Comparison of the Impact of Ship Emissions in Northern Europe and Eastern China

Abstract

It is well known that ship emissions contribute significantly to atmospheric pollution. However, the impact on air quality can regionally vary, as influenced by parameters such as the composition of the regional shipping fleet, state of background atmospheric pollution, and meteorological aspects. This study compared two regions with high shipping densities in 2015. These include the North and Baltic Seas in Europe and the Yellow and East China Seas in China. Here, a key focal point is an evaluation of differences and similarities of the impacts of ship emissions under different environmental conditions, particularly between regions with medium (Europe) and high air pollution (China). To assess this, two similarly performed chemical transport model runs were carried out with highly resolved bottom-up ship emission inventories for northern Europe and China, calculated with the recently developed MoSES model, publicly available emissions data for nonshipping sources (EDGAR, MEIC). The performance of the model was evaluated against measurement data recorded at coastal stations. Annual averages at affected coastal regions for ${\mathrm{NO}}_2$, ${\mathrm{SO}}_2$, ${\mathrm{O}}_3$ and ${\mathrm{PM}}_{2.5}$ were modeled in Europe to be 3, below $0.3$, $2.5$, 1 and in China 3, 2, 2–8, $1.5$, respectively, all given in $\mathsf{\mu }\mathrm{g}/{\mathrm{m}}^3$. In highly affected regions, such as large harbors, the contributions of ship-related emissions modeled in Europe were 15%, $0.3$%, $-12.5$%, $1.25$% and in China were 15%, 6%, $-7.5$%, 2%, respectively. Absolute pollutant concentrations from ships were modeled slightly higher in China than in Europe, albeit the relative impact was smaller in China due to higher emissions from other sectors. The different climate zones of China and the higher level of atmospheric pollution were found to seasonally alter the chemical transformation processes of ship emissions. Especially in northern China, high PM concentrations during winter were found to regionally inhibit the transformation of ship exhausts to secondary PM, and reduce the impact of ship-related aerosols, compared to Europe.

1. Introduction

For economic reasons, the maritime sector has long been the first choice for freight transport across the globe. Although ships are a relatively environmentally friendly mode of transport, when considering cargo volume per fuel consumption, the steadily growing global merchandise turnover and the associated shipping activity have led to a deterioration of air quality in many regions worldwide [1,2]. The combustion of fuel oil in ships is responsible for large amounts of air pollutants such as sulfur oxides (${\mathrm{SO}}_{\mathrm{X}}$), nitrogen oxides (${\mathrm{NO}}_{\mathrm{X}}$) and fine particulate matter (${\mathrm{PM}}_{2.5}$) [3]. Since ship emissions can be transported several hundred kilometers in the atmosphere, they can impact coastal regions and cities as well as regions further inland. In particular, ${\mathrm{PM}}_{2.5}$ is well known to be responsible for respiratory and cardiopulmonary diseases that can lead to premature death [4,5,6,7]. ${\mathrm{NO}}_{\mathrm{X}}$ and ${\mathrm{SO}}_2$ can transform to secondary PM and acidify the environment. In addition, ${\mathrm{NO}}_{\mathrm{X}}$ is responsible for eutrophication and influences the formation of tropospheric ozone [8,9,10]. For these reasons, the regulation of ship emissions has been placed on the agenda by the International Maritime Organization (IMO) and many national authorities. The formulated legal framework is the International Convention for the Prevention of Pollution from Ships (MARPOL) Annex VI, in which global and regional regulations are adopted with the aim of preventing the negative effects from ship emissions [11]. In practice, the globally allowed fuel-sulfur content of marine fuels has been limited to $3.5$% m/m (mass/mass) by MARPOL Annex VI. This cap was tightened to $0.5$% at the beginning of 2020 [12,13].

Two regions that have been particularly affected by air degradation from ship emissions in recent decades are the North and Baltic Seas in northern Europe and the Bohai, Yellow and Eastern China Seas in East Asia. Both are regions with high shipping densities and large port cities that are tightly interconnected by global maritime freight transport.

Land-based sources of air pollution have been systematically reduced in Europe since the end of the last century, but air quality degradation from ship emissions has received necessary attention only in the last two decades [14,15,16,17,18,19]. Emissions from a growing shipping sector were met by the introduction of a sulfur emission control area (SECA) in the North and Baltic Seas, described in MARPOL Annex VI, in which the allowed fuel-sulfur content was limited to 1% m/m from 1. July 2010 and to $0.1$% from the beginning of 2015 [12]. Many studies have been conducted that investigate the impact of ship emissions in the North and Baltic Seas on regional air quality and the regulating effect of control measures [5,15,19,20,21,22,23,24]. Furthermore, interdisciplinary projects have been initiated by the EU to investigate impacts in a comprehensive approach, e.g., Clean North Sea Shipping (CNSS) and Sustainable Shipping and the Environment of the Baltic Sea Region (SHEBA) [25,26].

In East Asia, the recent and rapid economic growth of the main regional actor China made it the world’s largest emitter of air pollutants (18–35%) [27]. The Chinese government, well aware of the problem, set an annual average exposure limit for ${\mathrm{PM}}_{2.5}$ of 35 $\mathsf{\mu }\mathrm{g}{\mathrm{m}}^{-3}$ in 2012 [28]. From 2013 to 2017, the Chinese air degradation containment strategy was further tightened with the adopted “Plan on the Prevention and Control of Air Pollution” [29]. A monitoring system for air pollution was established, and industry, energy production and the transportation sector were adjusted to reduce ambient ${\mathrm{PM}}_{2.5}$ concentrations by 28–40% in 2017 [30]. However, regulations for the shipping sector, one of the most prominent contributors to ${\mathrm{PM}}_{2.5}$, remained incomplete [6]. China is home to seven of the ten largest container ports worldwide, and in 2019, almost 30% of the world’s container port throughput (TEU) occurred there [31,32]. This freight volume is expected to increase in the near future due to China’s 21st Century Maritime Silk Road Strategy and the recently ratified Free Trade Agreement Regional Comprehensive Economic Partnership (RCEP). A domestic emission control area (DECA), issued by the Chinese authorities, was enforced from the beginning of 2017 for berthing ships in 11 regional ports and from 2019 within 12 $\mathrm{NM}$ of the Chinese shoreline [33]. Within this DECA, the fuel-sulfur content is limited to a maximum of $0.5$%. From the beginning of 2020, regulations were tightened to a maximum allowed fuel-sulfur of $0.1$% for berthing ships. The effect of the DECA on air quality has been investigated in several studies. Liu et al. [34] determined that ${\mathrm{PM}}_{2.5}$ and ${\mathrm{SO}}_2$ concentrations in the Pearl River Delta (PRD) were reduced by $2.7$% and $9.5$%, respectively. Possible benefits from expanding the DECA have been modeled by Feng et al. [35]. Here, ships up to 96 $\mathrm{NM}$ from shore were found to contribute substantially to ship-originated ${\mathrm{PM}}_{2.5}$. This finding is in accordance with the results from Zhao et al. [36], which showed that the DECA decreases the contribution of ship emissions to ${\mathrm{PM}}_{2.5}$ by 71%, albeit a stricter fuel-sulfur limit and an expansion of the DECA to 100 $\mathrm{NM}$ from the coast may even achieve a reduction of 86%.

Although many studies exist concerning the impact of ships on air quality in China or Europe, no comparison between these two regions has been made to date. While the overarching problem is similar and many of the larger cargo ships can be encountered in either region, regional differences result in a different impact. For example, domestic shipping activities such as fishing, freight/passenger transport and leisure shipping differ between Europe and China. The legislation targeting the control of ship emissions remains different in their details. In addition, higher pollutant emissions for nonshipping sources are found in China than in Europe. These can modulate the chemical and physical processes that emissions from shipping undergo in the atmosphere. Furthermore, different climate zones and regional weather phenomena, such as the East Asian monsoon and the rainy season in summer, affect the transport and deposition of ship exhausts. This study aims to assess the impact of ship emissions under consideration of these factors for otherwise analogously conducted air quality model runs in China and northern Europe. By comparing the results, conclusions are drawn concerning the behavior and physical/chemical transformation of pollutant species under different atmospheric conditions. These results should contribute to an understanding of the underlying processes and help in the formulation of future legislative control measures.

To assess this problem, the three-dimensional Community Multiscale Air Quality model (CMAQ) was set up (Section 2.1) and fed with emissions data from 2015 for both regions (Section 2.2). A highly temporally and spatially resolved ship emission inventory for China was created for this study, by a first comprehensive application of the Modular Ship Emission Modeling System (MoSES). A description of this model and an analogously created ship emission inventory for the European domain can be found in [37]. The description of the inventory for China and applied emission factors, as well as a comparison of the European and Chinese shipping fleets, can be found in Section 2.2.5. The performance of the model was evaluated by a comparison with air quality data measured at coastal stations (Section 3). Finally, the modeled concentration patterns of ${\mathrm{NO}}_2$, ${\mathrm{SO}}_2$, ozone and primary and secondary fine particulate matter in both domains are discussed (Section 4).

2. Materials and Methods

2.1. Model Simulations

2.1.1. Regions of Interest

Model domains were prepared for the two regions of interest for which this study compared the air quality impact of ship emissions. The region of northern Europe was covered by a domain hereafter referred to as SC12NSBS. It includes the North and Baltic Seas and is shown in Figure 1 (left). China was represented by a domain that is located in East Asia and is hereafter referred to as CNC12 (Figure 1 right). It includes the coastline of China, including the Bohai Sea, the Yellow Sea, the East China Sea and parts of the South China Sea. Both domains have the same extent in kilometers and include not only the main local shipping routes but also densely populated coastal regions, whose air quality is affected by ship exhausts. Two $12\times 12$ ${\mathrm{km}}^2$ Lambert conformal grids were set up for the chemistry transport modeling (see Section 2.1.2 for details). Both grids consisted of $196\times 196$ cells.

Two chemical transport simulations were performed for each domain: One represented a “base” case that included emissions from all sectors, and the other represented a “no ships” case, omitting ship emissions (see Section 2.2). The pollutant concentrations originating from shipping were then determined by the zero-out method, in which the “no ships” case is subtracted from the “base” case.

2.1.2. Chemical Transport Model CMAQ-Setup and Forcing

The Community Multiscale Air Quality model (CMAQ) by Byun and Ching [38], Byun and Schere [39] in version 5.2, a chemistry transport model, was used and set up with the carbon bond 5 photochemical mechanism (CB05-TUCL) [40,41,42] and the AE6 aerosol mechanism. The CMAQ run was performed for 2015 after a previous spin-up time of 2 weeks to adapt to the initially homogeneous concentration fields and atmospheric conditions. The model considered 30 vertical layers from the ground surface up to a 100 $\mathrm{hPa}$ pressure level, which was located at approximately 8 $\mathrm{km}$. Twenty of these layers were below an altitude of 2000 $\mathrm{m}$. The lowest layer had a height of approximately 42 $\mathrm{m}$ and was located at approximately 21 $\mathrm{m}$.

The simulations for China and Europe were nested into preceding, overarching CMAQ simulations that covered larger areas at a coarser resolution (36 $\mathrm{km}$ for Europe and 24 $\mathrm{km}$ for China) but used the same emission sources as the 12 $\mathrm{km}$ runs. The chemical conditions at the boundaries of these larger domains were taken from the Integrated Forecast System-Copernicus Atmosphere Monitoring Service (IFS-CAMS) analysis [43]. The data were available from the MARS archive at the European Centre for Medium-Range Weather Forecasts (ECMWF) and the CAMS Atmosphere Data Store [44]. Particle and gas concentration fields of the IFS-CAMS data are usually provided on a T511 spectral grid with 137 vertical levels, but due to retrieval issues, data on a $0.5°\times 0.5°$ grid were used. The IFS-CAMS data were temporally and spatially remapped onto the boundary of the overarching domains. Chemical and particle species were converted to match the species used in CMAQ.

2.1.3. Meteorological Forcing

One important difference between the regions of interest was the characteristic of the atmospheric large-scale circulation. Therefore, reliable meteorological data are needed to calculate the transport and transformation of gas species and aerosol particles as realistically as possible. The meteorological forcing was realized by atmospheric simulations with the community model COSMO-CLM (version 5.0-clm15) [45], which is embedded into the COSMO model used for numerical weather prediction [46,47,48]. To simulate radiative transfer, an extension of the COSMO model for MACv2 transient aerosol climatology was used [49].

The atmospheric simulations were performed using forcing data from global models as initial and boundary conditions, i.e. the MERRA2 reanalysis [50] for the SC12NSBS domain and the JRA-55 reanalysis [51,52] for the CNC12 domain. The simulations for Europe were performed on a grid with a resolution of $0.06°\times 0.06°$. To ensure that the atmospheric fields in the transient model integration were in accordance with the observations over the whole period, a nudging technique was used as described in Petrik et al. [53]. The reader is referred to this publication to find more information about the setup of the atmospheric model (setup “CCLM-oF-SN”). Regarding the simulations over eastern Asia, a domain with a resolution of $0.11°\times 0.11°$ was used that consisted of $315\times 315$ grid points.

The initialization of the soil proved a challenging task since the forcing data from the global model did not provide sufficient information consistent with the soil parameterization of the regional model. In the southeastern Asia region, the characteristics of soil drying and soil wetting are very different for the various climates. Different spin-up experiments for the soil were performed. The results suggested that at least one complete monsoon season and one complete winter season were needed: only then could the water reservoir be reasonably filled up, and the solution converged to a balanced state. In particular, the soil levels below one meter required a considerable amount of time. Additionally, the region of northern China, where precipitation events occur only rarely.

2.2. Emissions Data

The emissions data fed to CMAQ were compiled from different sources for 2015 and preprocessed to an hourly resolution.

2.2.1. Anthropogenic Land-Based Emissions for Europe

For the SC12NSBS domain in Europe, land-based emission data were based on the CAMS-REGAP-EU emission inventory for 2015 in version 3.1. The data set comprises annual anthropogenic emission totals for 13 GNFR sectors [54] (https://permalink.aeris-data.fr/CAMS-REG-AP, last accessed: 16 September 2021). The emissions were prepared utilizing the internally available Highly Modular Emission Model (HiMEMO). A temporal distribution up to a one-hour resolution was achieved using temporal profiles from the LOTOS-EUROS model, which were used to simulate a daily schedule of emissions from the GNFR sectors.

2.2.2. Anthropogenic Land-Based Emissions for China

For the CNC12 domain, emissions data from the Multiresolution Emission Inventory for China (MEIC) were used for nonshipping, land-based emission sources [55,56,57,58,59,60,61,62]. The monthly, gridded emissions of the MEIC inventory had an original spatial resolution of $0.25°\times 0.25°$. The emission inventory (EI) contained data for the five emission sectors: “Agriculture”, “Industry”, “Power”, “Residential” and “Transportation”. The following chemical species and compound groups were included: black carbon (BC), $\mathrm{CO}$, ${\mathrm{NH}}_3$, ${\mathrm{NO}}_{\mathrm{X}}$, organic compounds (OC), ${\mathrm{PM}}_{2.5}$, ${\mathrm{PM}}_{\mathrm{coarse}}$, ${\mathrm{SO}}_2$ and VOC species that were precategorized into the functional groups considered in the CB05 chemistry mechanism.

For the other countries that were completely or partially included in the CNC12 domain, monthly, gridded emissions from EDGAR v5.0 were used at a $0.1\times 0.1$ spatial resolution. The EDGAR inventory covered 9 pollutants, i.e., BC, $\mathrm{CO}$, ${\mathrm{NH}}_3$, NMVOC, ${\mathrm{NO}}_{\mathrm{X}}$, OC, ${\mathrm{PM}}_{10}$, ${\mathrm{PM}}_{2.5}$ and ${\mathrm{SO}}_2$, of 27 activities that are based on IPCC 1996 and 2006 sector codes (see Section S1 of the Supplementary Material for the sector mapping). All nonshipping anthropogenic emissions were distributed from a monthly to an hourly resolution. The individual temporal patterns for countries and sectors were considered by applying the high-resolution temporal profiles developed by Crippa et al. [63]. For this, the five sectors of the MEIC inventory were mapped to fit EDGAR activities (Section S1). A sector-wise vertical distribution up to 1106 $\mathrm{m}$ was carried out by applying the height profiles described in Bieser et al. [64].

To be compatible with the application in CMAQ, ${\mathrm{NO}}_{\mathrm{X}}$ emissions were split into $\mathrm{NO}$ and ${\mathrm{NO}}_2$ at a ratio of 90 to 10 for road traffic and 95 to 5 for all other sectors. Furthermore, NMVOC emissions from EDGAR were split sector-wise according to a profile of The Netherlands Organisation for Applied Scientific Research (TNO) into the functional groups of the CB05 mechanism. ${\mathrm{PM}}_{2.5}$ splits were applied from the SMOKE for Europe emission model by Bieser et al. [65].

2.2.3. Biogenic Emissions

Emissions from biogenic sources were calculated with the Model of Emissions of Gases and Aerosols from Nature (MEGAN) in version 3 for both model domains [66,67]. MEGAN was driven by the same meteorological data preprocessed for CMAQ (see Section 2.1.3) Vegetation data tables were used unmodified from the official MEGAN resources [68]. SPOT/PROBA V LAI1 from GEOV1 products was chosen as an alternative input for leaf area index (LAI) data in MEGAN [69].

2.2.4. Ship Emissions in Northern Europe

The ship EI for northern Europe was created with the MoSES model using data from the automatic identification system (AIS) that were recorded by the European Maritime Safety Agency (EMSA), which was acquired from the Federal Maritime and Hydrographic Agency of Germany (BSH). The European inventory covered the area between the longitude and latitude ranges $-5.00$–$31.41°$ and $48.32$–$68.37°$, respectively. The spatial resolution was $0.069$ in the x (east–west) and $0.036°$ in the y (north-south) directions, which corresponded to approximately 4 $\mathrm{km}$. The temporal resolution was hourly. A detailed description of this ship EI and details about the bottom-up methods used for the calculation can be found in Schwarzkopf et al. [37]. To cover the whole computational SC12NSBS domain (Figure 1, left), the ship emission inventory was augmented at the northern and western boundary with data from the Ship Traffic Assessment Model (STEAM) described in detail in Jalkanen et al. [70,71], Johansson et al. [72,73]. The STEAM data were part of the CAMS-GLOB-SHIP dataset (v2.1) and can be downloaded at the Emissions of Atmospheric Compounds and Compilation of Ancillary Data website [74]. The emission inventory of MoSES was scaled up to fit the grid of the STEAM data (regular lon.-lat. with a $0.1°$ resolution) and then merged.

Table 1. Gas and aerosol species included in the ship emission inventories (EIs), with sources of the emission factors (EFs) used.

Pollutant	EF Source
Sulfur dioxide (${\mathrm{SO}}_2$)	—
Sulfate (${\mathrm{SO}}_4^{2-}$)	Schwarzkopf et al. [37]
Water associated with sulfate (${\mathrm{SO}}_4\times {\mathrm{H}}_2\mathrm{O}$)	Jalkanen et al. [71]
Nitrogen oxides (${\mathrm{NO}}_{\mathrm{X}}$)	Zeretzke [76]
Black carbon (BC)	Aulinger et al. [22]
Primary organic aerosols (POAs)	Jalkanen et al. [71]
Mineral ash excl. metal sulphates (MA)	Schwarzkopf et al. [37]
Carbon dioxide (${\mathrm{CO}}_2$)	IMO [3]
Carbon monoxide ($\mathrm{CO}$)	IMO [3]
Methane (${\mathrm{CH}}_4$)	IMO [3]
Nonmethane volatile organic compounds (NMVOCs)	EMEP/EEA [77]
Dinitrogen oxide (${\mathrm{N}}_2\mathrm{O}$)	IMO [3]
Particulate matter (${\mathrm{PM}}_{\mathrm{tot}}$)	EMEP/EEA [77]

lists the gas and aerosol species included in the ship EIs, together with a reference to the source of the EFs used. ${\mathrm{NO}}_{\mathrm{X}}$ emissions were split into $\mathrm{NO}$ and ${\mathrm{NO}}_2$ in the ratio of 92 to 8 for shipping. NMVOCs were split according to data from the Finnish Meteorological Institute (FMI), which was consistent with the results from CIMAC [75] PM_2.5 emissions from ships were split with data from the SMOKE for Europe emission model [65]. Finally, the merged data set was fed into the CMAQ model as emission input.

2.2.5. Ship Emissions in China

The ship EI for China covered the area between the longitude and latitude ranges $133.15$–$102.09°$ and $44.47$–$15.54°$, respectively. The AIS data used for this inventory were obtained through a collaboration with the Shanghai Maritime Department. The ship EI for China generated from these data, had a spatial resolution of $0.039°$ in the x (east–west) and the y (north–south) directions, corresponding to approximately 4 $\mathrm{km}$. The temporal resolution is hourly. For better comparability of the model results, the same parameters and EFs were used with MoSES for this EI as for the EI of northern Europe (Table 1, see also for details [37]).

The spatial distribution of ${\mathrm{CO}}_2$ emission fluxes for 2015 in the CNC12 domain and monthly, normalized emission totals for each considered species are illustrated in Figure 2. The monthly variation in ship emissions showed two minima. The first minimum was registered in February during the long public holidays of the Chinese Spring Festival, and the second minimum was seen during the summer months, which corresponded to the summer fishing moratorium. In contrast, higher fishing activities and corresponding emissions were found in spring and autumn. Similar observations were also made by Chen et al. [78] and Fan et al. [79].

For the calculation of ship emissions, 10 ship types were differentiated plus an “Undefined” class, used for ships for which no type information was present. These were Bulk, Cargo and container, Cruise, Fishing, Military, Passenger and Ro-Ro, Pleasurecraft, Tanker, Tug and Other. The annual emissions for 2015 are shown in

Table 2. Modeled ship emissions in $\mathrm{Gg}·{\mathrm{year}}^{-1}$ for China in 2015 in total and for each individual ship type.

Ship Type	SO2	SO4	SO4 xH2O	NOX	CO2	CO	CH4	NM VOC	N2O	BC	MA	POA	PMtot
All	486.55	14.38	11.22	1678.98	85,325.01	70.3	5.97	21.93	4.36	29.94	4.98	39.87	89.34
Bulk	6.43	0.13	0.10	219.43	12,144.84	9.95	0.20	1.95	0.60	2.81	0.07	4.53	6.80
Cargo	181.93	5.28	4.12	750.25	39,012.88	31.12	0.58	10.22	1.94	13.66	1.86	17.98	38.18
Cruise	0.63	0.02	0.02	8.33	363.08	0.34	0.01	0.06	0.02	0.10	0.01	0.18	0.29
Fishing	26.17	0.77	0.60	51.23	2445.1	2.01	0.03	0.67	0.13	1.03	0.27	1.42	3.68
Military	0.37	0.01	0.01	0.94	43.38	0.04	0.00	0.02	0.00	0.02	0.00	0.02	0.05
Passenger	10.69	0.35	0.27	45.98	1925.97	1.82	0.03	0.59	0.12	0.74	0.11	0.99	2.19
Pleasurec.	1.11	0.03	0.02	2.17	103.08	0.09	0.00	0.05	0.01	0.05	0.01	0.05	0.15
Tanker	31.89	0.99	0.77	177.59	8978.86	8.43	4.83	1.88	0.48	2.54	0.33	3.60	7.33
Tug	15.91	0.49	0.38	27.90	1600.8	1.36	0.03	0.59	0.09	0.74	0.16	0.85	2.33
Other	81.46	2.45	1.91	150.52	6679.75	5.89	0.10	2.94	0.38	3.54	0.83	3.78	11.13
Undef.	129.96	3.85	3.00	244.63	12,027.27	9.26	0.16	2.96	0.60	4.70	1.33	6.46	17.22

as totals and for the individual ship types. The majority of emissions could be accounted for by freight ships. Hereby, cargo ships, including container and Ro-Ro cargo ships, had the largest share, with an annual average of 38% of ${\mathrm{SO}}_2$, ${\mathrm{SO}}_4$ and MA, 45% of ${\mathrm{NO}}_{\mathrm{X}}$, ${\mathrm{CO}}_2$, $\mathrm{CO}$, NMVOC, ${\mathrm{N}}_2\mathrm{O}$, BC and POA and 10% of ${\mathrm{CH}}_4$ and 43% of ${\mathrm{PM}}_{\mathrm{tot}}$ emissions. A decreased activity for this ship type was modeled in February and March. Bulk freighters instead accounted for only 1% of ${\mathrm{SO}}_2$, ${\mathrm{SO}}_4$ and MA, 14% of ${\mathrm{CO}}_2$, $\mathrm{CO}$, ${\mathrm{NO}}_{\mathrm{X}}$ and ${\mathrm{N}}_2\mathrm{O}$, 3% of ${\mathrm{CH}}_4$, 9% of NMVOC and BC, 11% POA and 5% ${\mathrm{PM}}_{\mathrm{tot}}$ emissions. The share of tankers, the last type of freight ships considered, was 7% of the emissions of ${\mathrm{SO}}_2$, ${\mathrm{SO}}_4$, MA and ${\mathrm{PM}}_{\mathrm{tot}}$. They also accounted for 11% of the emissions of ${\mathrm{NO}}_{\mathrm{X}}$, ${\mathrm{CO}}_2$, $\mathrm{CO}$ and ${\mathrm{N}}_2\mathrm{O}$, 80% of ${\mathrm{CH}}_4$ and 9% of NMVOC, BC and POA. In total, freight ships accounted for approximately 50–70% of ship emissions and for almost all methane emissions. The emission share of fishing vessels in the model domain was 5% of ${\mathrm{SO}}_2$, ${\mathrm{SO}}_4$ and MA, 3% of ${\mathrm{NO}}_{\mathrm{X}}$, ${\mathrm{CO}}_2$, $\mathrm{CO}$, NMVOC, ${\mathrm{N}}_2\mathrm{O}$, BC, POA and ${\mathrm{PM}}_{\mathrm{tot}}$ and 1% of ${\mathrm{CH}}_4$ emissions. A seasonality for emissions from fishing vessels was observed, with increased emissions in April and May and October. An explanation for this was the summer fishing ban in the South China Sea enforced since 1999 (https://www.fao.org/fishery/en/facp/chn, last accessed: 14 December 2021). This resulted in a decrease in ${\mathrm{CO}}_2$ emissions to 1% in June, which increased to 5% during peak periods. Tugs and passenger ships accounted for only 2–3% of the emissions of the species considered. Cruise ships, military ships and pleasurecrafts accounted for a negligible share of all emissions species considered, with less than 1%. In this context, it should be noted that AIS is not a reliable source for tracking military vessels. Ships of other types, grouped under the type “Other” (e.g, dredging or drilling ships, patrol or research vessels) accounted for 17% of ${\mathrm{SO}}_2$, ${\mathrm{SO}}_4$, MA and ${\mathrm{PM}}_{\mathrm{tot}}$, 9% of ${\mathrm{NO}}_{\mathrm{X}}$, $\mathrm{CO}$, ${\mathrm{N}}_2\mathrm{O}$ and POA, 8% of ${\mathrm{CO}}_2$, 2% of ${\mathrm{CH}}_4$ and 13% NMVOC and BC emissions. Ships for which no type could be determined due to missing data were assigned the type “Undefined”. Various assumptions had to be made for this type, which usually included estimations of the engine power and fuel type. When vessel size information was not available, “Undefined” ships were considered small, and an approximated gross tonnage of 500 was found to be plausible [37]. This result was supported by the two assumptions that large and commercial vessels are considered more reliable in transmitting information via AIS and that the available vessel characteristics databases are more reliable for larger vessels.

In the CNC12 domain, undefined ships are responsible for approximately 26% of ${\mathrm{SO}}_2$, ${\mathrm{SO}}_4$ and MA, 14% of ${\mathrm{NO}}_{\mathrm{X}}$ and ${\mathrm{CO}}_2$, 13% of $\mathrm{CO}$, NMVOC and ${\mathrm{N}}_2\mathrm{O}$, 3% of ${\mathrm{CH}}_4$ and 16% of ${\mathrm{PM}}_{\mathrm{tot}}$, POA and BC emissions. Numerically, nearly 57% of the vessels had to be classified as “Undefined”; no information on installed main engine power was available for nearly 92% of them. This corresponds to a share of 67% of all vessels for which the installed main engine power had to be estimated. Similar findings were made by Zhang et al. [80] for the PRD, where unidentified ships accounted for 49% of ${\mathrm{CO}}_2$ emissions. This introduces an uncertainty to the calculated emissions due to the estimations that had to be made. In comparison, 11% of undefined ships by number were found in the European EI. Estimations about the main engine power had to be made for more than 56% of ships and for 31% of the total on ships installed main engine power.

In both regions, oil-based residual and distillate fuels are predominant, being used by more than 99% of registered vessels. However, in China, 65% of vessels were found using residual fuels, compared to 37% in Europe. The vast majority of the remaining ships use distillate fuels. The more frequent application of residual fuels is, in addition to the differing legal regulations, responsible for higher ${\mathrm{SO}}_{\mathrm{X}}$ and particle emissions in China.

Of the more than 110,000 of different vessels registered in China in 2015 and of the nearly 22,000 registered in Europe, 4090 vessels were found in both regions. Most of them, 95%, were freight ships, with a gross tonnage over 5000. This was plausible, as larger ships are usually deployed for this long voyage. They were distributed among the ship types considered: 38% bulk freighters, 35% cargo and container ships, 23% tankers, 2% other ships. The remaining ship types make up less than 1%.

3. Assessment of the Model Performance

To assess the reliability and performance of the model results for the European SC12NSBS and Chinese CNC12 domains, they were compared with publicly available air quality data from the European Environment Agency (EEA) and the Chinese authorities. The data for Europe were acquired from the Air Quality e-Reporting (AQER) repository, which contains data from 2013 onwards [81]. The data for China were acquired from the China National Environment Monitoring Centre (http://www.cnemc.cn, last accessed: 27 April 2022).

Since the focus was on the influence of ship emissions, measurement data from monitoring stations in coastal regions near the major shipping routes were evaluated. Note that a point measurement at the station site and an average concentration in a $12\times 12$ $\mathrm{km}$ grid cell enclosing the station were compared. Therefore, the measured concentrations may not be fully representative of the grid cell, as the resolution of the regional model does not allow for the consideration of small-scale measurement conditions and effects near the station. Although the uncertainties of this method are hard to estimate, it can be seen as a performance indicator for the model.

A comparison of the concentration time series was performed (for ${\mathrm{NO}}_2$ and ${\mathrm{SO}}_2$, emissions were included in the comparison), and statistical quantifiers were calculated from the hourly concentrations for each pollutant. These include the geometric mean for all stations, two arithmetic mean values calculated using only positive or negative normalized mean biases (NMB_pos, NMB_neg) and the arithmetic mean of the Spearman correlation coefficients. Only stations with data availability greater than 60% of hourly measurements for 2015 were considered for comparison. A total of 29 stations in the SC12NSBS domain and 20 stations in the CNC12 domain were considered (their locations are shown in Figure 1). A comparison of the modeled “base” case concentrations with the measured data is shown for 6 selected stations for each domain (

Table 3. Comparison of modeled ${\mathrm{NO}}_2$, ${\mathrm{SO}}_2$, ${\mathrm{O}}_3$ 8-h mean and ${\mathrm{PM}}_{2.5}$ concentrations of the modeled “base” case for selected stations in $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ with measurements in the SC12NSBS domain. Based on all 29 considered stations in Europe, the mean correlation coefficient (Mean Corr.) and two values, calculated as the mean of either positive or negative normalized mean biases are shown (NMB_pos, NMB_neg). For the latter two values, the number of stations used for the calculation is given in parentheses.

	NO2		SO2		O3 8-h Mean		PM2.5
Station	Meanmodel	Meanmeas	Meanmodel	Meanmeas	Meanmodel	Meanmeas	Meanmodel	Meanmeas
Narberth	$1.60$	$1.63$	$0.37$	$0.68$	$75.12$	$61.09$	—	—
Lull. Heath	$4.44$	$4.79$	$0.28$	$0.95$	$73.98$	$53.61$	—	—
De Zilk	$12.99$	$9.83$	$0.73$	$0.95$	$48.49$	$42.27$	$9.73$	$6.61$
Ostf. Inseln	$3.43$	$6.39$	$0.14$	$0.40$	$72.08$	$55.72$	—	—
Zingst	$2.95$	$3.73$	$0.21$	$0.39$	$72.69$	$57.10$	—	—
Helsinki	$5.90$	$14.01$	$1.00$	$0.61$	$58.39$	$44.94$	$4.84$	$4.22$
Mean NMBpos	$0.398\left(1\right)$		$0.365\left(10\right)$		$0.27\left(28\right)$		$0.164\left(9\right)$
Mean NMBneg	$-0.38\left(28\right)$		$-0.266\left(6\right)$		n.a.$\left(0\right)$		$-0.03\left(2\right)$
Mean Corr.	$0.617$		$0.279$		$0.659$		$0.377$

and

Table 4. Comparison of modeled ${\mathrm{NO}}_2$, ${\mathrm{SO}}_2$, ${\mathrm{O}}_3$ 8-h mean and ${\mathrm{PM}}_{2.5}$ concentrations of the modeled “base” case for selected stations in $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ with measurements in the CNC12 domain. Based on all 20 considered stations in China, the mean correlation coefficient (Mean Corr.) and two values, calculated as the mean either from positive or negative normalized mean biases are shown (NMB_pos, NMB_neg). For the latter two values, the number of stations used for the calculation is given in parentheses.

	NO2		SO2		O3 8-h Mean		$\mathbf{PM}2.5$
Station	Meanmodel	Meanmeas	Meanmodel	Meanmeas	Meanmodel	Meanmeas	Meanmodel	Meanmeas
Tianjin	$56.95$	$33.53$	$29.83$	$19.00$	$4.32$	$29.99$	$59.93$	$50.52$
Lianyungang	$15.05$	$24.31$	$4.75$	$22.07$	$57.19$	$60.35$	$42.26$	$39.82$
Nantong	$36.79$	$30.45$	$11.84$	$23.50$	$37.06$	$62.88$	$42.19$	$45.28$
Ningbo	$44.97$	$37.27$	$16.06$	$14.36$	$36.73$	$50.95$	$37.51$	$35.30$
Shantou	$15.01$	$16.98$	$6.05$	$11.64$	$78.48$	$61.95$	$42.11$	$28.22$
Zhuhai	$14.25$	$23.30$	$5.03$	$7.03$	$79.18$	$51.72$	$45.85$	$23.98$
Mean, NMBpos	$0.402\left(12\right)$		$0.512\left(9\right)$		$0.544\left(18\right)$		$0.439\left(17\right)$
Mean, NMBneg	$-0.283\left(8\right)$		$-0.38\left(11\right)$		$-0.1\left(2\right)$		$-0.06\left(3\right)$
Mean Corr.	$0.477$		$0.398$		$0.628$		$0.408$

). These include statistical quantifiers calculated with data from all stations (Table 3). The comparison data of all stations are shown in

Table A1. Comparison of modeled ${\mathrm{NO}}_2$, ${\mathrm{SO}}_2$, ${\mathrm{O}}_3$ 8-h mean and fine particulate matter (${\mathrm{PM}}_{2.5}$) concentrations of the modeled “base” case in $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ with measurements in the SC12NSBS domain. Additionally, the mean correlation coefficient (Mean Corr.) and two values, calculated as the mean of either positive or negative NMBs are shown (NMB_pos, NMB_neg). For the latter two values, the number of stations used for the calculation is given in parentheses.

	NO2		SO2		O3 8-h Mean		PM2.5
Station	Meanmodel	Meanmeas	Meanmodel	Meanmeas	Meanmodel	Meanmeas	Meanmodel	Meanmeas
Oismäe	$3.65$	$5.13$	$0.39$	$0.47$	$67.60$	$51.05$	—	—
Phare d’Ailly	—	—	—	—	$76.09$	$59.62$	—	—
Schoten	$17.79$	$21.26$	—	—	$45.04$	$29.50$	$10.87$	$10.79$
Utö	$1.16$	$2.25$	$0.10$	$0.30$	$74.56$	$66.32$	$4.00$	$3.45$
Newcastle	$6.06$	$25.56$	—	—	$66.52$	$36.56$	$7.96$	$7.80$
Århus	$3.23$	$9.61$	—	—	$72.37$	$50.36$	—	—
Westerland	$1.34$	$2.17$	$0.09$	$0.26$	$77.55$	$61.91$	—	—
De Zilk	$12.99$	$9.83$	$0.73$	$0.95$	$48.49$	$42.27$	$9.73$	$6.61$
Den Haag	$17.91$	$22.20$	—	—	$39.51$	$33.29$	—	—
Wieringerwerf	$5.30$	$8.55$	—	—	$68.96$	$44.15$	$8.93$	$6.04$
Pyykösjärvi	$3.00$	$6.61$	—	—	$51.92$	$43.39$	—	—
Råö	—	—	—	—	$74.34$	$62.74$	—	—
Brighton	$4.77$	$10.72$	—	—	$72.40$	$51.52$	—	—
Plymouth	$2.27$	$13.53$	—	—	$76.77$	$45.59$	$8.52$	$9.33$
Narberth	$1.60$	$1.63$	$0.37$	$0.68$	$75.12$	$61.09$	—	—
Blackpool	$2.47$	$11.46$	—	—	$68.55$	$49.39$	$8.17$	$5.78$
Dublin	$3.05$	$4.94$	—	—	—	—	—	—
Hamburg	$12.12$	$26.39$	$1.07$	$3.28$	—	—	—	—
Lahemaa	$1.33$	$1.61$	$0.23$	$0.53$	$65.52$	$50.24$	—	—
Ostfries. Inseln	$3.43$	$6.39$	$0.14$	$0.40$	$72.08$	$55.72$	—	—
Elbmündung	$5.33$	$9.56$	—	—	$67.33$	$47.54$	—	—
Virolahti	$1.56$	$2.74$	$0.26$	$0.21$	$67.24$	$47.51$	$3.87$	$4.19$
Vilsandi	$0.98$	$1.37$	$0.11$	$0.27$	$75.61$	$65.64$	—	—
Copenhagen	$7.13$	$12.32$	—	—	$66.14$	$53.76$	—	—
Ulborg	$1.54$	$2.77$	—	—	$73.55$	$61.99$	—	—
Kallio	$5.90$	$14.01$	$1.00$	$0.61$	$58.39$	$44.94$	$4.84$	$4.22$
Houtem	$7.19$	$7.12$	$1.19$	$1.07$	$64.51$	$47.15$	$9.65$	$8.12$
Gent	$13.51$	$23.40$	$0.94$	$1.23$	$50.48$	$32.78$	$10.57$	$11.15$
Lullington Heath	$4.44$	$4.79$	$0.28$	$0.95$	$73.98$	$53.61$	—	—
Zingst	$2.95$	$3.73$	$0.21$	$0.39$	$72.69$	$57.10$	—	—
Gdańsk Nowy Port	$5.80$	$10.44$	$1.67$	$2.17$	—	—	—	—
Mean NMBpos	$0.398\left(1\right)$		$0.365\left(10\right)$		$0.27\left(28\right)$		$0.164\left(9\right)$
Mean NMBneg	$-0.38\left(28\right)$		$-0.266\left(6\right)$		n.a. $\left(0\right)$		$-0.03\left(2\right)$
Mean Corr.	$0.617$		$0.279$		$0.659$		$0.377$

and

Table A2. Comparison of modeled ${\mathrm{NO}}_2$, ${\mathrm{SO}}_2$, ${\mathrm{O}}_3$ 8-h mean and ${\mathrm{PM}}_{2.5}$ concentrations of the modeled “base” case in $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ with measurements in the CNC12 domain. Additionally, the mean correlation coefficient (Mean Corr.) and two values, calculated as the mean of either positive or negative NMBs are shown (NMB_pos, NMB_neg). For the latter two values, the number of stations used for the calculation is given in parentheses.

	NO2		SO2		O3 8-h Mean		$\mathbf{PM}2.5$
Station	Meanmodel	Meanmeas	Meanmodel	Meanmeas	Meanmodel	Meanmeas	Meanmodel	Meanmeas
Dalian	$53.68$	$29.53$	$24.74$	$19.59$	$26.28$	$64.12$	$43.42$	$33.98$
Huludao	$16.78$	$29.73$	$7.85$	$33.10$	$52.85$	$48.38$	$31.79$	$40.38$
Qinhuangdao	$29.04$	$38.01$	$14.51$	$28.64$	$31.46$	$30.50$	$38.45$	$30.11$
Tianjin	$56.95$	$33.53$	$29.83$	$19.00$	$4.32$	$29.99$	$59.93$	$50.52$
Lianyungang	$15.05$	$24.31$	$4.75$	$22.07$	$57.19$	$60.35$	$42.26$	$39.82$
Yancheng	$18.57$	$19.60$	$3.75$	$15.53$	$59.14$	$72.76$	$41.12$	$35.52$
Nantong	$36.79$	$30.45$	$11.84$	$23.50$	$37.06$	$62.88$	$42.19$	$45.28$
Shanghai	$59.11$	$39.21$	$27.09$	$14.70$	$21.57$	$60.07$	$41.51$	$41.10$
Ningbo	$44.97$	$37.27$	$16.06$	$14.36$	$36.73$	$50.95$	$37.51$	$35.30$
Wenzhou	$30.72$	$38.75$	$9.05$	$11.96$	$51.90$	$33.98$	$36.03$	$37.29$
Fuzhou	$23.77$	$28.92$	$6.19$	$5.78$	$70.80$	$40.27$	$34.67$	$24.41$
Quanzhou	$24.96$	$21.22$	$9.87$	$8.60$	$63.49$	$46.82$	$44.64$	$23.66$
Shantou	$15.01$	$16.98$	$6.05$	$11.64$	$78.48$	$61.95$	$42.11$	$28.22$
Shenzhen	$47.27$	$30.07$	$15.55$	$7.93$	$27.79$	$48.69$	$48.70$	$25.08$
Guangzhou	$58.01$	$41.38$	$20.86$	$10.96$	$22.08$	$26.21$	$52.25$	$32.91$
Zhongshan	$20.33$	$23.39$	$7.03$	$9.99$	$64.65$	$33.83$	$48.02$	$26.30$
Zhuhai	$14.25$	$23.30$	$5.03$	$7.03$	$79.18$	$51.72$	$45.85$	$23.98$
Haikou	$6.92$	$11.34$	$2.98$	$4.84$	$85.70$	$44.20$	$37.56$	$17.73$
Beihai	$5.38$	$12.10$	$4.93$	$7.90$	$92.03$	$65.84$	$42.38$	$19.20$
Fangchenggang	$5.04$	$10.45$	$4.48$	$4.89$	$87.99$	$36.73$	$41.99$	$24.14$
Mean, NMBpos	$0.402\left(12\right)$		$0.512\left(9\right)$		$0.544\left(18\right)$		$0.439\left(17\right)$
Mean, NMBneg	$-0.283\left(8\right)$		$-0.38\left(11\right)$		$-0.1\left(2\right)$		$-0.06\left(3\right)$
Mean Corr.	$0.477$		$0.398$		$0.628$		$0.408$

. For station-specific NMBs and correlation coefficients for the modeled “base” and “no ships” case, refer to Section S2 of the Supplementary Material. A high correlation coefficient was an indicator of the representativeness of the time profiles used for the concentration patterns at the monitoring station. A correlation was considered for a coefficient greater than or equal to $0.5$. The correlation was considered “good” if the coefficient was equal to or greater than $0.7$.

All of the 29 stations in the SC12NSBS domain in Europe were background monitoring stations in less densely populated locations, e.g., suburban or rural stations were preferred in the selection. ${\mathrm{NO}}_2$ concentrations were often underpredicted by the model, indicated by 28 of 29 stations having a negative NMB. For 6 of the 29 stations a correlation coefficient of $0.7$ or greater was found. For 26 stations, the coefficient was greater than or equal to $0.5$. In general, the NMB was higher for stations where lower concentrations were modeled. When comparing the “base” with the “no ships” case, lower concentrations were found at all stations, which highlights the importance of including ship emissions in the model. Consequently, a higher absolute NMB was calculated, indicating a poorer agreement between the “no ships” and the measurements. This was consistent with the assumption that shipping is an important source of air pollution for many of these stations because they were located near major shipping routes. Considering the correlation, differences were not so clear since the disparities between the “base” and “no ships” cases were small.

In the SC12NSBS domain, ${\mathrm{SO}}_2$ concentrations were generally low at the stations considered, mostly below 1 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$. In comparison to the measurements, the modeled ${\mathrm{SO}}_2$ concentrations were often overestimated, as 10 of the 16 stations had an NMB greater than zero. A correlation coefficient greater than $0.5$ was calculated for only one station. The low correlation could be explained by the overall low ${\mathrm{SO}}_2$ concentrations, which are both difficult to measure and model. The correlation differences between the “base” and “no ships” cases were generally small.

The distinct diurnal concentration changes and seasonal trends of the ${\mathrm{O}}_3$ 8-h mean were well represented in the model results, as shown by correlations greater than $0.5$ for all 28 stations. For 9 stations, the correlation was greater than $0.7$. The average NMB of $0.27$ indicated that concentrations at all 28 stations were systematically overestimated. This overestimation was most concise in spring and autumn. Furthermore, large daily fluctuations were often not captured well by the model. In general, the correlation was better for the “base” case than for the “no ships” case. The performance gain of the model by including ship emissions was small for most stations. However, for stations close to the port cluster of Rotterdam and Antwerp, i.e., Schoten, De Zilk and Den Haag, the performance gain was significant. The observed ozone reduction by ship exhausts in this region is also visible in concentration patterns, which are shown and discussed later in Section 4.3.

Of the 29 stations in total, ${\mathrm{PM}}_{2.5}$ was only measured at 11 sites in The Netherlands, Belgium, England and Finland. According to the NMB, the modeled concentrations were mostly overestimated. For two stations, a correlation equal to or greater than $0.5$ was calculated; for none, a correlation equal to or greater than $0.7$. The low correlation was explained in terms of the complex nature of ambient ${\mathrm{PM}}_{2.5}$ concentrations. The chemical and physical transformations that generate particulate matter, and that particles themselves undergo mostly depend on location-specific parameters. Local events can generate high fluctuations in hourly measurements, which are difficult to reproduce with a regional model. However, the correlation coefficient increased at all regarded stations by the inclusion of ship emissions. The NMB decreased for some cases due to an overestimation of the concentrations.

The 20 stations in the CNC12 domain in China are located in or close to large coastal cities (Table 4).

${\mathrm{NO}}_2$ concentrations were overestimated at 12 of the 20 stations, but at 9 of these 12 stations, the NMB was only small (≤0.1). An underprediction was found for 8 stations, of which 7 had an NMB lower than $-0.1$. No general systematic was found for this observation. An analysis of the ${\mathrm{NO}}_2$ emissions that were used as input for the model run leads to manifold reasons. For stations bordering the Bohai Sea, emissions were either too high or too low. In Jiangsu Province, summer and winter trends were not well captured. ${\mathrm{NO}}_{\mathrm{X}}$ emissions were also too high for the stations in Shanghai and Ningbo; in the case of Shanghai, this may be related to the location of the monitoring station, which is on the Huangpu River, while the corresponding model cell covers a large part of the urban center. A comparison of the model results with the measurements in the PRD showed that the diurnal trend was more pronounced in the model than in the measurements. At the three southernmost stations, emissions were found to be too low. A correlation coefficient greater than $0.5$ was found for 8 stations; no station had a correlation greater or equal $0.7$. The low correlation might be explained by the urban environments in which most stations are located, which has a large influence on the hourly concentration patterns. Nevertheless, the main concentration features could be reproduced for most stations with the applied time profiles, although some concentration peaks were slightly shifted in time.

In general, the measured and modeled ${\mathrm{SO}}_2$ concentrations at stations in China were much higher than those in Europe. The model underestimated the concentrations of ${\mathrm{SO}}_2$ at 11 and overestimated them at the other 9 stations. The correlation coefficient was greater than or equal to $0.5$ for 7 stations and was not greater than or equal to $0.7$ for any station. In contrast to Europe, better correlations were calculated, which is probably related to the higher ${\mathrm{SO}}_2$ concentrations. An examination of the emissions at the measurement sites suggested that ${\mathrm{SO}}_2$ emissions were mainly too high at station sites in the YRD or PRD. Too low emissions were frequently found at northern stations, e.g., Huludao, Qinhuangdao, Lianyungang and Yancheng.

Similar to Europe, the diurnal ozone fluctuations were well modeled, but concentrations were overestimated at 17 of 20 stations. However, the correlations were greater than or equal to $0.5$ at 16 stations and greater than or equal to $0.7$ at 7 stations. In general, the correlation coefficients in China were higher for the northern stations than for the southern stations. In the north, overestimations were observed mainly during summer, while this trend was reversed in the south, which is presumably related to the regionally different climate zones. While the strong daily ${\mathrm{O}}_3$ fluctuations in Europe were often underestimated, the results for China were in better agreement. The largest improvements from the inclusion of ship emissions were found in southern China, e.g., Quanzhou, Fuzhou and Wenzhou. In the large port clusters in the YRD or PRD, the impact of ships on ozone concentrations was small.

Compared to the measurement data, the modeled ${\mathrm{PM}}_{2.5}$ concentrations were overestimated. A correlation coefficient greater than or equal to $0.5$ was calculated for 7 stations all located in the northern part of the domain, i.e., Wenzhou, and at stations further north with a steady increase in correlation up to $0.69$ at Qinhuangdao. Additionally, a small NMB was calculated for these stations. In southern China, the model performance for ${\mathrm{PM}}_{2.5}$ was worse. While the mean of the 10 stations, including Fuzhou and those farther south, had an average NMB of $0.66$, the average NMB for Wenzhou and the 10 stations further north was $0.11$. Based on this observation, it is assumed that the model performs better in the calculation of ${\mathrm{PM}}_{2.5}$ concentrations for the temperate climate zone.

In summary, the model performance was found to be satisfactory in both domains; however, different conclusions were drawn for Europe and China. The predominantly underestimated ${\mathrm{NO}}_2$ concentrations in Europe, although with good correlations, suggest a systematic bias. The under- and overestimates in China and the moderate correlations indicate inconsistencies in the emission inventory and occasional difficulties of the model in describing ${\mathrm{NO}}_2$ concentrations under the local meteorological conditions and atmospheric background. The better performance and correlation for ${\mathrm{SO}}_2$ in China is related to the difficulties in modeling and measuring the low concentrations that occur in Europe. The pronounced diurnal trends in ozone concentrations were well captured in both domains, as reflected by high correlation coefficients, but a systematic overprediction of ozone concentrations was found for both regions. Concentrations of ${\mathrm{PM}}_{2.5}$ were mostly overpredicted in both domains. Furthermore, it is noteworthy that in China the accuracy for modeled ${\mathrm{PM}}_{2.5}$ concentrations increased at sites further north, which might be related to the temperate climate. However, caution must be exercised in interpreting these observations, as consideration of the immediate environment of the measurement site and small-scale effects is only limited to $12\times 12$ $\mathrm{km}$ resolved model runs. It should also be emphasized that the stations selected for Europe are mostly nonurban background stations, while most of the monitoring stations in China are located in or near major cities. Moreover, differences in the methodologies and sources that were used to create the nonshipping anthropogenic EIs are hardly avoidable. For the CNC12 domain, a second model run was carried out, in which anthropogenic, nonshipping emissions for China were used from the EDGAR instead of the MEIC inventory. The EDGAR inventory has a higher resolution of $0.1°$, compared to $0.25°$ in MEIC. However, the results obtained with EDGAR showed higher overestimations, compared to MEIC. This was especially the case for ${\mathrm{NO}}_2$ and ${\mathrm{O}}_3$. Thus, it was decided that the MEIC inventory would be more reliable for China.

4. Results and Discussion

Pollutants from ship exhausts can be transported over land by onshore winds to degrade air quality in populated areas. During this transport, a portion is removed from the atmosphere by wet and dry deposition processes. Furthermore, chemical reactions are responsible for the formation of secondary pollutants and particulates. For the latter, ammonia emissions from agriculture play a significant role. Rates for the relevant chemical transformations generally increase with temperature. In contrast, nucleation and coagulation of the particulation processes are facilitated at lower temperatures. All of these processes are considered within CMAQ and result in concentration patterns that are discussed and compared for northern Europe and eastern China.

In general, primary pollutants from shipping had the largest impact close to densely shipped routes (e.g., the Yangtze River, the English Channel or the southern North Sea) and in the vicinity of large port clusters (e.g., in the YRD, PRD or Rotterdam/Antwerp/Amsterdam). In Europe, prevailing southwesterly winds transported pollutants towards populated areas. In China, prevailing south and southeasterly winds transported ship emissions onto coastal regions, especially over the flat terrain of the Lower Yangtze Plain (LYP) and North China Plain (NCP). Meteorological conditions proved to be more important for an understanding of the seasonal patterns of ship pollutant concentrations than fluctuations in ship traffic [82]. In particular, the East Asian monsoon is a dominant factor for emissions from a marine environment. Secondary pollutants generally have a higher atmospheric lifetime and can thus be encountered farther away from marine shipping lanes. They were found to have a higher impact on extended coastal regions in both domains.

For an interpretation of the results and an assessment of the impact ship emissions have on air quality, total regional pollutant concentrations must be considered, as well as the contribution from ships only. This was done for the pollutant species ${\mathrm{NO}}_2$ (Section 4.1), ${\mathrm{SO}}_2$ (Section 4.2), ozone (Section 4.3), $\mathrm{PM}2.5$ (Section 4.4) and the main components of $\mathrm{PM}2.5$ from shipping, ammonium (Section 4.4.1), sulfate (Section 4.4.2) and nitrate (Section 4.4.3). The contribution of shipping was determined by the zero-out method, i.e., by subtracting the results of a model run with omitted ship emissions from a model run that considers all emission sources. All concentration fields shown in the following are annual averages, unless otherwise specified. Seasonally averaged concentration fields are found in Section S3 of the supplementary material to avoid disrupting the reading flow by their large numbers. All concentrations are given in $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$. The contribution of shipping to the total pollutant concentrations is shown as a percentage.

4.1. NO₂

Nitrogen dioxide emissions are mainly the result of nitrogen oxidation during combustion processes. ${\mathrm{NO}}_2$ plays an important role in the formation of ground-level ozone (see Section 4.3) and as a precursor of PM (Section 4.4 and Section 4.4.3).

In Europe, annual average concentrations of 15–20 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ were modeled near major cities and industrialized areas, e.g., London, Paris, Prague, the Rhine-Ruhr or Rhine-Main metropolitan regions. The overall highest ${\mathrm{NO}}_2$ concentrations of 30–35 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ were modeled in the area of the Rotterdam/Antwerp/Amsterdam port cluster (Figure 3a).

In the coastal regions of mainland Europe bordering the North Sea and the southeastern coast of the United Kingdom, 1–4 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$${\mathrm{NO}}_2$ could be attributed to shipping, even up to 100–150 $\mathrm{km}$ from the coastline (Figure 3b). The contribution of shipping to total ${\mathrm{NO}}_2$ varied between 20 and 40% in these regions. In ports, almost 50% of ${\mathrm{NO}}_2$ could be attributed to ships (Figure 3c). Due to higher shipping activities in spring and summer, slightly higher concentrations were modeled for these seasons (Figure S2). In winter and autumn, the contribution of shipping to total ${\mathrm{NO}}_2$ was lower as ${\mathrm{NO}}_2$ emissions from residential heating increased and shipping activity decreased (Figure S3).

In China, the concentrations of ${\mathrm{NO}}_2$ were mostly in the range of 20 to 50 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$. In densely populated regions, large cities and heavily industrialized areas, ${\mathrm{NO}}_2$ concentrations reached 60 to 90 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$, e.g., in Jiangsu, Tianjin, Hebei, the PRD, Shanghai, Beijing and the Sichuan Basin (Figure 3d).

The impact of shipping on regional ${\mathrm{NO}}_2$ concentrations was most evident in the YRD, PRD, the associated rivers, the LYP and NCP. ${\mathrm{NO}}_2$ concentrations were in the range of 2 to 8 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ but could locally increase up to 13 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (Figure 3e). Vessel traffic was found to be responsible for 10–20% of the ${\mathrm{NO}}_2$ concentrations in the major port clusters and the LYP and NCP. The highest contribution of up to 50% to the total ${\mathrm{NO}}_2$ concentrations was modeled in parts of the Yangtze and Pearl Rivers (Figure 3f). In large parts of the Chinese coast, onshore pollutant transport and the impact of ${\mathrm{NO}}_2$ by ships were small. In autumn and particularly in winter, higher ship-induced ${\mathrm{NO}}_2$ concentrations and an extended dispersion were modeled, especially close to the Yangtze River (Figure 4a and Figure S6). This was consistent with the results of Shah et al. [83] and could be explained by the longer lifetimes of ${\mathrm{NO}}_{\mathrm{X}}$. High ${\mathrm{NO}}_2$ and ${\mathrm{SO}}_2$ concentrations from other sectors created oxidant-limited conditions during these seasons that reduced the conversion rate for ${\mathrm{NO}}_2$-depleting reactions, such as conversion to nitrate or nitric acid. Additionally, atmospheric oxidants such as ozone or hydroxyl radicals were less readily formed in winter due to lower solar irradiance. Furthermore, the frequent northerly and northwesterly winds during winter were able to transport more pollutants from shipping and other sectors from coastal regions to the open ocean. Combined with the reduced shipping activity, this explained the lower contribution of ships to ${\mathrm{NO}}_2$ in coastal areas during this season. The relatively low concentrations over the Yellow Sea in autumn could be related to dilution by cold air currents from the northeast [82].

When the two regions were compared, approximately $2.5$ times higher ${\mathrm{NO}}_2$ concentrations were found in China. With only slightly higher ship-related ${\mathrm{NO}}_2$ concentrations in China, the relative contribution of ${\mathrm{NO}}_2$ from ships was 30–40% and thus higher in European ports. Furthermore, ${\mathrm{NO}}_2$ concentrations from ships showed little seasonality in Europe, with slightly higher concentrations in spring and summer. In China, reduced ${\mathrm{NO}}_2$ concentrations were modeled in spring and summer, especially in the southern coastal regions (Figure 4b and Figure S5). Higher deposition rates during the summer monsoon were responsible for this observation. Furthermore, high overall ${\mathrm{NO}}_2$ concentrations in China during winter resulted in increased transport and a further dispersion of ${\mathrm{NO}}_2$ from shipping in inland areas (Figure 4b).

4.2. SO₂

Sulfur dioxide emissions are mainly generated by the combustion of fossil fuels. Similar to ${\mathrm{NO}}_2$, it can form secondary aerosols by oxidation, e.g., sulfate (Section 4.4 and Section 4.4.2).

In Europe, ${\mathrm{SO}}_2$ was emitted mainly from coal power plants, e.g., in the Rhine-Ruhr region in Germany or in the Polish provinces of Lodz, Opole and Silesan. Elevated concentrations of $12.5$ $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ were common near these coal power plants, and concentrations up to 4 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ were modeled in large parts of Poland (Figure 5a). In winter and autumn, ${\mathrm{SO}}_2$ concentrations were higher due to residential heating (Figure S7).

Since an SECA was introduced in the North Sea and Baltic Sea in early 2015, ${\mathrm{SO}}_2$ from shipping has become less important as an air pollutant. Nevertheless, small contributions of less than $0.3$ $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (15% of total ${\mathrm{SO}}_2$) were modeled in the vicinity of the ports of Rotterdam, Amsterdam, Antwerp and Hamburg. From outside of the SECA, ${\mathrm{SO}}_2$ concentrations of similar magnitude were transported from the major shipping routes to the coastal regions of Brittany (40%). Furthermore, concentrations of up to $0.8$ $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (50–70%) were modeled in port cities on the east coast of Ireland and the west coast of the United Kingdom (Figure 5d).

In 2015, 64% of the Chinese domestic energy consumption was produced using coal power plants (26% in Europe) [84,85]). This difference was reflected in higher atmospheric ${\mathrm{SO}}_2$ concentrations (Figure 5d). In densely populated and industrialized regions, annual averages of 30 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ were modeled with local concentration peaks of up to 100 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$. Similar to ${\mathrm{NO}}_2$, ship traffic contributed the most to ${\mathrm{SO}}_2$ in the major port clusters in the YRD, PRD and along the Yangtze River, with concentrations ranging from 4 to 8 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (30–40%). ${\mathrm{SO}}_2$ concentrations of 1 to 3 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (10–20%) were modeled along large parts of the coastline (Figure 5e,f). The seasonal patterns of ${\mathrm{SO}}_2$ were similar to those of ${\mathrm{NO}}_2$. Concentrations were higher in winter and autumn, especially in the YRD and PRD port clusters and along navigated rivers (Figure S11). It is assumed that, similar to ${\mathrm{NO}}_2$, chemical conversion and deposition of ${\mathrm{SO}}_2$ from shipping was inhibited due to high ${\mathrm{SO}}_2$ emissions from other sectors.

When comparing the two regions, atmospheric ${\mathrm{SO}}_2$ concentrations in China were found to be approximately 6 times those in Europe. For similar reasons as for ${\mathrm{NO}}_2$, reverse seasonal trends between the two regions were also modeled for ${\mathrm{SO}}_2$.

4.3. Ozone

Tropospheric ozone is known to be harmful to the respiratory tract and is also a potent greenhouse gas [86,87]. ${\mathrm{O}}_3$ is a photochemical product for which formation solar radiation is the most important factor. However, concentrations of tropospheric ozone are also controlled by the availability of reaction partners, such as ${\mathrm{NO}}_{\mathrm{X}}$ and VOCs. For this reason, high VOC/${\mathrm{NO}}_{\mathrm{X}}$ ratios can produce ${\mathrm{NO}}_{\mathrm{X}}$-limited conditions that favor ozone formation and usually occur in rural areas. Vice versa, low VOC/${\mathrm{NO}}_{\mathrm{X}}$ ratios and VOC-limited conditions are common in urban centers and promote ${\mathrm{O}}_3$ titration. By this means, ship emissions that contain ${\mathrm{NO}}_{\mathrm{X}}$ and VOCs can have a significant influence on regional ozone concentrations. With respect to the comparison with measurements in Section 3, it must be noted that ozone concentrations were systematically overestimated in both domains.

In Europe, annual averages of 8-h max. ozone concentrations over land were modeled in the range of 75 to 90 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (Figure 6a). Due to high cloud coverage in Europe during the summer months of 2015, especially over the North Sea, the United Kingdom and the Scandinavian countries, the modeled ${\mathrm{O}}_3$ concentrations were lower for summer than spring (Figure S13). However, this result needs to be interpreted with care, as the ozone overestimation was found to be higher in spring than in summer.

Due to high ${\mathrm{NO}}_{\mathrm{X}}$ emissions from shipping and VOC-limited conditions in the area of the harbors of Rotterdam, Antwerp and Hamburg, ${\mathrm{O}}_3$ concentrations were reduced to approximately 70–80 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ by ${\mathrm{NO}}_{\mathrm{X}}$-promoted ozone decomposition. This corresponded to an ozone reduction of 10 to 15 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ and a relative reduction of 10 to 15% (Figure 6b,c, illustrated ozone concentrations here represent annual averages instead of averages of the ${\mathrm{O}}_3$ 8-h max.). In other areas of the SC12NSBS domain, ship emissions increased ozone by 1–3 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (1–3%). The reduction potential was stronger in winter when the solar irradiation was lower (Figure 7a). During summer, a ship-related increase in ozone concentrations by 4 to 8 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (5–10%) was modeled in Denmark, southern Sweden, Norway and northern Germany, as well as in coastal regions bordering the Baltic Sea and the English Channel (Figure 7b, Figures S14 and S15).

In China, average values of 8-h max. ozone concentrations were modeled between 140 and 150 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$. Values were lower in the southern- and northernmost coastal regions, 100–120 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (Figure 6d). Regarding the seasonality of ${\mathrm{O}}_3$, different factors had to be considered for mainland China. While in the winter radiation was lower, especially in the northeast, high ${\mathrm{NO}}_{\mathrm{X}}$ emissions from residential heating promoted ozone decomposition. Furthermore, an increased aerosol load was able to take up gaseous precursors relevant for ozone formation [88]. Consequently, ${\mathrm{O}}_3$ concentrations were very low (e.g., in the LYP and NCP during winter, often less than 50 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$).

In China, an ozone reduction from ship emissions of 5–15 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ could mainly be observed along the Yangtze River. As in Europe, the reason for this was a smaller VOC/${\mathrm{NO}}_{\mathrm{X}}$ ratio, which decreased through ship emissions. At the large port cluster in the YRD and PRD, VOC-limited conditions were plausible; however, the impact of ${\mathrm{NO}}_{\mathrm{X}}$ from ships on ozone degradation was small. An explanation for this was the low relative contributions of ships to the total ${\mathrm{NO}}_2$ concentrations, which were lower than, for example, in Rotterdam/Antwerp. This was also consistent with the results in Section 3, which indicated that the inclusion of ship emissions has little effect on ozone concentrations in the YRD and PRD. Increased ozone concentrations due to ship emissions were found along the southern coastline by 4–8 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (5–8%) and by approximately 2–3 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ along the northern coastline (Figure 6e,f)). Here, it was plausible that ozone whose formation was promoted by ship emissions in marine environments might have been transported to coastal regions by onshore winds. During spring and especially in summer, higher ${\mathrm{O}}_3$ concentrations from ships were shifted towards the northeast, while in autumn and winter, high concentrations were shifted to the southeast (Figure 7c, Figures S17 and S18). Due to less irradiation during winter, a seasonal reduction in ${\mathrm{O}}_3$ concentrations through ship emissions was modeled in the PRD, the YRD and along the Yangtze River. An ozone reduction during winter was also evident in the Bohai and Yellow Seas, where it was plausible that ${\mathrm{NO}}_{\mathrm{X}}$ was transported from northern China and the Korean Peninsula to the marine environment to create VOC-limited conditions. Concentrations were reduced by approximately 10–20 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (10–20%) in the YRD, PRD and near the Yangtze and by 2 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (2%) at the northern coastline.

In contrast to Europe, ozone concentrations in China were less reduced by ship emissions, especially in the large port clusters. This could be attributed to the lower contribution of shipping to total ${\mathrm{NO}}_2$ concentrations in China. In addition, ship emissions increased ozone formation in many coastal regions that were close to the major shipping lanes. However, in China this observation was much more pronounced than in Europe, as significant amounts of ozone formed in marine areas were transported to land. A similar seasonality for ${\mathrm{O}}_3$ was modeled for northern Europe and northern China, with lower ${\mathrm{O}}_3$ concentrations in winter and autumn (Figure 7, Figures S14 and S17). A reversal of this seasonality was found for the subtropical climate zone of southern China.

4.4. Fine Particulate Matter (PM_2.5)

With respect to its source, ${\mathrm{PM}}_{2.5}$ is divided into primary and secondary particles. Primary ${\mathrm{PM}}_{2.5}$ is emitted directly from the emission source. With respect to ship emissions, these are sulfate, water associated with sulfate, BC, MA and POAs. However, the major impact of PM_2.5 from shipping usually stems from secondary nitrate and sulfate particles, which are formed in atmospheric chemical reactions from the precursors ${\mathrm{NO}}_{\mathrm{X}}$ and ${\mathrm{SO}}_2$. Normally, these form as the ammonium salts (Section 4.4.1): ammonium sulfate (Section 4.4.2) and ammonium nitrate (Section 4.4.3). The oxidation rate for these compounds, as well as nucleation and particle growth, depends on environmental parameters (e.g., temperature, solar radiation and humidity). Furthermore, the availability of reaction partners such as ammonia (${\mathrm{NH}}_3$), hydroxyl radicals (${\mathrm{OH}}^{·}$) and ozone (${\mathrm{O}}_3$) is of importance. The high atmospheric lifetime of aerosol species enables ${\mathrm{PM}}_{2.5}$ from shipping to be transported hundreds of kilometers away from the main shipping routes to inland areas [89].

In Europe, the highest annual average ${\mathrm{PM}}_{2.5}$ concentrations of 12–15 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ were modeled in neighboring states of the North Sea, close to the major shipping lanes, i.e., northern France, England, northern Germany and especially Belgium and The Netherlands (Figure 8a). In these regions, the formation of secondary aerosols was promoted by high ammonia emissions from agriculture, which were highest during spring, the main season for fertilization (Figure S19); see also Section 4.4.1, Section 4.4.2 and Section 4.4.3).

Ship emissions were relevant precursors for the formation of secondary particulates and contributed approximately 10% to the modeled concentrations (1 to $1.25$ $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$, Figure 8b,c). The main spatial patterns were preserved during all seasons; however, the most impacted region shifted eastwards during winter to northern Germany and Denmark. In summer, the field with the highest concentrations was stretched along the English Channel and the North Sea coastline of the European mainland. This observed shift corresponded to the frequent southwesterly wind direction in winter and the westerly wind directions in summer. During spring, it is evident that ship emissions, combined with the higher ammonia emissions, result in a strong increase in ship-related ${\mathrm{PM}}_{2.5}$ concentrations (Figure 9c).

In China, the highest annual average concentrations of 100–125 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ were modeled in megacities and regions with high traffic and industry densities, e.g., Chongqing and Chengdu in the Sichuan Basin, Wuhan and the Beijing-Tianjin-Hebei city cluster (Figure 8c). During winter, ${\mathrm{PM}}_{2.5}$ concentrations were highest, with concentrations up to 150–200 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (Figure S22). Although China has a large agricultural sector, a concise increase in ${\mathrm{PM}}_{2.5}$, promoted by high ammonia emissions from agricultural fertilization, which has its main season here during summer, could not be observed.

${\mathrm{PM}}_{2.5}$ concentrations of 1–$1.5$ $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ from shipping were modeled along large parts of the Chinese coastline and in the LYP and NCP; concentrations of 2 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ were found in the vicinity of the port cluster in the YRD and PRD and in parts of the Yangtze River. The share of shipping on total ${\mathrm{PM}}_{2.5}$ concentrations in China was 2–3% small due to the high ${\mathrm{PM}}_{2.5}$ contributions from other sectors (Figure 8e,f). Remarkably, almost no ship-related impact on ${\mathrm{PM}}_{2.5}$ concentrations was modeled in northeastern China during winter (Figure 9e). It is assumed that in winter, the ratio of aerosol precursors from other sectors was too high to allow significant particle generation from ship emissions. Furthermore, lower concentrations were modeled for spring and summer along the southern coastline due to the higher temperatures and a higher deposition during the rainy season (Figure 9f,g).

In summary, it can be said that the modeled ${\mathrm{PM}}_{2.5}$ concentrations were approximately 6 times higher in China than in Europe. The ${\mathrm{PM}}_{2.5}$ concentrations due to ship traffic were on a similar level, resulting in a lower contribution from ships to total ${\mathrm{PM}}_{2.5}$ concentrations. Notable differences in seasonality were found between the two regions (Figure 9).

To gain better insight into these aspects, the main components of ${\mathrm{PM}}_{2.5}$, ammonium, sulfate and nitrate concentrations were analyzed for both regions in the following sections. Hereby it was to consider that particle formation is strongly dependent on microclimatic conditions, such as temperature, humidity, concentrations of oxidants and precursors, and the presence of pre-existing aerosols [90,91].

4.4.1. Ammonium (NH₄)

Ammonium is a typical secondary aerosol component originating mainly from agricultural ammonia emissions. The modeled ammonium concentrations presented in this chapter refer to the ${\mathrm{NH}}_4^{+}$ mass component in secondary aerosols, i.e., approximately the sum of the mass of ammonium sulfate and ammonium nitrate particles minus the mass of sulfate and nitrate.

In both regions, a strong seasonality and temperature dependency could be seen for the formation of ammonium particles with higher concentrations during winter and lower concentrations during summer. Seasonality was less pronounced for ammonium from shipping.

In Europe, ammonium originating from ship emissions increased aerosol pollution by $0.2$ $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ in most coastal regions bordering the main shipping routes (Figure 10b). Analogous to the total ${\mathrm{PM}}_{2.5}$, elevated ammonium concentrations of $0.4$ $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ were modeled in coastal regions of mainland northern Europe in spring. These are generated by ammonia from agricultural fertilization in combination with ship emissions (Figures S25 and S31). The lowest impact of ammonium from shipping was observed in summer, when high temperatures impeded the nucleation of particles. An approximate contribution of 20% to total ammonium concentrations in Europe could be ascribed to ships (Figure 10c). This contribution could rise up to 40–50% during summer, when fewer precursor species from other emission sectors were present.

In China, ammonium from shipping was modeled in concentrations of $0.2$ $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ along the coastline. Higher concentrations of $0.3$–$0.4$ $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ were found in the YRD, along the Yangtze, and in the LYP and the NCP (Figure 10e). The contribution of shipping to ammonium was 12% on the southeast coast and 8% in the northeast coastal regions (Figure 10f). Distinct regional and seasonal differences were evident in the model results. Similar to ${\mathrm{PM}}_{2.5}$, no contribution of shipping to ammonium was modeled in northeast China during winter. Due to the rainy season in spring and summer, almost no ammonium contributions from shipping were modeled at the southern coastline (Figure S29).

A comparison of the two regions showed that ammonium concentrations in China were approximately 5 times higher than those in northern Europe, which was in agreement with the results of the modeled ${\mathrm{PM}}_{2.5}$ totals (Figure 10a,d). The contribution from shipping, however, was in both domains on a similar scale. In China, the spatial pattern of ${\mathrm{NH}}_4$ concentrations corresponded better to ammonia emissions, which was probably related to the overall higher ${\mathrm{NH}}_3$ emissions (Figures S28 and S32). Despite the large agricultural sector in China and the high ammonia emissions during summer, no peak ${\mathrm{NH}}_4$ concentrations from fertilization were modeled, as in Europe during spring. A reason for this could be found in the high deposition rates during the rainy season in summer. Similar to ${\mathrm{PM}}_{2.5}$ concentrations, the remarkable seasonal concentration differences for ammonium from shipping in China were very different from the more uniform seasonality observed in Europe. They could be attributed to the high atmospheric background pollution in China and different weather phenomena.

4.4.2. Sulfate (SO₄)

Sulfate particles form from ${\mathrm{SO}}_2$ in the atmosphere via a photoinduced gas-phase oxidation over the intermediate ${\mathrm{SO}}_3$ or in an aqueous phase reaction. Sulfate is deposited as sulfuric acid or irreversibly combines with ammonium in a reaction favored over the formation of ammonium nitrate [92].

In both domains, sulfate concentrations corresponded to the spatial and seasonal patterns of ${\mathrm{SO}}_2$ concentrations, although they were more diffuse due to a longer particle lifetime. Accordingly, sulfate from shipping played a subordinate role in the area of the northern European SECA (see also Section 4.2). However, ${\mathrm{SO}}_2$ emitted in the Celtic Sea, outside the SECA, was able to form aerosols that were transported by westerly winds to the United Kingdom and the French coastline in concentrations of approximately $0.1$ to $0.15$ $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (Figure 11b). Sulfate concentrations were highest during summer due to increased shipping activity (Figure S43).

In China, sulfate from shipping impacts the air quality along the whole coastline and along the Yangtze River. The highest concentrations of approximately $0.5$ $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (5–10%) were found near the port cluster in the YRD and PRD (Figure 11e). Similar to ${\mathrm{PM}}_{2.5}$ and ammonium, almost no ${\mathrm{SO}}_4$ concentrations from shipping were modeled in northeastern China during winter, while in summer, concentrations were reduced in southeastern China due to the rainy season (Figure S43). When compared to other particulates, ${\mathrm{SO}}_4$ from shipping was found closer to shipping lanes and coastal regions, which could be explained by its higher affinity to ammonium compared to nitrate.

4.4.3. Nitrate (NO₃)

Nitrate particles form from ${\mathrm{NO}}_2$ in a gas-phase oxidation with hydroxyl radicals during the day or via the intermediate ${\mathrm{N}}_2{\mathrm{O}}_5$ in a nightly oxidation by ozone. Nitrate is deposited from the atmosphere as nitric acid or reacts with ammonia to form ammonium nitrate particles.

In Europe, especially the coastal regions bordering the English Channel, northern France, Belgium, The Netherlands, northern Germany and Denmark were affected by high ${\mathrm{NO}}_3$ loads (Figure 12b). In winter and autumn, higher concentrations were shifted towards the northeastern coastline of the European mainland, while in summer, they were shifted towards the southwest. Due to agricultural fertilization, ${\mathrm{NO}}_3$ concentrations were $1.5$ $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ highest during spring (Figure S34).

In China, the Yangtze River, the LYP and NCP were most affected by nitrate aerosols from shipping in concentrations of $0.5$–1 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (Figure 12e). The highest concentrations of 1 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ were modeled during spring (Figure S37).

In both regions nitrate concentrations were found to correlate with ammonium concentrations. Especially in China, high ${\mathrm{NO}}_3$ concentrations were modeled in areas with low ${\mathrm{SO}}_2$ concentrations but abundant ammonium (Figure 12a,d). In general, the formation rate of ${\mathrm{NO}}_3$ particles showed a stronger seasonality than sulfate. A comparison of the nitrate and ${\mathrm{NO}}_2$ ratios suggested that the formation of nitrate aerosols was limited by available ammonium. The ${\mathrm{NO}}_3$ and ${\mathrm{NH}}_4$ concentrations were approximately 5 times higher in China than in Europe, while the ratio for ${\mathrm{NO}}_2$ was only $2.5$. The highest ${\mathrm{NO}}_3$ concentrations were modeled in winter, when low temperatures enabled higher nucleation rates. Due to the SECA in the North and Baltic Seas, ${\mathrm{SO}}_2$ concentrations were smaller than those China, and emissions from regional shipping resulted in higher ${\mathrm{NO}}_3$ concentrations (Figure 12b).

5. Conclusions

In this work, an air quality modelling system was applied to calculate and study pollutant concentrations for northern Europe, including the North and Baltic Seas, and eastern China, including the Yellow and South China Seas for 2015 in a harmonized approach. However, a noticeable difference between these two regions was, generally, the background air pollution, which could be classified as high in China and medium in Northern Europe. The goal of this study was a comparison of the impact of ship emissions between both regions, with a focus on the obvious differences and similarities.

An accurate representation of ship emissions was achieved by using two temporally and spatially high resolved bottom-up ship emission inventories, both created with the MoSES ship emission model using comparable input parameters. Anthropogenic emissions from other sectors were covered by the MEIC and EDGAR emissions inventories for China and CAMS emissions for Europe, which were distributed and generated using the HiMEMO emissions model. The emission data were fed into the chemical transport model CMAQ, as well as the meteorologic forcing, which was calculated with COSMO-CLM. Species concentrations of interest were those for ${\mathrm{NO}}_2$, ${\mathrm{SO}}_2$, ${\mathrm{O}}_3$, PM and the secondary aerosol components ammonia, sulfate and nitrate. The contribution of ship traffic to air pollutant concentrations was determined by using the zero-out method.

The performance of the model was evaluated by a comparison of the modeled concentrations with measurement data from coastal air quality stations. Overall, the model performance proved to be satisfactory. However, ${\mathrm{NO}}_2$ concentrations were often predicted to be to low for Europe, and ozone concentrations were often predicted to be too high in both regions. A comparison between the results obtained with the MEIC inventory and the results obtained with the EDGAR v5 inventory for nonshipping land-based anthropogenic emissions in China pointed to an overestimation by the latter.

The comparison between China and Europe showed that air pollutant concentrations originating from ship emission were generally on a comparable scale, with slightly higher concentrations in China. However, a stronger seasonality was found for China, with the East Asian monsoon and different climate zones as an important factors. Greater similarity was found between temperate northeastern China and Europe, while in the subtropical climate zone, the observed trends were often reversed. Compared to Europe, the overall higher concentrations of background air pollution in China modified the atmospheric chemistry of ship exhausts, especially for PM. This resulted in a smaller relative impact of ship emissions in China.

The relative contribution of ships to ${\mathrm{NO}}_2$ concentrations was higher in Europe, with 30% in moderately affected regions and 50% in highly affected regions, compared to 10% and 30% in China. In addition, ${\mathrm{NO}}_2$ was transported further in the atmosphere during winter in China, which was related to an increased ${\mathrm{NO}}_2$ lifetime due to oxidant-limited conditions.

As a result of the SECA implemented in the North and Baltic Seas in 2015, the amount of ${\mathrm{SO}}_2$ emissions differed significantly between the compared regions.

Tropospheric ozone reductions through ship emissions in large ports were stronger in Europe than in China, by an average of 10–15 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (10–15%). In China’s large ports, the shipping influence on ozone concentrations was mainly neutral due to the lower contribution of ships to total emissions. Along the Chinese coast, ship emissions had a stronger promoting effect on ${\mathrm{O}}_3$ formation, particularly in the south, with 2–10 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$ (3–10%). In southern China, this could be attributed to a higher irradiation but also, an inland transport of ozone, whose formation is promoted by shipping in marine environments, was plausible.

In Europe, ${\mathrm{NO}}_{\mathrm{X}}$ and ${\mathrm{SO}}_2$ from shipping led in combination with high ammonia emissions from agricultural fertilization during spring to peaking secondary PM concentrations (2 $\mathsf{\mu }\mathrm{g}·{\mathrm{m}}^{-3}$, 13%). A similar seasonal peak of secondary PM could not be observed in China, despite high ammonia emissions from its large agricultural sector. Another notable observation was that during winter, no significant ${\mathrm{PM}}_{2.5}$ concentrations from shipping were modeled in the eastern and northeastern parts of China. High aerosol precursor emissions from other sectors could deplete the reaction partners and oxidants necessary for secondary particle formation and inhibit the transformation of aerosol precursors from ship exhausts.

A comparison of the obtained model results with former studies by Aulinger et al. [22] on the North Sea and Karl et al. [23] on the Baltic Sea, both for 2011, showed a similar impact of shipping on ${\mathrm{NO}}_2$ concentrations. The results for ${\mathrm{SO}}_2$ were not comparable due to differences in the sulfur regulation between these years. The impacts of ships on ozone concentrations were modeled slightly higher in this study for both, a promotion of ozone formation and depletion. ${\mathrm{PM}}_{2.5}$ concentrations from ships were slightly lower in this study, which was reasonable due to the stricter fuel sulfur limits in 2015. Former studies on the impact of ship emissions on air quality for the YRD and PRD in 2015, by Feng et al. [35] and Chen et al. [93], respectively, were for most pollutants in agreement with the concentrations found in this study. However, the ozone reducing effect of ship emissions in the PRD was also found to be stronger in the present study.

The results from this study show the importance for a regulation of ship emissions in concert with other emission sectors, particularly in China. Otherwise, potential achievements from reducing ship emissions could be offset by emissions from other sectors. Indicators for such effects were the prolonged ${\mathrm{NO}}_{\mathrm{X}}$ lifetime and the impeded formation of secondary PM from ship emissions found in northern China. Furthermore this was indicated by the fact that ship-related ${\mathrm{PM}}_{2.5}$ concentrations were in a similar range between Europe and China although sulfur-containing ship emissions, as precursors to secondary PM, were drastically reduced in Europe due to the implemented SECA. Since ammonia is an important precursor for ship-related PM, a regulation of ${\mathrm{NH}}_3$ emissions could help in reducing PM concentrations. Such regulations could include a motivation to limit agricultural fertilization to specific seasons or by fertilization recommendations based on the current meteorologic situation to mitigate PM formation.