Shipping activity has increased considerably over the last century and currently represents a significant contribution to the global emissions of pollutants and greenhouse gases. Despite this, information about the historical development of fuel consumption and emissions is generally limited, with little data published pre-1950 and large deviations reported for estimates covering the last 3 decades. To better understand the historical development in ship emissions and the uncertainties associated with the estimates, we present fuel-based CO2 and SO2 emission inventories from 1925 up to 2002 and activity-based estimates from 1970 up to 2000. The global CO2 emissions from ships in 1925 have been estimated to 229 Tg (CO2), growing to about 634 Tg (CO2) in 2002. The corresponding SO2 emissions are about 2.5 Tg (SO2) and 8.5 Tg (SO2), respectively. Our activity-based estimates of fuel consumption from 1970 to 2000, covering all oceangoing civil ships above or equal to 100 gross tonnage (GT), are lower compared to previous activity-based studies. We have applied a more detailed model approach, which includes variation in the demand for sea transport, as well as operational and technological changes of the past. This study concludes that the main reason for the large deviations found in reported inventories is the applied number of days at sea. Moreover, our modeling indicates that the ship size and the degree of utilization of the fleet, combined with the shift to diesel engines, have been the major factors determining yearly fuel consumption. Interestingly, the model results from around 1973 suggest that the fleet growth is not necessarily followed by increased fuel consumption, as technical and operational characteristics have changed. Results from this study indicate that reported sales over the last 3 decades seems not to be significantly underreported as previous simplified activity-based studies have suggested. The results confirm our previously reported modeling estimates for year 2000. Previous activity-based studies have not considered ships less than 100 GT (e.g., today some 1.3 million fishing vessels), and we suggest that this fleet could account for an important part of the total fuel consumption (∼10%).
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Over the last 100 years the total fuel consumption and emissions by the oceangoing civil world fleet has significantly increased as the fleet expanded by 72,000 motor ships to a total of 88,000, with a corresponding increase in tonnage from 22.4 to 558 million gross tonnage (GT) (Lloyd's Register of Shipping, statistical tables, 1964 (year 1900), and world fleet statistics and statistical tables, 2000). This growth has been driven by increased demand for passenger and cargo transport, with 300 Mt cargo transported in 1920 [Stopford, 1997] and 5,400 Mt in 2000 [Fearnleys, 2002]. There is a significant delay in building up the concentrations of some of the greenhouse gases (e.g., CO2) and in the climate impact. Knowledge on how ship emissions have developed over time is required to quantify climate effects and trends, and to implement effective regulations. Only limited information has been published relatively to the historical development of fuel consumption and emissions by the world fleet, and even for the last 3 decades the estimates being presented differing significantly.
Smith et al.  have reported an emission inventory covering the period before 1950. The validity of these estimates may be questioned as the annual coal consumption figures appear to be substantially lower than reported in the literature (coal used for shipping purposes in 1915 was taken to be only 80 Kt instead of the 80 Mt that was published by Annin ). Eyring et al.  have reported simplified activity-based inventories from 1950 up to 1992. After 1992, different approaches and assumptions have been applied to estimate global shipping emissions, but significant differences are apparent among these reported emissions inventories. This study has therefore established an historical time series of marine fuel sales to oceangoing ships from 1925 to 2002, and used a fuel-based approach to estimate the emissions of CO2 and SO2.
 It is an ongoing scientific debate regarding the reliability of marine bunker sale statistics to be used for estimates of fuel-based ship emissions [Corbett and Koehler, 2003, 2004; Eyring et al., 2005; Endresen et al., 2003, 2004, 2005]. Activity-based estimates reported by Corbett and Koehler  and Eyring et al.  are significantly higher compared to historical sales data covering the last decades. Moreover, the variation over time between reported sales and estimated consumption does not correspond. It has been argued that underreporting of sales explains the large differences [Corbett and Koehler, 2003; Eyring et al., 2005]. An alternative explanation could be that changes in structure, technology and activity of the expanding world fleet have to be better captured in the activity-based models. It has recently been questioned if the assumed activity level for the fleet is representative [Endresen et al., 2004], and especially for medium and smaller ships (which dominate by number). This study argues that any activity-based approach must take into account variation in the demand for sea transport and operational and technical changes over the years, to better represent the real fuel consumption and corresponding emissions. To better examine these factors this work has established an improved activity-based modeling approach to estimate fuel consumption for the oceangoing civil world fleet larger or equal to 100 GT. This model is applied to the period 1970 to 2000, and limited to the main engines of the nonmilitary fleet. The estimated fuel consumption is then compared with historical marine sales data, to investigate if marine sales data reported over the last 30 years are representative and could be used as proper bases for emissions estimates.
 In this study we first present a description and discussion of the developed CO2 and SO2 emissions inventories for the period 1925 to 2002 (section 2). We describe the activity-based model developed, the input data applied and the modeled fuel consumption for the last 3 decades in section 3. Section 4 compares the modeled fuel consumption for the last 3 decades with marine sales data. In section 5, major findings are summarized and discussed.
2. Fuel-Based Emissions Inventory: 1925–2002
 This section presents estimates from 1925 up to 2002 of global CO2 and SO2 emissions from oceangoing civil ships by combining estimates for marine fuel sales of coal and oils with their respective fuel-based emissions factors.
where Eg,j denotes the emissions of individual exhaust compound g (g = 1 = CO2, g = 2 = SO2) from burning fuel type j (j = 1 = diesel, j = 2 = heavy fuel, j = 3 = coal), kg emissions/year; Sj denotes the total sales of fuel type j, kg fuel/year; and eg,j denotes the emission factor for exhaust compound g in relation to fuel type j, kg emissions/kg fuel.
 The emission estimates presented in this study are based on bunker sales data (marine oil and coal) obtained from several sources (section 2.2), and average fuel-based emission factors assumed representative for the different time periods (section 2.3).
2.2. Marine Sales
 Data exists for sales of fuel to foreign bound ships from 1925 to 1970, but assumptions have to be made for the national sales. Data for the total marine sale exists from 1971 up to 2002, except for the fishing fleet for which separate sale estimates are made. The sources, assumptions and inventories are presented below for two periods (1) 1925–1970 and (2) 1971–2002. The sales figures given in bold in Table 1 are used to estimate the emissions for the period. Note that no data is available for the World War II period.
Table 1. Reported Worldwide Marine Sales of International Marine Bunkers (Oil and Coal) for the Period 1925–2000 and Estimated SO2 and CO2 Emissions (Equation (1))a
Oil equivalents, using 1/1.416 as conversion factor from coal to oil equivalents [UN, 1998].
2.2.1. Period I: 1925–1970
 The international marine sale figures from 1925 up to 1970 are based on coal and oil statistics reported by Darmstadter et al.  and United Nations (UN) [1957–1979]. These data cover sales to foreign bound ships and aircrafts, irrespective of flag [Darmstadter et al., 1971; UN, 1957–1979]. It is expected that sales to oceangoing military vessels in international operations are included in the presented inventory. However, no data is available for oceangoing military vessels in national services. This sale is small compared to sales to the nonmilitary fleet, according to estimates reported by Endresen et al. , and is not included in this work. As national fuel sales to the nonmilitary fleet are not included, the total oil and coal sale for this period is estimated on the basis of ratios between national sales versus international, which can be established for years when data is available. The calculated average ratio between IEA sales figures of national bunker (category: “internal navigation”) and international bunker (category: “international marine bunkers”) figures for the period 1971 to 2002 is calculated to be 0.27 (i.e., national sales is on average 27% of the international sales). However, the estimated national sales do not include the fishing fleet as IEA reports this sale under the “agriculture” category for this period. A separate estimate for the fishing fleet is given in section 2.2.2 and indicates that the fishing fleet consumes some 10% of the total sales. Assuming that these ratios also are representative for the period 1925 to 1970, the total sale (Total) is calculated by means of
Table 1 shows the estimated total figures for the period 1925 up to 1970, as well as the reported international sales (Int) used as input to equation (2).
2.2.2. Period II: 1971–2002
 Data for the total marine fuel sales exists for 1971 up to 2002, but this does not include the fishing fleet. The applied sales figures from 1971 up to 2002 are based on oil statistics reported by International Energy Agency (IEA) [1971–2002]. Our data is based on IEA categories “international marine bunkers” and “internal navigation.” The sales to oceangoing military vessels are expected to be included, as IEA defines the “international marine bunkers” to cover those quantities delivered to seagoing ships of all flags, including warships [IEA, 2001]. The fishing fleet sale is reported in the IEA category “agriculture” [IEA, 2001]. It is not possible to extract sales to fishing vessels from this overall category and these sales figures are therefore estimated from the 34,379 MW main engine power that was reported to be installed in the 1.3 million decked fishing vessels in 1998 [Food and Agriculture Organization of the United Nations (FAO), 2006a]. This represents nearly twice the power reported by Corbett and Koehler  (18,474 MW) for the fishing fleet larger than 100 GT, and about 13% of the total main engine power of the oceangoing civil world fleet larger than 100 GT. We may assume that the main engine power installed is roughly proportional with the sales to the fishing fleet. The entire fishing fleet consumption in 1998 is then calculated to be of the order 10% of the total sales by assuming less days at sea for the fishing fleet compared to the cargo ships [Endresen et al., 2004] and by taking into account that some of the reported IEA sales include military vessels. This study assumes that the estimated 10% consumption by the entire fishing fleet in 1998 is representative for the period 1925–2002. Lloyd's fleet data for the years 1970 and 2000 shows little change in the relative fraction by numbers of fishing vessels, but the relative tonnage fraction was higher and the average vessel size was lower in 1970 than in 2000. Global fishing fleet data up to 1970 is not available, but the fish catch statistics indirectly support our assumption. Fish catches in 1938, 1950, 1970 and 2000 within marine waters was 19.3 Mt (reported by FAO, figures given by Aschehougs Konversasjonsleksikon ), 17.3 Mt, 59 Mt and 88Mt [FAO, 2006b], respectively and this development corresponds well with the total fleet development over this period (Figure 1). On the basis of this assumption, the total sale (Total) is calculated by equation (2). Table 1 shows the estimated total figures for the period 1971 to 2002, as well as the reported international (Int) and national (National) IEA sales data used as input to equation (2).
2.2.3. Discussion of the Data Sources
 We find a good match between international oil fuel sales reported by Darmstadter et al.  and IEA [1971–2002] with reference to 1965 and 1971, respectively. The match between international oil fuel sales reported by UN and IEA is generally good for the period 1971 to 1978, and by UN and EWE from 1953 to 1965 (Table 1). However, the 1977 and 1978 sales reported by UN are lower than IEA data. The IEA data does not include fuel sales to aviation, indicating that for this time period the reported fuel sales to aircraft is likely to be insignificant in both the EWE and the UN data. This is supported by the fact that airplanes do not use coal or heavy fuel oils, and that EWE states that the reporting of aviation fuel sales appear to be incomplete. Thus we assume that the aviation fuel included in UN and EWE data is small or negligible, and that the sources give representative data for sale to international shipping. It is important to recognize that IEA and UN define international bunker differently. UN includes only sales to foreign bound ships, while IEA defines the “international marine bunkers” to cover those quantities delivered to seagoing ships of all flags, including warships [IEA, 2001]. The good match between international bunker sales reported by UN and IEA indicates that the IEA “international” bunker category mainly includes sale to ships in international operations.
2.3. Emission Factors for CO2 and SO2
 We have used a factor of 3.17 tonnes CO2 per tonne oil consumed [EMEP/CORINAIR, 2002] and 2.58 tonnes CO2 per tonne coal consumed to model emissions to air. The emission factor for coal is calculated by combining the molecular weight ratio of CO2 to C (44/12), with 0.704 tonnes C/tonne fuel [Society of Naval Architects and Marine Engineers (SNAME), 1983]. The sulphur content varies over time, as pointed out by Endresen et al. . We have assumed a global average of 2.5% sulphur for marine oils up to 1970. From 1971 up to 1995, a global average of 2.7% sulphur is assumed for heavy fuel and 0.5% for distillates. Global values reported by Endresen et al.  are applied from 1996 to 2001. For year 2002 global weighted sulphur contents (heavy fuel and distillates) are applied [Endresen et al., 2005]. The globally weighted average content for heavy fuels is found to be 5% higher than the average (arithmetic mean) sulphur content commonly used. The likely reason for this is that larger bunker stems are mainly of high-viscosity heavy fuel, which tends to have higher sulphur values compared to lower-viscosity fuels [Endresen et al., 2005]. Smith et al.  reports a sulphur content in coal of 1.1%, and we assume this value to be representative for the entire period. The relation between burned sulphur and generated SO2 is 2.0 kg SO2/kg S (derived from the chemical equation [Lloyd's Register of Shipping, 1995; EMEP/CORINAIR, 2002]).
2.4. Modeling of Ship CO2 and SO2 Emissions
Figure 2 shows the calculated CO2 and SO2 emissions (equation (2)) using the sale numbers (section 2.2) in combination with the emission factors per tonne oil and coal consumed (section 2.3). Ship emissions are estimated at 229 Tg (CO2) in 1925, growing to about 634 Tg (CO2) in 2002. The corresponding SO2 emissions are 2.5 Tg (SO2) and 8.5 Tg (SO2).
2.4.1. Factors Determining Development in Worldwide Fuel Sales
 This study expects that the demand for sea transport, technical and operational improvements as well as changes in the fleet composition and size will explain most of the development in fuel consumption by the fleet during the last 100 years. The coal sale peaked in 1913 (80 Mt coal is reported by Annin ), and dominated up to around 1920, as the fleet grew and steamers replaced sail ships (Table 1) [Fletcher, 1997]. After 1920, the oil sale started to dominate the emissions, as coal gradually was replaced by marine oils, following the shift to diesel engines and oil-fired steam boilers (Table 1) [Fletcher, 1997; Corbett, 2004]. Increased focus on fuel economy [Kofoed, 1926], and a shift from coal to oil, combined with depressions in both the world economy and the sea borne trade in the 1930s [Stopford, 1997] (e.g., some 21% of the fleet was out of service in 1932) partly explain the gradual reduction in sales from around 1925 (Table 1) toward the Second World War.
Table 1 illustrates the significant increase in marine sales after the Second World War and up to 1973. The main reasons for this significant growth was the demand for sea transportation, as the sea borne trade grew from around 500 Mt in 1949 to 3,233 Mt in 1973 [Stopford, 1997] (i.e., more than a sixfold increase). The growth in sea borne transport was not reflected by a corresponding growth in the fleet by vessel numbers (i.e., a twofold increase) or tonnage (3.5-fold increase), indicating the influence of modern, larger and more efficient cargo ships, with improved cargo handling in ports. For instance, the volume of cargo transported per tonne fuel sold significantly increased in this period (see Figure 3). Note that the transports of passengers, important up to around 1960, have had an influence on the volume transported per tonne fuel sold. We expect that passenger ships, the largest ship type in the fleet up to around 1960 [Aschehougs og Gyldendals store Norske Leksikon, 1999], account for some of the coal consumption for this period and before. For instance, the annual number of European emigrants transported to the US increased from 350,000 around 1890 to around 1.4 million in 1910 [Aschehougs Verdenshistorie, 1982]. The largest of the old passenger liners such as the Olympic and the Titanic burned on average 620 tonnes of coal per day at 21.7 knots (Encyclopaedia Titanic, Daily fuel consumption for Titanic and Olympic, available at http://www.encyclopedia-titanica.org/discus/messages/5919/6509.html, visited 2005).
 The marine sales decreased from around 1973 to 1983, followed by a nearly steady growth up to 2002 (Table 1 and Figure 3). The main reasons for the decrease were the slowdown in world sea borne trade (Figures 3 and 4) , the reduction in sailing distances (Figure 4), improved energy efficiency of the fleet (e.g., phasing out of steam ships) and a reduction in speed (and installed power) within some dominating segments (Figure 5). World economy generates most of the demand for sea transport, through either the import of raw materials for manufacturing industry, or trade in manufactured products [Stopford, 1997]. Figure 3 illustrates that development in bunker sale follow development in seaborne trade with a correlation of r = 0.92 for the period 1975 to 2000. The correlation between transport work (measured in tonne miles, Figure 1) and fuel sale for the same period is even better (r = 0.97) (Figure 6, right), indicating that the average length of haul, not surprisingly, is affecting the sales. However, the correlation between these variables is lower (r = 0.88) for the period 1971 to 2000 (Figure 6, left), indicating that other factors also were important, especially between 1971 and 1975. For instance, the transport work by the old passenger fleet is not included, as well as effects of changed operational speed and shift to diesel powered ships. The typical operational speed has also varied widely over time, which significantly influences the power requirements. For example Very Large Crude oil Carriers typically operated at 10 knots when freight rates were low in 1986, but this increased to 12 knots when the rates were higher in 1989 [Stopford, 1997]. A reduction in the average operating speed by 2–3 knots below design speed may halve the daily fuel consumption of the cargo fleet [Stopford, 1997; Wijnolst and Wergeland, 1997]. Lloyd's fleet data [Lloyd's Register Fairplay (LRF), 2005–2006] also indicates a reduction in installed power and operational speeds for diesel powered crude oil carriers built after 1980, followed by a significant reduction in fuel consumption (Figure 5). The annual fuel consumption by the fleet is also strongly affected by the installed propulsion systems (engine, gear, shaft, propeller arrangement), as modern diesel engines have about half the daily fuel consumption compared to the old inefficient steam engines with the same power outtake (Table 2). The shift to modern marine diesel engines has typically occurred in periods with high oil prices. For instance in 1961, there were still over 10,000 steam engine powered ships and 3,536 steam turbine powered ships in operation (36% by number) (Lloyd's Register of Shipping, statistical tables, 1961). By 1970 these numbers had decreased to 4,425 and 3,534 ships respectively (15% by number) (Lloyd's Register of Shipping, statistical tables, 1970). By 1984 only 1,213 ships were steam engine powered and 1,743 turbine powered ships (4% by number) remained in service (Lloyd's Register of Shipping, statistical tables, 1984).
Table 2. Reported Specific Fuel Consumption (SFC) for Different Engine and Fuel Types
 The decrease in marine sales from around 1973 is also explained by the decline in both sea borne transport of oil (represented about 50% of the sea borne trade) and the average sailing distances (Figure 4, middle). For instance, crude oil tankers reached a peak in productivity in 1972 (measured in tonne miles per deadweight (total carrying capacity)). By 1985 this had nearly halved, and a few years later it increased by 40% [Stopford, 1997]. More efficient and specialized ships have also pushed their way into the marked (e.g., the first deep sea cellular container ship in 1965 [Stopford, 1997]). The specialized ships have different operational and technological characteristics, which results in a particular logistic efficiency and energy and emission profiles. For example, passenger ships have on average 2.2 main engines per ship, while the large passenger ships have 5.7 main engines per ship (greater than 100,000 GT) [LRF, 2005]. The civil world fleet have on average 1.3 main engines per ship (above or equal to 100 GT) [LRF, 2005]. Engine capacity is not likely to be fully exploited at all times, resulting in a lower fuel consumption in practice than what might be expected on the basis of the power installed. It is reasonable to expect that these effects are likely to have different impacts on the total fleet within the period 1970–2000 and so we have included most of them in the fleet modeling approach outlined in section 3.
 The uncertainties in our estimated sales figures are significant. Reliable inventories are probably best developed by comparing the different modeling approaches and the different data sets that are available. This extends our knowledge base and improves our understanding of the governing processes. The inventory presented in this study aims to cover sales in oil equivalents to all oceangoing ships worldwide (Table 1). However, the marine sales for the period 1925 to 1970 is likely to be slightly underestimated, as sales to the oceangoing military fleet in national services are not included. Our best estimate today is that the data after 1970 should be within a range of ±15%, while the data before 1970 should be within a range of ±25%. Underreporting by some countries has been [Endresen et al., 2003] and may still be a problem. The national sales reported by IEA from 1971 to 2002, also include sales to smaller ships operating on inland waterways. This fleet is reported to account for 42,000 engine powered ships and 38,000 push-towed vessels in 1992 [Organisation for Economic Cooperation and Development, 1997]. The engine powered ships are small (∼300 Dwt) and represent around 2% of the cargo capacity of the oceangoing fleet. Thus it is assumed that the sale to this segment is small.
 The variation in carbon content for marine oil products is small [Energy Information Administration, 1994]. The uncertainty in the average CO2 emission factor for marine oils is less than ±5%. This is in line with Skjølsvik et al. . The assumed carbon content in coal for marine purposes is reported to be 70.4% [SNAME, 1983]. We expect the error in the assumed carbon content for coal to be within ±10%. Taking into account uncertainties in sales numbers, it is expect that the uncertainties in the CO2 emissions after 1970 should be within a range of ±20%, while the estimates before 1970 should be within a range of ±30%. Limited data exists for average sulphur content of marine oil up to 1990. This is supported by the fact that the use of residual fuel in marine diesel engines dates to the 1940s. Prior to the 1940s, residual fuels for navigational purposes had been used by steam ships [Cullen, 1997]. We judge the error in sulphur content in oil to be within ±20% up to 1990, and within ±10% after 1990. It is expected that the total uncertainties in the SO2 emissions after 1990 should be within a range of ±20%, while the estimates before 1990 should be within a range of ±30%. We realize that the modeling is based on a number of assumptions regarding average sulphur and carbon content (coal) and the results should therefore be interpreted carefully.
3. Activity-Based Fleet Modeling, 1970–2000
 This section presents an improved activity-based modeling approach that uses high-resolution time series as input data to estimate fuel consumption for the oceangoing civil world fleet larger or equal to 100 GT. Modeling is made for the period 1970 to 2000, and only for the main engines. Historical data available is limited, and do not allow for detailed modeling, such as the baseline approaches reported by Corbett and Koehler , Endresen et al.  and others (e.g., breakdown on ship types and sizes). Simplifications and assumptions are therefore made. However, compared to past activity-based modeling studies [Eyring et al., 2005; Corbett and Koehler, 2003], this study has developed and applied a more detailed approach, which includes the variation in the demand for sea transport, as well as operational and technological changes.
3.1. Modeling Approach
 The average ship size approach applied in this study assumes an equal size for the total number of ships in the world fleet of oceangoing civil ships larger or equal to 100 GT, and calculates the average size of the ships by dividing total tonnage with the total number of ships. The fuel consumption for an average ship is estimated on the basis of average characteristics of installed main engine power, main engine load, bunker fuel consumed per power unit (kW) (depends on propulsion and fuel type) and days at sea (based on demand for sea transport). The fuel consumption is calculated separately for the diesel and steam ships, as steam ships have a significantly higher fuel consumption (Table 2). The main engine fuel consumption for the period 1970–2000 for all oceangoing civil ships above or equal to 100 GT is then estimated by
where Fi,s denotes the total fuel consumption of average ships with main engine type i burning fuel type s, kg fuel/year; bi,s denotes the average specific fuel consumption for an average ship with main engine type i burning fuel type s, kg fuel/kWh; t denotes the average number of operating hours at sea per year for an average ship (see equation (5), below), h/year; p denotes average installed main engine power for an average ship, kW; m denotes the average main engine load for an average ship; ni,s denotes the fraction of the average ships in the fleet with main engine type i burning fuel type s (see equation (4), below); and N denotes the total number of active average ships in the fleet (i.e., not laid up and used as storage).
 Input data for the different terms in equation (3) is described below and given in Table 3. It should be noted that table figures are only given every tenth year, while the numbers applied in our modeling are given year by year.
Table 3. Input Data to Equation (3), Modeling Main Engine Fuel Consumption for the World Fleet of Oceangoing Civil Ships Larger or Equal to 100 GT, Period 1970–2000 (Presented per Decade)a
Note that yearly input data is used in the modeling.
3.1.1. Number of Active Average Ships, Diesel and Steam Powered
 As outlined above, the number of average ships corresponds to the total number of ships reported in the world fleet of oceangoing civil ships above or equal to 100 GT (Figure 1, left). The total number of ships and GT in the fleet from 1970 to 2000 is based on fleet statistics (Lloyd's Register of Shipping, world fleet statistics and statistical tables, 1992 (year 1970–1992), 1995 (year 1993–1995) and 2000 (year 1996 and 2000)). The fraction of the average ships (ni,s) in the fleet with main engine type i burning fuel type s is calculated on the basis of the following equation:
where Di,s denotes the total tonnage in the fleet with engine type i burning fuel type s, GT.
 The tonnage in the fleet with engine type i burning fuel type s is based on yearly fleet data from Lloyd's Register of Shipping (statistical tables, 1972–1975 and 1977–1984). From 1985 to 1992 the yearly fleet data is only available for the steam powered tank and bulk fleet (Lloyd's Register of Shipping, statistical tables, 1985–1990, and world fleet statistics and statistical tables, 1991, 1992). However, we have assumed that the tanker and bulk fleets are representative for the total steam tonnage, as they were the dominating ships by tonnage. Fleet data is not available for the steam powered segment from 1993 onward, and the changes in the steam tonnage are estimated by interpolation between the 1992 tonnage and the actual tonnage in 2004 [LRF, 2005]. The percentage of the coal fired tonnage is available from Corbett  up to 1960. Detailed fleet data from Lloyd's is available for the years 1961, 1962 and 1963 (Lloyd's Register of Shipping, statistical tables, 1961–1963), but no data is available thereafter according to our information. A linear reduction from 3.3% of the coal fired tonnage in 1963 to zero in 1979 is assumed, on the basis of the development of coal sales (Table 1). It should be recognized that other engine types are represented in the oceangoing civil world fleet (e.g., gas turbine), but these types are negligible by tonnage and number when compared with steam and diesel. Thus we have not taken them into account in this study.
 The number of average active ships (N · ni,s) is calculated by subtracting the number of average ships corresponding to the nontrading tonnage. The number of ships out of service is distributed on engine and fuel categories according to the relative tonnage of the fleet with main engine type i burning fuel type s. The development in laid up tonnage and tonnage used for storage from 1970 to 1995 is based on Stopford , and on data from Fearnleys  thereafter. The number of active average sized steam ships (N · n2,s) will then reflect the total amount of tonnage (in service) for the steam powered fleet (D2,s), while number of active average sized diesel ships (N · n1,1) will reflect the total amount of tonnage (in service) for the diesel powered fleet (D1,1).
3.1.2. Time at Sea
 The operational profile is derived by combining yearly fleet and trade data. This deviates from recent modeling studies that estimate days at sea on the basis of engine manufactures data for large engines and tracking studies. The development of the yearly total number of days at sea sailing with cargo for the fleet is estimated on the basis of number of voyages required for an average cargo ship (with an average utilization of cargo capacity) to transport the yearly reported worldwide total cargo volumes, combined with average voyage time calculated from reported average length of haul and assumed average operational speed. We have only considered the cargo carrying fleet when estimating days at sea, and assumed that days at sea for cargo and noncargo ships are equal. Noncargo ships normally have less days at sea as indicated by Endresen et al. , and this assumption may bias our fuel consumption estimates toward being too high, compared to the real situation. The development of the yearly average number of days at sea (t) (with cargo and in ballast condition) of the active average cargo ships is calculated by means of
where q denotes the total yearly sea borne trade of ships in the fleet, tonnes/year; l denotes of average length of haul (with cargo), nautical miles (nm); d denotes average dead weight tonnage (Dwt) of ships in the fleet, Dwt; v denotes the average operational speed of ships in the fleet, nm/h; η denotes the average utilization of cargo capacity of ships in the fleet, tonnes/Dwt; and α denotes the ballast factor, defined as the average number of days in ballast relative to days sailing with cargo.
 Input data for the different terms in equation (5) is described below and given per decade in Table 4. Average operational speed (v) is derived from data collected for 37,193 cargo and passenger ships in the Lloyd's 2006 fleet database [LRF, 2005–2006]. The average operation speed for a given year is based on data for all ships the actual year and all older ships in the database. The total yearly reported sea borne trade volumes (q) for 1970 to 1995 are based on Stopford , while data for the period 1996 to 2000 are based on Fearnleys  (Figure 3). The average sailing distances (l) are calculated by dividing the total transport work (tonne miles) by the total sea borne trade volumes. From 1970 to 2000 these data are based on Fearnleys  and Stopford .
Table 4. Input Data to Equation (5), Modeling Time at Sea for the World Fleet of Oceangoing Civil Ships Larger or Equal to 100 GT, Period 1970–2000 (Presented per Decade)a
Note that yearly input data is used in the modeling.
 Relatively little information is available about the average time spent in ballast. Ideally, a ship should complete all voyages with cargo. However, many trades require return voyages without cargo. For example, a crude oil tanker typically transports a single cargo load between two ports, then returns to its point of origin or another port without cargo. Wijnolst and Wergeland  indicates that it is not likely for the tanker fleet that the average ballast factor will be less than 0.8, and it will hardly ever exceed 1. Tracking data reported for 453 Very Large Crude oil Carriers in 1991 illustrated a ballast factor of 0.81. Other ship types such as general cargo and container ships, often sail with some cargo and some ballast and have limited time in ballast, but are frequently not fully laded. For about 100 smaller cargo ships operating on a regional basis (most ships around 3,500 Dwt), the ballast factor is reported to be around 0.2 [Wilson EuroCarriers, 2005]. In the modeling, these ships are “forced” to transport all cargo nearly fully laden, and then spend time in ballast. On an annual basis, it is expected that this simplification, may give representative number of days at sea, as well as cargo volumes transported. Taking into account the fact that larger ships dominate the transported volumes at sea [Stopford, 1997], the average ballast factor (α) is assumed to be 0.7 (i.e., days with ballast is 70% of the number of days with cargo) for all ship types.
Wijnolst and Wergeland  reports that utilization of cargo capacity in practice hardly exceeds 0.95, but may become as low as 0.65. The utilization of cargo capacity of bulk ships and oil tankers larger than 50,000 Dwt in 2001 is reported by Behrens et al.  to be 0.91 and 0.87, respectively. These ships account for almost 55% of the sea borne trade in 2000 [Fearnleys, 2002], and will of course then have a large impact on the average factor. We acknowledge that general cargo and container ships have a lower utilization of the cargo capacity, and that the average density of cargo could make volume the limiting factor. For instance, data reported by Johnsen  for a dry cargo ship with carry capacity of about 5,200 tonnes, indicate that the ships on average were loaded at 4,000 tonnes. Of this, the average exploited capacity is reported to be 76% (considering different trading routs). Data for container ships serving U.S. trade [PIERS/Journal of Commerce, 2005], indicate a utilization rate of the container capacity of about 70% on average. Important to note is that the utilization rate of the container capacity will typical be higher than the average utilization of the cargo capacity. Clearly the utilization of cargo capacity varies for different ship types, as well as for year considered (depending on the marked). We assume that 0.8 is representative for the average utilization of the cargo capacity (η) for all cargo ships types.
3.1.3. Specific Fuel Consumption and Engine Load
 The main engine specific fuel consumption (bi,s) is based on data reported in the literature and engine data reported for individual ships. The main engine specific fuel consumption for an average diesel ship (b1,1) is estimated for different periods (up to 1970; up to 1980;…; up to 2000), combining installed main engine power and the daily fuel consumption reported for 16,465 diesel powered oceangoing civil ships in the Lloyd's 2004 fleet database [LRF, 2005]. A similar calculation is made for the 770 steam turbine powered cargo and passenger ships running on oil [LRF, 2005–2006]. The daily fuel consumption is normally given at full power (85% MCR: Maximum Continuous Rating). We have therefore assumed in the calculation of specific fuel consumption (Table 3) that utilized power is 85% of installed power.
 It is important to recognize that no differentiation is made between steam engines and turbine engines. We have defined average steam ships using oil (b2,1) or coal (b2,2). Steam engines have higher fuel consumption compared to turbine engines, as illustrated in Table 2. However, limited data is available on specific fuel consumption for old steamers, and especially for the steam engines. Both steam engines and steam turbines burning coal were phased out of the fleet around 1970. We have therefore established typical constant averages specific fuel consumption for these ships. Riksheim  compared the fuel consumption of a diesel powered ship of 142,000 Dwt delivered in 1981, with a coal fired steam turbine ship with the same hull dimension, shaft power and services speed. He reported that the coal fired solution used more than 3 times the fuel by weight, compared to the diesel solution (64 tonnes oil per day versus 198 tonnes coal per day or 140 tonnes of oil equivalents). The data in Table 2 also illustrates that turbines running on coal typically use 3–3.5 times more fuel per unit power production, compared to the diesel engines. Thus we assume for all years that turbine engines fuelled by coal consume 800 g/kWh.
Table 2 indicates that steam engines running on oil consume 700 g/kWh, while steam engines using coal have slightly higher fuel consumption than turbine engines using coal. Thus, for steam engines using oil and coal we have assumed a fuel consumption of 700 g/kWh and 900 g/kWh, respectively. For average steam ships using oil (b2,1) or coal (b2,2), we have estimated the specific fuel consumption using these specific consumptions with a weighting according to the tonnage. In 1961, the steam engines accounted for 41% of the steam tonnage (Lloyd's Register of Shipping, statistical tables, 1961). This is reduced to 8% in 1970 (Lloyd's Register of Shipping, statistical tables, 1970), 2% in 1984 (Lloyd's Register of Shipping, statistical tables, 1984) and assumed 0% in 1990. The estimated weighted specific fuel consumptions (b2,1) and (b2,2) are then applied in the modeling. Table 3 shows the applied averages (only given per decade).
 We assume an average main engine load of 70% MCR, when including slow cruise, port maneuvering and ballast sailings. This assumption is based on recommendations by Endresen et al. [2003, 2004] and Corbett and Koehler . The average engine load (m) is assumed for all ships and all years. However, the average main engine load and speed varies a lot for different ship types. For instance, Flodström  reports an average load of 80% MCR based on data from 82 ships. Bulk vessels tend to have slightly lower average values (72% MCR), while tank vessels have higher (84% MCR). The load was ranging from about 60% MCR up to 95% MCR for the 82 different ships. This illustrates large variations in required engine output and average operational speed.
3.1.4. Installed Main Engine Power
 From 1978 to 2000 the yearly average main engine power (p) is estimated from Lloyd's fleet database [LRF, 2005]. The average installed power for a given year is calculated using a similar approach as for operational speed. Up to 1978 the change in average power is assumed to follow the development for ships larger that 2,000 Dwt reported by Eyring et al.  (based on data from UK's trade magazine “The Motor Ship”). We believe that this better represents the development of average power in the fleet up to 1978, compared to application of detailed fleet data available, where larger merchant ships could be underrepresented because of scrapping after some 25–30 years (not included in the fleet database). Our calculated average power figures per ship correspond to the data reported for 1990, 1995 and 2000 by Eyring et al. . However, our estimate for 1980 is about 10% lower, compared to Eyring et al. .
3.2. Modeled Fuel Consumption
Figure 7 shows the modeled fuel consumption by means of equation (3) for the world fleet from 1970 to 2000. Our results show that the fuel consumption can be modeled by including the major changes in technology, fleet structure and operational factors. Our modeling differentiates on engine and fuel types, and Figure 7 (as well as Figure 8) illustrates that these effects are important around 1970, and should be included in historical fleet modeling. The increased fuel consumption from 1970 to 1973 (1974) is explained with increased demand for sea transport (i.e., number of days at sea, Figure 9), as well as long sailing distances with cargo (Figure 4, middle). The latter have reduced number of port calls and consequently reduced total time in port (more time at sea). For the tanker fleet, transporting 50% of the sea borne trade around 1970, the average distance was about 6,600 nautical miles (nm) in the mid 1970s, before rapidly falling to a level of 4,600 nm in the mid 1980s (Figure 4, middle). The reasons was that after the oil crisis in 1973, the USA in particular, but also European importers, started to import more from other, Non-OPEC sources, or increased their own production as was the case for United Kingdom and Norway [Wijnolst and Wergeland, 1997]. The variation in trade patterns for oil is therefore an important factor determining demand for crude oil transportation, as well as fuel consumption by this fleet segment and the entire fleet.
 The results show that growth in the fleet is not necessarily followed by increased fuel consumption. For instance, the stagnation and decline from around 1973 to 1983 in fuel consumption is explained with decrease in number of days at sea (Figure 9), combined with reduction of average sailing distances (oil tankers), increasing number of ships laid up (Figure 4) and the rapid shift to diesel powered ships. The stagnation and decline around 1973 in oil transportation (Figure 4), the largest individual commodity trade by sea, is explained with the very high oil prizes resulting in shift back to coal as fuel for power stations [Stopford, 1997]. Around 1985, when the laid up tonnage was reduced and most of the ships in the fleet were diesel powered, the stagnation in fuel consumption can mainly be explained by a slow down in sea borne trade (Figures 3 and 4). After 1990, the fuel consumption has almost followed the development in sea born trade and fleet growth (Figure 4 versus Figure 7).
 Our results indicate that better activity data on a yearly basis over time is required when fleet modeling is used to determine the actual fuel consumption for the entire world fleet. If our method for estimating days at sea is extended to cover the main cargo shipping segments separately, we expect that the uncertainty will be significantly reduced. However, the method fails for service ships (noncargo). Yearly tracking data (e.g., movement data available) for such vessels would then increase the reliability in model results. Lack of actual service speed data for the fleet significantly influences the uncertainty as the specific consume is speed-dependent and because the service speed defines number of days at sea required to carry out the transport demand. It is recommended that reported days at sea are applied with care since the model results are particularly sensitive to this parameter.
 We expect that in the future the actual service speed will be estimated on the basis of AIS (automatic identification systems) tracking data for individual oceangoing ships. Such data also makes it possible to indirectly estimate the engine power utilization per ship (and for fleet segments) by combining recorded service speed with installed main engine power for each individual ship (available from Lloyd's fleet databases). AIS is primarily an anticollision system, and is designed to be capable of automatically providing position and identification information about the ship to other ships and to coastal authorities [U.S. Coast Guard, 2002]. The International Maritime Organisation (IMO) requires AIS to be fitted aboard all international ships of certain size. Dalsøren et al.  also report that in the future local and regional ship emission inventories (geographical distribution of emissions) will be based on AIS statistics.
 The model has been applied to periods before 1970 as well. However, the presented approach fails around 1960, when the world fleet still transported large numbers of passengers (see equation (5)). It was not until 1958 that airplanes transported more transatlantic passengers than large passenger ships [Hansen, 2004]. Another problem is the fact that about half the US fleet of 28 million GT was laid up in 1949. This was the US reserve fleet and represented some 17% of the total world fleet by tonnage [Aschehougs Konversasjonsleksikon, 1951]. Also, historical shipping data for this period is limited. This illustrates some of the challenges that have to be considered when modeling fuel consumption before 1970.
3.2.2. Comparison With Other Studies
 Our new estimate for year 2000 is some 30 Mt higher than previously reported by Endresen et al.  (Table 5). The main reason for the deviation, is that our new modeling estimates also include about 45,000 noncargo ships, not considered by Endresen et al. . It is interesting that these two models using different approaches and data sets, give nearly similar results, if the consumption by noncargo ships is taken into account. The detailed activity-based estimates for the world fleet (civil) reported by Corbett and Koehler  and Eyring et al.  are still a factor of 1.25 (∼50 Mt) higher (Table 5). However, Corbett and Koehler  also considered alternative input data to the activity-based modeling, and pointed out that the fuel consumption could be up to 16% lower. This implies that the two studies are within the same range when the uncertainty bounds are taken into account.
Table 5. Comparison of Reported Fuel Consumption for the Oceangoing World Fleet
Estimates based on equation (2). The coal sales converted to oil equivalents by using 1/1.416 as conversion factor [UN, 1998]. Estimates for sales to the fishing fleet (section 2.2) are included.
Estimates based on equation (3). Cover fuel consumption by the main engines in the oceangoing civil world fleet above or equal to 100 GT. The coal consumption is converted to oil equivalents by using 1/1.416 as conversion factor [UN, 1998].
Simplified activity-based modeling, covers oceangoing ships above or equal to 100 GT (unclear if fuel consumption by the large military ships and auxiliary engines are included).
Fleet modeling, covers the world civil cargo fleet (oceangoing) above or equal to 100 GT.
Estimated sales of marine fuel, based on IEA and EIA data.
Detailed activity-based modeling, covers oceangoing ships above or equal to 100 GT (the civil fleet 254 Mt (includes noncargo fleet), the noncargo fleet 46.2 Mt, the military fleet 9.4 Mt (1300 navy ships), and auxiliary engines 16.3 Mt).
Detailed activity-based modeling, covers oceangoing ships above or equal to 100 GT (the civil fleet 248 Mt (includes noncargo fleet), the noncargo fleet 45 Mt, and the military fleet 41 Mt)).
Eyring et al.  also reported simplified activity-based fuel consumptions estimates for 1950, 1960, 1970, 1980 and 1995, and assumed that interpolation between these periods would reflect the development. They also interpolated between the simplified 1995 estimate and the detailed estimate for year 2001. Our modeled fuel estimates are significantly lower than reported by Eyring et al.  for the period 1980–2000. The main reasons for the large deviation are probably that they have assumed the number of days at sea significantly higher than we estimate and include less key influencing factors, compared with our model. Note that their assumed number of days at sea is about 90 days higher for year 1995 than the estimate provided by this study (179 days). The sensitivity modeling indicates more than two times higher fuel consumption for some years (e.g., 1983), if the major effects are not included and if days at sea are assumed equal to 270 (Figure 8). This shows that results from simplified activity-based models are sensitive to key input factors. Our model estimates for 1970, 1980, and 1995 are about 10–15 Mt higher than reported by Eyring et al. , if the days at sea is increased to 270 and held constant, and the laid up tonnage is set to zero (Figure 8). This illustrates that the two simplified models gives nearly the same response with similar input data. However, Eyring et al.  have not reported detailed input data for these reference years (e.g., assumed engine load, specific fuel consumption) and some consumption by military ships and the auxiliary engines could have been included. In addition, their model do not take into account laid up tonnage, and do not differentiate on engine and fuel types. Direct comparison is therefore difficult.
3.2.3. Uncertainty in the Estimates
 The uncertainty in each parameter of equation (3), including main assumptions, is addressed below with an estimate for the uncertainty in total fuel consumption. The weakness of the average ship modeling approach compared to more detailed modeling approaches (separates on ship sizes and types) is that our simplified approach uses average characteristics, separating on only steam and diesel powered ships. The uncertainties in our estimates arise from both uncertainties in the applied averages, as well as the simplified method. The methodology does not include second-order effects among parameters in equation (3). Moreover, it should be noted that our model do not take into account the technical development on antifouling systems [Evans, 2000] which likely have influenced fuel consumption and emissions over the past 100 years.
 The detailed activity-based approaches, identify size and type categories with common characteristics. The fuel consumption is then calculated for each category by applying characteristic factors for each category in combination with total power installed within this category. The total fuel consumption is based on a sum over all categories. Detailed input data is required for this approach and such data is generally not available to derive a historical inventory. We then have to simplify by establishing averages that are representative for one large category covering all ships in the fleet. The applied averages are not weighted with power installed or energy production, introducing an uncertainty. For example, small cargo ships have compared to large cargo ships relatively less installed power, less days at sea with a higher specific fuel consumption and lower average engine load (e.g., large number of voyages, requires more often part loads in port areas). The actual lower specific fuel consumption for larger ships implies that our model results in a too high overall fuel consumption as the larger ships dominate trade, tonnage and installed effect (Figure 10). Number of days at sea is based on a demand covered by a number of ships of average size. Larger ships will be more energy efficient and smaller ships less energy efficient. We therefore expect that our model results will overestimate the total consumption as the larger ships actually have more days at sea at a higher efficiency than the smaller ships (less days at sea with a corresponding lower efficiency). The average engine load is actually higher on larger ships than smaller ships. Our model will then likely underestimate the overall fuel consumption as larger ships dominate by trade and tonnage. The net effect of the nonlinear effects is consequently not straight forward to estimate.
 The assumed number of days at sea is important, and we have therefore performed a sensitivity analysis based on modeling the energy consumption (kWh) of the year 2004 civil world fleet [LRF, 2005]. The production within defined size categories is calculated by combining the installed power (Figure 10) with the number of days at sea reported for the different size categories. We used the number of days at sea reported for the 5 categories by Endresen et al.  for cargo ships (199 days for <9999 GT; 196 days for 10,000–24,999 GT; 205 days 25,000–49,999 GT; 219 days for 50,000–99,999 GT; 240 days for >100,000 GT). The total energy consumption for the entire fleet is then calculated by summing up for all categories. A corresponding modeling was made by using a calculated average (arithmetic mean) for these categories (= 212 days). The deviation was only 3%, and the constant average profile gave the highest estimate. However, most of the installed power for the noncargo ships originates from the category less than 9,999 GT (Figure 10), and they have typically on average less days at sea. The contribution from noncargo ships on the overall average operational profile was taken into account by assuming 150 days (some data given by Endresen et al. ) at sea on average (instead of 199 days) for the lowest category. A similar modeling was repeated. This exercise resulted in a deviation of 7% compared with our baseline model, and the constant average profile gave the highest estimate.
 It is important to recognize that the cargo fleet (including passenger ships), that account for 80% of the installed power (Figure 10), normally have higher engine utilization (load) and number of days at sea, compared to noncargo ships [Endresen et al., 2004]. The energy production (kWh) by the cargo fleet will then be higher than 80%, and could be as high as 90%. Corbett and Koehler  reported typical average specific fuel consumption by cargo and noncargo ships to be respectively 206 and 221 g/kWh, with reference to year 2001. By weighting according to installed power, a weighted average of about 209 g/kWh is obtained. We have applied a value of 220 g/kWh for year 2000, indicating that our model will overestimate the fuel consumption around year 2000. However, the fleet composition with respect to old ships with old engines varies over time and they are normally in operation for several decades (e.g., the average age for civil fleet was 22.4 years in 2006 [LRF, 2005–2006]), with unknown status and maintenance condition. Our somewhat higher specific fuel consumption takes the effect of old engines into account (see section 3.1.3).
 It is difficult to estimate the uncertainty because of limited technical and operational data over the decades. It is important to notice that some of the effects (e.g., engine load versus specific fuel consumption) will drive results in opposite directions, thus giving a representative average although the uncertainty in each parameter is significant. Our applied averages seem to reflect the cargo fleet and we therefore claim that the simplified approach will give reasonable estimates. Including the noncargo fleet in the modeling adds to the uncertainty as these ships have most of the power installed below 5,000 GT (Figure 10) and operate significantly different from cargo ships (e.g., engine load). However, the energy production in this fleet segment is significantly less than for the cargo ships (see above). Taking this into account, we expect ±10% as representative for the error of using a simplified model versus a more detailed approach (uncertainty related to the nonlinear effects).
 The uncertainty in the applied averages arises from the potential bias in data applied and error in average figures due to a limited data material. Our estimates for the uncertainty focus on the variable that account for the highest potential uncertainty. It is assumed that the uncertainty figures are independent and representative estimates for standard deviations. The uncertainties in the calculated averages are based on variation ranges reported in the literature.
 The modeled fuel consumption is sensitive to the applied installed engine power. Our year 2000 modeling of total installed engine power for the entire fleet of oceangoing civil ships above or equal to 100 GT (281,000 MW) corresponds very well with numbers reported by Corbett and Koehler  (280,000 MW). This result indicates a low uncertainty in the calculated average power around year 2000. We expect the uncertainty in average engine power per ship to be up to ±10% around 1970 decreasing to nearly zero for year 2000.
 The model results are also sensitive to the assumed number of days at sea per ship. The uncertainty bound for days at sea is calculated to be ±15% considering typical variations in the main input data. The largest contribution results from the assumed cargo capacity utilization, followed by the assumed ballast factor. We have evaluated our 2000 estimate (181 days at sea) with a calculated average activity profile covering 3,431 AMVER cargo ships [Endresen et al., 2004]. By weighting the reported days at sea for each category (see above) by the actual power installed in the year 2004 world fleet (for the same size categories) (Figure 10), the weighted average number of days at sea is calculated to 205 days. This is some 20 days higher than the applied average. The deviation could be explained with the fact that the AMVER fleet typically covers cargo ships, mostly larger than 3,000 GT in international trade [Endresen et al., 2003]. The average number of days would decrease if smaller cargo ships also were included. An alternative explanation is that the world sea borne trade may be 11% higher in year 2000 than assumed. Our calculated 181 days at sea is based on a trade estimate of 5434 Mt [Fearnleys, 2002], while Review of Maritime Transport [United Nations Conference on Trade and Development, 2006] reports a somewhat higher figure (5983 Mt). Our calculated average number of days at sea then corresponds with the weighted AMVER operation profile. This result indicates that we probably slightly underestimate the average days at sea, if we only consider the cargo fleet. However, operational data indicates less days at sea for the noncargo fleet [Endresen et al., 2004]. The expected bias will decrease up to 2000, as the noncargo fleet have doubled by numbers from 1970 to 2000 (Figure 4). Our best estimate is that the combined uncertainty related to the various input data is within ±25%. The total uncertainty in fuel consumption is estimated to be about ±30%, when the methodological uncertainties are added.
 Auxiliary engines will also contribute to the total fuel consumption and emissions for oceangoing civil ships above or equal to 100 GT, but this is not included in equation (3). Estimates presented indicate that the fuel consumed by auxiliary engines in port and at sea may amount to less than 10% of the total [Whall et al., 2002; Corbett and Koehler, 2003]. Consequently, if auxiliary engines were included, the modeled fuel consumption should have been in the order of +5% higher. The model estimates presented in this study are thus likely to be approximately 15% too low because auxiliary engines and all military ships and vessels less than 100 GT are excluded (see section 4).
4. Marine Sales Versus Modeled Consumption
 Several activity-based studies have reported fuel consumption without including oceangoing ships less than 100 GT. The fuel consumption by these ships is not addressed in the literature, and could be significant. For instance in 1998, the global number of engine powered fishing vessels (decked) was about 1.3 millions vessels [FAO, 2006a], while only some 23,000 of these vessels were larger than 100 GT in year 2000 (Lloyd's Register of Shipping, world fleet statistics and statistical tables, 2000). The fishing fleet less than 100 GT represents nearly half of the installed power for the entire fishing fleet (see section 2.2.2). We may therefore assume that these vessels account for half of the fuel demand of the fishing fleet (∼10 Mt fuel). Norway has approximately 3,000 cargo and service ships between 25 and 100 GT in coastal trade [Statistics Norway, 2000]. We do not have data for the rest of the world fleet less than 100 GT operating mainly in national waters, but we assume that this part represents a consumption of at least the same order of magnitude as the fishing fleet less than 100 GT. We also have to take into account the consumption by the military fleet which consumed some 5 Mt in 1996 [Endresen et al., 2003] and the consumption by auxiliary engines (∼5% of the total, see section 3.2.3). Consequently we expect our activity modeling estimates for the period 1995 up to 2000 to be some 25–35 Mt less than the actual fuel consumption for the entire worldwide oceangoing fleet. We find that the estimated consumption is about 10% too high for this period compared with total estimated sales data. This indicates that the reported sales number for this period may be representative, and not significantly underreported. However, the uncertainty in the activity-based estimates is significant (section 3.2.3), as well as the assumptions used to derive the added consumption of 25–35 Mt fuel.
 Several studies have questioned the reduction in sales for given time periods without considering the important changes in the fleet. This has led to the assumption that significant underreporting of sales can have occurred. However, this study illustrates that improved modeling, with the use of high-resolution time series as input data, gives corresponding trends in modeled fuel consumption and sales numbers. Our results are supported by the general changes in trade and the fleet activity (Figure 4).
 From 1910 to 2000, the oceangoing world fleet of civil ships above or equal to 100 GT grew by number from around 22,000 to 88,000 motor ships, by gross tonnage from 37 to 558 millions, and by cargo transported from about 300 Mt (year 1920) to 5,400 Mt. Oceangoing ships had a yearly consumption of about 80 Mt of coal (corresponding to 56.5 Mt heavy fuel oil) before the First World War. This increased to a sale of about 200 Mt of marine fuel oils in 2000 (including the fishing fleet), i.e., about a fourfold increase in fuel consumption. Of this sale, international shipping accounts for some 70–80%. The fuel-based ship emissions are estimated to 229 Tg (CO2) in 1925, growing to about 634 Tg (CO2) in 2002. The corresponding SO2 emissions are about 2.5 Tg (SO2) and 8.5 Tg (SO2). The CO2 emissions per tonne transported by sea have been significantly reduced as a result of larger and more energy efficient ships.
 We find that the development of fuel consumption from 1970 up to 2000 can be modeled by including the major changes in the fleet size, shift in fuels and propulsion, technical improvements, changes in average operating speed, average sailing distance and demand for sea transport. It is suggested that these key factors are included when performing historical activity-based fleet modeling. The variation in trade patterns over the years for oil is an important factor determining demand for crude oil transportation (e.g., average sailing distances), as well as fuel consumption by this fleet segment and the entire fleet. This study concludes that the growth in the fleet is not necessarily followed by increased fuel consumption, as the complex interaction among the key influencing variables will determine the fuel consumption. We find that the estimated fuel consumption corresponds fairly well with the reported fuel sales from 1970 to 2000, when especially the consumption by auxiliary engine power, ships less than 100 GT and all military ships are included. It is not possible to conclude on the actual uncertainty or bias in the marine sales data on the basis of our findings, but our results and other studies indicate that underreporting may occur. However, our results indicate that the reported sales number for this period may be representative and not significantly underreported, as previous activity-based studies have suggested. Fuel consumption by ships less than 100 GT (e.g., about 1.3 millions fishing vessels today) is important to include when comparing fuel sales with activity-based estimates.
 Interestingly, our results here agree well with our previous activity-based estimates for the year 2000 (if consumption by 45,000 noncargo ships is taken into account) that used an alternative approach and different data sets. However, our simplified model estimates of fuel consumption from 1980 to 2000 are significantly lower than previously reported activity-based studies. By considering alternative input data to our simplified activity-based model, we conclude that the main reason for the large deviation between activity-based fuel consumption estimates is the number of days assumed at sea. Our results indicate that improved activity data on a yearly basis are needed to determine the actual energy demand for the entire world fleet if a simplified activity-based model is to be used. Such data will also significantly reduce the uncertainty for estimates based on more detailed activity-based modeling.
 The preparation of this paper was cofunded by the EU-project QUANTIFY (contract 003893). We sincerely acknowledge Kristin Rypdal at CICERO, Norway, for her support and input during the work with this paper. We would also like to acknowledge Stephen McAdam at DNV for significantly improving the paper.