Work

Considering alternative input parameters in an activity‐based ship fuel consumption and emissions model: Reply to comment by Øyvind Endresen et al. on “Updated emissions from ocean shipping”

Year: 2004
Type: letter
Abstract: [1] The comments of Endresen et al. [2004a] offer differing assumptions and clarifying information that may reduce the estimate of ship emissions by Corbett and Koehler [2003]. The authors claim to substantiate a lower estimate for marine fuel consumption that more closely agrees with published fuel statistics and therefore agrees with all previous studies that have assumed that reported marine fuel statistics are essentially accurate and complete [Benkovitz et al., 1996; Corbett and Fischbeck, 1997; Corbett et al., 1999; Endresen et al., 2003; Olivier and Peters, 1999; Skjølsvik et al., 2000]. Importantly, Endresen et al. do not argue fundamentally against the accuracy of an activity-based methodology; their comments entirely focus on whether the input data obtained or developed for our model are reasonable. In this regard, they focus on exactly the issue identified in our uncertainty analysis, namely, the need for accurate engine activity parameters such as operating hours, load profile, and specific fuel consumption. Their revised parameters indeed reduce the estimate of maritime fleet fuel consumption, but not as much as claimed and certainly within our uncertainty bounds. Here we incorporate their assumptions into our activity-based model to confirm our earlier conclusions that ships listed in international registries use more fuel than reported in international marine fuel statistics. [2] The amount of fuel used by internationally registered vessels is directly related to the magnitude of emissions (mostly uncontrolled and unregulated) generated during energy consumption on either domestic or international voyages. Quantifying ship emissions and characterizing them geographically is important when modeling atmospheric fate and transport [Capaldo et al., 1999; Kasibhatla et al., 2000; Lawrence and Crutzen, 1999], when designing field experiments to study the role of ship plumes in marine boundary layer chemistry [Song et al., 2003], and when defining policy to reduce emissions from these vessels [Corbett, 2003]. By obtaining better vessel activity characteristics, uncertainties in ship emissions inventories can be minimized, thereby improving both atmospheric science and policy insights with regard to ship pollution. [3] Consistent with our uncertainty analysis, Endresen et al. [2004a] agree that activity patterns, such as average engine power (duty cycle), time in operation (engine hours), and fuel consumption rates, are critical to improve the accuracy of emissions estimates for ships. They present results of a model [Endresen et al., 2003, paragraph 9] “based on the number of hours at sea …, statistical relations between size … and engine power for the ship types (container, bulk, general cargo, etc.), distribution of engine types on ship types, bunker fuel consumed per power unit (kW) …, and an assumed average engine load.” This model differs from ours in two important aspects: (1) The Endresen et al. [2003] model depends upon an indirect statistical relationship between known vessel size and inferred main engine power where our model directly considers installed engine power characteristics by vessel type, and (2) their work estimates the number of vessel hours at sea where our model directly inputs engine operating hours, including auxiliary engines (i.e., onboard diesel generators). While these indirect measures are valid, their accuracy relies on correlations of parameters that our model inputs explicitly. (See auxiliary material for additional discussion.) [4] The importance of obtaining empirical and valid measures of vessel activity is also a point of agreement. These data are fundamental to understanding engine load profiles, engine operating hours, and resultant fuel consumption. Moreover, deriving these directly from geographically resolved data is very useful, but introduces important new uncertainties (see section 4). Endresen et al. [2003] have made a significant contribution to the geographic characterization of ship activity in recent work comparing traffic densities derived from the publicly available International Comprehensive Ocean-Atmosphere Data Set (ICOADS) with those derived from a U.S. Coast Guard data set (Automated Mutual-Assistance Vessel Rescue System (AMVER)), made available to Norway under special arrangement. We simply think the power-based approach, focused directly on the installed power, marine fuel properties, and applying specific engine activity levels (duty cycles) that correspond to marine engine data, can achieve better accuracy. Of course, we agree with Endresen et al. that the level of aggregation detail will affect the accuracy; that is why our best estimate applied some 132 different main-engine subgroups according to propulsion type, engine size, fuel type, and vessel service (with an average of only 890 main propulsion engines in each) and treated auxiliary engines as a separate subgroup. [5] Importantly, we agree with Endresen et al. [2004a, section 2.1] that a reasonable estimate for average engine load for cargo vessels is closer to 70%. In fact, this helped identify a typographical error with regard to average engine load in Table 3 of our paper [Corbett and Koehler, 2003]; they reasonably concluded from that table that we assume average engine loads of 80% (as mislabeled), where those data represent typical loads at sea speeds (cruise loads). We produce a corrected version (Table 1). We agree that cruising loads are significantly higher than composite average load profiles over the range of vessel operating conditions. Total fleet fuel consumption is largely determined by ships with the most installed power (large ships); these vessels have rather longer duty periods at high engine output, reducing the potential impact of low loads (including idling). The baseline average load factors that we use in our calculations are nearly the same as those recommended by Endresen et al., typically 70% to 75% average load, based on data provided by operators. The composite average load factors used for our uncertainty analysis are 65% of rated power for cargo vessels and 48% of rated power for noncargo vessels (calculated by obtaining the sum product of power and weighting factors in Table 6 and Table 7 of Corbett and Koehler [2003], respectively). We appreciate this opportunity to discover and recognize this typographical error in our original manuscript. This removes their concern that we assumed a “high utilization of engine power,” since both approaches agree. [6] Some points of clarification are also merited. First, fuel consumption of mostly cargo ships engaged in international voyages, and depicted in geospatial representations of commercial shipping lanes, would be a subset of the activity-based inventory we presented. As shown in Figure 3 of Corbett and Koehler [2003], our best estimates for fuel consumption totaled 289 megatons (Mt) for the world fleet (including cargo (203 Mt), noncargo (45 Mt), and diesel-powered military vessels (41 Mt)). Second, the fleet we analyze represents registered vessels of 100 gross tons or greater that are listed in international ship registries (considered capable of routine ocean voyages); this includes large vessels sailing under all national registries but effectively eliminates most small domestic craft not typically in ocean service such as inland towboats, recreational boats, and harbor ferries. The cargo fleet includes 43,852 vessels with some 53,125 main engines totaling some 215,715 MW of installed power; Endresen et al. [2004a] appear to confuse engine counts and vessel counts. The registry of vessels does not indicate which vessels are engaged in transoceanic service, coastwise ocean transport (sometimes considered domestic), international transportation along rivers and canals, or other specific voyages; in section 5 of Corbett and Koehler [2003] we discuss the importance of continuing research in this regard. Summing the data presented in Table 2 of our study, cargo ship activity (activity most consistent with geospatial data sets used to depict shipping lanes) represents an estimated 65% of total fleet energy demand. Third, the military vessel inventory included only vessels with diesel engines (some 16,109 of 19,646 military vessels), substantially the same data as used in previous inventories; no nuclear-powered vessels (or other naval propulsion systems) were considered in our analysis. [7] Endresen et al. [2004a] were successful in reproducing our estimates for fuel used by cargo ships; by adopting the input data we chose, they estimate ∼216 Mt compared to our best estimate of 203 Mt. We think this agreement lends validity to both models and incorporate their revised assumptions in our model. Our model cannot produce results as low as published fuel statistics, despite adopting their inputs. We summarize and compare results using the alternative inputs suggested by Endresen et al. with our estimates. [8] The major difference between our input parameters is the estimated hours of engine operation. Endresen et al. [2004a] use estimates of vessel days in service, indicating that cargo ships are in service (including port time) between 240 and 300 days per year. At first glance, if we assume that engine operation corresponds to in-service time, this roughly suggests typical annual operating hours between 5700 and 7200 hours, similar to the data we used; however, Endresen et al. develop an argument supporting lower engine hours that assumes no engine activity for much of the time in port and assumes a large number of days (60–120 days) that ships are laid up. According to their comment [Endresen et al., 2004a, section 2.1], “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.” Some of this time includes maneuvering operations for the main engine, included in our estimates of engine operating time and apparently also included by Endresen et al. In Table 2, we show that the engine hours they present could imply that main engines would not operate between 49% and 95% of turnaround time in port, depending on ship size. [9] We do not fundamentally disagree with Endresen et al.'s [2004a, Table 3] discussion of average voyage cycles. We find support in a recent study at the Port of Los Angeles that suggests that ship engines may be stopped during 60–75% of the time in port [Starcrest Consulting Group LLC et al., 2004]; however, turnaround times have decreased by an average of 89% from 1970 to 1998, varying by vessel type from ∼0.5 days for container ships to 2 days for dry bulk vessels [International Labour Organization, 2001; Kahveci, 2000], with an average for all ships of 1.16–1.27 days [United Kingdom, 2002; Whall et al., 2002]. [10] Lower turnaround time statistics may be consistent with annual engine hours we used, but only by increasing the total days in service claimed by Endresen et al. [2004a]. Using the average 248 at-sea days (corresponding to 5952 engine hours at cruise loads) for “very large crude carriers” (VLCCs), reported by Endresen et al. [2004a, section 2.1], and using their estimated noncruising period for petroleum products in their Table 3, we estimate a total of 299 total days in service (248 divided by 83%), representing 51 days in port “turnaround time.” (Note that large tankers have the largest period of time in noncruising mode, per Table 3 of Endresen et al.) If we apply an average of 6500 total engine hours, per information we obtained from industry, then engines operate during turnaround time only 45% of the time (23 of the 51 in-port days); this would be inconsistent with the Endresen et al. claim that engines do not operate during 95% of large vessel turnaround time, unless more days per year were in service. For example, if the vessel days in service were at least 323 days per year, then main engines on this average VLCC would operate for at least 6500 hours even with engines off for 95% during in-port periods. [11] Even accounting for these reductions, their engine hours suggest very substantial periods when the vessel is laid up (i.e., not in active service, out of commission for repair, or in dry dock (see the following sources for these definitions: http://www.nextlinx.com/tools_gloss6L.html, www.ninnescah.org/library/glossaryl.htm, and http://www.trans-inst.org/seawords.html)) or being used as floating storage, such as can be the case for large tankers [Wijnolst and Wergeland, 1997]. We think that it is unrealistic that on average, most ships (except perhaps tankers) would be out of service for two to four months each year; this is certainly inconsistent with our understanding of transportation economics and our own experience. Our study cited industry data that suggested ∼1% of the fleet was laid up or under repair; Lloyds Maritime Information System (LMIS) [2002] listed a total of 9944 two-stroke, low-speed marine engines made by MAN B&W, with 56 engines laid up, 69 engines under repair, and 62 sold for scrapping. Including engines scrapped, this is consistent with the literature cited by Endresen et al. [2004a] regarding fleet productivity. The United Nations Review of Maritime Trade reported that only ∼2.6% of the fleet was considered surplus in 2001, a value that has remained below 3% since 1999; furthermore, the capacity of the active world cargo fleet continues to increase at ∼2% per year [United Nations Conference on Trade and Development (UNCTAD), 2002, 2003]. While this revises our estimate about the fraction of laid-up ships, they would still be in service some 350 days per year on a fleetwide basis; it is impossible for 2.6% of the fleet being out of service to result in a fleetwide average of 60–120 days laid up per ship (see Table 2). [12] For noncargo ships (excluding military), Endresen et al. [2004a] also claim that our estimates of engine hours are too high. Table 3 summarizes assumptions for noncargo at-sea days and engine hours according to their comment. In particular, they claim that we based our estimate of 6000–7000 engine operating hours on activity levels from cargo ships; we did not. However, it may be that their activity information for noncargo ships is more complete than the information we obtained. We acknowledge that some fishing vessels may not operate throughout the year because of the seasonal nature of certain fisheries; however, we understand that many oceangoing fishing vessels move among fisheries to maximize utility. We note that noncargo ships account for only ∼20% of total fleet fuel consumption because they have less installed power, used at lower loads, than cargo ships, so the impact on fuel estimates may be less than implied by Endresen et al. Industry information we obtained produced nearly the same estimates for average load (48%) in the noncargo fleet as reported by Endresen et al. [2004a] for supply vessels (<50%). For military vessels, the Endresen et al. comment does not substantially amend our approach (largely the same as J. J. Corbett's earlier work). Again, we agree with Endresen et al.'s main claim that any model of vessel fuel use and emissions must explicitly consider engine and fuel characteristics and that improvements in the understanding of noncargo vessel activity, particularly engine hours and load profiles, are needed; our methodology explicitly incorporates these parameters and can be adjusted with better activity information. [13] A second set of assumptions we modified in our attempt to produce Endresen et al.'s [2004a] annual fuel consumption estimate of 170–200 Mt related to specific fuel consumption (sfc) rates. Although not explicitly included in their comment, other feedback we have received suggested that the fuel consumption rates we used were different from those typically provided by engine manufacturers' specification sheets. They are different. Engine data sheets represent new-engine conditions, which are well understood to deteriorate during service. Moreover, specific fuel consumption is much higher at loads different from the typically 85% optimal; toward full load and lower loads there is an increase in specific fuel consumption rates for most marine engines. Published sfc rates based on standard test conditions (typically using a distillate fuel rather than marine residual fuels) simply do not represent the existing fleet of vessels with an average age greater than 15 years [Koehler, 2003a, 2003b]. Although we believe that sfc rates in our original analysis are correct, we adopt average sfc rates of 195 g/kWh for main engines on cargo vessels and 215 g/kWh for main engines on noncargo vessels and for auxiliaries in this analysis, because it helps reduce our model estimates. [14] Model results using the alternative assumptions are presented in Table 4 and Figure 1. The alternative assumptions indeed reduce the estimates, but not more than ∼14–16%. To fall within the fuel consumption ranges claimed by Endresen et al. [2004a], reductions would have to be more than 23% for the registered fleet, more than 26% for cargo ships, and more than 31% for the world fleet including military vessels. Clearly, these results are consistent with the idea, and support the conclusion, that our model predicts more fuel consumption than can be accounted for in international fuel sales statistics, even when some 17% (25 Mt) are added for domestic usage (per Endresen et al. [2004a, Appendix A]). This finding under Endresen et al. inputs continues to suggest significant domestic fuel consumption (either during “domestic voyages” according to United Nations Framework Convention on Climate Change (UNFCCC) definitions but poorly documented as marine vessel consumption, or with significant potential error in classification of vessel service, per our discussion in section 4.3 of Corbett and Koehler [2003]). [15] Importantly, our revised model results show that the fleet of registered ships (not including military) may consume between 175 Mt (cargo ships only) and 209 Mt (cargo and noncargo ships) under alternative activity and fuel consumption assumptions. This results in rather good agreement with the 170–200 Mt annual fuel use claimed by Endresen et al. [2004a] for the entire registered fleet. (However, this leaves open the question of military fuel use; nonetheless, our estimate for military vessels at ∼14% of world fleet energy demand may not be detailed enough to make strong inferences.) [16] Points of emphasis in the comment to our paper [Endresen et al., 2004a] are that engine characteristics are essential to evaluating fuel consumption and emissions and that these characteristics vary by engine type and size, vessel type and size (including age), and fuel type; these points reinforce the conclusions of our paper. Moreover, these points support the use of activity-based models that directly evaluate the parameters that determine fuel usage and emissions; in fact, Endresen et al. [2003] consider different vessel type and size parameters for the vessels represented geospatially in data sets such as AMVER and ICOADS, with interesting results. We note that the inputs we used are not entirely assumptions, but average engine running hours and corresponding average engine load across a year for characteristic ship types provided by engine manufacturers and operators. While we think that the average engine hours for cargo vessels documented in our earlier paper [Corbett and Koehler, 2003] and discussed again above represent industry practice, we acknowledge that the Endresen et al. [2004a] data for noncargo vessels (especially fishing vessels) may be more complete. Furthermore, we would expect of input parameters if studies were with such as in or other However, the most important point is that input parameters are to the model and can be revised if better data Engine load factors specific fuel consumption account for less difference than claimed by Endresen et al. [2004a]. As our model used nearly the same average engine load factors as Endresen et al. By adopting specific fuel consumption rates similar to those in the we estimate only about a in total fuel estimates. We believe that as engine activity data are included explicitly among vessel operating characteristics, fleet will better this will policy internationally to develop [International Maritime and 3] methodology to the of a ship in of a Endresen et al. [2004a] use AMVER statistics to lower engine operating hours, and there is very good between AMVER data and other however, perhaps they are a in the of these data in representing all internationally registered ships at all In fact, the AMVER at-sea days are not to the engine assumptions we used for two First, vessels in AMVER may their activity, a point in the comment by Endresen et al. At AMVER data represent a lower where engines ships operate at least this Second, AMVER data may many ships or voyages, particularly since [Endresen et al., 2003, paragraph in AMVER is to ships over on a voyage of hours or This that the data, more would vessels engaged in coastwise or sea shipping because (1) these vessels fall below the applied and (2) coastwise and sea shipping may be less than the Importantly, the additional fuel to operate vessels or not represented well by AMVER data (or other geospatial data for ship may account for the additional fuel consumption, consistent with the discussion in section 4 of Corbett and Koehler [2003]. The AMVER data set may the distribution of cargo ships during international very and an activity-based methodology to estimate fuel use and emissions can be used to the geographic information derived from data sets AMVER and We simply suggest that it is important to consider the of the AMVER data as well as any when the is to use those sets to of any such data set an important by the that AMVER data are not available AMVER As we recommended in section 5 of Corbett and Koehler [2003], the toward large ships represented by the AMVER data suggests a need for research to evaluate the uncertainty in ship traffic at a using both and if only to where data may be to geographic general insights for atmospheric from the documented in the recent by and by Endresen et al. [2003], of which specific input assumptions are First, inventories can apply the same general activity-based modeling used to estimate energy consumption and emissions for and but data assumptions duty engine and engine operating hours are the most important more critical than the uncertainty in factors for certain This suggests that models where these parameters are and adjusted according to vessel type and size may the best results and inventories at to be modeling can emissions from oceangoing vessels in international that operate in both international and domestic Second, the geospatial of vessel traffic used to represent emissions in atmospheric models best represent vessels using major shipping This that Corbett et al. Table 2, section and Figure total fleet emissions by vessel type by on published fuel statistics of the world fleet of ships greater than 100 gross registered it also that total fleet emissions cargo, and military ships) as estimated by our activity-based model to either AMVER or traffic would emissions in those [Corbett and Koehler, 2003]. In other these data sets (either or are of total ship they represent a of total fleet activity, to noncargo vessels of AMVER vessels per Table of Endresen et al. [2004a] but of ships in the registered may not represent vessels, and may vessel activity and major In this regard, the recent work by Endresen et al. to represent vessel size and type is an important to earlier as by Endresen et al. [2004a], emissions to resolved may be with a fuel usage value to or lower than our cargo ship estimate of 203 Mt. As the emissions to data sets be adjusted to represent the inventories for the vessel types and Third, more research is to represent vessel traffic activity and in domestic activity, coastwise voyages, and inland or sea Clearly, the marine is not to or international for vessels, but to fuel consumption during international between atmospheric and policy need is information about total energy consumed and whether or not a vessel's fuel is considered international or In this regard, some of fuel consumption in the world fleet is not accounted for in published marine fuel statistics on whether the national fuel consumption estimates by Endresen et al. [2004a] or the estimates in section 4 of Corbett and Koehler [2003], respectively). Endresen et al.'s 2004a] work and our work agree on this general and disagree on the magnitude of fuel consumed but not documented in international marine fuel statistics. the general insights from atmospheric models will not be fundamentally to improve their accuracy. In this regard, we believe that an inventory that includes not only best estimates but also and lower explicitly related to input parameters both the and policy We this with Endresen and and as these estimates and are to data that can use; this can reduce large uncertainty particularly where and data may be more and better The most result of the ranges in inputs is not that the as high as they but that lower may not estimates based on fuel statistics By more inventories of ship understanding of the of in ship plumes [Song et al., 2003] and by emissions along shipping et al., the result better of whether and where ship emissions to at and material for this a of indirect based on fuel and and a table AMVER data that only of vessels in AMVER are and The is not for the or of any supporting information by the than be to the corresponding for the
Language: English
Open Access status: bronze
OpenAlex ID: W2093211856
Cites: 17
Cited by: 65
Related to: 10