Fine particulate matter (PM) in the atmosphere is of concern due to its effects on health, climate, ecosystems and biological cycles, and visibility. These effects are especially important in the Mediterranean region. The western Mediterranean basin experiences high gaseous pollution levels originating from Europe in particular during summer, when photochemical activity is at its maximum. Furthermore, the western Mediterranean basin is impacted by various natural sources: Saharan dust, intense biogenic emissions in summer, oceanic emissions and biomass burning, all of which emit gases (e.g., volatile organic compounds (VOC), nitrogen oxides (NOx)) and/or primary particles . During the TRAQA 2012 and SAFMED 2013 measurement campaigns, observed that aerosols in the western Mediterranean basin are strongly impacted by dust outflows and continental pollution. A large part of this continental pollution is secondary, i.e., it is formed in the atmosphere by chemical reactions e.g.,. These reactions involve compounds, which may be emitted from different sources (e.g., biogenic and anthropogenic). Using measurements and/or modeling, several studies have shown that as much as 70 to 80 % of organic aerosol in summer in the western Mediterranean region is secondary and from contemporary origins .

Air quality models are powerful tools to simulate and predict the atmospheric chemical composition and the properties of aerosols at regional scales. In spite of the tremendous efforts made recently, the sources and transformation mechanisms of atmospheric aerosols are not fully characterized nor fully understood. For organic aerosols, modeling difficulties partly lie in the representation of volatile and semi-volatile organic precursors, which can only take a limited number of compounds or classes of compounds into account. Difficulties in modeling atmospheric particles are strongly linked to uncertainties in meteorology and emissions . For example, turbulent vertical mixing affects the dilution and chemical processing of aerosols and their precursors , clouds affect aerosol chemistry and size distribution and photochemistry , and precipitation controls wet deposition processes . Over the Mediterranean region, uncertainties due to meteorology and transport may strongly impact pollutant concentrations. This is due to the fact that the basin is influenced by pollution transported from different regions, such as dust from Algeria, Tunisia and Morocco as well as both biogenic and anthropogenic species from Europe . and have shown that although organic aerosol concentrations at a remote marine site in the western Mediterranean are mostly of biogenic origin, they are strongly influenced by air masses transported from the continent and by maritime shipping emissions.

In addition to the meteorological uncertainties, uncertainties in emission inventories are also important. There are uncertainties in biogenic emissions , as well as in anthropogenic emission inventories. For anthropogenic emissions, uncertainties concern not only the emissions themselves, but also the pollutants that are to be considered in the inventory and the spatial and temporal distributions of the emissions. For example, intermediate and semi-volatile organic compounds are missing from emission inventories, even though they may strongly affect the formation of organic aerosols . The spatial distribution of ships and harbor traffic differs depending on emission inventories; however, over the Mediterranean Sea, ships and harbor traffic emissions may strongly affect the formation of particles. found that the minimum ship emission contributions to PM10 were 11 % at Lampedusa Island, and 8 % at Capo Granitola on the southern coast of Sicily. showed that ship emissions in the Mediterranean may contribute up to 60 % of sulfate concentrations, as SO2 is a major pollutant emitted from maritime transport. However, in comparison to on-road vehicles, ship emissions are still poorly characterized . Furthermore, the multiplicity of Mediterranean pollution sources and their interactions makes it difficult to quantify ship contributions to aerosol concentrations.

Seas and oceans are a significant source of sea-spray aerosols (SSA), which strongly affect the formation of cloud condensation nuclei and particle concentrations. However, according to , sea-spray aerosols (SSA) have one of the largest uncertainties among all emissions. The modeling of sea-salt emissions is based on empirical or semi-empirical formulas. There is a tremendous amount of parameterization of the SSA emission fluxes . The SSA emission parameterization of is commonly used to model sea-salt emissions of coarse particles e.g.,. However, the strong non-linearity of the source function versus wind speed (power law with an exponent of 3.41) may lead to an overestimation of emissions at high-speed regimes, as suggested by and . Many studies have shown that wind speed is the dominant influence on sea-salt emissions . However, other parameterizations use different power laws with different exponents for the wind speed (e.g., 2.07 for ) and have introduced other parameters like sea-surface temperature and water salinity . Although the influence of marine emissions on primary organic aerosols is low for the Mediterranean , their influence on inorganic aerosols is not .

The aim of this work is to evaluate some of the processes that strongly affect inorganic and organic aerosol concentrations in the western Mediterranean in summer (transport and emissions), and to establish how the data/parameterizations commonly used in air quality models affect the concentrations. To that end, sensitivity studies relative to transport (meteorology) and emissions (anthropogenic and sea salt) are performed using the Polyphemus air quality model. The model results are then compared to measurements performed at the remote marine Ersa super-site (Cap Corsica, France) during the summer campaigns of 2012 and 2013, and to airborne measurements performed above the Western Mediterranean Sea in summer (July) 2014.

This paper is structured as follows. The Polyphemus air quality model setup is first described for the different input datasets/parameterizations used, as well as the measurements. Second, the meteorological fields used as input for the air quality model are evaluated. Third, the model is evaluated by comparisons to the measurements and comparisons of the sensitivities studies to meteorology, sea-salt emission parameterizations and anthropogenic emissions are performed to determine the main aerosol sources and sensitivities.

In order to simulate aerosol formation over the western Mediterranean, the Polair3d/Polyphemus air quality model is used, with the setup described in and summarized here. For parameters/parameterizations that are particularly related to uncertainties (anthropogenic emissions, meteorology, sea-salt emissions and modeling of condensation/evaporation), the alternative parameters/parameterizations that are used in the sensitivity studies are also detailed for emissions and meteorology. For computational reasons, alternative parameterizations for the modeling of condensation/evaporation are only used in the comparisons to airborne measurements in Sect.  (where they are also detailed).

Aerosol phenomenology on the Corsica Cape is influenced by diverse meteorological situations as well as transport of pollutants from a number of sources. Therefore, it is crucial to estimate the input meteorological data used in the air quality model as accurately as possible. The four meteorological datasets (ECMWF, WRF-Lon-Lat, WRF-Lambert and WRF-Lambert-Obsgrid) are compared to observations of air temperature and wind at Ersa in Fig.  for the summer campaign periods of 2012 and in Fig.  for the summer 2013 (Appendix ).

The observed and simulated temperature, wind speed, wind direction and relative humidity at Ersa during these summers, the statistical scores defined in Table  of Appendix  and the comparison of the four model results to measurements (hourly time series) are shown in Tables –, respectively.

As mentioned in the 2007 EPA report, proposed benchmarks for temperature (mean bias (MB) within ±0.5 K and a gross error (GE) of 2.0 K), wind speed (MB within ±0.5 m s-1 and a RMSE < 2 m s-1) and wind direction (MB within ±10∘ and a GE < 30∘). suggested an alternative set of benchmarks for temperature (MB within ±1.0 K and a GE < 3.0 K).

The four meteorological simulations reproduce the ground temperature measured at Ersa well. Whilst only the ECMWF temperature in 2012 verifies the US EPA criteria, all simulations verify the criterion from for the GE. Statistically, the correlation to temperature measurements is high (between about 54 and 96 % for all models), and the root mean square error (RMSE) is low (below 3.4 K). The best model differs depending on the year: the correlation of ECMWF to measurements is the highest (96 %) and the RMSE the lowest (1.5 K) in 2012, but in 2013, the correlation of ECMWF is the lowest (70 %) and its RMSE the highest (3.2 K). The mean fractional biases and errors (MFB and MBE) of the simulated temperatures are almost zero.

For wind speed, ECMWF systematically leads to better statistics than WRF, despite the fine horizontal resolution of WRF (0.125∘ × 0.125∘). ECMWF agrees best with the measurements, with the highest correlation (between 69 and 87 %) and the lowest errors (MFE is between 33 and 47 %). It also verifies the US EPA criteria for both the 2012 and 2013 summers. WRF-Lon-Lat also performs well with correlations between 60 and 65 % and MFEs between 47 and 64 %. WRF-Lambert and WRF-Lambert-Obsgrid have poorer statistics with negative correlations and MFEs between 71 and 74 %.

The average wind direction is quite similar for the 2012 and 2013 summers (202 and 186∘, respectively). The mean wind direction is best represented by ECMWF for the 2013 summer, and WRF-Lon-Lat for the 2012 summer. However, the modeled wind speed does not respect the US EPA criteria. Errors are higher with the two models using the Lambert projection, which tend to underestimate the wind direction angle. For relative humidity, the observed mean relative humidity is 0.65 in 2012 and 0.70 in 2013. It is relatively well reproduced by the models (between 0.70 and 0.77 in 2012 and between 0.69 and 0.78 in 2013). All models perform well with a MFE below 32 % and a MFB below 18 %. WRF-Lon-Lat leads to the best statistics in 2012 and WRF-Lambert-Obsgrid leads to the best statistics in 2013.

As ECMWF and WRF-Lon-Lat show better overall performance than the other two models (Tables –), they are used for the meteorological sensitivity study.

The model performances presented above compare well to other studies . In this study, for ECMWF and WRF-Lon-Lat during the summers of 2012 and 2013, the RMSE ranges between 1.5 and 3.2 K for temperature, between 1.3 and 3.9 m s-1 for wind speed and between 58 and 118∘ for wind direction. At Ersa, for the summer 2013 (not exactly the same period – 10 July to 5 August 2013), found a RMSE between 1.5 and 2.3 K for temperature, between 1.6 and 1.9 m s-1 for wind speed and between 92 and 117∘ for wind direction using the mesoscale WRF model. Moreover, reported a RMSE ranging between 1 and 4 K for temperature, and 0.6 to 3.0 m s-1 for wind speed over Greater Paris during May 2005 using the WRF model with a longitude–latitude map projection.

This section focuses on the evaluation of the reference simulation (S1) against aerosol measurements (PM10, PM1, OM1 and inorganic aerosols (IA) species), in addition to the factors controlling simulated aerosol concentrations (meteorology, sea-salt and anthropogenic emissions). This evaluation is performed against ground-based measurements during the 2012 and 2013 summers, and against airborne measurements from the ATR-42 flight on 10 July 2014. The criteria of are used to evaluate the model-to-measurement comparisons. The performance criterion is verified if |MFB| ≤ 60 % and MFE ≤ 75 % (MFB and MFE stand for the respective mean fractional bias and the mean fractional error and are defined in Table  of Appendix ), while the goal criterion is verified if |MFB| ≤ 30 % and MFE ≤ 50 %. To evaluate the sensitivity of the modeled concentrations to input data, the different simulations summarized in Table  are compared to the reference simulation S1 by computing the normalized root mean square error (RMSE of the concentration differences between a simulation and S1, divided by the mean concentration of S1).

Section  was dedicated to explaining how the simulated concentrations of particles at Ersa are influenced by the different input data used (meteorology, sea salt and anthropogenic emissions) and the modeling hypotheses. This section generalizes the sensitivity study of Sect.  by investigating how the concentrations over the Mediterranean domain are influenced by the input data.

Figure  of Appendix  shows maps of the concentrations of PM10, OM1, sulfate and other secondary inorganic aerosols (nitrate, ammonium and chloride)over the Mediterranean domain from simulation S1 during the 2013 summer. The highest PM10 concentrations correspond to high OM1, sulfate or ammonium, nitrate and chloride concentrations. OM1 concentrations are high near locations with high biogenic emissions such as Italy and Corsica (Fig.  of Appendix ). Sulfate concentrations are particularly high over the Mediterranean Sea, near main shipping routes (Fig. ). Ammonium nitrate concentrations are high in places of high anthropogenic emissions, such as the north of Italy, as well as in major cities. Hereafter, the term VIA (volatile inorganic aerosol) is used to refer to chloride, ammonium and nitrate aerosols.

Figure  of Appendix  shows maps of the relative difference of the concentrations of PM10, OM1, sulfate and VIA between S2 and S1 (sensitivity to meteorology). VIA concentrations show the highest sensitivity to meteorology, with relative concentration differences between S2 and S1 ranging between -90 and -60 % locally over Italy. Sulfate shows the lowest sensitivity with relative concentration differences mostly ranging between -20 and 20 %. The larger influence of meteorology on VIA than on sulfate concentrations is partly explained by the influence of temperature on the partitioning of VIA between the gas and particle phases, as VIA is highly semi-volatile. OM1 concentrations are quite sensitive to meteorology over the whole Mediterranean domain, with relative concentration differences mostly between -60 and -20 %, especially near regions where the biogenic emissions are the highest. The regions of the highest OM1 concentrations also correspond to the areas where VIA concentrations are the most sensitive to meteorology. By influencing biogenic emissions, meteorology influences the formation of organics (OM1) and in turn the formation of VIA by the formation of organic nitrate. The influence of meteorology on sulfate concentrations is limited in this study, because the formation of organosulfates is not modeled in our simulations.

Figure  of Appendix  shows maps of the relative differences of the concentrations of PM10, OM1, sulfate and VIA between S3 and S1 (sensitivity to sea-salt emissions).

As sulfate is assumed to comprise only 4 % of sea-salt emissions (Sect. ), the influence of sea-salt emissions on sulfate concentrations at Ersa is low (the relative concentration difference is between 0 and 20 %). The effect is stronger over the western part of the Mediterranean domain (with relative concentration differences between S3 and S1 of between 20 and 60 %). Chloride concentrations are also strongly influenced by sea-salt emissions, as it is directly emitted (it is assumed to make up 25 % of sea-salt emissions). Furthermore, nitrate and ammonium concentrations are strongly impacted by sea-salt emissions, due to thermodynamic exchanges between the gas and particle phases of chloride, nitrate and ammonium.

The influence of sea-salt emissions on OM1 concentrations is also important, but it is less important than VIA (the relative concentration differences of VIA are between 90 and 180 %) over the western Mediterranean part of the domain, compared to between 20 and 60 % for OM1 and between 40 and 60 % for sulfate. The increase of OM1 concentrations when sea-salt emissions are high is due to the hydrophilic organic compounds in OM1, which are absorbed onto inorganic concentrations. The organic concentrations originating from sea-salt emissions are very low, as discussed in ; therefore, they are not taken into account here.

Figure  of Appendix  shows maps of the relative difference of the concentrations of PM10, OM1, sulfate and VIA between S4 and S1 (sensitivity to anthropogenic emissions). Sensitivities to sulfate and VIA concentrations are more spatially localized than sensitivities to OM1 concentrations, and are higher, with relative concentration differences between S4 and S1 of between -40 and 20 % for OM1 and between -40 and 60 % for VIA. Sulfate concentrations are strongly sensitive to anthropogenic emissions near main shipping routes, with negative (S4 - S1) concentrations between -60 and -40 %. This is due to the fact that shipping routes are not well represented in the EMEP emission inventory (simulation S4). For VIA concentrations, the influence of anthropogenic emissions can either be negative or positive (increase or decrease of concentrations); this is owing to the different spatial distributions of the two emission inventories, which directly affect nitrate formation.

This work presents a sensitivity study to different input data and model parameterizations to better understand aerosol sources over the Mediterranean and the parameters influencing the aerosol concentrations. Aerosol sources are different depending on aerosol chemical compounds. Comparisons to observations are performed at the Ersa station to estimate how realistic the concentrations simulated with the different parameters (meteorological fields, anthropogenic and marine emissions, intermediate/semi-volatile organic compounds (I/S-VOC) emissions and different options for condensation/evaporation modeling) are. For most pollutants, the best model performance is obtained when the meteorological fields that represent the best wind direction are used together with the emission inventory with the most accurate spatial description of ship emissions (EDGAR-HTAP).

Using ECMWF and WRF to model the meteorological fields, secondary pollutants (inorganics and organics) show a high sensitivity to meteorology. This highlights the importance of accurate meteorological modeling and the potential strong influence of climate change on the concentrations of these secondary pollutants.

The influence of meteorology on concentrations is due to its impact on sea salt and biogenic emissions, which directly influence the formation of ammonium, nitrate, chloride and OM; furthermore, temperature, humidity and radiation influence secondary aerosol formation. Sulfate is less sensitive to meteorology than volatile inorganic aerosols (VIA), because it is not volatile. However, this low sensitivity may change if the formation of organosulfates are modeled (not carried out in this study). Both inorganic and organic concentrations are highly sensitive to sea-salt emissions, although great discrepancies exist between different published parameterizations. The commonly used Monahan parameterization of sea-salt emissions leads to an overestimation of all particulate concentrations, especially sodium concentrations. A parameterization with a lower exponent in the wind speed power law is chosen to model sea-salt emissions and leads to better model performance. The overestimation of the modeled sea-salt concentrations using Monahan parameterization has an incidence on the overestimation of the modeled concentrations of inorganic compounds such as nitrate, which replaces chloride in the particles when the thermodynamic equilibrium approach is used to model condensation/evaporation. This assumption (the thermodynamic equilibrium approach) was shown not to be accurate at Ersa and over the Mediterranean Sea. At Ersa, the gas / particle ratio was too high for nitrate and too low for chloride if the thermodynamic equilibrium approach was used, as the exchange between the gas and particle phases was dynamic not instantaneous. This dynamic exchange is strongly influenced by the particle composition, and comparisons to measurements over the Mediterranean Sea suggest that sea-salt particles are not mixed with background (transported) particles. Overall, secondary pollutants such as nitrate, ammonium and chloride in the particle-phase are strongly influenced by the gas / particle phase partitioning, as a high percentage of their concentrations are in the gas phase. This underlines the need to develop aerosol models able to accurately represent this gas-phase partitioning.

Sulfate primarily originates from maritime traffic. Shipping emissions lead to the formation of oxidants that in turn enhance the formation of biogenic aerosols, with the potential formation of organic nitrate and organosulfates. Organics are mostly from biogenic origins during summer. Even if the contribution of sea-salt emissions to organic concentrations is low, organic concentrations are strongly influenced by sea-salt emissions because they partition between the gas and particle phases and they are hydrophilic. This underlines the need to better characterize the properties (affinity with water) of secondary organic aerosols. The emissions of I/S-VOC played a limited role in OM1 concentrations during the 2013 summer, suggesting that the influence of ship emissions on OM1 is mostly due to anthropogenic VOC precursors (aromatics) and NOx emissions. These substances lead to the formation of oxidants that may oxidize biogenic aerosol precursors (and form organic nitrate, for example).