Projections of shipping emissions and the related impact on air pollution and human health in the Nordic region

. International initiatives have successfully brought down the emissions, and hence also the related negative impacts on environment and human health, from shipping in Emission Control Areas (ECAs). However, the question remains as to whether increased shipping in the future will counteract these emission reductions. The overall goal of this study is to provide an up-to-date view on future ship emissions and provide a holistic view on atmospheric pollutants and their contribution to air quality in the Nordic (and Arctic) area. The ﬁrst step has been to set up new and detailed scenarios for the potential developments in global shipping emissions, including different regulations and new routes in the Arctic. The scenarios include a Baseline scenario and two additional SO x Emission Control Areas (SE-CAs) and heavy fuel oil (HFO) ban scenarios. All three scenarios are calculated in two variants involving Business-As-Usual

air pollution concentrations resulting from all emissions and the share related to shipping. It is expected that models with different physical descriptions and setups give some differences in modelled air pollution and by using two models an 85 indication of the related uncertainties is displayed. The included emissions represent current day conditions and future projections towards 2050. For the Nordic area, we focus mainly on total PM2.5, while for the Arctic we focus on the deposition of nitrogen and black carbon. Furthermore, the modelled concentration maps serve as input to the health impact assessment system EVA, in order to assess the overall impacts of air pollution on the human health in the Nordic area and the changes in health impacts resulting from the different ship emission scenarios. 90

Materials and methods
The overall modelling setup will be described in the following. The study area and the included model domains are shown in

Setup of shipping emission inventories
The background data for the emission scenarios is, traffic data for the area north of 60N, emission factors and scenario specific emission inventories from the ship emission model developed at the Danish Centre for Environment and Energy (DCE) at Aarhus University (Winther et al., 2017). In order to obtain a spatial coverage of the entire Nordic area and the Arctic, the 100 DCE emission inventories are combined with a global CO2 ship emission inventory produced with the Ship Traffic Emissions Assessment Model (STEAM) for 2015 (Johansson et al., 2017).

STEAM model
Global emission data from the STEAM model (developed at the Finnish Meteorological Institute) from an earlier study 105 (Johansson et al., 2017) was applied in this work. The model uses global Automatic Identification System (AIS) data to describe shipping activity and it applies vessel level modelling using technical description of each ship in the global fleet (Jalkanen et al., 2009;Johansson et al., 2013Johansson et al., , 2017. Details of the method and the numerical results for emissions used d01 https://doi.org/10.5194/acp-2020-1274 Preprint. Discussion started: 9 February 2021 c Author(s) 2021. CC BY 4.0 License. in this study can be found in Johansson et al. (2017). The AIS data from terrestrial and satellite networks was purchased from Orbcomm Ltd. (Orbcomm: 395 West Passaic Street, Suite 325 Rochelle Park, NJ 07662 USA) and the technical details 110 database was acquired from IHS Markit (IHS Markit Global Headquarters 4th floor Ropemaker Place25 Ropemaker Street, London EC2Y 9LY). Emissions for ships were modelled without considering weather effects (wind, waves, ice, currents) and thus represent ideal conditions, which may introduce uncertainties when compared to real emissions and fuel consumption.
Regardless, average absolute deviation of the STEAM predicted fuel consumption for any single ship is around 19%, whereas inventory level totals are equivalent to those reported in the EU Monitoring, Reporting and Verification system (EU, 2015). 115 These comparisons were made for more than 2550 vessels reporting their fuel consumption during the reporting year 2019.

Traffic activity data in the DCE ship emission model
The ship activity data used in the DCE ship emission model are provided by the Danish Maritime Authority (DMA) based on AIS signals received from terrestrial base stations and from satellites equipped with AIS receivers for the area north of 60N.
The data represent the years 2012-2016, divided into 0.5° longitude x 0.225° latitude grid cells, with a monthly resolution. The 120 ships are classified into 14 ship types and 16 ship length categories and data for total sailed distance and average sailing speeds are provided stratified into the different ship types/length/average speed combinations that have been recorded in the individual cells.
A weighted and consolidated ship activity data set for a base year was prepared for the DCE ship emission model based on the five-year ship activity data provided by DMA in order to avoid inexpedient temporal and spatial specific fluctuations in traffic 125 records, and in order to achieve a uniform grid cell reference system for the emission projection calculations.
The traffic scaling factors used in the DCE ship emission model for traffic projections are derived from traffic growth factors in the Corbett et al. (2010) business as usual (BAU) and high growth (HiG) scenarios by referring the DCE ship types to the Corbett et al. (2010) ship types and by using Corbett traffic growth factors evolved from the base year. Fuel efficiency improvements for future ships are modelled from EEDI fuel efficiency regulations agreed by Marpol 83/78 Annex VI and 130 mandatory from 1 st January 2013 for new built ships larger than 400 GT. For further explanations regarding ship activity data, traffic scaling factors and EEDI factors, see Winther et al. (2017).

Scenarios
The current study includes a Baseline emission projection scenario and two additional emission projections; a SECA (sulphur emission control area) scenario and an HFO (heavy fuel oil) ban scenario (see table 1). The Baseline scenario forms the basis 135 for the SECA and HFO ban scenarios.
All three scenarios use the BAU and HiG growth traffic activity projections explained above in section 2.1.2 and further the scenarios assume an increasing amount of liquefied natural gas (LNG) fuel being used as a substitution for heavy fuel oil in the inventory area throughout the projection years. The scenarios use the "low case" LNG fuel share of total marine fuel consumption being 2 %, 4 % and 8 % in the years 2020, 2030 and 2050, respectively, as described in the IMO 3 rd GHG study 140 published by IMO (2015).
The Baseline scenario assumes an increase in the use of Exhaust Gas Cleaning Systems (EGCS) for SO2 emission abatement in the case of ships using HFO with a high content of sulphur. In the Baseline scenario inside the existing SECA zones (i.e. America and North Sea/Baltic Sea SECA's) the fuel type switches from HFO to marine diesel/marine gas oil (MDO/MGO) for ships using HFO and not having an EGCS installed (c.f. Winther et al., 2017). Outside the existing SECA's the latter ships 145 use 0.5 % HFO after the global sulphur cap introduction in 2020.
In the SECA scenario, the existing SECA zones (i.e. America and North Sea/Baltic Sea SECA's) are expanded to cover the entire inventory area. The SECA scenario takes on board the Baseline shares of LNG fuel consumption and EGCS installations. https://doi.org/10.5194/acp-2020-1274 Preprint. Discussion started: 9 February 2021 c Author(s) 2021. CC BY 4.0 License.
Further in the SECA scenario, the fuel is shifting from HFO to MDO/MGO outside the existing SECA's by ships using HFO and not using EGCS. 150 In the HFO ban scenario, no use of HFO by ships is allowed at all in the inventory area. The HFO ban scenario includes the consumption of LNG as assumed in the Baseline scenario. The remaining part of the HFO consumption not being substituted by LNG is assumed to switch to MGO/MDO in the entire inventory area.

Emission factors
The specific fuel consumption factors (SFC) and NOx emission factors (g/kWh) for HFO and MDO/MGO used in the 155 calculations are classified according to engine type and engine production year (Ministry of Transport, 2015;MAN Energy Solutions, 2012). The CO2 emission factors (g/kg fuel) comes from Nielsen et al., (2019 (ICCT, 2015;Lack and Corbett, 2012;Johnson et al., 2016). For LNG, a BC emission factor of 0.00155 g/kg fuel is used, derived as 1 % of the BC emissions for HFO (ICCT, 2015).
The full set of basis emission factors is not shown here, but they are explained in more detail in Winther et al. (2017). However, 170 aggregated from ship type, engine type, fuel type and engine production year, fuel related emission factors for 2015 and the forecast years 2020, 2030 and 2050 are shown in Figure 2 derived from the fuel and emission results presented above. The development of the emission factors reflects the emission technology improvements (for NOx) and fuel sulphur content and the Baseline shares of LNG fuel consumption and EGCS installations (for SO2 and BC). The emission factors shown in Figure   2 include the emission factor adjustments made in the calculations in order to account for engine load variations (c.f. Winther 175 et al., 2017).

Scaling of the global CO2 ship emission inventory
In order to cover the entire Nordic area and the Arctic, and make use of already well defined and elaborated scenarios for shipping in the Arctic, the global CO2 ship emission inventory produced with the STEAM model for 2015 is used in combination with the Arctic fuel consumption and emission scenarios calculated with the DCE ship emission model for , 180 2020, 2030and 2050(Winther et al., 2017. The STEAM data stops at 74N, so the area beyond this is covered by the DCE model. STEAM related emission results are subsequently obtained as the product of 1) the aggregated emission factors obtained with 190 the DCE ship emission model for each scenario, scenario year and STEAM ship type and 2) the corresponding fuel consumption. The final emission data set for 2015 and the scenario years are monthly files with a spatial resolution of 0.1º×0.1º.

The DEHM model
The Danish Eulerian Hemispheric Model (DEHM) is a state-of-the-art three-dimensional, Eulerian, atmospheric chemistry transport model (CTM) originally developed in the early 1990's to study the atmospheric transport of  Sulfate into the Arctic (Christensen, 1997). The model has been modified, extended and updated continuously since then and now includes a comprehensive chemical scheme, detailing 80 chemical species and 158 chemical reactions including a Volatile The meteorological data, driving DEHM, is calculated using the WRF model (Skamarock et al., 2008), set up with the same domains and nests as DEHM and forced with ERA-Interim meteorology (Dee et al., 2011). Natural emissions like sea salt and biogenic volatile organic compounds (VOCs) are calculated online in the model as a function of meteorological parameters 205 (described in Soares et al., 2016and Zare et al., 2014. The anthropogenic emissions for the current and future periods are based on the global 0.5° x 0.5° ECLIPSE V5a data sets including sectoral emissions (https://www.iiasa.ac.at/web/home/research/researchPrograms/air/Global_emissions.html). We apply the future Baseline scenario assuming Current LEgislation (CLE) for the air pollution components (see Klimont et al., 2017 for an overview). The projected change in the land-based emissions in the applied domains in DEHM, are given in table S1 in the Supplement to this 210 paper. The shipping emissions in ECLIPSE V5a have been replaced with the new shipping emissions described in Section 2.1.
We have run the model for the meteorological year 2015 (and December 2014 as spin up) with a combination of land based ECLIPSE V5a and new shipping emissions representing the years 2015, 2030 and 2050. This is done in order to isolate the impact from emission changes. Additionally, we have made simulations, with and without a new polar diversion route and simulations where the shipping emissions have been reduced by 30% (i.e. multiplied by 0.7). By scaling the results afterwards, 215 the impact from shipping alone can be analyzed, but non-linear effects of atmospheric chemistry are still included. In total 22 simulations have been made with the DEHM model (an overview of model runs is given in Table S2 in the Supplement). The Baseline simulation with 2015 BAU emissions has been evaluated by comparison to European observations of the components relevant for the health assessment (PM2.5, NO2, and O3) and a sufficiently good agreement between model and observations is seen both in terms of level and variability (see the Supplement for details). Furthermore, the DEHM model is one of the core 220 models in the Copernicus Atmosphere Monitoring Service (CAMS) providing daily air pollution forecasts and analyses, which are continuously evaluated on-line against European observations (see https://www.regional.atmosphere.copernicus.eu/).

The MATCH model
The MATCH (Multi-scale Atmospheric Transport and CHemistry) model (Robertson et al., 1999;Andersson et al., 2007; is a state-of-the-art Eulerian chemistry and transport model, including wet-, thermal-and photochemical reactions, to 225 describe the sulfur and nitrogen cycle as well as tropospheric ozone chemistry, and particle formation and transformation. The version used in the current set up is the same as was used in the EURODELTA-trends exercise (Colette et al., 2017). It includes online emissions of sea salt as described in Soares et al. (2016). Secondary organic aerosol formation is modelled through a volatility basis set and biogenic VOC emissions are modelled online, both as described by Simpson et al. (2012). Further details on the model configuration are described in Colette et al. (2017). The driving meteorological forcing data used was the 230 https://doi.org/10.5194/acp-2020-1274 Preprint. Discussion started: 9 February 2021 c Author(s) 2021. CC BY 4.0 License.
HIRLAM operational weather data for 2015 for the domain covering Fennoscandia (11 km resolution). The lateral and top boundaries of this domain were fed every 6 hours by results from the DEHM model, domain d01 (see Figure 1). The anthropogenic and shipping emissions were the same as for DEHM, but with MATCH we simulate the current (2015) and a selection of the 2050 shipping scenarios. The MATCH grid is smaller than the DEHM d03 grid with a finer horizontal resolution (11 km x 11 km). The shipping attribution was conducted in the same manner as with the DEHM model, by reducing 235 the shipping emissions by 30% and using the corresponding DEHM simulation on the boundary of the MATCH grid. In total 12 simulations have been made with the MATCH model ( Table S2 in the Supplement).
The current model configuration is extensively evaluated with in situ observations and compared to the performance of other models in numerous papers in similar set ups as in this study for ozone, particles, and nitrogen and sulfur deposition in Europe (Otero et al., 2018;Theobald et al., 2019;, Ciarelli et al., 2019a. The conclusion is that MATCH in the 240 current configuration, performs among the best models for near-surface ozone as well as nitrogen and sulfur deposition, and for particles, it displays high correlation with observations while PM concentrations are somewhat underestimated. An evaluation of MATCH for the Fennoscandia region for the year 2015 is included in the Supplement of this paper. MATCH is also a core model in the operational CAMS (Marécal et al., 2015; https://www.regional.atmosphere.copernicus.eu/) including daily updates of daily air pollution forecasts, fused measurement and modelling of atmospheric concentrations (through data 245 assimilation), as well as model performance scores. MATCH is a building block in the MATCH Sweden system for environmental surveillance , including measurement model fusion of total atmospheric deposition (MMF-TDEP) of ozone, nitrogen, sulphur and base cations.

The EVA system
The EVA (Economic Valuation of Air pollution; Brandt et al., 2013a;Geels et al., 2015, Im et al., 2018 model system is 250 based on the impact pathway chain (Friedrich and Bickel, 2001), where modelled air pollution levels are coupled to population data for calculation of human exposure, health impacts (both mortality and morbidity), and related external costs. In the current study, we focus on the health impacts and are not including the assessment of the cost. The health impacts are calculated using linear exposure-response functions, which in the applied model version (EVAv5.2) are based on the HRAPIE recommendations (WHO, 2013). The number of premature deaths in the system is calculated from short-term exposure to O3, 255 NO2, SO2 and PM2.5 (acute deaths) as well as long-term exposure to PM2.5 and NO2 (chronic deaths). The EVA model system can be used to estimate the health impacts due to the total air pollution levels or due to contributions from various model scenario runs, where e.g. specific emission sectors are reduced, or the weather/climate scenarios are changed. In this study, the population exposure has been assessed for the d03 DEHM domain by combining the concentration maps from both DEHM and MATCH with gridded population data from EUROSTAT for 2015 (see the Supplement for a table (S4) with the national 260 total populations and a map of the population distribution in the applied 16.67 km x 16.67 km grid). In order to do so, the MATCH 11 km x 11 km gridded data has been aggregated to the d03 grid. The EVA model system has been compared to other similar models (Anenberg et al., 2015;Lehtomäki et al., 2020), is part of the Danish monitoring programme (Ellermann et al., 2020) and has been used routinely in numerous advisory projects for public authorities. In Section 3.2 the current results for 2015 are compared with EEA's results for the same year (EEA, 2018). 265

Results
In this section we first describe the projected developments in the shipping emissions in the different scenarios within the Nordic area. Thereafter we describe the modelled current day air pollution levels with focus on total PM2.5 and the overall health impact related to air pollution in the Nordic area. We then move on to the future developments in the PM2.5 levels as simulated by the two models based on the shipping scenarios as well as the projected development in the land-based emissions. 270 https://doi.org/10.5194/acp-2020-1274 Preprint. Discussion started: 9 February 2021 c Author(s) 2021. CC BY 4.0 License.
The related impacts on the number of premature deaths are then analyzed. Next we focus more directly on the contributions from shipping in terms of the PM2.5 levels and health impacts. Finally we move to the Arctic and demonstrate, how the deposition of BC and Nitrogen will be effected by the projected developments in emissions. Table 1 shows the shipping emissions for the domain area d03 for the base year 2015, and the Baseline, SECA and HFO ban 275 BAU scenarios in 2020, 2030 and 2050. The percent changes between Baseline and SECA/HFO ban scenario results in 2020, 2030 and 2050 are also shown in Table 1.

Baseline results
The emission development relies on the development in fuel consumption and the corresponding emission factors. Figure 3 shows the emissions of CO2, NOx, SO2 and BC for 2015 and Baseline scenario results for 2020, 2030 and 2050, split into the current ECA and non ECA parts of domain d03. 285 In terms of total fuel consumption, the projected growth in ship traffic during the forecast period more than outweighs the future fuel efficiency improvements for ships.   For SO2 and BC, the major reason for the emission reductions outside SECA from 2015 to 2050 is the shift from HFO with a Sulphur content of 2.45% in 2015 to HFO with 0.5% Sulphur from 2020 onwards and the consequently reduced emission factors. 305  fuel being replaced. The SO2 emissions increase slightly less than fuel consumption due to the gradually increased 310 consumption of liquefied natural gas (LNG) in the Baseline scenario.

Emission and fuel consumption results across scenarios
The fuel consumption and NOx emission totals calculated for the SECA and HFO ban scenario equal the results obtained in the Baseline scenario (Table 1; Figure 4 (NOx)). The main reason for this is that the engine specific fuel consumption and NOx  In all scenario years for SO2, the calculated emissions for the SECA and HFO ban scenarios are close to 30% lower than the emissions calculated for the Baseline scenario (Table 1). In the SECA scenario, HFO is only used by ships with EGCS, with 325 a Sulphur removal efficiency equivalent Fs = 0.1 (Section 2.1.3). The 0.5% fuel oil used outside the original SECA area, which is not being replaced by LNG, is replaced by MDO/MGO (Fs = 0.08) according to the scenario definitions. In the HFO ban scenario, all HFO consumption by ships not being replaced by LNG is replaced by MGO/MDO.  Apart from LNG with similar fuel consumption shares assumed in all three scenarios, in the HFO ban scenario, only the fuel type MDO/MGO with the smallest BC emission factor (0.053 g/kg fuel, before load adjustment) is used. In the Baseline and SECA scenarios similar shares of EGCS are used with a higher BC emission factor (0.093 g/kg fuel, before load adjustment). 335 In the SECA scenario, however, the BC emissions from MDO/MGO fuel that replaces 0.5% fuel oil are smaller due to the level of the BC emission factors.

Results for diversion routes
Potential changes in polar sea ice distribution and quantity due to a future warming, can open up for additional ship traffic in the Arctic diverted from current shipping routes. Therefore scenario estimates of CO2 (proxy for fuel consumption), BC, NOx 340 and SO2 for so called 'diversion traffic' are made based on the Business As Usual (BAU) and High Growth (HiG) scenario emission results for the diversion routes 1 from Corbett et al. (2010). Based on the CO2 emissions for the diversion routes and the fuel related emission factor for CO2 from Corbett et al. (2010), the diversion route fuel consumption is calculated. Fuel consumptions are then further modified by taking into account future fuel efficiency improvements for ships.
Next, the diversion route emissions related to the Baseline, SECA and HFO ban scenarios for BAU traffic and the Baseline 345 scenario for HiG traffic, respectively, are calculated as the product of the diversion route fuel consumption and the emission factors derived from each of the four scenarios. Table 2 shows the estimated emissions of CO2, SO2, NOx and BC for the polar diversion routes passing through domain d03 in 2020, 2030 and 2050 for the Baseline, SECA and HFO ban emission scenarios based on the BAU traffic growth scenario and the Baseline emission scenario based on the HiG traffic growth scenario, respectively. The diversion emission results are 350 also shown in Figure 5 together with the totals without including the diversion emission contribution (Table 1) for each scenario case.

Current day Nordic air pollution and health effects 365
The annual mean PM2.5 concentration as modelled with the two models in the Baseline current situation (2015)  The total number of premature deaths attributable to short term exposure to O3, NO2, SO2 and PM2.5 (acute effects) as well as long-term exposure to PM2.5 (chronic effects) for the base year 2015 are shown in Figure 7 for Norway, Finland, Denmark and Sweden. Concentration fields from both the DEHM and the MATCH model have been used as input to the EVA model and the results are given as box plots in order to show the central (mean) estimate as well as the range between the two models. 375 The upper estimate is for all countries obtained by the DEHM-EVA setup, while the MATCH-EVA setup gives a slightly lower estimate. The total number of premature deaths in the Nordic region in 2015 is estimated to be between ca. 9400 (MATCH-EVA) and ca. 10.400 (DEHM-EVA) (or 9900 ± 10% -given as the average of the two models and the difference as percent). The number is lowest for Norway (1300±8%) and Finland (1700±2%), while highest for Sweden (3600±7%).  For Denmark the total number of premature deaths is only slightly lower (3300±3%). Part of the difference between the countries is of course related to difference in the population numbers, where Sweden with a population of about 9.9 million (in 2015) by far is the largest in the Nordic region (see table S4). The high number of premature deaths in Denmark, where the 385 total population is only slightly higher than in Norway, can be explained by the higher air pollution levels across Denmark (see Figure 6a and 6b). In all countries the number of premature deaths attributable to long-term exposure to PM2.5 is a factor of 2-4 higher than the number of premature deaths attributable to acute effects. The DEHM-EVA setup also covers Iceland https://doi.org/10.5194/acp-2020-1274 Preprint. Discussion started: 9 February 2021 c Author(s) 2021. CC BY 4.0 License. and the Faroe Islands, where the total number premature deaths are estimated to be approximately 40 and approximately 15, respectively, in 2015. MATCH-EVA excludes these regions, to the benefit of a higher resolution in a smaller domain. 390

Future developments in Nordic air pollution and health effects
The annual mean PM2.5 concentration as simulated by both models for the BAU Baseline 2050 scenario is seen in Figure 6c and 6d. The land-based emissions of e.g. NOx and SOx are projected to decrease in most of Europe in the applied ECLIPSE V5a scenarios (see Table S1 in the Supplement). This leads to significant general reductions in the annual mean PM2.5 levels 410 in most parts of the Nordic region towards 2050 (see Figure 6e and 6f for the change in %). Parts of Russia (e.g. around https://doi.org/10.5194/acp-2020-1274 Preprint. Discussion started: 9 February 2021 c Author(s) 2021. CC BY 4.0 License. concentration is projected to increase. In the MATCH results, the concentration in the Oslo region is projected to increase by a few percent. This is not seen in the DEHM results. Potential causes to this are the slightly lower resolution in the DEHM setup, differences in chemical scheme and differences in long-range transported (LRT) component from continental Europe 415 (the LRT-component is weaker in MATCH), possibly partly due to differences in meteorological forcing. Thus, although there is a general decrease in exposure to PM2.5, some areas are projected to possibly experience increased exposure. on the %-difference maps given in Figure 9. 425 With the SECA scenario (9a and 9b), a small decrease in the PM2.5 concentration is seen outside the current Baltic/North Sea SECA areas compared to the BAU scenario. Along the Norwegian coast the two models project a decrease in PM2.5 concentration between 0.6-1.5%, whereas smaller changes are seen across the other countries.
In the HFO ban scenario (9c and 9d) a slightly larger decrease is seen in most of the domain and the PM2.5 levels along shipping routes in the Baltic and around Denmark are up to 1.5% lower than in the Baseline, and like in the SECA scenario the largest 430 decrease (ca 2%) is seen along part of the Norwegian coast in the MATCH simulation. These changes in the PM2.5 levels are as described in Section 3.1.2 linked to decreased SO2 emission in both the SECA and HFO ban shipping scenario, which leads to a decrease in the formation of secondary inorganic aerosols and hence in the total PM mass. The HFO ban scenario also includes a decrease in the BC, which leads to small decrease in the primary PM part in the models.
The HiG scenario will on the other hand, result in an increase in the PM2.5 concentration compared to the BAU scenario ( Figure  435 9e and 9f). Here an increase in the PM2.5 levels of up to 1-5% are seen across Finland, Norway and Sweden and even higher over Denmark, according to the DEHM results. Somewhat lower values are seen in the MATCH results.
The resulting development in the number of premature deaths in the Nordic region is seen in Figure 10a

The contributions from shipping
The specific model simulations with shipping emissions reduced by 30%, can for each scenario be used to quantify the 450 contributions from shipping to both the air pollution levels and the related health impacts. In terms of the annual PM2.5 concentration, the contribution from shipping for all scenarios (given as % of total PM2.5 in Figure 11) is highest along shipping routes and in coastal regions giving a similar spatial pattern for both the DEHM and the MATCH models. For the current 2015 Baseline about 7-15% of the annual mean PM2.5 concentration over the majority of the Nordic land areas is linked to shipping in the DEHM results, while the MATCH results point to 1-10% for the same areas. This is in line with a recent study based on 455 the EMEP model, where the contribution from shipping to the averaged PM2.5 concentration in the Nordic countries ranged from about 5% in Finland to about 13% in Denmark (Jonson et al., 2020). The estimated number of premature deaths attributable to shipping can be seen in Figure 12. As a mean over the two models this amounts to approximately 850 premature deaths under current day conditions, decreasing to just below 600 cases in the 470 2050 BAU Baseline scenario and to about 580 and 550 in the BAU SECA and BAU HFO ban scenarios. In the HiG Baseline scenario this number is projected to increase to almost 900 cases, and hence to a value slightly higher than today. Like for PM2.5 described above, the contribution from ships is around 11% of the total number of premature deaths, when based on the DEHM-EVA simulation for 2015. When based on the parallel MATCH-EVA results, the estimated contribution from shipping is ca. 6%. In the DEHM BAU Baseline, BAU SECA and BAU HFO ban simulations for 2030 this fraction increases somewhat 475 to about 13%, before it decreases to ca. 9% in the 2050 scenarios. In the HiG Baseline scenario the fraction is ca 15% in 2030, with a decrease to 13% in 2050.
The future scenarios discussion here all include the polar diversion as described in Section 3.1.3. To isolate the effect of this, simulations have been made with the 2050 BAU Baseline scenario with and without the polar diversion. In terms of health 480 effects the additional emission leads to 10-30 premature deaths in the Nordic area.

Shipping and related depositions in the Arctic
The second domain in the DEHM model (d02) covers the Arctic region and in the following, the impact of the shipping scenarios are briefly analysed for two components (BC and total Nitrogen (N) deposition) within the Arctic area, where we 495 here focus on the modelled deposition of these components.
A map of the BC deposition is given in Figure 13a for the DEHM BAU Baseline 2050 simulation (for the d02 domain with a horizontal resolution of 50 km x 50 km). The deposition of BC under both present day and future conditions show the same overall pattern, with a gradient towards the Arctic area, where the deposition level is low (<1 mg/m 2 ). With the included emissions in the Baseline scenario (including both land-based and shipping emissions), the deposition is projected to decrease 500 by -1 % to -10 % across the Arctic towards 2050. The overall contribution from shipping to the BC deposition in the Arctic is very low and below 1% in most of the Arctic area, as can be seen from Fig 13b (the pattern is very similar for the 2015 simulation). In the simulation including the new diversion shipping routes (Fig 13c), contributions of up to 3% can be seen along the ship routes across the Arctic Sea and in the Baffin Bay, along the coast of Ellesmere Island, Canada and Northwest Greenland. The contribution will be higher during summer, when the shipping activity along these routes peaks. In 2050 the 505 emission along the new diversion routes will constitute approximately 50% of the BC related to shipping in the area.
As described in Section 3.1 for the Nordic region, the future mitigation scenario with an HFO ban will have the largest impact on the BC emissions and for the Arctic this scenario leads to a decrease of about 12 % compared to the BAU Baseline for 2050 (see Winther et al., 2017 for details). When comparing the simulations, where the polar diversion route is included, the results show that an HFO ban will lead to reductions of a few percent in the BC deposition in the Arctic; mainly along the shipping 510 routes (Fig 13d).
In the scenario describing a high growth in the shipping traffic (HiG), the contribution can increase to 5-15 % along the shipping routes as seen in Fig. 13e. Also in areas further away from the routes, a general increase in the contribution from shipping is seen. In an earlier study using a global model and a similar HiG scenario, Browse et al. (2013) found comparable increases in the contribution from shipping towards 2050 and they conclude that shipping could have a significant impact on 515 the albedo in this area. The deposition of BC on snow and ice decreases the albedo and increases the absorption of incoming solar radiation, which can lead to earlier melting (AMAP, 2011).

525
Globally, observations and modelling results indicate that radiative forcing (RF) induced by BC on snow and ice is highest in the mid-latitudes (Bond et al., 2013;Kang et al., 2020). While BC in the atmosphere can lead to a direct RF of +0.71 W m 2 , its semi-and indirect effects can be up to +0.23 W m 2 . Sand et al. (2013) showed the Arctic (>60 N) sources of BC leads to a https://doi.org/10.5194/acp-2020-1274 Preprint. Discussion started: 9 February 2021 c Author(s) 2021. CC BY 4.0 License. surface warming of 2.3 K, 1.6 K of which is attributed to the effect from BC deposited on snow and sea-ice. Sand et al. (2016) estimated that BC in the atmosphere and snow leads to an Arctic warming of 0.48 K. Both Sand et al. (2013) and (2016)  530 showed that per unit emissions, sources in the Arctic have a factor of two higher impact on the Arctic warming compared to sources outside Arctic. These findings imply that increased BC emissions from shipping activities in the Arctic can have a large effect on the Arctic climate even though these emissions are small compared to global emissions.
Overall, the BC emissions in the Arctic are project to increase towards 2050 in the current study. The SECA scenario will limit the BC emission by about 3% in the Arctic (North of 60N) in both traffic scenarios, while the HFO ban will have a higher 535 impact and limit the emissions by up to about 14% in the HiG traffic scenario (Winther et al., 2017). A HFO ban can thereby limit part of the emissions increase following the HiG traffic scenario, but not all.
The map of total nitrogen (N) deposition is given in Figure 14a  In a previous modelling study with focus on the Canadian Arctic, Gong et al. (2018) estimated the present day annual N deposition to be on the order of 0.2-1 kg N/ha within the Canadian sub-Arctic, and 0.05-0.2 kg N/ha over the Canadian high Arctic, which overall is similar to the levels estimated in the present study.
The shipping scenarios setup in the current study, only impact the NOx emissions through the development in ship traffic and 555 the new diversion routes. The contribution from shipping in 2050 without (Fig 14b) and with the new routes (Fig 14c), is seen to be slightly higher than for BC and more "widespread" due a lower lifetime of the N-components. The contribution is on the order of 1-10% without the new routes, whereas the contribution reaches up to about 15-20% along the new diversion route in the Baffin Bay, and hence along the coast of Ellesmere Island, Canada and Northwest Greenland. If the traffic develops along the HiG traffic scenario (Fig 14c), the contribution from shipping increases to more than 20% in large parts of the Arctic and 560 to even above 40% in the Baffin Bay and in part of the Arctic sea. The Arctic ecosystems are adapted to nutrient-poor conditions and critical loads for N depositions are hardly exceeded during present day conditions (Forsius et al., 2010). But as also outlined by Forsius and colleagues, the N cycle in ecosystems are very complex and highly sensitive to increasing N depositions that can lead to significance changes in e.g. inter-species relationships.

Discussion 565
Future projections and health assessments are inherently associated with uncertainties. In the following some of the uncertainties and limitations of the presents study are discussed.
The future developments in shipping traffic is in this study based on Corbett et al. (2010), which to our knowledge was the only dedicated shipping traffic forecast for the Arctic available, when our scenarios were developed. They assume a strong 570 growth of shipping in the Arctic areas. There are, however, several factors, which will have an impact on the popularity of the Arctic ship routes and hence on the future traffic growth. First is the extent of receding sea ice, which makes Arctic routes more manoeuvrable and viable for ships. Second factor involves the geopolitical situation, because local icebreaker assistance may be necessary for safe navigation in Arctic waters. Third are the expenses, fuel, personnel and insurance (Sarrabezoles et al., 2016) costs, which complicate the evaluations of economic viability of the Arctic sea routes. However, the assumed traffic 575 scenarios of Corbett are used here for consistency with other previous studies, and to study the magnitude of changes concerning the fuel restrictions in the scenarios.
We then combine the shipping scenarios with the ECLIPSE V5a (current legislation) land-based emission scenario to project future impacts on air pollution levels and human health. However, not only emission changes have an impact on future 580 developments in the air pollution levels. A changed climate will also impact future levels, both directly through altered transport patterns and changed temperature, precipitation and sunshine impacting chemical transformations and removal processes as well as indirectly through changed biogenic emissions and deposition due to changes in growing season of vegetation (Andersson and Engardt, 2010); an important sink of many reactive tropospheric trace gases such as O3. Numerous previous studies have been conducted to investigate the impact of climate change on air pollution levels in Europe, e.g. Colette 585 et al. (2015) summarizes the climate penalty of O3 in Europe. The studies are conclusive: although climate change until 2050 has an impact on near-surface O3 concentrations (e.g. Langner et al., 2012), particle concentrations (Hedegaard et al., 2013;Lacressoniere et al., 2017), and deposition of N (Simpson et al., 2014), the main factor of future evolution until the midcentury, is overall the emission changes. There are differences in the results for the current and future scenarios from the two CTMs. MATCH projects a lower exposure in general and the changes to the future differs. For example, MATCH projects a wider spread increase in PM2.5 in the Nordic area than DEHM. The fact that we are using two models should be seen as a benefit to illustrate uncertainties in the projections.
The main message is robust, but spatial patterns differ and are more uncertain. There may be various reasons for these differences, ranging from the higher resolution in the MATCH setup, differences in chemical scheme and differences in the 595 long-range transported (LRT) component from continental Europe. The LRT-component is weaker in MATCH, which partly can be due to differences in meteorological forcing, but also due to differences in parameterised deposition processes. It is expected that models with different physical descriptions and setups give some differences in the resulting air pollution distributions and the inclusion of several models hence provides valuable insight on CTM uncertainty. An ensemble of models usually performs better than individual models (Marecal et al., 2015), and an interval gives an indication of the uncertainties 600 involved. This has been used in other studies, e.g. HTAP/AQMEII (Solazzo et al., 2012a,b;Im et al., 2015a,b) and EDTRENDS (Colette et al., 2017;Ciarelli et al., 2019b) and ENSCLIM (e.g. Simpson et al., 2014;Soares et al., 2015).
As illustrated in Figure 11 and 12 the differences between the two models is also clearly seen in the %-contribution from shipping to the overall PM2.5 level and the estimated number of premature deaths attributable to shipping. This was also seen in a recent study focusing on the Baltic Sea region. Here three different CTMs were compared and significant differences were 605 found in the estimated contribution from shipping, which mainly was linked to differences in the schemes for inorganic aerosol formation (Karl et al., 2019).
The health assessment in the current study is based on the modelled air pollution maps with a resolution of 16.67 km x 16.67 km grid across the Nordic area. The estimated number of premature deaths related to air pollution can therefore be somewhat 610 underestimated as the high air pollution levels in urban areas with high population density will not be fully resolved at this resolution. National assessments of health effects for Denmark (see e.g. Ellermann et al., 2020) and the Nordic area (Lehtomaki et al., 2020) made with the same EVA system as applied here, but based on higher resolution air pollution modelling (1 km x 1 km with the Danish UBM model) points towards higher number of premature deaths than in the present study. Additionally, as described earlier, the EVA system employs a linear exposure-response relationship based on the recommendations of WHO 615 (WHO, 2013). However, recent studies show that the shape of the relative risk vs pollutant concentration can change from region to region, and may not be linear (e.g. Burnett et al., 2018). Some studies have shown that in low pollutant regions, such as the Nordic region, non-linear functions give higher negative impact from air pollution than linear functions, while in high pollutant regions, i.e. China/Africa, linear functions give the highest numbers (Im et al., 2018;Bauer et al., 2019).
Finally, it should also be noted that we in the projected health assessment keep the population data constant (at the 2015 620 number and distribution), even if previous studies have shown that e.g. an aging population in both the Nordic area and in Europe in general, will lead to a higher sensitivity to air pollution Tarín-Carrasco, submitted).
Our focus has been on the Nordic and Arctic region, where we e.g. find increased deposition of N and BC in the Arctic due to new shipping routes in the future. But these new routes also have the potential to limit the overall CO2 emissions compared to 625 longer routes (Corbett et al., 2010) and air pollution levels and negative health impacts could be lowered in other parts of the world. The study by Sofiev et al. (2018) has e.g. previously shown that especially the Asia population would benefit from ship emission reductions.
The full climatic effects of BC emissions are very complex. As e.g. described in Kühn et al. (2020) BC in the Arctic atmosphere 630 and on snow/ice can lead to a general warming of the climate. But part of this warming can be counteracted by other processes, as aged BC particle e.g. can impact cloud dynamic that have a cooling effect. It has been outside the scope of the current work to make an assessment of the different climate feedbacks related to BC in the Arctic. https://doi.org/10.5194/acp-2020-1274 Preprint. Discussion started: 9 February 2021 c Author(s) 2021. CC BY 4.0 License.

Conclusion
We have setup new global shipping emission scenarios including potential mitigation options either based on additional fuel 635 quality requirements (Heavy fuel Oli ban) or on an expansion of the existing ECA areas. The scenarios are considered in terms of two alternatives for the development in traffic: a Business As Usual (BAU) and a High Growth (HiG) traffic growth scenario and for the years 2015 (our base year), 2020, 2030 and 2050. Overall, the projected increase in traffic is to some degree counteracting the effects of technological developments and fuel requirements leading to different trends for the different components. When focusing on the Nordic area the projections display that Baseline NOx emissions will go down by 640 approximately 40% (HiG traffic growth) to 60% (BAU traffic growth) towards 2050 due to improved technology. The fuel consumption and NOx emission totals calculated for the SECA and HFO ban scenario equal the Baseline scenario. For SO2 the Baseline projections includes a decrease on the order of 40-50% in 2050 compared to 2015. Here the largest drop is seen from 2020 due to the global 0.5% Sulphur cap. In all scenario years for SO2, the calculated emissions for the SECA and HFO ban scenarios are close to 30% lower than the emissions calculated for the Baseline scenario within the Nordic domain. Due 645 to a different development in BC emission factors, the BC emissions increases are 20-70% in the Baseline for 2050 and only a small (1-2%) reduction is obtained in the future years in the SECA scenario in both traffic growth cases. For the HFO ban scenario in 2020 [2030,2050] with BAU traffic growth, the BC emissions are 9%[12%, 16%] smaller than the Baseline results, while slightly larger emission reductions are obtained in the HiG traffic growth scenario.

650
In combination with a scenario for the land-based emissions the two chemistry transport models DEHM and MATCH have been used to simulate the developments in the overall air pollution towards 2050 and also the contribution related to shipping alone. By using the health assessment model EVA, we estimate that for the Nordic area the number of premature deaths related to air pollution will decrease from approximately 9900 (9400-10.400) in 2015 to 7900 (7700-8200) in 2050. The range of the numbers given in brackets, represents the results for the two models and illustrates the uncertainties related to this kind of 655 assessments e.g. related to differences in the setup of the two CTMs. Changes in climate and population demography are disregarded here, where meteorological and population data for 2015 are used in all simulations.
Shipping contributes to 1-15% of the PM2.5 levels in the Nordic, with highest contribution along the shipping routes and in coastal areas. In terms of health impact, we estimate that shipping emissions leads to about 850 (560-1100) premature deaths during current day conditions (as a mean over the two models). With the 2050 Baseline BAU scenario this number will decrease 660 to approximately 600 (480-720) cases. Introducing additional mitigation options as assumed in the heavy fuel oil (HFO) ban scenario, the number of premature deaths attributable to shipping emissions will be on the order of 550 (440-670) in 2050. The SECA scenario will have less impact in the Nordic area. If the shipping traffic follows a high growth (HiG) path and no additional mitigation options are introduced, the negative health impact will increase, and the number of premature deaths will be on the order of 890 (700-1080) in 2050. In terms of health impacts in the Nordic area, the HFO ban BAU traffic scenario 665 can be regarded as the "best case", while the Baseline HiG traffic scenario must be regarded as the "worst case".
When moving the focus further towards north, we investigate the impacts of new potential ship traffic routes in the future Arctic and the effect of expanding requirements from current SECA areas as well as a HFO ban to the full Arctic region. The new diversion routes will make the shipping traffic a more important source for BC and N in parts of Arctic, especially along 670 the shipping routes. If the development in traffic follows the high growth path, the simulations show significant increases in the deposition. The mitigation scenarios will limit the contribution from shipping slightly, but the deposition is in general low in the Arctic.
This study addresses the topic, which was recently debated at the International Maritime Organization and reports the results 675 for various emission mitigation options for ships operating in Arctic areas. Health benefits of Arctic HFO ban and an Arctic https://doi.org/10.5194/acp-2020-1274 Preprint. Discussion started: 9 February 2021 c Author(s) 2021. CC BY 4.0 License.
SECA have been quantified, considering the health effects in Nordic countries. It is very likely that full benefits of fuel restrictions in Arctic areas will be delayed, because of the recent decision of IMO MEPC75 to relax the fuel requirements for some ships belonging to the Arctic fleets and postpone the full HFO ban until 2029. By setting up these new shipping emissions scenarios that follows on-going discussions for mitigation options and evaluate the impacts in terms of health impacts in the 680 Nordic region and depositions in the Arctic, this study can add to the science based evaluation of potential mitigation strategies for shipping emissions. It also adds to the recent work of the IMO Fourth GHG study (Faber et al, 2020) and updates regional Arctic emission scenarios and BC emissions from ships, considering the atmospheric transport of pollutants.