The chemical composition of the submicron fraction of the aerosol was studied using three different analysers. Concentrations of non-refractory species, SO42-, nitrate (NO3-), ammonium (NH4+), chloride (Cl⁻), and OA, were measured at the PEB station, with a high-resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS, Aerodyne) with a 30 s time resolution (DeWitt et al., 2015). The metal composition of the submicron particles was also determined at the PEB station using an online energy-dispersive X-ray fluorescence (EDXRF) spectrometer (Xact 625i, Cooper Environment) (Tremper et al., 2018) with a 30 min time resolution. Equivalent black carbon (BC) measurements at a 1 min time resolution were performed with a multiangle absorption photometer (MAAP 5012, ThermoFischer) (Petzold and Schönlinner, 2004) at the PEB station and with a dual-spot seven-wavelength aethalometer (AE33, Magee Scientific) (Drinovec et al., 2015) at the MAJOR station.

Particle number (PN) concentrations were measured by ultrafine condensation particle counters (CPCs) in the size range of 2.5 nm to 3 µm at PEB (CPC 3776, TSI) and from 7 to 2500 nm at MAJOR (Envi CPC 200, Palas), with a temporal resolution of 1 s. The aerosol number size distribution was measured at both sites (i) in the range of 15–660 nm using scanning mobility particle sizers (SMPSs 3936, L-DMA, CPC, TSI) with a scan time of 2 min for 105 channels and (ii) in the range of 250 nm to 3.2 µm using an optical particulate counter (OPC model 1.109, Grimm Aerosol Technik) with a scan time of 1 min for 31 channels. Particle mass concentrations (PM₁, PM_2.5, PM₁₀) were also estimated by the OPC.

Non-methane volatile organic compounds (NMVOCs) were monitored with a proton transfer reaction time-of-flight mass spectrometer (PTR-ToF-MS 8000, Ionicon Analytik, Austria) at a 10 s time resolution. An overview of the PTR-ToF-MS operation and data analysis can be found in Marques et al. (2022). The main organic molecules detected during the measurement period are listed in Table S3. CO₂, CO, methane (CH₄), and ammonia (NH₃) were measured at a temporal resolution of 5 s by a cavity ring-down spectrometer (model G2103 for NH₃ and model G2401 for CO₂, CO, CH₄, H₂O; Picarro) (Martin et al., 2016). Concentrations of regulatory gaseous pollutants were measured at a 10 s time resolution by a chemiluminescence analyser (model 200E, Teledyne API) for the combined measurement of nitrogen oxide (NO_x, NO, and NO₂), by an absorption spectrometry monitor (model 400E, Teledyne API) for ozone (O₃), and by a fluorescence analyser (model AF22 Environment SA at PEB station and model 100E Teledyne API at MAJOR station) for sulfur dioxide (SO₂).

AIS data records from all ships during the measurement period (31 May–3 July 2021) that were within a 10 km × 10 km area surrounding the port of Marseille were purchased commercially from MarineTraffic (2022). The requested AIS database contained 326 590 records, each containing the following information: the vessel's identification number (Maritime Mobile Service Identity, MMSI), position (latitude and longitude), date and time, status, heading and course angles, speed, and the last and next ports visited. To improve the quality of the AIS data, the dataset was (i) pre-processed to exclude sailing vessels and pleasure craft and to remove data redundancy and noise, (ii) interpolated at a time step of 90 s to remove trajectory outliers and recover lost AIS data using the PyVT tool developed by Li et al. (2023), and (iii) cross-referenced with the ship arrival and departure data supplied by GPMM. Additional vessel parameters such as name, category, year the vessel was built, and engine and engine power (in kW) were retrieved from the MMSI provided in the database.

Plume identification was achieved by cross-referencing measurement data with meteorological and AIS data (Ausmeel et al., 2019; Celik et al., 2020; Eger et al., 2023; Krause et al., 2023). The selection steps and criteria used are detailed below. 1.

Plume pre-selection was done using four typical tracers of ship activity (CO₂, NO_x, BC, PN) and O₃. i.

Calculation of atmospheric background was performed using a lowpass-filtered time series in the form of a rolling median with a 60 min window size for each of the five selected pollutants. This lowpass filter describes the variability in the background concentration due to atmospheric physico-chemical processes and regional transport of pollutants but excludes the short-term variation caused by passing ships (Krause et al., 2023).

The most common approach to characterize the chemical composition of ship plumes is that of emission factors (Ausmeel et al., 2019; Celik et al., 2020; Pirjola et al., 2014; Van Roy et al., 2022). This approach, described in detail by Celik et al. (2020), is based on the carbon balance method, which rules out plume dilution and background contributions. For each plume that was successfully assigned to a single ship or a ship category, the emission factor of each pollutant x (EF_x), expressed in grams or particle number per kilogram of fuel used, was derived from Eq. (1) (Diesch et al., 2013). 1EFx=∫SEx(t)dt∫SECO2(t)dt⋅MCMCO2⋅wc, where wc represents the carbon mass fraction of fuel used (fixed at 0.865 kg C kgfuel-1; Celik et al., 2020; Grigoriadis et al., 2021b); MC/MCO2 is the mass fraction of carbon in CO₂; x is the excess (above the atmospheric background) concentration of pollutant x in µg m⁻³ for mass concentrations and in 10¹² particles cm⁻³ for number concentrations; CO2 is the excess CO₂ concentration in mg m⁻³; and the indices S and E denote the start and end of the plume, respectively.

Plume start and end dates (indices S and E) were retrieved from the point of inflection of the concentration to the time curve, i.e. when the derivative crosses zero, to avoid any subjectivity. EFs were calculated separately for each pollutant measured, accounting for the response times of each instrument (Ježek et al., 2015). Then the background to be subtracted from the signal to determine excess concentrations was defined by the mean concentrations of two 30 s background intervals before and after the peak. A toolkit has been developed in the Igor Pro 8 environment (WaveMetrics, USA) to systematically perform these calculations.

Furthermore, particle size distributions in ship emission plumes were computed by applying the same method to SMPS measurements. However, the start and end times for defining the background and plume intervals were set identically for all size classes based on those defined for the CPC analyser. This is because the CPC analyser counts the total number of particles regardless of size and provides a more accurate time resolution (10 s). As the plume duration can approach the 2 min scan time of the SMPS, additional checks were applied. For each plume, the total number of particles measured by the SMPS and the CPC were compared. Only plumes with a Pearson correlation coefficient greater than 0.7 were selected for the analysis.

The short-term impact of shipping plumes on ambient air concentrations levels was assessed for each single plume using Eq. (2). 2[x]‾=1E-S∫SEXtdt, with [x]‾ the average excess concentrations and x(t) the excess concentrations over the plume duration period (E–S). Plume start and end dates, as well as the background corrections, have been determined using the method described in Sect. 2.3.2.

This approach has certain limitations. It excludes a significant number of ship plumes from the analysis, particularly during periods of heavy maritime traffic, which could skew the estimated additional average concentrations. However, this exclusion ensures that the results are not biased by non-ship sources resembling ships emissions (Eger et al., 2023).

Plume detection is based on both meteorological and ship traffic conditions. The meteorological conditions observed during the campaign highlight the complexity of air mass circulation in the study area. According to Fig. S1, wind directions vary considerably within the port area, especially during land and sea breezes, which are common at this time of year. Wind direction analysis during the campaign enabled us to estimate the probability of the measurement stations being downwind of Marseille port's main ship emission areas (located in Fig. S2). Combining all areas, this probability was 35 % at the PEB station and 20 % at the MAJOR station. Detailed probabilities for each zone are detailed in Table S5.

In June 2021, approximately 800 ships arrivals and departures were recorded by the GPMM. That excludes pilot boats that systematically escort vessels from the lane entrance to the mooring berth and back, as well as pleasure craft and passenger shuttles that mainly access the marinas of Vieux Port, Estaque, and the Frioul Islands (located in Fig. S2), which are located outside the GPMM sector. Most ships operating in the GPMM were dedicated to passenger transport (40 %), including ferries and cruise ships (35 % and 5 %, respectively). Cargo ships represent 25 % of the activities, while tugs and tankers accounted for 20 % and 10 % of port movements, respectively. Other vessels primarily used for sea rescue made up the remaining 5 %. On average, ship arrivals peaked early in the morning from 04:00 to 06:00 UTC, and departures were most frequent late in the afternoon from 16:00 to 18:00 UTC. This schedule varies slightly according to ship category, as shown in Fig. S3.

A total of almost 110 chemical components were measured during the campaign, including 45 trace-metal elements (or metals) and 41 NMVOCs. The main statistics of pollutant concentrations measured at both PEB and MAJOR stations are reported in Table S6. Time series of the key substances and the particle size distribution over the whole campaign are given in Figs. S4 and S5.

To investigate the impact of port activities on local air quality, concentration levels of pollutants measured simultaneously at the two stations were compared with those from the MRS-LCP station (Table S6), part of the regional air quality network. This monitoring station, located in the city's centre (Fig. 1), was chosen because it serves as a reference for urban background pollution (Chazeau et al., 2021). Average concentrations of PN, PM_2.5, PM₁, and NO_x at PEB and MAJOR were 1.5 to 2 times higher than those measured at MRS-LCP, indicating a significant influence of port activities. In contrast, average levels of other compounds were similar across all stations. Additionally, analysis of maximum concentrations and the 75th percentile reveals that pollutants such as SO₂ and some metals (As, Cd, Co, Fe, Ni, Sb, Se, Sn, V, Zn, and Zr) are 2 to 10 times higher near the port than they are downtown.

To identify the sources responsible for the high concentrations in the port area, the conditional bivariate probability function (CBPF; Uria-Tellaetxe and Carslaw, 2014) was computed for measured species. The CBPF is a polar coordinate graphical method commonly used to highlight wind speeds and directions associated with high concentrations of a pollutant in order to identify emission sources (Adotey et al., 2022; Ryder et al., 2020; Toscano et al., 2022). It estimates the probability that measured concentrations exceed a predetermined threshold (in this case the 80th percentile) for a given range of wind sectors and wind speeds. As shown Fig. 4, CBPF indicates that the highest concentrations typically occur when the measurement sites are downwind of the mooring berths or of the ships' access lanes to the port (located in Fig. S2), except for OA, which also had high concentrations associated with land breezes (to and from the city). These findings are supported by analysis of the daily profiles of concentrations (Fig. S6), showing a correlation with ship arrival and departure profiles (Fig. S3) depending on the pollutants.

This correlation is particularly strong for NO, BC, PN, vanadium (V), and nickel (Ni). The increased dispersion of concentrations and maxima and the differences between mean and median concentrations during ship movements underscore the significant impact of their emissions on the concentrations measured in the port. For the other substances, such as SO₂ and NMVOCs (e.g. toluene) in the gaseous phase and SO42- or OA in the particulate phase, correlations exist but to a lesser extent. For OA and NMVOCs (e.g. toluene), average concentrations slightly increase during ship movements but rise more notably at night (with greater dispersion and extrema) due to the nighttime land breeze regime bringing urban emissions back to the measurement station. In the case of SO₂, average concentrations and peaks also slightly increase during ship movements but are enhanced over the morning and are longer than for other compounds, suggesting contributions from other sources. This could be due to sea breezes lifting emissions from the large industrial areas located 25 km north-west of Marseille, which had been previously pushed away to sea by nighttime land breezes (Chazeau et al., 2021). Unlike SO₂, SO42- does not follow the same daily cycle, instead showing concentrations increasing in the afternoon, mainly driven by its photochemical formation cycle. It is worth noting that the reduction in sulfur content in fuels in 2020 could lead to lower contributions from ships compared to other sources. Indeed, as shown in Fig. 3b by the flat pattern of the daily profile observed after the new regulations, SO₂ may become a less effective tracer of ship emissions.

From the measurements taken at the two stations (PEB and MAJOR) and following the procedure described in Sect. 2.3.1, 118 plumes were attributed to ships at berth and 235 plumes to ships manoeuvring (ships at speeds below 2.5 m s⁻¹ (5 kn) or navigating (ships at speeds between 2.5 and 14 m s⁻¹) in or near the port area (less than 750 m from the coast).

The yield of plumes attributed to ships manoeuvring and/or navigating is 29 % of the number of passages recorded by the AIS data (see Sect. 3.1.1). This relatively low yield is primarily due to the dependence on wind direction at the time of the ship's passage. The plume can only be captured when the wind direction positions the measurement station downwind of the ship's emissions. This dependency, combined with the stringent criteria of the plume assignment algorithm, resulted in the exclusion of many plumes, particularly during periods of heavy ship traffic. All these precautions ensure the robustness of the emission factors (EFs), especially when analysing EFs as a function of ship characteristics. These restrictions do not affect the EF values themselves, as the number of plumes detected does not influence the EF values, which are defined through normalization by CO₂.

The plume samples span eight different ship categories, among which are the three main categories operating at GPMM (ro-ro ferries, cargo ships, and cruise ships). However, plume frequency varies by category, with ro-ro ferries comparable to cruise ships, together accounting for 70 % of the sample plumes. The remaining categories in descending order of frequency include pleasure craft, cargo ships, tankers, passenger shuttles, tugs, and rescue vessels. The sample also encompasses the various operational phases observed in a port: at berth (33 %), manoeuvring (16 %), and navigating (51 %), with ships generally maintaining speeds below 5 m s⁻¹ when not docked due to regulatory speed limits in port. This reduced speed generally corresponds to an engine load of less than 25 % (Jeong et al., 2023; Knudsen et al., 2022; Lack and Corbett, 2012). However, the breakdown of operational phases varies by ship category. For cruise ships, the at-berth phase is most common, accounting for 90 % of their time. In contrast, the navigation phase predominates for most other categories of ship (80 %) except for tankers, which show an even distribution between navigation and manoeuvring.

Figure S7 shows the example of two successive plumes from different ro-ro ferries arriving at the port. As expected, these emissions mainly consist of a blend of gaseous compounds (such as CO₂, CO, NO_x, SO₂) and particulate compounds (BC, SO42-, OA), accompanied by a decrease in O₃ levels. The particle size distribution from these ships shows a bimodal pattern (20 and 80 nm) for the first ship, with sulfur predominantly emitted as SO42-, and an unimodal pattern (20 nm) for the second ship, where sulfur is mainly emitted as SO₂. Regarding metals, the 30 min time resolution of the analyser prevents us from distinguishing emissions between the two ships but reveals the presence of V, Ni, calcium (Ca), and iron (Fe). In terms of NMVOCs, only fragments of unspecified hydrocarbons are detectable.

Figure 5 shows the distribution of EFs across all gaseous compounds (see Table S10 for detailed statistics). NMVOCs and NH₃ are excluded from these plots because their median and percentile values are below the detection limits, except for aromatic C₈ compounds and toluene. These exceptions are discussed in more detail below. The large variability in EFs observed is consistent with findings in the literature (Table S8) and is further accentuated when EFs are compared without considering the specific characteristics of the ships or the fuels used.

For NO_x, the median EF of the plumes studied (37 g kgfuel-1 (28–48)) is comparable to the low range of values reported in the literature for ships using fuel oil (average of 57 ± 26 g kgfuel-1) and 2 times lower than the value from the regional emissions (80 g kgfuel-1) inventory, which lies in the high range of the values in the literature. Only 2 % of the 353 EFs determined from ship plumes exceed this value.

The median EF_NO (14 g kgfuel-1 (9–16)) is significantly lower than the values reported in the literature (average of 70 ± 34 g kgfuel-1), as most of these values stem from measurements taken directly from the exhaust of ship engines, where NO_x emissions are more than 90 % NO (Zhao et al., 2020). This balance is quickly altered in the atmosphere, where NO oxidizes rapidly to NO₂. In this case, the oxidation reaction is driven by O₃ because the initial NO concentrations in the exhaust gases are much higher than the ambient O₃ concentrations, which leads to a local reduction in O₃ concentrations and a decrease in the NO/NOx ratio (see Fig. S7a). For this reason, most studies explicitly report only the EF_NOx. Nevertheless, field campaign studies that take this oxidation into account report EF_NO on the same order of magnitude as that determined in this study (7 ± 1 g kgfuel-1 (Celik et al., 2020) and 16 ± 11 g kgfuel-1 (Diesch et al., 2013)).

Our results showed that the main influence factor is the plume age (or residence time) and to a lesser extent the ship category. Beside these factors, other studies have also pointed out the influence of the operational phase, ship speed/engine load, and engine (Celik et al., 2020; Grigoriadis et al., 2021a; Huang et al., 2018; Peng et al., 2020; Sugrue et al., 2022).

As shown in Fig. 6, as plume age increases, both EF_NOx and the NO/NOx ratio decrease during daytime (differences between adjacent classes are not statistically significant but become so with the most widely separated classes), while remaining stable during nighttime (all the groups have the same central tendency). During daytime, the NO/NOx ratio decreases rapidly, from 0.9 at emission (Zhao et al., 2020) to 0.4 for the youngest plumes and 0.3 for plumes older than 15 min. This latter value is close to the one corresponding to photochemical equilibrium (0.2) suggested by Celik et al. (2020) for plumes older than 30 min. In addition, the diurnal 2-fold decrease in EF_NOx between the shortest and longest plume age suggests the existence of NO_x sinks involving photochemical reactions, with radicals such as OH, HO₂, and RO₂ or with NMVOCs (Celik et al., 2020). These reactions could lead to the formation of nitric acid and, through heterogeneous reactions, to the production of aerosols containing nitrates or organo-nitrates.

In addition, as shown in Fig. 7, the categories grouped under the other category exhibit statistically lower EF_NOx compared to cruise ships, cargo ships, and ferries. Specifically, the two categories within this group, pleasure craft and passenger shuttles, have the lowest emissions: EF_NOx is 1.3 times lower than that of all the other categories combined (31 g kgfuel-1 (23–35) vs. 40 g kgfuel-1 (32–51)). This difference is likely due to the low engine power of these ships (Sinha et al., 2003) or the type of fuel used (petrol can be used for pleasure craft). All other categories – including those within the other category group except for pleasure craft and passenger shuttles – show a similar central trend. This figure also suggests that the operational phases do not significantly affect EFs and confirms that EF_NOx values for vessels at berth or operating with an engine load below 30 % are of a similar magnitude (Grigoriadis et al., 2021a) (similar central trend). It should be noted that for ferries at berth, the observed EFs are lower due to the limited number of plumes identified, coupled with a higher proportion of plumes older than 10 min detected during the day.

Finally, analysis of the ship construction year retrieved from the AIS ship tracking data indicates that the EFNOx is not affected by the tier regulations on NO_x production imposed by MARPOL (Tier 0 ships built before 2000, Tier I before 2011, Tier II before 2016, and Tier III after 2016) (similar central trend). This result suggests, as previously pointed out by Knudsen et al. (2022) and Sugrue et al. (2022), that these regulations have little influence on NO_x emissions, particularly in the case of low engine loads and for the Tier 0 and Tier 1 ship categories (the most represented categories (90 % of the plumes studied for which the ship construction year was known)).

The median SO₂ emission factor of the plumes studied (0.4 g kgfuel-1 (<0.1–0.7)) is comparable to the average value reported in the literature for ships using fuel oil with sulfur contents <0.1 % (1.1 ± 1.0 g kgfuel-1). It is nevertheless more than 5 times lower than the value provided by the regional emissions inventory (2.0 g kgfuel-1) for the same fuel sulfur content category.

Analysis of the parameters likely to influence SO₂ EFs reveals that its emissions depend mainly on the ship category and operational phase. As shown in Fig. 8, ships equipped with engines of more than 10 000 kW (cruisers, ferries, and cargo ships) emit statistically significantly more SO₂ (p value < 0.05) than ships with engines of less than 4000 kW (the other category). In addition, ships at berth using their auxiliary engines generate less SO₂ than ships manoeuvring/navigating using mainly their main engine. Regarding the manoeuvring/navigation phases, a significant statistical difference in SO₂ emissions is observed between arrivals and departures, with higher emissions during arrivals. This distinction is more pronounced when the number of plumes identified in these phases is high, as is the case for ferries. For cruise ships, cargo ships, and tankers (included in the other category), the same trend is observed, although the number of plumes on arrival and departure is lower. The difference in SO₂ emissions between arrivals and departures could reflect the transitional period when the ship makes the required fuel change in the Marseilles harbour (switching from fuels with a sulfur content of 0.5 % to 0.1 % in response to regulations) and/or could be linked to the use of open-loop scrubbers, which are required to be shut down in the Marseilles harbour. When this type of scrubber is shut down, a fuel switch from HFO to MGO is typically performed beforehand. However, a temporary increase in SO₂ emissions can be observed (Teinilä et al., 2018), attributed to engine system purging during the transition process.

In conclusion, as highlighted by numerous studies (Grigoriadis et al., 2021a; Huang et al., 2018; Zhao et al., 2020), SO₂ emissions depend mainly on the sulfur content of the fuel used. During the campaign, all the plumes from ships at berth had sulfur contents < 0.1 %. For ships manoeuvring/navigating, sulfur levels were systematically <0.5 %, and only 10 % of the plumes measured had sulfur levels above 0.1 %, mainly including incoming ships.

The median of the CO emission factors (EF_CO) for the plumes studied (5.4 g kgfuel-1 (<1.3–9.3)) is comparable to the values reported in the literature for ships using fuel with a sulfur content of <0.1 % (average of 5.7 ± 8.6 g kgfuel-1), as well as being comparable to the value used in the regional emissions inventory (7.5 g kgfuel-1) for this type of fuel.

Analysis of the parameters likely to influence EF_CO reveals that CO emissions depend mainly on the operational phase (Fig. 9). Ships at berth emit statistically less CO than ships manoeuvring/navigating (p value < 0.05). As fuel type has little or no influence on CO emissions (Petzold et al., 2011), the variation in EF_CO within the operational phases is attributable to engine load. An increase in engine load leads to a rise in combustion temperature, making it more efficient and thus reducing CO (Agrawal et al., 2010; Zetterdahl et al., 2016). Ships at berth, which mainly use their auxiliary engines operating at a stable and optimal engine load, thus emit less CO than ships manoeuvring/navigating using their main engine at lower and less stable loads. The effect of combustion temperature is also reflected within the manoeuvring/navigation phase, with higher emissions observed on departure than on arrival, probably due to the cold start of the main engines and the resulting incomplete-combustion conditions. In addition, plumes with particularly high emission factors (30–100 g kgfuel-1) have been observed, likely corresponding to changes in engine speed during acceleration or deceleration phases (Bai et al., 2020; Huang et al., 2018; Jiang et al., 2021), as these plumes are systematically captured at the port exit.

The methane median of the EFs of the plumes studied (0.4 g kgfuel-1) is comparable to the EFs reported in the literature for ships using fuel oil (average of 0.2 ± 0.4 g kgfuel-1) and aligns with the value used in the regional emissions inventories (0.3 g kgfuel-1). However, the mean of the EFs (1.3 ± 0.3 g kgfuel-1) is higher because 10 % of the plumes exhibit EFs >1.0 g kgfuel-1, with some reaching up to 23 g kgfuel-1. Analysis of the various ship parameters did not reveal (p value >0.05) any significant cluster driver for EFs >1.0 g kgfuel-1. Furthermore, the distribution analysis of NO_x and SO₂ EFs does not indicate the use of methane-based fuels, such as liquefied natural gas (LNG) or gas to liquids (GTLs), which are typically associated with higher CH₄ EFs but lower NO_x and SO₂ EFs. For further details on the other hypotheses considered based on our knowledge of the study area that could potentially be combined to explain the higher observed EFs, please refer to Table S12.

For the majority of NMVOCs, the mean, median, and percentile of the EFs observed with the plumes are below the detection limits (DLs), which vary between 5 and 200 mg kgfuel-1 depending on the compounds (Table S10). These results are consistent with EFs reported in the literature, where values derived exclusively from direct emission measurements are generally below 30 mg kgfuel-1 (Agrawal et al., 2008a, 2010; Huang et al., 2018; Timonen et al., 2022).

However, for certain compounds such as C₈ aromatics and toluene, the 90th percentiles (P90) exceed the detection limits, suggesting that these compounds may occasionally be emitted in greater quantities. Nevertheless, due to the detection limits and measurement uncertainties for these compounds, reliably determining the parameters associated with higher emissions remains statistically challenging. However, among the parameters examined, the operational phase – and consequently the type of engines and fuel used – appears to exert the most significant influence on P90. Indeed, a detailed examination of EFs by operational phase, as shown in Table S11, indicates the following.

For toluene, EFs from ships in the manoeuvring phase are higher than in other operating phases. Huang et al. (2018) and Timonen et al. (2022) have also highlighted this when studying the emissions of ships (cargo ships and ro-ro ferries) during different operating phases. The study by Huang et al. (2018) further showed that, in the manoeuvring phase, the quantity of toluene emitted was 4 times higher when low-sulfur fuel (0.4 %) was used compared to when higher-sulfur fuels (1.1 %) were used.

The ammonia emission factors of the plumes analysed are systematically below the DL of 0.1 g kgfuel-1, except for two plumes with EFs close to this threshold, with values of 0.12 and 0.15 g kgfuel-1. These values are in line with the literature, which reports an average of 0.07 ± 0.14 g kgfuel-1 (across all fuel types).

Figure 10 provides an overview of the median global emission profile of gaseous phases and illustrates its variability according to the operational phases of ships, which most commonly account for the variations in emission factors (EFs) of different compounds. Ship gaseous emissions are primarily composed of NO_x (86 %) and CO (12 %), while SO₂ and CH₄ each represent about 1 %. Other compounds, such as NMVOCs, constitute less than 0.1 % of the gaseous phase but can account for up to 10 % under certain operational conditions only identified when ships were at berth or manoeuvring, which may significantly impact the formation of secondary pollutants.

EF_PN obtained from particle counter measurements (CPC) and EF_PM1 obtained from particle size measurements (SMPS) are illustrated using boxplots in Fig. 11 (for detailed statistical analysis, refer to Table S10).

The median EF_PN (6.7×1015 particles kgfuel-1 (4.2–10.8)) is comparable to the values reported in the literature for ships using fuel oil with sulfur contents <0.1 % (mean of 8.1 ± 14.1×1015 particles kgfuel-1).

Regarding the overall submicron particle mass (PM₁) emission factors, the use of the SMPS analyser was preferred over the OPC one for the calculation. Inter-comparisons showed that the OPC analyser could underestimate PM₁ concentrations by up to a factor of 3, particularly when the measurement sites were downwind of ship plumes, due to its inability to measure particles smaller than 250 nm in diameter. The median of the plume EF_PM1 thus obtained (1.0 g kgfuel-1 (<0.4–2.25)) is comparable to the high range of values reported in the literature for ships using fuel oil with sulfur contents <0.1 % (average of 0.6 ± 0.2 g kgfuel-1). Considering the PM1/PM2.5 ratio (average 0.8) derived from the OPC measurements, the estimated EF_PM1 is also in line with the EF_PM2.5 (1.4 g kgfuel-1) considered in regional emissions inventories.

Analysis of the parameters likely to influence EF_PN and EF_PM1 reveals a statistically significant dependence of emissions on the operational phase. As shown in Fig. 12, ships at berth generate more PN and less PM₁ than when they are manoeuvring/navigating (9.9×1015 particles kgfuel-1 (7.9–12.1) vs. 5.2×1015 particles kgfuel-1 (3.4–8.1) for EF_PN and 0.6 g kgfuel-1 (<0.4–0.9) vs. 1.7 g kgfuel-1 (0.8–3.3) for EF_PM1). It should be noted that no systematic bias related to the age of the plumes has been observed. The distribution of plumes in the different plume age classes is similar for ships at berth and those manoeuvring/navigating.

These variations can be explained mainly by the use of auxiliary engines operating at a stable and optimum engine speed with low-sulfur-distillate fuel (<0.1 %) at berth, whereas ships manoeuvring/navigating within the port area use their main engine at a lower (<25 %) and less stable engine load. These observations indicate that PN and PM₁ emissions are influenced by engine load (Anderson et al., 2015a; Grigoriadis et al., 2021a; Zetterdahl et al., 2016) and engine speed (Diesch et al., 2013). The contrasting evolution of the mass and number of particles observed in Fig. 12 between ships at berth and manoeuvring/navigating has also been highlighted by Anderson et al. (2015b) and Chu-Van et al. (2018). This unusual development is attributable to the particle size distribution (see Sect. 3.2.4) and not to particle formation. When ships are docked, particles are mainly smaller than 50 nm in diameter, whereas when manoeuvring/navigating, a mode around 100 nm appears and may even become predominant, thus contributing more to the total mass of PM₁ than finer particles do.

In addition, as is the case with SO₂, a difference was observed between emissions of PN and PM₁ during arrivals and departures, with arrivals showing statically higher emissions (Fig. 12). However, for cruise ships, this trend does not hold, but the small number of plumes identified for this category during these operational phases makes this result uncertain. In the analysis of the evolution of SO₂, two hypotheses were considered to explain the differences between arrivals and departures: fuel transition and the shutdown of open-loop scrubbers. The fact that this distinction between arrivals and departures is equally marked for both the number and the mass of particles suggests that fuel switching is the most likely hypothesis. Indeed, emissions of PN and PM₁ decrease with the sulfur content of the fuel (Celik et al., 2020; Diesch et al., 2013; Grigoriadis et al., 2021a) but also as fuel quality improves (from residual fuel oil to distillate fuel oil) (Gysel et al., 2017; McCaffery et al., 2021).

Finally, Fig. 12 also indicates that manoeuvring/navigating ferries and cargo ships equipped with engines of more than 10 000 kW emit more particles, in mass and number, than other ships with engines of less than 4000 kW (p value < 0.05). For cruise ships, which are equipped with engines similar to those of ferries and cargo ships, it is challenging to draw definitive conclusions due to the limited number of plumes identified during these operational phases, especially since the emission factors for cruise ships at berth are comparable to those of ferries.

The age of the plumes also seems to affect the number and/or mass of particles, as shown in Fig. 13. Regarding the number of particles (PN), the youngest plumes (less than 5 min old for ships at berth and less than 2 min old for ships manoeuvring/navigating) exhibit statistically higher emission factors (by factors of 1.5 and 2, respectively) compared to older plumes (p value < 0.05). This observation suggests particle accumulation and/or coagulation processes, which reduce the number of particles but increase their average size (Celik et al., 2020; Lack et al., 2009). For particulate mass (PM₁), an upward trend was observed only for ships at berth. However, the results of the Kruskal–Wallis statistical tests indicate that all the groups show a similar central tendency, signifying a stability of the aerosol mass for the age range of the plumes studied in this study (<30 min). Thus, the increase in total aerosol mass due to photochemical ageing observed by several authors using reactors simulating atmospheric oxidation over periods of 2 to 6 d (Lanzafame et al., 2022; Timonen et al., 2022) was not noticeable for plumes less than 30 min old. However, the wide variability in PM₁ emission factors in the different age classes precludes any definitive conclusion.

Figure 14 shows the boxplot analysis of the PM₁ component EFs (for detailed statistical analyses, refer to Table S10). Cl⁻ is not included in analysis, as its median and percentiles are below the DL.

The emission factor values determined in this study are generally comparable to those documented in the literature: for BC, OA, and SO42-, the median EFs are 298 mg kgfuel-1 (163–592), 863 mg kgfuel-1 (543–1742), and 50 mg kgfuel-1 (<28–174), respectively. These values are comparable to the averages reported in the literature for ships using fuel oil with sulfur contents <0.1 % ( 238 ± 305, 624 ± 335 and 120 ± 50 mg kgfuel-1, respectively; Table S8). For NO3- and NH4+, the EF medians are below the DLs (5.4 and 5.0 mg kgfuel-1, respectively) but remain in agreement with the values reported in the literature for ships using fuel oil with sulfur contents <0.1 % (averages of 3 ± 6 mg kgfuel-1 and 2 ± 3 mg kgfuel-1, respectively; Table S8).

Analysis of the parameters likely to influence BC emissions reveals that they mainly depend on the operational phase. Ships manoeuvring/navigating generate more BC (479 mg kgfuel-1 (261–801)) than ships at berth (165 mg kgfuel-1 (105–247)) (Fig. 15a). As for many compounds, this is linked to the low engine load and its reduced stability during navigation and manoeuvres within the port (Sugrue et al., 2022; Zhao et al., 2020). As for SO₂, for these operating phases, a distinction is also observed between arrivals (601 mg kgfuel-1 (349–1039)) and departures (389 mg kgfuel-1 (230–621)), probably due to the change in fuel required when entering the port, switching to a more refined fuel in response to regulations (Grigoriadis et al., 2021a; Huang et al., 2018; McCaffery et al., 2021). Finally, for this compound, the analysis of EFs as a function of the ship construction year demonstrated the influence of the tier regulations imposed by MARPOL on BC production (Tier 0 ships built before 2000, Tier I before 2011, Tier II before 2016, and Tier III after 2016). A statistically significant decrease was observed (Fig. S8) between Tier 0 and Tier I class vessels (the categories that were most represented (90 % of the plumes studied for which the ship construction year was known)). The same trend holds for the Tier II and Tier III classes, even if it is not statistically relevant due to the small number of plumes for these categories. This result corroborates the results of Sugrue et al. (2022) and suggests that these regulations influence BC emissions, even for low engine loads. It is important to note that the distribution of the four tier classes is similar regardless of the operational phases and the age of the plumes, thus eliminating any systematic bias that might alter the previous interpretations.

The statistical tests carried out on the other parameters (category of vessel (Fig. 15a) and plume age (Fig. 16a)) indicate that all groups show a similar central tendency.

Numerous studies have been carried out on the relationship between quantities of BC and CO emitted from various combustion processes. In most cases high correlations were observed (Guo et al., 2017; Taketani et al., 2022; Zhou et al., 2009), as both components arise from incomplete combustion of carbon-based fuels, and the slope of the linear regression (ΔBC/ΔCO) was often used to identify sources such as petrol/diesel vehicles or biomass combustion (Guo et al., 2017). The correlation between the EF_BC and EF_CO of the plumes identified in the present study is negligible (R2<0.1), indicating that the emissions of these compounds evolve in an independent way, probably due to the use of different fuels. The study by Zhao et al. (2020) showed a very strong correlation for a cargo ship and an HFO fuel (R2=0.91) obtained at different engine loads, but the analysis of the correlation for this same ship with another type of fuel (MDO) shows a weaker correlation (R2=0.4). The correlation becomes negligible (R2<0.1) when all the various ships with different fuels (the set of EFs compiled in Table S8) are considered.

OA and SO42- emissions also mainly depend on the operational phase. Ships manoeuvring/navigating emit statistically more OA and SO42- than ships at berth (1603 mg kgfuel-1 (1095–3382) vs. 611 mg kgfuel-1 (470–800) for OA (Fig. 15b) and 171 mg kgfuel-1 (55–466) vs. <28 mg kgfuel-1 (<28–50) for SO42- (Fig. 15c)). These variations can be explained mainly by the use of auxiliary engines operating at a stable and optimum engine speed with a distilled fuel with a low sulfur content (<0.1 %) at berth, whereas ships manoeuvring/navigating within the port area use their main engines at a lower engine load (<25 %), which results in less stable engine performance, with fuels potentially containing a little more sulfur.

For OA, the same as for BC, a difference was observed between EF on arrival and departure of ships (2365 mg kgfuel-1 (1241–4141) vs. 1399 mg kgfuel-1 (470–800)), with arrivals showing emissions 1.7 times higher (Fig. 15b). This statistical distinction probably results from the change in fuel required upon entering the port, involving a switch to a more refined fuel (Gysel et al., 2017; McCaffery et al., 2021). However, this phenomenon is not observed for SO42- (Fig. 15c). This could be related to the reduction in SO42- emissions due to a reduction in the sulfur content of fuels, particularly for fuels with an already low sulfur content (Gysel et al., 2017).

In addition, when they are manoeuvring/navigating, ferries have an EF_OA 2 times higher than the ships in all other categories (3102 mg kgfuel-1 (1479–4933) vs. 1303 mg kgfuel-1 (807–2221)) (Fig. 15b) and an EF_SO4 4 times higher than ships in all other categories (358 mg kgfuel-1 (195–582) vs. 86 mg kgfuel-1 (<28–174)) (Fig. 15c)).

As this observation is not the same for vessels with equivalent engine power, the hypothesis considered is that this vessel category would use different fuel and/or after-treatment devices than ships in all other categories during navigation and/or manoeuvres. The validity of this hypothesis is strengthened by the fact that some ferries in the port of Marseille are equipped with scrubbers, a notable feature for this category of vessel.

The statistical tests carried out on the age of the OA and sulfate plumes (Fig. 16b, c) indicate that all the groups show a similar central trend.

The emission factors for NO3- are often below the DL (more than 60 % of measurements), and this remains true even at night. As a result, statistical analyses of the influence of the various parameters indicate a similar central tendency for all groups, regardless of the parameter analysed. The low-nitrate EFs and the absence of any significant variation in NO3- as a function of the age of the plume (Fig. 16d) do not support the hypothesis previously put forward about NO_x sinks due to photochemical reactions leading to the production of nitrate aerosols. Moreover, according to Celik et al. (2020), the high ambient temperatures observed during the measurement campaign limit the presence of this species in the particulate phase.

The emission factors for NH4+ are often below the DL (more than 60 % of measurements). The analysis of the ratio between NH4+_measured and NH4+_predicted (Fig. S9) – indicating particle acidity (Zhang et al., 2007) – for ship plume and background conditions suggests that while background particles are fully neutralized, those from ship are not or are only partially neutralized. The neutralization level depends on the sulfate emission and consequently on the operating phase. When sulfate emissions are high (EF >250 mg kgfuel-1), the near-zero slope, similar to that observed by Fossum et al. (2024), suggests that the sulfate measured is mainly in the form of sulfuric acid. For sulfate emission under this threshold, partial neutralization occurs. In these cases, ammonia concentration levels (3 ppb (2.3–3.7)) (Table S6) from city road traffic and agricultural activities are insufficient to neutralize the sulfate emitted by ship, or plumes are too young to reach equilibrium.

In summary, Fig. 17 depicts the median chemical mass composition of PM₁ emissions across different ship operational phases, which is the parameter most frequently found to affect EFs. The amount of PM₁ emitted by ships can vary by a factor of 3 depending on the operational phase and tends to be more variable during the manoeuvring/navigation phases compared to when ships are docked. Particles emitted by ships across all operational phases are primarily composed of organic matter (OA, 75 %), black carbon (BC, 21 %), and sulfate (SO42-, 4 %). However, this composition can change with the operational phase: the proportion of black carbon increases to 34 % during manoeuvring, while the proportion of sulfate rises to 8 % during navigation and decreases to 2 % when at berth.

The PM₁ chemical composition characteristics (low sulfate but high organic content) found in the present study share similarities with the organic-rich PM₁ composition recently identified by positive matrix factorization (PMF) analysis of measurements made in Dublin port (Fossum et al., 2024). This suggests that the global ship plumes signature in Marseille port is dominated by ships using VLSFO (very-low-sulfur fuel oil), ULSFO (ultra-low-sulfur fuel oil), or MGO (marine gas oil) rather than HFO (heavy fuel oil) combined with a scrubber system, for which sulfate makes up 60 % of the PM₁ (Fossum et al., 2024). It is noteworthy that in Marseille, the PM₁ has 2-fold higher relative BC content than the organic-rich PM₁ detected in Dublin port (21 % vs. 9 %). One reason for its higher value is that the field campaigns experienced different climatic conditions (summertime for Marseille vs. wintertime for Dublin). The higher temperature at Marseille (average ambient temperature of 24 °C vs. 8 °C in Dublin) could favour the evaporation of organics from the non-volatile black carbon core of the aerosol once the ship emissions are released into the air.

To ensure the comparability of the particle size distributions between the different plumes, the emission factors for each class of particles were normalized with respect to the maximum emission factor observed in each plume. Significant variations in EFs between plumes require this standardization.

Among the ship parameters examined, the operational phase has the most significant impact on the particle size distribution of the 158 plumes selected. For ships at berth, which operate their auxiliary engines at a stable and optimal load using low-sulfur distillate fuel (<0.1 %), the particle size distribution was found to be unimodal and centred around 30 nm (Fig. 18a). In contrast, vessels manoeuvring/navigating, which use their main engines at lower (<25 %) and less stable loads, display a bimodal distribution with modes at 35 and 100 nm (Fig. 18b). Unlike the number of particles, no significant differences are observed between arrivals and departures. The 35 nm mode is generally more prevalent, while the intensity of the 100 nm mode varies among plumes. Notably, the 100 nm mode is particularly pronounced among the ferries, a category of vessels in the port of Marseille that are known to be partially equipped with scrubbers.

These findings are supported by analysis of the daily profiles for mean particle diameter (Fig. S6), which shows a shift towards smaller particles (around 50 nm) during periods of ship movement. These observations are also consistent with the literature. While the specific causes of a unimodal or bimodal distribution can vary between studies, the 30 nm mode is consistently observed. According to particle morphology analyses conducted using transmission electron microscopy (Aakko-Saksa et al., 2023; Alanen et al., 2020), the 30 nm particles are spherical, non-volatile, and originate from the combustion of fuel and lubricating oils. The mode around 100 nm could be attributed to (1) incomplete combustion during manoeuvring phases, which promotes soot formation and particle coagulation (Diesch et al., 2013); (2) the use of scrubbers when the 100 nm mode is dominant (Jeong et al., 2023; Kuittinen et al., 2021; Winnes et al., 2020); or (3) the use of heavy fuel oil types (HFO, VLSHFO, ULSHFO) when the 100 nm mode is not dominant (Anderson et al., 2015a; Fossum et al., 2024). Fossum et al. (2024) demonstrated that (1) plumes from ships at berth, which used marine gas oil (MGO) fuel, had a unimodal particle size distribution centred around 30 nm, whereas (2) plumes from ships manoeuvring/navigating within port, which used ultra-low-sulfur heavy fuel oil (ULSHFO), exhibited a bimodal distribution consistent with the two modes observed in the present study. Since the type of fuel and the use of scrubbers were not specified in the AIS database, these hypotheses could not be confirmed.

The analysis of particle modal diameter evolution with plume ageing, as shown in Fig. 19, indicates a marginal increase in diameter. This finding suggests that, aside from natural dilution that gradually reduces particle concentrations within the plumes, no other significant physico-chemical processes occur over the short timescales studied (less than 30 min). However, while the increase in modal diameter is not statistically significant, this observation, along with a statistically significant decrease in EF_PN, suggests that condensation or coagulation phenomena could occur and could contribute to the increase in particle modal diameter.

Furthermore, scatterplots of the EF_PN measured using the CPC analyser (2.5 nm (PEB) or 7 nm (MAJOR) to 2.5 µm) vs. the SMPS analyser (15 to 660 nm) yield a slope greater than 1.3 for 30 % of the plumes analysed. This finding suggests that, for these plumes, about 30 % of the particles had a diameter smaller than 15 nm. Interestingly, the percentage of plumes associated with this feature increases as plume age decreases, consistent with the expected plume evolution in the atmosphere (e.g. 100 % of plumes from vessels at berth with an age of less than 2 min exhibit this bias compared to 40 % of plumes aged between 2 and 5 min).

The analysis of particle size distribution emitted by ships emphasizes the importance of specifically monitoring the PM₁ fraction, particularly particles smaller than 150 nm. This range includes the two likely particle size modes, which could be characteristic of the fuels used by ships.