[1] A reliable and up-to-date ship emission inventory is essential for atmospheric scientists quantifying their impact and policy makers implementing regulations and incentives. However, significant differences are apparent among the various global ship emission inventories. The amounts emitted from ships to the atmosphere are largely determined by the bunker fuel consumption. Corbett and Koehler [2003] have a model estimate that is ∼100 megatons (Mt) higher (∼289 Mt), compared with a recent estimate by Endresen et al. [2003]. The main reasons for the discrepancy between the model estimates are identified in this study, and the effect of the differences in input data are discussed and quantified. This study also substantiates our previous suggested estimates that correspond with updated total bunker sale statistics. [2] The global bunker fuel sales and the figures reported by Endresen et al. [2003] are all significantly lower than those reported by Corbett and Koehler [2003]. The world marine bunker sale statistics for 2000 indicate a world bunker fuel consumption of around 180–190 Mt (see Appendix A). Of this, ships in international trade are responsible for 140–150 Mt. The Energy Information Administration (EIA) total bunker figures correspond well with the International Energy Agency (IEA) sales, except for a few important countries (EIA, International petroleum information, in “Crude oil imports and exports, with most countries and world total 2000,” available at http://www.eia.doe.gov/emeu/international/petroleu.html#IntlTrade; IEA, International bunker statistics year 2000 and 2001, available at http://data.iea.org) (see Appendix A). Endresen et al. [2003] modeled the year 2000 fuel consumption by the main engines as 144 Mt for the total cargo world fleet (∼45,000 vessels), separating on vessel type and size categories. We also presented a year 1996 total bunker estimate for the whole world fleet of 170–200 Mt (∼105,000 vessels), separating national and international fuel. Our estimate is supported by unpublished model results provided by R. Meech (personal communication, 2004). Corbett and Koehler [2003] modeled the year 2001 bunker consumption as 289 Mt, covering the cargo (203 Mt), the noncargo (45 Mt), and the military fleet (41 Mt). They have also made uncertainty analyses, illustrating some 190 Mt as a lower bound and ∼350 Mt as an upper bound. [3] To clarify the main reasons for this significant difference between the fleet modeling results, we have calculated the fuel consumption by means of the model reported by Endresen et al. [2003] and the input data given by Corbett and Koehler [2003]. The fuel consumption by the main engines for cargo ships (including passenger vessels) above 100 gross tonnage (GT) is then found to amount to 216 Mt. This corresponds to the amount reported by Corbett and Koehler [2003] for the same segment (203 Mt) and indicates that the models give corresponding results with common input data. However, their estimate for the cargo and passenger fleet includes 52,222 vessels, compared to ∼46,000 vessels greater than 100 GT reported by Lloyd's Register of Shipping (LR) [2000]. A fuel consumption of 203 Mt is 50 Mt higher than estimated by Endresen et al. [2003] with alternative operational profiles depending on ship type and size. The difference corresponds to the modeled fuel consumption for all oil tankers and general cargo ships [Endresen et al., 2003]. The main source for the observed deviation between the estimates is the assumed average number of operational hours while cruising at sea for medium and smaller vessels (which dominate by number) and the average engine load. Endresen et al. [2003] separated the activity profile on vessel size categories. Ships less than 5000 deadweight tons (Dwt) were assumed to have 4000 operational hours per year, while the larger ships above 100,000 Dwt were assumed to have 6000 operational hours per year. Ships between 5000 Dwt and 100,000 Dwt were assumed to have 5000 operational hours per year. The average utilization of main engines, while in operation, was assumed to be 70% for all ships, independent of size. Corbett and Koehler [2003] assume that the cargo ships are in operation 74% of the year (6500 hours) with 80% engine utilization. The tracking studies presented below do not support their assumed profiles, especially for small and medium-sized vessels (dominating by number). [4] The world cargo fleet (oceangoing) is dominated by a number of vessels less than 6000 Dwt (assuming 3000 GT is 6000 Dwt). For instance, ∼13,000 general cargo vessels are in this size range [LR, 2000]. Corbett and Koehler [2003] claim that the general cargo fleet dominated by small vessels stands for 22% of the energy demand (main engines) by the world fleet total and 31% of the world cargo and passenger fleet. This is in conflict with modeling results presented by Endresen el al. [2003, Figure 4] that clearly illustrated that medium and large vessels dominate the fuel consumption inventory. This deviation contributes to the higher total fuel consumption estimates and may be explained with their assumption of 6500 hours in use for the main engines for small vessels, while Endresen et al. [2003] assume 4000 hours. [5] Statistics Norway holds detailed information for the ∼400 smaller vessels between 100 and 3000 GT in coastal trade in Norway [Statistics Norway, 2000a, 2000b] (available at http://www.ssb.no/emner/10/12/40/nos_sjofart/arkiv/nos_c582/nos_c582.pdf and http://www.ssb.no/emner/10/12/40/nos_sjofart/nos_c633/nos_c633.pdf, respectively). The largest cargo ships (500–3000 GT) are typically in service (including port time) for 300 days, decreasing to around 240 days for the smallest categories (100–199 GT). This illustrates that service time decreases significantly with decreasing vessel size (or trade), mainly caused by the higher number of days laid up. This is also illustrated by Institute of Shipping Economics and Logistics (ISL) [2001], which shows that 206 of the 299 laid-up vessels were in the lowest size category (300–4999 Dwt). The annual time used in ports normally increases with decreasing vessel size. Operational data for ships in short sea, coastal, and regional trade indicates that ∼60% of the service time is in port [Norway, 1994]. We then only have 100–120 days at sea. Note that this estimate is based on Norwegian statistics for cargo vessels of 100–3000 GT and may not be representative for the world fleet. However, Norway has a very long coast, and the operation pattern and vessel size may be comparable with regional international traffic (e.g., North Sea trade). [6] From the individual operational profiles in the ship movement database Seasearcher (Lloyd's Maritime Information Services, http://www.seasearcher.com/), the annual number of port calls varies from, typically, 20 for very large cargo vessels up to 100 or higher for small and medium vessels, depending on trade. [7] Data submitted to the International Maritime Organisation (IMO) [United Kingdom, 2002] for time in port indicate great variability between ship types with an average of 1.17 days in port for all vessels (Table 3). Whall et al. [2002] reported that time in port varies between 0.46 and 3.8 days (median values), depending on vessel type. An average of these median values is calculated to be 1.27 days. Assuming 100 port visits and 1–1.3 days on average per port visit as representative for the small vessels (100–3000 GT) gives us ∼100–130 days in port annually. With 240–300 service days [Statistics Norway, 2000a, 2000b] and 100–130 days in port, the number of days at sea for the smaller vessels is probably in the range of 100–200 days (2400–4800 hours). We acknowledge great variability and uncertainty in the data. However, the Corbett and Koehler [2003] assumption of 6500 hours in use for the main engines for small vessels seems significantly too high. [8] Automated Mutual-Assistance Vessel Rescue System (AMVER) holds detailed voyage information (daily reports during a voyage) (E. Carroll, U.S. Coast Guard, personal communication, 2002) for ∼7100 medium and large cargo vessels mainly larger than 3000 GT in international trade. The AMVER data show that the main bulk of reporting frequencies is found to be in the interval of 100–250 days at sea, for cargo vessels, decreasing at lower and higher size intervals, neglecting the category of 1–50 days (Table 1). The AMVER data illustrate that the reporting frequency variability is reduced and the average time at sea is increased with ship size. The activity profiles presented below (and the AMVER data) indicate that on average, vessels spend around 220 days at sea with variation upward to ∼290 days depending on size. The standard error of ∼70 days illustrates vessels that report less than 150 days, report infrequently during a voyage, or do not report for the voyages shorter than 24 hours. We have assumed that all these vessels underreport, and the average profiles are then calculated by removing all vessels that report less than 150 days (Table 1). The average number presented in Table 2 illustrates some 60 days of variation between vessel types and up to 100 days more time at sea for large vessels compared to small-sized vessels. Clearly, size and type dependencies have to be considered when performing fleet modeling (see also Table 3). The data in Tables 1 and 2 indicate that an average of 6500 hours (270 days) at cruising speed as assumed by Corbett and Koehler [2003] is a high estimate. The AMVER fleet average profile at sea for different vessel sizes corresponds well with the proposed operational profiles suggested by Endresen et al. [2003] (see above). However, vessel type dependencies have to be included. The AMVER average reporting frequencies for very large crude oil carriers (VLCCs; ∼150,000 GT) are in the same range (Table 2) as the average 248 days per year reported for 453 VLCCs tracked in 1991 [Wijnolst and Wergeland, 1997]. Uncertainty and limitations related to the AMVER data are discussed by Endresen el al. [2003]. [9] A study that followed 1580 cargo vessels during a period of 16 months supports the above findings and illustrates significant variations in operation profile related to different vessel types/sizes and type of trade [United Kingdom, 2002]. They reported a mean turnaround and total sea time by vessel type, as shown in Table 3. They also reported mean turnaround times by different main bulk carrier sizes, ranging from 38 hours (minibulker) to 120 hours (Panamax). The ship turnaround time is the duration of the vessel's stay in port and is calculated from the time of arrival to the time of departure. Table 3 illustrates large variation in the voyage cycles. On average, 29% of the in-service time is related to the noncruising mode (assuming two turnarounds per voyage), where the main engine normally is not in use. This distribution is supported by a tracking study reporting the annual time fraction in port to be 37% for vessels operating in worldwide trade [Liberia, 1996]. However, this will largely depend on vessel type/size and trade. For instance, the annual time fraction in port is reported to be 60% for 605 Norwegian vessels operating in European waters [Norway, 1994]. In addition, detailed activity data for several ships operating in short sea areas (less than 10,000 Dwt) show that time in port constitutes 60–70% of the year [SAFESHIP, 2004]. [10] It is difficult to identify reliable and detailed statistics for nonservice time. Several sources report tonnage laid up, but little information on incidents and ship waiting, storage, and repair time is available. On the basis of the operational pattern for 453 VLCCs, the average number of days not in service was found to be 76 days per year [Wijnolst and Wergeland, 1997]; 32.4 days were reported for long- and short-term storage, 4.9 days laid up, 7.3 days waiting, 6.8 days related to incidents, and 24.7 days for repair. The waiting time is important to consider for cargo vessels. For example, Fremantle Ports [2003] (available at http://www.freport.wa.gov.au/about/report/docs/Yearend30June3020011.pdf) report that 20% of the bulk vessels waited outside the port for more than 2 days. [11] The average engine load, which is the ratio of actual power output to rated output based on maximum continuous rating (MCR: engine output power available for long periods without stop), will significantly influence the modeled fuel consumption. In open sea, the ships will usually run the engines on 80–85% MCR [LR, 1995; Environmental Protection Agency (EPA), 2000] (the latter is available at http://www.epa.gov/otaq/models/nonrdmdl/c-marine/r00002.pdf). However, periods with slow cruise (typically 10–40% MCR), port manoeuvring (typically 10–20% MCR), and ballast cruise reduce the average engine load (depending on trade and vessel type). Normally, ship speed and fuel consumption increase with increasing engine load. Operational data illustrate that ballast voyages at the same speed use significantly less bunker (typically, a reduction of 20% for large vessels), compared to normal cargo voyages [Wijnolst and Wergeland, 1997; Frontline Management AS, Vika Oslo, Norway, Technical and operational information for Frontline vessels, available at http://www.frontline.bm/fleetlist/index.php3; Clarkson Research Studies, 1998]. In addition, depending on the market situation, vessels operating in the spot market may slow steam in ballast condition. Bulk carriers, tankers, and general cargo vessels cruise at 14–16 knots, while reefers, RO-ROs, and container ships cruise at 20–22 knots [EPA, 2000; Cooperative Programme for Monitoring and Evaluation of the Long Range Transmission of Air Pollution in Europe/Core Inventory Air Emissions (EMEP/CORINAIR), 2002]. However, there is a wide range of cruising speeds within ship types [EPA, 2000; Lloyd's Register, 2002]. Speed variations were reported for 453 very large crude oil carriers tracked in 1991 [Wijnolst and Wergeland, 1997]. The reported average speed was 11 knots, but the variations were substantial, ranging from 10 to 13.6 knots. Assuming 10–12 knots as slow cruising speed, and the percent power required by the cube of the ratio of 12 knots to actual cruise speed, only 14% of MCR for container ships and 40% of MCR for bulk carriers are used. For maneuvering, the actual power applied will be lower. This affects the assumed annual average power rating for the main engines and will reduce it below 85% MCR, depending on type of trade (length of voyages, restricted waters, etc.), bunker prices, and freight rates. Also taking into account that cargo vessels sail in ballast close to 50% of the time [Wijnolst and Wergeland, 1997], our best average estimate is 70% MCR. This corresponds to the average weight load factor of the test cycles defined for main propulsion in the IMO technical code, when applied for verification of compliance with the NOx emission limits in accordance with regulation 13 of Annex VI [International Maritime Organisation (IMO), 1998]. [12] Corbett and Koehler [2003] report a fuel consumption for the noncargo fleet that is 100% higher than the estimate provided by Endresen et al. [2003]. We find a large discrepancy between the simplified activity profiles assumed by Corbett and Koehler and activity data reported for offshore vessels, large and medium fishing vessels, and tugs [Statistics Norway, 1996, 2000a, 2000b; SAFESHIP, 2004; Eidesvik, Technical and operational data for supply vessels, provided by K. Sandaker, Norway, 2004]. For the fishing fleet of ∼23,000 vessels, Corbett and Koehler [2003] assumed 6500 hours (270 days) in operation. However, Statistics Norway [1996] reports 162 days (20–25-m length) to 235 days (25–45-m length) days in operation (at sea) for Norwegian fishing vessels on the basis of activity data. These segments of the world fishing fleet represent by numbers more than 80% of the vessels in the fleet. Statistics Norway [2000b] reported, for tugs operating in Norwegian waters, 119 days with tugging at sea and 64 days with port towing services. There are some 9300 tugboats in the world fleet [LR, 2000]. The AMVER noncargo vessel voyage data (Table 1) illustrate that the number of days at sea is found to be less than 200 days for ∼90% of the ships. This clearly illustrates that on average, 270 days at sea is a very high estimate. [13] Noncargo vessels commonly have redundant engine capacity and allocated engine power for service operations. For example, tugs and ice class vessels have extra power installed that is only partly utilized on a yearly basis. Although the energy demand in the navy fleet may be significant, the actual bunker fuel consumption will be lower because of several very large nuclear-powered navy vessels [Adcock and Stitt, 1995; Haze Gray and Underway, World Navies Today, a database of all the world's navies and naval ships, by A. Toppan, available at http://www.hazegray.org/worldnav/]. Adcock and Stitt [1995] report that more than 30% of the large aircraft carriers are nuclear powered. An estimate of 41 Mt for the navy fleet as claimed by Corbett and Koehler [2003] seems to be unrealistic, considering the variable activity levels for the national navy fleets (e.g., Russia) and the fact that many vessels are in standby (laid-up) mode for military contingency [European Union (EU), 2000] (available at http://europa.eu.int/comm/enterprise/maritime/maritime_industrial/ship_scrapping_study/ship_scrapping_study_21feb01.pdf). Modeling the 759 submarines included in the navy fleet, of which 276 are nuclear powered, is also a problem. Endresen et al. [2003] argue that the navy fleet consumption is of the order of 4–5 Mt. This estimate is based on size characteristics for the fleet and assumed operational profiles. The significant deviation compared with Corbett and Koehler [2003] for this fleet segment can explain some of the discrepancy in the overall estimate for fuel consumption. Present estimates are uncertain and should be given attention in future research. [14] The main operation modes vary a lot depending on vessel types. For example, offshore and service vessels often operate several days at sea in the dynamic position mode only utilizing part of the available power (or engines). Data for the two offshore supply vessels “Viking Dynamic” and “Viking Energy” show that the ships are 30–40% of annual time in dynamic position mode, using 2 out of 4 engines (50% of installed power) with ∼20% load on each engine. In transit mode (at cruising speed), representing 30–40% of the year, these ships normally use 2 out of 4 engines, with 85% load on each engine (Eidesvik, Technical and operational data for supply vessels, provided by K. Sandaker, Norway, 2004). For other supply vessels, detailed activity data show that they are only at sea sailing at cruising speed for 41% of the year [SAFESHIP, 2004]. Noncargo fleet activity varies a lot depending on geographical area, vessel type, and size and needs to be better analyzed and described if the accuracy of models is to be improved. We claim that a fleet model based on installed engine power, without engine and fuel type adjustments, and with activity data from commercial trade will greatly overestimate the fuel consumption. [15] A detailed study of the main input data used by Corbett and Koehler [2003] to model bunker oil consumption shows that adopting 6500 hours in average operation for all ship types and size categories will result in overestimates of the bunker consumption. Our study shows that available operational data indicate a strong dependency on ship type and size, with the average number of operational hours decreasing from 6000 hours for large vessels to 4000 hours for small vessels. In addition, we argue here that the assumed engine power rating of 80% MCR for the 270 days assumed to be at sea is very high, especially for noncargo vessels. The main reason for the significant differences in estimated fuel consumption can be explained by engine load factor and operational profiles applied. We firmly believe that our estimates for fuel and operational profiles [Endresen et al., 2003] are representative. We find that our estimates are supported by detailed operational data and updated bunker inventories. However, vessel type and size dependency needs to be better analyzed and described if the accuracy for fleet segments is to be improved. For comparison of modeling results, it is important to have a common definition of world fleet segments. Some of the deviations between the fleet segments may be explained by differences in fleet numbers applied. We provide evidence here to suggest that the model of Corbett and Koehler [2003] results in an overestimate that may lead to an exaggerated view of the impact of shipping on the environment as well as wrong fleet performance characteristics. [23] We would like to acknowledge Doug Horton, Elissa Carroll, and Travis Hessenauer of AMVER and Kjell Sandaker of Eidesvik (Norway) for providing traffic and operational data. We would also like to acknowledge the reviewers and Per Olaf Brett at DNV, for improving the paper. [16] All fuel supplied to ships engaged in international operations is, irrespective of the flag of the carrier, referred to as “international marine bunkers.” The revised 1996 Intergovernmental Panel on Climate Change (IPCC) [1997] guidelines requested countries to estimate emission from international bunker fuel separately and to exclude these emissions from national totals. The world fleet of ∼105,000 oceangoing vessels [Endresen et al., 2003] will use international and national/domestic marine bunkers, depending on the type of trade. It is therefore difficult to distinguish between national consumption and consumption related to international trade (subject to IMO regulations). This “problem” is discussed in several studies [Endresen et al., 2003; Skjølsvik et al., 2000; EPA, 2000; EMEP/CORINAIR, 2002]. [17] The Lloyd's fleet statistics includes all oceangoing merchant vessels greater than 100 GT [LR, 2000], independent of flag and register status. The term “international marine bunkers” is not linked to national, open, or international registers for vessels, but to bunker consumption by vessels in international operations (transport between nations). [18] The International Energy Agency (IEA) defines “international marine bunkers” to cover those quantities delivered to seagoing ships of all flags, including warships. Consumption by ships engaged in transport in inland and coastal waters is not included. National and coastal navigation (including small craft and coastal vessels not purchasing their bunker requirements under “international marine bunker” contracts) is included in their “internal navigation” category. Ocean, coastal, and inland fishing should be included in their “agriculture” category (all deliveries to users are classified as agriculture). For the year 2001, IEA reports 139 Mt, 31 Mt, and 89 Mt marine bunkers sold (heavy and diesel fuel) for the “international marine bunker,” “internal navigation” and “agriculture” categories, respectively (IEA, International bunker statistics year 2000 and 2001, available at http://data.iea.org). Only a fraction of the bunker fuel within the “agriculture” category is related to fishing vessels. The world marine bunker sale in 2001 is then 170 Mt according to IEA, excluding bunker fuel to fishing vessels. The Norwegian fishing fleet, counting some 8292 vessels (which includes 7424 vessels of length less than 15 m), reported a yearly bunker fuel consumption of 0.43 Mt in 1993 [Statistics Norway, 1996]. Norway has some 600 fishing vessels larger than 100 GT, representing 2.5% by numbers and 4% by GT of the world fishing fleet [LR, 2000]. Modeling reported by the government agency Statistics Norway [1996] indicates that the ∼600 large fishing vessels consume 0.25 Mt of fuel annually. Assuming that the Norwegian fleet is representative of the world fleet in terms of activity level and size, we find that the 25,000 fishing vessels in the world consume around 10 Mt of fuel. Consequently, it is unlikely that the fishing fleet can explain the difference of ∼100 Mt between the world bunker fuel statistics and the modeled consumption reported by Corbett and Koehler [2003]. [19] The Energy Information Administration (EIA) reports bunker fuel sales to be ∼172 Mt for the year 2000 (EIA, International petroleum information, in “Crude oil imports and exports, with most countries and world total 2000,” available at http://www.eia.doe.gov/emeu/international/petroleu.html#IntlTrade). Although this corresponds very well with IEA data, the EIA data should be carefully interpreted as recommended by Skjølsvik et al. [2000]. EIA and IEA define “bunkers” differently. IEA gives the fuel consumption of international marine bunkers including consumption by warships, while EIA does not separate international bunkers and also includes jet fuel in its figures. The EIA data should thus be expected to give a higher estimate compared with those of the IEA. [20] However, Figure A1 illustrates a good agreement for most of the countries reported by both EIA and IEA (year 2000). The 22 countries with zero reporting, either by IEA or EIA, consume small amounts of fuel and therefore have a low impact on the total inventories. Table A1 illustrates some differences between the EIA and IEA data. The largest deviation is obtained for Russia and the United States. The total consumption for the United States is approximately the same for IEA and EIA but varies for the main fuel types. The Russian consumption is reported to be some 16 Mt by EIA, while IEA does not report sales at all. This deviation explains most of the difference between the total IEA and EIA numbers. Figure A1 clearly illustrates that these bunker statistics mainly overlap and that only a few countries generate the error in the total bunker inventory. However, improvements should be made in harmonization of common definitions and figures reported by countries. [21] The higher marine diesel consumption reported by EIA, especially in Asia (Table A1), indicates that national consumption is included for some countries according to EIA's definition of bunker. The EIA and IEA total figures are ∼175 Mt, with 120–130 Mt of heavy fuel and 50–60 Mt of marine diesel fuel. Our best estimate is 180–190 Mt, including the fleet of fishing vessels. The national consumption represents some 17% of the total usage. [22] The estimate of 289 Mt reported by Corbett and Koehler [2003] is significantly higher than IEA and EIA worldwide sales numbers for marine bunkers (year 2000). It is not likely, however, that the difference of some 100 Mt can be explained by deficiencies in the statistical sales reports.