Wind direction, wind speed, air temperature, air pressure, relative humidity (RH), rainfall, and solar radiation measurements were made using a Casella weather station (model Nomad, UK) mounted to the top of the main container. The measured wind speed and direction compared well with the data available from the nearest Met Éireann meteorological station located at Dublin Airport, less than 10 km north of Dublin Port.

The wind speed and wind direction were compiled in the Igor software compatible tool for geographical origins of atmospheric pollution, ZeFIR (Petit et al., 2017), to plot air pollution roses aided by the openair package (v2.8-3; Carslaw and Ropkins, 2012).

A PM1 quadrupole aerosol chemical speciation monitor (Q-ACSM) from Aerodyne Inc. (Billerica, MA, USA) measured the mass concentrations of non-refractory species including organic aerosol (OA), sulfate (SO42-), nitrate (NO3-), ammonium (NH4+) and chloride (Cl-) (Ng et al., 2011). While the intensive campaign ran from 16 December 2022 to 7 February 2023, Q-ACSM data are only available through to 27 January 2023. The Q-ACSM used in the study had a standard vaporiser and was calibrated and maintained following the standard protocol developed by the Cost Action CA16109, COLOSSAL (2021). Details of the Q-ACSM instrument can be found in previous studies (e.g. Ng et al., 2011; Pieber et al., 2016). In this study, the Q-ACSM was installed with a PM2.5 URG-2000-30ED cyclone connected to 3/8 in. stainless steel tubing and operated using a carrier flow rate of 2.5 (±0.2) L min-1 with a distance from the inlet to the Q-ACSM of approximately 2 m, to keep particle losses to a minimum. A monotube Nafion^® membrane dryer was installed to maintain RH of the sample air in the range 20 %–40 %. The instrument was operated at a time resolution of just over 5 min (five sets of one sample and one filter measurement scans). The response factor (RF) of NO3- and relative ionisation efficiencies (RIEs) of NH4+ and SO42- were determined following standard operating procedures (COLOSSAL) for ammonium nitrate and ammonium sulfate calibration. The OA RIE was experimentally determined through comparison with another PM1 Q-ACSM combined with use of a state-of-the-art organic RIE calibration with organic alcohols recommended by the Q-ACSM manufacturer. An RF of 2.81×10-11, an NH4+ RIE of 4.15, an SO42- RIE of 0.61, and an organic RIE of 1.9 (default is 1.4) were applied after validation during data ratification in the standard Q-ACSM data analysis process. Composition-dependent collection efficiency (CDCE) was applied following the (Q-ACSM-modified) methods of Middlebrook et al. (2012). The uncertainty in the mass concentration of the non-refractory species is considered ±30 %. The 30 min average limits of detection for the Q-ACSM were calculated to be 0.110 µg m-3 for NO3-, 0.175 µg m-3 for SO42-, 0.662 µg m-3 for NH4+, 0.561 µg m-3 for OA, and 0.105 µg m-3 for Cl-, following the methods of Ng et al. (2011).

The dual-spot aethalometer (Model AE33, Magee Scientific) operates seven different wavelength channels (370, 470, 520, 590, 660, 880, and 950 nm) to provide optical absorption coefficients by measuring light attenuation every minute through a filter tape that has collected aerosol at a flow rate of 5 (±0.4) L min-1. The AE33 Dual Spot™ measurement technique allows for the correction of filter-loading effects by aerosol in real time (Drinovec et al., 2015). The 880 nm wavelength channel is classically used to measure light absorbing equivalent black carbon (eBC) (Petzold et al., 2013; Bond et al., 2013), using the standard mass-specific absorption cross section (MAC) of 7.77 m2 g-1 (Magee Scientific Inc., 2018; Drinovec et al., 2015). Multiple scattering effects of the collection tape are accounted for with the correction value (C) of 1.57 that is based on experimental investigations into TFE-coated glass fibre filter tape material (part no. 8050) (Drinovec et al., 2015). The rolling 15 min average was calculated from the 1 min data to reduce noise. This rolling average was used to interpolate eBC concentrations that matched Q-ACSM data points in time.

The Xact 625 (Xact from this point onward) can measure up to 24 elements between silicon and uranium at hourly time resolution and has been evaluated and described in previous studies (Furger et al., 2017; Tremper et al., 2018). The instrument has a flow rate of 1 m3 h-1; the inlet tube is heated to 45 °C when the ambient relative humidity (RH) exceeds 45 %, which was usually the case. The samples are collected onto Teflon tape and subsequently analysed using energy dispersive X-ray fluorescence (EDXRF). The X-ray source used is a rhodium anode (50 kV, 50 W), and the X-ray fluorescence is measured using a silicon drift detector. In this study, the instrument measured the elements As, Ba, Ca, Cd, Ce, Cl, Cr, Cu, Fe, K, Mn, Mo, Ni, Pb, Pt, S, Sb, Se, Si, Sr, Ti, V, and Zn in PM2.5. Daily automated quality assurance checks were performed at midnight. Further quality assurance checks, such as flow checks and external calibration checks, were performed at the start and end of the campaign.

Nitrogen oxides (NOx) and sulfur dioxide (SO2) were measured throughout the campaign using automated gas analysers. The NOx is measured by the Teledyne Instruments Model 200A chemiluminescent NO/NO2/NOx analyser, which measured NO and NOx and by calculation NO2 at 5 min time resolution. The total NO (NOx) can be measured in parts per billion (ppb), and NOx is converted to µg m-3 as NO2 ppb ⋅ 1.9125 = NO2 µg m-3, and NO ppb ⋅ 1.28 = NO µg m-3 (20 °C, 1 atm). The SO2 was measured at 1 min time resolution by a Teledyne API Model T100 UV fluorescence SO2 analyser that was used throughout the PortAIR project. A small drift in the SO2 baseline was observed over the yearlong campaign, so the measurements were subsequently corrected using a polynomial function for baseline drift derived from laboratory tests conducted at the end of the campaign.

The scanning mobility particle sizer (SMPS) characterises the number-size distribution of the ambient aerosol particles. Particles passing through the system are charge-neutralised (Fuchs, 1963) (electrical ioniser model 1090, MSP) and then sized by electrical mobility through a differential mobility analyser (DMA, TSI Inc. model 3080) and finally counted by a condensation particle counter (CPC; TSI Inc. model 3010). The SMPS was operated by passing sample air through a multi-tube Nafion^® membrane and into the DMA at a sample flow of ∼ 1 L min-1 with a sheath flow of 5 L min-1 (Collins et al., 2004). The SMPS was operated at 3–5 min scan duration with TSI Inc. AIM software (release version 9.0.0.0) with charge correction applied.

Unconstrained PMF solutions with 2–10 factors were tested as a first step (see discussion in Sect. S1.1 of the Supplement). Only the two-factor solution resulted in reasonable profiles, comprised of generic hydrocarbon-like organic aerosol (HOA) and an aged organic aerosol. The rest of the solutions resulted in some particular ions separating out as individual factors that, however, are not physically meaningful. Despite the measurements being conducted in a port environment, no sea-salt factor (see Supplement) was resolved by the unconstrained PMF solution for the intensive campaign period. This was supported by evaluation of elemental Cl (Xact) that showed comparable contributions from the directions of the city and of the sea.

To direct the PMF model towards a physically meaningful solution, the mass spectra of reference primary OA factors were constrained using the ME-2 algorithm (Canonaco et al., 2013). The organic-sulfate combined PMF was run by combining the two constrained ship type factors with other constrained factors expected to be present. These include a traffic HOA derived from a previous curb-side study in Dublin, Ireland (see Fig. S10) (Lin et al., 2020); individual solid fuel burning (SFB) factors for peat, wood, and coal from a previous Irish study (Lin et al., 2021); a sea-salt factor; and the S-Ship and O-Ship factors discussed above. While finding the most reasonable solution is somewhat subjective, the best solution occurs when increasing the number of factors leads to avoidable splitting of the factors or when reducing the number of factors leads to avoidable mixing of factors. Whether factors are split or mixed in a solution was evaluated by looking at the solution residuals, correlations to other factors, and external time series (e.g. NO3-, NH4+, eBC, and metals) and checking if diurnal patterns looked representative of real port or city activities (e.g. traffic patterns). The best solution was determined to be with six factors, two unconstrained and four constrained factors: S-Ship emissions, O-Ship emissions, peat, and traffic HOA. Increasing the number of factors for the ME-2 solution beyond six could not resolve any more reasonable solutions, with extra factors being separated into unrealistic profiles with poor correlation to external tracers. The six-factor solution was then run with bootstrap resampling (50 runs) and found to be very stable, where the standard deviation of the profiles or time series was 2 %–23 %. The factor profiles derived from the six-factor bootstrap solution for the organic-sulfate source apportionment are shown as factor profiles in Fig. 3a. The time series and diurnal trends for the four constrained factors (S-Ship, O-Ship, HOA, peat) and the two unconstrained factors (oxygenated OA (OOA) and X-Ship) are presented in Fig. 3b and c, respectively.

The diurnal variations (Fig. 3c) between the S-Ship and O-Ship factors were similar, while the time series shows differences in the patterns observed for the factors, as well as some periods where the factors overlap but peak at slightly different times. This may indicate intrinsically linked emissions from different emission sources. The HOA factor had a diurnal pattern with small peaks occurring at the same times as the peaks in the ship factors, which is not surprising given the flow of vehicular traffic linked with ship arrivals and departures. However, the HOA factor also had an evening peak, which could be caused by the HOA traffic mass spectra being very similar to those for home heating oil at UMR and m/z < 120 (Lin et al., 2020). The correlation matrices (Fig. S11) showed HOA correlating with SFB-related factors (peat, OOA) and elemental tracers (As and K), as well as with shipping-related tracers (i.e. SO2 and the X-Ship factor). The HOA factor seemed split between traffic from the port and a city source that peaks in the evening, likely oil burning for residential heating. Peat showed time trends that match the regional pollution episodes in Fig. S7 and diurnal patterns that are dissimilar to the ship emissions, with a clear evening peak. The increase in the evening is expected for factors associated with residential SFB for home heating.

One of the unconstrained factors is attributed to oxygenated OA (OOA). This factor followed a very similar diurnal pattern and time series to peat, which points to OOA being present or produced during SFB pollution episodes. The factor correlation matrix (Fig. S11) also supports this assertion, with strong correlations for OOA with peat (r=0.74), elemental As (r=0.64), elemental K (r=0.62), and Cl- (r=0.62), potentially related to the build-up of local pollution, which can be dominated in winter by SFB-related pollution (e.g. Lin et al., 2023). On the other hand, the correlation of OOA with NO3-, NH4+, and Cl- could be related to regional pollution during stagnant weather, so all or some portion of OOA may be independent of SFB. The strong correlation of OOA with NO3- (r=0.77) could indicate the OOA was semi-volatile, freshly formed secondary OA, which would point to the formation of OOA along the route to the port. As a result, OOA is probably a combination of contributions from regional secondary production, including from SFB sources, and freshly formed secondary aerosol.

The second unconstrained factor contained heavier ions, without fragmentation at lower m/z, and lacked m/z 44, indicating no ageing. This factor time series is well correlated with O-Ship and to a lesser extent with HOA and S-Ship. It is unlikely to be a split factor due to association with both shipping factors, which points to a source related to ship engines, mostly to vessels using low-sulfur fuels. For these reasons, we call this factor X-Ship. The ions at m/z 81 and 95 are typical of exo-sulfur aromatics, while there are also CnH2n-1+ ions for m/z 41 and 55 and for m/z 105 and 119 that could be carboxylic acids, possibly naphthenic acid. Some of these ions point towards this factor being an indicator of engine oil lubricant (Anders et al., 2023; Czech et al., 2017). The X-Ship OA ions are poorly correlated with the S-Ship (r2=0.007) and O-Ship (r2=0.089) factors (see Fig. S8b and c). In both cases, the majority of X-Ship ions, especially at higher m/z, have stronger relative intensities. The major difference is that X-Ship does not contain m/z 18, 41, 43, 44, 55, and 57, indicating a lack of hydrocarbon content and ageing. The X-Ship factor also appeared in unconstrained PMF runs of the matrices; therefore, it was mathematically divergent and found in most solutions with a few unconstrained factors.

The organic-sulfate PMF resolved 58 S-Ship plumes and 190 O-Ship plumes over the intensive campaign. The average chemical breakdown of PM1, along with NOx and SO2, is shown in Fig. 4a for S-Ship and Fig. 4b for O-Ship. The S-Ship factor plumes were comprised mostly of SO42- (57 %), followed by OA (35 %), eBC (6 %), and NO3- (1 %), with negligible contributions from NH4+ and Cl-. There is slightly more SO42- and less OA than the average of the exemplary S-ship plumes, caused by the inclusion of plumes with lower PM1 concentrations. O-Ship plumes were comprised mostly of OA (77 %), followed by eBC (9 %), SO42- (7 %), NO3- (3 %), NH4+ (3 %), and Cl- (1 %). The increased contribution from inorganic species compared to the average of the exemplary O-Ship plumes is caused by the presence of plumes on top of regional secondary aerosol. Overall, 7 % of the S-Ship plumes and 27 % of O-Ship plumes had PM1 concentrations less than 15 µg m-3. The 99th percentile of PM1 statistically represents extreme pollution episodes, which was PM1 < 53.5 µg m-3 during the PortAIR intensive campaign. The O-Ship factor included 33 plumes where PM1 reached at least 53.5 µg m-3 (99th percentile of PM1), whereas the S-Ship factor only had 10 plumes where PM1 reached 53.5 µg m-3.

Using the PMF solution to identify S-Ship and O-Ship plumes, particle mass concentration and other components of these two ship factor types can also be compared. In the PMF solution, despite the higher frequency of occurrences of O-Ship pollution plumes, S-Ship and O-Ship plumes had nominal average PM1 concentrations of 29 ± 22 and 32 ± 26 µg m-3, respectively. This was due to S-Ship pollution events having significant fractions of SO42-, such that mean OA was 10 ± 10 µg m-3 and mean SO42- was 17 ± 12 µg m-3 for S-Ship. O-Ship however had 25 ± 24 µg m-3 of OA and 2 ± 2 µg m-3 of SO42- on average. S-Ship PM1 ranged from 9–135 µg m-3, and O-Ship ranged from 4–252 µg m-3.

Average particle number concentrations were (1.52 ± 1.55) × 104 and (3.75 ± 2.28) × 104 cm-3 for S-Ship and O-Ship, respectively. Although the number concentration varied widely across the ship plumes, on average O-Ship had more particles associated with this plume type. Additionally, while there were variations in size modal distributions over the duration of the plume and larger variation from plume to plume, the number-size distributions of O-Ship emissions were shifted to smaller sizes than S-Ship (Fig. 5). The combined smaller diameter particles at higher number concentration from O-Ship could have more adverse health impacts as these particles (dm<100 nm) penetrate into the bloodstream and translocate to all organs in the body (Schraufnagel, 2020). While it seems that some variation in number-size distribution and absolute number concentration may be due to fuel type, it is important to note that additional factors such as variations in engine loadings and the use of lubricating oil may also influence the emissions patterns. Furthermore, the measurements in this study are expected to represent diluted and mixed plumes after undergoing transport, since these are ambient plume detections rather than measurements directly from the stack.

In terms of NOx and SO2, both median and mean values were similar for the two types, with S-Ship plumes having an average concentration of (117.6 ± 118.0) µg m-3 of NOx and (3.5 ± 3.4) µg m-3 of SO2 and O-Ship having an average of (132.3 ± 108.7) µg m-3 of NOx and (3.9 ± 3.2) µg m-3 of SO2. The eBC concentrations were also similar, with S-Ship having (1.8 ± 1.4) µg m-3 and O-Ship having (3.0 ± 1.8) µg m-3 of eBC on average. The V/Ni median ratio for S-Ship events was 3.41 (range 2.7–3.9) in line with the literature but was 0.74 (range 0–2.3) for O-Ship, which is in line with a study that found V/Ni = 0.6–1.1 after the Global Sulfur Cap 2020 regulation (Tauchi et al., 2022). A summary of the mean characteristics of the two types of ship plumes can be found in Table S2.

Factors from the PMF, inorganic species, and eBC were used to quantify the relative contribution of sources to measured PM1 (Fig. 6). Since the PMF only included the directly measured SO42- ions, the SO42- fragments that are normally calculated from the measured ones were added back in after to get the true SO42- mass concentration. The S-Ship factor and OOA factor contained all significant fractions of the SO42- ions (Fig. 3a), so these factors were affected by the re-addition of the non-measured SO42- fragments (see Sect. 2.3 and discussion in the Supplement). The PM1 breakdown using these factors from the PMF is shown in Fig. 6a.

However, the factors from the PMF are not true representations of the mass contribution of sources since they only take into account OA–SO42- input. Therefore, we use the time series of factors from the PMF and other ambient observations to quantify the relative contribution of sources, separating them into primary and regional sources where possible. As outlined in detail below, the procedure for quantification of the sources included separating regional SO42- from S-Ship SO42- and included attributing a fraction of eBC to the ship emissions. The result of this procedure is the source contribution estimates shown in Fig. 6b.

The PMF solution has trouble separating the regional SO42- from S-Ship SO42-. This is partly because the measured SO42- ions have low variability in the ACSM mass spectra, with ammonium sulfate and acidic sulfate having similar relative intensities (Chen et al., 2019) and fragmentation patterns (see discussion in the Supplement and Fig. S12). The overall effect on the data is that the relative fraction of S-Ship was overestimated by about 8 % of the total owing to regional SO42-. This includes various nighttime periods of elevated regional sulfate that were not attributed to solid fuel burning because coal is the only solid fuel used in Dublin that produces appreciable amounts of sulfate (Trubetskaya et al., 2021), and it was found to have a negligible contribution to PM1 during PMF runs.

While the separation of regional and ship sulfate is a limitation of the organic-sulfate PMF, we can still isolate the real S-Ship emissions by manually subtracting the contribution of regional SO42- from the S-Ship factor. This adjusted S-Ship is called S-Ship* and reflects S-Ship emissions with regional SO42- subtracted out. The pollution rose of S-Ship* indicates an emission source from a nearby ferry terminal (Figs. S13 and S14) and no longer includes a nighttime increase in the diurnal profile (Fig. S14).

Since eBC is observed to be part of the ship emissions, it was apportioned to the S-Ship and O-Ship factors using the estimated eBC / (OA + SO42-) ratios (eBCr). The ratios from the observed five exemplary plumes for S-Ship (eBCr = 0.066) and O-Ship (eBCr = 0.047) were compared to those derived from the source apportioned S-Ship and O-Ship plumes. The S-Ship was comparable with an average eBCr of 0.068, close to 0.066 from the S-Ship five exemplary plumes. However, the O-Ship eBCr differed from 0.047 for the five exemplary plumes to an average of 0.113 for the source apportioned plumes. This difference was caused by the much larger OA contribution during the exemplary plumes than for the average taken over all O-Ship factor plumes. Therefore, one ratio was applied for all O-Ship plumes where PM1 > 53.5 µg m-3, and for the rest of the period another ratio was applied, eBCr = 0.063 and eBCr = 0.141, respectively. These eBCr ratios were used to re-apportion eBC, amounting to 11 % of total eBC to S-Ship* and 13 % to O-Ship*, which denotes O-Ship and the eBC fraction. The eBC shown in the pie chart (Fig. 6b) is adjusted (eBC*) and reflects a subtraction of the 24 % of eBC (only 2.9 % of PM1) already accounted for in the ship fuels. Nevertheless, the pollution rose of eBC* (Fig. S13) suggests a significant influence from the east–south-east that could reflect shipping-related traffic through Dublin Port and the wider channel.

Factors and chemical components that belong to regional sources were combined and called the regional source, which includes the OOA, NO3-, NH4+, Cl-, and regional sulfate subtracted from S-Ship. It should be noted that an estimated 1 % of PM1 from the combined NO3-, NH4+, and Cl- could be associated with O-Ship fuel emissions rather than regional; however, there was enough uncertainty in this estimation that the 1 % has been left associated with the regional source. The pollution rose of the regional source shows two major source origins, one from the west and one from the north-east, and very little from the ocean or nearby ferry terminals, which also confirms that most of the inorganics (other than sulfate) did not originate from shipping emissions. The S-Ship* and O-Ship* contain the apportioned eBC, as explained above. Across the campaign, the source contributions to the measured PM1 were regional (46 %), O-Ship* emissions (14 %), S-Ship* emissions (12 %), HOA due to traffic or oil burning (10 %), eBC* (9 %), peat (6 %), and X-Ship emissions (2 %). The Dublin Port ship-related factors made up 28 % (S-Ship*, O-Ship*, and X-Ship) of PM1, not counting ship-traffic-related HOA and associated port traffic and shipping lane eBC. It was difficult to attribute the HOA factor in Dublin Port to either ship-related traffic, city traffic, or oil burning for residential heating, as the pollution roses only indicate a local and often westerly source (Fig. S13). Therefore, we estimate that shipping-related emissions in Dublin Port contributed 28 %–47 % of PM1 (Fig. 6b), where the upper range represents the entire contribution of hydrocarbon-like organic aerosol (HOA) and traffic-related eBC added to the estimate (from activities such as ferry traffic, vehicles for moving containers, and crane engines).