A comprehensive inventory of ship tra ffi c exhaust emissions in the European sea areas in 2011

Introduction Conclusions References

variation, the geographical distribution of emissions, and their disaggregation between various ship types and flag states. The total ship emissions of CO 2 , NO x , SO x , CO and PM 2.5 in Europe for year 2011 were estimated to be 131, 2.9, 1.2, 0.2 and 0.3 million tons, respectively. The emissions of CO 2 from Baltic Sea were evaluated to be more than a half (58 %) of the emissions of the North Sea shipping; the combined contribu-10 tion of these two sea regions was almost as high (96 %) as the total emissions from ships in the Mediterranean. As expected, the shipping emissions of SO x were significantly lower in the SO x Emission Control Areas, compared with the corresponding values in the Mediterranean. Shipping in the Mediterranean Sea is responsible for 39 and 49 % of the European ship emitted CO 2 and SO x emissions, respectively. In particular,

Introduction
The traffic activities of shipping in Europe are nowadays well known, as compared with vehicular traffic; this was not the case previously. The introduction of automatic vessel position reporting systems, such as the Automatic Identification System (AIS), have 25 significantly reduced the uncertainty concerning ship activities and their geographical 7460 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | distribution. Nowadays, all vessels larger than the 300 ton size limit globally report their position with few second intervals; this has resulted to an availability of information on ship activities at an unprecedented level of detail. The ship emission inventories, which are based on such automated identification systems, have several significant advantages over the previously developed approaches. Such inventories are based 5 on time-dependent, high-resolution dynamic traffic patterns, which can also allow for the effects of changing conditions, such as, e.g., marine and meteorological conditions (e.g., harsh winter conditions and sea ice cover) or weather routing.
Previous studies concerning the ship emissions in Europe have been based on statistics of cargo volumes (Schrooten et al., 2009), vessel arrival/departure times 10 (Whall et al., 2002), voluntary weather reports from ships (ICOADS, Corbett et al., 2007) or search and rescue services (AMVER, Endresen et al., 2003;Wang et al., 2007). None of these data sources is able to reflect the total ship activity with full flexibility of traffic activity and temporal changes. Inconsistencies can exist between geographical emission inventories and satellite observations of pollutants (Vinken 15 et al., 2014). Furthermore, important emission sources, like harbours have been often neglected from regional emission studies.
In this work, we have used the Ship Traffic Emission Assessment Model Johansson et al., 2013) which combines the vessel activity (AIS data) with vessel specific information of main and auxiliary engines. This allows the 20 determination of vessel specific emissions, which are based on the detailed technical information of fuel consuming systems onboard. Fuel type used during harbour stays or open seas will be determined from actual vessel activity and engine characteristics taking possible sulphur restrictions in specific regions into account. The fuel type assignment (residuals/distillates) is determined from technical specifications of ships' 25 engines which can provide a more realistic description of the use of various marine fuels than fleet wide adoption of residual fuels.
We present emissions for European sea regions, which are covered by the terrestrial network of AIS base stations. In general, European seas are relatively densely traf-7461 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ficked, especially in regions, in which intercontinental ship traffic intersects with busy short sea shipping routes. The vessel activity data from this area have been collected to operational Vessel Traffic Services center at the European Maritime Safety Agency. This centralized data archive allows one of the most comprehensive high resolution sources of vessel activity on a continental scale. The modelling approach of the present 5 study can be largely automated, which facilitates annual updates of large-scale ship emissions. This allows, e.g, for the inclusion of the impacts of policy changes, such as sulphur reductions, to be included in the emission inventories used in air quality applications.
The aim of this study is to present a comprehensive inventory of ship traffic exhaust 10 emissions for a number of contaminants (CO 2 , NO x , SO x , PM 2.5 , CO) in European sea areas, utilizing the STEAM ship emission model Johansson et al., 2013). A more specific aim is to geographically present and discuss the high-resolution spatial distributions of shipping emissions for selected species, and the shares of emissions in terms of the various ship types and flag states. We have 15 also identified a few tens of the highest emission intensities in the European sea and harbour areas; these regions contain the highest amounts of predicted shipping emissions of CO 2 within a radius of 10 km. We aim also to compare the numerical values of this new emission inventory with the corresponding values presented in some previous inventories on the emissions originated from European shipping.

Geographical domain and input datasets
This modelling approach uses as input values the position reports generated by the automatic identification system (AIS); this system is globally on-board in every vessel that weighs more than 300 tons. The AIS system provides automatic updates of the 25 vessel positions and instantaneous speeds of ships at intervals of a few seconds. For 7462 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | this paper, we used the AIS messages received by the terrestrial AIS network and provided by the European Maritime Safety Agency (EMSA). We extracted the data that corresponded to the year 2011; the data contained more than 10 9 archived AIS messages. The data has been collected from the terrestrial AIS base station network of the EU member states. The coverage of this network is illustrated in Fig. 1. 5 Most of the European sea areas are well represented in this data. However, the Arctic Ocean has not been included. Extensive open sea areas, such as the Atlantic Ocean, are also not completely represented, due to the limited reception range of the terrestrial AIS base station network. There are also spatial gaps of the data in the southernmost parts of the Mediterranean, especially near the northern African coastline. The data did 10 not include position reports from any of the African countries; however, the ship activity in this area is significantly lower than in the northern parts of the Mediterranean. This was shown with an independent investigation of satellite AIS datasets obtained from the Norwegian Coastal Administration (detailed results not shown here). The data from inland waterways in Europe has been included, but cannot be taken to fully reflect the 15 inland shipping, as the IMO SOLAS regulation does not require the use of AIS from these vessels.
The model requires as input also the detailed technical specifications of all fuel consuming systems on-board and other relevant technical details for all the ships considered. Such technical specifications were therefore collected and archived from various 20 sources of information; the data from IHS Fairplay (IHS, 2012) was the most significant source. The technical data was supplemented with material from several other companies and agencies. These included the following: Det Norske Veritas, Nippon Kaiji Kyokai, Bureau Veritas, Germanischer Lloyd American Bureau of Shipping, publicly available ship registers (such as the Korean, Norwegian and Russian ship registers), 25 ship owners and engine manufacturers. Fuel type was determined based on the properties of engines, such as power output, angular velocity and stroke type. The sulphur content was assigned based on the current regulations in European sea areas, such as the MARPOL Annex VI (IMO, 1998) and the EU Sulphur Directive.

7463
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | The technical specifications were collected and archived for more than 65 000 vessels that have an International Maritime Organization (IMO) number. This set of ships represents a majority of the global commercial fleet. In addition to these vessels, the AIS position reports were received from more than 35 500 vessels, for which the technical data could not be determined based on the information from classification societies, 5 such as the Lloyds Register. In addition to the IMO number, the vessel Maritime Mobile Service Identity (MMSI) code was used as a secondary key in searching vessel data from ship databases.
However, the vessel data was not received for a vast majority of vessels that transmitted the MMSI code (and no IMO number) in AIS data. An additional attempt to 10 identify these vessels with internet search engines using MMSI code was made for 5000 vessels, which had the largest fuel consumption. This revealed some potentially large vessels, but the impact of this step on overall CO 2 emissions was just over one percent. Clearly, the default method of assuming those vessels small, which do not transmit IMO registry number, introduces uncertainty to overall results, but the impact 15 is negligible.

The STEAM model and its application
The emissions presented in this paper were evaluated using the Ship Traffic Emission Assessment Model (STEAM). A brief overview of this model is presented in the following; for a more detailed description, the reader is referred to Jalkanen et al. (2009, 20 2012) and Johansson et al. (2013). This study does not introduce any refinement of the model.
The STEAM model was used to combine the AIS based information with the detailed technical knowledge of the ships. This combined information is used to predict vessel water resistance and instantaneous engine power of main and auxiliary engines. The model predicts as output both the instantaneous fuel consumption and the emissions of selected pollutants. The fuel consumption and emissions are computed separately for each vessel, by using archived regional scale AIS data results in a regional emission 7464 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | inventory. The STEAM emission model allows for the influences of the high-resolution travel routes and ship speeds, engine load, fuel sulfur content, multiengine set-ups, abatement methods and the effects of waves Johansson et al., 2013). The STEAM model includes a possibility to model some environmental effects on 5 ships, such as the effects of waves and the influence of sea currents. However, for simplicity these factors were not taken into account in this study. The waves increase fuel consumption and emissions, whereas the direct effects of the wind and sea currents can be negative or positive. In considering long time scales and extensive regions, the net influences of direct wind effects and sea currents are expected to be fairly small.

10
It would be possible also to use satellite-based AIS messages as input values of the model; however, for simplicity these were not used in this study, except for the above mentioned confirmation of lack of significant vessel activity in southern Mediterranean Sea.
The emissions of NO x were modelled as a function of crankshaft angular velocity 15 (revolutions per minute, rpm), according to the IMO Tier I and II rules (IMO, 2008). Tier I rule was applied also to those ships, which were built before 2000; this assumption can result in a slight underestimation of emissions originated from these vessels. The effects of emission abatement techniques were also modeled, and certified emission factors have been used whenever possible. The emission certificates were provided 20 by a group of ship owners, the emissions of these vessels had been measured by an accredited laboratory, in order to obtain a discount in the system of Swedish fairway dues. However, the vessels that were equipped with emission abatement techniques or had been subject to certified emissions represented less than 1 % of the ships included in this study. 25 We have included in the modelling most of the various engine setups, such as gas turbines, diesel electric and mechanical power transmission, nuclear vessels and sailboats. We allowed for the fact that the operation of a shaft generator is possible for vessels, which have been indicated to have geared drives or power take-off systems.

7465
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | The modelled values of engine loads also take into account multi-engine setups and load balancing of operational engines.
The STEAM model simulates the required power of the main and auxiliary engines, by determining the required power level set up that corresponds to the speed value in the AIS messages. All ships are modelled individually, and the modelling takes into 5 account the differences in hull form, propeller efficiency, shaft generators and auxiliary engine usage. The sulphur content of the fuel has been modelled explicitly for each vessel and its engines. We have allowed for the sulphur reduction techniques and the influences of the regulations regarding fuel sulphur content in various regions and during various time periods .

10
In cases, in which more detailed information could not be obtained from engine manufacturers, the Specific Fuel Oil Consumption (SFOC) has been modeled based on the methods in the second IMO GHG report (Buhaug et al., 2009). The SFOC is modelled as a function of engine load. In the model, low engine load levels can increase SFOC up to 25 %. Operating engines outside their optimal working range (without de-rating) 15 will lead to increased SFOC and emission factors. The emissions of particulate matter, sulphate and water are modeled as a function of the fuel sulphur content. All vessels have been treated as single displacement hulls; catamarans and hydrofoil vessels were not separately modeled. The currently modelled pollutants are NO x , SO x , CO, CO 2 , EC, OC, ash and hydrated SO 4 . The model can also be used to generate vessel-specific 20 emission inventories, and to predict the amount of consumed fuel. The transport work (cargo payload) is described as the product of the weight of cargo transported and the distance travelled (commonly in units of ton km). In this work we adopted the scheme reported by the second IMO GHG study (Buhaug et al., 2009). 25 We have evaluated in more detail the emissions from locations with an especially high emission intensity, which we refer to as shipping emission "hotspots". The STEAM model has been executed on a resolution of approximately 2.5 km × 2.5 km in the EU-7466

Detection of locations with the highest shipping emissions of CO2
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | region. For the evaluation of hotspots the resulting CO 2 emission grid has subsequently been evaluated using the following rules: 1. The sum of emissions in the vicinity of each grid cell has been calculated within a radius of 10 km (such a domain contains approximately 44 closest grid cells).
2. The sum (if high enough) along with center coordinates are placed in the list of 5 top 30 highest ranking CO 2 hotspots.
3. The first and second steps are repeated until each cell in the emission grid has been once the candidate emission hotspot.
This analysis also indicates the areas with the highest ship fuel consumption, whether this occurs in harbour areas or along shipping lanes.
10 3 Results and discussion

Summaries of total emissions and their geographical distribution in Europe
A compilation of computed emissions, payloads, numbers of ships and distances travelled has been presented in Table 1. The geographical distribution of ship CO 2 emis- and Bosporus, and in the Danish Straits. In addition, there are localized high amounts of CO 2 emissions near several major ports. These ports include, in particular, the ones in the Netherlands (e.g. Antwerp, Rotterdam and Amsterdam), Gibraltar, St. Petersburg and some ports in the UK, Germany, Italy and Spain. The relative geographical distribution of the shipping emissions is similar also for the other modelled compounds, 5 and those results have therefore not been presented here. The international cargo traffic contributes significantly to the emissions at the most densely trafficked shipping lanes; a prominent example is the ship route in the Mediterranean Sea that extends from Suez Canal to Gibraltar. The route patterns of passenger traffic are different; these occur more frequently via shorter routes. For example, there 10 are a lot of routes between the islands in Greece and the mainland, and between Italy and the islands of Sardinia, Corsica and Sicily. There is a dense network of shorter passenger vessel routes in numerous sea regions in the Mediterranean. The routes of cargo and passenger traffic intersect also in several regions of the Baltic Sea and the North Sea. For example, in the English Channel passenger traffic takes mainly place 15 across the channel, whereas most of the cargo routes are aligned along the Channel.
We have also analysed the areas that have the highest CO 2 shipping emissions in Europe. These areas were defined as circular domains with a radius of 10 km. We have presented the results for 30 areas that had the highest estimated emissions. These domains are called in the following as the emission hot spots. The results have been 20 presented in Table 2 and in Fig. 3. The combined CO 2 emissions of these 30 hotspot areas correspond to approximately 7 % of total CO 2 emitted by ships in Europe.
The area including the Netherlands and the English Channel has the highest density of these hot spots; there are in total ten domains in these regions amongst the top 30 shipping CO 2 hotspots in Europe. There are also hot spots at numerous locations 25 in the Mediterranean, some in Germany, and a few in the Baltic Sea region. Harbour areas dominate the list of highest CO 2 hotspots. Besides harbour locations, some shipping lanes and some major coastal cities are associated with very high CO 2 emissions. Clearly, a major part of emissions in these coastal cities are also due to harbour activi-ties. Several of the largest harbours in Europe reside in the Netherlands and along the English Channel.
In some sea regions, busy shipping traffic is focused in geographical bottlenecks with high CO 2 emissions; prominent examples of these in southern Europe are the strait of Gibraltar, the channel between Malta and Sicily, and the Bosporus Strait. However, the 5 emissions originated in the Bosporus Strait are not well represented in the data, as the data from the Turkish national AIS network were not available for this study. The data from Greece and Romania include part of vessel activity from this area, but not a sufficient coverage.
Emissions of CO 2 originated from Mediterranean shipping were found to be about 10 36 % of the total CO 2 emissions from shipping. Emissions from ships in the North Sea and the Baltic Sea constituted approximately one quarter and one seventh of the total emissions of CO 2 from shipping, respectively. The emissions of NO x from the ships in the Mediterranean Sea (1 186 000 tons, calculated as NO 2 ) are almost as high as those in the Baltic Sea (359 000 tons) and the North Sea (664 000 tons) combined.

15
Emissions of NO x from other areas considered in this study are of the same magnitude as contribution from the North Sea shipping. The share of the Mediterranean sea traffic is even larger in case of the SO x emissions, compared with the corresponding emissions for CO 2 and NO x . The emissions originated from the other sea areas except for the three specifically 20 mentioned three sea regions (Baltic Sea, North Sea and Mediterranean Sea) have also been reported in Table 1 and Fig. 4. These areas include the western parts of the Black Sea, Canary Islands, Celtic Sea, Barents Sea and North-East Atlantic Ocean. The emissions from shipping in these other regions were estimated to produce almost one quarter of CO 2 ; however, this value is probably an underestimation, as the coverage of 25 AIS reception in remote sea areas, such as the Atlantic Ocean, is incomplete. These results have obvious policy implications. Reductions of ship exhaust emissions in areas with high emission levels and a surrounding dense population is likely to yield major health benefits (e.g., Corbett et al., 2007). However, policy changes for 7469 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | reducing shipping emissions may have significant cost impacts (e.g., Johansson et al., 2013;Kalli et al., 2013), which necessitates thorough assessments of both the costs and the benefits. The identified emission hot spots, especially those which are in the vicinity of major cities, are prime candidates for enhanced emission control measures. The low fuel sulphur requirement of the EU directive has already addressed some as-5 pects of this issue.

Analysis of emissions in terms of the flag state and the ship type
The AIS signals include a Maritime Mobile Service Identity (MMSI) code that contains information that specifies the flag state of the ship. We have selected 20 flag states that had the highest total fuel consumption in Europe in 2011, and evaluated their annual We have allocated the emissions to IMO registered (referred here also as "large") and unidentified (referred to as "small") ships in Table 1, as the IMO registered ships constitute most of the commercial marine traffic. According to the values in Table 1, the contribution of unidentified vessels is only 2.2 % of the total CO 2 emissions, although the number of such small vessels is 43 % of all vessels. The significance of the con-5 tribution of unidentified ships to total emissions varies in terms of pollutants; for CO it is the largest (4 %). The unidentified ships travel 7 % of the distances travelled by all vessels. For some countries, such as the Netherlands, Germany, France and Sweden, the share of large vessels is less than one third of the total number of ships. This may indicate the different practices in including the small vessel movements in overall traffic 10 image of various countries. It is very likely that small vessel traffic is underestimated by AIS, because for these vessels AIS voluntary, in contrast to the requirements for large vessels. In this context, the Dutch fleet is an extreme case, in which only 13 % of 7530 vessels are considered large. In the Dutch case, the share of CO 2 emitted by small vessels is 43 %, which is the largest fraction for all of the studied fleets. Already   15 in Finland, there are over 190 000 boats (Trafi, 2014) and 525 Finnish vessels were picked up by AIS in Europe. Clearly, the representation of small vessel traffic activity is incomplete in AIS.
The descriptions of the technical details for small vessels in the emission inventory are limited. These are significantly less accurate than the corresponding descriptions 20 for large vessels, for which the engine setup and technical data are readily available. Model results for the fuel consumption of small vessels are further complicated by an incomplete inclusion of the activities of small vessels; a fraction of the small vessels do not carry AIS equipment on board.
The shares of emissions and payload (i.e., transport work) for various ship types 25 have been presented in Fig. 6. A comparison of CO 2 emissions and payload reflects the energy efficiency of various ship types. We used the approach described by Buhaug et al. (2009). The unit emissions (the mass of CO 2 emitted, divided by transport work) are lowest for the tanker class (7.3 g ton −1 km −1 ), slightly higher for con-7471 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | tainer (10.2 g ton −1 km −1 ) and cargo vessels (10 g ton −1 km −1 ), and significantly higher for passenger traffic (175 g ton −1 km −1 ). However, the values for passenger traffic are not directly comparable, as the above mentioned transport work of passenger traffic has been calculated as a function of cargo capacity, which does not take the number of passengers into account. There are large variations of unit emissions between vari-5 ous vessels in the cargo class, as this class includes both dry bulk and palletized cargo vessels, for which there are large differences in the use of their cargo carrying capacity.

Seasonal variation of the emissions
There were clear seasonal variations in the emissions of all pollutants; the variations in case of CO2 have been presented in Fig. 7. For example, the emissions of CO2 in June 10 are 30 % larger than the corresponding values in January. During the summer months (June, July and August), both the numbers of passenger vessels and small vessels is the largest, especially in the Mediterranean Sea. This is mainly caused by the increased recreational travel; in summer the number of small vessels is at maximum in all sea areas. The emissions of container ships are also higher in the summer than  Table 3. The five more general level categories (cargo, container, tanker, passenger and other) have been divided to more detailed categories. The division of vessel activity to operational modes (cruising, maneuvering and hoteling) has not been predetermined; it has been defined by vessel activity data. Based on AIS data, it is possible to determine these explicitly, which will significantly decrease the large uncertainties that habe previously been associated with vessel activities.
The shares of fuel used by the main engines have also been presented in Table 3, these have also been evaluated by the model. The amounts of fuel used in main and auxiliary engines depend not only on vessel specifics, but also its operational profile.
However, there is a major uncertainty in the predictions of the fuel consumption of the auxiliary engine, as the use of an auxiliary engine varies greatly, even for ships of the same type. The use of auxiliary power cannot be determined from tank tests of ship resistance, unlike the power needed for propulsion, for which various theories exist for performance prediction. In this study, we have used the methodology presented 10 previously Johansson et al., 2013). This method combines the information on cargo capacity and auxiliary engine power profiles. However, there are also other modelling approaches, which are based on extensive vessel boarding programs (Starcrest, 2013), local knowledge and pre-assigned contributions (Dalsoren, 2009).

Comparison of the predictions of various emission inventories
The comparison of the numerical results of various European-scale emission inventories can be challenging, as pointed out, e.g., by EEA et al. (2013). The main reasons for this are that the methodologies and various modelling selections used for evaluating shipping emissions vary substantially in various published studies. E.g., the various 20 studies may define differently the geographical domain, and some studies address only international ship traffic.
The current work reports emissions for the year 2011. Significant reductions were therefore in force regarding the sulphur content of marine fuel in the North Sea and the Baltic Sea area, as well as the requirement for low sulphur fuel in EU harbour 25 areas. The effects of these regulations were included in the current work, and it is therefore not possible to directly compare the predicted SO x and PM 2. 5

emissions with the corresponding values during previous years. Changes in international regulations 7473
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | also concern NO x , but to a lesser extent, as the IMO Tier II NO x limits for marine diesel engines affect all engines built since 1 January 2011. The ships constructed after this date will have to conform to Tier II NO x requirements (15 % less NO x produced when compared with Tier I engines), but such new ships constitute only 3 % of the fleet of IMO registered vessels in this study. Significant policy changes are expected to be 5 implemented in 2015, regarding the sulphur content of marine fuel. The emissions of NO x for the Mediterranean Sea reported in this work are lower than in the EMEP inventory, qualitatively the same result was reported by Marmer et al. (2009). Marmer et al. also concluded that their methodology yielded lower SO x emissions than the corresponding EMEP values. The prediction of the STEAM inven-10 tory for the Mediterranean shipping in the case of NO x is about 40 % of the correspond-ing value in the EMEP inventory. Vinken et al. (2014) used satellite observations of NO x from the OMI instrument to constrain top-down emissions from ships. The study area of this study (defined by AIS coverage illustrated in Fig. 1) and Vinken et al. (2014) are the same (N,W,S boundaries are same), except that the domain used by Vinken than either of the above mentioned two inventories. The reported total NO x emission for all European sea areas in this study is 2.94 million tons, which corresponds to 0.9 million tons of reduced nitrogen (N). This is close to the corresponding value reported by Vinken et al. The annual SO x emissions reported in this study for various sea regions are 88, 154 and 613 thousand tons for the Baltic Sea, North Sea and the Mediterranean, respectively. The corresponding SO x emissions of the EMEP inventory for the above 10 mentioned sea areas are 82, 192 and 1367 thousand tons. For the Baltic Sea, the inventories are approximately in agreement; there is a larger difference (20 %) between the inventories in the North Sea. However, the predicted emissions of SO x in the Mediterranean Sea by the two inventories are drastically different. The SO x emissions predicted in this study for the Mediterranean are less than a half of the corresponding 15 values in the EMEP inventory. As discussed above, this was also the case for the NO x emissions.
The reasons for such major differences in the predictions of these two inventories could be caused, for example, by the neglect of the impacts of relevant legislation, such as the EU sulphur directive (2012/33/EC). This directive limits the sulphur content of 20 marine fuels to 0.1 % (by mass) in harbour areas and to 1.5 % (by mass) for passenger vessels on a regular schedule. It is possible that not all passenger ships comply with the requirement of 1.5 % fuel sulphur content, as assumed in the STEAM model. However, a possible non-compliance by a fairly small fraction of ships would explain only a minor portion of the differences between the STEAM and EMEP inventories. More information 25 on the compliance with EU regulations can be obtained either during Port State Control checks, or via relevant compliance monitoring schemes (Balzani et al., 2014;Berg et al., 2012;Beecken et al., 2014a, b;Pirjola et al., 2014).

7475
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | It is not possible to perform a similar satellite-based comparison for SO x , due to the technical limitations of currently available satellite instruments; these cannot accurately determine ship emitted SO x near the sea surface. Such instruments can detect stationary SO 2 sources that have an emission higher than approximately 70 kilotons (Fioletov et al., 2013); however, this value is too high for the shipping lanes in Europe. 5 The inventory of Cofala et al. (2007) includes an estimate for ship CO 2 emissions, which is based on the same methodology as the EMEP inventories. According to Cofala et al. (2007), the predicted CO 2 emission in 2010 from ships in the Mediterranean is approximately 76 million tons (obtained by a linear interpolation between the values in 2000 and 2020), which corresponds to 24 million tons of fuel burned. In this work, the 10 Mediterranean shipping was responsible for 51 million tons of CO 2 emitted.
The differences in PM 2.5 emissions between this work and Cofala et al. (2007) in all sea areas are less than 20 %. A large variation could be expected in the PM 2.5 emissions predicted by the various methods, due to the substantial variability of experimentally determined emission factors and the differences in PM 2.5 sampling methods (e.g., Jalkanen et al., 2012). Clearly, the PM 2.5 emissions are associated with the SO x emissions and the sulphur content of the fuel, as SO 4 is one of the main constituents of atmospheric PM 2.5 .
The range of European shipping emissions of CO 2 reported in the review by EEA et al.

Conclusions
The comparison of emitted pollutants with existing ship emission inventories revealed that there are major differences between the estimates of the various inventories for the emissions of ships sailing the Mediterranean Sea, whereas the results were better in agreement for the North Sea and the Baltic Sea regions. The NO x , SO x and CO 5 emissions evaluated in this study for the Mediterranean Sea were roughly half of the corresponding values in EMEP and IIASA inventories, whereas differences in PM 2.5 emissions were less than 20 % between these different inventories. Satellite observations using the OMI instrument also indicated smaller annual emissions of NO x in the Mediterranean, compared with the EMEP inventory (EMEP, 2014). The reasons for these deviations should be investigated further and confirmed with independent experimental datasets, as these can have significant policy implications concerning health and environmental impact assessments.
Despite the wide geographical extent, the ship emission data can also be segmented in terms of the various properties of vessel categories or individual vessels. This makes 15 it possible to classify the emissions using several criteria. The disaggregation of ship emissions into individual vessels on a fine temporal resolution also allows fine resolution air quality and health impact assessment studies. A specific advantage of an inventory based on individual vessel data is that it facilitates comparisons with experimental stack measurements. 20 According to this study, the vessels carrying an EU flag were responsible for 55 % of CO 2 emissions in the EU, whereas the states with flags of convenience and other states constitute the remaining share. The CO 2 hotspot mapping indicate that the English Channel constitutes a large source of ship emitted CO 2 , both from harbour areas and densely trafficked shipping lanes. 25 The emissions from ships have a clear seasonal variation; the emission maximum occurs during the summer months. This concerns especially passenger traffic, but also containerships have the same seasonal pattern. However, the emissions originated 7477 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | from oil tankers and other cargo ships do not have a clear seasonality. Temporal vari-ation of ship emissions has mostly been neglected in previous emission inventories, due to inherent limitations of the activity data used as a basis for these inventories. Seasonal variations can be of the order of 30 %; these features should therefore be included in emission and health impact assessments. 5 The current work also facilitates studies of ship energy efficiency, as all emissions and fuel data are generated on the ship level. There were substantial differences be-tween fuel burned and transport work carried out by various ship types. The unit emis-sions were the lowest for the oil tankers and largest for passenger vessels. However, the description of transport work of passenger vessels currently considers cargo oper-10 ations and does not completely cover passenger cargoes.
The availability of the shipping activity data for research can be a challenging task; however, there are several options for data acquisition. Data collected by maritime au-thorities are rarely available for research purposes. However, there are networks of volunteers maintaining AIS base stations; activity data can therefore either be shared 15 or is commercially available. Most satellite AIS datasets are available from commercial service providers, but also national space programs may provide access to these. Auto-matic AIS data collection facilitates annually updated ship emissions in the EU waters; however, the coverage area should be expanded to the North-East Atlantic Ocean. This could be done with the inclusion of other activity data sources, such as, e.g., the 20 satellite AIS data, which could be used to extend the AIS coverage, e.g., to fully cover the EMEP modelling domain.

7481
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Table 1. Emissions and shipping statistics in the SafeSeaNet area in 2011. The section "All ships" includes also emissions from unidentified vessels. "IMO registered" refer to commercial ships with specified IMO number. In the section "top flags", twenty countried are presented that have the largest contributions to emitted CO 2 . Together these flag states represent over 76 % of the total CO 2 emitted. The NO x emissions have been calculated as NO 2 . The details of ship type aggregation is further explained in Annex II.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |  Table 3. Summary of average operational features of some selected ship types. The first column indicates the aggregated ship type, whereas the second column contains a more detailed description of vessel type. The time spent in each operation mode (cruising, maneuvering, hoteling) is indicated by the next three columns as percentages. "ME of Fuel" refers to the fraction of fuel used in main engines from total fuel consumption.