Impact of a nitrogen emission control area (NECA) on the future air quality and nitrogen deposition to seawater in the Baltic Sea region

Air pollution due to shipping is a serious concern for coastal regions in Europe. Shipping emissions of nitrogen oxides (NOX) to air on the Baltic Sea are of similar magnitude (330 kt y−1) as the combined land-based NOX emissions from Finland and Sweden in all emission sectors. Deposition of nitrogen compounds originating from shipping activities contribute to eutrophication of the Baltic Sea and coastal areas in the Baltic Sea region. For the North Sea and the Baltic Sea a nitrogen emission control area (NECA) will become effective in 2021; in accordance with the International Maritime Organization 5 (IMO) target of reducing NOX emissions from ships. Future scenarios for 2040 were designed to study the effect of enforced and planned regulation of ship emissions and the fuel efficiency development on air quality and nitrogen deposition. The Community Multiscale Air Quality (CMAQ) model was used to simulate the current and future air quality situation. The meteorological fields, the emissions from ship traffic and the emissions from land-based sources were considered at a grid resolution of 4× 4 km2 for the Baltic Sea region in nested CMAQ simulations. Model simulations for the present-day (2012) 10 air quality show that shipping emissions are the major contributor to atmospheric nitrogen dioxide (NO2) concentrations over the Baltic Sea. In the business as usual (BAU) scenario, with the introduction of the NECA, NOX emissions from ship traffic in the Baltic Sea are reduced by about 80 % in 2040. An approximate linear relationship was found between ship emissions of NOX and the simulated levels of annual average NO2 over the Baltic Sea in year 2040, when following different future shipping scenarios. The burden of fine particulate matter (PM2.5) over the Baltic Sea region is predicted to decrease by 35– 15 37 % between 2012 and 2040. The reduction of PM2.5 is larger over sea, where it drops by 50–60 % along the main shipping routes, and smaller over the coastal areas. The introduction of NECA is critical for reducing ship emissions of NOX to levels that are low enough to sustainably dampen ozone (O3) production in the Baltic Sea region. A second important effect of the NECA over the Baltic Sea region is the reduction of secondary formation of particulate nitrate. This lowers the ship-related PM2.5 by 72 % in 2040 compared to present-day, while it is reduced by only 48 % without implementation of the NECA. The 20 effect of a lower fuel efficiency development on the absolute ship contribution of air pollutants is limited. Still, the annual mean ship contributions in 2040 to NO2, sulphur dioxide and PM2.5 and daily maximum O3 is significantly higher if a slower fuel efficiency development is assumed. Nitrogen deposition to the seawater of the Baltic Sea decreases on average by 40–44 % between 2012 and 2040 in the simulations. The effect of the NECA on nitrogen deposition is most significant in the western


Introduction
Air pollution due to shipping is a serious concern for coastal regions in Europe (Viana et al., 2014;Matthias et al., 2010).
Globally, nearly 70 % of the exhaust emitted from ship traffic occurs within a corridor of 400 km along the coastline (Endresen 5 et al., 2003). Since emissions from ships can be transported in the atmosphere over several hundreds of kilometres, they have the potential to diminish the air quality in coastal areas. In addition to the primary emitted particles in the ship exhaust, secondary particles are formed in the atmosphere by oxidation of emitted gaseous precursors -nitrogen oxides (NO X ) and sulphur dioxide (SO 2 ) -during the dispersion of the ship exhaust. Mainly by contributing to the ambient levels of fine particulate matter, PM 2.5 (particles with diameter less than 2.5 µm), emissions from ship traffic are responsible for a large number of premature deaths 10 globally (Corbett et al., 2007). According to Sofiev et al. (2018b) the worldwide use of cleaner marine fuels with a lower content of sulphur will strongly reduce the ship-related premature mortality and morbidity, by 34 % and 54 % respectively. In northern Europe, the health-related external costs from international shipping in the Baltic Sea and North Sea are expected to decrease by 36 % between 2000 and 2020 (Brandt et al., 2013). This reduction is mainly a consequence of the introduction of the sulphur emission control area (SECA) for Baltic Sea (enforced 2005) and North Sea (enforced 2006), which step-wise 15 reduced the sulphur content in ship fuels.
However, air emissions of NO X from ship traffic remained almost constant throughout the last decade, and the impact of NO X will remain a concern for health. Shipping emissions of NO X on the Baltic Sea are of similar magnitude as the combined land-based NO X emissions from Finland and Sweden in all emission sectors . While EU air quality legislation will lead to a decline of land-based emissions of NO X in the future, ship emissions -without more stringent emission 20 control measures on NO X -will rise with the projected annual growth of maritime traffic in the Baltic Sea of about 5 % (Stipa et al., 2007). As a consequence the relative importance of shipping emissions compared to land-based emission sources of NO X is expected to increase. A review of model studies on ship emissions showed that NO X emissions from international shipping on European seas could be equal to land-based emission sources in Europe (EU-27) from 2020 onwards and confirmed that the contribution of the shipping sector to future air pollution in Europe will increase (EEA, 2013). 25 The atmospheric transformation of emitted NO X from shipping is especially relevant for the formation of ozone (Eyring et al., 2010). Shipping emissions are estimated to play an important role on ozone (O 3 ) levels compared to the road transport sector near the coastal zone in Europe (Tagaris et al., 2017). A regional impact study by Huszar et al. (2010) found that the contribution of shipping emissions to surface NO X levels causes an increase of surface O 3 by up to 4-6 ppbv over the eastern Atlantic and western Europe. O 3 can damage vegetation, reduce plant primary productivity and agricultural crop yields 30 (Chuwah et al., 2015) and is also a serious concern for human health (EEA, 2015).
Ship exhaust emissions of NO X are further converted to gaseous nitrous acid (HNO 3 ) through atmospheric oxidation. This conversion of nitrogen dioxide (NO 2 ) to HNO 3 takes place at a rate of approximately 5 % per hour, causing an atmospheric tential emission reductions. A delayed introduction of the NECA by 5 years (in 2021), would cause concentration increases of these pollutants by 10-15 % compared to today . The study by Matthias et al. (2016) assumes an increase in ship number by 1 % p.a., an increase of transported cargo of 2.5 % p.a. and a ship renewal rate of 2.5 % p.a. independent of ship size. The study considered no gains in fuel efficiency of new built ships. Clearly, predicted consequences of the Tier III NO X emission regulation on future shipping emissions depend critically on the projected growth of transported volume, the 5 increase in ship number and the share of new ships in the future fleet. In a similar study, Jonson et al. (2015) investigated the effect of the NECA introduced in 2016 on the air quality in 2030, assuming a moderate increase in ship activity. According to their future scenario, total NO X emissions in the Baltic Sea and the North Sea will almost be unchanged in 2030 compared to 2010, if the NECA is not implemented. However, implementation of the NECA in 2016 will lead to significantly lower NO X emissions from ships in 2030, resulting in slight reductions in the burden on health due to shipping (Jonson et al., 2015). The 10 emission study by Kalli et al. (2013), which calculates the emissions separately for every ship taking into account expected traffic growth and fleet renewal, corroborates the strong decrease of NO X shipping emissions (by 11 % in 2020 and by 79 % in 2040) when the NECA is established in 2016.
The goal of the present study is to investigate the effect of the implementation of the NECA in 2021 on the air quality in the Baltic Sea region and on the total deposition of nitrogen to the Baltic Sea in 2040. In addition to the effect of the NECA 15 regulation, we also look into possible future developments which might diminish the beneficial effect of the NECA, such as failing to achieve increased fuel efficiency of ships.
Several future shipping emission scenarios for the year 2040 were designed. These scenarios were based on the projected development of the economic growth and ship traffic volume in accordance with the study by Kalli et al. (2013). Land-based emission sources are assumed to follow the emission reduction due to current EU legislation. Three cases with respect to future 20 air quality were considered: (1) implementation of the NECA in 2021; (2) no implementation of the NECA; and (3) alternative assumptions for the fuel efficiency of the ship fleet in combination with NECA.
A regional atmospheric CTM system using the Community Multiscale Air Quality (CMAQ) model (Byun and Schere, 2006;Appel et al., 2013), similar to that used in the study by Matthias et al. (2016), was used to simulate the present-day and future air quality conditions in the Baltic Sea region. The advantage of the applied CTM system for the Baltic Sea compared 25 to previous studies in the same region Jonson et al., 2015;Hongisto, 2014) is the higher spatial and temporal resolution of all components driving the chemistry-transport calculations. The meteorological fields, the emissions from ship traffic and the emissions from land-based sources were considered at a grid resolution of 4 × 4 km 2 for the innermost model domain in the nested CMAQ runs. Higher resolution of shipping emissions, which are obtained based on ship positions acquired from 4-minute AIS (Automatic Identification System) records and detailed ship characteristics using the 30 Ship Traffic Emission Assessment Model (STEAM; Johansson et al., 2013; in combination with the higher resolution of the chemistry-transport computation allow for a better resolution of the individual ship's plumes. Moreover, the high resolution meteorology (0.025 • grid) resolves convective precipitation, which is expected to improve the timing and amount of predicted rainfall, crucial for the determination of the nitrogen inputs to the Baltic Sea.
The focus of the present study will be on the computational model results for summer, defined as the average of the period June-August (JJA), when assessing the changes of air quality and deposition between the future scenarios and the present-day situation. In summer, emissions from shipping are highest and the photochemical conversion of the ship exhaust constituents into compounds that are readily scavenged by precipitation is faster than in other seasons. Therefore, ship-originated oxidised nitrogen deposition to the sea is highest during the summer (Hongisto, 2014). In addition, the seasonal variation of air quality 5 indicators and of the accumulated nitrogen deposition to seawater is presented.
A first set of model runs was performed for the situation in year 2012. The present day model results on nitrogen deposition and the air quality situation is analysed. Modelled deposition of nitrogen was evaluated in two steps, first the predicted rainfall amount and frequency is compared to daily precipitation measurements from rain gauge stations in Sweden, and second the wet deposition of oxidised and reduced nitrogen is compared against measurements of the "Cooperative Programme for Monitoring 10 and Evaluation of the Long-range Transmission of Air Pollutants in Europe" (EMEP) programme. Present-day model results on air quality are evaluated with measurements from the regional background stations of the EMEP monitoring network in the Baltic Sea region. A companion paper by Karl et al. (2019) presents a more detailed comparison of the model results for the current air quality situation with land-based observations of air pollutant concentrations in the Baltic Sea region. The contribution of shipping emissions to the modelled concentration of air pollutants was determined from the difference between 15 a reference run that included all emissions and a "Noship" run that excluded emissions from ship traffic (zero-out method).
A second set of model runs was performed to assess the effect of projected emissions from shipping for the year 2040.
Future air quality and nitrogen deposition is analysed, in order to investigate: (1) the effect of establishing the NECA in 2021 compared to a future situation without NECA; and (2) the effect of a lower fuel efficiency increase than expected based on continuation of the current trend. Changes of the ship contribution to regulated air pollutants and to nitrogen deposition over 20 seawater between the present-day simulation and the future scenario simulations are presented. Finally, recommendations with respect to the future regulations and their possible impacts and side-effects are given.
2 Chemistry-transport modelling 2.1 CMAQ model description Regional chemistry/transport model simulations with the Community Multiscale Air Quality (CMAQ) model v5.0.1 (Byun 25 and Ching, 1999;Byun and Schere, 2006;Appel et al., 2013;Appel et al., 2017) were performed to assess the effect of emissions from ship traffic on the present-day and future air quality of the Baltic Sea region. The CMAQ model computes the air concentration and deposition fluxes of atmospheric gases and aerosols as a consequence of emission, transport and chemical transformation. The atmospheric chemistry of reactive species is treated by the Carbon Bond V mechanism (Yarwood et al., 2005), with updated toluene chemistry (Whitten et al., 2010) and chlorine radical chemistry (mechanism cb05tucl; Sarwar The aerosol scheme AERO5 is used for the formation of secondary inorganic aerosol (SIA). Aerosol growth and nucleation is simulated by three lognormal distributed modes, each represented by three moments (Binkowski and Roselle, 2003). The Aitken and accumulation modes represent PM 2.5 and the coarse mode represents particulate matter with diameter >2.5 µm (PM coarse ). The instantaneous gas phase/aerosol equilibrium partitioning of sulphuric acid (H 2 SO 4 ), HNO 3 , hydrochloric acid (HCl) and NH 3 on the fine particle modes is solved with the ISORROPIA v1.7 mechanism (Nenes et al., 1999). Dynamic mass transfer is simulated for the coarse particle mode because large particles often do not reach equilibrium with the gas phase for typical atmospheric time scales (Meng and Seinfeld, 1996). For the coarse mode, semi-volatile inorganic species are allowed to 5 condense and evaporate, while H 2 SO 4 does not evaporate again from the coarse mode. Because of the dynamic mass transfer to coarse particles it is possible to use CMAQ for the simulation of chloride (Cl − ) replacement by NO − 3 in mixed marine/urban air masses  which could be an important aerosol process in the Baltic Sea region.
Sea salt emissions were calculated inline by the parameterization of Gong (2003), as described in Kelly et al. (2010). Sea salt surf zone emissions were deactivated because of considerable overestimations in some coastal regions (Neumann et al.,10 2016b). The formation of secondary organic aerosol (SOA) from isoprene, monoterpenes, sesquiterpenes, benzene, toluene, xylene, and alkanes (Carlton et al., 2010;Pye and Pouliot, 2012) is included. SOA formation pathways include the traditional two product representation, reaction of volatile organic compounds (VOC) to give non-volatile products, oxidative ageing of primary organic aerosol, acid-catalysed enhancement of SOA mass, oligomerization reactions and in-cloud aqueous-phase oxidation. 15 Three types of clouds are modelled in CMAQ: sub-grid convective precipitation clouds, sub-grid non-precipitating clouds and grid-resolved clouds. CMAQ simulates the aqueous phase chemistry in all cloud types. For the two types of sub-grid clouds, the cloud module in CMAQ vertically redistributes pollutants and calculates in-cloud and precipitation scavenging.
Since the meteorological model provides information about the grid-resolved clouds, CMAQ subsequently does not apply further cloud dynamics for this cloud type. Sub-grid clouds are only simulated in CMAQ when the meteorological driver uses 20 a convective cloud parameterization. Hence sub-grid clouds are treated by CMAQ on the coarser outer resolution grids (16-km and 64-km) but not on the 4 × 4 km 2 model domain because the convective clouds are resolved for the fine grid resolution by the meteorological model. Wet deposition of gases and particles is computed by the resolved cloud model of CMAQ which estimates how much certain vertical model layers contributed to the precipitation. The precipitation flux for each model layer is computed as a function 25 of the non-convective precipitation rate, the sum of hydrometeors (rain, snow, and graupel) and the layer thickness (see Foley et al. (2010) for details).
Dry deposition is determined as the product of the atmospheric concentration and the deposition velocity. The dry deposition velocity is modelled in CMAQ using the resistance analogy, where resistances are defined along pathways from the atmosphere to the surface which act in parallel or in series. Details on the deposition pathways in CMAQ can be found in Pleim and Ran 30 (2011). The deposition velocity for particles is calculated based on the aerosol size distribution, as well as meteorological and land-use information. For large particles, the dry deposition transfer is by turbulent air motion and by direct gravitational sedimentation. The dry deposition algorithm for particles includes an impaction term in the coarse mode and the accumulation mode.
In the resistance method it is assumed that the surface concentration of the chemical species is zero. However, NH 3 can be both emitted from and deposited to surfaces depending on its atmospheric concentration. This bi-directional nature of the airsurface exchange can modify the atmospheric transport and environmental impact of ammonia. Bi-directional fluxes of NH 3 over marine surfaces have been documented in a review by Hertel et al. (2006). In fact, inclusion of the bi directional air-water exchange in a CTM resulted in lower overall dry deposition of NH 3 to coastal waters (Sorensen et al., 2003). However, until 5 now, the parameterization of the bi-directional flux has not been evaluated to a large extent for marine waters. Although the bi-directional flux of NH 3 is implemented in CMAQ v5.0.1, the option was not used in this study. Because we are mainly interested in the differences of total nitrogen deposition due to changes in emission alone, the outcome of this study will be less affected by the sensitivity of the modelled nitrogen deposition to bi-directional fluxes of ammonia.
2.2 Setup of the model 10 Nested simulations with CMAQ were performed on a horizontal resolution of 4 × 4 km 2 to simulate the current and future air quality situation for the entire Baltic Sea region. The model was set up on a 64 × 64 km 2 grid for entire Europe, subsequently on an intermediate nested 16 × 16 km 2 grid for Northern Europe, and finally on two nested 4 × 4 km 2 grids, one for the southern Baltic Sea (Baltic major) and one for the northern Baltic Sea (including Bothnian Bay and Gulf of Finland). The nesting is visualized in Fig. 1a and the geographic details of the high resolution domain is shown in Fig. 1b. The vertical dimension of 15 the model extends up to 100 hPa in a sigma hybrid pressure coordinate system with 30 layers. Twenty of these layers are below approximately 2 km; the lowest layer extends to ca. 36 m above ground. A spin-up period of one month (December 2011) was used for the initialization of the model runs, sufficiently long to prevent that initial conditions have an effect on the simulated atmospheric concentrations of the investigated period (year 2012). 20 The meteorological fields that drive the CTM were simulated with the COSMO-CLM, version 5.0, for the year 2012 (Geyer, 2014) using the ERA Interim reanalysis and spectral nudging technique to force the model. COSMO itself is the operational weather forecast model applied and further developed by a consortium of national weather services whereas COSMO-CLM stands for the climate mode used and developed by the limited area modelling community (clm-community; Rockel et al., 2008). 25 The meteorological runs were performed first on a 0.11 × 0.11 degrees rotated lat-lon grid using 40 vertical layers up to 22 km for entire Europe. The output was used as forcing of a high-resolution nested meteorology run on a 0.025 × 0.025 degrees grid; 50 vertical levels were used for this simulation for the Baltic Sea region. The convection permitting configuration is used on the high-resolution grid, e.g. only shallow convection is based on Tiedtke scheme, resolving convective precipitation clouds.

Meteorological fields
The meteorological fields were processed afterwards using a modified version of CMAQ's Meteorology Chemistry Interface Europe with 16 × 16 km 2 resolution (CD16, green) and the high-resolution grids of 4 × 4 km 2 for southern Baltic Sea (CD04a, dark red) and northern Baltic Sea (CD04b, dark blue). (b) Exemplary structure of spatial maps spanning from latitude 53.30 • N (south) to 65.80 • N (north) and longitude 9.85 • E (west) to 30.95 • E (east). Green shaded area is the high-resolution area which shows output from regional model runs with a grid resolution of 4 × 4 km 2 . Dark red outline marks the extent of the southern part of the Baltic Sea region and dark blue outline marks the extent of the northern part of the Baltic Sea region, for which model output from two high-resolution nests were used. For the overlap area, the arithmetic mean of results from both nests was used. In the post-processing of model results, the native Lambert conformal projection of CMAQ output was transformed to a regular lat-lon grid, therefore the two outlined areas do not fill complete rectangles. The entire domain shown in (b) was interpolated to a uniform resolution of 0.05 • in the post-processing. White areas of the map are covered by the output from the model nest with 16 × 16 km 2 resolution. and total precipitation (±25 mm) were closely aligned to the decadal average of the 2004-2014 period. The meteorological year 2012 was also used in CTM calculations of the future air quality situation to avoid complication of the interpretation of changes between present-day and the future. Hence, future changes of the air quality are solely due to changed land-based and shipping emissions.

5
The initial conditions for the simulation and the lateral boundary conditions for the 64 × 64 km 2 outer European domain (CD64) are taken from APTA global reanalysis (Sofiev et al., 2018a) and were provided by the Finnish Meteorological Institute (FMI).
The global boundary conditions results have been interpolated in time and space to provide hourly boundary conditions for the outer domain. Boundary conditions for the nested intermediate grid and the two inner grids were calculated on hourly basis from the output of the next-outer grid. For the model simulations with no shipping emissions, the full model chain was run calculations based on real world stack information were performed (Bieser et al., 2011b).
The annual total emissions were temporally and spatially redistributed individually for each emission sector and grid cell.
Emissions of residential heating were redistributed using the heating demand calculated from daily average temperatures (Aulinger et al., 2011). Emissions from agricultural activity and animal husbandry were disaggregated according to a fertilizer and plant growth model and meteorological parameters (Backes et al., 2016a). Finally, biogenic emissions were calculated off- 15 line with the biogenic Emission Inventory System BEIS version 3.4 (Schwede et al., 2005;Vukovich and Pierce, 2002). The SMOKE-EU emission datasets were calculated on a 5 × 5 km 2 grid for the whole of Europe and were subsequently interpolated to the respective CMAQ model grids.
3 Shipping emissions and scenario description

Ship emission inventory for the Baltic Sea and North Sea
Shipping emissions for the Baltic Sea and North Sea with high spatial and temporal resolution for this study were obtained 5 from STEAM Johansson et al., 2013;). STEAM combines the AIS-based information and the detailed technical knowledge of the world fleet with principles of naval architecture. This input information is used to predict the resistance of vessels in water and the instantaneous engine power of the main and auxiliary engines on a minute byminute basis, for each vessel that has sent AIS messages. The model predicts as output both the instantaneous fuel consumption and the emissions of selected pollutants. The dynamic modelling of shipping emissions also includes, e.g., the emission control 10 areas and regulations, emission abatement equipment on-board the ships as well as fuel sulphur content modelling separately for main and auxiliary engines (Johansson et al., 2017;Jalkanen et al., 2012). The shipping emission inventory consist of hourly updated 2 × 2 km 2 gridded data for NO X , SO X , CO, and particulate matter, which is further divided into Elementary Carbon (EC), Organic Carbon (OC), sulphate (SO 4 ) and mineral ash. For North Sea, ship emissions from 2011 were adopted for 2012; total ship emissions of NO X were almost unchanged between the two years.
For Baltic Sea ship emissions are from 2012 and were provided for two vertical layers (below 36 m, from 36-1000 m). In 20 CMAQ, SO X was attributed completely to SO 2 and a NO:NO 2 ratio of 95:5 was applied. Ship emissions below 36 m were attributed to the lowest vertical model layer. Ship emissions above 36 m were attributed to the second lowest layer; which appears to be justified based on findings with ship plume simulations (Chosson et al., 2008) showing that plume dispersion in the convective boundary layer (BL) is insensitive to the initial buoyancy flux.
3.2 Future scenarios for shipping emissions 25 Shipping in the Baltic Sea in the future is modelled in a number of scenarios taking into account the development of traffic and transport work, fleet development for different ship types (number and size), changes in fuel mixture and regulations influencing emissions and fuel consumption. Due to the long lifetime of ships it will take about 30 years after the NECA entry date until the entire ship fleet will be renewed  and follows the Tier III emission regulation for NO X . It was decided to perform the future regional CTM simulations for 2040 in order to see the full effect of the NECA. The baseline scenario for the future situation in 2040 is the so-called "business as usual" (BAU) scenario that is constructed as a reference scenario ("BAU 2040") forall other future scenarios. It accounts for current trends of economic growth and development of shipping and takes into account already decided regulations. Regarding regulations effecting emissions to air the following are the most important ones in BAU: 1. Sulphur regulation: The Baltic and North Seas are Sulphur Emission Control Areas (SECA) where the maximum allowed sulphur (S) content in marine fuel has been gradually lowered reaching 0.1 % S from 2015. For sea areas outside SECA the maximum fuel sulphur content will be 0.5 % S from 2020. These regulations directly influence the emissions of SO X and have a strong impact on the particulate matter emissions. will influence engine emissions in a similar way as the regulations on sulphur and NO X . 15 The BAU scenario assumes a share of ships driven by liquefied natural gas (LNG) of about 10 % in the ship fleet in 2040.
This is modelled as a fraction of new ships introduced each year that will use LNG since retrofitting of existing ships from fuel oil to LNG is assumed less likely due to high costs. Since LNG is used as a means to comply with the sulphur regulations ship types that operate mainly within SECAs are modelled as more likely to use LNG. The fuel efficiency for new ships in BAU is assumed to improve further than what is required from the EEDI regulation, following recent trends and assumption from 20 Kalli et al. (2013), assuming that further technical improvements and more efficient operation take place. The traffic volumes are expected to continue to grow with about 1 % p.a. on average (it varies with ship type); the current trend of using larger vessels is expected to continue as well.

Future scenario "NoNECA 2040"
The other two future scenarios, "NoNECA 2040" and "EEDI 2040", are deviations from the development given by the BAU 25 scenario. In the NoNECA scenario, the nitrogen emission control area is assumed not to be implemented, i.e. all new ships up to 2040 are assumed to follow the Tier II NO X standard. The difference to the BAU scenario is then that new ships from 2021 follows the Tier II standard rather than Tier III. The same introduction of LNG as in BAU is assumed since the use of LNG is mainly motivated by the SECA regulation. From the difference between BAU and NoNECA the effect on emissions of implanting the NECA can be deduced.

Future scenario "EEDI 2040"
In the EEDI scenario, improvements in fuel efficiency follow strictly the requirements of the EEDI regulation. Annual efficiency increases of 0.65 % to 1.04 %, depending on ship type, are assumed in the EEDI scenario while the corresponding values in the BAU scenario are 1.3 % to 2.25 %. From the difference between BAU and EEDI the effect of a lower fuel efficiency increase than expected based on continuation of the current trend can be deduced.
5 Table 2 provides emission scaling factors used in the three scenarios for future shipping emissions.

Future land based emissions
The three scenarios studied here ( Atmospheric deposition of nitrogen to the Baltic Sea seawater is mainly controlled by wet deposition . Since wet deposition of N-containing compounds is determined as the product of the concentration of N-containing compounds dis-5 solved in rainwater and the amount of rainfall, the accurate prediction of the amount, frequency and spatial distribution of precipitation is important. The precipitation amount and frequency from COSMO-CLM output is compared to daily precipitation measurements from rain gauge stations operated by the Swedish Meteorological and Hydrological Institute (SMHI).
The rain gauge network includes 1804 precipitation stations in Sweden which were recording daily precipitation sums during 2012. The precipitation data is available from the SMHI opendata portal (http://opendata-catalog.smhi.se/explore/). Details on 10 the methodology for comparing modelled precipitation data with these observations are given in Sect. S1 of the Supplementary Materials.
The model-observation comparison was done for the three different configurations of COSMO-CLM: 0.11 degree grid resolution with Tiedtke scheme for convection ("011"), 0.025 degree grid resolution with Tiedtke scheme for convection ("0025_Tiedtke"), and 0.025 degree grid resolution with convection-permitting configuration ("0025_convper").

5
Finer grid resolution ("0025_Tiedtke" versus "011") has a tendency to increase the rainfall over land in summer. In particular, more orographic rainfall occurs in Norway for "0025_Tiedtke" compared to "011" (Fig. S1). The finer resolution improves the agreement with measured rainfall in Svealand in August, but causes too high simulated precipitation in Norrland. The convection-permitting configuration ("0025_convper") yields only small changes compared to "0025_Tiedtke". Most notable differences are the higher precipitation amounts over the Danish islands in June and more convective rainfall over southern 10 Norway in July and August. It has been suggested that the observed inland precipitation intensity in the warm season in the southern part of Sweden is associated with convective rainfall forced by solar heating (Jeong et al., 2011). The slightly increased inland precipitation in June in "0025_convper" compared to "0025_Tiedtke" is in line with this suggestion.
However, COSMO-CLM predicts too low precipitation amounts in southern Sweden in June in all three configurations. Compared to the two other configurations, "0025_convper" has the highest percentage fraction of days with zero difference between where COSMO-CLM has a positive bias (Table S2). A possible reason for the dry bias in summer could be that south Swe-20 den receives too little precipitation due to its location in the lee of the Norwegian mountains, where humidity is lost through excessive orographic rainfall in the simulation.

Comparison of the modelled wet deposition of nitrogen with observations
Wet deposition of oxidised and reduced nitrogen was evaluated with measurements of regional background stations in the Baltic Sea region for the period of 1 March to 30 November 2012. The winter months were excluded from the analysis to 25 avoid possible artefacts associated with the collection of snow. Modelled wet deposition of nitrate, NO − 3 (WNO 3 ), representing oxidised nitrogen and modelled wet deposition of ammonium, NH + 4 (WNH 4 ), representing reduced nitrogen, were compared to data from the EMEP monitoring programme (Tørseth et al., 2012;EMEP, 2014) at the stations displayed on the map in Fig. 3a.
Observation data was obtained from the EBAS database (http://ebas.nilu.no/). Details on the methodology for comparing modelled wet deposition of nitrogen with these observations are given in Sect. S2. The comparison of the daily sum of wet 30 deposition was done in terms of mean values (µ Mod and µ Obs ), the Spearman's correlation coefficient (R Spr ) and the normalized mean bias (NMB). Only days with predicted and observed rain events in common were included in the comparison. Several stations in the Baltic Sea region had only few measurements during the period. Stations with less than seven model-observation pairs were excluded from the statistical analysis. CMAQ model data from the intermediate grid (CD16) and from the highresolution grid (CD04) are evaluated separately. Fig. 3b-g show the time series modelled and observed daily sums of WNO 3 at selected stations (all other stations are shown in Fig. S4). The 4-km resolution output gave higher WNO 3 than the coarser CD16 output in the southern part of the Baltic Sea region (e.g. stations Zingst, Preila and Keldsnor). For the more northern stations, simulated time series of WNO 3 5 from the two model grids are similar. The correlation between modelled and observed data improves for several stations when going from CD16 to CD04, supporting the use of finer resolution for chemistry and transport computations in combination with high-resolution precipitation modelling. WNO 3 is underestimated at all stations included in the statistical analysis (Table S3), most severely at the Finnish stations and at Zingst. WNH 4 is underestimated at all stations included in the statistical analysis (Table S4; corresponding time series are plotted in 10 Fig. S5). The underestimation is highest for Zingst and the Finnish stations, as for WNO 3 . The joint underestimation of WNO 3 and WNH 4 especially in the northern part of the Baltic Sea region could indicate missing formation of particulate ammonium nitrate or too slow conversion of NO X to HNO 3 in the model. The long-range transport of particulate ammonium to the remote parts of the Baltic Sea region is further limited by the availability of particulate nitrate and sulphate (Ferm and Hellsten, 2012).

Plots in
To account for the fact that the days with predicted rain often do not correspond to days with observed rain, seasonal averages 15 (spring, summer and autumn) were calculated for WNO 3 (Table S5) and WNH 4 (Table S6) independently for CD04 model data and observation data. The joint underestimation of WNO 3 and WNH 4 at Zingst and the Finnish stations is confirmed in this analysis.
The agricultural sector, including animal husbandry, is an important source of r educed nitrogen emissions to the atmosphere (e.g., Bouwman et al., 1997). NH 3 emissions from animal housing and application of manure on fields are highly relevant and 20 can influence the formation of ammonium nitrate particles (Backes et al., 2016b). Formation of ammonium sulphate is much less sensitive to agricultural NH 3 emissions because ambient background concentrations of NH 3 in the model simulations are high enough to saturate the reaction forming sulphate particles (Backes et al., 2016b). Too low emissions of gaseous NH 3 from agriculture in northern Germany might also explain the missing WNH 4 at Zingst. Annual emission totals of NH 3 reported by

Nitrogen deposition to the Baltic Sea region
Deposition of nitrogen includes particulate ammonium and nitrate as well as gaseous NO, NO 2 , NH 3 , nitrate radical (NO 3 ), HNO 3 , dinitrogen pentoxide (N 2 O 5 ), peroxy nitric acid (HNO 4 ) and peroxy acetyl nitrate (PAN). Figure 4a shows the spatial distribution of the annual total (wet and dry) nitrogen deposition in 2012 from the CMAQ simulation. A strong gradient from southwest to northeast is found for the annual total nitrogen deposition, both over land and over sea. Highest nitrogen deposition In coastal regions, nitrogen deposition is markedly higher compared to further inland. Sea-salt particles can considerably increase nitrogen deposition in coastal regions, although this effect is relatively small in the Baltic Sea region and only pronounced along the coast of Denmark (Neumann et al., 2016a). Reaction of HNO 3 with coarse mode sea-salt particles, when marine aerosol mixes with the polluted air from the continent, leads to a shift of fine mode nitrate to the coarse mode, through the formation of sodium nitrate (Brimblecombe and Clegg, 1988;Zhuang et al., 1999) which is essentially non-volatile in atmospheric conditions. Since coarse mode particles are prone to deposition through gravitational settling, the nitrate formation reaction on sea-salt particles may lead to enhanced deposition of nitrogen in the coastal zone (Spokes et al., 2000;Neumann et al., 2016a).

5
The injection of reactive nitrogen through shipping activities contributes to increased input of nitrogen to the Baltic Sea.
The annual nitrogen deposition related to ship emissions (ship-related deposition) is on average 52 mg(N) m −2 over the Baltic Sea (Fig. 4b). The absolute contribution of shipping emissions (seasonal cycle shown in Fig. S8) is highest during summer; amounting to 20 mg(N) m −2 (JJA) in the Baltic Sea on average.

Present-day air quality
CMAQ model results for surface air concentrations of O 3 , NO 2 , SO 2 and PM 2.5 from the 4-km resolution grid were evaluated against measurements at regional background stations of the EMEP monitoring programme available from the EBAS database.

Seasonality of ozone and comparison with measurements
Ozone is generated in the troposphere involving two classes of precursor compounds, VOC and NO X , in photochemical reaction cycles, initiated by the reaction of the OH radical with organic molecules. The precursors of O 3 have anthropogenic and natural (or biogenic) sources, both are considered in the CTM simulation. At the continental scale, the formation of O 3 is sustained by the oxidation of methane (CH 4 ) and CO. In the present-day CMAQ simulation, highest seasonal averages of the daily maximum O 3 concentration were found in spring (MAM), with levels up to 50 ppbv in the southern part of the Baltic Sea region (Fig. S9), which are a consequence of the inflow of ozone-rich background air masses from the Atlantic. Photochemical production in summer leads to elevated ozone concentrations over the southern Baltic Sea (range 36-44 ppbv). In autumn and winter daily  (Table S7) when the entire year is considered. In summer, ozone is slightly underestimated at the stations in the southern part of the Baltic Sea region.

Seasonality of nitrogen dioxide and comparison with measurements
The main sources of nitrogen oxides are traffic and combustion processes. Emissions of NO X and the derived oxidation prod-10 ucts strongly influence concentrations of ozone and particulate matter (Seinfeld and Pandis, 2005), the latter directly through formation of nitrate aerosols and indirectly by influencing the oxidation of secondary aerosol precursors.
In spring and summer, average NO 2 concentrations in proximity of the main shipping routes several times exceed the concentrations in the regional background (Fig. S10). In autumn and winter the spatial distribution of modelled seasonal averages show a gradient from south to north. High values are predicted in northern Germany, Poland and over the Danish Straits 15 (range: 3.5-7.5 ppbv) with hotspots in the large cities (> 9 ppbv). The wider spread of elevated NO 2 concentrations in winter compared to summer is in accordance with a longer lifetime of NO X in winter (up to one day) compared to summer (a few hours) (Schaub et al., 2007). The evaluation of modelled NO 2 based on daily concentrations for the entire year and for summer (Table S8) indicates a better performance of CMAQ over the entire year than over summer alone.
In contrast to a previous study with the CMAQ model in the North Sea region by Aulinger et al. (2016) and other multi-model 20 air quality studies in Europe (e.g., Giordano et al., 2015), the simulations for the Baltic Sea region did not show substantial underestimation of observed NO 2 daily means. The improved performance for NO 2 compared to the previous study by Aulinger et al. (2016) is partly attributed to the high spatial resolution, as NO X emissions are injected into a smaller grid box volume and consequently less diluted initially.

25
The main atmospheric sources of SO 2 are fossil fuel combustion and metal producing industries. The atmospheric lifetime of SO 2 based on the reaction with the OH radical is about one week (Seinfeld and Pandis, 2005). SO 2 is removed efficiently by dry deposition; the lifetime towards dry deposition is typically about one day. Overall, the average lifetime of SO 2 in the troposphere is a few days. SO 2 is converted to sulphate aerosols either via gas-phase oxidation to H 2 SO 4 and subsequent nucleation or condensation or by uptake into cloud droplets followed by aqueous phase oxidation. SO 2 is a major air pollutant 30 and linked to air quality and human health issues.
SO 2 shows higher concentrations in autumn and winter than in spring and summer (Fig. S11). The main reason is the stable boundary layer connected with stagnant air and frequent inversions during the colder season which causes emissions of SO 2 to accumulate in the surface layer. Residential heating emissions and power plant emissions for district heating strongly contribute to the higher SO 2 concentrations in winter as compared to summer. Highest SO 2 concentrations in autumn and winter are simulated over Poland, where levels in the cities exceed 3 ppbv. In spring and summer elevated SO 2 levels over the Baltic Sea (0.9-1.8 ppbv), confined to the main shipping routes, are a sign of the influence from shipping activities. Another factor leading to lower concentrations in summer is the faster oxidation of SO 2 by OH compared to other seasons.

5
Observed SO 2 concentrations are generally overestimated (Table S9), indicating that the oxidation of SO 2 in the background air is not efficient enough in the simulation. The overestimation of both SO 2 and NO 2 by the model corroborates the hypothesis of too slow conversion of the primary gaseous precursors given in Sect. 4.1.2 to explain the underestimated nitrogen deposition, but it is also possible that the anthropogenic emissions of these pollutants are too high in the model. Ambient PM 2.5 comprises primary emitted and secondary PM that formed in the atmosphere. Primary PM includes OC and EC particles from anthropogenic sources such as traffic and industrial activities, as well as wind-blown soil dust and sea-salt particles from natural sources. Secondary PM includes secondary inorganic and organic particles from the homogeneous and heterogeneous chemical transformation of primary gaseous precursors such as NO X , SO 2 , NH 3 and NMVOC in the atmosphere.

15
PM between 0.1 µm and 1 µm in diameter can remain in the atmosphere for days or weeks and thus be subject to long-range transport. PM 2.5 is known to have adverse health effects; short-term exposure to PM 2.5 is associated with respiratory and cardiovascular diseases (e.g., Pope and Dockery, 2006), while long-term exposure to PM 2.5 is associated with an increase in the long-term risk of cardiopulmonary mortality (Beelen et al., 2008).
Modelled PM 2.5 is highest in winter, exceeding 6 µg m −3 in most parts of the Baltic Sea region, which is attributable to the For the entire year CMAQ performs quite well in the prediction of daily mean PM 2.5 , but in the summer period, PM 2.5 is underestimated (Table S10). This is partly due to the underestimation of secondary organic aerosols by the CMAQ model.
Although the capability of CMAQ to predict SOA has been improved compared to earlier versions of the model, the predicted

Summer mean ship contribution of air pollutants
The influence of shipping emissions on the present-day air quality was evaluated for the summer months. The results for the 5 impact of shipping emissions were calculated as difference between the reference run and the run with no ship emissions (in the North and Baltic seas) in 2012. Results for the absolute and relative ship contributions in summer (as JJA average) are shown in Fig. 5 for the daily maximum O 3 , NO 2 , SO 2 as well as PM 2.5 , and discussed in the following.
In the proximity of the main shipping routes, ozone concentrations are reduced by 10-20 % on spatial average in summer compared to a situation with no shipping emissions. This reduction is due to local scale titration of O 3 by NO emitted in the

20
Ships emit NO X mainly in the form of nitrogen oxide (NO). When ozone entrains into the ship's exhaust plume, NO is however quickly converted to NO 2 , so atmospheric NO X will be mainly in the form of NO 2 .
Over the Baltic Sea, shipping emissions have a high contribution to atmospheric SO 2 concentrations in the present-day situation. The summer mean ship contribution to SO 2 is 2.5 ppbv (about 80 %) or more in a wide area around the main shipping routes of the Baltic Sea (Fig. 5c). The EU has implemented a sulphur emission control area (SECA) for the North and Baltic 25 seas, which means that in the present-day situation for the model (year 2012), fuels burned on ships in these areas must not contain more than 1.0 % S. After 1 January 2015, not more than 0.1 % S in the fuel is allowed in the SECA, which drastically decreases SO 2 concentrations along the shipping routes (Kattner et al., 2015).
The ship contribution to summer mean PM 2.5 shows a gradient from south to north with highest concentrations over the Belt Sea/Kattegat and over the sea south of Sweden with maximum values up to 1.4 µg m −3 (Fig. 5d)  particles in the ship exhaust plume during its transport away from the shipping route. The production of secondary particles via the oxidation of NO 2 and SO 2 emitted from ships happens over a longer time scale, during which the plume is advected.
In addition, the aerosol formation rates critically depend on ambient temperature, humidity, solar radiation and the level of atmospheric oxidants (OH and NO 3 radicals) and reaction partners such as NH 3 . 5 Future scenario model results

15
Over most parts of the Baltic Sea region, the summer mean of daily maximum O 3 in "BAU 2040" decreases by 10-25 % compared to 2012, as consequence of the NECA and reduced land-based emissions of NO X (Fig. 6a). The future change of ozone is similar in "EEDI 2040", implying, that the effect of increased fuel efficiency is less pronounced and that the NO X reduction through establishing the NECA has a much greater influence on future ozone levels in the Baltic Sea region. In the NoNECA scenario, daily maximum O 3 over land will decrease less than in the BAU scenario, but still an average ozone 20 reduction by 15 % in 2040 is predicted for large parts of Sweden and the Baltic Sea, compared to present day.
In the "BAU 2040" scenario, summer mean NO 2 concentrations are drastically reduced, by ∼80 % over most parts of the Baltic Sea and by up to ca. 90 % in the northern Baltic Proper, compared to 2012 (Fig. 6b). This appears to be a result of the combined emission reductions through the NECA and the regulation of land-based emissions (Sect. 3.3), leading to a shift in the overall atmospheric photochemical regime due to the lower abundance of NO X in the future. Strong reduction is also seen 25 in "EEDI 2040", where NO 2 levels over the Baltic Sea decrease by ∼80 %, compared to 2012. "NoNECA 2040" results in a reduction of NO 2 by ∼50 % over the entire Baltic Sea.
"BAU 2040" adopts the agreed SO X emission reduction measures; i.e. the SECA limit of 0.1 % S in fuel from 2015 onwards and the global limit of 0.5 % S in fuel from 2020 onwards. The other two future scenarios also implement the two sulphur regulations. In 2040, summer mean SO 2 levels drop by 80-90 % over the entire Baltic Sea compared to present day.   O 3 concentrations is highest over the Gotland Basin (range: 5-6 ppbv), while it is smaller for all over parts of the Baltic Sea region, not exceeding 4.5 ppbv. Overall, the model simulations predict that shipping emissions will still influence ozone levels over the Baltic Sea and in the coastal areas in 2040, with relative contributions in the range of 10-20 % to daily maximum O 3 .
The absolute ship contribution to summer mean NO 2 concentrations in 2040 drop substantially compared to 2012. The shiprelated NO 2 concentration decreases from ca. 3 ppbv in the present-day situation to 0.5-1.5 ppbv in the BAU scenario, along 5 the main shipping routes. Even with the NECA established, emissions from ship traffic remain the dominant contributor to atmospheric NO 2 over the Baltic Sea in 2040.
The absolute ship contribution to SO 2 concentrations in summer 2040 is less than 0.1 ppbv. However, the ship influence on ambient SO 2 concentrations has not completely vanished in 2040. Along the main shipping routes throughout the Baltic Sea, the relative contribution remains high. 10 The absolute ship contribution to PM 2.5 in summer 2040 is predicted to be ≤ 0.2 µg m −3 over most parts of the Baltic Sea region, with higher values over the Belt/Kattegat (0.4 µg m −3 ). The ship influence substantially weakens compared to the present-day situation: the relative contribution peaks along the shipping routes (15-25 %) and is below 10 % over land.  The ship contribution to NO 2 decreases by 80-85 % over the Baltic Sea, slightly more than linear with the reduced NO X emissions from shipping. The decrease is smaller (∼77 %) in some port cities like Gdansk and St. Petersburg and in areas with high density of ship traffic. The reduced NO X emission from ships causes an increase of the ratio of [NO 2 ] to [NO] (short: NO 2to-NO ratio) in the ship plumes. Although the NO 2 -to-NO ratio at the ship stack is the same (equal to 5:95), it becomes higher, as NO 2 from the background air entrains into the plume, than in the present-day situation. According to the photostationary 25 state relation, the increased ratio causes a higher steady-state O 3 concentration in the ship plume. With the local increase of O 3 , the reaction of NO with the hydroperoxyl (HO 2 ) radical giving NO 2 starts to compete with the titration reaction (reaction of NO with O 3 ). In the reaction of NO with HO 2 an additional ozone molecule is produced, as the resulting NO 2 molecule photolyses, amplifying the ozone production in the plume. Hence the smaller decrease of the NO 2 ship contribution is due a change of the photochemistry regime in the ship plumes accompanied with a higher conversion of NO to NO 2 .

30
For the ship contribution to SO 2 , a uniform decline by around 90 % is seen for the entire Baltic Sea, in accordance with a linear decrease following the reduction of SO X emissions from shipping by 91.2 % between 2012 and 2040 in "BAU 2040".
Note that ship emissions of SO X were attributed completely to SO 2 . As for the NO 2 ship contribution, the decrease is slightly higher than expected due to the reduction of ship emissions. Due to the drastic decrease of nitrogen oxides, the atmospheric oxidation capacity increases in the future scenario simulation leading to more efficient oxidation of pollutants and higher availability of photo-oxidants (OH and HO 2 radicals). Hence, the removal rates of SO 2 and NO 2 by reaction with photooxidants and the rate of SO 2 oxidation in clouds are slightly increased in 2040 compared to 2012.
The ship contributed summer mean PM 2.5 between 2012 and 2040 ("BAU 2040") reduces by 75-90 %, with largest reductions over the southern part of the Baltic Sea and in the coastal regions. This is more than can be explained by the reduction 5 of primary PM emissions (by 65 %) from shipping. Thus a substantial fraction of the changed ship contribution is caused by changes of the secondary aerosol production. The future ship contribution to PM 2.5 is affected by reduced SO X emissions from ships, as a result of the regulations for lower sulphur fuel content and by reduced NO X emissions due to the NECA.
Together, the regulations lead to a decline of the atmospheric formation of sulphate and nitrate particles related to shipping.
In the southern part of the Baltic Sea region, especially over Denmark and northern Germany, the ship-related formation of sec-10 ondary aerosol is also affected by the lower NH 3 emissions from agriculture. Decreasing atmospheric ammonia concentrations reduces the formation of ammonium nitrate particles, since their formation is limited by the availability of NH 3 .
For the other two future scenarios, "NoNECA 2040" and "EEDI 2040", changes of the ship contributed pollutant concentrations compared to present day are smaller than in "BAU 2040". In the scenario without implementation of NECA, "NoNECA 2040", the ship contribution to NO 2 in 2040 decreases by 50-60 % over the Baltic Sea (Fig. S13). The ship contribution to ozone increases widely by more than 10 % compared to present-day, indicating enhanced ozone production due to shipping activities in 2040, mainly over sea and the coastal areas of Sweden, Denmark and Poland. The EEDI scenario, with lower 5 fuel efficiency, results in a significantly smaller reduction of ship contributed PM 2.5 than the BAU scenario. Still, the ship contributed summer mean PM 2.5 between 2012 and 2040 reduces by 65-80 % over the impacted areas (Fig. S14).

Future air quality: effect of the NECA
The difference in the two future scenarios "BAU 2040" and "NoNECA 2040" is the higher emission reduction of NO X from shipping in the BAU scenario through establishment of the NECA. Figure 9 illustrates the effect of introducing the NECA 10 in 2021 on major air quality components compared to a future situation without NECA, determined based on the difference between modelled concentrations in the "BAU 2040" and "NoNECA 2040" scenarios. Land-based emissions are the same in both scenarios, therefore changes are solely due to different ship emissions in the two future scenarios.
The result of the NECA in 2040 is a reduction of NO X emissions from shipping by 59 % on average, corresponding to the difference between a Tier III dominated ship fleet with the NECA and Tier II dominated ship fleet without the NECA. The 15 reduction of NO X emissions from shipping primarily translates into a ∼60 % decrease of NO 2 summer mean concentrations within a wide corridor of the ship routes. In addition, the population in coastal areas in northern Germany, Denmark and western Sweden will be less exposed to NO 2 in 2040 due to the introduction of the NECA. Due to the lower atmospheric NO X levels, less ozone is formed, and daily maximum O 3 concentration over the Baltic Sea in summer 2040 is on average 6 % lower than without the NECA. In the areas close to the main shipping routes, ozone is almost unchanged despite the sharp reduction 20 of NO X emissions, probably due to compensating effects between changed titration losses and changed photochemical ozone production. As expected, levels of atmospheric SO 2 are largely unaffected by the NECA (< ±2 %).
A secondary effect of the NECA is a reduction of the formation of particulate nitrate. Due to the non-linearity of the atmospheric particle mass formation, i.e. photochemistry and gas-to-particle conversion depend on precursor concentrations and existing particulate matter in a non-linear fashion, the impact of reducing gaseous precursors does not result in a linear 25 reduction of future PM 2.5 levels. Fig. 9d shows the change of summer mean PM 2.5 concentration pattern due to the NECA. Note that primary emissions of PM 2.5 are the same in BAU and NoNECA, thus changes are solely attributed to modified particulate nitrate concentrations. Largest decrease of PM 2.5 , by up to 8 %, occurs over the Danish islands, where the abundance of ammonium nitrate is highest.

Future air quality: effect of lower fuel efficiency 30
The BAU scenario assumes an improvement of the marine fuel efficiency beyond that required by the EEDI regulation for new ships. With the difference between the "EEDI 2040" and "BAU 2040" scenarios (land-based emissions are the same in both scenarios), the effect of a slower rate of fuel efficiency improvement compared to the projections in the BAU scenario on the air quality in 2040 is determined. The lower fuel efficiency affects the ship engine emissions and leads to NO X , SO 2 and PM 2.5 emissions from ships that are on average 37.9 %, 36.8 % and 39.6 % higher in 2040, respectively, compared to the BAU scenario. As a consequence of the lower fuel efficiency, modelled summer mean concentrations of NO 2 and SO 2 along the main shipping routes in 2040 are higher by 40 % and 25 % than in BAU, respectively (Fig. 10).
The lower fuel efficiency has little influence on daily maximum ozone concentrations over the Baltic Sea. Further, the 5 influence of the changed fuel efficiency on atmospheric secondary particle formation is rather limited (not shown). For PM 2.5 , the higher primary particle emissions compared to BAU do not fully propagate into surface air concentrations (increase by less than 10 %). A large fraction of the ship-related PM 2.5 is from secondary formation, which does not increase proportionally with the increased primary PM emissions, for example due to the limited availability of NH 3 .

Future nitrogen deposition
10 Summer-accumulated total nitrogen deposition to seawater in 2040 according to "BAU 2040" is below 100 mg(N) m −2 in most parts of the Baltic Sea, with highest deposition remaining in the Belt Sea (Fig. 11a). The average summer deposition rate to the Baltic Sea is 48 mg(N) m −2 . The ship contribution to total nitrogen deposition in summer is massively reduced (by more than 60 %) in the coastal areas of the Baltic Sea region compared to 2012 (Fig. 11b). Over sea, largest reductions of the ship contribution take place in an area extending from Kattegat to the Arkona basin.
Introduction of the NECA causes a maximum reduction of the summer-accumulated nitrogen over seawater by 18 %, compared to not introducing the NECA in 2021 (Fig. 11c). This means that the Tier II fleet in "NoNECA 2040" already accomplishes a large reduction in nitrogen deposition compared to today. The effect of the lower fuel efficiency in 2040 (according to 5 "EEDI 2040") is an increase of nitrogen deposition compared to BAU, mainly over the Northern Baltic Proper and over coastal areas. The relative increase is up to 12 % (Fig. 11d). Table 4 shows the "BAU 2040" annual and seasonal nitrogen deposition sums to the entire Baltic Sea seawater surface, for total, oxidised and reduced nitrogen. The ship-related annual nitrogen deposition reduces by 17.6 kt N, while the overall nitrogen deposition reduces by 70.3 kt N, compared to 2012. Thus the reduction of NO X emissions over the continent, in 10 accordance with a current legislation scenario for land-based emissions in the Baltic Sea region, has a larger impact on the future nitrogen input to the Baltic Sea than the shipping fleet. The BAU scenario adopts the agreed SO X emission abatement regulations: the already established SECA limit of 0.1 % S 10 in fuel from 2015 onwards followed by the global limit of 0.5 % S in ship fuels from 2020 onwards. On average, annual mean SO 2 decreases by ∼60 % between 2012 and 2040, independent of the shipping scenario. Consequently, particulate sulphate Table 4. Future (2040) annual and seasonal nitrogen deposition amounts (kt N) to the seawater of the Baltic Sea and ship-related nitrogen deposition according to scenario "BAU 2040", taken from the CD04 grid. Values in brackets denote the change (in kt N) compared to 2012.
Amounts refer to a Baltic Sea surface area of 431390 km 2 , including the western part of Skagerrak. where it drops by 50-60 % along the main shipping routes, and smaller over the coastal areas. The large drop over sea is due to the reduction of particulate matter emissions from ships and the lower production of sulphate and nitrate related to reduced emission of primary precursor gases (NO X and SO X ) from ship traffic. In most coastal areas the decreased PM 2.5 is mainly a 5 consequence of the abatement measures on land.
On annual average, the daily maximum O 3 decreases only slightly over the Baltic Sea region, but the summer average decreases by 10-25 %, depending on the shipping scenario, in large parts of Sweden and the Baltic Sea, compared to present day.
Overall, a lower fuel efficiency increase than in BAU has only marginal implications on the future air quality in the Baltic 6.2 Changes of the ship contribution in the future scenarios The absolute ship contribution to ambient levels of NO 2 and SO 2 between 2012 and 2040 changes slightly more than expected due to the reduction of ship emissions. The lower abundance of NO X in the future atmospheric background increases the oxidation capacity of the atmosphere and leads to a more efficient oxidation of pollutants via gas-phase reactions and incloud processing. Table 6 presents the relative changes of the annual mean absolute ship contributions in the Baltic Sea region 5 between 2012 and 2040. The effect of the lower fuel efficiency on the absolute ship contribution of air pollutants is limited. Still, the annual mean ship contributions in 2040 to the four pollutants is significantly higher than in the BAU scenario.

Contribution of ship emissions to nitrogen deposition
A previous study (Bartnicki et al., 2011) estimated the contribution of airborne nitrogen from international ship traffic to the oxidised nitrogen deposition in the Baltic Sea basin to be about 8 to 11 % (period: 1997-2006) on annual average. The contribution from ships with a range from 12 to 14 % has been reported for the period 2008 to 2011 (Hongisto, 2014). In the present study the relative ship contribution to the deposition of oxidised nitrogen is 24 % (Table 3), about twice as high as 20 the previous estimates. However, the total annual nitrogen deposition for 2012 in the present study is 29 % lower compared to the EMEP-MSC/W model used by HELCOM (Bartnicki et al., 2017). Taking the literature value of 14 % and the oxidised nitrogen deposition flux in 2012 reported by HELCOM (128.9 kt N y −1 ; Bartnicki et al., 2017), an absolute ship contribution of 18 kt N y −1 is derived, only slightly lower than our estimate of 22.5 kt N y −1 .
The relative ship contribution to the total nitrogen deposition is 14 % on annual average and 21 % in summer in the presentday situation (Table 3). The ship contribution drops to 5.6 % in 2040 (9 % in summer) when following the BAU scenario (Table 4). Between 2040 and 2012 the ship-related deposition of oxidised nitrogen decreased by 78 %. In "BAU 2040" the ship 5 contribution to the annual deposition of oxidised nitrogen over the Baltic Sea is only 14 %.
Nitrogen deposition to the seawater of the Baltic Sea decreases on average by 40-44 % between 2012 and 2040 (Table 5).
Depending on the future shipping scenario, the decline of the ship-related nitrogen deposition varies between 46 % and 78 % (Table 6). In the EEDI scenario, when the NECA is established but fuel efficiency increase is lower than in BAU, nitrogen deposition in most ship-influenced areas decreases less than in the BAU scenario. The weakest reduction is found for the 10 NoNECA scenario, in which nitrogen deposition decreases by only 30 % over coastal areas of Denmark, Germany and west Finland. The western part of the Baltic Sea would be most affected if the NECA is not implemented (Fig. 11c).

Prognosis of the total nitrogen deposition to the Baltic Sea
A linear relationship was found between the emissions of NO X from the Baltic Sea ship fleet and the annual ship-related nitrogen deposition to Baltic Sea seawater (spatial average) based on the results of the present-day simulation and the future 15 scenario simulations (Fig. 12). Because the changes of the nitrogen deposition attributed to shipping (Fig. 11b) between 2012 and 2040 are mainly confined to the Baltic Sea and the surrounding coastal areas, it was expected that the changes of the ship-related deposition flux are proportional to the atmospheric input of oxidised nitrogen via ship emissions. An important link between the ship emissions and the deposition of nitrogen is the formation of HNO 3 , which constitutes the most important removal pathway for nitrogen in the atmosphere (Riemer et al., 2003). 20 The relationship presented above is useful for a quick evaluation of the ship-related nitrogen deposition in future shipping scenarios. Cumulative scenarios based on Shared Socioeconomic Pathways (SSPs) with respect to future ship emission in the Baltic Sea region were designed in the SHEBA project. In scenario SSP3 (regional rivalry), which represents a world with much less international trade and low mitigation capacity (Fujimori et al., 2017), future shipping deviates largely from the already decided regulations but growth of shipping is slower than in BAU by 0.5 % p.a.. The fuel efficiency development is

Discussion of uncertainties and limitations
The ship contribution to air pollutants and nitrogen deposition in the present study was computed using a zero-out method, i.e. the ship emissions were removed in one simulation. An alternative brute force method would be the perturbation of the emissions, for example reduction by 20 %, which might be more careful with respect to the non-linearity of the involved photochemistry. However, our goal was to derive the impact of shipping in different scenarios; while perturbing emissions is 5 mainly used to investigate short-term responses to expected (small) changes of a sectoral emissions. A previous study by Geels et al. (2012) applied the so-called tagging method to assess the ship contribution from each riparian state of the Baltic Sea.
Tagging requires adding auxiliary variables to the CTM itself to track pollution. While tagging for inert primary pollutants is straightforward; methods for addressing secondary pollutants requires an analysis of the limiting reagents to avoid tagging all possible follow-up products in the gas-phase, aerosol phase and cloud water. Differences between tagging and brute force 10 methods are usually observed in secondary processes involving precursors from different sources. Some comparison studies (Collet et al., 2014;Koo et al., 2009) indicate that tagging is advantageous for source allocation rather than for predicting responses to emissions changes.
European regions that are affected by high density of ship traffic, such as UK, France, western Germany, North Sea, the southern part of the Baltic Sea and along the ship tracks in the Mediterranean are currently in a NMVOC-limited regime with respect to ozone formation (Beekmann and Vautard, 2010). In northern Europe, except of the region of the English Channel and parts of the North Sea, a transition from NMVOC-limited to NO X -limited regime is projected until 2020 (Beekmann and Vautard, 2010) and the next decades (Lacressonnière et al., 2014). In a NMVOC-limited regime the production of ozone is sensitive to emissions of NMVOC, while increasing NO X leads to a reduction of ozone by titration. In the NO X -limited regime, ozone is sensitive to emissions of NO X while it is hardly affected by additional NMVOC emissions.
In the simulations for the future scenarios in 2040, most certainly a transition towards a NO X -limited regime happens in the currently NMVOC-limited areas of the Baltic Sea, in particular along the ship tracks in the southern part. This is clearly seen in the "BAU 2040" scenario, where a relative increase of the ship-related daily maximum ozone occurred (due to less titration) 10 over the central shipping routes, whereas the ship-related ozone decreased in the already NO X -limited areas outside the ship tracks and over the coastal regions. However, predicted changes of the daily maximum ozone concentrations due to shipping are uncertain because of the lack of data on NMVOC emissions from shipping in the STEAM inventory that was used in the CTM calculations.
We have reduced land based emissions in the future scenarios in order to obtain a more realistic estimation of the conse- 15 quences of regulations on shipping emissions on the future air quality in the Baltic Sea region. Based on the model results for the future ship contribution, it is obvious, that reduced land-side emissions of primary gaseous precursors amplified the decline of secondary aerosols related to shipping, in particular over the coastal areas. However, the reduction of land-side emissions has a very small effect on the determined ship contributions to NO 2 and SO 2 over the Baltic Sea (Fig. S15).
The reason for the underestimation of WNO 3 and WNH 4 in the CMAQ simulations, compared to observations of the regional 20 background monitoring stations of the EMEP network, could not be fully resolved. The formation of particulate nitrate involves complex chemistry of several compounds in the gas-phase and multicomponent solution systems on aerosols. The simulation of nitrate is highly uncertain because it requires accurate computation of the concentrations of the precursors, e.g. HNO 3 , NH 3 , dust and sea-salt. The joint underestimation of WNO 3 and WNH 4 was found in the statistical analysis of model-observation pairs and also in the comparison of modelled and observed seasonal averages. The most convincing explanation at the current 25 stage is, that the oxidative conversion of NO X to HNO 3 occurs at a too slow rate in the model, combined with too little particulate ammonium from the regional background that is advected into the Baltic Sea region.
An alternative explanation might be that the wet removal of NO − 3 and NH + 4 in CMAQ is not efficient enough. In addition, the evaluation of simulated precipitation amounts and frequency showed that the southern part of the Baltic Sea receives too little rainfall in summer. For the other seasons and in the northern part the precipitation bias is positive. Too low precipitation 30 in the southern part, where modelled concentrations of NO − 3 and NH + 4 are much higher compared to the northern part, could be responsible for an average underestimation of the total nitrogen wet deposition to the Baltic Sea.
Coarse mode particles are much faster removed than fine mode particles, therefore the deposition of particulate nitrate crucially depends on the uptake to larger particles. Heterogeneous chemical production of nitrate on coarse mode particles has been found to control the atmospheric nitrate production to a very large extent (Bian et al., 2017). The hydrolysis of N 2 O 5 to 35 produce HNO 3 is considered in CMAQ by uptake coefficients depending on temperature, RH and particle composition, using the parameterization by Davis et al. (2008), but only for fine-mode aerosols. The Davis parameterization tends to predict too high N 2 O 5 uptake coefficients near the surface, especially over marine and coastal areas where relative humidity is high (Chang et al., 2016). CMAQ allows for a dynamic mass transfer to coarse particles and therefore takes into account the reactive uptake of HNO 3 by sea salt particles. Meanwhile, resuspension of mineral dust was not activated in the simulations, and the missing 5 heterogeneous reaction on dust particles surfaces may have contributed to the underestimation of WNO 3 .

Conclusions
The impact of ship emissions on the present-day (2012) and future (2040) air quality and nitrogen deposition was evaluated with a regional atmospheric CTM. The meteorological fields, the emissions from ship traffic and the emissions from landbased sources are considered at a grid resolution of 4 × 4 km 2 for the inner-most model domain covering most of the Baltic Sea 10 region. Ship emissions from the STEAM model based on ship movements from AIS records and detailed ship characteristics in combination with solving atmospheric chemistry and transport at high resolution, enable a better treatment of the plumes from ship traffic, compared to previous CTM studies in the Baltic Sea region.
The effect of future legislation related to shipping and of future changes of the ship fuel efficiency of the ship fleet on air quality and deposition in 2040 in the Baltic Sea region was determined based on computational results from regional CTM 15 simulations. Future air quality and nitrogen deposition is analysed, in order to investigate: (1) the effect of establishing the NECA in 2021 compared to a future situation without NECA; and (2) the effect of a lower fuel efficiency increase than expected based on continuation of the current trend. A BAU scenario has been designed in which the NECA is implemented and the fuel efficiency for new ships improves more than required by IMO's Energy Efficiency Design Index regulation.
Establishing the NECA in 2021 has several benefits for the Baltic Sea environment. One important effect of the NECA is a 20 reduction of secondary formation of particulate nitrate. The introduction of the NECA reduces the ship-related PM 2.5 by 72 % in 2040 compared to present-day, while it is reduced by only 48 % without implementation of the NECA. A major consequence of establishing the NECA is a reduction of the ship contribution to daily maximum ozone in 2040 compared to the present situation. If the NECA is not implemented, an increase of the ship-related daily maximum ozone results compared to presentday. The introduction of NECA is thus critical for abating ship emissions of NO X to levels that are low enough to sustainably 25 dampen ozone production in the Baltic Sea region. Overall, the introduction of the NECA is expected to be beneficial for avoiding future health impacts of ozone and PM 2.5 in coastal areas of the southern part of the Baltic Sea region.
The effect of the lower fuel efficiency on the absolute ship contribution of air pollutants is relatively small. The implementation of the NECA in 2021 can be regarded as safeguard for the case that the fuel efficiency increase falls below the projected trend.

30
The decline of the ship-related nitrogen deposition to the Baltic Sea between 2012 and 2040 varies between 46 % and 78 % in the different future scenarios. When the NECA is established but the fuel efficiency increase is lower than expected, nitrogen deposition in most ship-influenced areas decreases less than in the BAU scenario. The weakest reduction is found for the scenario without implementing the NECA, in which nitrogen deposition decreases by only 30 % over coastal areas of Denmark, Germany and west Finland. The western part of the Baltic Sea would be most affected if the NECA is not implemented.
A prognostic relationship for a quick evaluation of the ship-related nitrogen deposition in future shipping scenarios was derived in this work. The relationship should be further modified to consider the inter-annual variability of atmospheric deposi-5 tion due to changing meteorological conditions in order to allow for more robust projections of the ship-related nitrogen input to the Baltic Sea. However, it may be used for estimating possible exceedances of critical loads for eutrophying substances that are based on annual nitrogen inputs.
A limitation of the model results for regional surface concentrations of the daily maximum ozone concentrations over the Baltic Sea region is the lack of data on NMVOC emissions from shipping in the STEAM inventory that was used in the CTM 10 calculations. Additional NMVOC emissions from shipping would serve as precursors of ozone and enhance photochemical ozone production in a NMVOC-limited regime. In the presented model simulations, NO X emissions from continental sources were reduced by 60 % between 2012 and 2040, following current legislation, i.e. already decided emission abatement regulations. The lower abundance of NO X in the future could lead to a shift in the overall atmospheric chemical regime. To predict more accurately how such change in the chemical regime will affect the future influence of ship emissions, a better handle on 15 NMVOC emissions from ships and their future development would be important.
As a consequence of SO X emission abatement regulations for shipping, annual mean SO 2 decreases on average by ∼60 % between 2012 and 2040, independent of the future scenario. With the reduction of SO 2 emissions, less NH 3 is required to neutralise the strong acid H 2 SO 4 . The excess NH 3 is available for the formation of NO − 3 and NH + 4 in the particulate phase. According to Tsimpidi et al. (2008), the trend of future particulate NO − 3 concentrations depends on whether NO X or NH 3 are 20 the limiting gas-phase compounds for nitrate formation. Measurements in southern Sweden have shown that the concentrations of NH 3 and HNO 3 are too low to form pure solid or aqueous ammonium nitrate particles (Ferm, 1992). Thus in a future background atmosphere over the Baltic Sea region, ambient levels of both gases might be too low for ammonium nitrate formation, and the fate of these gases would be the removal by dry and wet deposition. Meanwhile, the formulation of heterogeneous processes related to the production of nitrate are highly uncertain in the models, limiting the conclusions about the future 25 transition in the nitrate formation regime.
Use of the presented model data for health impact assessment in the densely populated coastal areas of the Baltic Sea region is connected to uncertainties arising from limitations of the chosen grid resolution. Despite the fine spatial resolution of the inner-most model grid, the concentration gradients between urban areas and their surroundings (urban increment) and within harbour cities are not adequately resolved by the simulations due to the large spatial and temporal variability of emissions 30 in urban areas. Ideally, a grid length of 1 km should be chosen to resolve the urban increments (Schaap et al., 2015) in the coastal areas. However, a finer resolution brings along the need for more accurate emission data in the urban areas, which is challenging because the compilation of urban emission inventories requires specific information for each emitting sector (Guevara et al., 2016).
A related study by Jutterström et al. (2018) assessed the extent of environmental damage related to shipping on the terrestrial ecosystems surrounding the Baltic Sea. Ecological impacts of air pollutants on land are evaluated in terms of critical load (CL) exceedance for eutrophication. Using the latest reported CL values for eutrophication together with the modelled deposition data of nitrogen for 2012 and the future scenarios for 2040 of the present study, Jutterström et al. (2018) find a significant improvement from 2012 to 2040. For the BAU scenario, the area where the CL (eutrophication) are exceeded due to ship-5 related nitrogen deposition decreased from about 20 % in 2012 to 5 % in 2040. If the NECA is not implemented, the exceeded area due to shipping is about 14 % in 2040, indicative for the relevance of the NECA for coastal ecosystems surrounding the Baltic Sea. We note, that the use of gridded model data of dry deposition in the estimation of CL exceedances has limitations.
In the model simulation, dry deposition to land surfaces is weighted for the different land use classes present in each grid cell.
This might lead to an underestimation of the eutrophication risk for forests in a grid cell which includes other land uses, as the 10 canopy resistance of forests is much higher than that of grassland or other low vegetation. The CMAQ deposition data is less affected by this problem due the high resolution of the gridded data.
The shipping sector is an important contributor to atmospheric nitrogen deposition to the Baltic Sea. The present study estimates a deposition flux of oxidised nitrogen in the order of 22.5 kt N y −1 due to shipping emissions for the year 2012, slightly higher than previous estimates (Hongisto, 2014: Bartnicki et al., 2017. Occurrences of high nutrient input to coastal 15 waters have been suggested to cause short-term algal blooms (Spokes et al., 2000). On the other hand, a study in the Kattegat showed that direct nitrogen inputs through atmospheric deposition could not be linked to any summer algal bloom observation, probably because the atmospheric input is considerably diluted through mixing in the surface water layer (Carstensen et al., 2005). The incidence of harmful algal blooms, which cause health damages to humans and animals in shallow coastal waters, has also been linked to atmospheric nitrogen inputs (Paerl, 1997). However, the relationships between high nutrient inputs and 20 the development of harmful algal blooms are still not well understood (Anderson et al., 2002).
Much stricter regulations for NO X emission from new built ships will be enforced in 2021. It can be expected that significant emission reductions will be the consequence of these regulations, however, this requires that the exhaust gas cleaning technologies that will be implemented on board of most the new built ships work properly. From the experiences with Euro 4 and Euro 5 diesel cars that frequently emit much more NO X than allowed, policy should pave the way for extended compliance control 25 measures. Several techniques exist how emissions from ships can be measured, including in-situ observations at coastlines, ground based remote sensing techniques, sniffers on board of aircraft or drones and sensors on board of the ships. The best technology needs to be tested now in order to be prepared for the implementation of the NECA.