According to the International Maritime Organization (IMO), more than 90 % of world trade is carried by sea, since maritime transport is the most cost-effective way to move mass goods and raw materials (International Maritime Organization, 2015). However, maritime transport is an important source of air pollutants on the global (Wang et al., 2008) and European level (Eyring et al., 2010) and can contribute significantly to local air quality (AQ) problems in European harbour cities of all sizes (Viana et al., 2009). Globally, ships are known to emit 5–7×109 kg yr-1 of nitrogen oxides (NOx), 4.7–6.5×109 kg yr-1 of sulfur dioxide (SO2) and 1.2–1.6×109 kg yr-1 of particulate matter (PM) into the atmosphere (Smith et al., 2014; Corbett and Koehler, 2003; Eyring et al., 2005); 70 % of these emissions occur near coastlines and therefore contribute to air pollution in both coastal areas and harbour cities (Andersson et al., 2009; Corbett et al., 1999; Endresen, 2003). Ships emit NOx mainly in the form of NO, which is quickly converted to NO2; thus atmospheric NOx from shipping is mainly in the form of NO2 (Eyring et al., 2010). The contribution of international shipping to the air quality over European seas reached up to 80 % for NOx and SO2 concentrations, up to 25 % for particles with a diameter of 2.5 µm and less (PM2.5), and up to 15 % for ozone (O3) in hotspot areas along coastlines in 2005 (EEA, 2013). In the North Sea region, the relative contribution of international shipping to NO2 concentration levels ashore close to the sea can reach up to 25 % in summer and 15 % in winter (Aulinger et al., 2016), while Karl et al. (2019c) showed average shipping contributions of 40 % over the Baltic Sea and 22 %–28 % for the entire Baltic Sea region. In the entire Baltic Sea region the average contribution of ships to PM2.5 levels is in the range of 4.3 %–6.5 %, (Karl et al., 2019a).

However, little is known about the impact of ship emissions in harbour cities of the North Sea and Baltic Sea region. Even if emissions of in-port ships account for only a few percent of the global emissions related to shipping (Dalsøren et al., 2009), they can have an important impact on local AQ in harbour cities due to additional emissions from manoeuvring, mooring and diesel-powered activities at berth, such as lighting, cooling, heating and sanitation (Meyer et al., 2008). Viana et al. (2014) performed a literature review with the aim of characterising and quantifying the contribution of the maritime transport sector to air quality degradation along European coastal areas. The reviewed studies agreed on the relevance of ship emissions in coastal areas for PM, NOx and SO2 and identified a large spatial variability, with maximal contributions in the Mediterranean Basin and the North Sea. On average, shipping emissions in the coastal North Sea region contribute 7 %–24 % to NO2 annual mean and 3 %–5 % to PM2.5 annual mean concentrations in the North Sea, while in the Mediterranean PM2.5 from shipping contributes 4 %–20 % (Viana et al., 2014).

Only few studies investigated the impact of in-port ship emissions on the AQ in harbour cities of the Baltic Sea. Saxe and Larsen (2004) showed the impact of local shipping activities in Copenhagen, Denmark, which connects the ship traffic between the North Sea and Baltic Sea. NOx from shipping was exceeded 200 µg m-3 of NOx and caused values of 50–200 µg m-3 over several square kilometres of central Copenhagen, while PM and SO2 contributed with insignificant mass concentrations of PM to populated areas near the harbour (Saxe and Larsen, 2004). Pirjola et al. (2014) measured particulate and gaseous emissions from ship diesel engines with different after-treatment systems using a mobile laboratory inside the harbour areas in Helsinki and along the narrow shipping channel near Turku, Finland, and concluded the need for additional regulation of shipping particulate emissions beyond controlling the fuel sulfur content. Also in Helsinki, Soares et al. (2014) investigated the impact of emissions from ship traffic in the harbours of Helsinki and in the surrounding area on concentrations and exposure, identifying a contribution of about 3 % to PM2.5 concentrations by shipping activities.

A more recent study by Ledoux et al. (2018) in the North Sea port of Calais showed the direct influence of in-port shipping on SO2, NO2 and PM10 average concentrations, at 51 %, 15 % and 2 %, respectively, and with substantial concentration peaks synchronised with departures and arrivals of ferries. In the harbour city Hamburg, Ramacher et al. (2018) identified maximum relative contributions from shipping to total NO2 and PM2.5 concentrations, at 23 % and 3 % in January and 45 % and 16 % in July 2012 and with the highest concentrations located in the port area of Hamburg. A study in preparation (Tang et al., 2019) modelled local NO2 shipping contributions to air pollution in the urban area of Gothenburg of about 14 % and a regional NO2 contribution of up to 41 % on average to the annual mean, indicating the same importance in controlling local shipping emissions as, for example, road traffic emissions, while SO2 and PM2.5 contributions are negligible.

Exposure to air pollution can lead to asthma, respiratory and cardiovascular diseases, lung cancer, and premature deaths according to the World Health Organization (WHO, 2006). Corbett et al. (2007) showed that shipping-related PM emissions are responsible for approximately 60 000 cardiopulmonary and lung cancer deaths annually, with most deaths occurring in coastal regions of Europe, eastern Asia and southern Asia. An update of this study shows that despite implemented regulations, low-sulfur marine fuels will account for 250 000 deaths annually in 2020 due to an increase in transport by sea (Sofiev et al., 2018b). Approximately 230 million people are directly exposed to these shipping emissions in the top 100 world ports (Merk, 2014). The large majority (95 %) of Europeans living in urban environments are exposed to levels of air pollution considered dangerous to human health. The average contribution of shipping emissions to the population exposure from primary PM2.5, NOx and SOx is 8 %, 16.5 % and 11 %, respectively, across Europe (Andersson et al., 2009). While exposure to PM2.5 was estimated to be a leading cause of the WHO environmental burden of disease in six selected European countries (Hänninen et al., 2014), the relationship between NO2 and health is scientifically not as well founded as for PM2.5 (WHO, 2006; Heroux et al., 2013). However, NO2 is usually regarded as an indicator of other pollutants, and long-term residential exposure to NOx is moving into focus due to rising evidence for severe health effects on the respiratory system (WHO, 2016; Wing et al., 2018; Hamra et al., 2015) and as risk factor for myocardial infarction (Rasche et al., 2018). In terms of exposure to shipping emissions, NO2 was found to be consistently associated with total non-accidental mortality and specific cardiovascular mortality in the Baltic Sea harbour city Gothenburg (Stockfelt et al., 2015). Thus, exposure to air pollution caused by shipping activities in harbour cities needs to be reduced and emissions regulated.

Regulations for the prevention of air pollution from ships was introduced in the Marine Pollution Convention (MARPOL) Annex VI by the IMO and entered into force in 2005. Many countries have ratified this protocol, particularly for limiting NOx and SO2 emissions from ships. The coastal areas of the North Sea and the Baltic Sea have been classified as sulfur emission control areas (SECAs), where the sulfur content in marine fuels has been limited to 0.1 % from 2015 on. Moreover, the European Union introduced a requirement limiting the sulfur content in fuels used by ships at berth to 0.1 % in 2010. The European Environment Agency (EEA) therefore estimated the decrease in SO2 ship emissions to be 54 % between 2000 and 2010, and a further decrease is expected from 2020 onwards due to changes in technology and global regulations (EEA, 2013). It is also expected that this will lead to a decrease in emissions of PM2.5. Nevertheless, NOx emissions from international maritime transport in European waters are projected to increase and could be equal to land-based sources by 2020. In order to reduce NOx emissions from shipping, an NOx Emission Control Area (NECA) will be implemented in the North Sea and Baltic Sea on 1 January 2021. The goal is to decrease nitrogen oxide emissions from maritime transport by 80 % compared to present levels in the long run. Besides this, an additional reduction in PM2.5 is expected in the future due to less NOx-induced secondary organic aerosol (SOA) formation, which would lower the ship-related PM2.5 by 72 % in 2040, compared to the present, while it would be reduced by only 48 % without implementation of the NECA (Karl et al., 2019c). Despite these regulations to reduce SOx (SECA) and NOx (NECA) emissions in Europe, ship traffic is still the least regulated sector in Europe compared to other types of anthropogenic emission sources such as road traffic, industrial sources, power generation or residential heating. Hence, shipping emissions are increasing in terms of the relative weight of shipping emissions to the total anthropogenic emissions on the regional and local scale in Europe (EEA, 2013). Taking into account the projected increase in maritime transport due to growth of the global-scale trade (Lloyds Register Marine, 2014; EC, 2012) as well as the simultaneous increase in population growth and urbanisation in coastal areas (Neumann et al., 2015), it is necessary to come up with pollution prevention efforts for ports in harbour cities.

The objective of this study is to identify the impact of emissions due to local shipping activities on air quality and population exposure to concentrations of NOx in three major Baltic Sea harbour cities: Rostock (Germany), Riga (Latvia) and the urban agglomeration of Gdańsk–Gdynia (Poland). To identify the impact of local shipping activities on AQ, an urban-scale chemistry transport modelling (CTM) system, was set up for the selected Baltic Sea harbour cities. Besides city-specific emission inventories for land-based emission sources, spatially and temporally high-resolution shipping-emission inventories have been modelled and applied. All study areas are located in the SECA, and the study was performed for 2012 conditions, when the sulfur content in marine fuels was limited to 1 % in the region and 0.1 % for ships at berth. Therefore, and because of the decreasing importance, we excluded SO2 from the study focus. We analysed concentrations of NO2, O3 and PM2.5 for 2012 conditions and evaluated these with local measurement network data of each harbour city. The impact of local shipping activities on urban air quality was determined with the perturbation method (zero-out scenario runs). We focus on the impact of local in-port shipping on the air quality in harbour cities while considering that the influence of ocean-going shipping on the Baltic Sea is beyond the scope of this study. Based on the simulated concentration fields, dynamic population-weighted outdoor exposure to NO2 and PM2.5 for all urban domains was calculated in different microenvironments using a newly developed generic exposure modelling approach based on publicly available data. This study mainly focusses on NO2 exposure, taking into account the high contributions of local shipping activities to NO2 in other harbour cities, the growing importance of NO2 as an indicator for health effects and the usage of NO2 as an indicator for health effects due to other pollutants.

To our knowledge, this study is the first one investigating the impact of emissions from local shipping activities on air pollutant concentrations and population exposure in Baltic Sea harbour cities since the 2010 commencement of the 0.1 % sulfur fuel requirement in harbours (European Parliament Directive 2005/33/EC), using a CTM system with high spatial and temporal resolution.

Section 2 of this paper describes the model and data set-up, introducing the urban-scale CTM EPISODE-CityChem in Sect. 2.1 and describing the set-up of each urban domain in Sect. 2.2 and 2.3. This is followed by the description of local emission inventories and their application in the CTM system (Sect. 2.4 and 2.5). Finally, a new generic approach to achieve outdoor exposure for different microenvironments will be introduced in Sect. 2.7. In Sect. 3, the simulated concentrations will be evaluated (Sect. 3.1), and total as well as ship-related concentration distributions of NO2 and PM2.5 will be presented for the city domains (Sect. 3.2). This is followed by the analysis and illustration of exposure results due to total and ship-related concentrations (Sect. 3.3). Section 4 discusses the exposure results with respect to the novel approach for generic dynamic population activity and is followed by conclusions in Sect. 5.

A CTM system with the EPISODE-CityChem model (Karl et al., 2019b; Karl and Ramacher, 2018) to simulate present day urban concentrations of NO2 and PM2.5 as well as the contribution of shipping activities to urban air quality was set-up for the Baltic Sea urban areas of Rostock, Riga and Gdańsk–Gdynia. City-specific meteorological fields, regional boundary conditions, and land-based emission and shipping-emission inventories have been gathered and modelled. The contribution of present shipping emissions to the modelled concentration of air pollutants was determined from the difference between “base” runs, which include all emissions, and “no-ship” runs, which exclude emissions from ship traffic (zero-out method). The concentration results are then evaluated and used to model dynamic population-level exposure in different microenvironments for each city (Fig. 1).

We evaluate and present results for simulated concentrations in the Baltic Sea harbour cities Rostock, Riga and the urban agglomeration of Gdańsk–Gdynia, focussing on NO2. For each city, we performed runs with and without shipping to determine the effect of local shipping on NO2 concentration levels as well as population-level exposure to NO2. Besides the exposure of all ME due to total concentrations and shipping activities, we analyse the exposure to shipping-related concentrations in ME_home, ME_work and ME_port.

We developed a generic approach to model population activity for exposure calculations (Sect. 2.6.2) to bridge the gap between static residency population numbers and very dynamic but specific population-activity data derived from surveys or gathered with mobile devices, which were both not available in the harbour cities of this study. Thus, we used generic data, and a set of assumptions which introduces spatial and temporal uncertainties in the exposure calculation, in addition to those of the applied CTM system. Exposure is the cross-product concentrations and population density. Therefore, all uncertainties that play a role for either of them have to be considered.

In terms of uncertainties within the applied CTM system that produce concentrations, the range of uncertainty can be identified by comparisons with measurements. The evaluation of measurements (Supplement Sect. SII; Table SII-2) shows a range of -26 % to +4 % for bias in annual measured vs. modelled NO2 concentrations at different stations in Gdańsk–Gdynia. In Rostock, there are higher underestimations of -56 % to -32 %, while in Riga the range is -60 % to -4 %. High underestimations in all cities mainly occur at or near traffic stations. Matthias et al. (2018) showed that the biggest uncertainty in CTM simulations is mostly due to emission data, which are a key driver and a major source of uncertainty in atmospheric chemistry transport models. Especially in urban areas, concentrations of NOx, for example, depend linearly on the local emissions. In emissions, modelling the amount, temporal distribution and spatial distribution of emissions is often uncertain and thus has a high sensitivity. For example, non-methane volatile organic carbon (NMVOC) emissions for ships in port areas were not available as output from STEAM. This restriction led us to estimate NMVOC emissions based on the carbon monoxide (CO) emissions provided. Products of incomplete combustion, like CO and NMVOC, are difficult to estimate because these emissions are very sensitive to engine load changes, engine control (mechanics and electronics), service history and fuel injection. Very little experimental information is available concerning NMVOC emissions from modern marine engines at a sufficient level of detail, and NMVOC emission factors based on measurements performed decades ago may not represent NMVOC emissions from modern marine diesel engines accurately. The lack of detailed measurement data is probably because emission measurement standards (ISO 8178) do not require NMVOC classification but report NMVOCs as total hydrocarbons instead, which makes evaluation of NMVOC species very difficult, hindering the CTM description of secondary aerosol formation with consecutive modelling effort. Nevertheless, in this study we used a CO–NMVOC emission ratio of 1.4, which is representative of emissions from auxiliary and main engines at an engine load of 70 %–80 % (Aulinger et al., 2016), to calculate NMVOC emissions from STEAM CO emissions in Rostock, Riga and Gdańsk–Gdynia. These uncertainties in emissions will translate to uncertainties in NOx concentrations due to the chemistry of ozone, NOx and volatile organic compounds (VOCs), which represent one of the major uncertainties in the field of atmospheric chemistry, especially in urban areas (Sillman, 1999). Another example of uncertainties due to emissions is traffic emissions, which play a major role in the overall urban emissions. The exposure in the ME_traffic is very likely to be under-predicted in Rostock, and probably also in Gdańsk–Gdynia and Riga, due to the following reasons. In Rostock, the traffic emission modelling is not based on actual traffic density data but was only spatially disaggregated based on road type classification and corresponding factors, which represent a national average. While in Riga and Gdańsk–Gdynia the traffic emissions are based on traffic counts, they also do not account for all the effects of traffic congestion, slowing down of traffic in certain locations and streets and the effects of idling, and the deceleration and acceleration of vehicles. Traffic congestion can increase emissions in streets during rush hours (Gately et al., 2017; Requia et al., 2018; Smit et al., 2008). The evaluation at traffic stations has also shown that NO2 was modelled with a high negative bias, although EPISODE-CityChem was run with the activated Street-Canyon-Module and therefore included treatment for dispersion in street canyons. The ME_port shows, in all urban domains, lower exposure to NO2 compared to ME_work. This is mainly due to the detailed allocation of people directly employed by the port to the ME_port, which are distributed to the comparably large port areas.

Besides emissions, also meteorological fields and regional boundary conditions are crucial inputs for correct CTM simulations. Nevertheless, Karl et al. (2019a) proved good agreement with measurements for the regional boundary conditions as calculated with CMAQ, and the performance of the meteorological module of TAPM shows very good agreements with measurements. Therefore having correct emissions is the highest priority in terms of improving the concentrations of NO2, which then will linearly improve the results of exposure calculations.

In terms of uncertainties within the population activity, which is the second part of the cross product to calculate population exposure, there are four major factors in the developed dynamic population-activity approach that need to be considered: the size of the population, the temporal distribution of the population, the spatial distribution of the population and the application of infiltration factors for different microenvironments. In the following, these will be discussed in detail.

In this study, the population in each urban domain was derived from a population density map, valid for the European Union, instead of national or municipal population counts. This introduces biases in terms of total population numbers and the spatial distribution of people in their home environments. We have shown that the total population number derived from population density maps in this study is altered by 9 %, 12 % and 8 % for Rostock, Riga and Gdańsk–Gdynia, respectively, compared to population counts valid for the cities of interest (Table 3). Nevertheless, the advantage of this approach is the detachment from municipal boundaries or statistical zones, which are often used in population counts; these could lead to blind spots in research domains, which exceed municipal boundaries or statistical zones. A future development will be the integration of “Population estimates by Urban Atlas polygon”, which is a Copernicus Land Monitoring Service product in preparation (https://land.copernicus.eu/local/urban-atlas/population-estimates-by-urban-atlas-polygon, last access: 7 July 2019). Besides this, we uniformly distribute the derived total population with UA2012 land use classifications to spatially disaggregate the total population. A future development of this approach will be the integration of population density maps as a proxy in the distribution of population to the home environment to integrate a weighted distribution of population to the UA2012 land use classifications. This will also lead to a clearer distinction of areas which are allocated to work and home environments at the same time.

We considered the UA2012 land use classification “continuous urban fabric” for both the home and work environment to have a 30 % and 70 % share due to the description of the UA2012 classification, which includes central business districts. To check the impact of this assumption, we changed the applied split of 30 % ME_home and 70 % ME_work into two tests: (1) 50 % ME_home and 50 % ME_work and (2) 70 % ME_home and 30 % ME_work in the Gdańsk–Gdynia domain. By changing the distribution of ME_home to 50 %, the contribution of ME_home to the total annual gridded mean increases by 0.7 %, while the total annual exposure increases by 1.8 %. Changing the distribution of ME_home to 70 % increases the contribution of ME_home to the total annual gridded mean by 1.2 %, while the total annual exposure increases by 3.2 %. In the same tests, the ME_work is changed to 50 % and 30 %, which results in a decrease in the ME_work contribution to the annual grid mean by 0.3 % and 0.5 %. Therefore, we evaluate the uncertainty of the applied split of 70 % ME_work and 30 % ME_home in the UA2012 land use class “continuous urban fabric” to have limited influence on the overall exposure results. Nevertheless, due to a lack of information about specific population activity in any of the urban domains, we cannot validate our assumptions in distributing population to the MEs and the connected UA2012 land use classifications. Based on the descriptions of the UA2012 land use classifications, we matched the best-fitting microenvironments but still introduced uncertainties, e.g. in the category “Industrial, commercial, public, military and private units”, which contains not only work environments but also non-work environments, e.g. schools, universities, museums or churches. When it comes to ME_work, we also considered the UA2012 class “continuous urban fabric” to mainly constitute indoor work environments in city centres and the UA2012 classes “Industrial, commercial, public, military and private units”, “Mineral extraction and dump sites” and “Construction sites” to account for mixed indoor and outdoor work environments. In future studies, a clearer distinction of the UA2012 categories in terms of numbers of workers and indoor–outdoor classification should be done; e.g. the number of workers in the category “Mineral extraction and dump sites” could be taken from city-specific statistics, and the category could be classified as an outdoor-only environment. Besides this, we considered the amount of commuters, taken from municipal statistics, in the ME_work and ME_traffic and thus accounted for people who are additionally exposed to pollution in traffic and work environments. The consideration of commuters in Gdańsk–Gdynia leads to a 4 % higher total annual population exposure and a 20 % higher annual exposure in ME_work. For a better distribution of the ME_work and ME_ other we plan to use the “point-of-interest” feature in OSM data as a proxy in future studies, which potentially allows for a better distribution between work and other activities and the identification of very busy city centres.

Besides uncertainties in the spatial distribution, we also introduced uncertainties regarding the temporal distribution, which is based on a temporal profile for the city of Helsinki (Soares et al., 2014). We adapted this profile and then added features which we found to appear in other European cities, such as traffic rush hours in the morning and evening. However, such a generic profile is not able to reflect the actual population activity throughout the day. Moreover, there are regional and national differences, e.g. the siesta in Mediterranean countries. Still this pattern emulates a dynamic population, which moves between environments and is exposed to different levels of pollution throughout the day. In comparison to traditional approaches, which assume people to be at their residence (home address) all the time, we believe that this approach is beneficial in particular for cities in European regions where data from surveys or positioning data from mobile devices are missing. We compared population exposure to NO2 based on our dynamic population-activity approach, with population exposure being based on a static approach to analyse the effect of a population moving in space and time on calculated population exposure. In this test, we allocated the total population all day (100 % of the time) to the home environment (ME_home) in order to simulate a static approach. The dynamic activity considers people “moving” diurnally between different MEs. Moreover, we ran simulations with and without infiltration factors to test the effect of outdoor concentrations infiltrating to indoor environments in the static and dynamic approach. The comparison between the static and the dynamic approach without the consideration of IFs (i.e. indoor air concentrations are the same as in the surrounding outdoor air) shows a decrease in total annual exposure in each city (Table 6). Therefore, the consideration of diurnal dynamic activity in different MEs leads to an increase in total population exposure. This is an effect of people moving to areas which are more polluted and, additionally, the effect of commuting inside or outside of the city.

Another assumption made in calculating exposure in different environments is the infiltration of outdoor pollutant concentrations into indoor environments. We have considered the influence of outdoor air pollution on the total population exposure. However, we have not addressed indoor sources and sinks of pollution, although indoor sources such as tobacco smoking, cooking, heating and cleaning, for example, might cause additional short-term concentration maxima in indoor environments. We have also assumed that infiltration is temporally constant, changing only with the seasons. Nevertheless, we took into account the infiltration of outdoor pollution into indoor environments (ME_work and ME_home) using IFs. To check the impact of IFs for the indoor environments, we increased and lowered the applied IFs in ME_work and ME_home in the city of Gdańsk. An increase in the IFs by 0.1 in both MEs leads to a linear increase of 10 % in ME_home and ME_work, respectively. The total exposure increases by 10 %. When it comes to the relative contribution of each ME to the total exposure, the relevance of ME_home increases to 57 % (+2.5 % points) and ME_work increases to 14 % (+0.4 % points). A similar decrease in IFs by 0.1 shows the same changes with the opposite sign. Thus, the impact of the adapted IFs on exposure in environments that are mostly indoor environments has a significant influence on the total exposure results with a linear response of the total exposure to changes of the IF. The MEs ME_other, ME_traffic and ME_ port are considered outdoor environments. When it comes to the ME_other, which is an outdoor-only environment in this study, the exposure is heavily dependent on the season due to more people spending their time outdoors in summer than in winter. This has not been considered in this study but should be taken into account in future studies. Nevertheless, the ME_other areas in the city centre are mainly green urban areas and are therefore potentially areas of high exposure in summer. In general, the applied IFs for NOx as derived from Borrego et al. (2009) represent an average of infiltration measurements in South Korea (Baek et al., 1997), Hong Kong (Chau et al., 2002) and the United Kingdom (Dimitroulopoulou et al., 2006). Thus, in future studies it is desirable to derive and use IFs, which are representative of the city-specific building infrastructure, to account for different air-intake techniques, building structures or different ventilation manners. Better parametrisation to derive more representative IFs could be derived from a combination of the EU Buildings Database, the UA2012 and climate data.

Taking into account all uncertainties and possibilities for improvement, we promote this approach for European regions in which actual data on population activity are not available, with the overall goal of improving existing exposure calculations for policy support. Nevertheless, the highest uncertainties and therefore possibilities to improve the results of the exposure calculations are as follows:

a better representation of emission inventories in CTM,

We have presented population exposure to total and shipping-related NO2 outdoor concentrations in different microenvironments of the Baltic Sea harbour cities Rostock and Riga and the urban agglomeration of Gdańsk–Gdynia. The population exposure was calculated as a product of (1) hourly varying surface concentrations of NO2 simulated with a global-to-local chemistry transport model chain and (2) a newly developed generic approach to account for dynamic population activity in European cities.

We simulated the surface concentrations with the urban-scale CTM EPISODE-CityChem using regional boundary conditions from CMAQ simulations and land-based and ship emissions and meteorological fields for 2012 in Rostock, Riga and Gdańsk–Gdynia. The evaluation of modelled versus measured NO2 time series showed good spatial correlations and slight underestimations of annual NO2 but an overall applicable performance for studies in urban areas with a FAC2 value above 0.3 at all stations of each domain. The simulated results show the contribution of NO2 from shipping to overall air quality as being 22 % for Rostock, 11 % for Riga and 16 % for Gdańsk–Gdynia.

We developed a generic dynamic approach to account for population activity in European urban areas which is applicable for exposure calculations. Our approach aims at filling the gap between traditional approaches of exposure calculations, which are based on static population counts at residential addresses, and approaches that take into account individual activities as derived from surveys or individual GPS data. Due to missing surveys and individual GPS data in the research domains of this study, we combined existing publicly available data to follow state-of-the art exposure modelling approaches in four steps. At first, we split the total population of each urban domain into several microenvironments (home, work, traffic, other and port). Second, we distributed these microenvironments to matching land use classifications of the Urban Atlas 2012. Third, we temporally distributed the total population to the different microenvironments diurnally for weekdays and weekends, adapted from existing diurnal patterns in other European cities. Fourth, we applied infiltration factors for indoor environments to account for outdoor concentrations infiltrating indoor environments. Following this approach, it is possible to compile gridded datasets containing the number and spatial distributions of a city's population for every hour in a diurnal cycle in each defined microenvironment. For this study, we generated these grids with a grid resolution of 100 m, following the resolution of the simulated surface concentration.

In the exposure calculation, we focussed on exposure to NO2 because the ship influence was shown to be high and the regulations for NOx emission reductions will propagate slowly into the ship fleet. Moreover, NO2 from ships adds to other local sources and therefore creates problems of obeying AQ directive targets of annual mean NO2. Besides this, outdoor NO2 pollution is a health concern with lot of recent attention by the WHO.

The relative contribution of each microenvironment to total NO2 exposure is highest for the home environment, at 59 %, 54 % and 55 % in Rostock, Riga and Gdańsk–Gdynia, respectively. Although the home environment has shown that it is very sensitive to applied infiltration factors, the vast amount of people spending their time at home during the day makes the home environment the most important environment in terms of exposure to outdoor NO2. When it comes to the influence of local shipping activities, shipping contributes 13 %, 6 % and 4 % to NO2 exposure in all microenvironments in Rostock, Riga and Gdańsk–Gdynia. The shipping contribution mainly focusses on MEs near the port in all cities. MEs, which are close to the port areas, can be influenced by shipping, with up to an 80 % contribution in Rostock and Riga and up to a 50 % contribution in Gdańsk–Gdynia. The lower contributions in Gdańsk–Gdynia are due to NO2 concentrations from shipping transported towards the open sea with the predominating southwesterly winds, while in Rostock and Riga the home and work environments north of the port are mainly affected from shipping for the same reason. The differences in relative contributions from shipping are determined by the magnitude of shipping activities in relation to activities in the rest of the domain and the domain size. The contribution of shipping in the port environment is considerably higher, at 46 %, 44 % and 26 %, respectively. Nevertheless, the port environment accounts for less than 1 % of the total exposure in all domains.

In general, the applied approach for exposure modelling is capable of showing the diurnal variation in population activity and therefore diurnal exposure in different microenvironments, although we focussed on total annual population exposure in this study. By introducing dynamic population activity instead of static population activity, the total exposure in Rostock, Riga and Gdańsk increases and therefore illustrates the need for considering dynamic population activity in exposure studies. In addition, we demonstrated the importance of microenvironment- and region-specific infiltration factors in considering outdoor concentrations infiltrating indoor environments. The lack of city-specific activity profiles, workplace addresses and infiltration factors introduces the biggest uncertainties in this study. In future studies we plan to improve the spatial allocation of population by applying population density maps in the spatial disaggregation of people in the home environment and by applying OSM points of interest as well as sector statistics on workers. Therefore, a better differentiation of infiltration factors in the work environments appears to be feasible. Moreover, we plan to integrate parametrisations for infiltration factors, which will take into account public national data on building structures and building regulations as well as climate data. When it comes to the traffic environment, we also aim to integrate region-specific measurements of outdoor-to-indoor concentration ratios. Besides these efforts, further studies to test the impact of different emission sectors, such as traffic or industry, in different microenvironments are planned.

The developed approach applied for the first time for generic dynamic population activity for calculating exposure to surface concentrations is superior to traditional static approaches and can be transferred to other cities in Europe, since no need for local activity profiles is involved. Although we used a global-to-local chemistry transport model chain, the presented generic dynamic population calculation can also be used with the surface concentration field created with other methods. Therefore, we promote this approach for European regions in which specific population-activity data derived from surveys or gathered with mobile devices are not available, with the overall goal of improving existing exposure calculations for policy support and providing the basis for health effect studies.