We report measurements of NOx, SO2, CO, and HCHO mass-based emission factors from more than 200 commercial vessel encounters in the Gulf of Mexico and the Houston-Galveston region of Texas during August and September, 2006. For underway ships, bulk freight carriers have the highest average NOx emissions at ∼87 g NOx (kg fuel)−1, followed by tanker ships at ∼79 g NOx (kg fuel)−1, while container carriers, passenger ships, and tugs all emit an average of about ∼60 g NOx (kg fuel)−1. Emission of NOx from stationary vessels was lower, except for container ships and tugs, and likely reflects use of medium-speed diesel engines. Overall, our mean NOx emission factors are 10–15% lower than published data. Average emission of SO2 was lower for passenger ships and tugs and tows (6–7 g SO2 (kg fuel)−1) than for larger cargo vessels (20–30 g SO2 (kg fuel)−1). Our data for large cargo ships in this region indicate an average residual fuel sulfur content of ∼1.4% which is a factor of two lower than the global average of 2.7%. Emission of CO was low for all categories (7–16 g CO (kg fuel)−1), although our mean overall CO emission factor is about 10% higher than published data. Emission of HCHO was less than 5% that of CO. Despite considerable variability, no functional relationships, such as emissions changes with engine speed or load, could be discerned. Comparison of emission factors from ships to those from other sources suggests ship emissions in this region cannot be ignored.
 Commercial marine vessels range in size from small fishing boats (20–30 m in length) to extremely large cargo vessels (>300 m in length). These ships almost without exception use diesel engines for propulsion and auxiliary power generation. The larger ships, comprising bulk carriers, tankers, container and vehicle carriers, utilize diesel engines that produce power in the 5 to 100 MW range. These engines typically consume heavy fuel oils (HFO) which are high in sulfur content (1–4.5% by weight). These engines are also extremely efficient, converting nearly all of the carbon in the fuel to CO2, but also emitting considerable amounts of NOx (NOx = NO + NO2), SO2, and particulate material (PM), and smaller quantities of CO and volatile organic compounds (VOCs) such as formaldehyde (HCHO). Emissions from these ships have been largely unregulated and uncontrolled and can be significant on local, regional, and global scales.
 In ports ship engines are left running not only to provide power for lights, electronics, and air conditioning, but also in many cases to power the cranes and pumps that perform loading and unloading operations. This causes local air quality degradation not only by direct emission of priority pollutants (e.g., CO and PM) but also from increased production of secondary pollutants, such as O3, due to emission of precursor species such as NOx and VOCs. For example, HCHO, both an HOx source for photochemical processing and a hazardous air pollutant according to the U.S. EPA, is directly emitted by a number of transportation sources [Herndon et al., 2006; Ban-Weiss et al., 2008a], including large marine vessels. Thus the Port of Los Angeles (http://www.portoflosangeles.org/environment/alt_maritime_power.asp) now requires that many ships connect to shore power and shut off their engines while at dock.
 On a regional scale ship emissions can impact efforts to reach air quality regulatory goals. For example, the county of Santa Barbara, CA, has performed inventory modeling that indicates little or no reduction of overall NOx emissions between 1999 levels and projected 2015 levels despite controls on land-based point and area sources. This is due to increased emissions from commercial shipping transiting the sea lanes just offshore (http://www.sbcapcd.org/itg/shipemissions.htm). And globally it has been estimated that commercial marine vessels account for as much as 30% of total anthropogenic production of NOx and up to 8% of total anthropogenic SO2 [Corbett et al., 2007; Eyring et al., 2005; International Council on Clean Transportation, 2007].
 International and national regulations are currently under consideration to restrict emissions from commercial vessels. The International Maritime Organization (IMO)  recently released a document that outlines new standards for emission of NOx, SO2, and PM globally. In addition, standards for establishment of emission control areas (ECA) have been revised which allow considerable local and regional reductions in ship emissions [Blenkey, 2009]. Sulfur ECAs (SECA) have been in place for a number of years for the North Sea and Baltic Sea regions. Recently, a proposal has been submitted to the IMO to establish a North American ECA to 230 miles offshore, although the California Air Resources Board has already established an ECA. This latter ECA requires all ships to switch to low sulfur fuel when within 24 miles of the coast.
 Emission inventories are commonly used to evaluate the significance of pollution sources. For shipping, emission inventories have been compiled on a global basis [IMO, 2000; Corbett and Köhler, 2003; Endresen et al., 2003; Eyring et al., 2005], for North America [Wang et al., 2007], for smaller regions such as the Baltic Sea [Stipa et al., 2007], and for ports located in or near major metropolitan areas [Corbett, 2002]. These inventories are calculated from 1) estimates or compilations of source activity (e.g., marine fuel consumption, vessel movement information, fleet makeup) and 2) emission factors that relate mass emission rates of NOx, SO2, and other species to the activity data. The most comprehensive data on emissions from marine engines comes from a Lloyd's Register study [Lloyd's Register Engineering Services, 1995] which used measurements from vessel exhaust stacks to compile emission factors based on fuel consumption and on power output. These emission factors are widely used in emission inventory modeling, though other data have been published [U.S. Environmental Protection Agency (U.S. EPA), 1997; Entec, 2002] and used in inventory work.
 A characteristic of virtually all studies of marine vessel emissions is the large variability observed in the emission factors. For example, limited observations from several studies [Hobbs et al., 2000; Sinha et al., 2003; Chen et al., 2005] show a range of NOx emission factors from 20 to over 100 g NOx (kg fuel)−1. The more extensive Lloyd's Register Engineering Services  study shows plots of NOx emissions from both slow-speed diesel (SSD) and medium speed diesel (MSD) engines as a function of engine load. The variation of measured emissions at any given load [Lloyd's Register Engineering Services, 1995, Figure 9] is from approximately 25% to well over a factor of two; however, their mass-based emission factors are reported as single (average) values without uncertainty bounds. The Lloyd's Register Engineering Services  study also shows that not all emitted NOx is NO; there is a small, but variable, direct emission of NO2. While it is clear that the variability in these data is much greater than the uncertainty of any given measurement, emission inventory models use average values of emission factors which translates variability in the measurements into uncertainties in the inventory outputs. One way of reducing these uncertainties is to parameterize the measured variability with observed functional relationships or categorizations. Since this approach requires a substantial data set, during the Texas Air Quality Study of 2006 (TexAQS 2006) we undertook to determine emission factors from a wide variety of ships under actual operating conditions, both underway and stationary, in the Houston Ship Channel and the Gulf of Mexico. In this paper we report the results of these measurements, compare our observations to previous work on commercial shipping emission factors, and compare emission factors from ships to those from other sources in the Houston-Galveston area (HGA).
 The measurements reported here were taken aboard the NOAA research vessel Ronald H. Brown (RHB) from early August through mid-September, 2006. The study area (Figure 1a) spanned the Gulf of Mexico, and included Galveston Bay and the Houston Ship Channel (HSC; Figure 1b). Excursions were made into the harbors of Point Comfort (Matagorda Bay), Freeport, Galveston, and Beaumont (all in Texas). In addition, considerable time was spent sampling at the container terminal at Barbours Cut just east of Houston, and along the upper reach (east-west section) of the HSC to the turning basin.
 Instruments were installed in standard 6 m (twenty foot) sea containers that had been modified into laboratory space. These laboratory vans were placed on a forward deck of the ship where adjacent towers were erected to hold sampling inlets at ∼20 m above water line. A PFA Teflon tube (15.9 mm i.d. by 10 m) was used to bring sample air from the tower top to the instrument van where continuous measurements of CO2, CO, and SO2 were made. The flow rate in the sample line was ∼100 standard (STP = 273 K and 1013 hPa) liters per minute (slpm) and provided a gas residence time of ∼1 s. Separate lines provided samples for measurement of the oxides of nitrogen (NOy) and for HCHO. Brief instrument descriptions are provided here and operating characteristics are summarized in Table 1. All data are 1 Hz measurements.
 Carbon dioxide was measured by a nondispersive IR technique using a commercial instrument (Li-Cor Model LI-7000). Air was drawn from the sampling manifold through a Nafion dryer (PermaPure) and then to the instrument. The instrument was operated in a high-precision mode with a reference gas mixture of 339 (±3) ppmv CO2 in synthetic zero air (Scott-Marrin). The instrument was zeroed hourly by overflow of the sample cell with reference gas. Sensitivity tests were done three times every 5 h by sample replacement with either midlevel (385 ± 4 ppmv CO2 in zero air; Scott-Marrin) or high-level (501 ± 5 ppmv CO2 in zero air; Scott-Marrin) calibration gases, or both sequentially in a calibration sequence. All CO2 gas mixtures were referenced to NOAA/ESRL/Global Monitoring Division standards before and after the field campaign. Instrument precision was estimated at 0.07 ppm at 1 Hz.
 Carbon monoxide was measured in dried sample air (see CO2 section above) by a UV fluorescence technique using a commercial instrument (AeroLaser Model AL-5002). Instrumental background was determined by passing the sample air through a catalyst (Sofnocat 514) that removed CO. Standard addition calibrations were performed every 2.5 h using a small flow (relative to the sample flow) of calibration gas (20.38 ppmv ± 1% CO in zero air; Scott-Marrin) that was referenced to NIST-certified and other calibration standards before and after the mission.
 Sulfur dioxide was measured by a pulsed UV fluorescence technique using a commercial instrument (Thermo Electron Corp. Model 43s). To improve time response the instrument was modified by use of a large vacuum pump to increase flow and by removal of the hydrocarbon trap to decrease sample gas residence time. The instrument background was checked every 5 h by routing the sample air through an annular denuder (URG, Inc. Model 591) coated with sodium carbonate solution. Calibration was performed every 5 h by standard addition of a small flow of 5.27 ppmv SO2 in zero air (±5%; Scott-Marrin) that was compared to other laboratory standards before and after the mission.
 Ship emissions of total reactive oxides of nitrogen (NOy) were measured by conversion to NO (325°C Au tube with 0.3% H2) followed by chemiluminescence NO-O3 detection. Instrument background was measured every 5 h by removal of NO in the sample gas with excess O3 prior to detection. Instrument calibration was performed every 5 h by standard addition of 4.98 ppmv NO in dry nitrogen (±2%; Scott-Marrin) that was referenced to other calibration standards before and after the mission. In addition, conversion by the gold tube of the principal NOy species (NO2, PAN, HNO3) was evaluated periodically by addition of calibrated gas standards to ambient air or zero air samples. While conversion of NO2 and PAN remained high (>90%), conversion of HNO3 degraded somewhat over the mission (as low as 50%) which required occasional cleaning and high temperature bake-out of the gold tube converter.
 Formaldehyde was determined by tunable infrared diode laser spectroscopy using a quantum cascade laser (QCL) instrument [Herndon et al., 2007]. The light source was a pulsed quantum cascade laser device with one instrument optical channel operating in the n2 absorption band of HCHO at 1765 cm−1 (5.67 μm) and a second optical channel using the C2H4 absorption line at 964.4 cm−1 (10.37 μm). The line strengths were taken directly from the HITRAN database [Rothman et al., 2005]. The total HCHO measurement systematic error is estimated at 9%, primarily from the line strength determination [Herndon et al., 2005].
 For the TexAQS study we installed an Automated Identification System (AIS) receiver (SeaCAS Model SafePassage) which continuously logged the data (Global Navigation Software Model NavPak with AIS logging). Most ships are equipped with AIS, which is designed as a collision avoidance system, and continuously broadcast information (identification, position, speed, etc.) over VHF (about a twelve mile radius). The logged vessel identification data were verified by reference to the Lloyd's Maritime Directory [Lloyd's Marine Intelligence Unit, 2006, 2008] and several online ship databases.
2.1. Measured Data
 The measured data consist of mixing ratios of NOy, SO2, CO, HCHO, and CO2 determined in exhaust plumes from vessels underway, anchored or docked. In many cases association of a ship with a particular plume was straightforward. For example, during transits along the HSC (Figure 1b) ships are required to stay within the approximately 150 m wide dredged channel. This requirement typically means that ships must travel in a single file both inbound and outbound. For this situation, with moderate oncoming wind, identification of the exhaust plume from a vessel was unequivocal. In some instances vessels will pass slower ships and identification of vessel plumes can be difficult. Perhaps the most difficult case is the transit along a long stretch of docks under light winds. Not only is it difficult to determine individual plumes and their sources, but the AIS transmissions from these docked vessels are considerably less frequent (if there is any transmission at all) so misidentification is much more likely. Of the more than 1100 plumes encountered during the study, 274 have been positively associated with a particular vessel.
2.2. Emission Factor Calculations
 The analysis used for determination of emission factors from the measurements is shown in Figure 2. Figure 2a shows the time series of mixing ratios of NOy, SO2, CO, and CO2 for a plume encounter that lasted about three minutes (1 s data are shown). Linear regression analysis is used to determine the slope and strength of the linear relationship between NOy and CO2 (Figure 2b), SO2 and CO2 (Figure 2c), and CO and CO2, (Figure 2d). The slopes from these fits (in ppbv compound per ppmv CO2) are then used to calculate the mass-based emission factor, which is mass of compound emitted per kilogram of fuel consumed. The calculation is based on two assumptions: 1) that all of the carbon in the fuel is converted to CO2 and 2) that the average mass fraction of carbon in the fuel is 0.865, or 3170 g CO2 per kg fuel. This latter value is the CO2 emission factor from the Lloyd's Register Engineering Services  study, and the 0.865 value is calculated from that emission factor, assuming all fuel carbon is converted to CO2. For NOy (which we report as NOx but calculated as equivalent NO2) the multiplier used to convert the measured slope to an emission factor is:
The ratio, umole/nmole, in the result accounts for the fact that most compounds are measured in ppbv, while CO2 is measured in ppmv. Similar calculations produce multipliers of 4.61 g SO2 (kg fuel)−1, 2.02 g CO (kg fuel)−1, and 2.16 g HCHO (kg fuel)−1. Also, there is a stoichiometric relationship between the SO2 emission factor and fuel S content: the emission factor divided by 20 is equal to the fuel sulfur content in percent by weight, assuming all fuel S is converted to SO2.
 There are two points to note here. One, since dilution affects both the compound of interest and CO2 equally this effect cancels out of the ratio (i.e., moles of air in both numerator and denominator of the slope). Second, these calculations will only be valid if there are no losses or the loss rates are equal for the compound of interest and CO2. With CO and CO2 chemistry and deposition occur very slowly, so we expect these values to be robust for the plume transit times encountered during this work. On the other hand, there is the possibility of conversion of reactive nitrogen into more oxidized forms during plume transit. Most reactive N in these plumes is emitted as NO, with a small fraction emitted directly as NO2. However, conversion of NO to NO2 can occur quite rapidly, depending on the levels of atmospheric oxidants such as O3 and HO2/RO2. Further conversion to other species such as HNO3 and PANs during the day or to NO3 and N2O5 during the night will be much slower. Spot checks of a few of the more intense plumes confirmed that NO3, N2O5, ClNO2, and PANs were absent. The alkyl nitrates and HNO3 instruments did not have sufficient time resolution to make this determination. Background levels of the NOy species varied considerably, but on average NOx was present at the highest levels (∼71% of NOy), followed by HNO3 (∼14%), PANs (∼3%), and all other measured species (∼2%; see above). Here we use the measurements of NOy, rather than NOx, to determine all of the reactive nitrogen species in the plumes. There also may be very small losses of some of these species via deposition to the surface or to aerosols, but we consider those effects to be negligible.
 It is a different situation with the emitted sulfur compounds. Sulfur compounds in the exhaust gas are derived wholly from the S content of the fuel. As noted above, combustion of fuel in these engines is virtually complete and the vast majority of sulfur emitted is in the form of SO2. However, there are data that show that “prompt” sulfate can form during combustion and emission from the exhaust stack [Lloyd's Register Engineering Services, 1995; Lack et al., 2009]. Moreover, emitted SO2 can be converted to H2SO4 as quickly as 20–30% per hour [Lack et al., 2009] by both homogeneous and heterogeneous pathways in the atmosphere. Thus our data must be considered lower limits for the S emission factors reported here. We note, though, that because of the close proximity of many of the ships encountered ∼80% of the plumes were less than 5 min old, and conversion of SO2 to sulfate will be a few percent at maximum. The emission of HCHO was small, and the number of plumes where the emission factor could reliably be calculated was considerably less than for the other species. Photochemical losses of this species for these short plume transit times were estimated to be < 1%.
2.3. Precision of the Results
 A nighttime experiment that was conducted during the TexAQS mission afforded us the opportunity to examine the reproducibility of the emission factor determinations. During the night (0500–0800 UTC; subtract 6 h for local standard time) of 8 August 2006, RHB was stationed off the Gulf Coast south of Freeport, TX. A crude oil tanker (Patriot; 53772 Gt; 248 m long) was anchored offshore. The vessel was producing power as evidenced by the lights on the ship and the exhaust plume. R/V Ronald H. Brown made a number of downwind transits across the plume from this vessel during the 3 h period (Figure 3a). Shown in Figure 3b and Table 2 are emission factors from seven plume encounters that were independent measurements made at different distances away from Patriot in a constant wind field. We assume that the target ship was operating under constant conditions so that the emissions were constant over time. Two determinations of CO emission factors were unreliable (r2 less than 0.5) and are not shown. Both NOx and CO emission factors show very little relative deviation which indicates that these species can be determined with good precision. Emission factors for SO2 show reasonably constant values for the first six hits and then deviates significantly higher for the last encounter. This was not an influence of the exhaust plume from RHB because low sulfur fuel was used by our ship. Since we assume that Patriot was operating under constant conditions, we use all of the data and calculate the precision for SO2 emission factors to be about 14%. These precision determinations are robust values since all factors that contribute to uncertainty, including instrument stability, sampling errors, and calculation errors such as slope determinations, are included. The only errors not accounted for are systematic errors, which we address in the next section.
Table 2. Emission Factors From the Patriot on 8 August 2006
 The total uncertainty in the emission factors is estimated via a propagation of errors calculation that includes all the known sources of error. These sources are: 1) instrumental uncertainties; 2) errors due to sampling; 3) errors in the least squares slope determinations; and 4) errors from assumptions associated with equation (1) above.
 For measurements of NOy, SO2, CO, and CO2 the total instrumental uncertainties (see Table 1) result from errors in calibration, errors in the background (i.e., zero) determination, and interferences. In all cases calibration was via standard addition with associated errors in the calibration standard (2–5%) and the flow controllers (1%). Background uncertainties contribute little since the measured levels in these plumes are typically much higher than the zero levels of the instruments. Interferences were only a problem with the SO2 instrument which has a 3% positive response to NO, based on laboratory measurements. This was corrected using the separately measured NO data. As noted above, to improve response time this instrument was run without the hydrocarbon trap, also called a kicker. Aromatic VOCs produce a nonnegligible response in the pulsed UV fluorescence technique and must be removed from the SO2 sample air [Luke, 1997]. However, independent measurements of VOCs (data not shown; J. de Gouw, personal communication, 2008) indicate that these compounds are not present in ship plumes, so this correction was neglected. For measurements of HCHO the uncertainty in the line strength determination is the major source of error in the QCL data.
 Sampling errors result from surface losses and chemical transformations during sample gas transit down the sampling lines as well as from different instrumental response times. Instruments that measured CO, SO2, and CO2 were connected to a common sampling manifold which provided a gas residence time of ∼1 s. Additional time, typically < 1 s, was required for the sample to reach the individual instruments in the van. For CO and CO2 the instrument response time was very fast, as shown by the high degree of correlation in the fine structure of these measurements during plume sampling (see Figure 2a). On the other hand, the SO2 instrument has a large analytical cell which results in somewhat slower response time and loss of correlated structure when sampling plumes, even when run at the higher flow and lower pressure during this study. There also appeared to be some memory effect of SO2 on the sampling lines, evidenced by the tailing structure when coming out of a large plume. The result is that correlation between SO2 and CO2 was somewhat degraded compared to the other measurements. The NOy inlet box was placed at the top of the tower near the inlet for the manifold. Within the inlet box the sample passed through the converter, a filter, and a mass flow controller before proceeding down to the chemiluminescence analyzer in the van. Since the pressure dropped considerably after the mass flow controller, the sample residence time was very short for NOy measurements; typically this was the first measurement to respond to a plume encounter. Also, the gold converter destroys O3 and other radicals in the sample air which eliminates conversion of NO during sample transit to the detector. Sampling and pressure reduction for the HCHO instrument was also performed at the top of the tower and resulted in response times similar to those of NOy. Lags in sample times between CO2 and the other measurements were obvious in the 1 s data and were corrected prior to analysis.
 For NOx, SO2, and CO emissions normal least squares regression analysis was employed to determine the slopes used to calculate the emission factors. Since the variance of the errors (uncertainties) in the CO2 data (x axis data in the regressions) was constant (see Table 1) and thus much smaller than the variance of the measured mixing ratio data, it was not necessary to use two-sided regression [Draper and Smith, 1981]. For the same reason, and because most of the correlation coefficients were near unity, the regressions were performed with unweighted data. The average relative uncertainty of all the slope determinations was approximately 5%, regardless of species.
 Analysis of HCHO emission factors was similar to the other species; however, since the emission of formaldehyde was considerably lower than other species there are fewer data points. The 1 s RMS noise in the HCHO measurement was < 0.250 ppbv. If the ship plume exhaust HCHO content (relative to CO2) were present at a molar ratio of 1 × 10−5, the “strength” of the observed plume would need to be enhanced by 25 ppmv of CO2 in order to be detected in the HCHO measurement at 1 s. When the enhancement in CO2 was below 5 ppmv the plume event was discarded from consideration. As a result, 85 plumes were included in the HCHO emission factor analysis. For 63 of the analyzed plumes, the concomitant increase in HCHO with CO2 resulted in measurable emission factors. When no apparent increase in HCHO was observed during a plume encounter with sufficient CO2 enhancement, an upper limit of the HCHO emission factor was estimated using the apparent instrument noise ratioed to CO2.
 The uncertainty in the fuel carbon content (0.865) is estimated at 1% based on the fuel carbon analyses of marine fuels (0.8661–0.8674) reported in the Entec  study. The uncertainty of the assumption that all of the fuel carbon is converted into CO2 is more difficult to estimate. After CO2, the largest emission of carbon we observed was due to CO, at an overall average of 13 g CO (kg fuel)−1. There were only minor emissions of other carbonaceous material: VOCs (<1 g (kg fuel)−1) and black carbon (<1 g (kg fuel)−1 [Lack et al., 2008]). From these data we estimate an uncertainty of about 1% from this assumption.
 To determine the total uncertainty in the emission factors, we include 1) the instrumental uncertainties shown in Table 1 as estimates of systematic error, 2) the 5% uncertainties from the slope determinations, 3) the (combined) 2% uncertainty in fuel carbon value and the assumption of all fuel carbon converted to CO2, and 4) the relative standard deviations from the Patriot encounter (see Table 2) as estimates of precision. From standard propagation of errors calculations, the total RMS uncertainties are: 14% for NOx, 18% for SO2, 12% for CO, and 14% for HCHO. These are the uncertainties shown as error bars on the plots of emission factors in the next sections. Due to the lower emissions of HCHO, the uncertainty in the slope determination dominates other measurement uncertainties so that overall uncertainty depends upon the magnitude and duration of the plume for each encounter. Evaluation of HCHO emission factors via a ratio of integrals approach was also done for those plumes with measurable HCHO. Linear regression of the emission factors by slopes versus EFs by integrals gave a slope of 1.06. The HCHO emission factors reported here were calculated by the slope method because the errors associated with the integration technique were somewhat larger due to difficulty in determination of the baseline.
3. Results and Discussion
 The logged AIS data provided the name, position, and speed of each ship encountered. Using this information in conjunction with available ship databases we determined the vessel type, dimensions, rated power of the main propulsion engine, and the vessel design speed. From this we calculate an approximate load factor as the ratio of speed from the AIS to the vessel design speed. This was not possible to do with tugboats and tow vessels because 1) the vessel information is not readily available and 2) the calculation, in general, would be in error because the loads on these vessels are not generally a function of speed. Also, the draft of a cargo vessel must be considered in the load calculation since a heavily loaded ship will require more power than an empty ship. Although the broadcast AIS data has a field for vessel draft, in many cases this was missing or clearly erroneous, as occasional visual sightings of vessels confirmed. Thus the load factor that we report is only approximate.
 In the discussion below the emission factor data are categorized by species emitted and by vessel type and are shown as a function of load factor except for tugs and tows, which are shown as a function of vessel speed. Our categories of vessel types are based on the types of ships listed in the ship databases, primarily the Lloyd's Maritime Directory [Lloyd's Marine Intelligence Unit, 2006, 2008], and on categorizations from current emissions models, such as STEEM [Wang et al., 2007]. While there are many different types of ships, we have defined six categories that account for all of the ships observed in this study.
 We address here the question of how representative our data are with respect to the global commercial ship fleet. Data for the 2007 world merchant fleet from Lloyd's Maritime Directory [Lloyd's Marine Intelligence Unit, 2008] provides the following relative composition by vessel type: bulk/cargo, 26.6%; container/ro-ro, 8.6%; tanker ships, 14.3%; passenger, 5.1%; reefer, 1.5%; other, 43.8%. Our data, grouped into similar categories, provides the following: bulk/cargo, 8.7%; container/ro-ro, 7.7%; tanker ships, 22.6%; passenger/supply, 7.7%, tugs/tows, 53.4%. We find a considerably higher proportion of tanker ships in the Houston-Galveston region than expected based on the world fleet average. This is not surprising since this area has a large number of refineries and chemical processing facilities. The high level of industrial activity in this area is also a reason for the high proportion of tugs and tows in our data (i.e., barge movements).
 Finally, the vessel speeds we encountered ranged from < 1 knot to over 20 knots. Despite the relatively narrow dredged channel of the HSC, pilots routinely navigate the large cargo ships at 15–20 knots unless visibility or other vessel traffic dictate otherwise. These vessel speeds were also typical for ships in the Gulf, where pilots were picked up or dropped off. For the large ships (container carriers, crude oil tankers, ro-ro) these speeds can be 40–80% of the vessel ocean cruising speed. For smaller ships (bulk freighters and smaller tanker ships), 15–20 knots can be 70–90% of cruising speed. Thus our data are typical of coastal and port traffic, but not representative of open ocean traffic.
3.1. NOx Emissions
Figure 4 shows a summary of the NOx emission factors. For ease of visual comparison all of the plots in Figure 4 have the same axes scales except for the x axis for the tugs and tows category, which shows speed not load. In general, the emission of NOx from these ships appears to be unrelated to the load on the vessel. Even for load factors of zero, which indicate a docked or anchored ship, there is a large amount of variability. The observed variability at almost any load or category is at least a factor of two and in some cases as much as a factor of six. The one exception to this is for passenger and supply ships (Figure 4e) which appear to have relatively constant emission of NOx regardless of load.
 The tugs and tows category is by far the largest in our data, and reflects the constant shipping activity not only in the HGA but all along the Gulf Coast and Intracoastal Waterway. This category is comprised of both vessel assist tugs as well as tugs pushing or towing lines of barges, and in either case the load on the tug cannot be estimated from the speed. However, the majority of these ships that were encountered in this study were under load, in most cases pushing or towing a line of barges in the HSC. Corbett and Robinson  report an average NOx emission factor of 70 ± 4.2 g NOx (kg fuel)−1 from a main (high-speed diesel) engine of an inland waterway towboat. Although somewhat higher than our average for underway tugs (61.2 ± 23.1 g NOx (kg fuel)−1), their value is well within the observed variability in our data.
 A summary of the data in these plots is shown in Table 3, where the data are segregated by category, emitted species, and vessel speed. The differences observed in mean NOx emissions for underway versus stationary ships are significant at the 95% level for the vessel categories of bulk freighters, crude oil tankers, and products tankers, but are not significantly different for container ships or tugs. For ships that are underway, bulk freight carriers emit the most average NOx (87.0 ± 29.6 g NOx (kg fuel)−1), followed by either category of tanker ships (∼79 ± 23 g NOx (kg fuel)−1), while the other three categories are all at ∼60 g NOx (kg fuel)−1. The lower NOx emission rates for stationary vessels are probably due to the use of auxiliary engines which are typically medium-speed diesel engines. For underway vessels the NOx emissions come predominantly from the main propulsion engines. However, if auxiliary engines are also in use then the exhaust streams cannot be distinguished and the resulting NOx emission factor will be a composite of engine exhaust emissions.
Table 3. Emission Factor Statistics by Vessel Categorya
NOx E.F., g NO2 (kg fuel)−1
SO2 E.F., g SO2 (kg fuel)−1
CO E.F., g CO (kg fuel)−1
Speed = 0
Speed > 0
Speed = 0
Speed > 0
Speed = 0
Speed > 0
E.F. is emission factors. Shown are the mean and standard deviation with the number of points in parentheses, followed by the median value.
One extreme outlying point (>250 g CO (kg fuel)−1) has not been included.
 Since SO2 emissions arise solely from combustion of sulfur in the fuel, we expect no relationship between the operating state of an engine (i.e., load) and emission of SO2. This expectation is borne out by the plots shown in Figure 5. The only apparent differences in these plots is that the overall magnitude of sulfur emissions is greater for the larger ships (Figures 5a−5d) than for the smaller vessels (Figures 5e and 5f). This suggests that the larger ships with slow-speed diesel engines generally burn residual fuels which are highest (1–5%) in sulfur content. The smaller ships that have medium-speed or high-speed diesel or diesel-electric engines generally burn lower (<1%) sulfur distillate fuels. Again, the SO2 emission factor divided by 20 equals the fuel sulfur content in weight percent. Finally, there appears to be no distinction in SO2 emissions from underway versus stationary ships. This is not a surprising result since in the HGA there are no regulations that require ships to switch to low-sulfur fuel, a transition that would increase operating costs for the ship. As shown by the data in Table 3 the mean SO2 emission rates are not different (t test; 95% level) between moving or stopped ships. The one exception to this is crude oil tankers, where mean SO2 emissions from stationary vessels are greater by almost a factor of two than for similar underway vessels. Examination of NOx and CO emissions for this category show behavior consistent with the other vessel types. We considered the possibility that high ambient levels of aromatic hydrocarbons (e.g., from loading operations) might cause a positive interference in the pulsed fluorescence SO2 measurement. In fact, this has been determined from laboratory tests with our SO2 instrument to be on the order of 5%, depending on the aromatic hydrocarbon(s) present. However, because the emission factor is evaluated over a narrow plume any interference from large background levels will cancel when the slope of SO2 versus CO2 is determined (see Figure 2). At this time we have no explanation for the higher SO2 emission factors observed for stationary crude oil tankers.
3.3. CO Emissions
 In general, emissions of CO are low from commercial vessels. Figure 6 shows the measured CO emission factors by vessel category, and, with the exception of a few outlying points, the data cluster in the 0–10 g CO (kg fuel)−1 range. Nor are there any differences (t test; 95% level) between stationary or underway ships, except for the crude oil tanker category. However, inspection of Figure 6b shows that for speeds greater than zero the vast majority of the points are in the 10 g CO (kg fuel)−1 range, consistent with the other vessels. The largest number of outlying points is in the tugs and tows category. In fact, two extreme points (>250 g CO (kg fuel)−1) were not used in the calculations for Table 3. Possible explanations for the scatter in this category are 1) larger sample size, 2) there are more older vessels, or 3) these vessels not well maintained. From visual contacts of these ships during the cruise, we speculate that possibilities 2 and 3 above have some merit. Finally, as noted above, overall average CO emission is 13 g CO (kg fuel)−1 which corresponds to approximately 0.5% of fuel carbon content.
3.4. HCHO Emissions
 Of the 85 plumes that were analyzed for HCHO, 63 exhibited HCHO signals that could be correlated with CO2 in the plume. Figure 7 shows the results from an encounter with a tugboat moving at 7 knots on August 12. Though the HCHO data exhibit more noise than the CO2 data, correlation in the fine structure is clearly evident. For these plumes the emission factors ranged from 0.10 to 0.72 g HCHO (kg fuel)−1, with a mean of 0.15 g HCHO (kg fuel)−1. The other 22 plumes had significant CO2 enhancement but no detectable HCHO. For these plumes an upper limit to the HCHO emission factor was calculated from the ratio of the noise level of the HCHO QCL instrument to the measured CO2 enhancement. These results ranged from 0.04 to 0.21 g HCHO (kg fuel)−1, with a mean of 0.11 g HCHO (kg fuel)−1. These are extremely low values and are unlikely to be important to overall air quality in this region. On the other hand, since emission from ships is constant and tends to be concentrated in port and dock areas, this source of HCHO may be important on a local scale. For example, a build-up of formaldehyde during the night from ship emissions can provide an additional morning source for the photochemistry that makes O3 because photolysis of HCHO is a direct source of the HOx radicals involved in that chemistry.
 We also examined the fraction of HCHO directly emitted versus the fraction produced from photochemical conversion during plume transit prior to our measurements. Our analysis indicates that most of the plumes sampled at night are at or below the median HCHO emission factor for the overall sample. The daytime plumes are at or above the median HCHO emission factor for the overall sample. There are two potential reasons for this apparent dependence: a different use/activity or in-plume photochemical processing. It is possible that daytime ship operational activity was sufficiently different than nighttime use in a manner that led to different HCHO emissions. The data are also consistent with some degree of atmospheric processing, leading to an in-plume production of HCHO. However, these plumes are typically < 5 min old when sampled so processing time is minimal. It is likely that the presence of HCHO in these plumes is due to a combination of direct emissions and in-plume processing.
 Since formaldehyde is not present in the fuel, a chemically reasonable explanation for the presence of HCHO in the exhaust plumes is that combustion of the fuel in the engine is incomplete. Another marker for incomplete combustion is CO, and shown in Figure 8 is a plot of the relationship between these two species. The majority of the HCHO emission factor data fall between 0.5% and 5% of the CO emission factors. This range in the HCHO/CO ratio may be related to ship engine operational status.
3.5. Emission Factors Versus Engine Type
 The widely cited Lloyd's Register Engineering Services  study of marine vessel emissions made a distinction between two broad categories of sources: medium speed diesel (MSD) engines and slow speed diesel (SSD) engines. This division appeared to be based on the observation that NOx emission factors were generally lower for medium-speed engines than for slow-speed engines, although the variability in both cases was large. No such distinction was observed for the other species measured; however, it was clear that at low engine loads emissions of both CO and VOCs increased and in some cases quite significantly for both types of engines.
 Our data, aggregated by the engine categories of SSD and MSD, are shown in Table 4 along with emission factor data from the literature. Based on data from the Lloyd's Register Engineering Services  study, our distinction was based on the power rating of the main engine, where all engines greater than 5 MW were classified as SSD and vessels with main power rating 5 MW or less are MSD engines, except for passenger and supply ships. Thus our SSD classification includes all of the bulk freighter, container vessel, and crude oil tanker categories and most of the oil product tanker category. The MSD classification includes all of the tug and tow category and some of the oil product tanker category.
Table 4. Comparison of Measured Emission Factors and Literature Dataa
NOx E.F., g NOx (kg fuel)−1
SO2 E.F., g SO2 (kg fuel)−1
CO E.F., g CO (kg fuel)−1
Data from this study are shown as mean, standard deviation, and numbers of points in the upper row, and the median value in the lower row of each cell. SSD, slow speed diesel; MSD, medium speed diesel; HSD, high speed diesel; HFO, heavy fuel oil (fuel sulfur = ∼2–3%); MDO, marine diesel oil (fuel sulfur = ∼1%).
Data are for underway vessels only. SSD emission from this work contain some MSD exhaust from auxiliary engines.
Two extreme outlying points have not been included.
NOx data from these sources are based on in-stack measurement and corrected to standard temperature and humidity and weighted according to a standard test cycle.
Table 4 shows the averages of our observed emission factors (not including stationary vessels), the average values from the Lloyd's Register Engineering Services  study, and data for combinations of engine/fuel types from the Entec  study. For SSD engines, our average NOx emission factor is lower by about 10% to that from the Lloyd's Register Engineering Services  study and lower by 15% to that (SSD/HFO) from the Entec  work. The lower average value in our (more recent) work may reflect an increase in the use of NOx emission controls since the previous two studies were published. For MSD engines our average NOx emission factor is within 9% of each of the previously reported values. Also, the Entec  report includes data on high speed diesel engines which is shown as the HSD/MDO entry in Table 4 and is comparable to the passenger/supply ship category in Table 3. The two data points essentially agree, being within 3% of each other.
 Emission factors for SO2 are a function of fuel S content and not engine load so no relationship is expected. Emission of SO2 from MSD and HSD engines is considerably lower, on average, than from the SSD engines. The average of our MSD data, 6.3 g SO2 (kg fuel)−1, translates to an average fuel S content of 0.32% which agrees well with an earlier survey [U.S. EPA, 1999] of distillate marine fuels in this region that showed the average fuel sulfur content to be 0.36%.
 On the other hand, for SSD engines the average of our SO2 data is 27.8 g SO2 (kg fuel)−1, which translates to an average fuel sulfur level of 1.4%. This is about a factor of two lower than the 2.7% average estimated for HFOs on a global basis [International Council on Clean Transportation, 2007]. There are a number of possible reasons for this discrepancy. First, the measurements may be in error. This is unlikely because of the agreement in expected versus measured fuel sulfur in distillate fuels (see above). Moreover. if the cargo vessel SO2 emission factors are about a factor of two low, then many of our highest data points would reflect fuel sulfur levels that are totally unreasonable, 7% or higher, since current IMO regulations have capped residual fuel sulfur content at 4.5%. Second, the fuels consumed by ships we encountered in this region may not reflect, statistically, the global average. That this is possible can be seen from a fuel survey [Entec, 2002] that shows an average fuel sulfur of 1.91% from 50 analyses of residual fuel oils from ships in northern EU waters. Clearly, there are regional differences from the global average. Third, many large cargo ships (perhaps most ships in general) have multiple fuel tanks and use different grades of fuel. This must be the case for ships operating in and out of EU waters where SECA regulations require switching to low sulfur (i.e., more expensive) fuel, while no such regulation is in effect for ocean transits. Although low sulfur fuel requirements were not in place during the TexAQS 2006 study, it is possible that vessels could burn lower sulfur fuel while in coastal waterways or ports.
 Our average data for CO show no distinction between SSD and MSD engines, and are 50–60% higher than the Lloyd's Register Engineering Services  average value. However, the presence of a few outlying points of high emissions skews the average; the median data values are within about 10% of the Lloyd's Register Engineering Services  average emission factor. Our median data are about a factor of two higher than the Entec  CO emission factors. Even if the outlying CO emission factors (>20 g CO (kg fuel)−1 were removed from the data, our measurements are still closer to the Lloyd's results than to the Entec data. But, in contrast to the Lloyd's Register Engineering Services  results, we find no increase in CO emission at low vessel load or speed.
3.6. Significance of Ship Emission Factors
 To provide context for the significance of emission factors for commercial vessels, in Table 5 we compare emission factors from commercial marine vessels to those from other mobile sources: on-road gasoline and diesel engines and jet aircraft engines. Clearly, the large diesel engines used in shipping emit significantly more NOx than the other sources. Although the data are not shown, this is also true for SO2 emissions because U.S. EPA regulations cap sulfur in motor fuels at less than 0.00005%, compared to 0.5−4.5% in marine fuels. Sulfur is limited in jet fuels due to the corrosive effects of sulfur on the turbine blades. Emission of CO is considerably greater from on-road engines than from ship engines, probably due to the higher combustion efficiency of the latter. Emission of HCHO is uniformly low across these sources, although an emission inventory from California found that diesel engines were the largest direct anthropogenic source of HCHO in 2006 [California Air Resources Board, 2007].
 Emission factors are a large part, but only one part, of the equation to estimate the impact of emissions from commercial shipping. Activity factors and fuel consumption information are critical for accurately assessing the strength of these sources. This was recently pointed out by Corbett  who compiled an inventory of shipping emissions in the northwest U.S. His analysis, which used a “bottom up” fuel consumption approach that included vessel and cargo movement information along with existing emission factors, showed that NOx emission from ships in this region was 2.6 times greater than previously estimated. The accuracy of the estimate was improved in large part because the resolution of the activity data was finer than used in previous inventories. A conclusion from this work was that inventory uncertainty could be further improved with better characterization of marine engine emissions. Our measurements tend to confirm that view in that we observe large variability in emission factors. Because emission inventories use average values as emission factor inputs, this variability translates into uncertainty in the inventory outputs. Although our data do not reveal parameterizations that might capture some of this variability, there are clearly reasons for it. Undoubtedly, this will be an ongoing effort that will benefit from further research.
4. Summary and Conclusions
 During TexAQS 2006 our measurements of gas-phase species on board the NOAA R/V Ronald H. Brown allowed us to characterize the emissions of NOx, SO2, CO, and HCHO from a large number of commercial marine vessels. These measurements provided the means to calculate mass-based emission factors. With the information broadcast by these vessels over the Automated Information System we have unequivocally determined emission factors for over 200 vessels both at dock and underway.
 Our measurements confirm that these ships are significant sources of NOx. However, our mean emission factors are 10–15% lower than previously published average values for NOx, although the observed variability is large (30–40% relative standard deviation). For most vessel categories the emission of NOx appears to be lower for vessels when docked or anchored. In general, though, no robust functional relationships were found between NOx emission factors and ship variables such as vessel speed or load.
 Emission factors for SO2 were also observed to have large variability, but since emission of this species depends strictly on the sulfur content of the fuel used by the ship the variability reflects differences in fuel composition. Our data indicate that, on average, smaller vessels such as tugs and passenger ships consume fuel with a sulfur content of ∼0.3%, while large cargo ships in the Houston-Galveston region consume fuel with an average sulfur content of ∼1.4% by weight. While this is consistent with the use of more refined fuel (e.g., MDO or MGO) in the smaller vessels and bunker fuels (e.g., HFO) in the large cargo ships, the average bunker fuel sulfur value from this study is approximately a factor of two lower than the global average of 2.7%. No significant differences in SO2 emissions were found between underway versus stationary ships.
 Emission factors for CO from these ships were generally lower than those for on-road sources. Mean values for ships were about a factor of two greater than median values due to a small number of data points with much higher emission rates. Our median emission factors agree to within about 10% to the average value from the Lloyd's Register Engineering Services  study, but are a factor of two higher than the Entec  results.
 Emissions of HCHO were very low: less than 5% that of CO. However, even these low emission rates may be significant given the reactivity of this compound, especially since photolysis of HCHO directly produces the radical species that promote O3 formation in the troposphere. In large ports where ship engines are constantly operating, emissions of HCHO through the nighttime hours may produce atmospheric levels that could provide an additional source of radicals upon sunrise.
 In conclusion, our data largely confirm published average emission factors, but also show significant variability especially with emission of NOx. Even with the large number of observations in our data, no functional relationships could be found to provide parameters or categories to reduce the observed variability. Simple categorizations by vessel type or engine type did not significantly reduce the variance compared to the overall data set. Because of this variability the use of average emission factors in inventory modeling can result in increased uncertainty in the emissions output by the models versus those models with finer resolution input parameters and emission factors.
 We thank the officers and crew of the NOAA R/V Ronald H. Brown for their cooperation and enthusiasm. We thank three anonymous reviewers for their comments and suggestions which improved this paper. One of us (E.J.W.) also thanks Robert Harley for reading an early draft and pointing out an error. This research was funded by the Air Quality and the Climate Research and Modeling Programs of the National Oceanic and Atmospheric Administration (NOAA) and by the Texas Commission on Environmental Quality (TCEQ) under grant 582-8-86246.