating conditions for the considered vessel). Much higher ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ $/$ $V$ values are reported when secondary aerosols (mainly ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$) are formed from the oxidation of the gaseous precursors (e.g., ${\mathrm{SO}}_{\mathrm{2}}$) (Becagli et al., 2012).

Temporal evolution of the “combustion emissions” and “mineral dust” sources; examples of trajectories from fires are evidenced in red, together with an example of advection from the Saharan region (green).

[Figure omitted. See PDF]

A clear seasonal evolution, with maxima during summer, is evident for this source; this is in agreement with the temporal evolution of ship emissions markers (Ni${}_{\mathrm{sol}}$, $V_{\mathrm{sol}}$, nssSO${}_{\mathrm{4}}^{\mathrm{2}-})$ observed by Becagli et al. (2012) on a larger data set of samples collected at the same Lampedusa site (years 2004–2008). The authors explained the temporal evolution of these markers as due to three possible causes: (a) an increased photochemical activity in summer leading to a faster production of secondary aerosols; (b) a stronger stability of the marine boundary layer during summer, leading to an increase of their concentrations at the ground level; (c) prevalent winds from the Strait of Sicily (one of the main ship paths) during summer. Due to the primary origin of this factor, such seasonal behavior appears to be more probably linked to the two latter phenomena.

Combustion emissions

Several compounds, as $\mathrm{K}$, $\mathrm{Cu}$, ${\mathrm{NO}}_{\mathrm{3}}^{-}$, ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$, and short carboxylic acids such as oxalates and glycolates, whose signal is mostly explained by this source, characterize this factor. $\mathrm{K}$ and ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ are commonly associated with biomass burning, and short carboxylic acids have been recently associated with coal/waste/biomass burning (Wang et al., 2007). Enhancements of $\mathrm{Cu}$ in the aerosol produced by wood combustion have also been reported by several studies (Dall'Osto et al., 2013; Nava et al., 2015). Therefore this factor has been interpreted as a generic combustion source, which includes biomass burning.

The temporal evolution (Fig. 4) of this factor shows no clear seasonality, even if important contributions during summer may be observed and may be linked to the more frequent occurrence of fires in the Mediterranean region. During spring and fall the contributions from this source are generally low, while several high contribution days may be observed during winter. Figure 6 shows the combined evolution of the desert dust and combustion emissions contributions to PM${}_{10}$ and some selected cases with air-mass trajectories and satellite observations of active fires. Most of the winter cases characterized by elevated contributions of combustion sources display also a high contribution from the mineral dust source (see Fig. 6). The analysis of the back trajectories shows that these days were mostly interested by African dust advection episodes. Several studies report that dust particles plumes from the Saharan area also export biomass burning particles, mainly during the dry winter season, as well as emissions from refineries (Tesche et al., 2011; Rodríguez et al., 2011). Further, small fires, often occurring in biomes as croplands, wooded savannas, and tropical forests, are usually below the detection limit of the current generation of surface reflectance/thermal imagery instruments (Randerson et al., 2012). Therefore, winter contributions from this source are likely to be mainly due to small-scale biomass burning activity in northern Africa; nonetheless, some winter peaks of this source appear also connected with fires and/or biomass burning activity in central-eastern Europe (e.g., 13–15 March 2007, Fig. 6).

Summer peaks have been clearly connected by back-trajectory analysis to fires mainly in South Italy and Sicily, thus supporting the interpretation of this factor as a combustion source, with a main contribution from biomass burning.

Comparison between EC and OC data and PMF estimates of the contributions of combustion emissions and primary particulate ship emissions in the period of EC/OC data availability. Results for EC and OC are reported with solid lines (black and red, respectively), referring to the secondary axis (empty rounds report the single sample result attributing it to the central sampling day).

[Figure omitted. See PDF]

Seasonal relative contributions to PM${}_{10}$ concentration (average over the years 2007–2008). Factors may sum to more than 100 % due to number rounding.

[Figure omitted. See PDF]

Seasonal absolute contributions to PM${}_{10}$ concentration (average over the years 2007–2008).

[Figure omitted. See PDF]

Combustion emission source identification: additional data

EC and OC are produced in combustion processes of both fossil fuels and biomasses, while OC may have also a biogenic origin. Therefore, EC and, to a smaller extent, OC are good markers of combustion sources, while additional information from other chemical tracers (such as $\mathrm{K}$, levoglucosan, glycolates, or radiocarbon; Bernardoni et al., 2013) may help for the assessment of the burnt fuel (biomass or fossil). Despite OC is a major aerosol component and EC is an important combustion tracer, a recent study has demonstrated that the lack of data for these species does not significantly alter PMF results provided that a complete elemental and ionic speciation is available, so that other markers tracing combustion emissions are available (Lucarelli et al., 2015). Nevertheless, a comparison of the results for the combustion source as derived by the PMF analysis with the available EC/OC data was performed. As previously reported, additional data on the carbonaceous fractions in PM${}_{10}$ are available for limited periods in the years 2007–2008. In particular, 24 samplings of PM${}_{10}$ on quartz filters for EC/OC analysis were performed from 16 May to 15 August 2007 with integration times ranging from 24 to 72 $\mathrm{h}$. The starting time of the sampling was set in the morning/early afternoon (between 08:00 and 14:00) due to manual filter change operations. In Fig. 7, the EC and OC contents are plotted together with the estimates of the contributions from the combustion emission source as obtained by PMF. As may be seen, EC and OC are highly correlated with the combustion emission contribution estimated by the PMF, thus strengthening the attribution of the seventh PMF factor to a combustion source. Small discrepancies, as the one occurring at the beginning of July, may be attributed to episodic important contributions to EC and OC from heavy oil combustion, that is from the primary particulate ship emission source, also reported in Fig. 7.

The availability of a complete EC/OC data set for the future campaigns will add valuable information on the sources and contributions of the carbonaceous aerosol component.

Relative annual contributions to the PM${}_{10}$ concentration (average over the years 2007–2008).

[Figure omitted. See PDF]

PM${}_{10}$ source apportionment

Figures 8 and 9 report the relative and absolute seasonal contributions of the seven identified sources (winter – December, January, February; spring – March, April, May; summer – June, July, August; fall – September, October, November). During all the seasons, natural sources (sea salt, mineral dust, and biogenic emissions) give the largest contribution to PM${}_{10}$, accounting for a minimum of 62 % in spring to a maximum of 76 % during fall. The maximum relative and absolute contribution of sea salt is observed in winter, while relative contributions during the other seasons are comparable. This is consistent with a higher wind velocity during the winter months (7.4 $\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}$, compared to 5.6 $\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}$ in summer, 6.5 $\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}$ in spring, and 6.1 $\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}$ in fall during 2007 and 2008).

The mineral dust maximum relative contribution is found during fall (37 %) and is around 20 % in the other seasons. As expected, the maximum relative contribution of biogenic emissions is during summer and is negligible in winter.

Among the anthropogenic sources, primary particulate ship emissions and secondary sulfate display a clear and similar seasonality, with maxima in summer. No significant dependence on the season is observed for the relative contributions of secondary nitrate and combustion emissions. It is worth noting that the not explained mass is around 10 % across the seasons.

The annual average relative contributions are reported with their uncertainties in Fig. 10: the main contribution to aerosol comes from sea salt, accounting for 40 % of the mass; the second contribution is from mineral dust (25 %). Taking into account also the biogenic emissions, accounting for 4 % of the mass, the contribution of natural sources to the PM${}_{10}$ in Lampedusa is estimated to be around 68 % on average. Concerning secondary nitrates and sulfates, they are estimated to contribute by 9 and 11 % to the total PM${}_{10}$, respectively. Finally, primary particulate ship emissions have been observed to give an average 4 % contribution, while 5 % of PM${}_{10}$ was found to be due to combustion emissions. Uncertainties on the contributions were estimated propagating the uncertainties obtained by the PMF model on the $\mathbf{G}$ matrix.

In absolute terms, the following contributions were estimated: sea salt, $12.9\pm 1.1$ $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$; mineral dust, $7.8\pm 0.5$ $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$; biogenic emissions, $1.1\pm 0.2$ $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{\mathrm{3}}$; secondary nitrate, $2.8\pm 0.3$ $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$; secondary sulfate, $3.4\pm 0.4$ $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{\mathrm{3}}$; primary particulate ship emissions, $1.2\pm 0.2$ $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$; combustion emissions, $1.5\pm 0.3$ $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$.

Correlation plot between the contributions from secondary sulfate and from primary particulate emitted by ships as obtained by the PMF analysis.

[Figure omitted. See PDF]

Evaluation of the total contribution from ship emissions to PM${}_{10}$ in Lampedusa

An estimate of the contribution of the shipping emissions to the secondary sulfate formation was performed following previous studies (Kim and Hopke, 2008; Viana et al., 2009), based on the observed correlation between the daily contributions from the primary particulate ship emission source and those from the secondary sulfate source as assessed by PMF analysis. The contributions from these two sources are plotted one against the other in Fig. 11: the solid line represents the minimum amount of secondary sulfate that is expected to be associated to the observed primary particulate from ship emissions. This line was graphically drawn aiming at excluding outliers and its slope indicates that, at the receptor site, at least 0.9 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ of secondary sulfate are expected per unit of detected primary particulate from ship emissions. Therefore, primary particles contribute by 50 % or less to the total PM${}_{10}$ originating from ship emissions. Kim and Hopke (2008) reported about 0.8 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ of secondary sulfate per 1 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ of primary oil combustion particles in PM${}_{2.5}$ in three sites in Seattle (WA, USA). A larger ratio (about 2.1) was found by Viana et al. (2009) for PM${}_{10}$ in Melilla (Spain). Both oil combustion particles and secondary aerosols are found mainly in the fine aerosol fraction; therefore the ratio between the two components (secondary sulfate and primary particulate) of the ship emissions source is likely to be roughly the same for PM${}_{2.5}$ and PM${}_{10}$. It must be pointed out, however, that we retrieved a minimum value, and the actual contribution of secondary sulfate depends on many factors and, in many cases, may be much higher than the limit.

Taking into account this minimum secondary sulfate contribution, we derive a minimum ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ $/$ $V_{\mathrm{sol}}$ ratio of about 190 for ship emissions (primary particulate and secondary sulfate). It is worth noting that a ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ $/$ $V_{\mathrm{sol}}$ ratio $=200$ was proposed as characteristic lower ratio for ship emissions from a previous study performed at Lampedusa (Becagli et al., 2012) on a more extended data set (June 2004–December 2008).

Further, ship emissions on the whole (primary particulate and secondary sulfate) are estimated to contribute to PM${}_{10}$ in Lampedusa by approximately 7 % (about 4 % primary particles and 3 % secondary sulfate). These values confirm the order of magnitude of the estimated impact of ship emissions in European coastal areas, which is reported to be around 1–7 % (EEA, 2013).

Comparison of contribution estimates by PMF and speciation

The PMF estimates of the contributions from sea salt and mineral dust were compared with those obtained according to the average composition of seawater and Earth's soil (Sect. 3.2).

The mineral dust estimate from the PMF analysis is about 30 % higher than the crustal content calculated as the metal oxides sum. It must be pointed out that no information on carbonates is available and that the used analytical techniques are not sensitive to the oxygen content; thus the estimated mineral dust mass is based on the assumption of an oxygen fraction according to the prevalent oxides of the crustal elements. This assumption is obviously an approximation as the real mineral dust composition may differ from it; as an example, this approach may underestimate the mineral dust contribution in presence of a significant amount of carbonates. Nevertheless, PMF may slightly overestimate the source contribution as a consequence of some profile interferences (Mazzei et al., 2008): in this case, the PMF retrieval associated a small amount of ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ and ${\mathrm{Cl}}^{-}$ to the crustal dust profile, probably due to their co-linearity in case of long-range transport episodes.

Coming to sea salt, the stoichiometric approach gives an estimate of the average contribution of 8.3 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$, to be compared with the PMF estimated contribution for the sea-salt source of 12.9 $\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$. Since all the sea-salt elements are measured, this discrepancy is likely to be due to other compounds not visible with the applied analytical techniques, such as water or, to a lesser extent, organic compounds. In fact, although gravimetric measurements are performed in a controlled atmosphere (50 % relative humidity, 25 ${}^{\circ }$C), some water may endure and it is likely to be bound to hygroscopic aerosol components such as sea salt; this is in agreement with recent studies that have shown water exists in ambient aerosol also at low relative humidity (Plinis et al., 2014; Engelhart et al., 2011). Therefore, taking into account for water, the PMF contribution from the sea-salt source may be possibly split in two contributions, namely “dry sea salt” and “${\mathrm{H}}_{\mathrm{2}}\mathrm{O}$”, accounting for about 26 and 14 % of the PM${}_{10}$ mass in Lampedusa, respectively. This estimate lays in the 10 to 20 % range reported in literature for unaccounted mass in Mediterranean background/rural sites (Koçak et al., 2007; Pey et al., 2009; Querol et al., 2009), which is generally attributed to water molecules of moisture.

In any case, it has to be pointed out that some variability in the chemical profiles of the sea-salt factors across the Mediterranean is reported in literature, due to possible reactions occurring in the atmosphere: for Mediterranean coastal sites, some papers report, as for this work, a sea-spray composition in agreement with the marine water one (e.g., Koçak et al., 2007; Amato et al., 2015); other papers report a ${\mathrm{Cl}}^{-}$ depletion (i.e., ${\mathrm{Cl}}^{-}$ $/$ ${\mathrm{Na}}^{+}$ ratios lower than in marine water) at coastal sites (Koulouri et al., 2008) as well as in open Mediterranean sea (Schembari et al., 2014).

Conclusions

On average, each of the sources “primary particulate ship emissions”, “biogenic emissions”, and “combustion emissions” contribute about 5 % to the total PM${}_{10}$ mass, “secondary nitrate” and “secondary sulfate” account for about 10 % each, and the “mineral dust” contribution is around 25 %, while “sea salt” constitutes about 40 % of the PM${}_{10}$. According to a stoichiometric approach, the latter may possibly split in two contributions, namely “dry sea salt” and “${\mathrm{H}}_{\mathrm{2}}\mathrm{O}$”, accounting for roughly 25 and 15 % of the PM${}_{10}$ mass in Lampedusa, respectively. Further, about one-third of the total contribution of secondary sulfate appears to originate from ship emissions. Large variations in absolute and relative contributions are found and appear to depend on the season and on transport episodes.

Improvements to this source apportionment will be possible taking into account for the carbonaceous fraction (no extensive data set is available for it for the considered years, 2007 and 2008).

The Supplement related to this article is available online at doi:10.5194/acp-15-13939-2015-supplement.

Acknowledgements

This work has been supported by the SNUMMASS and NextData projects, and the RITMARE Flagship Project funded by Italian Ministry of University and Research. The measurements presented here are preparatory to Chemistry-Aerosol Mediterranean Experiment project (ChArMEx), which is the atmospheric component of the French multidisciplinary program MISTRAL. Edited by: X. Querol