Environ. Sci. Technol. 2002, 36, 1299-1306
Emissions from Ships in the Northwestern United States JAMES J. CORBETT* Marine Policy Program, College of Marine Studies, 305 Robinson Hall, University of Delaware, Newark, Delaware 19716-3501
Recent inventory efforts have focused on developing nonroad inventories for emissions modeling and policy insights. Characterizing these inventories geographically and explicitly treating the uncertainties that result from limited emissions testing, incomplete activity and usage data, and other important input parameters currently pose the largest methodological challenges. This paper presents a commercial marine vessel (CMV) emissions inventory for Washington and Oregon using detailed statistics regarding fuel consumption, vessel movements, and cargo volumes for the Columbia and Snake River systems. The inventory estimates emissions for oxides of nitrogen (NOx), particulate matter (PM), and oxides of sulfur (SOx). This analysis estimates that annual NOx emissions from marine transportation in the Columbia and Snake River systems in Washington and Oregon equal 6900 t of NOx (as NO2) per year, 2.6 times greater than previous NOx inventories for this region. Statewide CMV NOx emissions are estimated to be 9800 t of NOx per year. By relying on a “bottom-up” fuel consumption model that includes vessel characteristics and transit information, the river system inventory may be more accurate than previous estimates. This inventory provides modelers with bounded parametric inputs for sensitivity analysis in pollution modeling. The ability to parametrically model the uncertainty in commercial marine vessel inventories also will help policy-makers determine whether better policy decisions can be enabled through further vessel testing and improved inventory resolution.
Introduction Recent inventory efforts have focused on developing nonroad inventories for emissions modeling and policy insights (1). Characterizing these inventories geographically and explicitly treating the uncertainties that result from limited emissions testing, incomplete activity and usage data, and other important input parameters currently pose the largest methodological challenges. Since directly monitored data for nonroad emissions, fuel consumption, and activity level generally are not available, the development of nonroad inventories must rely on a number of assumptions that introduce significant uncertainty. To date, most nonroad inventories have applied national or regional average values that suffer from two important weaknesses. First, these assumptions often are generated through engineering modeling or manufacturer test data rather than in-service sampling. This means that certain assumptions may not apply to actual nonroad vehicle operations. Second, emissions * Phone: (302)831-0768;
[email protected]. 10.1021/es0155985 CCC: $22.00 Published on Web 02/16/2002
fax:
(302)831-6838;
2002 American Chemical Society
e-mail:
factors, fuel consumption data, and activity level information typically represent national averages (or regional averages at best) and therefore cannot capture the potential variability of nonroad operations at the local level. Nonetheless, these inventories are demonstrating that nonroad emissions from certain sources and locations may be more important than previously considered (1). One important example is nonroad emissions from water transportation. This mode includes freight transportation by inland river towboats pushing barges, tugboats and other working harbor craft, deep-draft oceangoing vessels, passenger ferries, and fishing vessels. Commercial marine vessels (CMV) represent one of the least-studied nonroad transportation modes in terms of emissions inventories. However, recent research has begun to evaluate these emissions internationally and domestically (2-4). This paper briefly reviews current inventories and reports characterizing marine vessel emissions and explicitly compares two methodologies for estimating emissions from commercial marine transportation on a geospatial basis. The first inventory uses detailed statistics regarding fuel consumption, vessel movements, and cargo volumes in the Columbia and Snake River systems in Washington and Oregon. This inventory is compared to a national inventory of waterborne commerce NOx emissions (i.e., emissions directly resulting from cargo movements) from vessels operating in U.S. waterways (4). In addition to developing a best engineering estimate, this work developed upper- and lower-bound estimates that are geospatially resolved to enable sensitivity analysis in regional haze modeling to be conducted by the Pacific Northwest’s Regional Technical Center (RTC) later in 2002. Research and policy insights from the characterization of uncertainty are presented, particularly the need for focused research to characterize air emissions from marine engines on the most heavily traveled water routes. Review of Current Inventories. Current characterizations of marine vessel activity and emissions have tended to focus on vessel activity within ports and applied either power- or fuel-based emissions factors. The following studies are reviewed (with more discussion in the Supporting Information) as background to the work presented here: (i) An inventory of commercial vessel activity developed for deep-sea ports in the United States. (5). Although this inventory focuses on activity and does not report an emissions inventory (EI), it was developed for inventory purposes. However, it does not report results geographically and focuses exclusively on in-port (or near-port) emissions, limited in the northwest region to the Puget Sound. This prevents its direct application for the purposes of research reported in this paper. (ii) An analysis of emissions and fuel consumption data for CMV (6). This analysis provides detailed statistics regarding the emissions factors that could be applied to vessel activity data. The report does not address geographically specific information such as expected load profiles (or duty cycles) during maneuvering. It suggests regression equations that can be used to estimate emission factors for marine vessels, and these factors are within the parametric bounds applied in this paper. Where that report treats uncertainty more generally through regression, the methodology presented in this paper includes an explicit uncertainty assessment of the input parameters. (iii) Locally produced inventories for Oregon and Washington. Because of the focus on the northwestern U.S., this paper considers Oregon’s emission inventory of CMV for VOL. 36, NO. 6, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1299
TABLE 1. Comparison of Average CMV NOx Factors with Current In-Use Fleet Averages of Other Diesel-Based Transportation Modes
diesel-powered vehicle type medium-speed marine enginea slow-speed marine enginea truck diesel engineb locomotive engineb nonroad vehicle engineb EPA tier 2 standard for new CI marine enginesc
NOx EF (kg of NOx/ t of fuel)
NOx EF (g/kWh)
57 87 33 81 50 33
12 17 7 18 11 7.2
a Lloyd’s Marine Exhaust Emissions Programme (20). b AP-42: Compilation of Air Pollutant Emission Factors (16). c U.S. EPA (21).
1996 (7), and Washington’s inventory prepared for their state implementation plan (SIP) (8). The spreadsheet tends to report emissions that are comparable to the estimates in the CMV inventory reported here. These emissions estimates appear to be within 12-30% of the baseline estimates developed in this report. This is considered good agreement given the differences in how emissions are calculated and assigned to locations by county. (iv) An inventory of emissions from waterborne commerce data in the U.S. waterway network (4). This research describes a methodology that is essentially similar to the results that may be expected by combining CMV activity data (5) with fuel-based emissions factors (6). In this earlier work, waterborne commerce emissions were estimated for various U.S. regions and types of traffic, including oceangoing (international), coastwise (domestic), inland-river system, and Great Lakes (4). This inventory only estimated emissions from cargo movements reported publicly by the U.S. Army Corps of Engineers (USACE) (9) and did not attempt to characterize emissions from all commercial vessel movements (such as vessel-assist tugboats). This analysis showed that more than 90% of all emissions in U.S. waters occur in shipping channels outside of port regions, either on rivers or within 322 km of shore. NOx emissions from river commerce in the top 20 states with waterborne trade account for 65% of total waterborne commerce emissions in those states. In contrast, 72% of SOx emissions in these states occurs along coastal (ocean or Great Lake) areas where high-sulfur fuels are more commonly used. On a national basis, that study suggested that NOx emissions from waterborne commerce equaled 4% of all U.S. transportation emissions, more than double previous nationwide inventories of vessel emissions. This national inventory produced a geographically resolved depiction of emissions, using a geographic database of navigable waterways. Sources of uncertainty were also discussed (summarized below). This methodology is applied in this paper to estimate emissions in waterways other than the Columbia-Snake River system. Evaluation of Emissions Factors and Vessel Activity Data. CMV emission factors have been developed and reported based on a limited number of vessel tests. However, the comparisons between various sources are generally good (see Table 1). The uncertainties in emissions factors are dominated by the limited number of marine engines for which test data has been reported. However, agreement between average emissions factors is generally within similar confidence intervals. The exception to this may be the Analysis of Commercial Marine Vessels Emissions and Fuel Consumption Data (6) report. It suggests somewhat lower emissions values from its statistical regressions than reported by other sources. However, that study appears to ignore physical relationships from an engine-based perspective, reporting only post-hoc relationships that combine fuel types 1300
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 6, 2002
and engine types within a vessel class. Given the general good agreement overall, bounding can be performed; this is discussed in the Updated Methodology section. In short, the literature suggest that the uncertainty bounds presented in this paper are appropriate. Vessel type and activity data are more variable, and the current studies do not adequately distinguish between vessel types and activity. The Commercial Marine Activity for Deep Sea Ports in the United States (5) study attempts to report significant detail for the Puget Sound region but remains focused on port activity. Moreover, that report suggests loads (or duty cycle parameters) that may not represent vessels on particular routes between ports. In general, duty cycles are variable by vessel type and are not well understood, and assumptions can contribute significant errors if not validated (10). This uncertainty applies generally to all of the previous inventories. However, the TVA Barge Costing model has performed well in approximating engine loads by geographic location. For that reason, it was adopted here for the inventory of the Columbia-Snake River system. Significant Sources of Uncertainty, Bias, or Error. In general, CMV inventories suffer from the same gaps in understanding as other nonroad sources, arguably with even greater problems. Most nonroad inventories have applied national or regional average values that suffer from two important weaknesses. First, these assumptions often are generated through engineering modeling or manufacturer test data rather than in-service sampling. This means that certain assumptions may not apply to actual nonroad vehicle operations. Second, emissions factors, fuel consumption data, and activity level information typically represent national averages (or regional averages at best) and therefore cannot capture the potential variability of nonroad operations at the local level. Each of the studies reviewed above encounters to varying degrees these sources of uncertainty, bias, or error. Only the last study attempts to directly quantify these factors in terms that can be propagated through subsequent modeling or policy analysis. In the previous work by Corbett and Fischbeck (4), at least four sources of uncertainty were identified. They include (i) the use of average emission factors based on limited measurement of in-use marine engines; (ii) the simplified duty cycle assumptions; (iii) the uncertainties about the actual distance of cargo movements; and (iv) the use of “average vessel” assumptions, particularly in inland rivers. It was shown that these errors could contribute to an underestimate by as much as 70% in some river systems, when the “topdown” fuel consumption was compared to a “bottom-up” estimate provided by the Tennessee Valley Authority (TVA) Barge Costing Model (11, 12).
CMV Emissions in Northwestern U.S. Region In an attempt to address these uncertainties, two methodologies for estimating emissions for commercial marine transportation are compared. The first inventory uses detailed statistics regarding fuel consumption, vessel movements, and cargo volumes in the Columbia and Snake River systems in Washington and Oregon. This inventory is integrated with Northwest regional emissions included in a national inventory of waterborne commerce NOx emissions (i.e., emissions directly resulting from cargo movements) from vessels operating in U.S. waterways (4). For brevity in reporting, the discussion that follows applies directly to NOx estimates. A brief section that details additional methodologies to produce CMV inventories for PM and SOx follows and is further addressed in the Supporting Information. Updated Methodology and Calculations for River System. This analysis begins with the bottom-up fuel consumption estimates for 1999 from the TVA Barge Costing Model (11) to estimate NOx emissions for the Columbia and Snake
a Base case assumes 80% of max-rated horsepower at full throttle and 20% of rated horsepower during maneuvering through locks. Lower bound assumes 75% of max-rated horsepower at full throttle and 15% of rated horsepower during maneuvering through locks. Upper bound assumes 95% of max-rated horsepower at full throttle and 20% of rated horsepower during maneuvering through locks. Data provided by Chris Dager, Tennessee Valley Authority, March 16, 2001. b Data obtained for this analysis was provided by TVA in gallons and is reported in those units. Conversion to SI units was made during the volume-to-mass conversion.
45 820 627
7 520 4 302
27 867 116 17 953 511
3 218 5 934
35 973 283 21 891 090
3 396 2 537
14 082 193 38 119 868
6 562 3 770
23 010 323
2 792
15 109 545 totals
1 479 206 120 110 1 359 096 1 157 034 94 037 1 062 997 101 283 1 134 310
1 235 594
4 114 274 314 532 3 799 742 3 230 477 247 212 2 983 265 266 279 3 192 746
3 459 026
4 383 033 317 589 4 065 444 3 428 874 248 668 3 180 205 268 554 3 418 962
3 687 516
totals
2 527 632 1 576 342 91 164 640 2 582 711 29 058 104 665 532 1 440 914 13 811 558 886 914 24 102 855
all others towboats
1 862 101 135 428 77 353 82 1 695 797 4 955 249 1 979 135 1 237 901 71 897 502 2 031 831 22 829 698
totals all others
522 354 1 131 587 10 903 438 697 650 18 934 844 1 456 781 106 314 60 994 64 1 334 181 3 894 854
towboats totals
570 566 1 185 218 12 095 466 728 418 19 873 673
all others towboats
1 562 009 112 583 66 792 68 1 447 151 4 172 132
river name
Snake, OR/WA/ID Columbia entrance, OR/ WA Willamette above Portland and Yamhill, OR Columbia at Bakers Bay, WA Lower Willamette Columbia & Lower Willamette below Vancouver, WA/OR Columbia between Vancouver, WA, and Dalles, OR Columbia above the Dalles Dam or two McNary L&D Columbia tributaries above McNary L&D to Kennewick Columbia between Wenatchee & Kettle Falls, WA
2 132 576 1 297 801 78 887 534 2 175 568 24 045 805
upper-bound estimates lower-bound estimates model base case
TABLE 2. 1999 Fuel Use (gal) for Columbia and Snake Riversa,b
River systems. The methodology presented here relies on government statistics about waterborne commerce movements and applies an engineering approach to calculate vessel emissions. Like the top-down approach of earlier work, this methodology has certain advantages in that the statistics are geographically resolved, thus enabling a locally detailed inventory with less effort than traditional bottom-up approaches. As with all modeled inventories, this bottom-up approach also poses a potential for significant uncertainty. However, a carefully constructed analysis can describe this uncertainty sufficiently to provide a robust inventory for policy decision making at local, regional, national, and even international scales. Moreover, the techniques of uncertainty analysis can provide focused attention on those aspects of the inventory that require additional research attention. The TVA model was developed to evaluate fuel usage by Inland River segment for fuel tax purposes. It combines statistics about vessel characteristics and cargo movements to estimate fuel consumption for each inland waterway segment. Rather than estimating fuel consumption based on cargo ton-miles (as the previous study did), this model is based on actual distances transited by vessels on specific segments of the inland-river system. A comparison of fuel consumption using cargo ton-miles and actual vessel-transit distances for the Columbia, Snake, and Willamette Rivers showed that using cargo ton-miles as the basis for fuel consumption underestimated fuel usage by more than 50%. Moreover, the TVA model has very good accuracy in estimating national fuel consumption from operating characteristics of CMV transiting inland rivers (including underway speed and idle time at locks). Comparing model output with the Inland Waterways Trust Fund’s fuel tax receipts for 1999 show that, on a national scale, TVA’s Barge Costing Model is accurate to 0.8%; since 1996, the model errors have not exceeded 2.1% (13). Table 2 presents the fuel consumption data from the Barge Costing Model for the Columbia and Snake River systems. This represents the output of the model using the bestestimate input parameters in the TVA model. Specifically, the base-case scenario assumes that the vessel power at full throttle is 80% of maximum-rated horsepower and that a maneuvering vessel in the inland river (i.e., during transit through locks) operates at 20% of rated horsepower. (Additional vessel traffic detail is presented in Tables 1 and 2 in the Supporting Information.) However, it is possible that some vessels operate under different conditions. Therefore, the model was re-run using both upper- and lower-bound scenarios. The lower-bound scenario uses a full-throttle of load 75% of rated power and a maneuvering load of 15% of rated power (which is the lowest power expected to provide adequate rudder control). The upper-bound scenario uses the same 20% of rated power for maneuvering but increases full-throttle operations to 95% of rated horsepower. Using the bottom-up fuel consumption estimated by the Barge Costing Model, one can apply fuel-based emission factors to construct an emissions inventory. First, a conversion is made from volume of fuel (provided by TVA in gallons) to mass of fuel (in metric tons or t). This requires an assumption of the fuel density, which can vary within specifications (14-16). Second, the mass of fuel consumed annually is multiplied by a fuel-based emissions factor (representing NOx as NO2) to estimate the annual NOx emissions. This emission factor also can vary and is based on limited sampling of in-use marine engines. Table 3 presents best estimates, lower- and upper-bound estimates for fuel density and fuel-based NOx emissions factors. Towboats are assumed to be more homogeneous in both fuel selection and engine type, so they have narrower bounds on fuel density and NOx factors than other vessels (which include deep-draft self-propelled vessels, harbor tugboats,
VOL. 36, NO. 6, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1301
TABLE 3. Fuel Density and NOx Emissions Parameters Used in Calculationsa fuel density
best estimate (kg/L)
lower bound (kg/L)
upper bound (kg/L)
fuel density for towboats (16) fuel density for all other vessels (15)
0.84 0.84
0.84 0.84
0.94 0.99
emissions factors
kg of NOx/t of fuel
kg of NOx/t of fuel
kg of NOx/t of fuel
NOx emissions factor for towboats (20, 22) NOx emissions factor for all other vessels (16, 20, 22)
57 59
56 56
63 96
a
NOx emissions factors are reported in mass terms using the molecular weight of NO2.
TABLE 4. Bounded Inventory Results for Oxides of Nitrogen (kg of NOx/yr) model base case river name Snake, OR/WA/ID Columbia entrance, OR/WA Willamette above Portland and Yamhill, OR Columbia at Bakers Bay, WA Lower Willamette Columbia & Lower Willamette below Vancouver, WA/OR Columbia between Vancouver, WA, and Dalles, OR Columbia above the Dalles Dam or two McNary L&D Columbia tributaries above McNary L&D to Kennewick Columbia between Wenatchee & Kettle Falls, WA totals
towboats
all others
towboats
all others
totals
towboats
all others
totals
107 222 2
387 235 14
264 19 11
95 205 2
359 224 13
338 25 14
121 261 3
458 286 17
0.01 262 756
0.09 136 3 716
0.10 394 4 358
0.01 242 706
0.08 126 3 432
0.09 368 4 138
0.01 307 898
0.10 161 4 369
0.12 468 5 267
620
50
668
576
45
621
737
58
794
579
50
627
541
45
586
689
57
746
206
19
224
193
17
210
246
22
268
0.50
0.71
1.19
0.46
0.62
1.08
0.58
0.78
1.36
2 739
4 303
6 909
2 552
3 968
6 520
3 254
5 051
8 305
Results and Discussion Most mobile source inventories consider the emissions on a per vehicle basis. However, with defined waterway routes (like highways) these estimates can be represented as line sources on a per waterway kilometer basis. Calculating annual NOx emissions using these “stationary” units and representing them geographically provides a representation of the emissions intensity from commercial marine activity. Emissions are assigned to specific waterways geographically by multiplying the weighted-average of each waterway link by the total NOx estimated for that river segment. For example, the Columbia and Lower Willamette River segments below Vancouver contain 18 links in the USACE waterway network. The largest link account for 17% of the total length of this river segment, so the annual NOx emissions estimated for that link would equal 17% times the total NOx estimated for the waterway segment. Using a geo-referenced data set, such as the USACE waterway network (9), this can be represented in a GIS platform. A map of emissions intensity by waterway is shown for the base-case data in Figure 1. Please note that the scale presented is nonlinear and was calculated automatically by the software according to “natural breaks” in the data, which highlights waterway segments of greater and lesser emissions intensity. Clearly, the Columbia and Lower Willamette Rivers below Vancouver, WA (near Portland, OR), to the mouth of the Columbia River have the highest NOx emissions per waterway km. This is expected because this river segment can accept deepwater oceangoing ships in addition to towboats and other river craft. It is worth observing that 9
totals
upper-bound estimates
283 20 12
etc.). Table 4 presents the base-case, lower-bound, and upperbound estimates, respectively, of NOx emissions for the Columbia and Snake River systems.
1302
lower-bound estimates
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 6, 2002
annual NOx emissions from most (55%) of the waterway links in the Columbia and Snake River system are below 3 kg of NOx/waterway km; the numbers in parentheses represent the number of waterway links at each scale category. Uncertainty Analysis of River System Inventory. The parametric analysis developed above provides three scenario estimates that can be used in air pollution models and by policy-makers. However, it does not directly provide adequate information about which input parameters are most uncertain. To evaluate this, the input parameters are varied under a Monte Carlo simulation by sampling from distributions for each of the input parameters. Triangular distributions were defined for the best estimates, lower bounds, and upper bounds for each input parameter (fuel consumption, fuel density, and emissions factor) because values toward the middle of the range are considered more likely than values near either extreme (17). Figures 2 and 3 show results of 10 000 iterations under Monte Carlo simulation, with results ranging from 6818 to 9119 t of NOx/yr. The uncertainty analysis shows that the base-case estimate of 6909 t of NOx/yr would fall at the bottom of the uncertainty distributions. This is a result of more conservative assumptions (see Table 3) in the base-case scenario than in the uncertainty analysis. The upper-bound estimate of 8305 t of NOx/yr shown in Table 4, calculated parametrically, falls at the 92nd percentile in the uncertainty analysis, showing consistency between the parametric analysis and Monte Carlo simulation. A sensitivity analysis of the simulation identifies three primary inputs that could most improve the robustness of the inventory calculations. These inputs are as follows: (i) the fuel consumption for nontowboat vessels (mostly deepwater draft vessels with tug assist) on the busiest river segment; (ii) the fuel density factor for towboats; and (ii) the
FIGURE 1. Commercial marine vessel NOx emissions (1999), base-case scenario.
FIGURE 2. Monte Carlo simulation frequency chart of annual NOx totals for Columbia and Snake River systems.
FIGURE 3. Cumulative distribution of NOx uncertainty for Columbia and Snake River systems.
NOx emission factor. As seen in Figure 4, uncertainty in each of these input assumptions are positively correlated with the total annual NOx estimates; in other words, varying these assumptions from the base case tends to increase the NOx inventory estimates for the Columbia and Snake River systems. Comparison with Previous Inventory Work. These estimates are significantly higher than previous estimates for the Columbia and Snake River systems, which relied on cargo ton-miles to derive emissions (4). In fact the 6909 t of
NOx/yr estimated in this work for these river systems exceeds the total emissions for all waterborne commerce in Washington and Oregon estimated previously (4890 t of NOx/yr). This is expected because the current method uses actual vessel transits and distances traveled to estimate fuel consumption for all vessel traffic, where the previous analysis assumed towboats were the dominant commercial vessel type on all river segments. Comparing only the NOx from towboats in the current work (2739 t of NOx/yr) with the previous study (which VOL. 36, NO. 6, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1303
FIGURE 4. Results of sensitivity analysis showing top 10 inputs by rank correlation. estimated 1919 t of NOx/yr for the same river segments) provides much better agreement, although the earlier work still underestimates emissions by 43%. In other words, the previous inventory analysis appears to have defined a lowest bound case, at least in the Northwest region. This can be explained by summarizing the main differences between the methodologies. The methodology developed here captures both empty and full vessel movements, includes tug assist of deepwater vessels in the Columbia and Lower Willamette Rivers below Vancouver, and accounts for other vessels with higher horsepower (such as self-propelled bulk and tanker vessels).
that approximately 65% of the national waterway NOx emissions were attributed to rivers (4).
With these comparisons in mind, the previous inventory methodology was applied to all waterways (including coastal waterways) in Washington and Oregon other than the Columbia-Snake River system. Base-case estimates for waterways not in the Columbia-Snake River system use the same calculations as previously published (4) because coastal data was not available in the detailed format needed. Lowerbound NOx estimates were set at 30% of base-case values for these waterways. Similarly, upper-bound NOx estimates were set at 2.6 times the base-case for coastal waterways, representing the difference between prior for the river system and the current approach. These bounds may be somewhat arbitrary because they only represent differences reported previously between the base-case and bottom-up approach using the TVA data. However, this approach is consistent with previously reported uncertainty ranges for this data set (0.3-3 times the best estimate).
Particulate Matter and Oxides of Sulfur. The CMV inventory prepared for NOx was used to derive inventories for PM and SOx. Ratios of emissions between NOx and PM, and between NOx and SOx were calculated and the georeferenced data sets for NOx were multiplied by these ratios. For the Columbia and Snake River system, PM emissions are estimated to be 285 t/yr, and SOx emissions are estimated to be 3156 t/yr. For the two-state region overall, the basecase estimates for PM and SOx (including river and coastal segments) is 443 t of PM/yr and 4672 t of SOx/yr. Summary tables of bounded results and results of uncertainty analysis are presented in Tables 3 and 4 and Figure 1 of the Supporting Information. Because the PM emissions factors for marine engines are relatively less well understood currently, a single ratio was used to make this adjustment. The NOx to PM ratio is calculated simply by dividing the average of NOx factors by the average of PM factors reported for large, uncontrolled diesel engines. Three NOx to SOx ratios were estimated to correspond to the lower-bound, base-case, and upper-bound estimates. For the base case, the NOx to SOx ratio assumed a 1.5% fuel-sulfur content for nonriver segments. This assumption is considered to be conservative (low), but without actual fuel quality data the estimate is considered to be adequately realistic. The lower-bound NOx to SOx ratio used a fuel-sulfur content of 1.0%, and the upper-bound NOx to SOx ratio used a 2.7% fuel-sulfur content, respectively.
Some of the insights that this comparison offers are worth noting. For the two-state region overall, the base-case estimate (including river and coastal segments) is 9880 t of NOx/yr, with 6909 t of NOx/yr in the Columbia-Snake River system. This suggests that the river system, which accounts for 37% of waterway distances in these two states, produces about 70% of the NOx emissions (although better inventories for coastal waters in the future may adjust this ratio). While surprising, this result is similar to previous research that found 1304
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 6, 2002
Moreover, port regions may only account for 7-15% of NOx emissions from all waterways in the Washington-Oregon region; this suggests that most of the emissions are outside of port areas where inventories have focused previously. This rough estimate for port versus nonport waterways is made by assuming that each of the seven or so major ports in these states account for about 40 km of waterway (20 km above and 20 km below each port). The ratio of port to nonport waterways could be confirmed using more specific port data to estimate the fraction of waterway within port regions.
For both PM and SOx, the lower and upper bounds in nonriver segments include the same uncertainty ranges of 30% to 2.6 times the base case. For SOx, the combination of fuel-sulfur adjustments and other bounding results in a wider range between lower and upper bounds; this is considered appropriate given the lack of fuel quality data.
Discussion This analysis estimates that annual emissions from marine transportation in the Columbia and Snake River systems in Washington and Oregon equal 6909 t of NOx/yr, 285 t of PM/yr, and 3156 t of SOx/yr. This base-case estimate is 2.6 times greater than previous NOx inventories for this region (4). By relying on fuel consumption estimates modeled in a bottom-up calculation that includes vessel characteristics and transit information, this inventory is considered to be more accurate than previous estimates. For the two-state region overall, the base-case estimate (including river and coastal segments) is 9880 t of NOx/yr, 443 t of PM/yr, and 4672 t of SOx/yr. Estimating the emissions intensity on a per waterway km basis shows that the Columbia and Lower Willamette River systems have the greatest NOx emissions density. This is expected since this segment accounts for the greatest amount of vessel traffic (60% more than the next busiest segment). The NOx emissions per kilometer of waterway can be compared to on-road emissions per kilometer of highway. As a first-order assumption, an average automobile emission rate of 1 g of NOx/mi (∼0.6 g/km) and an average heavy-duty truck emission rate of 8 g of NOx/mi (∼5 g/km) is assumed, reflecting a rough fleet-average factor for automobiles and trucks built since the 1980s (16). According to 1999 data from the Southwestern Washington Regional Transportation Council (RTC), some 126 000 vehicles cross the Columbia River on Interstate 5 and 132 000 vehicles cross the river on Interstate 205 during an average weekday (18). The RTC also reports that the percentage of heavy-duty truck traffic is between 4% and 7.9% at peak PM hours on each of these sections Interstate 5 and Interstate 205 (19). According to this inventory, some 26 t of NOx/waterway km are emitted annually from vessel traffic along the Columbia and Lower Willamette River segments. (This would be roughly equivalent to the annual emissions from 116 000 automobiles/day or some 14 000 heavy-duty trucks/day.) Using the bridge traffic and diesel-truck percentage data from the RTC and the base-case estimates calculated in this study, commercial marine emissions would equal between 56% and 71% of the NOx emissions from either freeway crossing the Columbia River. (Assuming that an upper bound for the daily fraction of truck traffic crossing the bridge might be 20% of total traffic, this would correspond to a lowerbound comparison for waterborne to freeway-bridge NOx emissions of 37%.) This comparison qualitatively demonstrates the potential significance of waterborne commerce emissions as one of the important mobile sources of NOx in the region, on the same order as major freeway segments. Using a parametric analysis, lower and upper bounds were modeled by changing the input parameters to the fuel consumption and emissions estimates from the validated base-case scenario. These three scenarios are geographically resolved to be used as input data sets for pollution and regional haze modeling. By evaluating the impact of these inventories in a regional haze model, pollution impacts attributable to commercial marine transportation can be evaluated. For example, if the upper-bound inventory does not show significant contribution to regional haze, that would provide strong evidence that CMV do not contribute to the problem in this region. However, if the lower-bound inventory shows significant impact in the regional haze model, then initial policy action to reduce emissions may be warranted.
The third possibility is that either one or both of the lowerbound and base-case inventories show little impact on regional haze while significant impacts may result from modeling the upper-bound inventory. This would provide support for additional research efforts that reduce the uncertainty in these inventories. In this case, the uncertainty analysis presented here could provide guidance. The uncertainty analysis shows greatest improvement in inventory accuracy for the Columbia and Snake River systems could result from three activities. First, there is a need for improved understanding of the vessel, engine, and duty-cycle characteristics of nontowboat vessels (primarily deep-draft vessels and tugboats). Second, towboat fuel characteristics (specifically fuel density) should be reviewed. Third, emissions testing should include more towboats to better characterize the fuel-based emissions factors with greater accuracy.
Acknowledgments This research was partly supported by the Department of Ecology, State of Washington, Contract C0100086. In kind support and initial review was provided by Chris Dager at the Tennessee Valley Authority (TVA). This research also received support from the Graduate College of Marine Studies at the University of Delaware and benefited from support by NSF Grant SBR9521914, which funds the Center for the Study of the Human Dimensions of Global Change at Carnegie Mellon University.
Supporting Information Available Review of current inventories, methodology detail for PM and SOx inventories from CMV, vessel traffic detail, and uncertainty analyses for PM and SOx. This material is available free of charge via the Internet at http://pubs.acs.org.
Literature Cited (1) Kean, A. J.; Sawyer, R. F.; Harley, R. A. J. Air Waste Manage. Assoc. 2000, 50, 1929-1939. (2) Corbett, J. J.; Fischbeck, P. S. Science 1997, 278, 823-824. (3) Corbett, J. J.; Fischbeck, P. S.; Pandis, S. N. J. Geophys. Res. 1999, 104, 3457-3470. (4) Corbett, J. J.; Fischbeck, P. S. Environ. Sci. Technol. 2000, 34, 3254-3260. (5) Arcadis Gerharty & Miller, Inc. Commercial Marine Activity for Deep Sea Ports in the United States; Assessment and Modeling Division, Office of Mobile Sources, U.S. Environmental Protection Agency: Washington, DC, 1999. (6) Analysis of Commercial Marine Vessels Emissions and Fuel Consumption Data; Office of Transportation and Air Quality, U.S. Environmental Protection Agency: Washington, DC, 2000. (7) Jacobs, B.; Stocum, J.; Aalbers, S. Oregon 1996 Hazardous Air Pollutant Emission Inventory Project: Emission Estimate Methodology Documentation; Oregon Department of Environmental Quality (DEQ), Air Quality, Technical Services: 2000. (8) Otterson, S. Visibility SIP Review, Appendix C: Emissions Inventory; Washington Department of Ecology: 2000. (9) U.S. Army Corps of Engineers (USACE). NDC Publications and U.S. Waterway Data CD, version CEWRC-NDC; Water Resources Support Center, Navigation Data Center: Alexandria, VA, 1999. (10) Corbett, J. J.; Robinson, A. L. Environ. Sci. Technol. 2001, 35, 1343-1349. (11) Bray, L. G.; Burton, M. TVA Barge Costing Model; Tennessee Valley Authority: Knoxville, TN, 1994. (12) Bray, L. G.; Burton, M.; Dager, C.; Henry, R.; Koroa, M. C. The Undervalued Social Benefits of Water Transportation; Tennessee Valley Authority Water and Hydro Resources Projects: Knoxville, TN, 1998. (13) Dager, C. Tennessee Valley Authority, Personal communication, January-February 2001. (14) Osbourne, A., Ed. Modern Marine Engineer’s Manual; Cornell Maritime Press: Centreville, MD, 1943; Vol. II. (15) International Organization for Standardization (ISO). International Standard, Petroleum Products-Fuels (class F)-SpecificaVOL. 36, NO. 6, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1305
(16) (17) (18)
(19)
tions of Marine Fuels; International Organization for Standardization: 1987. U.S. Environmental Protection Agency. AP-42: Compilation of Air Pollutant Emission Factors; U.S. Environmental Protection Agency: Research Triangle Park, NC, 1997; Vol. I. Morgan, G. M.; Henrion, M. Uncertainty: A Guide to Dealing with Uncertainty in Quantitative Risk and Policy Analysis; Cambridge University Press: New York, 1990. Columbia River Bridge Crossings 1961-2020, http://www.rtc.wa.gov/tc/brdgawd.htm; Southwest Washington Regional Transportation Council, 2001; current as of March 2001. Southwest Washington Regional Transportation Council (RTC). Transportation System Monitoring and Congestion Management Report; Southwest Washington Regional Transportation Council: 2001.
1306
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 6, 2002
(20) Carlton, J. S.; Danton, S. D.; Gawen, R. W.; Lavender, K. A.; Mathieson, N. M.; Newell, A. G.; Reynolds, G. L.; Webster, A. D.; Wills, C. M. R.; Wright, A. A. Marine Exhaust Emissions Research Programme; Lloyd’s Register Engineering Services: 1995. (21) U.S. Environmental Protection Agency. Title 40 CFR Parts 89, 92, and 94: Control of Emissions of Air Pollution from New CI Marine Engines at or above 37 kW, Final Rule; Environmental Protection Agency: Washington, DC, 1999. (22) Wright, A. A. Lloyd’s Register Engineering Services, Personal correspondence, November 1996.
Received for review July 6, 2001. Revised manuscript received December 30, 2001. Accepted January 8, 2002. ES0155985
