A multimodel evaluation of the potential impact of shipping on particle species in the Mediterranean Sea

Shipping contributes significantly to air pollutant emissions and atmospheric particulate matter (PM) concentrations. At the same time, worldwide maritime transport volumes are expected to continue to rise in the future. The Mediterranean Sea is a major short-sea shipping route within Europe and is the main shipping route between Europe and East Asia. As a result, it is a heavily trafficked shipping area, and air quality monitoring stations in numerous cities along the Mediterranean coast have detected high levels of air pollutants originating from shipping emissions.

The current study is a part of the EU Horizon 2020 project SCIPPER (Shipping Contributions to Inland Pollution – Push for the Enforcement of Regulations), which intends to investigate how existing restrictions on shipping-related emissions to the atmosphere ensure compliance with legislation. To demonstrate the impact of ships on relatively large scales, the potential shipping impacts on various air pollutants can be simulated with chemical transport models.

To determine the formation, transport, chemical transformation, and fate of particulate matter < 2.5 µm (PM_2.5) in the Mediterranean Sea in 2015, five different regional chemical transport models (CAMx – Comprehensive Air Quality Model with Extensions, CHIMERE, CMAQ – Community Multiscale Air Quality model, EMEP – European Monitoring and Evaluation Programme model, and LOTOS-EUROS) were applied. Furthermore, PM_2.5 precursors (ammonia (NH₃), sulfur dioxide (SO₂), nitric acid (HNO₃)) and inorganic particle species (sulfate (SO${}_{\mathrm{4}}^{\mathrm{2}-}$), ammonia (NH${}_{\mathrm{4}}^{+}$), nitrate (NO${}_{\mathrm{3}}^{-}$)) were studied, as they are important for explaining differences among the models. STEAM (see “List of abbreviations” in Appendix A) version 3.3.0 was used to compute shipping emissions, and the CAMS-REG version 2.2.1 dataset was used to calculate land-based emissions for an area encompassing the Mediterranean Sea at a resolution of 12 × 12 km² (or 0.1^∘ × 0.1^∘). For additional input, like meteorological fields and boundary conditions, all models utilized their regular configuration. The zero-out approach was used to quantify the potential impact of ship emissions on PM_2.5 concentrations. The model results were compared with observed background data from monitoring sites.

Four of the five models underestimated the actual measured PM_2.5 concentrations. These underestimations are linked to model-specific mechanisms or underpredictions of particle precursors. The potential impact of ships on the PM_2.5 concentration is between 15 % and 25 % at the main shipping routes. Regarding particle species, SO${}_{\mathrm{4}}^{\mathrm{2}-}$ is the main contributor to the absolute ship-related PM_2.5 and to total PM_2.5 concentrations. In the ship-related PM_2.5, a higher share of inorganic particle species can be found when compared with the total PM_2.5. The seasonal variabilities in particle species show that NO${}_{\mathrm{3}}^{-}$ is higher in winter and spring, while the NH${}_{\mathrm{4}}^{+}$ concentrations displayed no clear seasonal pattern in any models. In most cases with high concentrations of both NH${}_{\mathrm{4}}^{+}$ and NO${}_{\mathrm{3}}^{-}$, lower SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentrations are simulated. Differences among the simulated particle species distributions might be traced back to the aerosol size distribution and how models distribute emissions between the coarse and fine modes (PM_2.5 and PM₁₀). The seasonality of wet deposition follows the seasonality of the precipitation, showing that precipitation predominates wet deposition.

1 Introduction

Exhaust particles emitted from shipping have a large share in total emissions from the transport sector (Corbett and Fischbeck, 1997; Eyring et al., 2005), thereby affecting the chemical composition of the atmosphere as well as the regional air quality. Particularly in coastal areas, maritime transport contributes a considerable fraction to air pollution (Viana et al., 2014).

High particulate matter < 2.5 µm (PM_2.5) concentrations can be caused by transported particles, desert dust, or the production of secondary particulate matter (Tomasi and Lupi, 2017). Previous studies have revealed that in Europe the PM_2.5 concentration increase caused by shipping emissions is small (Viana et al., 2009; Aksoyoglu et al., 2016). Nevertheless, in the Mediterranean region the relative ship impact on the PM_2.5 concentration is large, with a share of 5 % to 20 % of the total PM_2.5 concentration (e.g., Aksoyoglu et al., 2016; Nunes et al., 2020). The formation of secondary particulate matter from ship emissions is of particular importance. According to Viana et al. (2009), the secondary contribution of ship emissions is equivalent to double their primary contribution. Secondary particles in the atmosphere form from gaseous precursors, whereas primary particles are directly emitted and evolve within a short time to form secondary particles. To improve the air quality in coastal regions, it is important to identify the pollutant sources and make reliable estimations of their impacts on surrounding PM levels. It has been shown that the majority of secondary particles contributing to local PM in ports come from shipping (Song and Shon, 2014). Furthermore, according to Klimont et al. (2017), the proportion of international shipping's particulate matter primary emissions to global anthropogenic emissions is between 3 % and 4 %, which is comparable to road traffic. Additionally, shipping contributions to total PM_2.5 concentrations far from coastlines were found to be responsible for exceedances of the WHO air quality guideline values (Nunes et al., 2020). The annual mean PM_2.5 limit value in the EU is 25 µg m⁻³ (EU DIRECTIVE 2008/50/EC, 2008), whereas the annual mean PM_2.5 goal established by the WHO is 5.0 µg m⁻³ (WHO, 2021). Strong evidence has been found for the relationship between exposure to PM_2.5 and the occurrences of certain diseases affecting the lungs, cancer, or type 2 diabetes (Heusinkveld et al., 2016; Chen et al., 2016; Gao and Sang, 2020). According to the WHO, there is no safe level of PM_2.5; thus, the gap between the WHO and EU's PM_2.5 values is of real concern (Karamfilova, 2022).

The MEPC decided in December 2022 to establish a sulfur emission control area in the Mediterranean Sea by 1 January 2025. In this area, the limit for sulfur in fuel oils used on board ships is 0.10 % (IMO, 2022). The global sulfur cap for marine vessels came into effect in January 2020, which declares that the sulfur content of any fuel oil used from ships must not exceed 0.50 % m m⁻¹, except for ships using “equivalent” compliance mechanisms, such as scrubbers. Calculations show that this policy has led to PM_2.5 reductions ranging from 0.5 µg m⁻³ to more than 2.0 µg m⁻³ along the major shipping routes in the Mediterranean Sea (Jonson et al., 2020). These relatively strict 2020 regulations are expected to lower the number of PM_2.5-related premature deaths by on average 15 % (Viana et al., 2020).

Although the Mediterranean Sea contains one of the busiest shipping routes worldwide, only a few regional-scale chemical transport modeling studies have considered this region. Viana et al. (2014) reviewed studies concerning the impacts of shipping emissions on air quality in European coastal areas, noting that the highest PM_2.5 contributions were found in the Mediterranean Sea and North Sea. Aksoyoglu et al. (2016) studied PM_2.5 concentrations in the Mediterranean Sea followed by a comparison of two models. Marmer and Langmann (2005) investigated the Mediterranean Sea on a broader scale and without comparing different CTM systems. Nevertheless, other studies have concentrated on smaller domains, such as the Iberian Peninsula (Baldasano et al., 2011; Nunes et al., 2020), the eastern Mediterranean Sea with the Arabian Peninsula (Večeřa et al., 2008; Tadic et al., 2020; Celik et al., 2020; Friedrich et al., 2021), or the urban scale and harbor cities (Schembari et al., 2012; Donateo et al., 2014; Prati et al., 2015). None of these studies, however, analyzed the potential shipping impacts on PM_2.5 concentrations together with individual aerosol species on a regional basis while additionally comparing the results of five CTMs.

A wide range of gaseous pollutants, such as sulfur dioxide (SO₂) and nitrogen oxide (NO_x= NO + NO₂), coming from shipping emissions can be precursors for particle formation (Jägerbrand et al., 2019; Karl et al., 2019; Matthias et al., 2010). Sulfur dioxide is released mainly by human activities such as fossil fuel burning, petroleum refining, and metal smelting (Zhong et al., 2020). SO₂ is oxidized by dissolved oxidants such as ozone (O₃) and hydrogen peroxide (H₂O₂) in the aqueous phase and by hydroxyl (OH) in the gas phase to generate sulfuric acid (H₂SO₄) (Seinfeld and Pandis, 2006). H₂SO₄ and nitric acid (HNO₃) react with ammonia (NH₃) to form ammonium sulfate ((NH₄)₂SO₄) and NH₄NO₃ aerosols, with H₂SO₄ neutralization having preference due to its lower vapor pressure (Hauglustaine et al., 2014).

Nitrogen oxides are primarily removed during the day via the OH radical oxidation reaction to produce (HNO₃) (Seinfeld and Pandis, 1998). At night, the main NO_x removal method involves interacting with O₃ to produce the nitrate (NO₃) radical, which may then combine with nitrogen dioxide (NO₂) to form dinitrogen pentoxide (N₂O₅) and may subsequently undergo a heterogeneous reaction with water to produce HNO₃. As it is highly soluble, HNO₃ disperses quickly in water droplets or is neutralized by reaction with NH₃ to produce NH₄NO₃ aerosols. Increased emissions of NH₃ or HNO₃ formation and their deposition negatively affect the environment through eutrophication and acidification, thereby contributing to the loss of ecosystem biodiversity (Remke et al., 2009; Kleijn et al., 2009; Krupa, 2003).

Furthermore, the air pollution status should be assessed to investigate the consequences of new legislation.

The current work investigates and analyzes the predictions of five different CTMs for air pollutant dispersion and transformation. The intercomparison was carried out in two parts: part one included the photochemistry and differences among the models regarding NO₂ and O₃ (Fink et al., 2023). The present study is part two of the model intercomparison and evaluates the same CTM simulations but different air pollutants, namely aerosols. This paper is structured as follows: Sect. 3.1 and 3.2 consider the simulated overall PM_2.5 model performance and spatial distribution. In Sect. 3.3, precursors (NH₃, HNO₃, SO₂, and NO₂) are investigated as the basis for inorganic particle species. Inorganic aerosol concentration and wet deposition are regarded in Sect. 3.4.

To date, the present study is the first multimodel study designed to compare the potential impacts of shipping on PM_2.5 and particle species simulated by five regional-scale CTMs for the Mediterranean Sea.

2 Materials and methods

In this section the models participating in the intercomparison study are briefly described. More detailed information about the standard setup of the models and model internal mechanisms used in the present study can be found in part one of this intercomparison study (Fink et al., 2023), which focuses on nitrogen oxides and ozone.

2.1 Models

In this study, five different regional-scale CTM systems run by four institutions participated: CAMx and CHIMERE run by AtmoSud, CMAQ run by Helmholtz-Zentrum Hereon, EMEP run by the IVL Swedish Environmental Research Institute, and LOTOS-EUROS run by the TNO Netherlands Organization for Applied Scientific Research. In order to produce comparable results with respect to the impact of shipping emissions on PM_2.5 concentrations, the models were set up in a similar way. The same shipping emissions data from STEAM (version 3.3.0.; Jalkanen et al., 2009, 2012; Johansson et al., 2013, 2017) were used for all CTMs. Land-based emissions (CAMS-REG, v2.0), grid projection (WGS84_longlat), domain (Mediterranean Sea), grid resolution (0.1^∘ × 0.1^∘, 12 × 12 km), and the modeled year (2015) were also consistent (Table 1). The CTM systems were applied in their standard setup for other input data; i.e., the meteorological input data and the boundary and initial conditions differed.

The model domains covered the largest part of the Mediterranean Sea, with a spatial extent ranging in latitude from 33.8 to 44.95^∘ and in longitude from −0.95 to 29.95^∘ (Appendix A). The appointed grid cell size was 12 × 12 km² interpolated on a 0.1^∘ × 0.1^∘ grid nested in a 36 × 36 km² grid (except EMEP).

A reference run for present air quality conditions was performed using all models, including all emissions (base case). Furthermore, all models ran once without shipping emissions (no-ship case). The difference between the estimates with all emissions and the calculations without shipping emissions was then used to calculate the potential impact of ships on pollutant concentrations (zero-out method).

From the results of all models, the annual averaged ensemble mean was calculated based on the daily files. The model run outputs all contained PM_2.5 in micrograms per cubic meter (µg m⁻³) at a daily resolution on a 2-D grid from the lowest layer and provides this as a netcdf file following CF conventions. Concentrations in the lowest layer close to ground were used for the intercomparison. The CTM systems calculated PM_2.5 concentrations in different ways depending on the major physical and chemical mechanisms implemented. Table 1 summarizes the model setups.

The models used in the intercomparison are listed as follows:

• CAMx v6.50 (Ramboll Environment and Health, 2020),

• CHIMERE 2017r4 (Menut et al., 2013),

• CMAQ v5.2 (Byun and Schere, 2006; Appel et al., 2017),

• EMEP MSC-W (Simpson et al., 2012, 2020), and

• LOTOS-EUROS v2.0 (Manders et al., 2017).

Detailed descriptions of the models used can be found in the first part of the intercomparison study (Fink et al., 2023).

Table 1 Main model parameters and input data for the five chemical transport models.

2.1.1 Aerosol modules

CAMx includes algorithms for inorganic aqueous chemistry (RADM–AQ), inorganic gas–aerosol partitioning (ISORROPIA), and two organic gas–aerosol partitioning and oxidation approaches (VBS or SOAP). Using gas-phase processes, these approaches produce sulfate, nitrate, and condensable organic gases. The hybrid 1.5-D VBS is applied to provide a unified framework for gas–aerosol partitioning and the chemical aging of both primary and secondary atmospheric organic aerosols (Ramboll Environment and Health, 2020). One crucial assumption in PSAT is that PM is allocated to the primary precursor for each type of particulate matter (i.e., PSO₄ is apportioned to SO_x emissions, PNO₃ is apportioned to NO_x emissions, and PNH₄ is apportioned to NH₃ emissions).

A detailed description of CHIMERE's inorganic and organic modules can be found in Menut et al. (2013). CHIMERE's sectional aerosol module includes emitted TPPM and secondary species such as nitrate, sulfate, ammonium, and SOAs. Natural dust and sea salt aerosols can also be produced as passive tracers or interactive species in equilibrium with other ions. Organic matter and elemental carbon can be speciated if an inventory of their emissions is supplied. The utilized models include the aqueous, gaseous, and particulate phases of ammonia, ammonium, nitrate, and sulfate. For instance, in accordance with the ISORROPIA thermodynamic equilibrium model, the model species pNH₃ represents an equivalent ammonium in the particulate phase as the sum of the NH${}_{\mathrm{4}}^{+}$ ion, NH₃ liquid, NH₄NO₃ solid, and other salts (Nenes et al., 1998).

CMAQ represents aerosol formation and growth using three log-normal-distributed modes: the Aitken and accumulation modes are generally less than 2.5 µm in diameter, while the coarse mode contains significant amounts of mass above 2.5 µm. PM_2.5 and PM₁₀ can be obtained from the model-predicted mass concentration and size distribution information.

The CMAQ aerosol scheme AERO6 was employed; this scheme expands the chemical speciation of PM by the species Al, Ca, Fe, Si, Ti, Mg, K, and Mn. Sulfuric acid (H₂SO₄), nitric acid (HNO₃), hydrochloric acid (HCl), and ammonia (NH₃) gas-phase–aerosol partition equilibria are solved by the ISORROPIA II mechanism (Fountoukis and Nenes, 2007; Nenes et al., 1998). Contained within this scheme is the formation of SOA from isoprene, terpenes, benzene, toluene, xylene, and alkanes (Carlton et al., 2010; Pye and Pouliot, 2012). CMAQ allows for dynamic mass transfer of semivolatile inorganic gases to coarse-mode particles, which facilitates the replacement of chloride by NO${}_{\mathrm{3}}^{-}$ in sea salt aerosols (Foley et al., 2010).

The EMEP MSC-W model version used was rv4.34 with chemical mechanism EmChem 19a (Simpson et al., 2012; Simpson et al., 2020). The mechanism builds on surrogate VOC species (as in Simpson et al., 2012, but extended with benzene and toluene) and has 171 gas-phase and heterogeneous reactions. The model always assumes equilibrium between the gas and aerosol phases using the MARS equilibrium module of Binkowski and Shankar (1995). For SOAs a VBS approach is used (Robinson et al., 2007; Donahue et al., 2009; Bergström et al., 2012). The semivolatile ASOA and BSOA species are considered to oxidize (age) in the atmosphere via OH reactions, whereas all POA emissions are treated as nonvolatile to maintain the emission totals of both the PM and the VOC components from the official emission inventories (Simpson et al., 2012). The aerosol module of the EMEP model distinguishes five classes of fine and coarse particles (fine-mode nitrate and ammonium, other fine-mode particles, coarse nitrate, coarse sea salt, and coarse dust); for dry-deposition purposes, these particles are assigned mass median diameters (D_p), geometric standard deviations (σ_g), and densities (ρ_p). The aerosol components that are taken into account include sea salt, SO${}_{\mathrm{4}}^{\mathrm{2}-}$, NO${}_{\mathrm{3}}^{-}$, NH${}_{\mathrm{4}}^{+}$, and anthropogenic main PM. Aerosol water is also considered.

LOTOS-EUROS uses the TNO CBM-IV scheme, which is a modified version of the original CBM-IV scheme (Whitten et al., 1980). N₂O₅ hydrolysis is described explicitly based on the available (wet) aerosol surface area (Schaap et al., 2004). The aqueous phase and heterogeneous formation of sulfate are described by a simple first-order reaction constant (Schaap et al., 2004; Barbu et al., 2009). Aerosol chemistry is represented using ISORROPIA II (Fountoukis and Nenes, 2007).

2.1.2 Wet-deposition mechanisms

Wet deposition is the predominant removal process for fine particles. The CAMx wet-deposition model uses a scavenging method in which the local concentration change rate inside or under a precipitating cloud is determined by a scavenging coefficient. From the top of the precipitation profile to the surface, wet scavenging is estimated for each layer inside a precipitating grid column. The scavenging coefficients of gases and PM are calculated differently depending on the correlations given by Seinfeld and Pandis (2006) (Ramboll Environment and Health, 2020). The wet-deposition process in CHIMERE follows the scheme proposed by Loosmore and Cederwall (2004). In CMAQ, wet deposition is calculated in cloud chemistry treatments. The resolved cloud model calculates the contribution of each model layer to the precipitation. Based on a normalized profile of precipitating hydrometeors, CMAQ operates a simple algorithm to assign precipitation amounts to individual layers (Foley et al., 2010). The EMEP model's parameterization of wet-deposition processes covers both the in-cloud and the sub-cloud scavenging of gases and particles. The parameterization of wet deposition is described in Berge and Jakobsen (1998). There are two types of wet deposition in LOTOS-EUROS: below-cloud scavenging and in-cloud scavenging. The technique is described in Seinfeld and Pandis (2006), and Banzhaf et al. (2012) served as the foundation for the in-cloud scavenging module.

2.2 Emissions

2.2.1 Land-based emissions

All five models used anthropogenic land-based gridded emissions from the CAMS-REG v2.2 emission inventory for 2015, which is described in Granier et al. (2019) and is essentially a further development of the earlier TNO_MACC inventories (Kuenen et al., 2014). A more recent version, CAMS-REG-v4.2, is described in detail in Kuenen et al. (2022).

For each country, the gridded emission files included GNFR emission sectors for the air pollutants NO_x, SO₂, NMVOC, NH₃, CO, PM₁₀, PM_2.5, and CH₄. The spatial resolution of the emissions data was $\mathrm{1}/\mathrm{10}$^∘ × $\mathrm{1}/\mathrm{20}$^∘ in longitude and latitude (i.e., $\sim \mathrm{6}\times$ 6 km over central Europe). The CAMS-REG inventory also provides default information in order to apply the emissions in the CTMs. The height distribution of emissions per GNFR sector was prepared according to Bieser et al. (2011). Based on the assignment of PM and NMVOC components at a detailed subsector level, PM and NMVOC speciation profiles are provided for each country, year, and GNFR sector. The temporal distribution of emissions is based on the default temporal variation provided along with the CAMS-REG inventory. The NO_x splitting was performed according to Manders-Groot et al. (2016).

2.2.2 Shipping emissions

The shipping emission dataset produced with the STEAM model has a spatial resolution of 12 × 12 km² and a temporal resolution of 1 h. The STEAM v3.3.0 emissions are divided into two vertical layers (0 to 36 m; 36 to 1000 m) and are provided for mineral ash, carbon monoxide (CO), carbon dioxide (CO₂), elemental carbon (EC), NO_x, organic carbon (OC), PM_2.5, particle number count (PNC), sulfate (SO₄), SO_x (containing SO₂ and SO₃), and VOCs. To reduce the number of generated emission maps and the computational resources needed to run the STEAM model, VOC emissions were divided into four categories according to their properties as a function of the engine load. Emission factors for VOCs are based on the average values taken from various publications (Agrawal et al., 2008, 2010; Sippula et al., 2014; Reichle et al., 2015).

All shipping emissions are included in the lowest layer of CAMx. In CAMx, all gridded emissions are at the ground level except punctual and linear emissions. For CHIMERE, 88 % of the emissions below 36 m and all shipping emissions above 36 m were added to the second layer. Only 12 % of the emissions below 36 m were allocated to the model's lowest layer. The STEAM emission dataset, which included stack heights, was used for this procedure. In CMAQ, shipping emissions were split between the two lowest levels; those below 36 m were ascribed to the lowest layer, while those above 36 m were positioned in the second layer. The heights of the lowest and the second layer in CMAQ are 42 m for each. The STEAM emissions were summed from hourly to daily emissions and were attributed to the lowest layer (up to 90 m) in the EMEP simulations. In LOTOS-EUROS, emissions below 36 m were divided into two layers: the first layer was 25 m thick (∼ 70 % of emissions), and the second layer was 30 m thick (∼ 30 % of emissions). Over 36 m, emissions were separated into various height groups: 30 % were between 36 and 90 m, 30 % were between 170 and 90 m, 30 % were between 170 and 310 m, and 10 % were between 310 and 470 m. These emissions were placed in the second or third model layers because of the dynamic second model layer, which follows the meteorological boundary layer. All emissions were placed in this second layer when the meteorological boundary layer was well mixed and vertically extended (higher than 470 m), while some emissions were placed in the third layer when the boundary layer was shallow.

2.3 Observational data, statistical analysis, and analysis of model results

The model findings regarding the total surface PM_2.5 concentrations from the five CTM systems were compared with data from the air quality monitoring network obtained from the EEA's download service (https://discomap.eea.europa.eu/map/fme/AirQualityExport.htm, last access: 6 September 2023). The locations of the measurement stations are shown in Fig. A1, and detailed information on the stations can be found in Appendix B.

The stations were chosen based on the following criteria: (i) the station type was “background”; (ii) the station elevation was less than 1000 m; and (iii) the station recorded data for more than one of the following pollutants – NO₂, O₃, or PM_2.5. In the first part of this intercomparison study (Fink et al., 2023), NO₂ and O₃ were discussed. Since simulating the potential impact of ships was the main focus of this study, stations near the sea were the preferable choice.

The model findings regarding the total surface PM_2.5 concentrations from the five CTM systems were compared with existing observations. The RMSE, NMB, and correlation coefficient R were determined for each monitoring station to quantify the model performance, as described in the previous study (Fink et al., 2023).

A categorization scheme for the correlations was established as described in Schober et al. (2018), with weak (0.00–0.39), moderate (0.40–0.69), and strong (0.70–1.00) correlations.

To compare the predicted daily mean concentrations with the measurements recorded at representative sites, time series were employed. In addition, based on hourly data, the yearly mean potential ship impact was determined. Boxplots based on yearly values obtained from hourly data at each station were used to graphically compare the model performances using the R, NMB, and RMSE metrics. Annual mean values based on hourly data were utilized for the intercomparison maps. Based on hourly data, the correlations between models were determined for each grid cell.

3 Results

3.1 PM_2.5 model performance

Regarding the model performance, time series can give an overview of the performance throughout the whole year. Figure 1 displays the average values at all 28 measurement stations. CAMx, CMAQ, EMEP, and LOTOS-EUROS underestimate the actual measured data. The largest underestimations are found for CMAQ (NMB = −0.42) and LOTOS-EUROS (NMB = −0.54). Contrary to the other CTM systems, CMAQ does not consider the contribution of dust, which can cause underestimations of PM_2.5. However, the correlations between the modeled and measured data are strongest for these models (CMAQ: R= 0.50, LOTOS-EUROS: R= 0.54; Table 2). No correlation can be found between the measured and modeled data for CHIMERE (R= 0.02); on the other hand CHIMERE displays only a slight overestimation of the actual data (NMB = 0.06). The simulated potential impacts of ships at all measurement stations are between 5.7 % (CMAQ) and 13.8 % (CAMx; Table 2) as annual averages. The simulated ship impacts on PM_2.5 concentrations are within the ranges stated in other studies. In a review of studies regarding the impact of shipping emissions on coastal regions, Viana et al. (2014) reported PM_2.5 impacts of shipping to be between 5 % and 14 %. Aksoyoglu et al. (2016) found PM_2.5 concentrations between 10 % and 15 % along coastal areas due to ship traffic. Ship impacts of approximately 20 % in the southern coastal region of the Iberian Peninsula were found by Nunes et al. (2020). Although the models underestimated the actual measured total PM_2.5 concentrations in this study, they slightly overestimated the relative potential impact of ships on PM_2.5 compared with previous measurement studies. Donateo et al. (2014) measured a proportion of 7.4 % of ships to the total PM_2.5; Pandolfi et al. (2011) measured a proportion of shipping to PM_2.5 concentrations in the Bahía de Algeciras of between 5 % and 10 %. Agrawal et al. (2009) monitored PM_2.5 at the harbor of Los Angeles and found PM_2.5 contributions from ships of up to 8.8 %. Predominating secondary particles in PM_2.5 for potential ship impact in the present study can explain the deviations to the measurement studies.

Figure 1 Time series with daily mean PM_2.5 concentrations in 2015, averaged for all stations, and the respective grid cells of the models for (a) CAMx, (b) CHIMERE, (c) CMAQ, (d) EMEP, and (e) LOTOS-EUROS. The dashed gray lines indicate measured data, the colored lines indicate modeled data, and the solid gray lines indicate the modeled potential ship impacts.

The RMSE is very similar for all models with a value between 10.7 and 12.2 µg m⁻³. However, the RMSE is strongly determined by high concentrations and can be biased by outliers. This might explain the similar RMSE derived from CHIMERE despite the lack of correlation. The mean RMSE from different models for PM_2.5 in Europe found in the AQMEII intercomparison study by Im et al. (2015) was 6.19 for rural stations and 10.26 for urban stations and is similar to the RMSE calculated in the present study.

The underestimation of PM_2.5 concentrations by four out of five models is consistent with results by Im et al. (2015), who reported an underestimation of particulate matter for all participating models, with the largest underestimations observed in the Mediterranean region. They stated that the representation of dust and sea salt emissions had a large impact on the simulated PM concentrations and that uncertainties remain when trying to identify the reasons for the model bias (Im et al., 2015). Additionally, in a study by Gašparac et al. (2020), underestimations were also found when using EMEP and WRF-Chem to model PM_2.5 at rural stations in Europe. Solazzo et al. (2012) performed an operational model evaluation for 10 models and found that the models underestimated the monthly mean PM_2.5 surface concentrations in Europe in most cases.

Table 2 Correlation (R), normalized mean bias (NMB), root mean square error (RMSE), and observational (obs) and modeled (mod) mean PM_2.5 values for 2015 over all 28 stations. The observed mean value for all stations is 14.6 µg m⁻³.

3.2 PM_2.5 spatial distribution

The highest PM_2.5 values are simulated by all five models in northern Italy, the Balkan Peninsula, and northern Africa (Fig. 2). The PM_2.5 annual mean concentration results show that CHIMERE has the highest annual mean values of 13 to 15 µg m⁻³ for the eastern part of the domain and over water, whereas LOTOS-EUROS displays the lowest values with 2.0 to 4.0 µg m⁻³ in most regions (Fig. 2). CMAQ, CAMx, and EMEP show similar model PM_2.5 outputs with diverse values distributed between 2.0 and 11 µg m⁻³ over the domain. The ensemble mean value over the whole domain is 8.6 µg m⁻³ (Fig. 8a). All five models display high PM_2.5 concentrations of > 15 µg m⁻³ in the Po Valley. In this area, Kiesewetter et al. (2015) and Clappier et al. (2021) also simulated high values between 20 and 45 µg m⁻³ for 2015. As demonstrated in Table 3, the correlation between the base-run model results with all emissions is the strongest between EMEP and CMAQ (R = 0.59) and CAMx and CMAQ (R= 0.42). In Fink et al. (2023), a high correlation was found between CAMx- and CHIMERE-simulated NO₂ and O₃ concentrations because both models used the same meteorology. Nevertheless, the present study reveals that particle chemistry causes results that differ more due to a higher complexity in the calculations.

Figure 2 Annual mean PM_2.5 total concentrations for (a) CAMx, (b) CHIMERE, (c) CMAQ, (d) EMEP, and (e) LOTOS-EUROS, as well as for the (f) ensemble model mean. Below the domain figures are the respective frequency distributions displayed for the annual mean PM_2.5 concentrations, referring to the whole model domain.

Figure 3 Annual mean PM_2.5 relative potential ship impacts for (a) CAMx, (b) CHIMERE, (c) CMAQ, (d) EMEP, and (e) LOTOS-EUROS, as well as for the (f) ensemble model mean. Below the domain figures are the respective frequency distributions displayed for the annual mean PM_2.5 potential ship impacts, referring to the whole model domain.

The potential impacts of PM_2.5 from ships simulated by CAMx, LOTOS-EUROS, and EMEP have the largest areas with values of up to 25 % at the main shipping routes (Fig. 3). CMAQ and CHIMERE have a potential shipping impact of 15 % along the main shipping lines close to the African coast. This impact is lower than that shown in other studies. Aksoyoglu et al. (2016) found the highest impacts of 25 % to 50 % of total PM_2.5 concentrations when using CAMx along the main shipping routes. Sotiropoulou and Tagaris (2017) used CMAQ for simulations and stated that emissions from shipping are likely to increase PM_2.5 concentrations during winter by up to 40 % over the Mediterranean Sea, while during summer, they simulated an increase of more than 50 %. In both studies, the modeled year is 2006, which might explain the deviation from the present study as using a different year. Regarding coastal areas in the present study, potential shipping impacts reaching 12 % to 15 % are simulated.

Figure 4 Annual mean PM_2.5 absolute potential ship impacts for (a) CAMx, (b) CHIMERE, (c) CMAQ, (d) EMEP, and (e) LOTOS-EUROS, as well as for the (f) ensemble model mean. Below the domain figures are the respective frequency distributions displayed for the annual mean PM_2.5 potential ship impacts, referring to the whole model domain.

Regarding the absolute potential impacts of ships at the main shipping routes, CAMx, CHIMERE, and EMEP show values of 2.0 µg m⁻³, and the values simulated by CMAQ and LOTOS-EUROS are between 0.5 and 1.0 µg m⁻³ (Fig. 4). The median of the ensemble mean is 0.85 µg m⁻³ (Figs. 4 and 8). Aksoyoglu et al. (2016) simulated similar shipping impacts with CAMx, with values mainly between 0.5 and 1.0 µg m⁻³.

The sea salt concentrations might partly give an explanation for the differing PM_2.5 concentration distribution among the models. The annual mean sea salt (NaCl) concentration in fine and coarse PM showed the highest values for CHIMERE, which might be an explanation for the high PM_2.5 absolute concentration (Supplement Fig. S1). The LOTOS-EUROS sea salt displayed the lowest concentrations; the overall PM_2.5 concentration is also the lowest compared with the other CTMs. The sea salt concentration was the highest (up to 7.0 µg m⁻³) over the sea in areas with high surface wind speeds for CHIMERE, CMAQ, EMEP, and LOTOS-EUROS (Fig. S2). This can be confirmed by the correlation between wind speed and sea salt at several points over water for CMAQ, EMEP, and LOTOS-EUROS (Fig. S3; Table S1 in the Supplement). CAMx is excluded from the analysis since sea salt is only present in fine PM.

Solazzo et al. (2012) demonstrated that the chemical components SO${}_{\mathrm{4}}^{\mathrm{2}-}$, NO${}_{\mathrm{3}}^{-}$, and NH${}_{\mathrm{4}}^{+}$ were better reproduced by nine CTMs than the total PM_2.5. They concluded from this result that other components (e.g., organic aerosols) could be simulated with less accuracy than inorganic components.

Table 3 Correlations between models for the PM_2.5 base runs of the whole domain (all grid cells), based on daily PM_2.5 total concentration data.

3.3 Precursors

High amounts of NH₃, HNO₃, SO₂, and NO₂ are expected to lead to higher values of the aerosol particles composed of NH${}_{\mathrm{4}}^{+}$, NO${}_{\mathrm{3}}^{-}$, and SO${}_{\mathrm{4}}^{\mathrm{2}-}$. The modeled spatial distributions of these precursors can be found in the Supplement (HNO₃: Figs. S4–S6; NH₃: Figs. S8–S10; SO₂: Figs. S11–S13; and NO_x: Figs. S14–S16).

The highest annual mean HNO₃ concentration among the base runs is found in the CAMx and CHIMERE simulations over water (2.0 to 5.0 µg m⁻³); over land, the values are between 0.0 and 1.5 µg m⁻³, and those in coastal areas reached 2.0 µg m⁻³ (Fig. S4). The absolute potential ship impact is also the highest in CAMx and CHIMERE at the main shipping routes and over water areas (1.0 to 3.0 µg m⁻³). The relative potential ship impact on the total HNO₃ ranges from 60 % to 85 % along the main shipping routes simulated by CAMx, CMAQ, and EMEP (Fig. S4). These impacts are slightly lower for CHIMERE and LOTOS-EUROS (60 % to 75 %).

The high HNO₃ concentrations simulated by CAMx and CHIMERE might be traced back to the NO₂ concentrations; these two models also show higher NO₂ concentrations than the other CTMs (Fig. S14; Fink et al., 2023). This can be explained by the fact that HNO₃ is a major NO₂ sink, especially during daytime. NO₂ is primarily emitted from anthropogenic fossil fuel burning but also from natural sources (i.e., soil emissions, biomass burning, lightning). During daytime, the main NO₂ removal mechanism is oxidation by hydroxyl (OH) radicals to form HNO₃ (Seinfeld and Pandis, 1998). It can be concluded that in areas with shipping, more NO₂ enters the atmosphere; the total NO₂ concentration increases; and as a result of the subsequent reactions, the HNO₃ concentration also increases. The HNO₃ : NO₂ ratio can be used to normalize the data (Fig. S7). The ratio displays low values over land and along the main shipping routes, indicating that in these areas, both the HNO₃ and the NO₂ concentrations are high. A low HNO₃ : NO₂ ratio could also mean that only a small amount of OH is present, especially in areas with a low O₃ concentration.

After its formation, HNO₃ can react with NH₃ to be neutralized and to form particles when NH₃ is in excess. The annual mean NH₃ for the base case shows very similar patterns and values among all models (Fig. S8). The highest concentrations of NH₃ with all emission sources are located over land areas with values up to 2.5 µg m⁻³, which can be traced back to agriculture, the main source of NH₃ emissions (Behera et al., 2013). Over water areas, the NH₃ concentration is very small, typically between 0.0 and 0.3 µg m⁻³, except for the slightly higher results modeled by LOTOS-EUROS, with values between 0.2 and 0.8 µg m⁻³. Negative potential ship impacts (−0.01 to −1.0 µg m⁻³ and −2.5 % to −150 %; Figs. S9 and S10) are found for the whole domain in all five models. The relative ship impacts are the lowest at the main shipping routes for CAMx and EMEP. The spatial distribution of the NH₃ relative ship impact is opposite to the simulated HNO₃ values at the main shipping routes, with low NH₃ and high HNO₃ values. These results indicate that available NH₃ reacts directly with HNO₃ to form particles (i.e., NH₄NO₃). Thus, NO_x emissions from shipping lead to HNO₃ formations and subsequent NH₃ consumption; e.g., shipping impacts on NH₃ concentrations are usually negative.

The CAMx simulations show the highest SO₂ concentrations with more than 10 µg m⁻³ in some areas in western Turkey, in urban areas, and along major shipping lanes (Fig. S11). The results from the other four CTMs display high values around the Bosporus and in some areas over the Balkan Peninsula with values of 11 µg m⁻³ and much lower concentrations along the main shipping routes. The potential ship impacts are similarly high in CAMx and CHIMERE (1.0 µg m⁻³; 85 % of the total concentration; Figs. S12 and S13), with the highest values along the major shipping route north of the African coast. The CMAQ, EMEP, and LOTOS-EUROS results display similarly high values but only in small areas. The modeled year is 2015, so the global 0.5 % sulfur cap of marine fuels was not yet effective. Heavy fuel oils with sulfur contents reaching 3.50 % were used until 2020 to power ships; thus, the SO₂ emitted from ships in the present study is still high, and it can be expected that it has a large impact on secondary particle formation.

3.4 Inorganic aerosol species

3.4.1 Concentrations

In the Northern Hemisphere, secondary inorganic ammonium, sulfate, and nitrate aerosols represent a large fraction of the PM_2.5 composition (Jimenez et al., 2009). Ammonium preferentially binds to SO${}_{\mathrm{4}}^{\mathrm{2}-}$ in atmospheric aerosols in the form of (NH₄)₂SO₄. NH₄NO₃, on the other hand, is formed in areas characterized by high-NH₃ and high-HNO₃ conditions and low-H₂SO₄ conditions. The results of the CTMs with regard to these three particle species and their potential ship impacts are considered in the following section. The spatial distributions of the total concentrations and absolute potential ship impacts of the individual species can be found in the Supplement (NH${}_{\mathrm{4}}^{+}$: Figs. S17 and S18; SO${}_{\mathrm{4}}^{\mathrm{2}-}$: Figs. S19 and S20; and NO${}_{\mathrm{3}}^{-}$: Figs. S21 and S22). The spatial distribution of the relative potential ship impact is shown in Figs. 5 to 7.

Figure 5 Annual mean NH${}_{\mathrm{4}}^{+}$ relative potential ship impacts for (a) CAMx, (b) CHIMERE, (c) CMAQ, (d) EMEP, and (e) LOTOS-EUROS. Below the domain figures are the respective frequency distributions displayed for the annual mean NH${}_{\mathrm{4}}^{+}$ potential ship impacts, referring to the whole model domain.

Figure 6 Annual mean SO${}_{\mathrm{4}}^{\mathrm{2}-}$ relative potential ship impacts for (a) CAMx, (b) CHIMERE, (c) CMAQ, (d) EMEP, and (e) LOTOS-EUROS. Below the domain figures are the respective frequency distributions displayed for the annual mean SO${}_{\mathrm{4}}^{\mathrm{2}-}$ potential ship impacts, referring to the whole model domain.

Figure 7 Annual mean NO₃ relative potential ship impacts for (a) CAMx, (b) CHIMERE, (c) CMAQ, (d) EMEP, and (e) LOTOS-EUROS. Below the domain figures are the respective frequency distributions displayed for the annual mean NO₃ potential ship impacts, referring to the whole model domain.

The spatial distribution of NH${}_{\mathrm{4}}^{+}$ shows that the lowest total annual mean can be found mainly in the southwestern part of the domain (approximately 0.0 µg m⁻³) and the highest in the Po Valley and Bosporus (1.5 µg m⁻³, Fig. S17). The relative ship impacts are very similar for all models (0.25 % to 5.0 % over land, 10 % to 25 % over water; Fig. 5) as well as for the absolute ship impact (Fig. S15). Aksoyoglu et al. (2016) simulated NH${}_{\mathrm{4}}^{+}$ values between 0.0 and 0.2 µg m⁻³ in the Mediterranean region, with higher concentrations (0.4 µg m⁻³) in the Po Valley. This is within the same range of concentrations in the present study. Ge et al. (2021) used the EMEP model to simulate global particle species concentrations and compared them with measured concentrations. They showed in their study that the NH${}_{\mathrm{4}}^{+}$ concentrations simulated in Europe in 2015 were overestimated by a factor of 2 compared with the actual measured NH${}_{\mathrm{4}}^{+}$ concentrations. The measurements displayed a mean of 0.45 µg m⁻³. The ensemble mean for NH${}_{\mathrm{4}}^{+}$ in the present study (0.6 µg m⁻³, Fig. 8a) is in good agreement with these measurements. However, a previous study on measured compared with simulated aerosol distribution with the CMAQ model displayed a slight underestimation of NH${}_{\mathrm{4}}^{+}$ (Matthias, 2008).

Figure 8 (a) Boxplots for concentrations of PM_2.5 and the PM_2.5 components SO${}_{\mathrm{4}}^{\mathrm{2}-}$, NO₃, NH${}_{\mathrm{4}}^{+}$, and “others” as simulated by the five CTMs. The ensemble mean is “all_mean”. Others are calculated as PM_2.5 minus the sum of SO${}_{\mathrm{4}}^{\mathrm{2}-}$, NO₃, and NH${}_{\mathrm{4}}^{+}$. Data are based on the whole domain (all grid cells) and hourly data for all emission sources (“emisbase”). (b) The same as (a) but for ships only.

The NH${}_{\mathrm{4}}^{+}$ proportion to the total PM_2.5 is similar among all models (5.6 % to 7.8 %; Fig. 8a, Table 4), and only LOTOS-EUROS displayed a relatively high share (12.2 %). This pattern is similar for the ship impacts, where all models show proportions between 9.1 % and 12.6 %, but higher values are simulated by LOTOS-EUROS (23.5 %; Fig. 8b, Table 5).

SO${}_{\mathrm{4}}^{\mathrm{2}-}$ is the oxidation product of SO₂, which is primarily emitted by anthropogenic processes such as fossil fuel combustion, petroleum refining, and metal smelting (Zhong et al., 2020). In the present study, SO${}_{\mathrm{4}}^{\mathrm{2}-}$ is the main contributor to the total PM_2.5 mass (Fig. 8, Table 4). Especially in the model ensemble mean for the absolute ship-related concentrations, SO${}_{\mathrm{4}}^{\mathrm{2}-}$ makes up 44.6 % of PM_2.5 (Fig. 8b, Table 5). The annual mean SO${}_{\mathrm{4}}^{\mathrm{2}-}$ total concentration is the highest for CHIMERE in the eastern part of the domain, reaching 6.0 µg m⁻². EMEP displays a SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentration within the ranges of the other models, CAMx, CMAQ, and LOTOS-EUROS, in the western part of the domain. These models show very similar spatial distributions with concentrations of up to 2.0 µg m⁻³. The median ensemble mean for the run with all emission sources is 2.0 µg m⁻³. This ensemble mean is low in comparison with the results of Solazzo et al. (2012); they found a mean value of 6.0 µg m⁻³ but considered a larger European area that included the areas with the highest SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentrations in Europe. For this larger area, Solazzo et al. (2012) found that the models used underestimated SO${}_{\mathrm{4}}^{\mathrm{2}-}$ by 7 % to 17 %.

In the present study, the relative potential ship impact on the total SO${}_{\mathrm{4}}^{\mathrm{2}-}$ is the lowest over land, with 0 % to 3.0 %, and higher in coastal areas, with values from 6 % to 20 % (Fig. 6). Along the main shipping routes it is the highest, reaching 50 % for CAMx, EMEP, and LOTOS; for CHIMERE and CMAQ, it is lower, with values reaching 30 %. Aksoyoglu et al. (2016) showed similar relative potential ship impacts of 50 % to 60 % in the western Mediterranean. In their study, values were between 0.0 to 1.0 µg m⁻³ over land areas, but over water along the main shipping routes they were the highest at 2.2 µg m⁻³.

Mallet et al. (2019) traced back higher SO${}_{\mathrm{4}}^{\mathrm{2}-}$ in the eastern part of the domain due to westerly winds. In the present study, we found this higher concentration for SO${}_{\mathrm{4}}^{\mathrm{2}-}$ in the eastern part of the Mediterranean as well. In Lampedusa, they found ammonium sulfate contributed 63 % to PM₁ mass, followed by organics (Mallet et al., 2019). In our study, the organics/others had the highest share of total PM_2.5 when considering all emission sources, followed by sulfate and ammonium. In the present study, CTM systems simulated lower values for ship impacts; over land, they are 0.0 to 0.03 µg m⁻³, and along the main shipping routes, they reached 0.9 µg m⁻³. Regarding the absolute ship impacts on SO${}_{\mathrm{4}}^{\mathrm{2}-}$, the model simulations display similar concentrations and are slightly lower for CMAQ and LOTOS-EUROS (Fig. S20) compared with the other models. Especially over water areas, large areas with considerable SO₂ and SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentrations can be seen. Because NH${}_{\mathrm{4}}^{+}$ is preferentially bound to SO${}_{\mathrm{4}}^{\mathrm{2}-}$ in atmospheric aerosols to form (NH₄)₂SO₄, in areas over water, less NH₄NO₃ forms.

Im et al. (2014) suggested in their intercomparison study that over Europe, SO${}_{\mathrm{4}}^{\mathrm{2}-}$ levels were underestimated by most models; only a few models overestimated SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentrations in Europe. The underestimating models were WRF-Chem models, and the SO${}_{\mathrm{4}}^{\mathrm{2}-}$ underestimations were attributed to the absence of SO₂ oxidation in cloud water in the heterogeneous phase.

The highest annual mean NO${}_{\mathrm{3}}^{-}$ total concentrations are simulated over land areas, especially over Italy and in the Balkan states (> 2 µg m⁻³; Fig. S18), and the lowest concentrations are over the sea. CAMx, CMAQ, and LOTOS-EUROS show higher concentrations compared with results derived from CHIMERE. The concentrations over water are lower than those over land. The ensemble median of all CTMs over the whole domain is 0.63 µg m⁻³ (median value; Fig. 8a). The absolute potential impacts of ships on the total NO${}_{\mathrm{3}}^{-}$ concentrations are similar among all models, displaying values mainly between −0.005 and 0.15 µg m⁻³; only CMAQ demonstrates relatively low values along the main shipping routes (−0.5 µg m⁻³), and CAMx has higher values (1.0 µg m⁻³) in some coastal areas (Fig. S19). This can be explained by higher SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentrations derived from SO₂ emissions. Sulfate replaces nitrate as long as the ammonia concentration is low. In model simulations with ships, NO${}_{\mathrm{3}}^{-}$ can decrease because ammonia is already taken from sulfur emissions from ships. Aksoyoglu et al. (2016) found similar results for the Mediterranean Sea considering the NO${}_{\mathrm{3}}^{-}$ concentrations, with values between 0.0 µg m⁻³ and 0.2 µg m⁻³. Im et al. (2014) showed that simulated NO${}_{\mathrm{3}}^{-}$ levels were overestimated by most of the CTMs by more than 75 %. Higher concentrations over water than over land due to NH₄NO₃ formation are found in areas characterized by high-NH₃ and high-HNO₃ conditions and low-H₂SO₄ conditions. In the present study, the relative potential ship impacts on NO${}_{\mathrm{3}}^{-}$ display contradicting tendencies among the models (Fig. 7). The CAMx, EMEP, and LOTOS model results are similar, with relative potential ship impacts over land from 0.0 % to 5.0 % (in the Balkan states), those in coastal areas and Italy from 10 % to 25 %, and those along main shipping routes from 50 % to 65 % or even up to 85 %. CHIMERE and CMAQ display lower relative potential ship impacts. For CMAQ, the impact is negative along the main shipping routes, at −25 %. Sulfur dioxide or ammonia might lead to a negative NO${}_{\mathrm{3}}^{-}$ impact because the NO₂ emissions from ships would make a positive contribution to nitrate formation. Therefore, without ships, an (NH₄)₂SO₄ should be formed, which is more stable than NH₄NO₃. These low values in the aerosol species for CMAQ but higher values for EMEP, CAMx, and LOTOS represented the PM_2.5 ship impacts and might partly explain the deviations in PM_2.5. Furthermore, in CMAQ the coarse mode in nitrate and ammonium has a larger share compared with the other CTMs. A more detailed discussion is given in Sect. 4.

Regarding the PM_2.5 composition, the share of other particles, which contain mainly organics but also, e.g., sea salt, is the highest compared with the inorganic species (Fig. 8). Nevertheless, the particle composition revealed varying distributions in the ship-related PM_2.5 concentration. Here, inorganic particle species have relatively high percentages compared with organic aerosols. In some cases, sulfate has an even higher share of the total PM_2.5 than other particles.

The seasonal variability in particle species shows that NO${}_{\mathrm{3}}^{-}$ is more temperature dependent than SO${}_{\mathrm{4}}^{\mathrm{2}-}$ and NH${}_{\mathrm{4}}^{+}$. NO${}_{\mathrm{3}}^{-}$ is higher in winter and spring but lower in summer and autumn. This pattern can be found in all CTM simulations. For PM_2.5, on the other hand, no discernible pattern is found regarding seasonal variability. In particular, the ensemble mean PM_2.5 concentration remained within the same range in all seasons.

Table 4 Relative particle species of total PM_2.5 emissions.

Table 5 Relative particle species of total shipping-related PM_2.5.

Figure 9 Concentration of particle species and precipitation divided by seasons and CTMs. “all_mean” displays the model ensemble. Spring: March, April, May; summer: June, July, August; autumn: September, October, November; winter: December, January, February. Concentration is based on the annual median over the whole domain. Precipitation displays the seasonal sum (in mm).

3.4.2 Wet deposition

Wet deposition can provide indications of the fate of particles. EMEP does not deliver separate deposition files for individual particle species but for reduced and oxidized nitrogen. Thus, EMEP is not considered when analyzing wet deposition in this study.

Regarding the spatial distribution of NH${}_{\mathrm{4}}^{+}$ wet deposition, the highest annual sums are displayed by CMAQ and LOTOS-EUROS (up to 250 mg m⁻² yr⁻¹ over land; up to 50 mg m⁻² yr⁻¹ over water; Fig. S23). CAMx and CHIMERE show a similar spatial distribution with values mainly between 10 and 25 mg m⁻² yr⁻¹. CAMx and CHIMERE used the same meteorology data, but despite this the seasonal distribution of wet deposition differs (Fig. 10). An explanation for this differing behavior might be provided by the scavenging mechanisms. In CHIMERE the in-cloud mechanism for deposition of particles is assumed to be proportional to the amount of water lost by precipitation. In CAMx, the in-cloud scavenging coefficient for aqueous aerosols is the same as for the scavenging of cloud droplets. Below the cloud, CHIMERE uses a polydisperse distribution following Henzig et al. (2006), whereas in CAMx for rain or graupel the collection efficiency is calculated as in Seinfeld and Pandis (1998). The other possible explanation is that all the emissions in CAMx are emitted in the first layer and in CHIMERE it depends on the emissions distribution.

Regarding the wet deposition of sulfate, the annual totals for all emission sources are the highest over the Balkan Peninsula in the CMAQ and LOTOS-EUROS model outputs (300 to 800 mg m⁻² yr⁻¹; Fig. S24). For CAMx over land areas, the values reach 300 mg m⁻² yr⁻¹, and the lowest totals over land can be seen in the CHIMERE results (0.0 to 50 mg m⁻² yr⁻¹). Over water, these values are low in all model outputs (50 to 150 mg m⁻² yr⁻¹), except in CHIMERE, where, in contrast to the other models, the highest wet deposition was found over water.

The wet deposition of NO${}_{\mathrm{3}}^{-}$ is the highest for CMAQ (> 400 mg m⁻² yr⁻¹) over the whole domain (Fig. S25). For CAMx and LOTOS-EUROS, it is generally lower, with most areas displaying 25 to 50 mg m⁻² yr⁻¹. The lowest wet deposition of nitrate is shown in CHIMERE outputs with values not exceeding 50 mg m⁻². Regarding the sum for the whole year, the highest values are found for CMAQ (northern Italy and the Balkan Peninsula, where the urban-area values reached 400 mg m⁻² yr⁻¹). Over water, deposition is lower than over land in the results of all CTMs. Lower wintertime precipitation in CMAQ compared with the other models might lead to high particle concentrations as well as high deposition due to low dilution (Fig. 10).

Figure 10 Wet-deposition sum (mg per season) of particle species and precipitation divided by seasons and CTMs. “all_mean” displays the model ensemble. Spring: March, April, May; summer: June, July, August; autumn: September, October, November; winter: December, January, February. Wet deposition is based on the annual sum over the whole domain. Precipitation displays the seasonal sum (in mm).

Wet deposition depends mainly on the ability of the models to predict the amount, duration, and type of precipitation. The precipitation data show that the lowest values are found for CMAQ input data. CAMx and CHIMERE use the same meteorological input data and thus display the same precipitation results, with the highest values in winter. CMAQ and LOTOS-EUROS have precipitation values within a similar range, with the highest values occurring in autumn and winter.

Although the precipitation results in CAMx and CHIMERE are the same, wet deposition differed between these two models, indicating that the concentration as well as model internal mechanisms caused differences rather than the input data. Additionally, in CMAQ, a lower wet-deposition rate is expected for nitrate. There are usually two mechanisms that are important for scavenging in CMAQ: in-cloud and below-cloud scavenging. High wet deposition for nitrate in CMAQ outputs might be traced back to efficient below-cloud scavenging of coarse-mode particles containing nitrate, through which the wet deposition can be high despite precipitation in similar ranges to other models. Furthermore, the deposition of particulate nitrate crucially depends on the reactive uptake of HNO₃ to larger particles (Karl et al., 2019) because coarse-mode particles are removed much faster than fine-mode particles.

4 Discussion

Various reasons for deviations in PM_2.5 concentrations among regional CTM systems might be traced back to model-specific calculations.

Regarding PM (coarse and fine for sea salt), an uncertainty among models might be caused by the differences in the calculation of sea salt and dust emissions. Here, both are considered in all CTMs, except for dust in CMAQ. If sodium chloride and dust components are not considered, underestimations of PM and uncertainties in areas near coasts (sea salt) or where dust is important, e.g., Saharan dust in the Mediterranean region, occur, as described in Sect. 3.1. Furthermore, if sea salt and dust are omitted from the pH calculations, it might also cause deviations in sulfur chemistry, as this factor is very sensitive to pH.

In the CMAQ runs dust was considered at the model boundaries but dust emissions were not included. The Mediterranean region is frequently affected by Saharan desert dust (Palacios-Peña et al., 2019), but the main source region for this dust emission is not included in the model domain; thus the dust coming from the boundary can be seen as sufficient for the CMAQ model run. Generally, the boundary conditions for dust and sea salt in CAMx and CHIMERE were produced by offline models that run on meteorological fields from GEOS-5, GEOS DAS, and MERRA. For CMAQ and LOTOS-EUROS these boundary conditions were produced within the boundary condition calculations. Boundary conditions of EMEP are developed from climatological ozone-sonde datasets.

All models used offline meteorology in which the ABL heights were calculated. Annual medians of the atmospheric boundary layer heights at 16:00 and 04:00 UTC were compared among the models. The comparison of spatial distribution of ABL heights at 16:00 and 04:00 shows that over water, the ABL heights do not have much variability in all models (Figs. S26 and S27). The lowest ABL height over water was used for CHIMERE. This corresponds to the high PM_2.5 concentrations simulated by this model over water. Over land, the comparison of spatial distribution from 16:00 to 04:00 displays more variable ABL heights: during nighttime the ABL heights are up to 200 m, whereas during daytime the heights increase to 1000 m or higher (Figs. S26 and S27). Over land the input in CAMx, CHIMERE, CMAQ, and LOTOS-EUROS has a higher median ABL at 16:00, whereas in EMEP it is the opposite, with the highest median at 16:00 mainly being over water areas. However, there was not a large deviation in the PM_2.5 concentration simulated by EMEP from concentrations received from other models. Generally, due to ABL dynamics, deviations between measured and simulated data can be expected because measurement stations were chosen close to the coast, which leads to uncertainties. In these areas, the measurements are influenced by air masses either coming from water or coming from land. In addition, measured data were received from one measurement point, which is hardly representative of a whole grid cell of 12×12 km².

The treatment of dust and sea salt, as well as the boundary conditions used, has an effect on the analysis and comparison of PM results because these parameters are part of the PM_2.5 formation but differ among the models. Regarding the CTM performances, reasons for underestimations of PM_2.5 have already been discussed in previous studies: for CAMx, Pepe et al. (2019) linked these underestimations to meteorological parameters and to the overestimation of the vertical mixing in the lower atmosphere. Tuccella et al. (2019) found underestimations of PM_2.5 in the CHIMERE model and explained these by an excess of wet scavenging in the model. An excess of wet scavenging in CHIMERE compared with the other CTM systems is not found in the present study; thus it cannot be used as an explanation for deviations here. In EMEP, which differs from the other CTM systems, the MARS module was used to calculate the equilibrium between the gas and aerosol phases; this model does not treat sea salt or dust, leading to underestimations of PM_2.5. Kranenburg et al. (2013) linked the underestimation of particulate matter in LOTOS-EUROS to the missing descriptions of SOA processes in the model. Thus, various reasons and combinations of reasons can lead to underestimations of PM_2.5 in the CTM systems used herein. For a better understanding, the inorganic particle species are considered in the present study. Consideration of inorganic and organic particles could lead to more uncertainties. Moreover, in shipping emissions the inorganic aerosols display a higher share.

A large part of PM_2.5 is secondary; therefore underestimations can be linked to underestimations of precursors, e.g., NO₂. This has already been shown in the first part of this intercomparison study, where all five CTM systems underestimated measured NO₂ (Fink et al., 2023). But SO₂ is also usually underestimated by CTMs, as shown in previous studies (e.g., Eyring et al., 2007). Four out of five CTM systems underestimate the actual measured PM_2.5 concentration in the present study. Gaseous precursors like SO₂ and NO₂ need to be oxidized before they can form particles in reactions with ammonia. The hydroxyl radical (OH) is the main oxidant. The amount of available OH can be analyzed when the NO₂ concentration is set in relation to HNO₃ and NO${}_{\mathrm{3}}^{-}$ (Fig. S28). This gives an indication of the OH availability. In ship plumes OH is consumed fast; therefore values are low along the shipping lanes. In regions with lower NO₂ concentrations, more OH is available, and HNO₃ is efficiently formed. In the present study, the HNO₃ was similar within all five CTM systems (Fig. S5).

One reason for the differences in HNO₃ might be traced back to the amount of cloud droplets, since HNO₃ is resolved in them. The dissolution of gases in droplets is usually assumed to be irreversible for HNO₃ and NH₃ in CTMs; thus, the amount of formed ammonium nitrate mass depends on the amount of HNO₃ or the cloud droplets. This could in the end lead to the deviation among the CTM-simulated HNO₃.

The preference of NH${}_{\mathrm{4}}^{+}$ to bind to SO${}_{\mathrm{4}}^{\mathrm{2}-}$ in atmospheric aerosols to form (NH₄)₂SO₄ explains why in some models NO${}_{\mathrm{3}}^{-}$ displays relatively low values when the SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentration is high. CHIMERE, for instance, has an NO${}_{\mathrm{3}}^{-}$ share of 3.1 % to the total PM_2.5 and an SO${}_{\mathrm{4}}^{\mathrm{2}-}$ share of 27.0 %, whereas in the CAMx results, NO${}_{\mathrm{3}}^{-}$ has a share of 11.1 % to the total PM_2.5 and an SO${}_{\mathrm{4}}^{\mathrm{2}-}$ share of 14.6 %. This is confirmed by the low SO₂ concentration and high SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentration in CHIMERE (Figs. S11 and S19), indicating that sulfate is formed more efficiently compared with CAMx. Furthermore, this leads to lower NO${}_{\mathrm{3}}^{-}$ concentrations in the CHIMERE output (Fig. S21). For SO₂ and SO${}_{\mathrm{4}}^{\mathrm{2}-}$ the concentration of cloud water and the amount of cloud droplets also play an important role.

Regarding the thermodynamic equilibrium within the models, ISORROPIA and ISORROPIA II mechanisms are used in all CTM systems except EMEP, meaning similar results can be assumed to be obtained from this mechanism. Despite this similarity, differences in concentrations may be a result of differences in available cloud water, vertical mixing, the spatiotemporal distribution of emissions, or aerosol size distributions. EMEP uses the MARS module to calculate the equilibrium between the gas and aerosol phases. Although four of the five models use the ISORROPIA or ISORROPIA II mechanisms for inorganic secondary aerosol formation, many factors within these models still cause significant differences among the model outputs.

The aerosol size distribution also has an impact on the particle species distribution. As displayed in Table 1 (Sect. 2.1), there are two concepts for how the aerosol size distribution is represented within the models: the distribution is either in bins or in log-normal modes. As already discussed in Solazzo et al. (2012), the PM chemical composition differs greatly with the particle size. Consequently, differences in modeling the aerosol size distribution also affect the chemical composition. In CMAQ, for example, large fractions of nitrate and ammonium can be found in the coarse mode where they undergo different removal processes than in the fine mode.

Although there is harmonization in terms of the input emission data in the present study, the internal model mechanisms used to calculate particulate matter lead to differences in the particle species distribution, as discussed in Sect. 3.1. In addition, the calculations of how to determine PM_2.5 vary among CTM systems or even within one CTM. As an example, there are two possibilities for calculating PM_2.5 within CMAQ: either online during the model run with the PM_2.5 module or subsequently by calculating the value as the sum of two modes. These different options lead to different results (as shown by Jiang et al., 2006) and will also affect the particle composition. In the present study, the sum of two modes is used in CMAQ.

Model simulations with relatively high PM_2.5 concentrations display higher absolute shipping impacts on PM_2.5, as presented in Sect. 3.2. Consequently, relatively low variability in the relative potential ship impacts among the models compared with that of the absolute values could be expected. For a more quantitative evaluation, the relative potential ship impact is plotted against the absolute potential impact. A larger incline of the regression line can be explained by a higher background PM_2.5 concentration; thus the relative ship impact is lower for the same concentration increase (e.g., EMEP and CHIMERE) (Fig. S29).

From the ISORROPIA and ISORROPIA II mechanisms, it can be expected that the molar ratios between the acids on the one side (NO${}_{\mathrm{3}}^{-}$ and SO${}_{\mathrm{4}}^{\mathrm{2}-}$) and the base on the other side (NH${}_{\mathrm{4}}^{+}$) are in balance. However, the ratio between SO${}_{\mathrm{4}}^{\mathrm{2}-}$, NH${}_{\mathrm{4}}^{+}$, and NO${}_{\mathrm{3}}^{-}$ shows that the balance in all models except in LOTOS-EUROS is not given for PM_2.5; sulfate plus nitrate is much higher compared with ammonium (Fig. S30). This balance is almost perfectly given in LOTOS-EUROS, although both CMAQ and LOTOS-EUROS used the ISORROPIA II mechanism. An imbalance among the inorganic particle species is present, especially at the shipping lanes. Differences in the particle species ratio among the models can be traced back to the differences in the particle size distribution. In contrast to the other models, CAMx only has three species in the coarse mode: coarse others primary, coarse crustal, and reactive gaseous mercury. For NO${}_{\mathrm{3}}^{-}$ and SO${}_{\mathrm{4}}^{\mathrm{2}-}$, the ratio between the fine and coarse mode is calculated for the CTMs (Figs. S31 and S32). NH${}_{\mathrm{4}}^{+}$ was not considered here, since it is only in present in the coarse mode in CMAQ. These ratios show that CHIMERE and LOTOS-EUROS only have a small proportion of particles in the coarse mode. For SO${}_{\mathrm{4}}^{\mathrm{2}-}$ in LOTOS-EUROS the coarse particle concentration is zero and for EMEP no SO${}_{\mathrm{4}}^{\mathrm{2}-}$ is present in the coarse mode. In CMAQ a higher concentration of particles is assigned to the coarse mode and also for NH${}_{\mathrm{4}}^{+}$.

The present study has shown that different reasons can cause deviations among the simulated PM_2.5 CTM outputs. The major reasons are the differences in the size distribution and how models distribute chemical species between the coarse and fine modes (PM_2.5 and PM₁₀). Differences among the modeled PM_2.5 concentrations can also be a result of the differences in the height of the lowest model layer and the way in which ship emissions are distributed among the layers. As shown in Fink et al. (2023), the vertical distribution of PM_2.5 precursor emissions varies among the models; e.g., in CAMx all shipping emissions are assigned to the lowest layer. This leads to differences in chemical transformations because of different concentration levels close to the source and consequently to deviations among the particle distributions. Furthermore, precipitation differences lead to variations among the model outputs for wet deposition.

The limitations of the present study are that only the chemistry of the lowest layer is evaluated. The model input was standardized as far as possible, but meteorological input data varied and are not compared in detail here. Interactions between fine and coarse particles are only studied to a limited extent, and the same holds for aqueous chemistry, which has an impact on the oxidation mechanisms of sulfur species.

5 Summary and conclusion

The current work investigates and analyzes the predictions of five different CTM systems for PM_2.5 and inorganic particle species (NH${}_{\mathrm{4}}^{+}$, SO${}_{\mathrm{4}}^{\mathrm{2}-}$, NO${}_{\mathrm{3}}^{-}$) dispersion and transformation in the Mediterranean region. Additionally, the total concentration focus is on the potential ship impact. The results show that four of the five models underestimated the actual measured PM_2.5 concentrations at stations close to the European coastline. The relative ship impacts on PM_2.5 simulated by the CTMs at the measurement stations are between 5.7 % (CMAQ) and 13.8 % (CAMx). The potential impacts of PM_2.5 from ships simulated by CAMx, LOTOS-EUROS, and EMEP have the largest areas with values of up to 25 % along the main shipping routes in the Mediterranean Sea. CMAQ and CHIMERE simulated potential ship impacts of 15 % along the main shipping lines close to the African coast. These impacts are within the range of the ship impacts obtained in other studies.

The spatial distribution of ammonium displays a low total annual mean mainly in the southwestern part of the domain (approximately 0.0 µg m⁻³) and is the highest in the Po Valley and Bosporus (1.5 µg m⁻³ ). The ensemble mean of NH${}_{\mathrm{4}}^{+}$ (0.6 µg m⁻³ ) is in good agreement with the measurements provided in previous studies. The relative and absolute ship impacts are very similar for all models (0.0 to 0.06 µg m⁻³ over land, up to 0.15 µg m⁻³ over water; 0.25 % to 5.0 % over land, and 10 % to 25 % over water). This indicates that differences among the simulated PM_2.5 from ships result from differences in sulfate and nitrate.

The NH${}_{\mathrm{4}}^{+}$ proportion to total PM_2.5 is similar in all models (5.6 % to 7.8 %), and only LOTOS-EUROS shows a relatively high share (12.2 %). The ship impact pattern is similar; all models display proportions between 9.1 % and 12.6 %, but higher values are simulated by LOTOS-EUROS (23.5 %).

SO${}_{\mathrm{4}}^{\mathrm{2}-}$ is the main contributor to the total PM_2.5 concentration regarding shipping emissions only. In the model ensemble mean for the absolute ship concentration, SO${}_{\mathrm{4}}^{\mathrm{2}-}$ accounts for 44.6 % of PM_2.5. The annual mean sulfate total concentration is the highest for CHIMERE in the eastern part of the domain, reaching 6.0 µg m⁻³. CAMx, CMAQ, EMEP, and LOTOS-EUROS simulate a total SO${}_{\mathrm{4}}^{\mathrm{2}-}$ concentration within one range between 0.4 µg m⁻³ and 2.0 µg m⁻³ in the western part of the domain. The relative potential ship impacts on the total SO${}_{\mathrm{4}}^{\mathrm{2}-}$ are the lowest over land, with values of up to 3.0 %, and are higher in coastal areas, with values ranging from 6 % to 20 %. Along the main shipping routes, the impacts are the highest, reaching 50 % for CAMx, EMEP, and LOTOS-EUROS; for CHIMERE and CMAQ, they are lower, with values reaching 30 %. Regarding the absolute ship impacts on SO${}_{\mathrm{4}}^{\mathrm{2}-}$,, the model simulations display similar concentrations and are slightly lower for CMAQ and LOTOS-EUROS. Concentrations are in particular identified over water areas with relatively high SO₂ and SO${}_{\mathrm{4}}^{\mathrm{2}-}$. Because NH${}_{\mathrm{4}}^{+}$ preferentially binds to SO${}_{\mathrm{4}}^{\mathrm{2}-}$ in atmospheric aerosols to form (NH₄)₂SO₄, in areas over water less NH₄NO₃ forms.

The highest annual mean NO${}_{\mathrm{3}}^{-}$ total concentrations appear over land areas in the simulations by CAMx, CMAQ, and LOTOS-EUROS, especially over Italy and in the Balkan states (> 2 µg m⁻³). The lowest concentrations are simulated by CHIMERE. The concentrations over water are lower than those over land areas. The ensemble mean of all CTMs over the whole domain shows a median value of 0.63 µg m⁻³. Higher concentrations over land than over water are expected due to NH₄NO₃ formation in areas characterized by high-NH₃ and high-HNO₃ conditions and low-SO${}_{\mathrm{4}}^{\mathrm{2}-}$ conditions.

The relative potential ship impact on NO${}_{\mathrm{3}}^{-}$ differs among the models. The CAMx, EMEP, and LOTOS-EUROS results are similar; the relative potential ship impacts over land range from 0.0 % to 5.0 % (in the Balkan states), those in coastal areas and Italy range from 10 % to 25 %, and those along main shipping routes range from 50 % to 65 % or even reach 85 %. CHIMERE and CMAQ show lower relative potential ship impacts for NO${}_{\mathrm{3}}^{-}$. For CMAQ, the impacts are the lowest along the main shipping routes; nitrate is even reduced by 25 %. Low values in nitrate can be explained by the preference to form (NH₄)₂SO₄; thus nitrate stays in the gas phase or is transferred to the coarse mode. These low values for SO${}_{\mathrm{4}}^{\mathrm{2}-}$ and NO${}_{\mathrm{3}}^{-}$ in CMAQ but relatively high values for EMEP, CAMx, and LOTOS are reflected in the PM_2.5 ship impacts and partly explain the deviations in PM_2.5 among the models. As expected, the seasonal variabilities in particle species show that SO${}_{\mathrm{4}}^{\mathrm{2}-}$ and NH${}_{\mathrm{4}}^{+}$ are less temperature dependent than NO${}_{\mathrm{3}}^{-}$. Nitrate is higher in winter and spring but lower in summer and autumn. This pattern is found in all CTM simulations.

The spatial distribution of NH${}_{\mathrm{4}}^{+}$ wet deposition shows the highest annual sums by CMAQ and LOTOS-EUROS (up to 250 mg m⁻² yr⁻¹ over land; up to 50 mg m⁻² yr⁻¹ over water). CAMx and CHIMERE show a similar spatial distribution with values mainly between 10 and 25 mg m⁻² yr⁻¹. For wet deposition of SO${}_{\mathrm{4}}^{\mathrm{2}-}$, the annual totals for all emission sources are the highest over the Balkan Peninsula in the CMAQ and LOTOS-EUROS model outputs (300 to 800 mg m⁻² yr⁻¹). For CAMx over land areas, the values reach 300 mg m⁻² yr⁻¹, and the lowest totals over land can be seen in the CHIMERE results (0.0 to 50 mg m⁻² yr⁻¹). Over water, these values are low in all model outputs (50 to 150 mg m⁻² yr⁻¹), except in CHIMERE. Wet deposition of NO${}_{\mathrm{3}}^{-}$ is the highest for CMAQ (> 400 mg m⁻² yr⁻¹) over the whole domain. For CAMx and LOTOS-EUROS, it is generally lower, with most areas displaying 25 to 50 mg m⁻² yr⁻¹. The lowest wet deposition of nitrate is shown in CHIMERE outputs, with values not exceeding 50 mg m⁻². Over water, deposition is lower than over land in the results of all CTMs.

The complexity of particle treatments within the models, as well as the large number of causes for these changes, makes it difficult to find a single cause for the variable outputs. One point causing uncertainties is that the aerosol formation mechanisms differ among CTMs. The detailed investigation of PM_2.5 and its chemical composition has demonstrated that differences among the particle species might be traced back to the aerosol size distribution. This was shown in particular for CMAQ regarding the balance of the inorganic particle species nitrate and sulfate on the one side and ammonium on the other side. CMAQ and EMEP tend to assign a higher particle mass to the coarse mode compared with the other three CTMs. This has implications for particle deposition because both wet and dry depositions are more efficient for larger particles.

An ensemble mean with standard deviations based on several model results can provide a more reliable assessment of possible ship impacts on air concentration and deposition. Previous research has demonstrated that using only one chemical transport model resulted in underestimated model uncertainty and overconfidence in the conclusions (e.g., Solazzo et al., 2013; Riccio et al., 2012; Solazzo et al., 2018), indicating that a model ensemble should be used. Particularly in terms of the study's policy point of view, the ensemble mean is important: if model simulations are used to support decision-making regarding shipping regulations, the uncertainty in individual models must be considered.

The goal of this study was not to make model outputs as similar as possible but to show the discrepancies that occur among CTM systems despite using similar input data. Different CTM systems were asked the same question to find the impact of shipping, for which they got the same emissions as input data.

Nevertheless, to achieve less-varying results in future studies, the vertical emission distribution as well as the boundary conditions could be the same in all CTM inputs. This can help to make the modeled output more similar. Adjustments in using the same meteorology could also be helpful, yet difficult to realize, since the meteorology and meteorological driver within each CTM system are closely connected. To gain more insight into certain mechanisms, one model could be used with, e.g., changing vertical profiles, emissions, or meteorology. Furthermore, the present study does not use the same boundary conditions, and the models also do not use the same sea salt or dust emissions. For more consistent investigations of model results, future intercomparison studies should be carried out using the same boundary conditions as well as sea salt and dust emissions as input data.

Regional-scale models with relatively coarse grid resolutions do not account for chemical transformation mechanisms within a ship's exhaust gas plume. They typically assume direct dilution and neglect the in-plume chemistry at high-pollutant concentration levels. To obtain more precise information regarding the effects of shipping on particle concentrations, the particle size distribution and the interaction mechanisms from plume to background concentrations, as well as chemical transformations within ship plumes, should be considered in future studies.

Appendix A: List of abbreviations

Abbreviation	Description
1.5-D	1.5-dimensional
ASOA	Anthropogenic secondary organic aerosol
BSOA	Biogenic secondary organic aerosol
CAMx	Comprehensive Air Quality Model with Extensions
CB05	Carbon bond mechanism 05
CBM-IV	Carbon bond mechanism (version IV)
CMAQ	Community Multiscale Air Quality model
CTM	Chemical transport model

ECMWF	European Centre for Medium-Range Weather Forecasts
EEA	European Environment Agency
EMEP	European Monitoring and Evaluation Programme model
EU	European Union
GEOS-5	Goddard Earth Observing System Model, Version 5
GEOS DAS	Goddard Earth Observing System Data Assimilation System
GNFR	Gridded Nomenclature for Reporting
IFS-CAMS	Integrated Forecasting System – Copernicus Atmosphere Monitoring Service
LMDz-INCA	Laboratoire de Météorologie Dynamique General Circulation Model – INteraction with Chemistry and Aerosols
MARS	Model for an Aerosol Reacting System
MEGAN	Model of Emissions of Gases and Aerosols from Nature
MEPC	Marine Environment Protection Committee
MERRA	Modern-Era Retrospective Analysis for Research and Applications
MSC-W	Meteorological Synthesizing Centre – West
NMB	Normalized mean bias
NMVOC	Non-methane volatile organic compound
PM	Particulate matter
POA	Primary organic aerosol
PSAT	Particulate source apportionment technology
R	Spearman's correlation coefficient
RADM–AQ	Regional Acid Deposition Model−Aqueous Chemistry
RMSE	Root mean square error
SCIPPER	Shipping Contributions to Inland Pollution – Push for the Enforcement of Regulations
SOA	Secondary organic aerosol
SOAP	Secondary Organic Aerosol Processor
STEAM	Ship Traffic Emission Assessment Model
TNO	Nederlandse Organisatie voor Toegepast Natuurwetenschappelijk Onderzoek
TPPM	Total primary particulate matter
VBS	Volatility basis set
VOC	Volatile organic compound
WHO	World Health Organization

Figure A1 Domains and measurement stations. The red trapezoid displays the 12×12 km² domain, the blue icons are the locations of the measurement stations. On the bottom left the larger 36×36 km² domain is displayed. Map source: ArcGIS Pro 2.7.1, Esri (2020).

Appendix B

Table B1 Detailed overview of monitoring stations.